{"meta":{"query_hash":"d42cdce44dd2","filters":{"venue":"Medical Decision Making"},"cohort_total":325,"direct_labels_cover":1,"predictions_cover":325,"exported":325,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/d42cdce44dd2","api":"https://metacan.xera.ac/api/v1/cohort?venue=Medical+Decision+Making"},"results":[{"id":"W1454465745","doi":"10.1177/0272989x15591966","title":"Quantifying Parameter Uncertainty in EQ-5D-3L Value Sets and Its Impact on Studies That Use the EQ-5D-3L to Measure Health Utility","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Institute for Clinical Evaluative Sciences; Health Sciences Centre; Hospital for Sick Children; Public Health Ontario; Hamilton Health Sciences; SickKids Foundation; University of Toronto; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Agency for Healthcare Research and Quality; U.S. Department of Health and Human Services","keywords":"Statistics; EQ-5D; Mean squared error; Mathematics; Prediction interval; Econometrics; Sensitivity analysis; Regression; Bayesian probability; Population; Variance (accounting); Regression analysis; Uncertainty analysis; Medicine","score_opus":0.8282736855108838,"score_gpt":0.5426143829996953,"score_spread":0.28565930251118843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1454465745","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9438601,0.007338716,0.0047382754,0.041394033,0.0011726571,0.0012118083,0.00012388031,0.000049330378,0.00011117525],"genre_scores_gemma":[0.9713503,0.00036322876,0.0014936508,0.026474705,0.00017801847,0.00007117772,0.000007165073,0.000036173304,0.000025550624],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9922581,0.0015597057,0.0035416307,0.0010087735,0.0008022734,0.0008295377],"domain_scores_gemma":[0.9832166,0.013584053,0.001321559,0.00085421215,0.00016202666,0.00086157495],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.062409416,0.00037175338,0.0016209282,0.00048022956,0.00035977762,0.00022069826,0.0005325062,0.00026953453,0.00014917337],"category_scores_gemma":[0.085331134,0.00028799253,0.00016817995,0.00046923873,0.000121518475,0.00045591616,0.00031806796,0.00065142894,0.00046254674],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010253985,0.0005274528,0.7036615,0.00071205274,0.00039666254,0.00005199799,0.030775515,0.01222436,6.248226e-7,0.023389068,0.11222885,0.11500654],"study_design_scores_gemma":[0.0034242761,0.00078088883,0.6683146,0.004149372,0.000017636794,0.00004387438,0.007155394,0.25410786,0.0000018601354,0.041511316,0.019482993,0.0010099338],"about_ca_topic_score_codex":0.0012091602,"about_ca_topic_score_gemma":0.0017011943,"teacher_disagreement_score":0.24188352,"about_ca_system_score_codex":0.0013873135,"about_ca_system_score_gemma":0.00050336064,"threshold_uncertainty_score":0.9999572},"labels":[],"label_agreement":null},{"id":"W1599059832","doi":"10.1177/0272989x15598743","title":"What Is Insufficient about Validation?","year":2015,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science","score_opus":0.37076626170763816,"score_gpt":0.4802594566583135,"score_spread":0.10949319495067533,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1599059832","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001599206,0.010701532,0.028186949,0.94172734,0.011649617,0.0007746462,0.0002935531,0.00012340586,0.0049437485],"genre_scores_gemma":[0.0016820417,0.0009879501,0.0032255389,0.9829235,0.0075380034,0.000109815664,0.00026159128,0.00012267214,0.0031489043],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9881149,0.00058009574,0.0075058313,0.0014641756,0.0014716029,0.0008634036],"domain_scores_gemma":[0.9895874,0.0045091663,0.003732721,0.0014431556,0.00027661232,0.00045094328],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.03411198,0.0005051533,0.0020464037,0.00092846365,0.00026521567,0.0007654139,0.0013144139,0.0019630417,0.013125037],"category_scores_gemma":[0.02136544,0.0005763458,0.00035313197,0.00042515746,0.00015859217,0.0008476747,0.00036944935,0.0020814217,0.022715226],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000006156516,0.000044324668,0.00029102765,0.00020505975,0.000046556226,0.00005288401,0.00084605877,0.000024557443,1.0422922e-8,0.0014961876,0.98929423,0.0076929685],"study_design_scores_gemma":[0.0005362414,0.00005549099,0.00019394777,0.0020626322,0.000010995957,0.000026645026,0.00021473442,0.00451283,2.7451597e-7,0.040413838,0.951418,0.0005543765],"about_ca_topic_score_codex":0.00007620823,"about_ca_topic_score_gemma":0.000011698493,"teacher_disagreement_score":0.04119614,"about_ca_system_score_codex":0.0012916484,"about_ca_system_score_gemma":0.00075900514,"threshold_uncertainty_score":0.9996688},"labels":[],"label_agreement":null},{"id":"W1812765017","doi":"10.1177/0272989x15604158","title":"Lessons Learned from a Cross-Model Validation between a Discrete Event Simulation Model and a Cohort State-Transition Model for Personalized Breast Cancer Treatment","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; University of Alberta","funders":"","keywords":"Breast cancer; Medicine; Cohort; Quality-adjusted life year; Actuarial science; Computer science; Cost effectiveness; Cancer; Risk analysis (engineering); Internal medicine; Economics","score_opus":0.5411688471614783,"score_gpt":0.5412581120805278,"score_spread":0.00008926491904948097,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1812765017","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39194578,0.00024424412,0.59515804,0.009303896,0.00007878761,0.00062709814,0.002588125,0.000033180968,0.000020824831],"genre_scores_gemma":[0.9789276,0.00017363274,0.018205421,0.001527095,0.00022783474,0.0003474062,0.00031024328,0.000047801634,0.00023297625],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9957593,0.00015645074,0.0024457048,0.0008292352,0.0004264191,0.00038286007],"domain_scores_gemma":[0.99666756,0.001429955,0.0009572355,0.00032573618,0.000174809,0.00044468796],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.007945113,0.00025872755,0.00092759775,0.00021076512,0.00026085993,0.00017129793,0.00017239436,0.00025926332,0.00009575299],"category_scores_gemma":[0.0019732115,0.00027266302,0.00015995667,0.00010668136,0.00009112061,0.00056349457,0.000052279625,0.00014451527,0.000039106075],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025831218,0.000058024954,0.012687028,0.000040049326,0.000080864535,4.656922e-7,0.0040014335,0.96865946,0.0000014130327,0.001950555,0.00028165063,0.011980724],"study_design_scores_gemma":[0.0026207962,0.000040231374,0.0047438056,0.00020106812,0.00003617548,8.5675487e-7,0.00009711278,0.8363389,7.2727926e-7,0.15553248,0.00015197175,0.00023588515],"about_ca_topic_score_codex":0.0005677972,"about_ca_topic_score_gemma":0.00018734603,"teacher_disagreement_score":0.58698183,"about_ca_system_score_codex":0.0011783002,"about_ca_system_score_gemma":0.00053716905,"threshold_uncertainty_score":0.9999726},"labels":[],"label_agreement":null},{"id":"W1856101391","doi":"10.1177/0272989x15600708","title":"Does Introducing Imprecision around Probabilities for Benefit and Harm Influence the Way People Value Treatments?","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital; University of British Columbia; Memorial University of Newfoundland","funders":"","keywords":"Harm; Population; Actuarial science; Value (mathematics); Confidence interval; Sample (material); Psychology; Medicine; Econometrics; Statistics; Social psychology; Mathematics; Economics; Environmental health","score_opus":0.08853594138880048,"score_gpt":0.41667049148711355,"score_spread":0.32813455009831305,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1856101391","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.62846696,0.00012169708,0.36931247,0.0008248108,0.0002374898,0.0007317957,0.000008047834,0.00016294548,0.00013378166],"genre_scores_gemma":[0.7785849,0.000060707025,0.22040878,0.00025683083,0.00022050149,0.00024144354,0.0000023684522,0.00003338724,0.00019109772],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9974824,0.00007571987,0.0005843699,0.00048035377,0.0010172192,0.0003599093],"domain_scores_gemma":[0.99123347,0.0074995942,0.00018855958,0.00060368376,0.00027568947,0.00019898891],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0023705906,0.00024096465,0.00038714625,0.00011054717,0.00023949677,0.0001375036,0.00042152533,0.00017262743,0.000057043137],"category_scores_gemma":[0.024656182,0.0001118497,0.00007319613,0.00022697171,0.00017942613,0.00033494283,0.0003698972,0.00025312474,0.000004966637],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000311423,0.00017088784,0.0027320613,0.00011553639,0.00005124985,0.000010884344,0.004958636,0.0002353172,0.00027507058,0.16529357,0.0019892391,0.8238561],"study_design_scores_gemma":[0.0005082048,0.0002393355,0.00053474086,0.0005719722,0.00003951676,0.000023078705,0.0009610058,0.0062735775,0.0005935978,0.9885267,0.0015523462,0.00017589734],"about_ca_topic_score_codex":0.000023553404,"about_ca_topic_score_gemma":0.00008072232,"teacher_disagreement_score":0.8236802,"about_ca_system_score_codex":0.0001779924,"about_ca_system_score_gemma":0.00008009964,"threshold_uncertainty_score":0.98355955},"labels":[],"label_agreement":null},{"id":"W1930633860","doi":"10.1177/0272989x15593083","title":"A Decision Analysis of Percutaneous Left Atrial Appendage Occlusion Relative to Novel and Traditional Oral Anticoagulation for Stroke Prevention in Patients with New-Onset Atrial Fibrillation","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Sciences Centre; University of Toronto; University of Ottawa; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Apixaban; Left atrial appendage occlusion; Dabigatran; Warfarin; Rivaroxaban; Atrial fibrillation; Quality-adjusted life year; Stroke (engine); Cost effectiveness; Intensive care medicine; Internal medicine; Risk analysis (engineering)","score_opus":0.08093161917929763,"score_gpt":0.3684787931947798,"score_spread":0.28754717401548213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1930633860","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6564118,0.000026668895,0.3420925,0.00026196992,0.00023506972,0.0009008351,0.00004000027,0.000016794676,0.000014284547],"genre_scores_gemma":[0.9426265,0.000006059257,0.05648835,0.000045803667,0.00037776085,1.5008328e-7,0.00036094192,0.000018844023,0.000075624666],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9965668,0.00007209501,0.0008748001,0.00044024497,0.0018321739,0.00021392456],"domain_scores_gemma":[0.99744475,0.00149834,0.0003004989,0.00019164228,0.0002685815,0.00029616168],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010620543,0.00019789282,0.00070694013,0.0010700393,0.00008637366,0.000035348927,0.00008200205,0.00021567107,0.0001933004],"category_scores_gemma":[0.004519468,0.00015706367,0.00033894944,0.00083530834,0.000045186564,0.00027344178,0.00010134127,0.00012289658,0.0000046986843],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.018235015,0.000026631991,0.88879645,0.00001546056,0.00032838195,0.0000057701372,0.00034971407,0.015180204,0.000029974566,0.0002695045,0.00014913575,0.076613754],"study_design_scores_gemma":[0.011946984,0.001342137,0.899503,0.000401541,0.001078082,0.0000143885445,0.000059939244,0.08328077,0.0000033704835,0.0018005834,0.00039671254,0.00017251854],"about_ca_topic_score_codex":0.000020933852,"about_ca_topic_score_gemma":0.00015587092,"teacher_disagreement_score":0.28621462,"about_ca_system_score_codex":0.00016411203,"about_ca_system_score_gemma":0.00019407115,"threshold_uncertainty_score":0.6404871},"labels":[],"label_agreement":null},{"id":"W1951521253","doi":"10.1177/0272989x15599546","title":"Health Condition Impacts in a Nationally Representative Cross-Sectional Survey Vary Substantially by Preference-Based Health Index","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Center for Advancing Translational Sciences; National Institute on Aging","keywords":"Health Utilities Index; Quality-adjusted life year; EQ-5D; Index (typography); Quality of life (healthcare); Medicine; Gerontology; Visual analogue scale; Cross-sectional study; Statistics; Demography; Scale (ratio); SF-36; Environmental health; Cost effectiveness; Disease; Health related quality of life; Mathematics; Geography; Physical therapy; Computer science","score_opus":0.5486336747817878,"score_gpt":0.5338205859529802,"score_spread":0.014813088828807608,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1951521253","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.66565585,0.002152619,0.30500844,0.021938004,0.0015037531,0.0011685898,0.0016655014,0.00008919631,0.00081805646],"genre_scores_gemma":[0.97338045,0.00007170905,0.0025741076,0.022780666,0.00019816573,0.000064010936,0.0008324669,0.0000325543,0.00006588708],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9899276,0.0018005272,0.005678781,0.00091628305,0.0010175287,0.00065930607],"domain_scores_gemma":[0.99038106,0.0052388925,0.0028350344,0.00041057714,0.00029039383,0.00084403215],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06609686,0.00023846369,0.0010742923,0.00069360697,0.00024943822,0.00018591726,0.0004441331,0.00029082972,0.00095689925],"category_scores_gemma":[0.03157035,0.00028331214,0.0000958923,0.000644378,0.00014510727,0.0005829298,0.00008580815,0.0005193945,0.00045799091],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019546926,0.00020517672,0.9396218,0.00009096919,0.00001952445,0.0000039133297,0.0005928736,0.0013030906,1.3588111e-7,0.003940375,0.052966576,0.0010600679],"study_design_scores_gemma":[0.0025680978,0.0001936343,0.9441086,0.00036720635,3.4984433e-7,0.000007977713,0.000105102445,0.029811911,4.2485033e-7,0.020552607,0.0020480244,0.00023607118],"about_ca_topic_score_codex":0.00438605,"about_ca_topic_score_gemma":0.0057662576,"teacher_disagreement_score":0.3077246,"about_ca_system_score_codex":0.0024844662,"about_ca_system_score_gemma":0.004158485,"threshold_uncertainty_score":0.9999619},"labels":[],"label_agreement":null},{"id":"W1965692977","doi":"10.1177/0272989x13511706","title":"Blocks, Ovals, or People? Icon Type Affects Risk Perceptions and Recall of Pictographs","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Safety Warnings and Signage","field":"Psychology","cited_by":147,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"National Cancer Institute","keywords":"Icon; Recall; Health literacy; Perception; Psychology; Medicine; Computer science; Cognitive psychology; Health care","score_opus":0.018948772556517036,"score_gpt":0.33547349783710356,"score_spread":0.3165247252805865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1965692977","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98535305,0.00055868767,0.0084374165,0.00023156658,0.00071247836,0.00024876383,0.0000041313388,0.000059301008,0.0043945946],"genre_scores_gemma":[0.9959914,0.00033745015,0.0026706706,0.00028915584,0.000119315126,0.000019328529,0.0000028048858,0.000022006783,0.0005478851],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99821794,0.00021299132,0.00037352392,0.00035872008,0.00054355385,0.00029329793],"domain_scores_gemma":[0.9972963,0.0019082146,0.000153239,0.00033914272,0.00008938695,0.00021374229],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007665442,0.00015032243,0.00032879235,0.0001903662,0.00012848497,0.000029554745,0.0002335087,0.000273987,0.051528662],"category_scores_gemma":[0.0021871827,0.00010428874,0.00008665126,0.00044977566,0.00013009841,0.00005816099,0.00013413801,0.0003698997,0.00033945683],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023933548,0.00015905041,0.12503253,0.000017165637,0.00005266935,0.000040717623,0.0025032787,0.000002276157,0.0004148273,0.00080333656,0.025570313,0.8451645],"study_design_scores_gemma":[0.0010297925,0.00045325715,0.98707503,0.0003343684,0.00004793145,0.000082960454,0.0010299342,0.0010510893,0.0000045544075,0.0026765142,0.006024939,0.00018965272],"about_ca_topic_score_codex":0.00016350069,"about_ca_topic_score_gemma":0.00012540109,"teacher_disagreement_score":0.8620425,"about_ca_system_score_codex":0.000010362082,"about_ca_system_score_gemma":0.00003069201,"threshold_uncertainty_score":0.9493384},"labels":[],"label_agreement":null},{"id":"W1966720862","doi":"10.1177/0272989x03256004","title":"Fast and Frugal Models of Clinical Judgment in Novice and Expert Physicians","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Infection and Immunity","funders":"","keywords":"Vignette; Heuristic; Matching (statistics); Psychology; Value (mathematics); Clinical judgment; Logistic regression; Medicine; Applied psychology; Social psychology; Computer science; Statistics; Artificial intelligence; Mathematics; Medical physics; Pathology","score_opus":0.18905995399552714,"score_gpt":0.481086002958928,"score_spread":0.2920260489634009,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966720862","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94510704,0.00041566236,0.050832056,0.00028468636,0.0006392149,0.00012805867,0.0000055083133,0.000013767679,0.0025740159],"genre_scores_gemma":[0.98030335,0.00020936671,0.018402893,0.00095918996,0.00006996161,0.000005176241,3.131417e-7,0.0000148935405,0.00003488674],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99430287,0.00035555052,0.0021502385,0.00084429776,0.0020271605,0.00031990476],"domain_scores_gemma":[0.9907487,0.007851345,0.0004171309,0.000567534,0.00014984244,0.00026543395],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008880311,0.0002001188,0.0008394054,0.00033639683,0.00009329917,0.00014593874,0.00055283547,0.0002791235,0.0003800885],"category_scores_gemma":[0.013281549,0.00014716694,0.00013692041,0.0004979268,0.0003270941,0.00032807112,0.00035101775,0.0003987208,0.000031840114],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006894096,0.0001725252,0.011287985,0.0000017335692,0.0000035788566,0.000042735435,0.00035399044,0.00021727181,0.000013785259,0.0011180919,0.0007040941,0.98601526],"study_design_scores_gemma":[0.003918147,0.00046165014,0.043095853,0.0014355448,0.00002313929,0.00011842104,0.0035218818,0.1933836,0.00007855868,0.7379009,0.015361076,0.0007012368],"about_ca_topic_score_codex":0.000022899982,"about_ca_topic_score_gemma":0.00011637359,"teacher_disagreement_score":0.985314,"about_ca_system_score_codex":0.000027089502,"about_ca_system_score_gemma":0.000120020944,"threshold_uncertainty_score":0.99503},"labels":[],"label_agreement":null},{"id":"W1967017081","doi":"10.1177/0272989x10369001","title":"Developing a Tuberculosis Transmission Model That Accounts for Changes in Population Health","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Life expectancy; Medicine; Tuberculosis; Epidemiology; Population; Demography; Proxy (statistics); Psychological intervention; Population health; Environmental health; Transmission (telecommunications); Public health; Mortality rate; Incidence (geometry); Gerontology; Statistics; Surgery","score_opus":0.09373358339035236,"score_gpt":0.4427716200951453,"score_spread":0.34903803670479294,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967017081","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6342912,0.00023673478,0.30029607,0.06408818,0.00021755116,0.00072257343,0.0000053205276,0.000054816202,0.00008760437],"genre_scores_gemma":[0.9012682,0.00049130345,0.08699183,0.010893642,0.00017702048,0.00008760191,0.00004457544,0.000026202353,0.000019618268],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99744886,0.00009135211,0.0004953925,0.0004267414,0.00091841567,0.0006192657],"domain_scores_gemma":[0.9981478,0.0010333719,0.00008652518,0.00024047116,0.00007508482,0.00041675122],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031150605,0.0001615547,0.0005236282,0.00038354917,0.00013915301,0.000017891785,0.00017730636,0.00036024014,0.0002287059],"category_scores_gemma":[0.0040948703,0.00011975448,0.00009985785,0.00026655532,0.000043450807,0.000089807,0.00006063736,0.0006381166,0.00000986454],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038804655,0.00006848625,0.041383736,0.00014296304,0.000011479792,0.000013447715,0.00014209957,0.00004702615,0.00048738348,0.0009095476,0.0018520568,0.9545537],"study_design_scores_gemma":[0.0027779902,0.0002757024,0.32383883,0.0024383855,0.000015626143,0.000102289676,0.000054677443,0.6383019,0.00016522016,0.019140895,0.012627193,0.00026129762],"about_ca_topic_score_codex":0.00013407036,"about_ca_topic_score_gemma":0.0019401399,"teacher_disagreement_score":0.9542924,"about_ca_system_score_codex":0.00017295977,"about_ca_system_score_gemma":0.0003238653,"threshold_uncertainty_score":0.49022382},"labels":[],"label_agreement":null},{"id":"W1967979861","doi":"10.1177/0272989x0202200309","title":"Reduction Mammaplasty: Defining Medical Necessity","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Breast Implant and Reconstruction","field":"Medicine","cited_by":76,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Mammaplasty; Breast reduction; Medicine; Breast surgery; Physical therapy; Breast cancer; Set (abstract data type); Quality of life (healthcare); MEDLINE; Surgery; Cancer; Computer science; Nursing","score_opus":0.02587630598120863,"score_gpt":0.30614336422014865,"score_spread":0.28026705823894005,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1967979861","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.95237315,0.0005834047,0.02606943,0.004009792,0.0036150562,0.0001768155,0.0000031796767,0.00028189266,0.012887308],"genre_scores_gemma":[0.9920052,0.00013567525,0.005949676,0.00046283842,0.0010447544,0.000006088792,0.000009878667,0.000022871967,0.00036298833],"study_design_codex":"design_other","study_design_gemma":"case_report","domain_scores_codex":[0.9964229,0.00007437561,0.000543267,0.00038393534,0.0022076077,0.00036794314],"domain_scores_gemma":[0.9986104,0.00035907523,0.00011229363,0.00024686963,0.00007543636,0.0005959754],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00086849637,0.00017383324,0.00036559845,0.00022034922,0.0002028435,0.000032359585,0.00016640042,0.00050093763,0.020928958],"category_scores_gemma":[0.0014031765,0.0001309275,0.00013337596,0.0003517574,0.00018625308,0.00012146766,0.00010953634,0.00076658366,0.0008750996],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012338764,0.00008719936,0.003439775,0.000023617062,0.000021727767,0.00042520647,0.00006268531,9.727192e-7,0.00002079024,0.0003674225,0.009169554,0.9862577],"study_design_scores_gemma":[0.0088485675,0.0005656365,0.035104677,0.0096034035,0.00027648752,0.8099195,0.0005071933,0.09602269,0.00030371264,0.004809081,0.033101372,0.00093771046],"about_ca_topic_score_codex":0.000006760216,"about_ca_topic_score_gemma":0.0000066301973,"teacher_disagreement_score":0.98532,"about_ca_system_score_codex":0.000073499716,"about_ca_system_score_gemma":0.0001018885,"threshold_uncertainty_score":0.99990284},"labels":[],"label_agreement":null},{"id":"W1972763552","doi":"10.1177/0272989x06297397","title":"Bedside Rationing by Health Practitioners: A Case Study in a Ugandan Hospital","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Global Maternal and Child Health","field":"Medicine","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Rationing; Medicine; Health care rationing; Health care; Actuarial science; Business; Medical emergency; Economics; Economic growth","score_opus":0.01761382841030457,"score_gpt":0.39944017156823425,"score_spread":0.3818263431579297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1972763552","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9758781,0.00029602498,0.017739866,0.0046823616,0.0002950254,0.0005802548,0.0000024057426,0.000048690807,0.00047726693],"genre_scores_gemma":[0.9918113,0.000029567143,0.002136403,0.0057560103,0.00020625246,0.0000074818176,0.0000060019215,0.0000135053215,0.000033531705],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9973976,0.00008298031,0.0007032584,0.00030734087,0.0011002371,0.0004085854],"domain_scores_gemma":[0.99879,0.00035734128,0.00014633172,0.0001777,0.00006278077,0.0004658109],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002544869,0.00012766536,0.00032256698,0.00016575356,0.00016914791,0.000029967274,0.00007781561,0.00009482239,0.00024695447],"category_scores_gemma":[0.001180457,0.00010038953,0.00004433453,0.00034221308,0.000033094395,0.00010132116,0.000065129476,0.00046789137,0.000032291042],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005048077,0.002375039,0.44310474,0.0000770765,0.000028837785,0.05441595,0.004478214,0.0000038806115,0.000012203734,0.00016333445,0.011175961,0.48365995],"study_design_scores_gemma":[0.023830222,0.010846175,0.8080843,0.008611277,0.000087974426,0.044309217,0.06226201,0.002006816,0.000039247225,0.0023265202,0.03668053,0.00091568416],"about_ca_topic_score_codex":0.0011654604,"about_ca_topic_score_gemma":0.0017169286,"teacher_disagreement_score":0.48274428,"about_ca_system_score_codex":0.00025078238,"about_ca_system_score_gemma":0.00024459205,"threshold_uncertainty_score":0.40937662},"labels":[],"label_agreement":null},{"id":"W1978408850","doi":"10.1177/0272989x13501721","title":"Toward Minimum Standards for Certifying Patient Decision Aids","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":532,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Marie Curie; Ottawa Hospital Research Institute","keywords":"Certification; Decision aids; Checklist; Delphi method; Quality (philosophy); Delphi; Process (computing); Psychology; Computer science; Inclusion (mineral); Medical education; Medicine; Social psychology; Artificial intelligence; Alternative medicine; Political science","score_opus":0.2502511443071819,"score_gpt":0.4862352165077257,"score_spread":0.2359840722005438,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1978408850","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.47086012,0.0039414456,0.49627808,0.008675715,0.008120382,0.006845627,0.00016544253,0.00052655005,0.0045866636],"genre_scores_gemma":[0.88734424,0.00011622343,0.10456875,0.0059371716,0.0003208077,0.0015452898,0.000026992531,0.00007179774,0.00006871653],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9922831,0.00077556755,0.002141308,0.00066085486,0.0030926135,0.001046522],"domain_scores_gemma":[0.98103315,0.014528737,0.0005829848,0.0013288163,0.0017980753,0.00072825135],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017302383,0.00031313163,0.000583757,0.00032089034,0.0015089681,0.000054438326,0.0010621734,0.000652504,0.007295577],"category_scores_gemma":[0.029230272,0.00024958528,0.0002054072,0.00046723257,0.00011836074,0.00032918958,0.0010621453,0.0011882199,0.0007117199],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029141695,0.0000682235,0.0055675856,0.00025080243,0.000013958419,0.0000054555367,0.0049228873,0.000011506361,0.00002308648,0.00077988603,0.06433867,0.9237265],"study_design_scores_gemma":[0.0067467378,0.0014402305,0.009464709,0.016769279,0.00005097211,0.000016915932,0.025649276,0.04099018,0.000052142597,0.10740405,0.7903503,0.0010651541],"about_ca_topic_score_codex":0.00013453221,"about_ca_topic_score_gemma":0.00007051281,"teacher_disagreement_score":0.92266136,"about_ca_system_score_codex":0.00080844265,"about_ca_system_score_gemma":0.0012775501,"threshold_uncertainty_score":0.99999565},"labels":[],"label_agreement":null},{"id":"W1979458535","doi":"10.1177/0272989x03256008","title":"Preference-Based Measurement of Health-Related Quality of Life (HRQL) in Children with Chronic Musculoskeletal Disorders (MSKDs)","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":70,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Health Utilities Index; Intraclass correlation; Medicine; Quality of life (healthcare); Concordance; Rating scale; Health assessment; Physical therapy; State of health; Categorical variable; Scale (ratio); Gerontology; Psychometrics; Psychology; Clinical psychology; Health related quality of life; Statistics; Developmental psychology; Disease","score_opus":0.30305140833922795,"score_gpt":0.4174880318865219,"score_spread":0.11443662354729395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1979458535","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9199682,0.0053922553,0.069541514,0.0031873907,0.00021421086,0.0009088663,0.00003691367,0.0000233953,0.00072724326],"genre_scores_gemma":[0.9960725,0.00011255987,0.002992768,0.00070628227,0.00002663574,0.000047265996,0.000010941934,0.000025786967,0.0000053037184],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9891398,0.0013752196,0.0075849984,0.00059971237,0.000834587,0.0004656732],"domain_scores_gemma":[0.99409837,0.0013709341,0.003598469,0.00057596347,0.000100321064,0.00025596053],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04839954,0.00021586519,0.0015914467,0.00047249312,0.000089802044,0.000014228483,0.00037073248,0.0002247451,0.001234751],"category_scores_gemma":[0.022794923,0.00021889439,0.00016951602,0.00051651936,0.00017825494,0.00014685861,0.000040374354,0.00032776492,0.00006837051],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007923866,0.0005015735,0.93865114,0.00069896155,0.00012252144,2.624955e-7,0.00033001174,0.0026968063,0.0000013704198,0.029920923,0.0004468933,0.026550302],"study_design_scores_gemma":[0.004848993,0.0005451948,0.973547,0.0015612569,0.000006065181,0.0000015434913,0.00016542853,0.004781429,0.0000034911334,0.013475212,0.0007062214,0.0003581781],"about_ca_topic_score_codex":0.0016853871,"about_ca_topic_score_gemma":0.0024799325,"teacher_disagreement_score":0.076104246,"about_ca_system_score_codex":0.00076022005,"about_ca_system_score_gemma":0.0016723414,"threshold_uncertainty_score":0.99967825},"labels":[],"label_agreement":null},{"id":"W1981652493","doi":"10.1177/0272989x14539731","title":"Opportunity Cost of Funding Drugs for Rare Diseases","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Complement system in diseases","field":"Immunology and Microbiology","cited_by":66,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kingston General Hospital; Health Sciences Centre; Queen's University; University of Ottawa; Sunnybrook Health Science Centre","funders":"","keywords":"Eculizumab; Medicine; Cost effectiveness; Context (archaeology); Opportunity cost; Cost–benefit analysis; Health economics; Quality-adjusted life year; Actuarial science; Health care; Health care rationing; Intensive care medicine; Business; Economics; Public health; Risk analysis (engineering); Political science; Immunology","score_opus":0.04801332625254817,"score_gpt":0.35297551935410393,"score_spread":0.30496219310155576,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1981652493","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5491906,0.0026531534,0.4378992,0.0006669939,0.004102188,0.0015302711,0.0005350896,0.00023178666,0.003190743],"genre_scores_gemma":[0.99884087,0.000008856224,0.00038584275,0.00034692558,0.000085798594,0.000057538105,0.0001092347,0.000014144743,0.00015080068],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99852675,0.0002088486,0.0005187758,0.00025748627,0.0001883753,0.00029974885],"domain_scores_gemma":[0.99556154,0.0036980251,0.00019268486,0.00033415234,0.000098851815,0.00011476995],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0008490819,0.000127556,0.00034930307,0.00011757925,0.00017421185,0.000011410059,0.00040176266,0.00012985288,0.006139263],"category_scores_gemma":[0.008191719,0.00010402452,0.00015905552,0.00009683916,0.00014347586,0.000049337857,0.00021746942,0.0001074625,0.00007601033],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045593243,0.0002453473,0.018130865,0.00018825746,0.0000933459,0.000005215785,0.00011981809,0.000008790726,0.0011493699,0.00768472,0.06873884,0.9031795],"study_design_scores_gemma":[0.010450074,0.00075079396,0.066921145,0.005759241,0.0003888502,0.00006650009,0.0014651774,0.0027692134,0.0023867204,0.030895256,0.8772078,0.00093921734],"about_ca_topic_score_codex":0.0000050760023,"about_ca_topic_score_gemma":0.0000056316712,"teacher_disagreement_score":0.9022403,"about_ca_system_score_codex":0.000030674706,"about_ca_system_score_gemma":0.000102122576,"threshold_uncertainty_score":0.9947693},"labels":[],"label_agreement":null},{"id":"W1983489842","doi":"10.1177/0272989x03261568","title":"Estimating the Prognosis of Hepatitis C Patients Infected by Transfusion in Canada between 1986 and 1990","year":2004,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Hepatitis C virus research","field":"Medicine","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Toronto General Hospital","funders":"","keywords":"Medicine; Cirrhosis; Cohort; Natural history; Blood transfusion; Hepatitis C; Hepatitis C virus; Confidence interval; Cohort study; Liver disease; Markov model; Internal medicine; Intensive care medicine; Pediatrics; Markov chain; Virology; Virus; Statistics","score_opus":0.017048214030822448,"score_gpt":0.3134578114904411,"score_spread":0.29640959745961865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1983489842","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99305266,0.0000986494,0.005231944,0.0008350953,0.00007937326,0.0005037752,0.00002614857,0.0000118210855,0.00016053255],"genre_scores_gemma":[0.99678874,0.000105829524,0.0027408923,0.00026355888,0.000039126364,0.000022154527,0.000021103602,0.000016916507,0.0000016591528],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99687433,0.00010329825,0.00058640493,0.00024501892,0.0018917137,0.00029922323],"domain_scores_gemma":[0.9984755,0.0009382144,0.00007951249,0.00017645342,0.00010768765,0.00022262428],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00062733324,0.000120958575,0.00034422922,0.00011804275,0.00008553494,0.00001299883,0.00015704833,0.000105385436,0.0001749124],"category_scores_gemma":[0.0042265537,0.00008154775,0.000032895885,0.0004885178,0.00010554447,0.00005535243,0.00010504325,0.00044309086,0.000002601509],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000018002742,0.000036353023,0.5903518,0.00003210962,0.000006834373,0.000029024957,0.00009670144,0.000010089301,0.000025315197,0.0000026612609,0.00015608814,0.409235],"study_design_scores_gemma":[0.002316388,0.00021178671,0.9892792,0.0021839237,0.000018274683,0.0000042525826,0.000017388367,0.004965521,0.00029833408,0.0002346376,0.00037865044,0.00009161876],"about_ca_topic_score_codex":0.634612,"about_ca_topic_score_gemma":0.64197946,"teacher_disagreement_score":0.4091434,"about_ca_system_score_codex":0.00034371577,"about_ca_system_score_gemma":0.000906568,"threshold_uncertainty_score":0.5059885},"labels":[],"label_agreement":null},{"id":"W1984421215","doi":"10.1177/0272989x07312477","title":"The Option Value of Delay in Health Technology Assessment","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":76,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; University of Toronto","funders":"","keywords":"Reimbursement; Value of information; Stochastic game; Option value; Actuarial science; Value (mathematics); Health technology; Cost–benefit analysis; Opportunity cost; Economics; Microeconomics; Computer science; Incentive; Health care","score_opus":0.29609196175929686,"score_gpt":0.5059744187735876,"score_spread":0.20988245701429076,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1984421215","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.58843017,0.004564127,0.30205473,0.10196599,0.00096328335,0.0005870934,0.000016401109,0.00004659968,0.0013715958],"genre_scores_gemma":[0.97501755,0.0010850034,0.019558333,0.0041448916,0.00008994753,0.00004209923,0.0000024707283,0.0000140262355,0.000045659453],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99400955,0.00033270897,0.004615488,0.00035846312,0.00031728958,0.00036652092],"domain_scores_gemma":[0.9948586,0.0028873028,0.001644011,0.0004437371,0.000043715285,0.00012261627],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.027953679,0.00010615105,0.00067772454,0.00045172253,0.00031473354,0.000015214915,0.00043066218,0.00019845676,0.00020874661],"category_scores_gemma":[0.011190285,0.00009844227,0.000072095725,0.0004459508,0.00017587794,0.000117982985,0.00010876529,0.00033833197,0.00019896179],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000154653,0.0001053064,0.06418537,0.00007343112,0.000015246001,0.000009279253,0.0004082757,0.0007180355,5.363864e-7,0.8946757,0.005857112,0.033936273],"study_design_scores_gemma":[0.0019261361,0.0003171422,0.2670149,0.0013198333,0.0000018468528,0.00011959475,0.0013815939,0.27756304,0.0000022927572,0.31765065,0.13234246,0.00036051613],"about_ca_topic_score_codex":0.00017586384,"about_ca_topic_score_gemma":0.00015964179,"teacher_disagreement_score":0.577025,"about_ca_system_score_codex":0.00053586287,"about_ca_system_score_gemma":0.00044889306,"threshold_uncertainty_score":0.99713886},"labels":[],"label_agreement":null},{"id":"W1990237742","doi":"10.1177/0272989x09334419","title":"Incorporating Herd Immunity Effects into Cohort Models of Vaccine Cost-Effectiveness","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; GlaxoSmithKline (Canada); University of Guelph","funders":"GlaxoSmithKline","keywords":"Herd immunity; Cohort; Medicine; Immunity; Vaccination; Immunology; Immune system; Internal medicine","score_opus":0.2271922060979795,"score_gpt":0.4653424012989373,"score_spread":0.2381501952009578,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990237742","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.54775405,0.001393328,0.44531226,0.001995492,0.0005580038,0.0010493638,0.0000117798545,0.0000557109,0.0018700011],"genre_scores_gemma":[0.98177385,0.00006359988,0.013879019,0.003971446,0.00018821338,0.00007369344,0.000013484585,0.000024833465,0.000011857965],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9938571,0.0008106077,0.003978692,0.0005652359,0.0004178822,0.0003704742],"domain_scores_gemma":[0.98966324,0.007339691,0.0019104906,0.00067302625,0.00013553088,0.00027803634],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.035248477,0.0002331079,0.001457267,0.00039789116,0.00023177404,0.000049618695,0.00055661815,0.00030553035,0.00027534552],"category_scores_gemma":[0.022950634,0.00025901102,0.0001639864,0.00043095165,0.000060134484,0.00047871354,0.00014556936,0.0004174474,0.00024111486],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00050467585,0.0010127305,0.22560297,0.0019328237,0.000262469,0.000055134715,0.0022550777,0.010389796,0.00010435504,0.3822484,0.0062216325,0.36940995],"study_design_scores_gemma":[0.0019914368,0.00031367323,0.23530045,0.0021433595,0.0000128981,0.0000138249925,0.00011869547,0.169495,0.000052800082,0.58937174,0.0007769753,0.0004091101],"about_ca_topic_score_codex":0.00024730418,"about_ca_topic_score_gemma":0.000043430406,"teacher_disagreement_score":0.43401977,"about_ca_system_score_codex":0.00047367497,"about_ca_system_score_gemma":0.0001765944,"threshold_uncertainty_score":0.99998623},"labels":[],"label_agreement":null},{"id":"W1990801963","doi":"10.1177/0272989x0002000104","title":"Does Clinical Error Contribute to Unnecessary Antibiotic Use?","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Medical Malpractice and Liability Issues","field":"Health Professions","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; Mount Sinai Hospital","funders":"","keywords":"Sore throat; Medicine; Antibiotics; Pharyngitis; Throat; Intensive care medicine; Throat culture; Streptococcus; Internal medicine; Pediatrics; Surgery; Microbiology","score_opus":0.12319627528592199,"score_gpt":0.5354014388806843,"score_spread":0.4122051635947623,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1990801963","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9209297,0.000081333055,0.010354926,0.057364773,0.004620568,0.0010153096,0.000026558606,0.00026625462,0.0053405925],"genre_scores_gemma":[0.94419926,0.00026573273,0.005273063,0.043699697,0.002034601,0.00002382456,0.000011425578,0.000037411955,0.004454959],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9921322,0.0019675128,0.0020408619,0.0007870052,0.0020800186,0.0009924343],"domain_scores_gemma":[0.9673251,0.029864106,0.00023331477,0.0008786635,0.00023450749,0.0014643166],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.008749682,0.00026090146,0.0008212658,0.00012531495,0.0006371759,0.000048053924,0.00070309086,0.00083226635,0.08860797],"category_scores_gemma":[0.082093745,0.00014229098,0.00019208984,0.0005002352,0.00022068626,0.00031755204,0.0003470073,0.0021096843,0.010968975],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008521658,0.00037157172,0.17078353,0.00011342658,0.00003919343,0.00034302953,0.0005420946,0.0000075730554,0.000005094707,0.0006533761,0.13708612,0.68920285],"study_design_scores_gemma":[0.0013142575,0.00021692428,0.1401164,0.0030011216,0.000044146964,0.000005137872,0.00048780124,0.0011713989,0.0000022472618,0.0022485151,0.85113084,0.0002612132],"about_ca_topic_score_codex":0.00009284102,"about_ca_topic_score_gemma":0.00021653966,"teacher_disagreement_score":0.7140447,"about_ca_system_score_codex":0.00008756547,"about_ca_system_score_gemma":0.00043313397,"threshold_uncertainty_score":0.9898011},"labels":[],"label_agreement":null},{"id":"W1992099844","doi":"10.1177/0272989x15578127","title":"Mapping between 6 Multiattribute Utility Instruments","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Comparability; EQ-5D; Actuarial science; Goodness of fit; Econometrics; Ordinary least squares; Quality-adjusted life year; Estimator; Quality (philosophy); Index (typography); Statistics; Operations research; Medicine; Computer science; Economics; Mathematics; Cost effectiveness; Disease; Health related quality of life","score_opus":0.5828080783151686,"score_gpt":0.49555686313634423,"score_spread":0.08725121517882434,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1992099844","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.58026296,0.00068527664,0.39609626,0.015022079,0.001744828,0.00045173962,0.00014723756,0.00012082078,0.0054687886],"genre_scores_gemma":[0.9733174,0.000019041823,0.016671222,0.009116091,0.00069597183,0.00002668323,0.000028639599,0.000027207234,0.00009773098],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.99346846,0.00034442582,0.004395845,0.0006780587,0.00056812115,0.0005451127],"domain_scores_gemma":[0.9950874,0.0021377157,0.0013074126,0.0006183273,0.00009838239,0.0007507867],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.03590371,0.0002041907,0.0010219495,0.0003675984,0.00019079298,0.00009817986,0.0006057268,0.00031846392,0.0015606489],"category_scores_gemma":[0.041730024,0.00023580834,0.00013359971,0.0003619257,0.00008863568,0.00042194783,0.0002904962,0.0003793245,0.005446188],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000021277225,0.00008582731,0.81523514,0.000085808184,0.000061406856,0.000008698322,0.0012674116,0.000015125959,1.4037578e-7,0.0058818306,0.05384448,0.12349286],"study_design_scores_gemma":[0.0027502424,0.00008445212,0.33946627,0.0007155571,0.0000059107065,0.000013240388,0.00083212066,0.04252011,0.0000019481868,0.12088237,0.49214733,0.0005804432],"about_ca_topic_score_codex":0.0001811279,"about_ca_topic_score_gemma":0.000032397023,"teacher_disagreement_score":0.47576886,"about_ca_system_score_codex":0.0005390045,"about_ca_system_score_gemma":0.00026413047,"threshold_uncertainty_score":0.99935204},"labels":[],"label_agreement":null},{"id":"W1995702931","doi":"10.1177/0272989x13475718","title":"Scoring the 5-Level EQ-5D","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Computer science","score_opus":0.6224432138530742,"score_gpt":0.46638050969000533,"score_spread":0.1560627041630689,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1995702931","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4159735,0.0029896908,0.42020112,0.13096982,0.0040203934,0.0012916673,0.000046933586,0.00016442429,0.024342414],"genre_scores_gemma":[0.94977623,0.00008005886,0.011126537,0.0372114,0.00090275245,0.00013130075,0.0000036217307,0.000035126002,0.0007329553],"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9950866,0.00022419247,0.003297149,0.0005072576,0.00039785833,0.00048690333],"domain_scores_gemma":[0.9929142,0.005017801,0.0009940836,0.00071138825,0.0000705389,0.00029196785],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.019453632,0.00016880232,0.00062304124,0.00020616042,0.00036512388,0.00020370985,0.0007589017,0.00019593045,0.012284362],"category_scores_gemma":[0.02608452,0.0001436912,0.00013544217,0.00025289998,0.0001098511,0.00041211367,0.000221493,0.00037788472,0.020131795],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000016395277,0.00013934643,0.061234348,0.00016245266,0.00008792395,0.000011618765,0.0019900037,0.00021519394,0.0000037473937,0.27813184,0.4230628,0.23494431],"study_design_scores_gemma":[0.0012285803,0.00007762924,0.26984897,0.0010334422,0.0000072319294,0.00004585081,0.0012220613,0.085603386,0.0000037476154,0.37727162,0.26298428,0.0006731836],"about_ca_topic_score_codex":0.00034513747,"about_ca_topic_score_gemma":0.00006520897,"teacher_disagreement_score":0.53380275,"about_ca_system_score_codex":0.00024319478,"about_ca_system_score_gemma":0.000117926094,"threshold_uncertainty_score":0.98861855},"labels":[],"label_agreement":null},{"id":"W1995914824","doi":"10.1177/0272989x15575285","title":"Use of Bayesian Markov Chain Monte Carlo Methods to Estimate EQ-5D Utility Scores from EORTC QLQ Data in Myeloma for Use in Cost-Effectiveness Analysis","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Bayesian probability; Statistics; Econometrics; Markov chain Monte Carlo; Computer science; Frequentist inference; Context (archaeology); Missing data; Data mining; Bayesian inference; Mathematics","score_opus":0.7010650185572499,"score_gpt":0.5604106812754549,"score_spread":0.140654337281795,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1995914824","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4559002,0.000297115,0.5408376,0.00078542193,0.00030644718,0.0010031478,0.00083825336,0.0000143281295,0.000017485],"genre_scores_gemma":[0.6798819,0.000013463481,0.3185661,0.0011609825,0.000069176975,0.0001721171,0.000099989804,0.000027275035,0.000009001901],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99127966,0.00185316,0.0047281277,0.0012292434,0.00039194745,0.00051786436],"domain_scores_gemma":[0.97441983,0.022030368,0.0012501848,0.0016599372,0.00014070414,0.0004989655],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.072011866,0.00025598088,0.00205624,0.0013244104,0.000064247295,0.000127688,0.0008870121,0.00030917555,0.00017862303],"category_scores_gemma":[0.13242488,0.0002905011,0.00016736033,0.0012095845,0.000086893364,0.00081161445,0.00054379844,0.00027296753,0.00002796578],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00046727853,0.00016730717,0.887346,0.00012278184,0.00015785055,0.000012996434,0.0007992673,0.0076209,0.0000013063448,0.00087828044,0.0011992744,0.10122671],"study_design_scores_gemma":[0.0007667708,0.000033019292,0.44169447,0.00036634927,0.000021523381,5.4027896e-7,0.00009398015,0.55044425,0.0000021969076,0.004173284,0.0022331774,0.00017046226],"about_ca_topic_score_codex":0.010361126,"about_ca_topic_score_gemma":0.021675115,"teacher_disagreement_score":0.5428233,"about_ca_system_score_codex":0.00062429544,"about_ca_system_score_gemma":0.0002844409,"threshold_uncertainty_score":0.9999547},"labels":[],"label_agreement":null},{"id":"W1996073988","doi":"10.1177/0272989x10375853","title":"Presenting the Results of Cochrane Systematic Reviews to a Consumer Audience: A Qualitative Study","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":129,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Cochrane; University of Ottawa; McMaster University","funders":"","keywords":"Systematic review; Medicine; Psychology; MEDLINE; Political science","score_opus":0.6511906448066801,"score_gpt":0.6268698211041657,"score_spread":0.024320823702514427,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1996073988","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.836264,0.00293992,0.13604465,0.0024990023,0.0023216677,0.01288521,0.00002283897,0.000018254525,0.007004441],"genre_scores_gemma":[0.9873738,0.00002456896,0.010232929,0.00045732234,0.00010692804,0.00029157542,6.4081263e-7,0.000012863349,0.0014993713],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.85403174,0.066157244,0.039850425,0.002416792,0.036783174,0.0007606428],"domain_scores_gemma":[0.57361495,0.37158746,0.027659155,0.019322103,0.006606054,0.001210315],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.5276366,0.0003717998,0.006061259,0.0004798083,0.0003225367,0.0005762842,0.004770756,0.00013545755,0.005736897],"category_scores_gemma":[0.82945865,0.00012854826,0.0014230536,0.003147695,0.00022806716,0.00015681896,0.000764784,0.0006068822,0.0033135274],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037174905,0.0013678648,0.0049892073,0.007376044,0.0010520831,0.0001785366,0.39649826,0.00006186187,0.000491049,0.0053461054,0.33378473,0.2484825],"study_design_scores_gemma":[0.008089197,0.0015717883,0.019044587,0.090353884,0.0042952695,0.0004572576,0.5138942,0.0733045,0.00008234144,0.06399199,0.22202164,0.0028933524],"about_ca_topic_score_codex":0.000017547185,"about_ca_topic_score_gemma":0.00030070113,"teacher_disagreement_score":0.3054302,"about_ca_system_score_codex":0.000016994605,"about_ca_system_score_gemma":0.00011085183,"threshold_uncertainty_score":0.9974625},"labels":[],"label_agreement":null},{"id":"W1997843288","doi":"10.1177/0272989x14556676","title":"Exploring the Potential Cost-Effectiveness of Patient Decision Aids for Use in Adults with Obstructive Sleep Apnea","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Obstructive Sleep Apnea Research","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"St. Paul's Hospital; Centre for Advancing Health Outcomes; Ottawa Hospital; University of British Columbia; Vancouver Coastal Health Research Institute; Vancouver Coastal Health","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Cost effectiveness; Cohort; Obstructive sleep apnea; Randomized controlled trial; Incremental cost-effectiveness ratio; Decision aids; Physical therapy; Quality-adjusted life year; Alternative medicine; Surgery; Internal medicine","score_opus":0.04541206030204751,"score_gpt":0.3295506152217962,"score_spread":0.28413855491974865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1997843288","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6322469,0.00002313494,0.36581597,0.000028593811,0.0003721709,0.001449227,0.0000063706225,0.00001771378,0.000039893705],"genre_scores_gemma":[0.9707792,0.000009056603,0.028487151,0.0000637416,0.00013217588,0.00046710376,0.00000945233,0.000051026676,0.0000010903877],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9952026,0.00045286553,0.00068297365,0.000633348,0.0024934122,0.0005347852],"domain_scores_gemma":[0.98865503,0.009701527,0.00017731186,0.0006229148,0.0005753578,0.000267834],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0023339756,0.00026236466,0.0006222945,0.00041996813,0.0001426202,0.00003898885,0.00035850576,0.00015149858,0.0000725542],"category_scores_gemma":[0.012090951,0.00015525158,0.00017148544,0.0007302244,0.00033729282,0.0003169208,0.00030062677,0.0005847673,0.000007806809],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.016115163,0.00019871644,0.019951925,0.00015520256,0.00005805624,0.000080905134,0.00030635082,0.00065008365,0.0003149239,0.00049943826,0.000003481457,0.96166575],"study_design_scores_gemma":[0.021236552,0.0049319533,0.7647244,0.009921599,0.0001451118,0.00023913525,0.001852985,0.18580505,0.00525841,0.0047134142,0.0006807391,0.0004906644],"about_ca_topic_score_codex":0.00003760211,"about_ca_topic_score_gemma":0.000049489056,"teacher_disagreement_score":0.9611751,"about_ca_system_score_codex":0.00021849139,"about_ca_system_score_gemma":0.00005976343,"threshold_uncertainty_score":0.9962306},"labels":[],"label_agreement":null},{"id":"W1998078556","doi":"10.1177/0272989x09353947","title":"Informed Choice Assistance for Women Making Uterine Fibroid Treatment Decisions: A Practical Clinical Trial","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Uterine Myomas and Treatments","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Uterine fibroids; Medicine; Clinical trial; Gynecology; Clinical decision making; Obstetrics; Intensive care medicine; Internal medicine","score_opus":0.1436823406416142,"score_gpt":0.514288237355758,"score_spread":0.3706058967141438,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1998078556","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.91413593,0.00023882854,0.07105736,0.0064852335,0.0016769146,0.0028212748,0.000031137108,0.00024582536,0.0033075227],"genre_scores_gemma":[0.8770325,0.0004086822,0.11565792,0.0036053308,0.0017034425,0.00031749622,0.000031272793,0.00004565412,0.0011976653],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9949614,0.00011255457,0.0017290735,0.00080837525,0.0015373279,0.0008513021],"domain_scores_gemma":[0.9888142,0.009066243,0.0003729375,0.0007916456,0.00017004496,0.00078495534],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001302629,0.00043598522,0.0011513184,0.00022375169,0.00023450975,0.00012085664,0.0002516399,0.00042333605,0.0010244851],"category_scores_gemma":[0.029522024,0.00029681067,0.00051605166,0.00035094417,0.00010854459,0.00023505517,0.00013320499,0.00037253,0.0000936249],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.06629749,0.0021151968,0.0016768865,0.000027136934,0.00018941282,0.0008580757,0.00009066454,0.0000011041492,0.000010649112,0.00040728916,0.003689049,0.9246371],"study_design_scores_gemma":[0.16656363,0.025546378,0.06675754,0.002356395,0.00022766418,0.0006561969,0.00008283126,0.0031567153,0.000018412358,0.004780233,0.729346,0.00050801365],"about_ca_topic_score_codex":0.0000023504301,"about_ca_topic_score_gemma":0.00002723804,"teacher_disagreement_score":0.924129,"about_ca_system_score_codex":0.00065337756,"about_ca_system_score_gemma":0.0009445066,"threshold_uncertainty_score":0.9999484},"labels":[],"label_agreement":null},{"id":"W1998948430","doi":"10.1177/0272989x09336075","title":"Capacity Constraints and Cost-Effectiveness: A Discrete Event Simulation for Drug-Eluting Stents","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare Operations and Scheduling Optimization","field":"Health Professions","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Programs for Assessment of Technology in Health Research Institute","funders":"","keywords":"Discrete event simulation; Scarcity; Computer science; Psychological intervention; Resource (disambiguation); Resource allocation; Operations management; Cost–benefit analysis; Cost effectiveness; Operations research; Medicine; Economics; Simulation; Microeconomics; Mathematics","score_opus":0.12689367170748925,"score_gpt":0.5112392589186368,"score_spread":0.38434558721114753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1998948430","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29319835,0.000040631072,0.70328194,0.001046052,0.00037922733,0.0017463284,0.00001473458,0.000049986964,0.00024277689],"genre_scores_gemma":[0.97173864,0.000017544364,0.02646089,0.0013850442,0.00020184877,0.00012990138,0.00002359327,0.000014116048,0.000028410986],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9973359,0.00064658053,0.0007207751,0.00034908982,0.00054501597,0.0004026323],"domain_scores_gemma":[0.9942445,0.004832253,0.00018453655,0.00017115664,0.0003133981,0.0002541607],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003913877,0.00014162529,0.0002827055,0.00010386118,0.0011647274,0.000029461757,0.00010108127,0.00025260227,0.00021319283],"category_scores_gemma":[0.008609942,0.000115282004,0.00005739736,0.00017390934,0.00006665348,0.00014816975,0.000048234553,0.00043179633,0.000013625412],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002779112,0.000056376848,0.007986777,0.00013848409,0.000010909416,0.000004807376,0.0013133615,0.022057747,0.000036249698,0.003934627,0.00014069847,0.96404207],"study_design_scores_gemma":[0.0028496154,0.00011437542,0.043974902,0.0027824207,0.000016631042,0.0000031720704,0.0004400685,0.9429906,0.000004231562,0.005398266,0.0012516349,0.0001740952],"about_ca_topic_score_codex":0.000013835691,"about_ca_topic_score_gemma":0.000033230346,"teacher_disagreement_score":0.96386796,"about_ca_system_score_codex":0.00016272046,"about_ca_system_score_gemma":0.00023953189,"threshold_uncertainty_score":0.99974096},"labels":[],"label_agreement":null},{"id":"W1999005999","doi":"10.1177/0272989x12453504","title":"The Choice of a Noninformative Prior on Between-Study Variance Strongly Affects Predictions of Future Treatment Effect","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; Osteoporosis Canada; University Health Network; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Prior probability; Deviance (statistics); Statistics; Bayesian probability; Mathematics; Population; Deviance information criterion; Sample size determination; Econometrics; Bayesian inference; Medicine","score_opus":0.3606362378436365,"score_gpt":0.5202843149643358,"score_spread":0.15964807712069928,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1999005999","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.928691,0.00081011845,0.0637963,0.00023324018,0.001125716,0.0019820076,0.000025285572,0.0000081831095,0.0033280978],"genre_scores_gemma":[0.99804986,0.000018797786,0.0011027018,0.000027228858,0.0004597848,0.000080173006,0.0000015576896,0.000009175229,0.00025069664],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.97644013,0.006946691,0.00556309,0.0004723224,0.0101989005,0.0003788527],"domain_scores_gemma":[0.93417865,0.058894202,0.0037996538,0.0024389618,0.00041231528,0.00027623057],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07590284,0.00031713335,0.0026649607,0.00027582058,0.000342717,0.00015302276,0.0015295439,0.00013947801,0.0017695234],"category_scores_gemma":[0.067848176,0.00010785531,0.0011142658,0.001309216,0.00011696642,0.00023977025,0.00019042335,0.00025515325,0.00032069764],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000058553815,0.00030936982,0.16832678,0.000022768243,0.00039673378,0.0000019669665,0.0016527483,0.00013176657,0.0000018978388,0.0006091257,0.0032868038,0.8252015],"study_design_scores_gemma":[0.0016940661,0.002937266,0.92053545,0.0006393122,0.00078243786,0.000007851,0.0023136984,0.009155131,0.000056442732,0.00031761083,0.06134813,0.00021261125],"about_ca_topic_score_codex":0.0000068886384,"about_ca_topic_score_gemma":0.00003126664,"teacher_disagreement_score":0.82498884,"about_ca_system_score_codex":0.000075656506,"about_ca_system_score_gemma":0.000087460605,"threshold_uncertainty_score":0.999143},"labels":[],"label_agreement":null},{"id":"W1999959624","doi":"10.1177/0272989x06297389","title":"Costs of Rheumatoid Arthritis: New Estimates from the Human Capital Method and Comparison to the Willingness-to-Pay Method","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":53,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; McGill University Health Centre; McGill University","funders":"","keywords":"Willingness to pay; Contingent valuation; Context (archaeology); Actuarial science; Indirect costs; Medicine; Rheumatoid arthritis; Population; Valuation (finance); Health economics; Public health; Human capital; Environmental health; Economics; Geography; Finance","score_opus":0.2209451114246103,"score_gpt":0.49811101640975813,"score_spread":0.2771659049851478,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1999959624","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19733883,0.003775641,0.77416986,0.023360629,0.0005869733,0.0004848015,0.000053801614,0.000025334473,0.00020413051],"genre_scores_gemma":[0.7113372,0.0001241943,0.27577296,0.012299775,0.00039010233,0.000019416102,0.000008192261,0.000029042529,0.000019147625],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9942314,0.0004079981,0.003917999,0.00059133995,0.0004261617,0.00042509925],"domain_scores_gemma":[0.9752242,0.02233348,0.0012071381,0.0006641188,0.00007692617,0.00049413915],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04354057,0.00020571533,0.0010716716,0.00017600105,0.0003824506,0.00012889001,0.0006927408,0.00018472306,0.0006483594],"category_scores_gemma":[0.026125612,0.0001473099,0.00010357836,0.00032128536,0.000079641606,0.00013633676,0.00026759028,0.00036258495,0.0003680231],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000082698614,0.00011085659,0.17243381,0.000038872964,0.00012605553,0.0000068975537,0.015443211,0.0017083494,0.000020367226,0.10239981,0.05189715,0.6557319],"study_design_scores_gemma":[0.0024202634,0.0002942735,0.519301,0.0037244176,0.000017678918,0.000046869158,0.006095531,0.05557235,0.0001084174,0.2717593,0.13979194,0.00086799054],"about_ca_topic_score_codex":0.0027638732,"about_ca_topic_score_gemma":0.0022620047,"teacher_disagreement_score":0.65486395,"about_ca_system_score_codex":0.00016567759,"about_ca_system_score_gemma":0.00010871181,"threshold_uncertainty_score":0.9848763},"labels":[],"label_agreement":null},{"id":"W2003286188","doi":"10.1177/0272989x07312714","title":"The Cost-Effectiveness of Counseling Strategies to Improve Adherence to Highly Active Antiretroviral Therapy among Men Who Have Sex with Men","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Western University","funders":"National Institute on Drug Abuse","keywords":"Medicine; Psychological intervention; Men who have sex with men; Cost effectiveness; Population; Intervention (counseling); Transmission (telecommunications); Human immunodeficiency virus (HIV); Antiretroviral therapy; Cost-effectiveness analysis; Viral load; Physical therapy; Demography; Family medicine; Environmental health; Psychiatry","score_opus":0.03398645882041067,"score_gpt":0.3803610488315129,"score_spread":0.3463745900111022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2003286188","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8621391,0.00016166916,0.120182745,0.00019105982,0.000034138557,0.0016904826,0.000017431681,0.000035457517,0.015547935],"genre_scores_gemma":[0.9960288,0.0002075187,0.0005332095,0.000090516834,0.00009697844,0.00021347716,0.0000045518013,0.000027619575,0.0027973466],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9964571,0.0002935443,0.00044607316,0.00042051813,0.0018964502,0.0004863287],"domain_scores_gemma":[0.99680656,0.0017373067,0.000106160485,0.0004288966,0.00045904118,0.00046203154],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016038233,0.00020266806,0.0004746213,0.0001840562,0.0003311226,0.00006912018,0.0004057286,0.00011696923,0.0001813523],"category_scores_gemma":[0.0016992625,0.0001137878,0.000119674936,0.0004391853,0.00036401892,0.00019518063,0.00012702521,0.00046125049,0.00002616063],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0069767083,0.00021820876,0.010329246,0.00006871471,0.00022126059,0.00026603654,0.0024284343,0.000061607454,0.008469587,0.00025851835,0.009133365,0.9615683],"study_design_scores_gemma":[0.021793112,0.14291543,0.54569185,0.025783954,0.00017898741,0.00037222708,0.0404752,0.117557205,0.08581634,0.002275298,0.01550237,0.001638034],"about_ca_topic_score_codex":0.00034638983,"about_ca_topic_score_gemma":0.00013226375,"teacher_disagreement_score":0.9599303,"about_ca_system_score_codex":0.00012756707,"about_ca_system_score_gemma":0.00043591904,"threshold_uncertainty_score":0.46401316},"labels":[],"label_agreement":null},{"id":"W2003516918","doi":"10.1177/0272989x10397615","title":"Effect of Treatment Interruption and Intensification of Antiretroviral Therapy on Health-Related Quality of Life in Patients with Advanced HIV","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"HIV Legal Network; Centre for Advancing Health Outcomes; St. Paul's Hospital; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Ontario HIV Treatment Network","keywords":"Medicine; Quality of life (healthcare); Visual analogue scale; Randomized controlled trial; Antiretroviral therapy; Human immunodeficiency virus (HIV); Cohort; Clinical trial; Physical therapy; Internal medicine; Viral load; Family medicine","score_opus":0.07339193055823794,"score_gpt":0.4168533306335525,"score_spread":0.34346140007531456,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2003516918","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99645704,0.00017803401,0.0026384643,0.000066585366,0.00002081083,0.00047198308,0.000009105703,0.000006311058,0.0001516416],"genre_scores_gemma":[0.99941105,0.00012292428,0.0003881893,0.00002642472,0.0000037061536,0.000009230912,0.000015756219,0.0000074351533,0.000015295409],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99809533,0.00033302343,0.000758147,0.00016338374,0.00053117634,0.000118924516],"domain_scores_gemma":[0.9988623,0.00039534413,0.0002965636,0.000190571,0.00013504308,0.00012015942],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010595412,0.00008835186,0.00049495895,0.00024511226,0.000014597895,0.0000012899272,0.00005749254,0.00006585194,0.00016072087],"category_scores_gemma":[0.0020711983,0.000052128715,0.000075053555,0.00020110856,0.00013377375,0.000044821943,0.000016709795,0.00012674206,0.000001836646],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010141823,0.0005349792,0.3472812,0.00009872794,0.000032462387,9.514806e-7,0.000425321,7.1792635e-7,0.00018921372,0.000044307533,0.000023029063,0.64122725],"study_design_scores_gemma":[0.010414231,0.07702591,0.9070672,0.003303348,0.00000850978,7.801493e-7,0.00006978628,0.0010254016,0.0009833786,0.000051096136,0.000006753444,0.000043618285],"about_ca_topic_score_codex":0.00020992232,"about_ca_topic_score_gemma":0.000028621555,"teacher_disagreement_score":0.6411837,"about_ca_system_score_codex":0.00007194733,"about_ca_system_score_gemma":0.00006743101,"threshold_uncertainty_score":0.24795674},"labels":[],"label_agreement":null},{"id":"W2003533845","doi":"10.1177/0272989x13493970","title":"Health Care Costs for State Transition Models in Prostate Cancer","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute of Health Services and Policy Research; Muscular Dystrophy Canada; University of Toronto; Toronto Public Health","funders":"Canadian Institutes of Health Research; Institute for Clinical Evaluative Sciences","keywords":"Medicine; Prostatectomy; Comorbidity; Health care; Prostate cancer; Cancer registry; Cancer; Healthcare Cost and Utilization Project; Demography; Family medicine; Emergency medicine; Internal medicine","score_opus":0.2455579437203913,"score_gpt":0.4760636838137804,"score_spread":0.2305057400933891,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2003533845","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34820154,0.010313106,0.4829406,0.15173912,0.0012756626,0.0041194553,0.00066231645,0.00008828153,0.000659894],"genre_scores_gemma":[0.95593756,0.0005739291,0.011613797,0.03060163,0.00018001998,0.00093228096,0.00004372447,0.000041019503,0.00007604569],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99504864,0.0001998189,0.0034143652,0.00055319857,0.00024328573,0.0005406963],"domain_scores_gemma":[0.99735576,0.0010892728,0.00088008953,0.00027497864,0.00010566857,0.00029425815],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009270017,0.00016278539,0.0008340782,0.00034612272,0.0001543307,0.00008504478,0.00023598032,0.00012117455,0.0005907964],"category_scores_gemma":[0.0019185095,0.00018473071,0.00009572086,0.00020613823,0.000035606932,0.0006025088,0.000035104687,0.00021303231,0.00035424833],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018151462,0.00021117934,0.009096555,0.0016119024,0.000053823194,0.0000067677743,0.034092765,0.019908141,0.0000010257874,0.026882324,0.11957943,0.7883746],"study_design_scores_gemma":[0.004176611,0.00024534648,0.016457489,0.0028290178,0.000003371258,0.000006574482,0.0044460935,0.70244193,0.0000016731744,0.24913144,0.01959689,0.0006635331],"about_ca_topic_score_codex":0.0029999935,"about_ca_topic_score_gemma":0.0022611246,"teacher_disagreement_score":0.787711,"about_ca_system_score_codex":0.0013131528,"about_ca_system_score_gemma":0.00038366276,"threshold_uncertainty_score":0.75331},"labels":[],"label_agreement":null},{"id":"W2006934976","doi":"10.1177/0272989x0202200413","title":"How Robust Is the Health Utilities Index Mark 2 Utility Function?","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; McMaster Children's Hospital; Hamilton Health Sciences; Institute of Health Economics","funders":"","keywords":"Index (typography); Function (biology); Health Utilities Index; Quality-adjusted life year; Econometrics; Actuarial science; Economics; Computer science; Medicine; Operations management; Health related quality of life; Disease; Cost effectiveness","score_opus":0.41523929465081916,"score_gpt":0.42490674263786365,"score_spread":0.009667447987044497,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2006934976","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019954665,0.011880458,0.41029513,0.54315346,0.0030467007,0.0008673119,0.0001612522,0.00014752505,0.010493521],"genre_scores_gemma":[0.90128326,0.0004017618,0.0018759385,0.092465475,0.00068662805,0.00006111581,0.000006074059,0.000033741708,0.003185976],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99416953,0.0005556748,0.0032979096,0.0007483674,0.0005912288,0.0006372572],"domain_scores_gemma":[0.9930892,0.0040481566,0.0014437078,0.0009598379,0.00007700265,0.00038211548],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.024921026,0.00023739418,0.00090496696,0.00025563932,0.000683691,0.00028756258,0.00064215047,0.00026193834,0.017150791],"category_scores_gemma":[0.014970952,0.00021236118,0.00021268567,0.00037588744,0.00021872074,0.00042113318,0.00017466179,0.0005451993,0.0021400827],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000035189456,0.00011587768,0.034744516,0.0001630675,0.000055241588,0.00000273904,0.0017020998,0.000028824808,1.5935399e-8,0.028743707,0.8304067,0.10400199],"study_design_scores_gemma":[0.0006376116,0.00010099906,0.065143555,0.0002925056,0.0000040172117,0.000012016388,0.0018764833,0.16480543,8.3190365e-8,0.051266283,0.71557975,0.0002812818],"about_ca_topic_score_codex":0.00019026613,"about_ca_topic_score_gemma":0.00016466713,"teacher_disagreement_score":0.88132864,"about_ca_system_score_codex":0.00038299971,"about_ca_system_score_gemma":0.00013212764,"threshold_uncertainty_score":0.99863684},"labels":[],"label_agreement":null},{"id":"W2008594868","doi":"10.1177/0272989x10391671","title":"Overestimation Error and Unnecessary Antibiotic Prescriptions for Acute Cystitis in Adult Women","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Urinary Tract Infections Management","field":"Medicine","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Medical prescription; Medicine; Acute Cystitis; Antibiotics; Intensive care medicine; Urinary system; Internal medicine; Pharmacology","score_opus":0.015230183218209561,"score_gpt":0.35006711350243586,"score_spread":0.3348369302842263,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2008594868","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9275585,0.000017010696,0.06835139,0.0014412972,0.0005941702,0.0005773681,0.00000952133,0.000059888094,0.0013909055],"genre_scores_gemma":[0.9747107,0.00003061397,0.024222642,0.0005947764,0.00013152238,0.00006570565,0.000016669934,0.000017604872,0.00020976427],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9986401,0.000021028765,0.000341875,0.00027850526,0.0004615037,0.0002569651],"domain_scores_gemma":[0.9990482,0.0003826141,0.00006614458,0.00024209864,0.00008033803,0.00018060901],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000505871,0.00011634295,0.0002215727,0.00032280144,0.0001041067,0.000037589674,0.000090659145,0.00013244592,0.0005331904],"category_scores_gemma":[0.0019159658,0.00010195094,0.000039204253,0.0002466932,0.00007969669,0.00017555204,0.000095132535,0.00035380118,0.000024235056],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011699707,0.003779217,0.15677732,0.00054223265,0.00030685854,0.000873142,0.0009885265,0.000114973176,0.0032814296,0.019419841,0.009902749,0.80284375],"study_design_scores_gemma":[0.006930973,0.0019402178,0.82676065,0.002156934,0.00026146823,0.0007413802,0.00029716414,0.13283648,0.000105174215,0.016020054,0.011542323,0.00040719792],"about_ca_topic_score_codex":0.000016095226,"about_ca_topic_score_gemma":0.00008224596,"teacher_disagreement_score":0.80243653,"about_ca_system_score_codex":0.00006963404,"about_ca_system_score_gemma":0.00006348685,"threshold_uncertainty_score":0.5838061},"labels":[],"label_agreement":null},{"id":"W2008908609","doi":"10.1177/0272989x14536779","title":"A Review and Meta-analysis of Colorectal Cancer Utilities","year":2014,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":55,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; University of Toronto; Toronto General Hospital; University of Calgary; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Colorectal cancer; Medicine; Stage (stratigraphy); Disease; Cancer; Meta-analysis; Surgery; Internal medicine","score_opus":0.6537234416413402,"score_gpt":0.5737678733815369,"score_spread":0.07995556825980332,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2008908609","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000004104937,0.9912761,0.0036857333,0.0023290205,0.00023939348,0.0009101773,0.0009303984,0.00002103544,0.00060404715],"genre_scores_gemma":[0.000028750208,0.9874132,0.0014499833,0.010345622,0.00009291646,0.00041660285,0.000043547243,0.000040933464,0.00016842555],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9879867,0.0009525851,0.009446287,0.00089732005,0.00036304357,0.00035405857],"domain_scores_gemma":[0.9837108,0.00938557,0.005848979,0.0006797057,0.000096317126,0.00027859482],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03303728,0.00042456243,0.016640862,0.0010142322,0.00009960886,0.000046681464,0.00060889346,0.00050003675,0.01611726],"category_scores_gemma":[0.021205563,0.0003638551,0.0024976598,0.0009272337,0.00017619543,0.00011551153,0.0002301032,0.0004185635,0.00028215896],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000064219594,0.00009003567,0.00009347089,0.22376025,0.15451263,0.00000454936,0.0002603767,0.00000890928,7.1408074e-10,0.026733601,0.062824585,0.5317052],"study_design_scores_gemma":[0.00006703704,0.00001928951,0.000020052776,0.006971298,0.12724048,0.000005366164,0.000010397799,0.0005810561,1.4170738e-9,0.001309511,0.86350244,0.00027307257],"about_ca_topic_score_codex":0.00020279665,"about_ca_topic_score_gemma":0.0002135648,"teacher_disagreement_score":0.80067784,"about_ca_system_score_codex":0.00024264793,"about_ca_system_score_gemma":0.0003442381,"threshold_uncertainty_score":0.9998813},"labels":[],"label_agreement":null},{"id":"W2009301569","doi":"10.1177/0272989x11407203","title":"Development and Validation of a Utility Weighting Function for the Patient-Oriented Prostate Utility Scale (PORPUS)","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Baycrest Hospital; Cancer Care Ontario; University of Toronto; University Health Network","funders":"","keywords":"Weighting; Scale (ratio); Mean squared error; Function (biology); Statistics; Medicine; Computer science; Mathematics; Artificial intelligence","score_opus":0.30268985099678275,"score_gpt":0.4000088148782638,"score_spread":0.09731896388148104,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009301569","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.66209286,0.00040204875,0.33500013,0.0006017651,0.0005598109,0.00070134044,0.000033623906,0.00002203317,0.0005863821],"genre_scores_gemma":[0.96279943,0.000019971572,0.036018502,0.00093985774,0.00006230723,0.00011483297,0.000012930514,0.000013603065,0.000018593291],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9957518,0.00015019182,0.0031316457,0.00044914804,0.00025800645,0.0002592195],"domain_scores_gemma":[0.9959582,0.001980743,0.001467111,0.0003333915,0.00013051476,0.00013003686],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01829645,0.00012959124,0.0004683393,0.00013561141,0.00031704042,0.000025308324,0.0001631913,0.00013282779,0.000627747],"category_scores_gemma":[0.008301977,0.00011373572,0.00007243297,0.00016745021,0.00010883258,0.00022615463,0.0001075281,0.00013558437,0.00004750453],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006037432,0.00034400314,0.2563127,0.000566488,0.000112552516,9.2024095e-7,0.019651257,0.000007509187,0.0000044782455,0.0207167,0.0030174905,0.69866216],"study_design_scores_gemma":[0.0027150675,0.00034245919,0.6175973,0.0010105988,0.00004019477,0.0000100557645,0.0055760155,0.119086616,0.00044763242,0.07517926,0.17739485,0.00059998967],"about_ca_topic_score_codex":0.0000697745,"about_ca_topic_score_gemma":0.000059048994,"teacher_disagreement_score":0.6980622,"about_ca_system_score_codex":0.00010091018,"about_ca_system_score_gemma":0.00012443244,"threshold_uncertainty_score":0.9938842},"labels":[],"label_agreement":null},{"id":"W2009416806","doi":"10.1177/0272989x05275156","title":"Medical Decision Making with Incomplete Evidence—Choosing a Platelet Glycoprotein IIbIIIa Receptor Inhibitor for Percutaneous Coronary Interventions","year":2005,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Acute Myocardial Infarction Research","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"Canadian Institutes of Health Research","keywords":"Tirofiban; Abciximab; Medicine; Eptifibatide; Randomized controlled trial; Placebo; Clinical trial; Intensive care medicine; Conventional PCI; Internal medicine; Myocardial infarction; Alternative medicine","score_opus":0.12697254047013057,"score_gpt":0.4655656324144825,"score_spread":0.33859309194435194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009416806","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00024351655,0.7946757,0.19531552,0.002092529,0.0015309582,0.005523556,0.00012388059,0.00031945907,0.00017484603],"genre_scores_gemma":[0.0010519592,0.86959183,0.12088065,0.0025356414,0.004263854,0.0008358211,0.00023045516,0.00040813774,0.00020163285],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9824444,0.0008147654,0.004033524,0.0021626172,0.008965945,0.001578728],"domain_scores_gemma":[0.9803189,0.014479243,0.0011002641,0.001570528,0.00084709306,0.001683943],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.0066415532,0.0013148853,0.0046271137,0.0012714801,0.00079802464,0.00032306547,0.0016799436,0.0024096859,0.013018228],"category_scores_gemma":[0.034813922,0.00092990947,0.0025764303,0.0016354603,0.00071351195,0.0003716987,0.0015973754,0.004717911,0.0008334223],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002218362,0.00026375198,0.000007996006,0.0064516813,0.0005539931,0.0038977005,0.00005391607,0.000004239162,0.000002400189,0.000034620756,0.026932351,0.959579],"study_design_scores_gemma":[0.0012427494,0.0011461878,0.0000065302365,0.3549766,0.0006773576,0.01763584,0.000036838184,0.0010981462,5.8191586e-7,0.00010610501,0.6224769,0.0005962202],"about_ca_topic_score_codex":0.000016946658,"about_ca_topic_score_gemma":0.000051689345,"teacher_disagreement_score":0.95898277,"about_ca_system_score_codex":0.0019248489,"about_ca_system_score_gemma":0.003215043,"threshold_uncertainty_score":0.99996024},"labels":[],"label_agreement":null},{"id":"W2011300284","doi":"10.1177/0272989x12475092","title":"Predicting the ICECAP-O Capability Index from the WOMAC Osteoarthritis Index","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"WOMAC; Osteoarthritis; Physical therapy; Multinomial logistic regression; Psychology; Medicine; Statistics; Physical medicine and rehabilitation; Mathematics","score_opus":0.1530851830635764,"score_gpt":0.3981294862755075,"score_spread":0.24504430321193107,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2011300284","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88892454,0.003059591,0.038627893,0.063047655,0.0018470023,0.0011359209,0.000110745896,0.000093299175,0.0031533348],"genre_scores_gemma":[0.9676835,0.00005917309,0.0007703753,0.029922744,0.0012082921,0.00022797374,0.000010596968,0.000033743752,0.00008360858],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9932642,0.0007385195,0.0040475056,0.0007431954,0.00062303705,0.00058356114],"domain_scores_gemma":[0.98038596,0.016427109,0.0015080217,0.0012468285,0.0001144511,0.00031762032],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.02289807,0.00024001564,0.00076254626,0.00011025633,0.0007141272,0.0003300608,0.001196941,0.00033518422,0.011171039],"category_scores_gemma":[0.039128006,0.00017731205,0.00018672922,0.00032372389,0.00030568388,0.00047176465,0.00036534216,0.0007971405,0.0040519317],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020179603,0.000061825325,0.8364541,0.000032026484,0.000048805356,0.0000036511647,0.0028642637,0.000117438874,6.2201786e-7,0.005060701,0.049546912,0.1057895],"study_design_scores_gemma":[0.00092840695,0.00004851322,0.7472617,0.0004411385,0.0000056708122,0.000011972528,0.0024693578,0.07639582,5.597908e-7,0.12741783,0.044732805,0.00028618617],"about_ca_topic_score_codex":0.0051478255,"about_ca_topic_score_gemma":0.0014513372,"teacher_disagreement_score":0.12235714,"about_ca_system_score_codex":0.00031277904,"about_ca_system_score_gemma":0.00019336268,"threshold_uncertainty_score":0.99672353},"labels":[],"label_agreement":null},{"id":"W2015076798","doi":"10.1177/0272989x10379919","title":"Determining the Impact of Informed Choice","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Randomized controlled trial; Medicine; Confidence interval; Triage; Quality of life (healthcare); Physical therapy; Internal medicine; Emergency medicine; Nursing","score_opus":0.35454481730403103,"score_gpt":0.6550643630870604,"score_spread":0.30051954578302936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2015076798","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98558,0.00001717604,0.0023142712,0.0011036232,0.00034985415,0.0001946749,0.0000016300958,0.000024940447,0.010413781],"genre_scores_gemma":[0.99441624,0.000046042438,0.004253474,0.00076494203,0.00036133136,0.0000060845573,0.0000022366432,0.000016621201,0.00013305362],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9957438,0.00005047949,0.0006846726,0.00021298506,0.002981595,0.00032644678],"domain_scores_gemma":[0.89930767,0.09806717,0.00021981461,0.001087108,0.0007682432,0.0005499834],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0058502113,0.00011096915,0.00031151884,0.00011634157,0.00010315374,0.000027187787,0.0006112313,0.000516972,0.0033170686],"category_scores_gemma":[0.39605078,0.000054767945,0.0002515362,0.00033667847,0.0005431045,0.0000596113,0.00038873972,0.003647682,0.000077767545],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018331205,0.00013835084,0.3314309,0.00005646688,0.0000396639,0.00003840566,0.00015314847,0.000002989786,0.0009118054,0.0011172594,0.001237427,0.66469026],"study_design_scores_gemma":[0.0015329403,0.0005067795,0.97474694,0.001191852,0.000019516585,0.000046964946,0.000027842121,0.0047436436,0.000114526476,0.01475417,0.0022269958,0.00008782895],"about_ca_topic_score_codex":0.00003101251,"about_ca_topic_score_gemma":0.00019563845,"teacher_disagreement_score":0.66460246,"about_ca_system_score_codex":0.000042529835,"about_ca_system_score_gemma":0.0013046656,"threshold_uncertainty_score":0.99865097},"labels":[],"label_agreement":null},{"id":"W2017353404","doi":"10.1177/0272989x14525264","title":"Sample Size and Power When Designing a Randomized Trial for the Estimation of Treatment, Selection, and Preference Effects","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Selection (genetic algorithm); Estimation; Preference; Sample size determination; Randomized controlled trial; Statistics; Sample (material); Econometrics; Computer science; Medicine; Psychology; Mathematics; Economics; Artificial intelligence; Internal medicine","score_opus":0.07397537297448394,"score_gpt":0.2725037227193427,"score_spread":0.19852834974485878,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2017353404","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39899173,0.00022839349,0.5999698,0.000077977806,0.000106948835,0.00050742645,0.0000033133497,0.0000050126314,0.00010941752],"genre_scores_gemma":[0.9652187,0.00014192805,0.034362443,0.00008819751,0.00004018796,0.00011862447,0.0000015831577,0.000007486371,0.000020799886],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9991906,0.000051173174,0.00041987287,0.00020251508,0.000042782907,0.00009302239],"domain_scores_gemma":[0.986927,0.01271332,0.00022753667,0.000086412394,0.000006510936,0.0000392749],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0019968294,0.00008025024,0.00034782663,0.000042937652,0.000110342386,0.000032302192,0.000055584616,0.00007035391,0.00018232303],"category_scores_gemma":[0.009690284,0.00005808729,0.000055861805,0.000030089122,0.00007283329,0.00008061427,0.000026283931,0.000039910377,0.000004672742],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08479414,0.0001501851,0.014963728,0.00007914255,0.00022709026,1.7586125e-7,0.0018930254,0.0019878943,0.000029426674,0.09851185,0.00022332268,0.79714],"study_design_scores_gemma":[0.2399646,0.0005180377,0.008194104,0.000068686386,0.0000357688,0.0000018557394,0.00001491501,0.4514835,0.000032536143,0.29922637,0.00036204717,0.00009760528],"about_ca_topic_score_codex":0.00004239046,"about_ca_topic_score_gemma":0.0000071189143,"teacher_disagreement_score":0.7970424,"about_ca_system_score_codex":0.000045521807,"about_ca_system_score_gemma":0.000009059102,"threshold_uncertainty_score":0.9986515},"labels":[],"label_agreement":null},{"id":"W2019342141","doi":"10.1177/0272989x0102100504","title":"The Beguiling Pursuit of More Information","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Organ Donation and Transplantation","field":"Medicine","cited_by":57,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Salience (neuroscience); Medicine; Veterans Affairs; Decision aids; Cognition; Psychology; Alternative medicine; Psychiatry; Cognitive psychology","score_opus":0.01268041747478935,"score_gpt":0.3320592304837411,"score_spread":0.3193788130089517,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2019342141","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7861214,0.00021958078,0.19302085,0.009358822,0.0005300728,0.00028054664,0.0000027862127,0.000073117,0.0103927655],"genre_scores_gemma":[0.99704874,0.0002712866,0.0013199666,0.0012004324,0.00007034472,0.0000027515225,0.000011354865,0.0000046222713,0.000070527836],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9982336,0.000015779146,0.00036718702,0.0000617564,0.0012158726,0.00010579739],"domain_scores_gemma":[0.9991478,0.00044781045,0.000082464605,0.00013693252,0.000110246074,0.00007474611],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00066450756,0.000053429198,0.00010244945,0.00008770801,0.00009483754,0.000024635558,0.00012488298,0.00007348061,0.00052526186],"category_scores_gemma":[0.0014426006,0.00003271569,0.000046595727,0.00023047646,0.000061740255,0.00012539828,0.000022207638,0.00013446093,0.00005406068],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021136283,0.000020669879,0.006232463,0.00002051536,0.000010832202,0.000019880688,0.00029208124,0.000011943793,0.00007316464,0.0014978824,0.0006576111,0.9909516],"study_design_scores_gemma":[0.0065349564,0.00034784075,0.6321996,0.002639176,0.0001550594,0.0009779063,0.0015265668,0.05298808,0.0019181207,0.0060862103,0.29433826,0.00028824556],"about_ca_topic_score_codex":0.0000022876732,"about_ca_topic_score_gemma":0.0000055788864,"teacher_disagreement_score":0.99066335,"about_ca_system_score_codex":0.000019689429,"about_ca_system_score_gemma":0.000064126456,"threshold_uncertainty_score":0.57512486},"labels":[],"label_agreement":null},{"id":"W2021478579","doi":"10.1177/0272989x13509406","title":"Appropriate Evidence Sources for Populating Decision Analytic Models within Health Technology Assessment (HTA)","year":2013,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Usability; Computer science; Identification (biology); German; Data science; Variety (cybernetics); Health technology; Resource (disambiguation); Management science; Health care; Political science; Engineering","score_opus":0.6830636891595988,"score_gpt":0.5905819930201459,"score_spread":0.0924816961394529,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021478579","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000083957304,0.55318356,0.44021377,0.0028646023,0.0010428276,0.0023733755,0.000069791764,0.00009748022,0.00007063496],"genre_scores_gemma":[0.00066786265,0.7560502,0.23847325,0.0027742286,0.00063795876,0.0010764194,0.000053541702,0.00015754708,0.00010896715],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9794774,0.00078501,0.015637884,0.0020644485,0.00091019354,0.0011250473],"domain_scores_gemma":[0.9694497,0.015928293,0.012136606,0.0016210321,0.00020985954,0.0006545309],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.059667457,0.000780466,0.006914213,0.0021048917,0.0006826179,0.00035911013,0.0018050403,0.0013899017,0.0006292876],"category_scores_gemma":[0.048438083,0.00074406073,0.0008373751,0.0010771562,0.0001594276,0.0007030415,0.0005996172,0.0012826897,0.00094660843],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000067228134,0.000058299123,0.0000986681,0.008327111,0.00010672626,0.0000029776447,0.00015982865,0.00058168767,3.856168e-9,0.042773183,0.0039454065,0.9439394],"study_design_scores_gemma":[0.00057267805,0.00023054487,0.000020274794,0.102316506,0.000082313585,0.000053389074,0.00036243288,0.27540144,1.1762658e-8,0.27182683,0.34820622,0.0009273481],"about_ca_topic_score_codex":0.00016575398,"about_ca_topic_score_gemma":0.0000866801,"teacher_disagreement_score":0.94301206,"about_ca_system_score_codex":0.0023322434,"about_ca_system_score_gemma":0.0019645249,"threshold_uncertainty_score":0.9999065},"labels":[],"label_agreement":null},{"id":"W2021708092","doi":"10.1177/0272989x06297103","title":"Measuring Health Preferences for Health Utilities Index Mark 3 Health States: A Study of Feasibility and Preference Differences among Ethnic Groups in Singapore","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics","funders":"","keywords":"Ethnic group; Preference; Index (typography); Health Utilities Index; Psychology; Environmental health; Medicine; Actuarial science; Demography; MEDLINE; Economics; Statistics; Political science; Computer science; Sociology; Mathematics","score_opus":0.6418340745295452,"score_gpt":0.5040753170029215,"score_spread":0.1377587575266237,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2021708092","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9286795,0.0070606195,0.057639983,0.0034687864,0.00038698435,0.0025571815,0.00010964594,0.000041477477,0.000055814042],"genre_scores_gemma":[0.99338156,0.00080244,0.0025316537,0.0030722965,0.000085153006,0.00007384082,0.000012947455,0.000026070333,0.000014025256],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9880883,0.0011394704,0.00799531,0.0011112918,0.00067405094,0.0009915483],"domain_scores_gemma":[0.98819536,0.006809626,0.0037131454,0.00055591,0.00010484072,0.0006211159],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.08169451,0.00030709262,0.0021472925,0.0007227051,0.00035560393,0.000070208356,0.0004922369,0.00020046998,0.00011285848],"category_scores_gemma":[0.011015153,0.00031623986,0.00009051165,0.00041297622,0.00022947039,0.0003560212,0.00018424555,0.00050933176,0.0000051694656],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015765727,0.00047232473,0.92360044,0.0012858986,0.00002328041,7.2975206e-7,0.021030556,0.000032335633,2.0332688e-8,0.0009823608,0.00032860774,0.052085806],"study_design_scores_gemma":[0.0016170186,0.0011261384,0.9336142,0.0019342629,0.0000014464533,0.000001991339,0.021589806,0.007030238,9.450872e-8,0.032788604,0.000078514924,0.00021769074],"about_ca_topic_score_codex":0.0056471312,"about_ca_topic_score_gemma":0.06194607,"teacher_disagreement_score":0.07067936,"about_ca_system_score_codex":0.0010820953,"about_ca_system_score_gemma":0.00086322095,"threshold_uncertainty_score":0.99992895},"labels":[],"label_agreement":null},{"id":"W2022284177","doi":"10.1177/0272989x06288684","title":"Implications of Cancer Staging Uncertainties in Radiation Therapy Decisions","year":2006,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Alberta Cancer Foundation; University of Calgary","funders":"","keywords":"Breast cancer; Radiation therapy; Medicine; Medical physics; Cancer; Intensive care medicine; Value of information; Computer science; Surgery; Artificial intelligence; Internal medicine","score_opus":0.429800335296527,"score_gpt":0.6046876730360619,"score_spread":0.17488733773953485,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2022284177","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21917185,0.00054923166,0.77647054,0.001195689,0.00051457767,0.00037894727,0.000037780894,0.00005547073,0.0016258826],"genre_scores_gemma":[0.6156088,0.00045260278,0.3834051,0.00022481436,0.00019909083,0.000061525185,0.0000012422223,0.000022027481,0.000024775014],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9963512,0.00041539996,0.0015569085,0.00033999403,0.0010486504,0.00028785062],"domain_scores_gemma":[0.9087763,0.090220526,0.00034551098,0.00041782693,0.00014809135,0.000091759735],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003932885,0.00015087068,0.0005506576,0.00023095794,0.000076416356,0.000026273428,0.00041257092,0.0002112031,0.0017905455],"category_scores_gemma":[0.10852484,0.00011726332,0.00012480762,0.0006631389,0.0001661643,0.000061088285,0.00010542029,0.00030649867,0.000007370953],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007039202,0.00015911054,0.020109758,0.000009645748,0.000010337541,0.0000049437035,0.000046027453,0.00008184119,0.00009089507,0.11956537,0.0019791008,0.8578726],"study_design_scores_gemma":[0.0009067973,0.00003033421,0.091982454,0.0005453045,0.000011497839,9.846401e-7,0.000025150537,0.0019108139,0.0001583723,0.903423,0.00089311425,0.00011218752],"about_ca_topic_score_codex":0.00013104048,"about_ca_topic_score_gemma":0.00018298942,"teacher_disagreement_score":0.85776037,"about_ca_system_score_codex":0.00010541915,"about_ca_system_score_gemma":0.00015520168,"threshold_uncertainty_score":0.99912196},"labels":[],"label_agreement":null},{"id":"W2022286039","doi":"10.1177/0272989x10364849","title":"Treatment Evolution and New Standards of Care","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Pipeline (software); Probabilistic logic; Sensitivity (control systems); Markov model; Computer science; Econometrics; Markov chain; Risk analysis (engineering); Medicine; Machine learning; Mathematics; Artificial intelligence; Engineering","score_opus":0.22178154802326572,"score_gpt":0.4751929208879055,"score_spread":0.2534113728646398,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2022286039","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7881285,0.0053617316,0.18960251,0.008833582,0.0017312095,0.00045220362,0.00022880756,0.000045044453,0.005616409],"genre_scores_gemma":[0.9869755,0.000098647375,0.011824425,0.0006555845,0.00033236571,0.000009156884,0.000004847032,0.000013274598,0.00008619306],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99723685,0.00006425835,0.00190284,0.0003275868,0.00027493207,0.00019352046],"domain_scores_gemma":[0.9975919,0.0010894461,0.00066641707,0.00031527277,0.000076708646,0.0002602691],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0052360035,0.000108959575,0.00057777954,0.00019894615,0.00009654522,0.000029241055,0.00014572707,0.00018850478,0.0015691695],"category_scores_gemma":[0.011287049,0.000110932255,0.00007253553,0.0001062564,0.000073653595,0.00013446702,0.000052029238,0.00014819941,0.00012121587],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013444618,0.00013481866,0.19847623,0.0003344041,0.00007468352,0.00000937548,0.005752607,0.000052598603,0.000019933008,0.3667544,0.019952113,0.4083044],"study_design_scores_gemma":[0.005872757,0.0010303475,0.19209002,0.0009765208,0.000027453516,0.000067721834,0.0026387668,0.023370486,0.000027397007,0.21618311,0.55697846,0.0007369219],"about_ca_topic_score_codex":0.00032661692,"about_ca_topic_score_gemma":0.00058973907,"teacher_disagreement_score":0.5370264,"about_ca_system_score_codex":0.000382045,"about_ca_system_score_gemma":0.00044175782,"threshold_uncertainty_score":0.9993435},"labels":[],"label_agreement":null},{"id":"W2028788852","doi":"10.1177/0272989x12453499","title":"Prediction of Health Preference Values from CD4 Counts in Individuals with HIV","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ontario HIV Treatment Network; Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"Ontario HIV Treatment Network","keywords":"Akaike information criterion; Statistics; Medicine; Cohort; Preference; Regression analysis; Demography; Stepwise regression; Regression; Linear regression; Mathematics","score_opus":0.45235024493847775,"score_gpt":0.4467406981218301,"score_spread":0.005609546816647659,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2028788852","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.74058986,0.006430599,0.23721738,0.0067197904,0.001314894,0.0006152862,0.00077263627,0.00005547447,0.0062840683],"genre_scores_gemma":[0.9882464,0.00011657865,0.008653541,0.0026783373,0.00020328574,0.000018900733,0.000025948753,0.000013378018,0.000043652584],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9966103,0.00025139967,0.0023190207,0.0002475387,0.0003018691,0.00026990683],"domain_scores_gemma":[0.9971767,0.0013608236,0.0010338856,0.0002569881,0.00002431508,0.00014730044],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.015726505,0.00009266847,0.0005684221,0.00022576023,0.000052795036,0.000015507947,0.00019471691,0.00012764156,0.0020127029],"category_scores_gemma":[0.0036078566,0.000092673385,0.000030818785,0.00017009764,0.00005014424,0.00024748011,0.000038563867,0.00018343431,0.00075197016],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000076265114,0.00008571023,0.9653093,0.00006070668,0.000017087188,2.5007637e-7,0.0012008405,0.00003232968,8.686641e-8,0.0071202107,0.015842931,0.010322907],"study_design_scores_gemma":[0.00079587137,0.00005898134,0.95129216,0.0011229808,0.0000027717388,0.0000022638276,0.00026229818,0.0043432745,6.913624e-7,0.019227903,0.022768285,0.00012253443],"about_ca_topic_score_codex":0.00014823928,"about_ca_topic_score_gemma":0.000023159284,"teacher_disagreement_score":0.24765651,"about_ca_system_score_codex":0.00020877337,"about_ca_system_score_gemma":0.00016652983,"threshold_uncertainty_score":0.9988996},"labels":[],"label_agreement":null},{"id":"W2031455043","doi":"10.1177/0272989x11409240","title":"Accounting for Methodological, Structural, and Parameter Uncertainty in Decision-Analytic Models","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":139,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Centre hospitalier universitaire de Québec","funders":"","keywords":"Econometrics; Management science; Computer science; Accounting; Economics","score_opus":0.6928345373661658,"score_gpt":0.5235144714335428,"score_spread":0.16932006593262305,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2031455043","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.47637016,0.0009534821,0.5197914,0.0013311526,0.00044234787,0.0004945127,0.000031081578,0.000028798831,0.00055703084],"genre_scores_gemma":[0.74892306,0.0000837583,0.2446158,0.0061414856,0.00012280041,0.00007037376,0.0000038772096,0.000022539372,0.000016320557],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9940183,0.00033090872,0.004023563,0.0008329442,0.00027410808,0.00052016263],"domain_scores_gemma":[0.9760731,0.021996614,0.0011354288,0.00046349937,0.00007490356,0.00025647483],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03538065,0.00022186112,0.0012326549,0.00047625438,0.00019034196,0.00009791598,0.00045927364,0.00034805434,0.0014405473],"category_scores_gemma":[0.07220902,0.00021017251,0.00015730006,0.00026640936,0.0001240896,0.0004789525,0.00018805452,0.00030565794,0.00010510552],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007828043,0.00022707695,0.115003996,0.0005562886,0.00020617327,0.000042182433,0.006520823,0.007681884,0.0000018946488,0.47161654,0.01303341,0.38432693],"study_design_scores_gemma":[0.00061471853,0.000046443933,0.013381983,0.00022711433,0.000005087683,0.000008593353,0.00022705323,0.41030875,2.9223997e-7,0.5739706,0.0010248703,0.0001844865],"about_ca_topic_score_codex":0.00022334741,"about_ca_topic_score_gemma":0.00017940394,"teacher_disagreement_score":0.40262687,"about_ca_system_score_codex":0.00022780297,"about_ca_system_score_gemma":0.000090014306,"threshold_uncertainty_score":0.99947226},"labels":[],"label_agreement":null},{"id":"W2031805757","doi":"10.1177/0272989x10369006","title":"Effect of Arrangement of Stick Figures on Estimates of Proportion in Risk Graphics","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Safety Warnings and Signage","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"U.S. National Library of Medicine; Agency for Healthcare Research and Quality","keywords":"Graphics; Quarter (Canadian coin); Mathematics; Statistics; Numeracy; Psychology; Computer science; Geography; Computer graphics (images)","score_opus":0.011112921160223874,"score_gpt":0.36801611053538547,"score_spread":0.3569031893751616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2031805757","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98901194,0.00009372008,0.008895657,0.00003059625,0.0004916297,0.00021319404,0.000005323394,0.000010250957,0.0012477008],"genre_scores_gemma":[0.9988658,0.000017333969,0.0010327082,0.00002336888,0.000031534793,0.000010658574,0.0000023740588,0.000009927306,0.000006289289],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99813557,0.00016000638,0.0006044475,0.0001931593,0.0007611773,0.00014561533],"domain_scores_gemma":[0.9964788,0.0027601428,0.00039637822,0.000263613,0.000052034717,0.00004900594],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0026417512,0.00010107493,0.00034386897,0.00023705419,0.000019896708,0.0000024042554,0.0001969813,0.00018485212,0.0017930198],"category_scores_gemma":[0.005496315,0.000070156646,0.000089817004,0.00024071337,0.00014821014,0.00001731303,0.000042794003,0.00039892568,0.000009906558],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001351295,0.00036997674,0.19398268,0.00016962236,0.00004122983,0.00004427683,0.00075811165,0.00008025801,0.0055826614,0.0050744014,0.00048269183,0.7920628],"study_design_scores_gemma":[0.003427009,0.0033824039,0.95787144,0.002400652,0.00009166105,0.00000885795,0.000106873165,0.006049231,0.013245782,0.012753646,0.0004632891,0.00019915112],"about_ca_topic_score_codex":0.000067676185,"about_ca_topic_score_gemma":0.00003171794,"teacher_disagreement_score":0.7918637,"about_ca_system_score_codex":0.0000055737664,"about_ca_system_score_gemma":0.000016294382,"threshold_uncertainty_score":0.99911946},"labels":[],"label_agreement":null},{"id":"W2032181056","doi":"10.1177/0272989x09344747","title":"Measuring Acceptability of Clinical Decision Rules: Validation of the Ottawa Acceptability of Decision Rules Instrument (OADRI) in Four Countries","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":73,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada; Ottawa Hospital; Canadian Institutes of Health Research; University of Ottawa","funders":"","keywords":"Decision aids; Clinical decision making; Decision rule; Decision analysis; Management science; Medicine; Computer science; Family medicine; Artificial intelligence; Alternative medicine; Engineering; Statistics; Mathematics; Pathology","score_opus":0.06783647977659336,"score_gpt":0.4054685763401585,"score_spread":0.33763209656356513,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2032181056","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98062295,0.00024308667,0.016123265,0.0005941265,0.00087978895,0.00068352494,0.00002697733,0.00002992977,0.00079638336],"genre_scores_gemma":[0.9721666,0.0003022846,0.027048014,0.00029915426,0.00014576426,0.00000807219,0.0000071386035,0.000019053421,0.000003927128],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98901045,0.0006312877,0.004986063,0.0007821938,0.0041634566,0.0004265714],"domain_scores_gemma":[0.92278403,0.07248025,0.0015538076,0.0018196048,0.00089659996,0.0004657149],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.014071313,0.0003208511,0.0017228537,0.00026422038,0.00009118182,0.000023875364,0.00081593514,0.0006451127,0.0024394037],"category_scores_gemma":[0.43772322,0.00020677455,0.0007418536,0.00068295933,0.0006746461,0.00016506703,0.000435643,0.00082884816,0.000016464817],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011239129,0.00072467764,0.40425235,0.00004311574,0.000015903353,0.000011064098,0.00009319016,0.000068687885,0.000018187324,0.00015212163,0.00010694023,0.59338987],"study_design_scores_gemma":[0.003930331,0.0005504686,0.9644116,0.006650706,0.000097311095,0.000020423731,0.00010013392,0.0014673845,0.0017544299,0.020593768,0.00026402826,0.00015940184],"about_ca_topic_score_codex":0.00008045142,"about_ca_topic_score_gemma":0.000066764755,"teacher_disagreement_score":0.5932304,"about_ca_system_score_codex":0.00016549215,"about_ca_system_score_gemma":0.0007256152,"threshold_uncertainty_score":0.9984725},"labels":[],"label_agreement":null},{"id":"W2032651512","doi":"10.1177/0272989x02238299","title":"The Relationship between Risk Attitude and Treatment Choice in Patients with Relapsing-Remitting Multiple Sclerosis","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Montreal Neurological Institute and Hospital","funders":"Multiple Sclerosis Society","keywords":"Medicine; Relapsing remitting; Glatiramer acetate; Multinomial logistic regression; Logistic regression; Multiple sclerosis; Risk factor; Preference; Socioeconomic status; Ordinal regression; Internal medicine; Physical therapy; Psychiatry; Environmental health","score_opus":0.13167752041814887,"score_gpt":0.34605075282160136,"score_spread":0.2143732324034525,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2032651512","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99682283,0.0007313051,0.00064974633,0.00083218334,0.00006289678,0.0005919461,0.0000070664028,0.000041360396,0.0002606609],"genre_scores_gemma":[0.9961483,0.0006189473,0.0028786722,0.00006505742,0.0001438791,0.000044861554,0.000004711384,0.000025898113,0.00006969154],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9968453,0.00021588981,0.0005106116,0.00041900744,0.0015578563,0.00045135518],"domain_scores_gemma":[0.98037285,0.018714966,0.00014688703,0.0003811969,0.000106570245,0.0002775457],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0010799253,0.00019968834,0.00036161835,0.00015998361,0.00069599383,0.00006691297,0.00014795543,0.00013627789,0.00007464458],"category_scores_gemma":[0.039130818,0.000108743094,0.00005920034,0.00042402817,0.0002607539,0.00009540173,0.0001723795,0.00056656066,0.00002915036],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000088972956,0.00008918246,0.62755793,0.0000059976846,0.000029575738,0.0000034006666,0.0001542292,0.0000041833778,7.614277e-7,0.0000031553916,0.00013904506,0.37192357],"study_design_scores_gemma":[0.006232221,0.00036456386,0.98761356,0.0018445714,0.00004334889,0.0000010483028,0.000048288304,0.0028771295,0.0000019877066,0.00003250493,0.00083077024,0.00011000984],"about_ca_topic_score_codex":0.00010028659,"about_ca_topic_score_gemma":0.0005686247,"teacher_disagreement_score":0.37181354,"about_ca_system_score_codex":0.00025820825,"about_ca_system_score_gemma":0.000027990454,"threshold_uncertainty_score":0.96896297},"labels":[],"label_agreement":null},{"id":"W2033279830","doi":"10.1177/0272989x13501558","title":"The Half-Cycle Correction Revisited","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":40,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Discounting; Life expectancy; Computer science; Econometrics; Statistics; Mathematics; Medicine; Population; Economics","score_opus":0.19682844679071176,"score_gpt":0.43982383129914304,"score_spread":0.24299538450843128,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2033279830","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23788299,0.009659011,0.4642878,0.20803729,0.015384951,0.0025446985,0.000028176706,0.00042062928,0.061754473],"genre_scores_gemma":[0.9712185,0.00032802956,0.0035625533,0.022182848,0.0009075339,0.00014289506,0.000007667373,0.000036987094,0.0016129453],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9954152,0.00028357693,0.003135712,0.00043783634,0.00032449287,0.00040317868],"domain_scores_gemma":[0.99155205,0.006412942,0.0010967429,0.00058803667,0.00009309625,0.00025713188],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.018051447,0.0001399996,0.0005127161,0.00016328786,0.00055406854,0.00025469443,0.00047176264,0.00019256238,0.006040989],"category_scores_gemma":[0.03754917,0.00012096588,0.00011841929,0.00027582914,0.00009193198,0.00035513655,0.000113520095,0.0003151072,0.016351825],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015890088,0.000055164677,0.01094225,0.00004291949,0.000034794997,0.0000029323405,0.00037636785,0.00008297449,9.246083e-7,0.06658679,0.67353755,0.24832141],"study_design_scores_gemma":[0.00060490635,0.00006113836,0.05070067,0.00052493945,0.0000038245394,0.000026753642,0.0005094234,0.2399046,0.0000011948599,0.15455766,0.5527832,0.00032167297],"about_ca_topic_score_codex":0.00031058516,"about_ca_topic_score_gemma":0.00007821012,"teacher_disagreement_score":0.73333555,"about_ca_system_score_codex":0.00027560344,"about_ca_system_score_gemma":0.00007836307,"threshold_uncertainty_score":0.9948676},"labels":[],"label_agreement":null},{"id":"W2035994379","doi":"10.1177/0272989x06290494","title":"Context Changes Choices: A Prospective Study of the Effects of Hospitalization on Life-Sustaining Treatment Preferences","year":2006,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Palliative Care and End-of-Life Issues","field":"Medicine","cited_by":150,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Agency for Healthcare Research and Quality","keywords":"Context (archaeology); Prospective cohort study; Medicine; Psychology; Intensive care medicine; Internal medicine","score_opus":0.06572774310177203,"score_gpt":0.4063462662940456,"score_spread":0.34061852319227354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2035994379","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9955956,0.0015101954,0.00065067067,0.00008438612,0.00020701306,0.00091335026,0.0000024051046,0.000014488481,0.0010219008],"genre_scores_gemma":[0.999406,0.0001378532,0.00014087703,0.0000820245,0.00011528797,0.000046233992,0.0000021231303,0.000009854103,0.000059759837],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9982058,0.000106460895,0.00037298663,0.00020777422,0.0009797191,0.00012728735],"domain_scores_gemma":[0.9977173,0.0015053372,0.00029469104,0.00022277184,0.00020505807,0.000054796183],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0002134393,0.00013242444,0.00045955044,0.00010805144,0.000044357686,0.000004820903,0.00012694941,0.00008340691,0.00003729091],"category_scores_gemma":[0.0040076915,0.000070770344,0.000075124786,0.00029804348,0.000099397854,0.000024069355,0.000052177624,0.00008653974,8.73032e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005359536,0.0018516299,0.78875476,0.0002807867,0.0001744233,0.000026838175,0.010228793,0.000029716,0.00017639268,0.0005607624,0.0001275154,0.19725245],"study_design_scores_gemma":[0.008904134,0.015931863,0.92347413,0.020351386,0.00038463357,0.0000022661472,0.0128908325,0.0006204767,0.014810736,0.001397827,0.0010494035,0.00018232882],"about_ca_topic_score_codex":0.00020914892,"about_ca_topic_score_gemma":0.0003002214,"teacher_disagreement_score":0.19707012,"about_ca_system_score_codex":0.00007352051,"about_ca_system_score_gemma":0.00009528049,"threshold_uncertainty_score":0.47978705},"labels":[],"label_agreement":null},{"id":"W2036478249","doi":"10.1177/0272989x13517750","title":"Underestimation of Uncertainties in Health Utilities Derived from Mapping Algorithms Involving Health-Related Quality-of-Life Measures","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; McMaster University; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Estimator; Variance (accounting); Ordinary least squares; Statistics; Nonparametric statistics; Mean squared error; Parametric statistics; Minimum-variance unbiased estimator; Mathematics; Regression; Computer science; Algorithm; Econometrics","score_opus":0.46667541207976365,"score_gpt":0.46527452999341246,"score_spread":0.0014008820863511962,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2036478249","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.55537146,0.009827111,0.41628215,0.017030716,0.00069491105,0.00044298786,0.00007190556,0.000046300655,0.00023245504],"genre_scores_gemma":[0.9640499,0.00033839841,0.030245285,0.005166038,0.00010413429,0.000020583904,0.000034898094,0.00002645992,0.000014301063],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9855044,0.0018901795,0.010962568,0.00056820567,0.00059807143,0.0004765778],"domain_scores_gemma":[0.98308676,0.010156648,0.0057760645,0.00053400244,0.00011981825,0.00032669443],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05353812,0.00021291833,0.0021048952,0.00071932795,0.00017807967,0.000038547783,0.00042128217,0.00025240023,0.0005130039],"category_scores_gemma":[0.06254791,0.00024952902,0.00014736358,0.00041722716,0.00019269198,0.0002875544,0.000119904355,0.0003428882,0.000063306805],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021655379,0.00046108762,0.34954092,0.005100056,0.00025873503,0.0000017291184,0.040767282,0.010060654,0.000018684243,0.39949062,0.004610428,0.18947326],"study_design_scores_gemma":[0.002301527,0.00015260393,0.32385486,0.0060885213,0.0000022870463,0.0000016333908,0.010021807,0.33987904,0.000004049214,0.3155945,0.0016791648,0.0004200242],"about_ca_topic_score_codex":0.017059632,"about_ca_topic_score_gemma":0.002320668,"teacher_disagreement_score":0.40867844,"about_ca_system_score_codex":0.0006038437,"about_ca_system_score_gemma":0.00071761385,"threshold_uncertainty_score":0.9999957},"labels":[],"label_agreement":null},{"id":"W2038410295","doi":"10.1177/0272989x14564433","title":"The Neglected Topic","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Ottawa Hospital Research Institute","keywords":"Decision aids; Decision analysis; Optimal decision; Decision support system; Cost–benefit analysis; Presentation (obstetrics); Cost effectiveness; Medicine; Actuarial science; Computer science; Operations management; Risk analysis (engineering); Decision tree; Business; Economics; Surgery; Alternative medicine; Artificial intelligence; Pathology","score_opus":0.3343279648802716,"score_gpt":0.5118711489690768,"score_spread":0.1775431840888052,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2038410295","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.41911453,0.037453264,0.0832718,0.20306426,0.040220812,0.006204686,0.00001922844,0.0020982446,0.20855318],"genre_scores_gemma":[0.9876807,0.00012858814,0.003946479,0.006762993,0.00045605082,0.0001640227,0.0000051066627,0.000021867905,0.00083424035],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9958367,0.0012884799,0.00079065183,0.00021310942,0.0013819949,0.0004891002],"domain_scores_gemma":[0.9900404,0.007824376,0.00021414149,0.0010019208,0.0004799246,0.00043919106],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0009658724,0.00010608632,0.00017564681,0.00006339846,0.0013997457,0.000019139236,0.0009579017,0.00026120187,0.00064325874],"category_scores_gemma":[0.027927617,0.0000651977,0.000043834516,0.00037524465,0.00010720938,0.000073323456,0.0006512798,0.0011212863,0.0010734266],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016752208,0.000033980763,0.02491667,0.000035315545,0.000011055701,0.000016506468,0.00589919,0.000005559796,0.0000040001655,0.011572866,0.1626945,0.79464287],"study_design_scores_gemma":[0.0014064561,0.000096524695,0.02362868,0.0011841021,0.000006744391,0.0000068108325,0.0074590915,0.0070276754,0.000001516413,0.047216583,0.9118053,0.00016054347],"about_ca_topic_score_codex":0.00007375013,"about_ca_topic_score_gemma":0.000357053,"teacher_disagreement_score":0.7944823,"about_ca_system_score_codex":0.00025021323,"about_ca_system_score_gemma":0.0009849685,"threshold_uncertainty_score":0.9999003},"labels":[],"label_agreement":null},{"id":"W2041968597","doi":"10.1177/0272989x0002000404","title":"Decision Analysis with Cumulative Prospect Theory","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital","funders":"","keywords":"Cumulative prospect theory; Weighting; Prospect theory; Decision analysis; Markov chain; Expected utility hypothesis; Decision theory; Sensitivity (control systems); Statistics; Econometrics; Markov model; Mathematics; Markov process; Economics; Medicine","score_opus":0.19018345995470579,"score_gpt":0.45814137937153054,"score_spread":0.2679579194168248,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2041968597","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.44578356,0.0010836448,0.52907676,0.003305108,0.00028480816,0.0005073864,0.000054741042,0.0001012739,0.019802738],"genre_scores_gemma":[0.95853406,0.00010726847,0.030977972,0.0087944325,0.0002926658,0.000052401538,0.000015437658,0.000039720122,0.001186033],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99409866,0.00036743798,0.0035613358,0.00087247475,0.0006177717,0.00048230478],"domain_scores_gemma":[0.9911049,0.0067165233,0.0009836827,0.0007681285,0.00006737937,0.00035937535],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.024174197,0.00024602888,0.0012476329,0.0006788771,0.000299124,0.00012751161,0.00052032137,0.00024204544,0.044006724],"category_scores_gemma":[0.009545419,0.00022379593,0.00024946153,0.0012071172,0.00014155965,0.0003855646,0.000069974245,0.00033143943,0.0044923215],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010689008,0.00045423603,0.15862374,0.00009671771,0.0012815036,0.00009318518,0.0034901937,0.012215508,2.7832334e-7,0.25533,0.019018851,0.5483269],"study_design_scores_gemma":[0.0027398043,0.0003316782,0.21155994,0.0010111633,0.00014561438,0.0000374585,0.0005075323,0.17326178,0.000001566902,0.5249607,0.08446354,0.0009792651],"about_ca_topic_score_codex":0.00007058755,"about_ca_topic_score_gemma":0.00019063405,"teacher_disagreement_score":0.5473476,"about_ca_system_score_codex":0.00034466793,"about_ca_system_score_gemma":0.00013483607,"threshold_uncertainty_score":0.9987976},"labels":[],"label_agreement":null},{"id":"W2049624880","doi":"10.1177/0272989x04271535","title":"Neonatal Circumcision: Cost-Effective Preventive Measure or “the Unkindest Cut of All”?","year":2004,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Genital Health and Disease","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Agency for Healthcare Research and Quality; U.S. Department of Health and Human Services","keywords":"Measure (data warehouse); Medicine; Obstetrics; Computer science; Data mining","score_opus":0.040423483369984047,"score_gpt":0.3758652697726055,"score_spread":0.33544178640262146,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2049624880","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008474232,0.011590679,0.072014265,0.86899054,0.008176716,0.023858778,0.0006126263,0.00037134532,0.005910826],"genre_scores_gemma":[0.34063396,0.0000685053,0.00094540615,0.6514701,0.004472512,0.00055697793,0.00028599738,0.00013927417,0.0014272516],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9916539,0.000394449,0.0011922265,0.0008627729,0.005067966,0.0008286668],"domain_scores_gemma":[0.9934706,0.0037780171,0.0004981038,0.0010021222,0.00052068545,0.0007304456],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0014811888,0.0005641601,0.001165837,0.00034297412,0.00018593471,0.000035774596,0.0007269559,0.0016248876,0.0033417623],"category_scores_gemma":[0.009792773,0.00028965544,0.0005945425,0.00059582206,0.00042694472,0.00007990856,0.00031884963,0.0037048536,0.0002967672],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002159083,0.0001479077,0.00012125101,0.0009060304,0.000099888086,0.010051094,0.0001824348,0.000002885493,0.0000012671209,0.000018854185,0.395009,0.5913003],"study_design_scores_gemma":[0.0070526768,0.0015449289,0.0047225477,0.013271824,0.0009876223,0.0025024363,0.00013293969,0.00018034683,0.000037533493,0.0071449103,0.9617961,0.0006261148],"about_ca_topic_score_codex":0.000032174503,"about_ca_topic_score_gemma":0.00004377894,"teacher_disagreement_score":0.59067416,"about_ca_system_score_codex":0.00044605692,"about_ca_system_score_gemma":0.0027952942,"threshold_uncertainty_score":0.99995553},"labels":[],"label_agreement":null},{"id":"W2050780994","doi":"10.1177/0272989x06290492","title":"Impact of the Ottawa Decision Support Framework on the Agreement and the Difference between Patients' and Physicians' Decisional Conflict","year":2006,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":182,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; Hôpital Saint-François d'Assise","funders":"","keywords":"Intraclass correlation; Confidence interval; Medicine; Health care; Family medicine; Psychology; Clinical psychology; Internal medicine; Psychometrics","score_opus":0.09306166675939855,"score_gpt":0.4320147542665299,"score_spread":0.3389530875071314,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2050780994","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98112065,0.0005618401,0.009866514,0.0055751367,0.00055493985,0.0015535114,0.00007243147,0.000022288687,0.0006726801],"genre_scores_gemma":[0.9945645,0.00011475116,0.0015436872,0.003385025,0.0002082737,0.00010778719,0.000010269945,0.000023382927,0.000042337077],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9934026,0.00181606,0.0014376235,0.00041867982,0.0024394654,0.0004855718],"domain_scores_gemma":[0.93721646,0.060127586,0.0008015721,0.001345581,0.00037305828,0.00013573225],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0020676376,0.00027181683,0.0005106866,0.00009792426,0.0016586217,0.00003429337,0.0012291961,0.00033739657,0.0006169026],"category_scores_gemma":[0.015052165,0.00011243109,0.00018798867,0.00043082758,0.0007672814,0.000064183085,0.0012596189,0.001593222,0.000035172485],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032620376,0.000093987706,0.69300896,0.000021983496,0.000025603451,6.4875786e-7,0.0019383818,0.000007309782,0.000001738788,0.010814224,0.005741647,0.2880193],"study_design_scores_gemma":[0.001489682,0.00017235591,0.9324857,0.002816826,0.000025517798,7.169785e-7,0.00038873908,0.0007838056,0.000002433487,0.059841808,0.0018754448,0.000116978146],"about_ca_topic_score_codex":0.00033888838,"about_ca_topic_score_gemma":0.000119628465,"teacher_disagreement_score":0.28790233,"about_ca_system_score_codex":0.00015405822,"about_ca_system_score_gemma":0.00029693873,"threshold_uncertainty_score":0.99964106},"labels":[],"label_agreement":null},{"id":"W2051488971","doi":"10.1177/0272989x13480852","title":"Comparing EQ-5D Valuation Studies","year":2013,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":109,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Valuation (finance); Checklist; Respondent; EQ-5D; Medicine; Population; MEDLINE; Cochrane Library; Data extraction; Systematic review; Health care; Economic evaluation; Actuarial science; Meta-analysis; Psychology; Environmental health; Economics; Disease; Health related quality of life; Finance; Political science","score_opus":0.8894377924039364,"score_gpt":0.5889339519329109,"score_spread":0.3005038404710255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2051488971","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00005611954,0.9618603,0.02985651,0.0013919639,0.0031619666,0.001443853,0.000038419745,0.000106636864,0.0020842014],"genre_scores_gemma":[0.00015759707,0.98676836,0.0077080484,0.0029610342,0.001383837,0.00046625925,0.000056447134,0.000100815494,0.00039762596],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9853832,0.0010219131,0.010977972,0.0012158651,0.0007251823,0.00067587843],"domain_scores_gemma":[0.9799864,0.012429718,0.0059187696,0.0010693645,0.00017612916,0.00041958463],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04093306,0.00058593816,0.0063134814,0.0009213991,0.00034967478,0.0002164306,0.0010289078,0.0007887563,0.0030254417],"category_scores_gemma":[0.057646107,0.00057539967,0.00065196824,0.0004983477,0.00015025178,0.00042571418,0.0004926808,0.00086068787,0.023189351],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000001919163,0.000045818786,0.00014654512,0.009095362,0.0002573269,0.000006189405,0.00030654052,0.000013192443,1.1999224e-9,0.021558547,0.059206292,0.90936226],"study_design_scores_gemma":[0.0002902227,0.000030734154,0.000049744667,0.031162629,0.00007187004,0.000025237312,0.00015820342,0.0045537013,4.343097e-9,0.033050504,0.93008393,0.00052322657],"about_ca_topic_score_codex":0.000052356023,"about_ca_topic_score_gemma":0.000036967776,"teacher_disagreement_score":0.90883905,"about_ca_system_score_codex":0.0014532899,"about_ca_system_score_gemma":0.0004889443,"threshold_uncertainty_score":0.99966973},"labels":[],"label_agreement":null},{"id":"W2052111658","doi":"10.1177/0272989x0102100204","title":"Lack of Congruence in the Ratings of Patients' Health Status by Patients and Their Physicians","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":147,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Medicine; Physical therapy; Worry; Rheumatism; Health care; Rheumatoid arthritis; Disease; Fibromyalgia; Family medicine; Internal medicine; Psychiatry; Anxiety","score_opus":0.13263920121081194,"score_gpt":0.45440894255809877,"score_spread":0.32176974134728686,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2052111658","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9936306,0.0010552972,0.0010233031,0.002425634,0.00026418047,0.001057142,0.00009742078,0.000016084452,0.0004303333],"genre_scores_gemma":[0.9918467,0.00051486754,0.0006588516,0.00683336,0.000019117857,0.000057136607,0.000047724323,0.000014987715,0.000007294986],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.995388,0.0015003044,0.0014200214,0.00025038235,0.0009667084,0.00047461299],"domain_scores_gemma":[0.9929442,0.0050508226,0.0009299873,0.0005559381,0.000380348,0.00013873343],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010662965,0.00014215878,0.00042848918,0.00010692867,0.00032562902,0.0000047463627,0.000547338,0.00014883895,0.00011184973],"category_scores_gemma":[0.004414637,0.0000936882,0.000039899176,0.00047626544,0.00020680009,0.000107778156,0.00029744644,0.00068750785,0.000005670445],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008727096,0.0002386057,0.73370415,0.00015202316,0.0000033322317,2.6290104e-7,0.02090782,0.0000014678718,0.000003177477,0.000208891,0.004258024,0.24043496],"study_design_scores_gemma":[0.0033538917,0.0005998127,0.95140284,0.004175298,0.000004408223,2.901019e-7,0.020746091,0.0015487907,0.000008591645,0.0030451035,0.014934832,0.00018004837],"about_ca_topic_score_codex":0.0007790658,"about_ca_topic_score_gemma":0.00050574687,"teacher_disagreement_score":0.24025491,"about_ca_system_score_codex":0.00011837374,"about_ca_system_score_gemma":0.00031088875,"threshold_uncertainty_score":0.5285052},"labels":[],"label_agreement":null},{"id":"W2054085425","doi":"10.1177/0272989x03023002012","title":"Book Review: Epidemiology and the Delivery of Health Care Services. 2nd ed. Edited by Denise M. Oleske. New York: Kluwer Academic/Plenum Publishers, 2001, 401 pages.","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Food Security and Health in Diverse Populations","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cochrane","funders":"","keywords":"Plenum space; Gerontology; Library science; Medicine; Sociology; Computer science; Engineering; Mechanical engineering","score_opus":0.13322909705039676,"score_gpt":0.460704001934636,"score_spread":0.32747490488423925,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2054085425","genre_codex":"review","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00091206504,0.8976326,0.0027095703,0.093431786,0.0021795668,0.0017107198,0.0000701268,0.000076546916,0.001277015],"genre_scores_gemma":[0.0120998155,0.4335503,0.0031376465,0.5485708,0.0009040156,0.00014117126,0.0002204026,0.00006703801,0.0013087725],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98950917,0.0047813,0.0028092028,0.0006221816,0.0010772967,0.0012008643],"domain_scores_gemma":[0.9881516,0.008558793,0.001351568,0.00067929755,0.00033324264,0.0009255073],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012407935,0.000309263,0.0013113775,0.00017071607,0.0014060076,0.000014174574,0.00069127645,0.0009621086,0.008040035],"category_scores_gemma":[0.012833293,0.00021640224,0.00016315753,0.00051038526,0.00037686844,0.000335689,0.00038815496,0.0026523704,0.00010510058],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010171645,0.000021884978,0.0032724075,0.0017169362,0.000020343186,0.0000028309755,0.0020400127,0.0000022021104,2.0814691e-7,0.001196765,0.95873326,0.03289143],"study_design_scores_gemma":[0.0022931648,0.000059933194,0.0011033338,0.009559254,0.000054984914,0.000008328999,0.0021601194,0.00023012621,2.4799954e-7,0.0020879137,0.9822651,0.00017744959],"about_ca_topic_score_codex":0.00085766625,"about_ca_topic_score_gemma":0.0003840676,"teacher_disagreement_score":0.4640823,"about_ca_system_score_codex":0.0002195486,"about_ca_system_score_gemma":0.0014820836,"threshold_uncertainty_score":0.999894},"labels":[],"label_agreement":null},{"id":"W2056047092","doi":"10.1177/0272989x09351590","title":"Comparative Effectiveness Research Through the Looking Glass","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Psychological intervention; Government (linguistics); Corporate governance; Health care; Comparative effectiveness research; Management science; Public relations; Psychology; Engineering ethics; Political science; Medicine; Risk analysis (engineering); Nursing; Economics; Management; Engineering; Law","score_opus":0.5865535459400191,"score_gpt":0.5869153970078773,"score_spread":0.00036185106785824317,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056047092","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.42216745,0.006129456,0.4346281,0.07583742,0.00210065,0.001865207,0.000036329184,0.00014308051,0.057092335],"genre_scores_gemma":[0.98379856,0.00006361255,0.0032322463,0.012152691,0.00059121044,0.000054855926,0.000003841039,0.000016099691,0.00008688576],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9934425,0.0018092593,0.0026993067,0.00067529565,0.00074244244,0.00063124974],"domain_scores_gemma":[0.9782618,0.019999351,0.0007226214,0.0006807422,0.00015215177,0.000183277],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.06348906,0.00018024135,0.0009614992,0.00024610586,0.0007670212,0.00018440315,0.0008481825,0.00022929376,0.0010504067],"category_scores_gemma":[0.023418726,0.00015316663,0.00013564728,0.0006124831,0.00024316402,0.00038572418,0.00015639083,0.00079350895,0.0030548996],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015513334,0.00026509457,0.005979877,0.00012176809,0.00007933977,0.000027949034,0.007855072,0.0006549603,0.000006089531,0.87589645,0.075487815,0.03347047],"study_design_scores_gemma":[0.00110538,0.00025579418,0.087212816,0.001373884,0.0000045782044,0.000026251546,0.0018308753,0.023807568,0.000014161399,0.73033094,0.15366895,0.00036878005],"about_ca_topic_score_codex":0.00011191717,"about_ca_topic_score_gemma":0.00004930301,"teacher_disagreement_score":0.56163114,"about_ca_system_score_codex":0.0005310806,"about_ca_system_score_gemma":0.00019421185,"threshold_uncertainty_score":0.9998628},"labels":[],"label_agreement":null},{"id":"W2056843730","doi":"10.1177/0272989x08315238","title":"Development and Validation of a Risk Scoring Tool to Predict Respiratory Syncytial Virus Hospitalization in Premature Infants Born at 33 through 35 Completed Weeks of Gestation","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Respiratory viral infections research","field":"Medicine","cited_by":75,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Alberta Children's Hospital; Hospital for Sick Children; Mount Sinai Hospital; Abbott (Canada); McMaster Children's Hospital; Izaak Walton Killam Health Centre; University of Toronto; Dalhousie University; University of Calgary; McGill University","funders":"Abbott Laboratories","keywords":"Medicine; Logistic regression; Receiver operating characteristic; Cohort; Pneumovirinae; Cohort study; Pediatrics; Risk factor; Risk assessment; Prospective cohort study; Gestation; Internal medicine; Virus; Pregnancy; Immunology; Paramyxoviridae; Viral disease; Biology","score_opus":0.05848501833394392,"score_gpt":0.3743818404316903,"score_spread":0.3158968220977464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2056843730","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96168405,0.0001651158,0.037083097,0.000022276721,0.000117004834,0.0007102762,0.000015670872,0.00003315261,0.00016937575],"genre_scores_gemma":[0.98851824,0.00009241356,0.011119289,0.00011886971,0.00006278985,0.000038984817,0.000018943421,0.000023666584,0.0000068069544],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963287,0.00020231795,0.0010796357,0.00038159147,0.0017584877,0.00024923385],"domain_scores_gemma":[0.9984928,0.0004860036,0.00028642736,0.00030152153,0.00028845522,0.00014480484],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012610751,0.0001662849,0.00044475027,0.00052895426,0.0001658283,0.000012416995,0.00012443899,0.00023756719,0.0001353762],"category_scores_gemma":[0.0069689853,0.00014697986,0.00004670843,0.0008859104,0.00014230225,0.00022996862,0.00021837086,0.00033233408,0.0000098434875],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021878972,0.00041603122,0.7290774,0.00036270925,0.000066416505,0.0001701975,0.005685729,0.0010842707,0.023804564,0.00013219145,0.00047328227,0.23653932],"study_design_scores_gemma":[0.0023609134,0.00050958205,0.953822,0.0025641697,0.000023965547,0.000038741528,0.000078992285,0.0030032538,0.035555273,0.00013038819,0.0017435699,0.00016913016],"about_ca_topic_score_codex":0.000052679363,"about_ca_topic_score_gemma":0.00009866219,"teacher_disagreement_score":0.23637019,"about_ca_system_score_codex":0.0002629356,"about_ca_system_score_gemma":0.00032177957,"threshold_uncertainty_score":0.83430296},"labels":[],"label_agreement":null},{"id":"W2057806081","doi":"10.1177/0272989x03256005","title":"Validation of a Decision Regret Scale","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":1441,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Manitoba; University of Ottawa; Queen's University; Medical Council of Canada; Ottawa Hospital","funders":"","keywords":"Regret; Scale (ratio); Cronbach's alpha; Psychology; Feeling; Health care; Reliability (semiconductor); Consistency (knowledge bases); Social psychology; Psychometrics; Clinical psychology; Computer science; Statistics; Mathematics; Economics","score_opus":0.1960965971306938,"score_gpt":0.48440904128822826,"score_spread":0.28831244415753443,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2057806081","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8287716,0.0018071581,0.14169234,0.0013267384,0.0032003957,0.0013599598,0.00001704986,0.00018535809,0.021639405],"genre_scores_gemma":[0.9403559,0.00012221737,0.058083005,0.0011273357,0.00008232598,0.00008793917,0.0000112477055,0.000031117725,0.000098930774],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99398655,0.0016011765,0.0016460777,0.00038650376,0.0019058909,0.00047383155],"domain_scores_gemma":[0.98903424,0.008223669,0.00060750893,0.0012365917,0.0005788324,0.00031915517],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002299132,0.00017061233,0.00041347332,0.0002733148,0.0006363664,0.000008520133,0.0006593712,0.00047844963,0.0048296642],"category_scores_gemma":[0.024885824,0.00014245058,0.00010597723,0.0006773161,0.000116876865,0.00016088723,0.00032976066,0.0009236458,0.0004474375],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00056119217,0.00033903803,0.20245585,0.0004978755,0.000027327074,0.000027472015,0.009127537,0.00008798813,0.00031413755,0.018399227,0.027141428,0.7410209],"study_design_scores_gemma":[0.015415123,0.0010989239,0.10001807,0.049189497,0.00014275261,0.00011587477,0.026851462,0.015363905,0.0028492878,0.4183403,0.36865115,0.0019636755],"about_ca_topic_score_codex":0.000033169665,"about_ca_topic_score_gemma":0.00005362655,"teacher_disagreement_score":0.73905724,"about_ca_system_score_codex":0.00020294347,"about_ca_system_score_gemma":0.00066752755,"threshold_uncertainty_score":0.99608004},"labels":[],"label_agreement":null},{"id":"W2058043800","doi":"10.1177/0272989x13507340","title":"Health-Related Quality of Life in HIV-Infected and At-Risk Women","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Hepatitis C virus research","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AIDS Vancouver; Simon Fraser University","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development; National Institute of Allergy and Infectious Diseases; National Center for Research Resources; National Institute on Drug Abuse","keywords":"Medicine; Hepatitis C; Human immunodeficiency virus (HIV); Hepatitis C virus; Internal medicine; Heroin; Multivariate analysis; Viral load; Quality of life (healthcare); Immunology; Cross-sectional study; Drug; Virus; Psychiatry; Pathology","score_opus":0.05086090029211912,"score_gpt":0.3996182501727657,"score_spread":0.34875734988064655,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2058043800","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9945986,0.00055117306,0.00095931307,0.0022855997,0.00006337498,0.00047183887,0.000007221609,0.00003700042,0.0010258787],"genre_scores_gemma":[0.9973102,0.00044643457,0.0012847304,0.0007454016,0.000030266805,0.000033998553,0.000005946181,0.000018403442,0.00012456845],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.995844,0.00056004344,0.0012236428,0.00031586806,0.001581521,0.00047493336],"domain_scores_gemma":[0.9965224,0.001967854,0.00022588547,0.00033356587,0.0001317689,0.0008185176],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0037756301,0.00012090211,0.00073824736,0.0002960617,0.00006642181,0.000013175554,0.00012971087,0.0002224516,0.007420355],"category_scores_gemma":[0.026498126,0.000094239884,0.000047721587,0.0005227107,0.00018142333,0.00006761751,0.00028118448,0.0006045308,0.00028652183],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018194043,0.00008802751,0.5789088,0.00008678043,0.00001582133,0.000024760784,0.00051697704,0.0000011387846,0.00007811154,0.000040886433,0.0018632533,0.41819352],"study_design_scores_gemma":[0.0029931585,0.00022129246,0.9875706,0.0007261851,0.0000022911538,0.0000094995985,0.00008141791,0.006038739,0.000007844756,0.001227127,0.0010349374,0.00008693666],"about_ca_topic_score_codex":0.0019250597,"about_ca_topic_score_gemma":0.0004945044,"teacher_disagreement_score":0.4181066,"about_ca_system_score_codex":0.00029712202,"about_ca_system_score_gemma":0.00036558168,"threshold_uncertainty_score":0.993487},"labels":[],"label_agreement":null},{"id":"W2061622052","doi":"10.1177/0272989x09333120","title":"Should Clinicians Deliver Decision Aids? Further Exploration of the Statin Choice Randomized Trial Results","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Decision aids; Medicine; Randomized controlled trial; Statin; Odds ratio; Confidence interval; Odds; Family medicine; Physical therapy; Alternative medicine; Internal medicine; Logistic regression","score_opus":0.34714318099367986,"score_gpt":0.5220135957669084,"score_spread":0.17487041477322857,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2061622052","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6124839,0.0029355406,0.2923322,0.05301775,0.0137620475,0.016413756,0.00013125814,0.00043503763,0.008488497],"genre_scores_gemma":[0.9722489,0.00028348804,0.018751753,0.007769461,0.00053497154,0.00021722775,0.000020923564,0.000032840962,0.00014042677],"study_design_codex":"design_other","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98644054,0.0058282875,0.0039132955,0.00052576244,0.002758954,0.00053313276],"domain_scores_gemma":[0.94976205,0.0456048,0.0017888247,0.0017865874,0.000796287,0.00026146788],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0070526185,0.00026361158,0.00094121776,0.00022745646,0.0008506818,0.00001949223,0.0013189513,0.00065116875,0.00072834885],"category_scores_gemma":[0.11060416,0.00016261073,0.00043192305,0.00073846785,0.00022668656,0.000358098,0.00045691364,0.0016686623,0.00014905115],"study_design_candidate":"randomized_trial","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.3789897,0.00019396526,0.0003143644,0.000029262015,0.000021760889,0.0000025492423,0.008886924,0.00012509066,0.0000069343896,0.001134815,0.0069463463,0.6033483],"study_design_scores_gemma":[0.8839526,0.0007383407,0.0035737462,0.0066991374,0.000102918224,0.0000013914267,0.005302368,0.015241333,0.0000114831755,0.064853065,0.019163685,0.00035988947],"about_ca_topic_score_codex":0.00016460929,"about_ca_topic_score_gemma":0.00046929112,"teacher_disagreement_score":0.6029884,"about_ca_system_score_codex":0.00019589953,"about_ca_system_score_gemma":0.00064814737,"threshold_uncertainty_score":0.8968876},"labels":[],"label_agreement":null},{"id":"W2065003169","doi":"10.1177/0272989x07300603","title":"A Comparison of EQ-5D Index Scores Derived from the US and UK Population-Based Scoring Functions","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":63,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"EQ-5D; Index (typography); Statistics; Population; Mathematics; Preference; Medicine; Econometrics; Health related quality of life; Computer science; Environmental health","score_opus":0.40972891984436766,"score_gpt":0.4661064639734716,"score_spread":0.05637754412910395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065003169","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6914768,0.001145721,0.30417025,0.0022105165,0.00053118175,0.00020055335,0.000036083453,0.000019249692,0.00020961683],"genre_scores_gemma":[0.9905996,0.000012621248,0.0051562386,0.0038599097,0.0003099133,0.000012008298,0.000022120195,0.000016811253,0.000010820148],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9956221,0.00014907838,0.0032160447,0.00039857457,0.00033347064,0.00028075924],"domain_scores_gemma":[0.9886956,0.009329097,0.0013262848,0.00040651293,0.00005731438,0.00018518903],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012054903,0.00013732164,0.0006904953,0.00024250134,0.00030184767,0.00006242961,0.0002758647,0.00018417386,0.0010651556],"category_scores_gemma":[0.014318749,0.00012904273,0.0000874229,0.00024413763,0.000103324004,0.000121584046,0.00008858723,0.0002655504,0.00011406429],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003651929,0.000036169095,0.9849729,0.000023996005,0.000019025474,9.4953515e-7,0.00032576566,0.0007426027,0.0000026928942,0.0020559395,0.0009939221,0.010789549],"study_design_scores_gemma":[0.0005643107,0.000029697974,0.92284834,0.0004134373,0.0000061528685,0.0000011132807,0.00038213082,0.06696445,0.0000046970017,0.0073398403,0.0013206856,0.00012517406],"about_ca_topic_score_codex":0.0021780469,"about_ca_topic_score_gemma":0.002305925,"teacher_disagreement_score":0.29912275,"about_ca_system_score_codex":0.00017038263,"about_ca_system_score_gemma":0.0000999743,"threshold_uncertainty_score":0.999848},"labels":[],"label_agreement":null},{"id":"W2065598197","doi":"10.1177/0272989x0002000208","title":"Perception of Quantitative Information for Treatment Decisions","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":186,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Context (archaeology); Task (project management); Computer science; Perception; Statistics; Artificial intelligence; Mathematics; Psychology; Engineering","score_opus":0.05602459499976107,"score_gpt":0.40357311940104107,"score_spread":0.34754852440128,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065598197","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.013397969,0.000012920591,0.98519766,0.0002479816,0.0001132214,0.00012589166,0.00001499156,0.000038102262,0.00085128576],"genre_scores_gemma":[0.5622453,0.00023820011,0.4364512,0.0008331921,0.0000339519,0.000015792471,0.00005128346,0.000005167935,0.00012591363],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99870896,0.000031667485,0.00041756665,0.0001355913,0.00058937917,0.00011681166],"domain_scores_gemma":[0.9986587,0.0007968929,0.00008579639,0.00025119566,0.00012404699,0.00008337075],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00039933212,0.00007422993,0.00014722554,0.00013955629,0.000091394504,0.00007143749,0.00038183157,0.00006124837,0.00084038626],"category_scores_gemma":[0.0011684254,0.00005490231,0.00007561613,0.00032109453,0.000031613028,0.00062701904,0.000047115434,0.000025564163,0.00015029417],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020423735,0.0000479037,0.000025712068,0.000002380937,0.000004683886,6.8319156e-7,0.0004951944,0.00019298418,0.0000060336183,0.029321117,0.0012756223,0.96860725],"study_design_scores_gemma":[0.00075848517,0.00034534265,0.0010359553,0.00019808768,0.000008217024,0.0000055470864,0.0002130796,0.9123156,0.000019817826,0.011521142,0.073482916,0.0000957883],"about_ca_topic_score_codex":0.00000430837,"about_ca_topic_score_gemma":0.0000036666272,"teacher_disagreement_score":0.96851146,"about_ca_system_score_codex":0.000039222068,"about_ca_system_score_gemma":0.00007471088,"threshold_uncertainty_score":0.92016405},"labels":[],"label_agreement":null},{"id":"W2068301344","doi":"10.1177/0272989x12465354","title":"Generalized Additive Models for the Analysis of EQ-5D Utility Data","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; University of Toronto","funders":"","keywords":"EQ-5D; Generalized additive model; Computer science; Econometrics; Statistics; Medicine; Mathematics; Health related quality of life; Disease; Internal medicine","score_opus":0.7473666536778095,"score_gpt":0.5185906035936125,"score_spread":0.228776050084197,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2068301344","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.038503177,0.0042548454,0.94649863,0.0057868734,0.0008479516,0.0005911477,0.002501594,0.000022609636,0.0009931918],"genre_scores_gemma":[0.9697919,0.0002182522,0.022104012,0.0070414557,0.00047822087,0.00007795164,0.00020788728,0.000019107445,0.0000612137],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9949819,0.00028295963,0.0034750153,0.00049911684,0.00034228186,0.0004187151],"domain_scores_gemma":[0.9832163,0.013351683,0.0016193808,0.0014740712,0.00010142706,0.0002371753],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04380127,0.00014231924,0.0010912715,0.00034480693,0.00020733016,0.00003681461,0.0010378076,0.00018022578,0.0033833894],"category_scores_gemma":[0.02811762,0.00012235,0.00025165195,0.00053640804,0.00012052202,0.0005462338,0.00033621927,0.0001605476,0.000121782825],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00027368029,0.0005634468,0.04561722,0.00023594483,0.0033224057,0.0000010970492,0.0039646463,0.0047641885,8.850354e-7,0.49129352,0.25068215,0.19928083],"study_design_scores_gemma":[0.0004559152,0.000011841193,0.025824767,0.00005784246,0.00013002659,7.930795e-7,0.0002725808,0.8889852,6.3482275e-7,0.025214754,0.058911357,0.00013429447],"about_ca_topic_score_codex":0.00021195834,"about_ca_topic_score_gemma":0.00015157067,"teacher_disagreement_score":0.9312887,"about_ca_system_score_codex":0.00012159776,"about_ca_system_score_gemma":0.00011216784,"threshold_uncertainty_score":0.99752766},"labels":[],"label_agreement":null},{"id":"W2074177020","doi":"10.1177/0272989x0102100101","title":"Randomized Trial of a Patient Decision Aid for Choice of Surgical Treatment for Breast Cancer","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":183,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University of Ottawa; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Decision aids; Randomized controlled trial; Regret; Medicine; Breast cancer; Mastectomy; Psychological intervention; Physical therapy; Surgery; Alternative medicine; Cancer; Nursing; Internal medicine","score_opus":0.21748743549257993,"score_gpt":0.516672963087418,"score_spread":0.299185527594838,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2074177020","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.90703356,0.0035624595,0.05823979,0.0062523833,0.004561198,0.019130811,0.0006351853,0.00009198174,0.0004926384],"genre_scores_gemma":[0.96671915,0.001720013,0.024262201,0.000435992,0.0005565653,0.0061432705,0.00003698936,0.000049082726,0.0000767573],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.99387753,0.0011134973,0.0027969193,0.0004219491,0.0012599793,0.00053011446],"domain_scores_gemma":[0.9454275,0.051450588,0.0012025218,0.0007759573,0.00085102534,0.00029244914],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0018088107,0.0002396543,0.0013664026,0.00023666723,0.00043508373,0.00000530032,0.00049504166,0.00043656252,0.0011276719],"category_scores_gemma":[0.0103452075,0.00016319045,0.000536926,0.00033147837,0.00017126469,0.00008094509,0.00022396373,0.00029694545,0.000006759666],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.5720782,0.00033888096,0.001422877,0.00012926888,0.000043129323,0.0000016937438,0.0008965103,0.000026047004,0.0000016271639,0.00054296,0.00078237604,0.4237364],"study_design_scores_gemma":[0.88869464,0.000977929,0.00017729409,0.003772875,0.000097092576,0.0000053564645,0.0005993682,0.009302681,0.0000114696095,0.0052985633,0.09089811,0.00016460329],"about_ca_topic_score_codex":0.00042625642,"about_ca_topic_score_gemma":0.00035229186,"teacher_disagreement_score":0.4235718,"about_ca_system_score_codex":0.0003763796,"about_ca_system_score_gemma":0.00090410525,"threshold_uncertainty_score":0.9997854},"labels":[],"label_agreement":null},{"id":"W2076003266","doi":"10.1177/0272989x12454577","title":"Modeling Good Research Practices—Overview","year":2012,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":511,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Task (project management); Computer science; Management science; Task force; Work (physics); Health care; Best practice; Event (particle physics); Data science; Risk analysis (engineering); Operations research; Systems engineering; Engineering; Medicine; Political science","score_opus":0.926401954458632,"score_gpt":0.6814429894895374,"score_spread":0.24495896496909464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2076003266","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000063193643,0.9574464,0.027650554,0.002674441,0.002318108,0.0013435946,0.000101762365,0.0000792255,0.008379592],"genre_scores_gemma":[0.000109328306,0.98621494,0.0071813473,0.002618805,0.0029452562,0.0003226236,0.000053966913,0.00015876863,0.00039495382],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9808977,0.0033284237,0.011391538,0.0015144667,0.0014475598,0.001420292],"domain_scores_gemma":[0.9666611,0.02311395,0.0071592275,0.001871877,0.00023295164,0.00096089276],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.17506087,0.0005639254,0.005001637,0.0014552437,0.00054417475,0.00032495317,0.001892329,0.0014818566,0.0076685986],"category_scores_gemma":[0.115627356,0.00057624327,0.00073435693,0.0011050411,0.00012325388,0.0007828346,0.0008210801,0.0027015137,0.03236534],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004307187,0.0001177542,0.000024437422,0.015095538,0.00010753983,0.000012775277,0.0001701809,0.000017261578,4.4912293e-10,0.057606295,0.009046795,0.9177971],"study_design_scores_gemma":[0.00019687554,0.00003055491,0.0000029306923,0.031332027,0.000064314714,0.00006336997,0.00008440514,0.014936703,8.0346657e-10,0.014308323,0.93848586,0.00049465266],"about_ca_topic_score_codex":0.000232117,"about_ca_topic_score_gemma":0.00008881809,"teacher_disagreement_score":0.92943907,"about_ca_system_score_codex":0.0015671054,"about_ca_system_score_gemma":0.0013417393,"threshold_uncertainty_score":0.99981445},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2080933952","doi":"10.1177/0272989x06290498","title":"The Impact of Unmeasured Clinical Variables on the Accuracy of Hospital Report Cards: A Monte Carlo Study","year":2006,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Monte Carlo method; Medicine; Emergency medicine; Medical emergency; Statistics","score_opus":0.23079714505280916,"score_gpt":0.556669291106792,"score_spread":0.32587214605398285,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2080933952","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98700976,0.00013636959,0.003208643,0.001963163,0.0018086623,0.0013735719,0.00001132029,0.00004067452,0.0044478136],"genre_scores_gemma":[0.99866414,0.00006213852,0.0001683543,0.0003245922,0.00059698464,0.000069687165,0.0000028219808,0.0000150163105,0.00009627626],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9917381,0.0015848551,0.0033273401,0.0002556764,0.0026054874,0.00048855774],"domain_scores_gemma":[0.9662606,0.03033507,0.0017009306,0.00088630617,0.0005891803,0.00022790323],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.023757417,0.00017169619,0.0005854964,0.00008231069,0.0009269996,0.000014749508,0.00061169144,0.00034093394,0.00035771687],"category_scores_gemma":[0.08614408,0.000075057156,0.00026519998,0.0003724964,0.00022943832,0.00007381818,0.00021454431,0.00144393,0.00003151966],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001204755,0.0013911705,0.61181384,0.00015984308,0.00018484387,0.000121555626,0.0039536473,0.00040255487,0.0000026173827,0.0029406552,0.18517677,0.19264776],"study_design_scores_gemma":[0.0024654858,0.002196597,0.95933896,0.003106524,0.00006293263,0.0000090269505,0.0045265304,0.015453952,0.000001681833,0.0047455933,0.007899615,0.00019307845],"about_ca_topic_score_codex":0.0013652836,"about_ca_topic_score_gemma":0.0001577885,"teacher_disagreement_score":0.34752515,"about_ca_system_score_codex":0.00013728208,"about_ca_system_score_gemma":0.0014001135,"threshold_uncertainty_score":0.92155373},"labels":[],"label_agreement":null},{"id":"W2082058463","doi":"10.1177/0272989x0202200213","title":"A Comparison of Bayesian Methods for Profiling Hospital Performance","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":61,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Frequentist inference; Bayesian probability; Ranking (information retrieval); Medicine; Statistics; Outlier; Standardized mortality ratio; Mortality rate; Cohort; Econometrics; Emergency medicine; Actuarial science; Computer science; Bayesian inference; Mathematics; Machine learning; Internal medicine","score_opus":0.1371093887246456,"score_gpt":0.43515864610790855,"score_spread":0.29804925738326293,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2082058463","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014239651,0.002395274,0.9752757,0.0022352787,0.00053440937,0.00035019,0.000012055969,0.000026099056,0.00493134],"genre_scores_gemma":[0.73846465,0.00010256965,0.26085234,0.00036463205,0.00008224834,0.000051043135,0.0000012143558,0.000011426207,0.00006987142],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9984087,0.0000251892,0.0009066841,0.00028012597,0.000099301076,0.0002800349],"domain_scores_gemma":[0.9987847,0.00053920253,0.00029547903,0.00023679266,0.00003094752,0.00011288359],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0019179465,0.000103792954,0.0005377236,0.00018025078,0.00009358562,0.000018642619,0.0002665675,0.00011866111,0.0010780897],"category_scores_gemma":[0.0018713047,0.00009881548,0.00015779141,0.00019325339,0.000041794432,0.00008455534,0.00009041722,0.00014723642,0.000054340064],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000012305789,0.00013901695,0.005617925,0.00016633172,0.000071084294,8.8944796e-7,0.0004510841,0.00004773176,9.742317e-7,0.10657808,0.001372997,0.88554156],"study_design_scores_gemma":[0.00052342715,0.0003573376,0.0014742393,0.00014908037,0.0000255929,0.0000012919685,0.000068411624,0.82169086,0.00012778434,0.031009171,0.14435945,0.00021333469],"about_ca_topic_score_codex":0.000018341132,"about_ca_topic_score_gemma":0.0000035153569,"teacher_disagreement_score":0.88532823,"about_ca_system_score_codex":0.0000391281,"about_ca_system_score_gemma":0.000008944238,"threshold_uncertainty_score":0.9998351},"labels":[],"label_agreement":null},{"id":"W2082104905","doi":"10.1177/0272989x0002000405","title":"A Comparison of HU12 and HU13 Utility Scores in Alzheimer's Disease","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":104,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Health Utilities Index; Psychology; Medicine; Gerontology; Disease; Internal medicine; Health related quality of life","score_opus":0.41433521058841966,"score_gpt":0.5164683261063964,"score_spread":0.10213311551797671,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2082104905","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97746766,0.0113511225,0.0036948833,0.0046822457,0.00021129662,0.00032867363,0.00005403988,0.000021770755,0.0021882977],"genre_scores_gemma":[0.9952557,0.00016383751,0.0020891882,0.0023439082,0.000088540095,0.000019840374,0.0000055564733,0.00001305073,0.000020378453],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99510854,0.00024814953,0.0036353914,0.00046528908,0.00026197327,0.00028066442],"domain_scores_gemma":[0.9962917,0.0023111897,0.0006658476,0.00038313016,0.000023350394,0.00032481473],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011240133,0.00013041467,0.0009516865,0.00024755963,0.00008123874,0.00003253579,0.00025978585,0.00013768517,0.006583437],"category_scores_gemma":[0.007945797,0.00014781773,0.00007820519,0.00020781417,0.00015126048,0.00019596794,0.00007236356,0.00020808428,0.0003743687],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000103716,0.00019273977,0.8123827,0.00008200322,0.000020287169,0.0000058260534,0.00083099183,0.00009874706,1.0728366e-7,0.007103439,0.004223458,0.174956],"study_design_scores_gemma":[0.0007647272,0.000040888695,0.8630477,0.0005876422,0.000007243456,0.0000018832822,0.00015659383,0.09049843,8.458669e-7,0.03381509,0.010898633,0.00018032943],"about_ca_topic_score_codex":0.00023481257,"about_ca_topic_score_gemma":0.00019627607,"teacher_disagreement_score":0.17477566,"about_ca_system_score_codex":0.000066019835,"about_ca_system_score_gemma":0.00010549731,"threshold_uncertainty_score":0.9943247},"labels":[],"label_agreement":null},{"id":"W2083386980","doi":"10.1177/0272989x0002000109","title":"Can the Standard Gamble Be Used to Determine Utilities for Uncertain Health States?","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Mount Sinai Hospital","funders":"","keywords":"Medicine; Visual analogue scale; Logistic regression; Risk factor; Psychological intervention; Surgery; Internal medicine; Psychiatry","score_opus":0.4423482648309067,"score_gpt":0.5014882948136984,"score_spread":0.059140029982791664,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2083386980","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34220085,0.001545497,0.1406785,0.5096202,0.0010045818,0.0021554402,0.0014904329,0.00010438581,0.0012001651],"genre_scores_gemma":[0.7906883,0.0002583611,0.018922923,0.18684673,0.0007228416,0.00039791834,0.00006366408,0.000074838776,0.002024429],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99457675,0.00029118152,0.0034426341,0.0005804807,0.00043160978,0.00067734136],"domain_scores_gemma":[0.9921397,0.006025637,0.00069070776,0.0006085268,0.00006674462,0.00046871277],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.026454562,0.0001978457,0.00094407523,0.00020035924,0.0004928234,0.00013927948,0.0005444635,0.0001301793,0.005964155],"category_scores_gemma":[0.010557847,0.00017757699,0.00014098217,0.00024516153,0.00009195272,0.00013658385,0.00006658183,0.00020603446,0.0003955823],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00039247374,0.000108639724,0.0042793523,0.00039980645,0.00007260509,0.000007514449,0.014609185,0.0021693131,2.7304878e-7,0.044228487,0.5049911,0.42874125],"study_design_scores_gemma":[0.0010951307,0.0003461895,0.0018977429,0.0004298475,0.0000030614567,0.000008109919,0.0014898021,0.036790244,0.0000012572395,0.06631155,0.89135325,0.00027382566],"about_ca_topic_score_codex":0.00074358936,"about_ca_topic_score_gemma":0.0023530019,"teacher_disagreement_score":0.44848743,"about_ca_system_score_codex":0.00068170234,"about_ca_system_score_gemma":0.0004909408,"threshold_uncertainty_score":0.9977766},"labels":[],"label_agreement":null},{"id":"W2084402960","doi":"10.1177/0272989x06297101","title":"Further Insight into the Perception of Quantitative Information: Judgments of Gist in Treatment Decisions","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":109,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cancer Care Ontario; Queen's University","funders":"","keywords":"Context (archaeology); Referent; Scale (ratio); Perception; Statistics; Psychology; Computer science; Mathematics; Geography; Cartography","score_opus":0.21743835805732956,"score_gpt":0.49552677253745137,"score_spread":0.2780884144801218,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2084402960","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92062265,0.0012015166,0.063005455,0.0038001256,0.0010549164,0.0017769336,0.000016571696,0.000040071223,0.008481767],"genre_scores_gemma":[0.9812963,0.0003402083,0.017348059,0.00084302307,0.00003711636,0.000081658,0.000018600113,0.000011419592,0.000023630308],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99543357,0.00074058236,0.0020777597,0.00016801352,0.0012676411,0.00031241786],"domain_scores_gemma":[0.9872714,0.010615913,0.0007372451,0.00075782253,0.000480531,0.00013704288],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00206495,0.00014420421,0.00036643705,0.00041282858,0.0003660313,0.0000051488364,0.0005476002,0.0002646981,0.0008385188],"category_scores_gemma":[0.0075468984,0.000090909765,0.00008803543,0.0007367713,0.0002132865,0.00026148357,0.00030120925,0.00051593076,0.00013488677],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000538057,0.0001860141,0.064155884,0.00007332064,0.000018735493,0.0000030015162,0.23336859,0.00004377261,0.000048448408,0.0038547576,0.00028078462,0.69742864],"study_design_scores_gemma":[0.005037358,0.00093666295,0.6572892,0.007903215,0.000034230496,0.0000036906943,0.26316503,0.006522729,0.000054432316,0.010345021,0.048396796,0.0003116532],"about_ca_topic_score_codex":0.00094001315,"about_ca_topic_score_gemma":0.0027086148,"teacher_disagreement_score":0.697117,"about_ca_system_score_codex":0.00048283246,"about_ca_system_score_gemma":0.00042928697,"threshold_uncertainty_score":0.9181193},"labels":[],"label_agreement":null},{"id":"W2085826762","doi":"10.1177/0272989x0002000102","title":"What Questions Do Patients with Curable Prostate Cancer Want Answered?","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":128,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Regional Cancer Foundation","funders":"","keywords":"Respondent; Medicine; Set (abstract data type); Prostate cancer; Family medicine; Second opinion; Cancer; Computer science; Pathology; Internal medicine","score_opus":0.08099655193544686,"score_gpt":0.4396937906940709,"score_spread":0.358697238758624,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2085826762","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9637557,0.008487271,0.000962414,0.014565765,0.0037305742,0.0032567147,0.00010057954,0.00043241112,0.0047085653],"genre_scores_gemma":[0.98374313,0.0038381435,0.005039905,0.0050804685,0.00018520743,0.0008168193,0.00005388377,0.000064253334,0.0011781766],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9950635,0.00082688493,0.0010539693,0.00051914674,0.0017628081,0.0007736872],"domain_scores_gemma":[0.9957105,0.0017952675,0.00030408875,0.0010773459,0.00061731116,0.00049551897],"candidate_categories":["sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.00053766917,0.00024274716,0.0003663457,0.00014948283,0.0014775279,0.00006781586,0.0006690686,0.00031544932,0.01936169],"category_scores_gemma":[0.00114265,0.00017447931,0.000055266933,0.0005681051,0.00015633475,0.0007072207,0.00022933233,0.0013215612,0.0008014737],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00037858798,0.00015733636,0.17420366,0.00010984572,0.000016018768,0.000010093596,0.0061274236,0.000098095035,4.995586e-7,0.0002867621,0.0070706955,0.81154096],"study_design_scores_gemma":[0.007401149,0.00058860047,0.29156506,0.04436092,0.000067640154,0.0000062169643,0.009333736,0.004154656,0.0000030568922,0.012250817,0.62926924,0.0009989417],"about_ca_topic_score_codex":0.0003465932,"about_ca_topic_score_gemma":0.00065105455,"teacher_disagreement_score":0.81054205,"about_ca_system_score_codex":0.00037832445,"about_ca_system_score_gemma":0.00090682117,"threshold_uncertainty_score":0.9999765},"labels":[],"label_agreement":null},{"id":"W2086352097","doi":"10.1177/0272989x08327114","title":"Development of a Clinical Prediction Model to Calculate Patient Life Expectancy: The Measure of Actuarial Life Expectancy (MALE)","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"University of Waterloo","keywords":"Life expectancy; Actuarial science; Comorbidity; Population; Actuary; Computer science; Psychology; Medicine; Psychiatry","score_opus":0.13146673586044802,"score_gpt":0.39338777883720993,"score_spread":0.2619210429767619,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086352097","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8578812,0.00048469933,0.13635847,0.0006090792,0.00073225796,0.00079609593,0.000008696306,0.000065554355,0.0030639698],"genre_scores_gemma":[0.98144835,0.000075007265,0.017077576,0.00086951064,0.00038347868,0.00006279947,0.000009156205,0.000031601703,0.00004254024],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.993336,0.00013543101,0.0022418804,0.00052815705,0.0033863797,0.00037210484],"domain_scores_gemma":[0.9973831,0.0005321984,0.00044584819,0.00068360165,0.00028019995,0.00067506044],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0012965557,0.0002619934,0.0008210011,0.00020018537,0.00017828894,0.00001193794,0.00037250156,0.00023419481,0.0009643429],"category_scores_gemma":[0.008227836,0.00017462573,0.00031695256,0.00043089446,0.0002307328,0.00008643537,0.00036720876,0.00038913012,0.00002697649],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013382654,0.0040887943,0.019365206,0.00064708205,0.0015550725,0.0007166252,0.026936855,0.006741655,0.0006468733,0.0025112403,0.062366232,0.8610417],"study_design_scores_gemma":[0.036953162,0.004008036,0.3395616,0.021508303,0.0016691166,0.0003128529,0.013023115,0.54870164,0.0009689024,0.0024587403,0.028749045,0.0020854888],"about_ca_topic_score_codex":0.0000051667585,"about_ca_topic_score_gemma":0.000014374293,"teacher_disagreement_score":0.8589562,"about_ca_system_score_codex":0.00011432806,"about_ca_system_score_gemma":0.0024387108,"threshold_uncertainty_score":0.9999489},"labels":[],"label_agreement":null},{"id":"W2086720844","doi":"10.1177/0272989x0102100301","title":"Risk-Adjusted Monitoring of Binary Surgical Outcomes","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":88,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Actua; University of Waterloo","funders":"","keywords":"Chart; Surgical procedures; Medicine; CUSUM; Control chart; Computer science; Cumulative risk; Statistics; Surgery; Mathematics; Internal medicine","score_opus":0.1256849586185509,"score_gpt":0.4730050120026749,"score_spread":0.34732005338412403,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2086720844","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6672402,0.00038928678,0.32864064,0.00019358617,0.0017336558,0.00011155571,0.000009222751,0.000087823624,0.0015940474],"genre_scores_gemma":[0.95580053,0.0001591454,0.043298908,0.00003244177,0.0003787725,0.000008695579,6.7447553e-7,0.00002761722,0.00029323672],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98898494,0.0003602104,0.0019039743,0.00080802373,0.0073607923,0.0005820552],"domain_scores_gemma":[0.9710898,0.026396442,0.0006298661,0.00085150194,0.00053349853,0.00049890945],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0051540197,0.0002824442,0.00083311484,0.00048076297,0.00028095974,0.00010658568,0.0015819181,0.00027151132,0.0020263086],"category_scores_gemma":[0.08854188,0.0001907179,0.00024323507,0.001731953,0.00029899544,0.000392858,0.00066128763,0.00070777955,0.00032558036],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012372812,0.00006610244,0.44195515,0.00000363332,0.000009746433,0.00058366003,0.000052816267,0.0004764222,0.000016602717,0.00041062647,0.0001620494,0.55613947],"study_design_scores_gemma":[0.0018978764,0.0002138276,0.8355974,0.00069712824,0.000036504698,0.00016015703,0.0008019065,0.020010196,0.00014570203,0.12609944,0.01383479,0.00050510856],"about_ca_topic_score_codex":0.000013386548,"about_ca_topic_score_gemma":0.0000044799194,"teacher_disagreement_score":0.5556344,"about_ca_system_score_codex":0.00007598808,"about_ca_system_score_gemma":0.00009946586,"threshold_uncertainty_score":0.998886},"labels":[],"label_agreement":null},{"id":"W2089134311","doi":"10.1177/0272989x0202200214","title":"A Note on the Estimation of Confidence Intervals for Cost-Effectiveness When Costs and Effects Are Censored","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Censoring (clinical trials); Confidence interval; Statistics; Econometrics; Correlation; Cost effectiveness; Point estimation; Mathematics; Medicine","score_opus":0.31198339051714963,"score_gpt":0.46750089246329984,"score_spread":0.15551750194615022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2089134311","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.24340539,0.0012929662,0.7348136,0.016632099,0.00068959943,0.0024786184,0.00009170306,0.000030267269,0.00056577666],"genre_scores_gemma":[0.98610514,0.000052018655,0.0052886093,0.008161422,0.000087012435,0.00026287176,0.0000029560804,0.000018371838,0.000021599106],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9969239,0.00045549465,0.0017627074,0.00039542405,0.00023124806,0.00023119763],"domain_scores_gemma":[0.9687403,0.029531904,0.0011901686,0.00034006976,0.000073592186,0.0001239675],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.015370226,0.00014028202,0.0007117299,0.00017191375,0.00015538582,0.000055993325,0.00026601178,0.00015924251,0.0003915732],"category_scores_gemma":[0.06404649,0.00012263954,0.00009126307,0.00010482494,0.00011473738,0.0001512211,0.00006316986,0.00016334039,0.00019687234],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038033645,0.00035908914,0.010586205,0.0024913598,0.00013058771,0.000010646285,0.0039042411,0.0012159453,0.000015461714,0.45062,0.049930267,0.4803559],"study_design_scores_gemma":[0.0022668764,0.00034519096,0.04321529,0.0065107956,0.0000139975955,0.000011434875,0.00017403591,0.7960065,0.00007952817,0.14462529,0.006441489,0.0003095844],"about_ca_topic_score_codex":0.00003310116,"about_ca_topic_score_gemma":0.000022556767,"teacher_disagreement_score":0.79479057,"about_ca_system_score_codex":0.0002354426,"about_ca_system_score_gemma":0.000023083125,"threshold_uncertainty_score":0.94383746},"labels":[],"label_agreement":null},{"id":"W2090476682","doi":"10.1177/0272989x05280560","title":"Comparing the Standard Rating Scale and the Magnifier Scale for Assessing Risk Perceptions","year":2005,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Promotion and Cardiovascular Prevention","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Health Services Research and Development; Heart and Stroke Foundation of Canada","keywords":"Scale (ratio); Rating scale; Perception; Risk perception; Event (particle physics); Psychology; Medicine; Statistics; Mathematics; Developmental psychology; Geography","score_opus":0.03413117811737237,"score_gpt":0.3917487377443799,"score_spread":0.35761755962700753,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2090476682","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.41170052,0.0004043514,0.58164173,0.004616898,0.00014677555,0.0005111906,0.0000016128514,0.000033427183,0.00094347954],"genre_scores_gemma":[0.97125304,0.00026170645,0.02635193,0.0012849305,0.000656376,0.00006196789,0.0000040662694,0.00001677376,0.00010921768],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9972959,0.0007497942,0.00047712616,0.00025040744,0.0009693075,0.00025743604],"domain_scores_gemma":[0.9981762,0.001112105,0.00011325938,0.000301628,0.00013171165,0.000165086],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0126743205,0.000110798406,0.00033835517,0.000053269687,0.00097244035,0.000124538,0.000118704855,0.000115520386,0.00038512363],"category_scores_gemma":[0.0023057992,0.000057925383,0.00024926,0.00014833287,0.00023884744,0.000112503454,0.00009897984,0.00049965427,0.000020085105],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008659962,0.000025570362,0.008327397,0.000047911242,0.000022128626,0.0000016864111,0.0006121248,0.000045307712,0.000013144767,0.000111501024,0.0007169373,0.9899897],"study_design_scores_gemma":[0.018689552,0.00017748255,0.45358685,0.0034471657,0.00078433455,0.0005933361,0.004740125,0.4478983,0.000046614015,0.0053356783,0.064359315,0.00034122792],"about_ca_topic_score_codex":0.000007723903,"about_ca_topic_score_gemma":0.00007514275,"teacher_disagreement_score":0.98964846,"about_ca_system_score_codex":0.000071597286,"about_ca_system_score_gemma":0.00010311969,"threshold_uncertainty_score":0.7479323},"labels":[],"label_agreement":null},{"id":"W2092038984","doi":"10.1177/0272989x0002000304","title":"Comparison of Finnish and U.S.-based Visual Analog Scale Valuations of the EQ-5D Measure","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics","funders":"","keywords":"EQ-5D; Time-trade-off; Visual analogue scale; Scale (ratio); Dimension (graph theory); Preference; Ordinary least squares; Quality of life (healthcare); Econometrics; Index (typography); Logistic regression; Measure (data warehouse); Statistics; Health related quality of life; Actuarial science; Medicine; Mathematics; Computer science; Economics; Geography; Data mining","score_opus":0.42760834112273943,"score_gpt":0.49110773798635793,"score_spread":0.0634993968636185,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092038984","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.956986,0.001406947,0.031216998,0.007356282,0.0003084851,0.00035895454,0.00008019974,0.000016008578,0.0022701232],"genre_scores_gemma":[0.99358714,0.000023595408,0.003753002,0.002444283,0.000083766674,0.0000143729585,0.0000049369846,0.00001365235,0.0000752541],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9952631,0.00036630727,0.003342551,0.00033538774,0.00047774252,0.00021492434],"domain_scores_gemma":[0.99553114,0.0026334978,0.0012034366,0.0004025469,0.00009330564,0.00013609762],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01302675,0.000120785815,0.0008338084,0.00018348303,0.0001783096,0.000025068253,0.00035200847,0.0001867755,0.0037690978],"category_scores_gemma":[0.008445861,0.00011175921,0.00013369873,0.00033972648,0.00022409392,0.0001083439,0.000058527017,0.00021206822,0.000118845936],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008083437,0.0005025648,0.829214,0.0002824239,0.00006184357,7.252218e-7,0.0015560358,0.0031013435,0.000013324747,0.006835401,0.018088976,0.14026253],"study_design_scores_gemma":[0.001680156,0.00017520913,0.60243076,0.0012790385,0.000025653884,0.0000041722824,0.00042676422,0.3650728,0.00005560406,0.017391698,0.011171051,0.0002871153],"about_ca_topic_score_codex":0.00009738827,"about_ca_topic_score_gemma":0.00015614412,"teacher_disagreement_score":0.36197144,"about_ca_system_score_codex":0.00009021158,"about_ca_system_score_gemma":0.0002183543,"threshold_uncertainty_score":0.9999064},"labels":[],"label_agreement":null},{"id":"W2092725101","doi":"10.1177/0272989x03258421","title":"Validity of Standard Gamble Utilities as Measured by Transplant Readiness in Lung Transplant Candidates","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University Health Network","funders":"","keywords":"Concordance; Medicine; Quality of life (healthcare); Population; Transplantation; Listing (finance); Lung transplantation; Actuarial science; Intensive care medicine; Surgery; Internal medicine; Environmental health; Finance; Nursing","score_opus":0.21076160389631227,"score_gpt":0.42332864785834584,"score_spread":0.21256704396203357,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092725101","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.86307824,0.0037447782,0.12335298,0.0025096473,0.00086852885,0.0005556802,0.0010127856,0.00004258068,0.0048347972],"genre_scores_gemma":[0.9965831,0.0005203444,0.0013461236,0.0012882032,0.000049162416,0.000038886006,0.00003231776,0.000026603735,0.000115269555],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99379694,0.00060259603,0.004028167,0.0005497688,0.0005404148,0.00048210606],"domain_scores_gemma":[0.99529415,0.0031701145,0.0008101983,0.00040000875,0.000065651155,0.0002598735],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.025698146,0.00021833178,0.0012571603,0.0003507029,0.00011089639,0.000041739833,0.00034975415,0.00027917488,0.002731263],"category_scores_gemma":[0.009188039,0.00023725675,0.00013284023,0.00028684005,0.000116561,0.00024012743,0.000016647167,0.00029331163,0.0001307546],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007273673,0.00047284647,0.82857805,0.0018775965,0.00012291623,0.00008504166,0.009666023,0.00023496809,0.00004134895,0.11023871,0.043570936,0.004384178],"study_design_scores_gemma":[0.022302888,0.001302373,0.116878495,0.019367594,0.00014080374,0.00036335835,0.005954674,0.065500885,0.002337861,0.42659897,0.3348529,0.0043992116],"about_ca_topic_score_codex":0.00060434843,"about_ca_topic_score_gemma":0.0006008703,"teacher_disagreement_score":0.7116996,"about_ca_system_score_codex":0.00035971403,"about_ca_system_score_gemma":0.00033963178,"threshold_uncertainty_score":0.99915797},"labels":[],"label_agreement":null},{"id":"W2092756424","doi":"10.1177/0272989x07312726","title":"Technical Note: Acceptability Curves Could Be Misleading When Correlated Strategies Are Compared","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Econometrics; Psychology; Medicine; Computer science; Statistics; Mathematics","score_opus":0.3817359241108896,"score_gpt":0.4748204836448817,"score_spread":0.09308455953399214,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2092756424","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23905766,0.008341618,0.65880626,0.071287155,0.0029523857,0.0016468706,0.00029357593,0.000625759,0.016988698],"genre_scores_gemma":[0.9606946,0.00047749572,0.018159382,0.020051138,0.00031673635,0.00006238184,0.000035992154,0.000045759487,0.00015649755],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9922747,0.00042249897,0.0050815153,0.0009273622,0.00066798273,0.0006259642],"domain_scores_gemma":[0.99137956,0.0052603087,0.0019019897,0.00085637945,0.00014312677,0.00045865466],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.016032659,0.00030994348,0.001555397,0.00031404174,0.00049695076,0.00010878591,0.00083278154,0.0005404709,0.00989577],"category_scores_gemma":[0.03228042,0.0003464808,0.00023757423,0.00037113056,0.00037616165,0.0006415896,0.0002319731,0.00083257916,0.0013503062],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014567486,0.0006572432,0.19143826,0.0010431334,0.00007419964,0.00014322107,0.0029175864,0.00073235657,0.000008687246,0.020421678,0.77315223,0.009265716],"study_design_scores_gemma":[0.004251079,0.00022037154,0.50436276,0.0057000685,0.000039356815,0.00041749972,0.003227298,0.16865626,0.000007890823,0.09373877,0.21731542,0.0020632478],"about_ca_topic_score_codex":0.00018961325,"about_ca_topic_score_gemma":0.00025322966,"teacher_disagreement_score":0.72163695,"about_ca_system_score_codex":0.0006201187,"about_ca_system_score_gemma":0.0004214812,"threshold_uncertainty_score":0.99989873},"labels":[],"label_agreement":null},{"id":"W2093104222","doi":"10.1177/0272989x03258443","title":"The Use of Fixed-and Random-Effects Models for Classifying Hospitals as Mortality Outliers: A Monte Carlo Assessment","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":101,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"","keywords":"Outlier; Logistic regression; False positive paradox; Statistics; Random effects model; Monte Carlo method; Odds; Medicine; Econometrics; Mathematics; Internal medicine; Meta-analysis","score_opus":0.20527028849510268,"score_gpt":0.5113356817594668,"score_spread":0.3060653932643641,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2093104222","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.71137035,0.0005472979,0.28229284,0.00049335667,0.0020771867,0.0027019286,0.000022880437,0.00005536182,0.00043879708],"genre_scores_gemma":[0.9836983,0.0003036232,0.014008618,0.0010142743,0.00007590246,0.000762935,0.000002126758,0.000033310705,0.000100878504],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9946378,0.0015795977,0.001309831,0.00043151475,0.0014405138,0.00060074666],"domain_scores_gemma":[0.9700112,0.028172227,0.00057316286,0.0005435337,0.00036096538,0.00033895235],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0039499486,0.00021164444,0.0005523597,0.00010184029,0.0013332401,0.000031891283,0.00020987848,0.00038529863,0.00008497531],"category_scores_gemma":[0.022472719,0.00014029646,0.00014637467,0.00019080541,0.00015324393,0.0002648714,0.00015091813,0.00077736937,0.0000067333594],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060913543,0.00006434036,0.7281302,0.0011330902,0.0001697056,0.000022209417,0.0027526917,0.0016106408,0.000015918306,0.03572468,0.0063522984,0.22341506],"study_design_scores_gemma":[0.01620442,0.00086106715,0.41804013,0.011129001,0.00024912946,0.000013301195,0.004407959,0.4130885,0.000033381064,0.108814865,0.026354143,0.0008041113],"about_ca_topic_score_codex":0.00019945715,"about_ca_topic_score_gemma":0.00021276942,"teacher_disagreement_score":0.41147786,"about_ca_system_score_codex":0.00023173592,"about_ca_system_score_gemma":0.0005979094,"threshold_uncertainty_score":0.99996686},"labels":[],"label_agreement":null},{"id":"W2096589265","doi":"10.1177/0272989x11426484","title":"An Equity Framework for Health Technology Assessments","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Equity (law); Deliberation; Health technology; Scope (computer science); Checklist; Health care; Business; Process management; Management science; Risk analysis (engineering); Political science; Computer science; Psychology; Economics","score_opus":0.6360861316845253,"score_gpt":0.6119078534373453,"score_spread":0.024178278247179996,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096589265","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021343429,0.00070613396,0.9569615,0.016805626,0.0013202265,0.0006182125,0.00007587798,0.00013376698,0.0020352234],"genre_scores_gemma":[0.57292306,0.0000554246,0.4007392,0.025790233,0.00030261674,0.00012219271,0.000009329728,0.000032623706,0.000025290401],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99453956,0.0001908458,0.0036555862,0.00070846546,0.00027667073,0.00062887697],"domain_scores_gemma":[0.9952395,0.0018287662,0.0016582409,0.00077465863,0.00006838646,0.00043046835],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.027692264,0.00017401697,0.00093401544,0.0004932481,0.00031209673,0.000057243233,0.0008242233,0.0004697276,0.0026522137],"category_scores_gemma":[0.018522838,0.0002007584,0.00010864352,0.0003145148,0.00010178757,0.0003140097,0.00018980309,0.0003775189,0.0007740529],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000038552953,0.00021379809,0.016994718,0.000115047544,0.000026022039,0.000002530777,0.00054505677,0.0000044840863,1.8522806e-7,0.8346617,0.008465234,0.13893263],"study_design_scores_gemma":[0.0005328536,0.00041816998,0.0121037355,0.0004642416,0.0000023808982,0.0000075570256,0.0004439422,0.011914246,0.0000014512974,0.9469075,0.026967473,0.00023645531],"about_ca_topic_score_codex":0.00006899395,"about_ca_topic_score_gemma":0.000061544335,"teacher_disagreement_score":0.5562223,"about_ca_system_score_codex":0.00045570053,"about_ca_system_score_gemma":0.0003429378,"threshold_uncertainty_score":0.9982595},"labels":[],"label_agreement":null},{"id":"W2097820688","doi":"10.1177/0272989x14547915","title":"Societal Preferences for Distributive Justice in the Allocation of Health Care Resources","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":59,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Capital District Health Authority","funders":"Health Research Board","keywords":"Distributive justice; Equity (law); Life expectancy; Rationing; Economics; Quality-adjusted life year; Actuarial science; Health care; Maximization; Redistribution (election); Preference; Economic Justice; Public economics; Microeconomics; Medicine; Cost effectiveness; Operations management","score_opus":0.11115335346204629,"score_gpt":0.31420633945608395,"score_spread":0.20305298599403765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2097820688","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7289303,0.0010116978,0.2666625,0.00093595637,0.00019499379,0.00022557582,0.000040628405,0.0000046774735,0.0019936897],"genre_scores_gemma":[0.99772316,0.000063713276,0.0015838076,0.0005006287,0.00006421959,0.000025260484,0.000027818278,0.000004334863,0.000007032754],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99903566,0.00004753091,0.0005261335,0.00018558504,0.000083772866,0.000121323894],"domain_scores_gemma":[0.99892586,0.000642025,0.0002757467,0.00012447732,0.000008542273,0.000023342725],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0024727122,0.000056191377,0.00019315848,0.000041445634,0.000088308174,0.000015675287,0.00020705583,0.00006306077,0.000106621774],"category_scores_gemma":[0.0009340402,0.000045876808,0.000052311527,0.0000671211,0.000055830573,0.000058477024,0.000033240478,0.00007621904,0.000021376152],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004762305,0.0001321592,0.15569948,0.00018765214,0.000018499224,1.945266e-7,0.012989975,0.00074040995,0.0000014839078,0.13223648,0.0021785724,0.69576746],"study_design_scores_gemma":[0.0008482646,0.0002471029,0.8804101,0.00016641834,0.000008407838,9.485407e-7,0.0060359198,0.02198233,0.000005537389,0.0780104,0.012150053,0.0001345087],"about_ca_topic_score_codex":0.000030476587,"about_ca_topic_score_gemma":0.000040891842,"teacher_disagreement_score":0.72471064,"about_ca_system_score_codex":0.0000839662,"about_ca_system_score_gemma":0.0000141827795,"threshold_uncertainty_score":0.18708019},"labels":[],"label_agreement":null},{"id":"W2098099927","doi":"10.1177/0272989x09357476","title":"Time Lost by Driving Fast in the United States","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Older Adults Driving Studies","field":"Health Professions","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Actuarial science; Psychology; Demography; Gerontology; Computer science; Economics; Sociology","score_opus":0.025680176216391036,"score_gpt":0.41335689956327315,"score_spread":0.38767672334688213,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098099927","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98294246,0.00003841287,0.003541506,0.0085581625,0.0009603642,0.0003810154,0.000009575678,0.00008665813,0.0034818638],"genre_scores_gemma":[0.99106705,0.00005523314,0.00070891576,0.006951813,0.00032255964,0.00003909223,0.000022467832,0.000030542276,0.0008023059],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9965585,0.0005694146,0.0006700591,0.0003212602,0.0012508205,0.0006299758],"domain_scores_gemma":[0.98846763,0.010688185,0.00014766907,0.00044681167,0.00011343438,0.0001362894],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.003246333,0.00017971889,0.00028270768,0.00020785956,0.00068033225,0.000026921596,0.0007240568,0.00027707143,0.004613006],"category_scores_gemma":[0.010325767,0.00010778865,0.000047956633,0.0006594586,0.00021491936,0.00008171838,0.00044201707,0.0022445943,0.0019361412],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000025105974,0.00021899058,0.33994392,0.0000669932,0.000025756459,0.00012809783,0.018544136,0.000017680064,0.0003880214,0.00030751992,0.54189646,0.09843731],"study_design_scores_gemma":[0.0024758535,0.000117220276,0.4189624,0.0053070905,0.000029679915,0.000022361535,0.0121536665,0.041142996,0.000010850495,0.006749078,0.5124273,0.00060156855],"about_ca_topic_score_codex":0.000100976205,"about_ca_topic_score_gemma":0.0025615916,"teacher_disagreement_score":0.09783574,"about_ca_system_score_codex":0.00004973866,"about_ca_system_score_gemma":0.00009410547,"threshold_uncertainty_score":0.998841},"labels":[],"label_agreement":null},{"id":"W2098574716","doi":"10.1177/0272989x15579172","title":"Patients or Volunteers? The Impact of Motivation for Trial Participation on the Efficacy of Patient Decision Aids: A Secondary Analysis of a Cochrane Systematic Review","year":2015,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Feeling; Clinical trial; Subgroup analysis; Psychology; Medicine; Meta-analysis; Family medicine; Clinical psychology; Social psychology; Internal medicine","score_opus":0.3520907661060999,"score_gpt":0.5660942976032334,"score_spread":0.21400353149713347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098574716","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007240391,0.9629428,0.0017320752,0.0000828489,0.0007800365,0.026632426,0.00051906053,0.000024456878,0.0000459156],"genre_scores_gemma":[0.055402927,0.93727964,0.00068344554,0.00040345316,0.00010779121,0.0054015135,0.0006156829,0.00009336339,0.000012167335],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.97958237,0.007807234,0.008229706,0.0005136457,0.0034142681,0.00045276622],"domain_scores_gemma":[0.87799454,0.10701103,0.009127073,0.0028391152,0.0027999666,0.00022829523],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008050391,0.0005099682,0.005613359,0.00081438303,0.000393496,0.000010281678,0.0016332924,0.00063769234,0.0007195202],"category_scores_gemma":[0.2040325,0.00021544109,0.0019574002,0.0030893781,0.00017974625,0.00009287542,0.000496261,0.0011476986,0.000017062934],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005534462,0.00065329164,0.0001039466,0.38688186,0.0014279434,4.547082e-7,0.002399657,0.00005332915,1.6962074e-8,0.00012882438,0.0023825732,0.60043365],"study_design_scores_gemma":[0.007053104,0.0027119508,0.00024148168,0.97446066,0.008684619,4.898231e-7,0.0005588894,0.0019331373,8.370236e-8,0.00021006457,0.0039157085,0.00022982489],"about_ca_topic_score_codex":0.000096046315,"about_ca_topic_score_gemma":0.000046625963,"teacher_disagreement_score":0.6002038,"about_ca_system_score_codex":0.00072539307,"about_ca_system_score_gemma":0.0027081915,"threshold_uncertainty_score":0.87854326},"labels":[],"label_agreement":null},{"id":"W2098643226","doi":"10.1177/0272989x15598540","title":"The CREATE Method for Expressing Continuous Outcome Data in Absolute Terms for Use in Patient Treatment Decision Aids","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hamilton Health Sciences; Montreal Heart Institute; McMaster University; University of Toronto; University Health Network","funders":"","keywords":"Absolute risk reduction; Outcome (game theory); Statistics; Kurtosis; Computer science; Medicine; Mathematics; Confidence interval","score_opus":0.4356436343332973,"score_gpt":0.5447473935557257,"score_spread":0.10910375922242838,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2098643226","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0920723,0.00010783089,0.9059069,0.000100338366,0.00026063385,0.0013558163,0.000049016646,0.00008959545,0.00005755722],"genre_scores_gemma":[0.14641085,0.000029864173,0.85287094,0.00015012972,0.00005665948,0.00038029888,0.000016045633,0.0000460898,0.000039141494],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9965252,0.00019687932,0.001294439,0.0005929512,0.00086917833,0.00052133045],"domain_scores_gemma":[0.97131556,0.0268252,0.00031242147,0.0012287988,0.00013389051,0.00018411575],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0047339494,0.0002706665,0.0006294021,0.00019362765,0.00012424312,0.00015593991,0.00080655684,0.00022376855,0.000011564611],"category_scores_gemma":[0.046572313,0.00016129213,0.00010013346,0.00020123649,0.00007147396,0.00037981058,0.0005754362,0.00021757795,0.0000024251083],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00055479456,0.00015616014,0.0014232205,0.000012483364,0.000009562721,0.000053530246,0.00030964302,0.000028822336,0.000030586634,0.0028058237,0.0011442137,0.99347115],"study_design_scores_gemma":[0.0037962005,0.0008951638,0.00040230353,0.001842065,0.0000406553,0.000027401833,0.00043196956,0.1266396,0.00035160393,0.8373538,0.027858773,0.00036044727],"about_ca_topic_score_codex":0.000044858523,"about_ca_topic_score_gemma":0.00066449033,"teacher_disagreement_score":0.9931107,"about_ca_system_score_codex":0.00037359446,"about_ca_system_score_gemma":0.000096051444,"threshold_uncertainty_score":0.9614588},"labels":[],"label_agreement":null},{"id":"W2100668956","doi":"10.1177/0272989x0202200212","title":"Bayesian Extensions of the Tobit Model for Analyzing Measures of Health Status","year":2002,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Global Health Care Issues","field":"Health Professions","cited_by":45,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Tobit model; Heteroscedasticity; Econometrics; Homoscedasticity; Bayes' theorem; Health Utilities Index; Bayesian probability; Variance (accounting); Statistics; Bayes factor; Actuarial science; Psychology; Mathematics; Medicine; Economics","score_opus":0.3499028085436311,"score_gpt":0.5650245885806403,"score_spread":0.2151217800370092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2100668956","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000008964237,0.8397403,0.15479173,0.00035155012,0.0014394702,0.0030034094,0.0002897384,0.00004442938,0.0003304477],"genre_scores_gemma":[0.001662172,0.98479307,0.011989032,0.00080119865,0.00022919939,0.00019441066,0.000021821335,0.00009926547,0.00020980703],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9883215,0.0022842973,0.004856885,0.0006856168,0.0025383588,0.001313357],"domain_scores_gemma":[0.98527634,0.008777676,0.003289778,0.0013271741,0.0007682631,0.0005607632],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.006654281,0.00049586344,0.0038759983,0.00040716052,0.00091820204,0.000006625016,0.0011568846,0.0010988439,0.00046085948],"category_scores_gemma":[0.01737582,0.00029789368,0.0010422808,0.0010188979,0.00019799465,0.000051957744,0.0006623477,0.0017126106,0.000035504112],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015886922,0.00005581959,0.000053972122,0.020685673,0.000043267508,0.0000015749912,0.00050878583,0.00004160016,2.3935689e-8,0.0006264797,0.02562579,0.95234114],"study_design_scores_gemma":[0.0004039382,0.00007968757,0.00005187717,0.22179261,0.00024664254,0.0000041080184,0.00017143984,0.013891315,2.7849813e-8,0.0020625687,0.76106983,0.00022593055],"about_ca_topic_score_codex":0.00017958302,"about_ca_topic_score_gemma":0.0005280265,"teacher_disagreement_score":0.9521152,"about_ca_system_score_codex":0.0006920762,"about_ca_system_score_gemma":0.0050686477,"threshold_uncertainty_score":0.9999473},"labels":[],"label_agreement":null},{"id":"W2100680479","doi":"10.1177/0272989x03256007","title":"Switching Therapy in Health Economics Trials: Confronting the Confusion","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Confusion; Clinical trial; Medicine; Comparator; Economic analysis; Randomized controlled trial; Intensive care medicine; Economics; Psychology; Surgery; Internal medicine; Engineering","score_opus":0.5105565727209648,"score_gpt":0.5181118890973229,"score_spread":0.00755531637635809,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2100680479","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5951245,0.01804158,0.2777371,0.09116027,0.0046800366,0.0024593025,0.00004000089,0.00011645993,0.010640769],"genre_scores_gemma":[0.9387637,0.0027850934,0.0073593524,0.050446767,0.000423557,0.00007388404,0.0000050945637,0.000048868882,0.00009368085],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.98695034,0.0025466094,0.008860448,0.00071290205,0.00026705387,0.0006626603],"domain_scores_gemma":[0.97964686,0.015276349,0.0040077353,0.0007175863,0.000044490953,0.00030698229],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.16946213,0.00024486092,0.0017321436,0.0003843296,0.0004337567,0.00018284899,0.0005574919,0.00024979716,0.002405839],"category_scores_gemma":[0.069925874,0.00021259849,0.00022156112,0.0002707243,0.00006808334,0.00028953055,0.00008035194,0.0005741549,0.00055587257],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000100542224,0.00020829638,0.033567388,0.00013257543,0.00007463132,0.000007857435,0.005064608,0.0011471063,0.0000021591168,0.6298155,0.015234628,0.31464472],"study_design_scores_gemma":[0.0045697195,0.00017856431,0.0127558075,0.0012724067,0.0000028975378,0.00003673139,0.0029817526,0.066075936,0.0000042186507,0.25414532,0.6573572,0.0006194525],"about_ca_topic_score_codex":0.00043000915,"about_ca_topic_score_gemma":0.00042348655,"teacher_disagreement_score":0.64212257,"about_ca_system_score_codex":0.00057421834,"about_ca_system_score_gemma":0.0005642967,"threshold_uncertainty_score":0.99850607},"labels":[],"label_agreement":null},{"id":"W2102688457","doi":"10.1177/0272989x03258438","title":"A Comparison of the Health Utilities Indices Mark 2 and Mark 3 in Type 2 Diabetes","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Type 2 diabetes; Medicine; Type (biology); Actuarial science; Statistics; Econometrics; Gerontology; Computer science; Mathematics; Economics; Diabetes mellitus; Endocrinology","score_opus":0.30993960717105595,"score_gpt":0.48373390875153405,"score_spread":0.1737943015804781,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102688457","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9734422,0.013824393,0.00086012116,0.008126889,0.0008120899,0.00031959542,0.00002632114,0.0000111572135,0.0025772406],"genre_scores_gemma":[0.99214596,0.00019691743,0.0020702942,0.0054668416,0.000043762317,0.000011137029,0.0000017318185,0.000011424242,0.000051922478],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99565256,0.00056309236,0.0029354,0.0003082638,0.00023753142,0.00030317446],"domain_scores_gemma":[0.9949812,0.0033245466,0.0012231297,0.00031247304,0.000026584763,0.00013202996],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.019691572,0.00010910642,0.0008199074,0.00022501587,0.00012603347,0.00003200891,0.00025497636,0.00013036467,0.00069968076],"category_scores_gemma":[0.02092253,0.00009859199,0.00005176834,0.00030970256,0.00013649084,0.00011360585,0.00008531879,0.00023895585,0.00006040396],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000011245623,0.00006068172,0.9466263,0.00021685546,0.000010644073,2.7909806e-7,0.0016230218,0.0000144668375,1.5169886e-7,0.030766452,0.009432989,0.01123691],"study_design_scores_gemma":[0.0008599188,0.00015408674,0.8258328,0.0015369945,0.000002766926,0.0000030823574,0.0028048053,0.012031764,0.0000042662928,0.07351025,0.08302258,0.00023670933],"about_ca_topic_score_codex":0.00011115591,"about_ca_topic_score_gemma":0.00028527385,"teacher_disagreement_score":0.12079354,"about_ca_system_score_codex":0.0001537662,"about_ca_system_score_gemma":0.00020695468,"threshold_uncertainty_score":0.98732466},"labels":[],"label_agreement":null},{"id":"W2103454095","doi":"10.1177/0272989x07306779","title":"Shared Decision Making in the Medical Encounter: Are We All Talking about the Same Thing?","year":2007,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":229,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"CINAHL; EconLit; MEDLINE; Term (time); Phrase; Computer science; Epistemology; Medicine; Psychology; Artificial intelligence","score_opus":0.38029480385156433,"score_gpt":0.5542990010819531,"score_spread":0.17400419723038874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2103454095","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00003247728,0.96920174,0.0118129505,0.006832523,0.0052032876,0.00439585,0.0000666497,0.00022951946,0.0022250146],"genre_scores_gemma":[0.0034601467,0.9612939,0.0036520439,0.028152674,0.0018425355,0.0012215668,0.00010009719,0.00022600318,0.000051083938],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9690286,0.009745324,0.0070168655,0.0016363834,0.010409592,0.0021632703],"domain_scores_gemma":[0.8874758,0.10303016,0.0036119982,0.0046606814,0.00053040223,0.0006909861],"candidate_categories":["metaresearch","metaepi_narrow","sts","open_science","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.021919841,0.0011602249,0.0030688287,0.001099112,0.002616791,0.0002197178,0.010341436,0.003760511,0.0046872757],"category_scores_gemma":[0.05683716,0.0006226851,0.0009405401,0.002380043,0.0004923215,0.0003218799,0.004414504,0.015104154,0.00093922677],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006878389,0.00009380756,0.0003335126,0.0047263755,0.000047156464,0.00048514112,0.007998792,0.0000019034451,4.374229e-9,0.00062032434,0.01613092,0.96949327],"study_design_scores_gemma":[0.00050105475,0.000034973353,0.00026587906,0.29875293,0.00010550758,0.00015825209,0.004494355,0.0010531083,2.1823092e-9,0.0033513128,0.69086146,0.000421169],"about_ca_topic_score_codex":0.00015692576,"about_ca_topic_score_gemma":0.0028717164,"teacher_disagreement_score":0.9690721,"about_ca_system_score_codex":0.0016370252,"about_ca_system_score_gemma":0.0033259375,"threshold_uncertainty_score":0.99983865},"labels":[],"label_agreement":null},{"id":"W2103642833","doi":"10.1177/0272989x11398489","title":"Use of a Catalytic Model to Estimate Hepatitis A Incidence in a Low-Endemicity Country","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Hepatitis Viruses Studies and Epidemiology","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Toronto General Hospital; Toronto Public Health; University Health Network","funders":"","keywords":"Incidence (geometry); Seroprevalence; Confidence interval; Demography; Medicine; Hepatitis A; Cumulative incidence; Cohort; Hepatitis; Virology; Immunology; Mathematics; Internal medicine","score_opus":0.12874883077390986,"score_gpt":0.3978675051832317,"score_spread":0.26911867440932186,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2103642833","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8451352,0.00006804945,0.15361364,0.00017909326,0.00012574917,0.00023121094,0.000010385025,0.0000230985,0.00061354396],"genre_scores_gemma":[0.9557554,0.00043654314,0.039435007,0.004272004,0.000031419208,0.000024918141,0.0000027693427,0.000017187387,0.000024758567],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99766254,0.00006577417,0.00085364707,0.00036261187,0.000651183,0.00040424732],"domain_scores_gemma":[0.9979199,0.0011808657,0.00012505462,0.00035962465,0.00011211786,0.00030241927],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0009976751,0.00015875707,0.00069518556,0.00019842497,0.000042523672,0.000004760067,0.00018767311,0.00020530981,0.00025922415],"category_scores_gemma":[0.021773411,0.00012701743,0.00007485232,0.0003804513,0.00014239125,0.0000974547,0.00033817225,0.00033420764,0.000049916278],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044430775,0.00016797685,0.95132023,0.00010561696,0.000023136434,0.00054537173,0.0011490404,0.0009289901,0.0000440424,0.00068941934,0.0019367912,0.04264506],"study_design_scores_gemma":[0.0006518628,0.00027110428,0.38909838,0.0044821906,0.000036228314,0.00010980906,0.000020742149,0.6019262,0.000036369256,0.0029346407,0.00025505654,0.00017741951],"about_ca_topic_score_codex":0.013013548,"about_ca_topic_score_gemma":0.021888366,"teacher_disagreement_score":0.6009972,"about_ca_system_score_codex":0.00014063131,"about_ca_system_score_gemma":0.00020552728,"threshold_uncertainty_score":0.99595964},"labels":[],"label_agreement":null},{"id":"W2105531965","doi":"10.1177/0272989x13492017","title":"Physicians’ Tacit and Stated Policies for Determining Patient Benefit and Referral to Cardiac Rehabilitation","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cardiac Health and Mental Health","field":"Medicine","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"York University","funders":"Canadian Institutes of Health Research","keywords":"Referral; Rehabilitation; Medicine; Family medicine; Scale (ratio); Psychology; Physical therapy","score_opus":0.02090035465025003,"score_gpt":0.3704648645458608,"score_spread":0.3495645098956108,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105531965","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99237764,0.00021331693,0.001861185,0.0035483637,0.00033778007,0.0013413985,0.000018926106,0.000033762164,0.00026762052],"genre_scores_gemma":[0.975097,0.00007053082,0.0178662,0.006559538,0.00016983882,0.00017442607,0.000011034818,0.00001899601,0.000032459026],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99836427,0.000037132937,0.0004144483,0.00029510487,0.000523573,0.00036545956],"domain_scores_gemma":[0.9980245,0.00104113,0.000063294356,0.00015232357,0.00012171047,0.00059703365],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00045550347,0.000122030986,0.00037197804,0.00013409666,0.00014258601,0.000028597884,0.000031069903,0.0000987818,0.000036769667],"category_scores_gemma":[0.0014798035,0.00009998107,0.00006107907,0.00014218896,0.00005387557,0.00008454089,0.0000726538,0.00013136711,0.000009180343],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018992621,0.000025609244,0.008369503,0.00018054072,0.0000054860507,0.0000018996913,0.001769216,0.0000011218588,0.00018165649,0.00008358402,0.0030743415,0.9861171],"study_design_scores_gemma":[0.0031665098,0.0057820645,0.92524946,0.00433469,0.000077072415,0.00006667332,0.005240419,0.004164318,0.00014892509,0.003674537,0.047706988,0.00038831568],"about_ca_topic_score_codex":0.00013441058,"about_ca_topic_score_gemma":0.000021056763,"teacher_disagreement_score":0.9857288,"about_ca_system_score_codex":0.00010099028,"about_ca_system_score_gemma":0.0000732113,"threshold_uncertainty_score":0.407711},"labels":[],"label_agreement":null},{"id":"W2106584438","doi":"10.1177/0272989x0002000113","title":"An Assessment of Methods to Combine Published Survival Curves","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":67,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Regional Cancer Foundation; University of Ottawa","funders":"","keywords":"Censoring (clinical trials); Covariate; Statistics; Survival analysis; Medicine; Mathematics; Oncology","score_opus":0.5056615645167237,"score_gpt":0.6965193766699368,"score_spread":0.19085781215321307,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106584438","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0136656435,0.000043336997,0.97525597,0.0011739513,0.0010231535,0.00042971756,0.000022748782,0.000103617946,0.008281847],"genre_scores_gemma":[0.028940538,0.00006751102,0.96905327,0.0015468469,0.00023521097,0.000028255163,0.0000019314728,0.00003979452,0.00008662943],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9906761,0.0038206938,0.0021394705,0.0006228829,0.0023199802,0.00042088295],"domain_scores_gemma":[0.8560016,0.14155526,0.00028999313,0.0011709777,0.0002742587,0.000707889],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03169734,0.00022873537,0.0012385757,0.00013893528,0.000071925584,0.000063998086,0.0010520088,0.00027661648,0.046198063],"category_scores_gemma":[0.3404889,0.00017238407,0.00018625516,0.00068348553,0.00017135637,0.00012759017,0.0002168478,0.00055082253,0.00003427034],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000108035376,0.0004997255,0.00032439426,0.00011080165,0.0000317755,0.000019313777,0.000022074308,0.000007630205,0.00007554073,0.04486774,0.0070525063,0.94688046],"study_design_scores_gemma":[0.0010412233,0.00048780435,0.0066789254,0.001900926,0.00006803302,0.000006524569,0.000018553541,0.007823421,0.00010765885,0.9787047,0.0029150913,0.00024711888],"about_ca_topic_score_codex":0.000006409715,"about_ca_topic_score_gemma":0.0000052507776,"teacher_disagreement_score":0.94663334,"about_ca_system_score_codex":0.00004752913,"about_ca_system_score_gemma":0.0001525538,"threshold_uncertainty_score":0.9970714},"labels":[],"label_agreement":null},{"id":"W2106835183","doi":"10.1177/0272989x10370815","title":"Comparative Effectiveness Research: Challenges for Medical Journals","year":2010,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Comparative effectiveness research; Medical research; Alternative medicine; Medical journal; Medicine; MEDLINE; Medical education; Family medicine; Political science; Pathology","score_opus":0.7167079717686192,"score_gpt":0.6301479999372894,"score_spread":0.0865599718313298,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2106835183","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011602029,0.030024122,0.055699702,0.023175843,0.8824978,0.0023583323,0.00080531806,0.00010460325,0.0041740914],"genre_scores_gemma":[0.009327446,0.0066508753,0.0062341727,0.001972294,0.97405434,0.0011211107,0.00017463739,0.0001971801,0.00026792064],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98156506,0.003182108,0.007992307,0.0018438997,0.004080791,0.0013358216],"domain_scores_gemma":[0.74044913,0.2507676,0.0041559404,0.0014867026,0.0015273558,0.0016132786],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.2776988,0.00054486777,0.004057441,0.0014343462,0.0007674193,0.0003368307,0.0023002187,0.004792672,0.0052968455],"category_scores_gemma":[0.48795664,0.0005731961,0.00050922536,0.00039334572,0.0005320505,0.00035466335,0.0005538067,0.006008456,0.0026326717],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016490062,0.00016317204,0.00003732272,0.0010036337,0.0001417386,0.000021992331,0.00086600723,0.0000045132283,2.0877323e-7,0.010820251,0.97369653,0.01307975],"study_design_scores_gemma":[0.0016205169,0.00023010506,0.0003592902,0.006024441,0.000010238794,0.000010392019,0.00034978782,0.002667405,8.3214513e-7,0.10652659,0.88171166,0.000488763],"about_ca_topic_score_codex":0.000079283294,"about_ca_topic_score_gemma":0.00062694476,"teacher_disagreement_score":0.24758548,"about_ca_system_score_codex":0.0010009831,"about_ca_system_score_gemma":0.0028069944,"threshold_uncertainty_score":0.99967194},"labels":[],"label_agreement":null},{"id":"W2107324201","doi":"10.1177/0272989x04267008","title":"Refining Estimates of Major Depression Incidence and Episode Duration in Canada Using a Monte Carlo Markov Model","year":2004,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Mental Health Research Topics","field":"Psychology","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Incidence (geometry); Duration (music); Depression (economics); Statistics; Markov chain Monte Carlo; Markov model; Monte Carlo method; Econometrics; Computer science; Medicine; Markov process; Mathematics","score_opus":0.05948164965937775,"score_gpt":0.4307853803543425,"score_spread":0.37130373069496475,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107324201","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9409169,0.0007011077,0.0576524,0.00020270872,0.00017592622,0.00013192225,0.0000043151927,0.00000838123,0.00020632769],"genre_scores_gemma":[0.94545096,0.00001754891,0.05423519,0.00023720402,0.000021271204,0.000012951735,9.727364e-7,0.000012642166,0.000011267949],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99759215,0.00010187335,0.0006236937,0.000311566,0.0010163272,0.00035438078],"domain_scores_gemma":[0.99870455,0.00066881743,0.00014596953,0.000226064,0.000047952108,0.00020664181],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010010584,0.00010956383,0.00024351901,0.00015098913,0.000089555935,0.00001191062,0.00019546076,0.00012471896,0.00015936313],"category_scores_gemma":[0.0017707959,0.00009588435,0.000017071527,0.00021538889,0.000060425817,0.000101166086,0.00019580526,0.00034698116,0.0000017845787],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004649963,0.000059908747,0.24500526,0.00016015241,0.00000979005,0.0006999498,0.0007915442,0.017047873,0.000529471,0.00040505308,0.00011982872,0.73470616],"study_design_scores_gemma":[0.0014615541,0.000050633025,0.10108833,0.002933335,0.0000065565537,0.000111489455,0.00024502646,0.890203,0.00016607593,0.003579676,0.000012404983,0.00014194194],"about_ca_topic_score_codex":0.6219445,"about_ca_topic_score_gemma":0.7765346,"teacher_disagreement_score":0.8731551,"about_ca_system_score_codex":0.0007377015,"about_ca_system_score_gemma":0.00094296306,"threshold_uncertainty_score":0.39100504},"labels":[],"label_agreement":null},{"id":"W2107667787","doi":"10.1177/0272989x13491463","title":"Validation of SURE, a Four-Item Clinical Checklist for Detecting Decisional Conflict in Patients","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":161,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université du Québec à Trois-Rivières; Université Laval","funders":"","keywords":"Checklist; Medicine; Spearman's rank correlation coefficient; Population; Clinical trial; Gold standard (test); Internal consistency; Physical therapy; Clinical psychology; Internal medicine; Psychometrics; Psychology; Statistics","score_opus":0.3313651932412289,"score_gpt":0.5151991827952819,"score_spread":0.183833989554053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2107667787","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9804988,0.00018731573,0.013738318,0.0006655766,0.001645924,0.0026078438,0.000022365897,0.000059875514,0.00057396677],"genre_scores_gemma":[0.97498554,0.000031746127,0.022781966,0.0013222464,0.00019904715,0.00054535357,0.000044208562,0.000041173993,0.000048727747],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99230653,0.0015170558,0.0034671323,0.0005051427,0.0016036003,0.00060055294],"domain_scores_gemma":[0.96060157,0.035518613,0.0011673339,0.00088066136,0.001471029,0.000360772],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0038767625,0.00020723527,0.00064914464,0.0003395604,0.00045803515,0.000017104005,0.0008031713,0.0005593089,0.0012110878],"category_scores_gemma":[0.075508304,0.00017838462,0.00018500519,0.00049844163,0.00012782175,0.00025115954,0.0005451408,0.0012004568,0.0001671605],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002751835,0.0001606166,0.71313363,0.00013508678,0.000010107204,0.0000010008152,0.0006448715,0.000010751914,0.0000054540005,0.00020840584,0.0028629722,0.28255188],"study_design_scores_gemma":[0.006324146,0.0003944325,0.95050645,0.004224048,0.000014593396,0.0000012888553,0.0011171536,0.022332016,0.000020348138,0.007980114,0.006805692,0.00027973027],"about_ca_topic_score_codex":0.00019149008,"about_ca_topic_score_gemma":0.00010525064,"teacher_disagreement_score":0.28227216,"about_ca_system_score_codex":0.00023240695,"about_ca_system_score_gemma":0.00056795677,"threshold_uncertainty_score":0.9997019},"labels":[],"label_agreement":null},{"id":"W2109622137","doi":"10.1177/0272989x12443311","title":"Decision Coaching to Prepare Patients for Making Health Decisions","year":2012,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":162,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre hospitalier universitaire de Québec; University of Saskatchewan; Ottawa Hospital; University of Ottawa","funders":"","keywords":"Coaching; Clinical decision making; Medical decision making; Decision analysis; Psychology; Medicine; Management science; Process management; Business; Medical emergency; Intensive care medicine; Psychotherapist; Engineering; Economics","score_opus":0.47208703101640753,"score_gpt":0.5822630102371843,"score_spread":0.11017597922077677,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2109622137","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000017507657,0.8422899,0.13454247,0.00045209576,0.008806045,0.012206674,0.0005566074,0.00036075723,0.0007679044],"genre_scores_gemma":[0.0011071602,0.8462601,0.1371455,0.0075279656,0.0015201364,0.005100774,0.000736978,0.00047540403,0.00012602752],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.97925663,0.0043732794,0.0072996286,0.0019123557,0.004286738,0.0028713504],"domain_scores_gemma":[0.9278329,0.061325986,0.0034852833,0.0038751853,0.0012043407,0.0022763168],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.006697687,0.0012562386,0.0044099498,0.0015650064,0.004570663,0.00009581305,0.003943602,0.0022163745,0.0010616733],"category_scores_gemma":[0.06506738,0.0010078459,0.0011215389,0.0017298601,0.00010436828,0.0003696736,0.004445672,0.0036420997,0.0015636316],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013979492,0.00018931123,0.0003930315,0.008000688,0.000050479703,0.0000023228527,0.0019921914,0.0000037787718,2.8396436e-9,0.0004917367,0.044506047,0.9442306],"study_design_scores_gemma":[0.0008437033,0.00021494651,0.00010935679,0.21976371,0.00016437501,0.0000073471397,0.00043516615,0.00018195903,3.659826e-9,0.0012238022,0.77641076,0.0006448925],"about_ca_topic_score_codex":0.000046789322,"about_ca_topic_score_gemma":0.00055144454,"teacher_disagreement_score":0.9435857,"about_ca_system_score_codex":0.0032936085,"about_ca_system_score_gemma":0.0046949526,"threshold_uncertainty_score":0.99985147},"labels":[],"label_agreement":null},{"id":"W2111227431","doi":"10.1177/0272989x05280558","title":"Preference-Based Antithrombotic Therapy in Atrial Fibrillation: Implications for Clinical Decision Making","year":2005,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":74,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Canadian Institutes of Health Research; McMaster University; Institute for Clinical Evaluative Sciences; Élisabeth Bruyère Hospital; St. Thomas Hospital; Ottawa Hospital","funders":"Agency for Healthcare Research and Quality","keywords":"Warfarin; Medicine; Atrial fibrillation; Antithrombotic; Aspirin; Stroke (engine); MEDLINE; Intensive care medicine; Medical prescription; Clinical Practice; Physical therapy; Internal medicine","score_opus":0.2659753272242949,"score_gpt":0.4873222501553039,"score_spread":0.221346922931009,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111227431","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5348606,0.0009125342,0.44967452,0.011029099,0.0010349009,0.0017499393,0.000007178549,0.00015458345,0.00057663425],"genre_scores_gemma":[0.9011865,0.00055095495,0.09233282,0.0013284226,0.004479108,0.0000067449173,0.000025215586,0.00003521243,0.00005500921],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99640775,0.00012189394,0.0015191187,0.000606858,0.0009549455,0.00038942578],"domain_scores_gemma":[0.9923489,0.0064634057,0.00028745545,0.00052112906,0.00018255426,0.00019656186],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0022855247,0.00023751985,0.0006832168,0.00040383864,0.00018076868,0.00007440888,0.00025958556,0.00035842057,0.0010731919],"category_scores_gemma":[0.007077635,0.00018788585,0.000582105,0.00061333773,0.00009475091,0.00017208015,0.00010180269,0.00027145818,0.000087037544],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012594605,0.00001444018,0.32844427,0.000011550647,0.00002348731,0.0000031491381,0.000015047073,0.0010087321,0.000005416328,0.0007220341,0.0011323072,0.6673601],"study_design_scores_gemma":[0.009079036,0.00042318538,0.8304365,0.0011074453,0.00006737416,0.000009734951,0.000013273963,0.06392459,0.000005479346,0.006290714,0.08839217,0.00025044815],"about_ca_topic_score_codex":0.0000014974576,"about_ca_topic_score_gemma":0.00003011468,"teacher_disagreement_score":0.66710967,"about_ca_system_score_codex":0.00012595802,"about_ca_system_score_gemma":0.000280925,"threshold_uncertainty_score":0.99983996},"labels":[],"label_agreement":null},{"id":"W2113500562","doi":"10.1177/0272989x07312721","title":"Modeling the Logistics of Response to Anthrax Bioterrorism","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Bacillus and Francisella bacterial research","field":"Biochemistry, Genetics and Molecular Biology","cited_by":72,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"U.S. Public Health Service","keywords":"Software deployment; Medicine; Medical emergency; Risk analysis (engineering); Business; Computer security; Environmental health; Intensive care medicine; Computer science","score_opus":0.04295761122448166,"score_gpt":0.3497121022199391,"score_spread":0.30675449099545743,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113500562","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6333094,0.00016373541,0.36570784,0.00042504052,0.00017834695,0.00007410413,0.000005481324,0.0000034833959,0.00013252432],"genre_scores_gemma":[0.99644434,0.00021336944,0.0025892009,0.00042541418,0.00022622911,0.0000045789125,0.0000031026384,0.000012520302,0.00008124079],"study_design_codex":"bench_or_experimental","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9984593,0.00012285143,0.00029480873,0.00023568589,0.00065617816,0.00023121972],"domain_scores_gemma":[0.9991115,0.00018824618,0.00003226393,0.0003935792,0.0001311673,0.00014329144],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012722102,0.0000919164,0.0001431174,0.000051736617,0.00014136531,0.000009986918,0.00041968666,0.00017841879,0.00013571317],"category_scores_gemma":[0.0055181733,0.000058848702,0.00007289226,0.00015528739,0.00016529328,0.0000014967484,0.000321639,0.00014404501,0.00002255981],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007142194,0.00014686488,0.0004999187,0.000018657633,0.00004206872,0.00015494956,0.00042340427,0.0056213485,0.6947864,0.00014052185,0.0055559184,0.2854677],"study_design_scores_gemma":[0.0071567576,0.006159447,0.019076936,0.0016032849,0.000069695845,0.0010173857,0.0007957885,0.50319415,0.14779644,0.0043446305,0.30665132,0.0021341443],"about_ca_topic_score_codex":0.000011753252,"about_ca_topic_score_gemma":0.000011130044,"teacher_disagreement_score":0.54699004,"about_ca_system_score_codex":0.000009817155,"about_ca_system_score_gemma":0.0001510634,"threshold_uncertainty_score":0.66061676},"labels":[],"label_agreement":null},{"id":"W2113807678","doi":"10.1177/0272989x08327067","title":"Decisional Conflict in Patients and Their Physicians: A Dyadic Approach to Shared Decision Making","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":197,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Université Laval; Hôpital Saint-François d'Assise; Centre hospitalier universitaire de Québec","funders":"","keywords":"Medical decision making; Psychology; Group decision-making; Management science; Social psychology; Medicine; Family medicine; Economics","score_opus":0.15019996944445355,"score_gpt":0.4396322377587697,"score_spread":0.2894322683143161,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113807678","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88406247,0.0023871765,0.10091701,0.0026470947,0.0013253903,0.0034288692,0.00007221394,0.00024044109,0.0049193096],"genre_scores_gemma":[0.9380649,0.00007602953,0.04271086,0.018672682,0.0001787235,0.00018700048,0.000041356892,0.00004784576,0.000020644848],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9929906,0.0009714066,0.0019770814,0.0009876498,0.0020541886,0.0010190379],"domain_scores_gemma":[0.9896546,0.0076971552,0.000461617,0.0011852433,0.0004653523,0.0005360512],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0015930473,0.00042370538,0.00078845076,0.0007346838,0.00090925803,0.00005871914,0.001280769,0.000589566,0.00035807487],"category_scores_gemma":[0.011585997,0.0003416577,0.00011921289,0.0011836968,0.00009689494,0.0003338447,0.0009990747,0.0016549156,0.0001985052],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00078653276,0.0003934507,0.043959867,0.000051096325,0.00000835016,0.000007008879,0.009091759,0.000061864994,0.000006828887,0.0009958558,0.0066524916,0.9379849],"study_design_scores_gemma":[0.0048872163,0.00048750188,0.90519714,0.014813149,0.000010891094,0.000006565865,0.0038435191,0.029027881,0.0000018319809,0.025012344,0.016049964,0.00066197535],"about_ca_topic_score_codex":0.000023523346,"about_ca_topic_score_gemma":0.000059038415,"teacher_disagreement_score":0.9373229,"about_ca_system_score_codex":0.00047725695,"about_ca_system_score_gemma":0.00034687127,"threshold_uncertainty_score":0.99990356},"labels":[],"label_agreement":null},{"id":"W2115767227","doi":"10.1177/0272989x11424401","title":"Patient and Public Involvement in Clinical Practice Guidelines","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":164,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Canadian Arthritis Patient Alliance; Ministère de la Santé et des Services Sociaux (Québec)","funders":"Canadian Institutes of Health Research; Centre Hospitalier Universitaire de Québec","keywords":"Data extraction; Inclusion (mineral); Public health; Medicine; Variety (cybernetics); Family medicine; Focus group; MEDLINE; Medical education; Nursing; Psychology; Political science","score_opus":0.588202314041336,"score_gpt":0.5769600471350957,"score_spread":0.011242266906240306,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115767227","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9634416,0.0013108726,0.0069223926,0.0034893653,0.0042149043,0.0014837505,0.000007109063,0.00007362651,0.019056357],"genre_scores_gemma":[0.83145225,0.0011492432,0.07290093,0.09383241,0.00037867832,0.0001884578,0.000019634095,0.00002657317,0.000051851745],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9935275,0.0010749124,0.0028618614,0.00046380688,0.0013870399,0.0006849054],"domain_scores_gemma":[0.99479187,0.003203605,0.00048402307,0.0003503797,0.00044810912,0.00072204485],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.008263669,0.00016797654,0.00034930214,0.00018753287,0.00035448195,0.000009088938,0.00022124419,0.00032918493,0.0025928363],"category_scores_gemma":[0.014727552,0.00012595361,0.0000525075,0.00026630537,0.00008832599,0.00020879583,0.0005591008,0.00093027845,0.00029766784],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024528013,0.00043116463,0.24252404,0.00013261041,0.0000075429703,0.00008203962,0.002304233,9.5809376e-8,2.3704614e-7,0.002347507,0.010420438,0.7415048],"study_design_scores_gemma":[0.005326699,0.0019852116,0.11302285,0.010771079,0.000029652854,0.000016447988,0.030632451,0.0035953312,0.000002814513,0.015329878,0.81873584,0.0005517417],"about_ca_topic_score_codex":0.00033628204,"about_ca_topic_score_gemma":0.0008470498,"teacher_disagreement_score":0.8083154,"about_ca_system_score_codex":0.00014659629,"about_ca_system_score_gemma":0.00046856853,"threshold_uncertainty_score":0.9983189},"labels":[],"label_agreement":null},{"id":"W2115838114","doi":"10.1177/0272989x12451056","title":"Use of Natural Health Products","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Complementary and Alternative Medicine Studies","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Guelph; University of Toronto","funders":"","keywords":"Laddering; Naturalness; Context (archaeology); Product (mathematics); Situational ethics; Psychology; Quality (philosophy); Marketing; Perception; Applied psychology; Social psychology; Medicine; Business","score_opus":0.14763258900715653,"score_gpt":0.4414239782614098,"score_spread":0.29379138925425324,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2115838114","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9837625,0.0040636747,0.005275048,0.004576507,0.0017000855,0.00030412688,0.000003866573,0.000035306017,0.00027890078],"genre_scores_gemma":[0.9856487,0.00013855624,0.011075651,0.0020642087,0.00090628926,0.0000037796508,0.00000769073,0.000009729362,0.00014538501],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9979027,0.00006052169,0.00044404902,0.00016818247,0.001100259,0.00032428524],"domain_scores_gemma":[0.9986603,0.0006686431,0.00012264238,0.00021424466,0.00011046083,0.00022367729],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00080297206,0.00011168125,0.0003845502,0.000116257266,0.00007348453,0.000003800061,0.00008556116,0.000010693298,0.00072022376],"category_scores_gemma":[0.0049731806,0.00007251414,0.00006070803,0.0002411173,0.00012016518,0.00011573728,0.00015903683,0.00024475754,0.000017236438],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015950085,0.00020308293,0.18738478,0.00010857936,0.000055387758,0.000023410703,0.00048016154,2.6161766e-7,0.00014737416,0.0006271404,0.013321106,0.7974892],"study_design_scores_gemma":[0.0018015645,0.0007233759,0.8153735,0.0034291353,0.00005753971,0.0002415227,0.0003593956,0.00041934484,0.0007162927,0.00022531663,0.1764773,0.00017575154],"about_ca_topic_score_codex":0.000022500815,"about_ca_topic_score_gemma":0.0000076028286,"teacher_disagreement_score":0.79731345,"about_ca_system_score_codex":0.000046241643,"about_ca_system_score_gemma":0.00008579864,"threshold_uncertainty_score":0.7885945},"labels":[],"label_agreement":null},{"id":"W2116396363","doi":"10.1177/0272989x10393284","title":"Calculating Utility Decrements Associated With an Adverse Event","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":27,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Programs for Assessment of Technology in Health Research Institute","funders":"","keywords":"Tobit model; Econometrics; Statistics; Mathematics; Event (particle physics)","score_opus":0.4643728791633878,"score_gpt":0.46888997858030834,"score_spread":0.0045170994169205225,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2116396363","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8280078,0.00014700582,0.1630215,0.000640206,0.0005030021,0.00041270338,0.000044157037,0.00009436931,0.007129211],"genre_scores_gemma":[0.9793615,0.000006722716,0.015157226,0.005216727,0.000119063414,0.000031788804,0.000018689649,0.000031027208,0.000057309517],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9946009,0.00039996495,0.0033855718,0.00065779116,0.00047243093,0.00048335234],"domain_scores_gemma":[0.9960495,0.0012455889,0.0015706138,0.00056657597,0.00009104818,0.00047665124],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.024566887,0.00019547243,0.0007294117,0.0002150362,0.00025971112,0.000032108575,0.00042436543,0.00023726832,0.007217523],"category_scores_gemma":[0.02267436,0.00020026215,0.00010608426,0.0002545081,0.0000966536,0.00044876747,0.000110973255,0.00028465534,0.0007509653],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023501166,0.0009930115,0.9158856,0.00010817354,0.00021707607,0.00006947729,0.0050380654,0.00010837518,6.359954e-7,0.030898487,0.003965691,0.042480346],"study_design_scores_gemma":[0.0021039173,0.00033516597,0.8113577,0.0011685033,0.000016154981,0.0000120454015,0.0012794968,0.13951737,0.0000025484724,0.0404376,0.0032097267,0.0005598087],"about_ca_topic_score_codex":0.00026949908,"about_ca_topic_score_gemma":0.0004922828,"teacher_disagreement_score":0.15135361,"about_ca_system_score_codex":0.00039311647,"about_ca_system_score_gemma":0.00017838186,"threshold_uncertainty_score":0.99369},"labels":[],"label_agreement":null},{"id":"W2118322498","doi":"10.1177/027298902237706","title":"Measurement of Health Preferences among Patients with Tuberculous Infection and Disease","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University","funders":"","keywords":"Disease; Medicine; Internal medicine","score_opus":0.2843799878558878,"score_gpt":0.396803924049546,"score_spread":0.1124239361936582,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2118322498","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9597527,0.0026470989,0.03242733,0.00389636,0.00028850458,0.0004595937,0.000024604777,0.000026152376,0.0004776795],"genre_scores_gemma":[0.996937,0.00019098297,0.0008639638,0.0019043777,0.000052876454,0.000024857378,0.0000033479,0.0000126084105,0.000009992844],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996352,0.00024103277,0.002294309,0.00037059493,0.0005118193,0.00023020252],"domain_scores_gemma":[0.9976438,0.0004052621,0.0012306449,0.00024871057,0.00009766352,0.00037389595],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.010137016,0.00011980015,0.00057423464,0.00021160793,0.00014729968,0.000034530065,0.00011367412,0.000074997166,0.00079038093],"category_scores_gemma":[0.009841131,0.00011250626,0.00004245267,0.00015084165,0.00009456457,0.00023180337,0.000048402257,0.00012600108,0.00008494848],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00001461061,0.00013820636,0.9452182,0.00018769242,0.000019517716,4.4660084e-7,0.00031192886,0.00004035969,9.79229e-9,0.0012476507,0.0027727212,0.050048623],"study_design_scores_gemma":[0.0007059764,0.00014225552,0.98279685,0.0009864932,0.000003869043,5.110886e-7,0.000030825653,0.007859937,6.239193e-8,0.005754655,0.0015928145,0.00012576468],"about_ca_topic_score_codex":0.000286707,"about_ca_topic_score_gemma":0.00019674141,"teacher_disagreement_score":0.049922857,"about_ca_system_score_codex":0.0002322655,"about_ca_system_score_gemma":0.00008635139,"threshold_uncertainty_score":0.9984994},"labels":[],"label_agreement":null},{"id":"W2120517142","doi":"10.1177/0272989x15594369","title":"A Practical Application of Value of Information and Prospective Payback of Research to Prioritize Evaluative Research","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vancouver Coastal Health Research Institute; Vancouver Coastal Health","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Prioritization; Value of information; Computer science; Management science; Value (mathematics); Risk analysis (engineering); Decision aids; Operations research; Data science; Actuarial science; Medicine; Artificial intelligence; Business; Economics; Machine learning; Alternative medicine","score_opus":0.6234322006084808,"score_gpt":0.631423106902669,"score_spread":0.007990906294188194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2120517142","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5585464,0.0005490276,0.4145874,0.017245494,0.00022528236,0.0027892506,0.00009334499,0.00001650291,0.0059472816],"genre_scores_gemma":[0.94923127,0.000035592122,0.050170586,0.0003210907,0.00009072245,0.00012113842,0.000004644911,0.000010461313,0.0000144699825],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.992567,0.00122181,0.0039022297,0.0003697347,0.0016037977,0.0003354087],"domain_scores_gemma":[0.9854315,0.010485878,0.0012855077,0.00047411394,0.0019829266,0.00034007087],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12841369,0.00009205573,0.00072594866,0.0009830865,0.00009020887,0.00002914834,0.00029907125,0.00020705283,0.000100818856],"category_scores_gemma":[0.19330172,0.000097675,0.000046306122,0.0010321685,0.00038844283,0.0005040547,0.0003078181,0.0004273108,0.00026066092],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00060421927,0.00031944556,0.064432785,0.0008126282,0.0000508458,0.00000111136,0.017489137,0.00016892278,0.000020613586,0.8648694,0.014909003,0.0363219],"study_design_scores_gemma":[0.003182178,0.0017651032,0.15200353,0.0018091241,0.000010685418,0.00001658712,0.015592337,0.12503852,0.0002303845,0.6865722,0.013461938,0.00031742855],"about_ca_topic_score_codex":0.00040821126,"about_ca_topic_score_gemma":0.000035507746,"teacher_disagreement_score":0.39068487,"about_ca_system_score_codex":0.00045062933,"about_ca_system_score_gemma":0.00079320156,"threshold_uncertainty_score":0.8974815},"labels":[],"label_agreement":null},{"id":"W2120962896","doi":"10.1177/0272989x12458457","title":"The Impact of Using Informative Priors in a Bayesian Cost-Effectiveness Analysis","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Programs for Assessment of Technology in Health Research Institute","funders":"","keywords":"Prior probability; Bayesian probability; Bootstrapping (finance); Frequentist inference; Nonparametric statistics; Computer science; Econometrics; Statistics; Machine learning; Bayesian inference; Artificial intelligence; Mathematics","score_opus":0.33291262287449563,"score_gpt":0.5373039471839247,"score_spread":0.20439132430942908,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2120962896","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7522225,0.0008071951,0.24527493,0.00026896546,0.0002804797,0.00046908017,0.000021726364,0.000008974282,0.0006461644],"genre_scores_gemma":[0.99749297,0.000039244405,0.0020066525,0.0003205712,0.00008699787,0.000033921347,0.0000035422777,0.000011565748,0.000004545794],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9952278,0.00055476394,0.0032808979,0.0002081965,0.0002664015,0.0004619407],"domain_scores_gemma":[0.99056613,0.0072135977,0.0015610674,0.0003799236,0.000056880563,0.00022239973],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.041807387,0.00014006432,0.00092500093,0.00068301323,0.000162711,0.00004620099,0.0003204437,0.00016008526,0.00047068816],"category_scores_gemma":[0.017913686,0.00011345337,0.00029010762,0.0010678754,0.000098373515,0.0005163647,0.000104416766,0.00023430411,0.00010288967],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000065302826,0.00008906447,0.9641504,0.000053600193,0.00026029535,9.383808e-7,0.0027262298,0.00721839,4.616029e-7,0.009791446,0.00021784399,0.015425999],"study_design_scores_gemma":[0.00048022196,0.000027710988,0.6961164,0.00023917112,0.000014636902,0.0000034724765,0.00060532434,0.297242,9.1101504e-7,0.004632624,0.0005011902,0.0001363315],"about_ca_topic_score_codex":0.00067887356,"about_ca_topic_score_gemma":0.00013186096,"teacher_disagreement_score":0.2900236,"about_ca_system_score_codex":0.0008966251,"about_ca_system_score_gemma":0.00019090303,"threshold_uncertainty_score":0.9903588},"labels":[],"label_agreement":null},{"id":"W2121161686","doi":"10.1177/0272989x14563082","title":"The Need for Transparency and Efficiency in Reimbursement Decisions Relating to Drugs for Rare Diseases","year":2014,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Complement system in diseases","field":"Immunology and Microbiology","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kingston General Hospital; Health Sciences Centre; Queen's University; University of Ottawa; Sunnybrook Health Science Centre","funders":"","keywords":"Transparency (behavior); Reimbursement; Actuarial science; External validity; Medicine; Psychology; Risk analysis (engineering); Social psychology; Business; Economics; Computer science; Health care","score_opus":0.031043579037847612,"score_gpt":0.3294900109060802,"score_spread":0.2984464318682326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2121161686","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.01928808,0.018915026,0.46487015,0.47742683,0.0070275106,0.010636441,0.0011922824,0.00022547363,0.00041821922],"genre_scores_gemma":[0.67787784,0.00026660855,0.0041866936,0.30652505,0.0023140605,0.005207765,0.001520139,0.0002428265,0.0018590019],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9963545,0.00035035546,0.0012659848,0.0008247665,0.0004007859,0.0008036097],"domain_scores_gemma":[0.9724501,0.026432743,0.00027233377,0.00063158193,0.00012505817,0.0000882349],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0015703827,0.00038149103,0.0006896171,0.00030505107,0.00076954794,0.00008240463,0.00093059137,0.00064550946,0.0002880708],"category_scores_gemma":[0.012180442,0.00025945524,0.00028066456,0.0002102179,0.00016667414,0.000036643734,0.00025466978,0.0006973764,0.000036478326],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030828835,0.00004347978,0.0005150136,0.00010922126,0.00005133795,0.000015649732,0.00020331313,0.000015568092,0.000020407459,0.00022730124,0.81767726,0.18081316],"study_design_scores_gemma":[0.002532856,0.00028823953,0.0009500818,0.0047507877,0.00014355677,0.000015682359,0.00026767002,0.0004612699,0.0000062885147,0.00524546,0.9849485,0.00038956568],"about_ca_topic_score_codex":0.00001031902,"about_ca_topic_score_gemma":0.000042080945,"teacher_disagreement_score":0.6585898,"about_ca_system_score_codex":0.00010263599,"about_ca_system_score_gemma":0.00015650968,"threshold_uncertainty_score":0.99998575},"labels":[],"label_agreement":null},{"id":"W2121206416","doi":"10.1177/0272989x14520719","title":"Response to “The Life Table Method of Half-Cycle Correction: Getting It Right”","year":2014,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Table (database); Computer science; Statistics; Mathematics; Data mining","score_opus":0.23034348677672625,"score_gpt":0.4611532655172139,"score_spread":0.23080977874048766,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2121206416","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00015656448,0.00048129153,0.39494765,0.5966068,0.0043971003,0.0005007278,0.00005530679,0.000042675587,0.002811868],"genre_scores_gemma":[0.0025573897,0.000027265021,0.041976508,0.945396,0.006193739,0.00013756212,0.000031393603,0.0001057812,0.0035743483],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98637384,0.0028892392,0.007687276,0.0012320665,0.0009963823,0.00082120666],"domain_scores_gemma":[0.95737696,0.03605284,0.0043122857,0.0016088265,0.00022152583,0.00042758754],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.09605688,0.00043340618,0.0022471964,0.0008566522,0.00047652784,0.0001726698,0.0014758239,0.0012962833,0.007300444],"category_scores_gemma":[0.13896888,0.00040948685,0.00035426376,0.0006494366,0.00011301462,0.00019429528,0.0003568528,0.0019353268,0.0048387917],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000088415414,0.000021047872,0.00011082725,0.00016424849,0.000063808766,0.000020162914,0.00043445814,0.00060818146,2.5255042e-7,0.0006163779,0.99219435,0.0056778532],"study_design_scores_gemma":[0.00041548218,0.000094169336,0.0001760497,0.0013963195,0.000018134757,0.000057519777,0.00011673788,0.039243322,0.0000014440602,0.0048405607,0.9532252,0.00041505924],"about_ca_topic_score_codex":0.00040312592,"about_ca_topic_score_gemma":0.00008383421,"teacher_disagreement_score":0.35297114,"about_ca_system_score_codex":0.00052178797,"about_ca_system_score_gemma":0.00059133576,"threshold_uncertainty_score":0.99983567},"labels":[],"label_agreement":null},{"id":"W2122615848","doi":"10.1177/0272989x0102100305","title":"What Should Be Reported in a Methods Section on Utility Assessment?","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Cronbach's alpha; Computer science; Scale (ratio); Reliability (semiconductor); Section (typography); Operations research; Risk analysis (engineering); Medicine; Statistics; Psychometrics; Mathematics","score_opus":0.6577698216773527,"score_gpt":0.6160409168506167,"score_spread":0.04172890482673597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2122615848","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.35439807,0.0009710411,0.5498486,0.07238773,0.0068257665,0.0008011783,0.000014857771,0.00013282614,0.014619885],"genre_scores_gemma":[0.8850187,0.000790889,0.05592419,0.056810103,0.0008052978,0.00012493346,0.000024079858,0.000051322004,0.00045050593],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9920861,0.0009991681,0.0050736824,0.00088850746,0.0004800225,0.00047250462],"domain_scores_gemma":[0.9916062,0.0057727527,0.0015050803,0.00077483244,0.00006055773,0.00028059538],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.066547826,0.00020927608,0.0009815202,0.00063467195,0.00017236455,0.00019334277,0.0003377474,0.00042901744,0.0053241043],"category_scores_gemma":[0.030950326,0.00023699984,0.00015103545,0.00060775,0.00006741123,0.0007646309,0.00010876536,0.0006920969,0.0002755236],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002966644,0.0010394696,0.24736224,0.00021528438,0.00012771082,0.00024146498,0.0020610194,0.0005833888,0.000006822928,0.05183125,0.062392224,0.63384247],"study_design_scores_gemma":[0.0017422489,0.00017427721,0.28996348,0.0012960939,0.0000075224434,0.00008517497,0.0013634264,0.20338945,0.0000026967336,0.11016316,0.3912873,0.00052516954],"about_ca_topic_score_codex":0.00023119112,"about_ca_topic_score_gemma":0.0004097701,"teacher_disagreement_score":0.6333173,"about_ca_system_score_codex":0.0007888534,"about_ca_system_score_gemma":0.0002063231,"threshold_uncertainty_score":0.99558514},"labels":[],"label_agreement":null},{"id":"W2123466484","doi":"10.1177/027298902236943","title":"What Accounts for the Appeal of Complementary/Alternative Medicine, and What Makes Complementary/Alternative Medicine \"Alternative\"?","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Complementary and Alternative Medicine Studies","field":"Medicine","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Appeal; Alternative medicine; Medicine; Dimension (graph theory); Psychology; Mathematics; Pathology","score_opus":0.1512226850698555,"score_gpt":0.42547322137361376,"score_spread":0.2742505363037583,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2123466484","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5657834,0.07964141,0.1753804,0.14401229,0.021775218,0.009379509,0.00039031752,0.00022762339,0.0034098923],"genre_scores_gemma":[0.950591,0.03329597,0.001992684,0.009169324,0.004339306,0.00018311324,0.00012295794,0.00007860268,0.00022698683],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9920093,0.00027500596,0.0020120922,0.0011175327,0.0037014643,0.00088460604],"domain_scores_gemma":[0.9873767,0.009897677,0.00087367295,0.00067035336,0.0007006918,0.0004809213],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002451491,0.00077717204,0.0017396648,0.0005877395,0.0006913912,0.000119095756,0.0009951857,0.0000680045,0.011960833],"category_scores_gemma":[0.0023988385,0.0004595411,0.0002618745,0.00056181836,0.0025285766,0.0011331601,0.0008415978,0.0007156548,0.00001890283],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012525822,0.00039376935,0.009364993,0.00033722416,0.001642649,0.0002470914,0.015558986,0.000020362302,0.00013405975,0.0037667335,0.026315678,0.9409659],"study_design_scores_gemma":[0.07221507,0.017566863,0.02630706,0.056809634,0.0027687314,0.0009873586,0.28586406,0.1445079,0.00087006885,0.04852969,0.34148297,0.0020905964],"about_ca_topic_score_codex":0.0005191395,"about_ca_topic_score_gemma":0.00043045444,"teacher_disagreement_score":0.93887526,"about_ca_system_score_codex":0.00020914075,"about_ca_system_score_gemma":0.00004325494,"threshold_uncertainty_score":0.9997856},"labels":[],"label_agreement":null},{"id":"W2123512569","doi":"10.1177/0272989x14559729","title":"Psst, Have I Got a Model for You . . .","year":2014,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Simulation Techniques and Applications","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science","score_opus":0.11254584941191156,"score_gpt":0.49519890766808483,"score_spread":0.38265305825617324,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2123512569","genre_codex":"methods","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000014523245,0.0000755704,0.7357211,0.0005498552,0.2606371,0.00056192384,0.00021517205,0.00016300204,0.0020617172],"genre_scores_gemma":[0.01134942,0.00007097328,0.12925732,0.001251405,0.84241194,0.00065102,0.0002243236,0.00015898186,0.014624638],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98514,0.00015559432,0.0022134415,0.0015717761,0.010351154,0.0005679994],"domain_scores_gemma":[0.95314676,0.041320194,0.0010051331,0.0021508273,0.0018746143,0.00050249055],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.010452385,0.00046303342,0.0010492746,0.0005929755,0.00044783787,0.00059199886,0.0033631928,0.0020875337,0.0014081121],"category_scores_gemma":[0.11894402,0.0003378498,0.00063232327,0.0005566822,0.00018616243,0.00015427434,0.00068288454,0.001086452,0.00042134002],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004253352,0.000028641554,0.0000036867093,0.0000060780044,0.00000576203,0.0000016418147,0.000030239984,0.00068556145,7.0352377e-7,0.0014041149,0.67988217,0.31790885],"study_design_scores_gemma":[0.00026655136,0.000030311236,0.0000017644909,0.00016187069,0.000013968556,7.996253e-7,0.000008216287,0.37779796,0.000001317444,0.17497155,0.4465697,0.00017596882],"about_ca_topic_score_codex":0.000005673626,"about_ca_topic_score_gemma":0.000041265244,"teacher_disagreement_score":0.6064638,"about_ca_system_score_codex":0.00013663433,"about_ca_system_score_gemma":0.00067602325,"threshold_uncertainty_score":0.9999074},"labels":[],"label_agreement":null},{"id":"W2124095688","doi":"10.1177/0272989x14565820","title":"Value of Genetic Testing for Hereditary Colorectal Cancer in a Probability-Based US Online Sample","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"BRCA gene mutations in cancer","field":"Biochemistry, Genetics and Molecular Biology","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"National Cancer Institute","keywords":"Genetic testing; Hereditary Cancer; Colorectal cancer; Sample (material); Medicine; Value (mathematics); Statistics; Oncology; Computer science; Cancer; Internal medicine; Mathematics","score_opus":0.04497405348664674,"score_gpt":0.35821223145774767,"score_spread":0.31323817797110093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124095688","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8859131,0.00063373067,0.11248297,0.00015434298,0.00034453248,0.00031127833,0.00012746248,0.000007847099,0.000024742389],"genre_scores_gemma":[0.7542374,0.000004188809,0.24511102,0.00024379783,0.00025634648,0.000093431256,0.00003549556,0.000016353522,0.0000019399215],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9985256,0.00006698824,0.00042778338,0.0003383303,0.00043034216,0.00021094907],"domain_scores_gemma":[0.9988041,0.00052288006,0.000103977414,0.00020584311,0.00025354602,0.00010967993],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0006150212,0.000111624504,0.00018312527,0.0000657426,0.000028371014,0.0000069628845,0.00023948326,0.0001790086,0.00003973111],"category_scores_gemma":[0.009761175,0.00010241682,0.000058284233,0.00023587477,0.000107695276,0.000002389498,0.00010683689,0.000101780075,5.756062e-7],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010216926,0.000562957,0.24803479,0.00015142921,0.00004461902,0.00002336436,0.00013366836,0.13165477,0.018666208,0.00003367621,0.0022568442,0.597416],"study_design_scores_gemma":[0.010294682,0.0038530543,0.2618368,0.0018796758,0.00009561412,0.00006912914,0.00024921304,0.6461895,0.033853978,0.027459983,0.0132261105,0.0009922324],"about_ca_topic_score_codex":0.00011958637,"about_ca_topic_score_gemma":0.0009982962,"teacher_disagreement_score":0.59642375,"about_ca_system_score_codex":0.0001241796,"about_ca_system_score_gemma":0.0013191211,"threshold_uncertainty_score":0.99858004},"labels":[],"label_agreement":null},{"id":"W2124392780","doi":"10.1177/0272989x10395029","title":"Information for Decision Making by Patients With Early-Stage Prostate Cancer","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Queen's University","funders":"","keywords":"Developing country; Medicine; Psychology; Demography; Family medicine; Economic growth; Economics; Sociology","score_opus":0.12380627893514994,"score_gpt":0.4328348681376491,"score_spread":0.30902858920249915,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124392780","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8079497,0.00064889895,0.17957278,0.00057993515,0.0024062174,0.0056234933,0.00048142343,0.00032211276,0.002415447],"genre_scores_gemma":[0.95834523,0.0000846941,0.036423404,0.0035279214,0.000079146324,0.0013003208,0.00008990278,0.00005472892,0.00009467265],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99523765,0.0003510782,0.0015288702,0.0003461593,0.0017462375,0.00078998104],"domain_scores_gemma":[0.99337757,0.0034378653,0.0009124398,0.00086857483,0.0010632279,0.0003403463],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00088714703,0.00027956933,0.00039033563,0.00026455178,0.0012303469,0.000034720626,0.0008015016,0.00041944603,0.001986588],"category_scores_gemma":[0.0050878935,0.0002099451,0.00007549079,0.0004398284,0.000103341896,0.0009019978,0.00048866845,0.00089770445,0.00020597938],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019991577,0.000096155396,0.4121907,0.0002613693,0.00001935208,0.0000013294384,0.01602578,0.000008959591,6.2876705e-7,0.00046328406,0.012590334,0.55634296],"study_design_scores_gemma":[0.011884155,0.0013463036,0.6976509,0.02618531,0.000079320664,0.000001620303,0.008836614,0.008825213,0.000015746678,0.008901642,0.23500492,0.0012682348],"about_ca_topic_score_codex":0.00026243512,"about_ca_topic_score_gemma":0.00027051513,"teacher_disagreement_score":0.5550747,"about_ca_system_score_codex":0.00038887688,"about_ca_system_score_gemma":0.00053210073,"threshold_uncertainty_score":0.99892575},"labels":[],"label_agreement":null},{"id":"W2124968614","doi":"10.1177/0272989x0002000403","title":"Patient Preference-based Treatment Thresholds and Recommendations","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital","funders":"National Institute of Neurological Disorders and Stroke","keywords":"Aspirin; Stroke (engine); Medicine; Myocardial infarction; Minor stroke; Physical therapy; Internal medicine","score_opus":0.044109027906295915,"score_gpt":0.3231085061340767,"score_spread":0.2789994782277808,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2124968614","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92344946,0.00031320014,0.011537069,0.008375182,0.00018184417,0.00055373093,0.000009735299,0.00011255785,0.055467233],"genre_scores_gemma":[0.97712255,0.00030013348,0.01828637,0.0033011192,0.00008674076,0.000052956475,0.000026862392,0.00001615751,0.00080713356],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9984056,0.00003002709,0.00033572243,0.0003212416,0.00068913295,0.00021828376],"domain_scores_gemma":[0.9990743,0.00029915205,0.0000443208,0.0003147134,0.000030626943,0.00023692515],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00017097729,0.00014903724,0.00023370677,0.00010155621,0.00009288812,0.000023708426,0.00007444358,0.000102000035,0.01587356],"category_scores_gemma":[0.00022479071,0.000104804916,0.00006318375,0.00014493792,0.000064992855,0.00003704228,0.000054293338,0.00013656142,0.0001247025],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016249441,0.00017546663,0.0030212395,0.000008014493,0.000028473072,0.00008084899,0.00008991264,0.000007637409,0.000013177518,0.000038464124,0.023562865,0.9728114],"study_design_scores_gemma":[0.004551975,0.001432909,0.014546428,0.0015436113,0.00016563402,0.000119800796,0.00014352667,0.015616516,0.00029662458,0.00036395303,0.96096206,0.0002569489],"about_ca_topic_score_codex":0.000007746342,"about_ca_topic_score_gemma":0.000006594352,"teacher_disagreement_score":0.97255445,"about_ca_system_score_codex":0.00012534192,"about_ca_system_score_gemma":0.000076481854,"threshold_uncertainty_score":0.98502606},"labels":[],"label_agreement":null},{"id":"W2125389120","doi":"10.1177/0272989x04265483","title":"From Diagnostic Accuracy to Accurate Diagnosis: Interpreting a Test Result with Confidence","year":2004,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Statistics; Confidence interval; Logit; Logistic regression; Statement (logic); Transformation (genetics); Computer science; Sample (material); Positive predicative value; Test (biology); Econometrics; Diagnostic accuracy; Pre- and post-test probability; Mathematics; Data mining; Predictive value; Medicine","score_opus":0.202435274144168,"score_gpt":0.48219096307565,"score_spread":0.279755688931482,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2125389120","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4003288,0.00032294003,0.59241724,0.004393571,0.00093073753,0.00031086005,0.00002263647,0.00013367123,0.0011395396],"genre_scores_gemma":[0.85422707,0.000068110574,0.14127271,0.00391722,0.00037422165,0.000090709436,0.0000016776296,0.00002719168,0.000021074407],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9901219,0.000471482,0.0017766065,0.0014783348,0.0053952644,0.00075640064],"domain_scores_gemma":[0.19334711,0.8016982,0.0010122436,0.0019818149,0.00084029586,0.0011202998],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.010527752,0.0003686668,0.00074801204,0.0007365003,0.00036172275,0.0008680262,0.0029593573,0.00023513305,0.0019759838],"category_scores_gemma":[0.9380154,0.00022833073,0.0001443811,0.004300513,0.00021018158,0.00048518024,0.0010863113,0.00070323906,0.00088946125],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017117639,0.00011617738,0.054590635,0.000003207943,0.000015559572,0.00075029494,0.0010039174,0.0014587262,0.00004561034,0.0002918347,0.005796375,0.9357565],"study_design_scores_gemma":[0.0050507183,0.0019737156,0.36676586,0.020477114,0.00008363756,0.00035513646,0.006240654,0.007926827,0.00094866444,0.55587614,0.03244492,0.0018566188],"about_ca_topic_score_codex":0.0002591902,"about_ca_topic_score_gemma":0.00015297292,"teacher_disagreement_score":0.9338999,"about_ca_system_score_codex":0.00012506027,"about_ca_system_score_gemma":0.00029604963,"threshold_uncertainty_score":0.9998885},"labels":[],"label_agreement":null},{"id":"W2127429441","doi":"10.1177/0272989x0202200315","title":"“You’re Perfect for the Procedure! Why Don’t You Want It?” Elderly Arthritis Patients’ Unwillingness to Consider Total Joint Arthroplasty Surgery: A Qualitative Study","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":115,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Western Hospital; Women's College Hospital; Health Sciences Centre; Sunnybrook Health Science Centre; University of Toronto; SickKids Foundation; Hospital for Sick Children; St. Michael's Hospital","funders":"","keywords":"Candidacy; Medicine; Joint arthroplasty; Deliberation; Arthroplasty; Osteoarthritis; Physical therapy; Surgery; Alternative medicine","score_opus":0.26448530451583124,"score_gpt":0.46124874926034554,"score_spread":0.1967634447445143,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2127429441","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9238037,0.0028764484,0.017500514,0.040850073,0.0039958465,0.010304368,0.00018421204,0.00026115804,0.0002236746],"genre_scores_gemma":[0.9844984,0.00019609282,0.0018512326,0.009680369,0.00034932885,0.0031562939,0.000029407127,0.00009316456,0.00014573641],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.98936826,0.0034484412,0.0025592812,0.00096042024,0.0024813365,0.0011822762],"domain_scores_gemma":[0.9654036,0.030353948,0.00076666486,0.0014036661,0.001425574,0.00064652594],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0038997952,0.00047571235,0.0009832047,0.00032992425,0.00306366,0.000074458985,0.0006235305,0.0004103156,0.003094465],"category_scores_gemma":[0.0385593,0.00033252948,0.00028223457,0.0007849075,0.00024104168,0.0002358973,0.0007178119,0.001524791,0.00036193765],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00056836876,0.0012588596,0.03626136,0.0003838336,0.00013053397,0.000031808264,0.39168337,0.000059498023,0.000001659995,0.00026511066,0.19394173,0.37541386],"study_design_scores_gemma":[0.00874569,0.0034551441,0.029067168,0.012068242,0.00012274628,0.000025355348,0.84499216,0.012914658,0.000004239061,0.0031589302,0.084073566,0.001372092],"about_ca_topic_score_codex":0.00022614337,"about_ca_topic_score_gemma":0.0016229751,"teacher_disagreement_score":0.4533088,"about_ca_system_score_codex":0.00037478955,"about_ca_system_score_gemma":0.000610312,"threshold_uncertainty_score":0.9999127},"labels":[],"label_agreement":null},{"id":"W2128221807","doi":"10.1177/0272989x07302127","title":"Extrinsic Goals and Time Tradeoff","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Computer science","score_opus":0.27433173765214913,"score_gpt":0.4680319024412106,"score_spread":0.1937001647890615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2128221807","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7080038,0.00599831,0.23571312,0.023755657,0.001265499,0.00060110644,0.00004297849,0.00014777022,0.02447174],"genre_scores_gemma":[0.96468765,0.00012311085,0.014615807,0.019112097,0.00067321275,0.000011341572,0.0000054665634,0.000035796806,0.0007355046],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9951921,0.0001212849,0.0033856472,0.0005346725,0.00030584753,0.00046042335],"domain_scores_gemma":[0.9934338,0.004956941,0.0008012756,0.0003632966,0.000029928335,0.00041476972],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.034895636,0.00015701045,0.0007364578,0.0003660999,0.00017524953,0.00007442748,0.00028532092,0.00025821608,0.0041383435],"category_scores_gemma":[0.016437553,0.0001768727,0.00008559231,0.00022373786,0.00010133014,0.0002445952,0.00010613365,0.0002568808,0.0042993845],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013108118,0.00026259528,0.090448715,0.00027409158,0.00009713585,0.0001152861,0.0023246184,0.000025225567,0.000018094344,0.3275305,0.09958079,0.47919187],"study_design_scores_gemma":[0.0029314675,0.00022849256,0.3101925,0.0011639376,0.000012872736,0.00016527901,0.0006879569,0.030738436,0.0000112855005,0.23449652,0.41833815,0.0010330868],"about_ca_topic_score_codex":0.000043901317,"about_ca_topic_score_gemma":0.000038408027,"teacher_disagreement_score":0.47815877,"about_ca_system_score_codex":0.00019095305,"about_ca_system_score_gemma":0.00007266929,"threshold_uncertainty_score":0.996772},"labels":[],"label_agreement":null},{"id":"W2129285512","doi":"10.1177/0272989x03256883","title":"The Impact of Ignoring Population Heterogeneity when Markov Models are Used in Cost-Effectiveness Analysis","year":2003,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Markov model; Psychological intervention; Econometrics; Quality-adjusted life year; Population; Markov chain; Cost-effectiveness analysis; Cost effectiveness; Cost–benefit analysis; Actuarial science; Statistics; Medicine; Economics; Mathematics; Environmental health","score_opus":0.3765691349258113,"score_gpt":0.4999965294025286,"score_spread":0.1234273944767173,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2129285512","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.82100827,0.0006141577,0.17705226,0.00030115148,0.00019873313,0.00048653528,0.00003335834,0.000012826622,0.00029272502],"genre_scores_gemma":[0.9978692,0.000057419402,0.0017314168,0.00019807745,0.000042124924,0.00006898616,0.000008727618,0.000017552,0.000006507026],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99453133,0.0011112986,0.003162064,0.00046812487,0.0003264967,0.0004007114],"domain_scores_gemma":[0.9917877,0.0058370964,0.0015895955,0.000543262,0.000065556625,0.00017684024],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03843368,0.00016578076,0.0010772521,0.0005665439,0.00017943436,0.00006758278,0.00031900822,0.00019395903,0.00031223774],"category_scores_gemma":[0.01691886,0.00014680794,0.00035281543,0.00067519385,0.000048157057,0.0003100694,0.0000608392,0.00021668708,0.000037241927],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000038411596,0.0000626602,0.93268657,0.000038775,0.00015332235,0.00000208644,0.00018806242,0.056013037,6.6703717e-7,0.006745292,0.00013925832,0.003931865],"study_design_scores_gemma":[0.00053057796,0.000023847328,0.65116173,0.00019867261,0.000011804764,0.0000016347096,0.00010350323,0.29253855,0.0000018976319,0.055213377,0.000082586586,0.00013184318],"about_ca_topic_score_codex":0.001348645,"about_ca_topic_score_gemma":0.0017743368,"teacher_disagreement_score":0.28152487,"about_ca_system_score_codex":0.0009029896,"about_ca_system_score_gemma":0.00010481314,"threshold_uncertainty_score":0.99136204},"labels":[],"label_agreement":null},{"id":"W2130358353","doi":"10.1177/0272989x14528382","title":"Identifying Best-Fitting Inputs in Health-Economic Model Calibration","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"National Cancer Institute","keywords":"Calibration; Pareto principle; Set (abstract data type); Statistics; Measure (data warehouse); Goodness of fit; Computer science; Data set; Mathematics; Econometrics; Mathematical optimization; Data mining","score_opus":0.3501431477793351,"score_gpt":0.4898896560167661,"score_spread":0.139746508237431,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2130358353","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1475186,0.0007539284,0.8212623,0.026579285,0.0011107477,0.00041017926,0.000021916532,0.000071896706,0.002271127],"genre_scores_gemma":[0.94157225,0.00011859779,0.03293186,0.02457759,0.00053122616,0.00005353298,0.000014380732,0.000048126298,0.00015245158],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9915016,0.00048331142,0.006278848,0.000806007,0.0003125459,0.00061763613],"domain_scores_gemma":[0.99434555,0.0026630515,0.0020507686,0.0005463117,0.000024011048,0.0003703078],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03913379,0.00021922658,0.0010751837,0.0006635638,0.0002565928,0.00016817861,0.0004969364,0.00028092318,0.00084114185],"category_scores_gemma":[0.014707318,0.00027827354,0.00012374605,0.00021769154,0.000058549605,0.00069489627,0.00017731632,0.00042267732,0.0017096964],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007806444,0.00031797716,0.08795909,0.0011683616,0.000054446373,0.000011304094,0.0062729474,0.12826392,0.0000058430574,0.60596454,0.02899355,0.14090995],"study_design_scores_gemma":[0.0006700591,0.000036948717,0.0020149276,0.0008725489,0.0000010373341,0.000005760197,0.00012864359,0.8783966,0.0000010584116,0.11241114,0.00522887,0.00023245368],"about_ca_topic_score_codex":0.00044125973,"about_ca_topic_score_gemma":0.00064948865,"teacher_disagreement_score":0.7940536,"about_ca_system_score_codex":0.0009860791,"about_ca_system_score_gemma":0.0003414845,"threshold_uncertainty_score":0.9999669},"labels":[],"label_agreement":null},{"id":"W2132708603","doi":"10.1177/0272989x12455463","title":"State-Transition Modeling","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":342,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Computer science; Microsimulation; Markov model; Decision model; Psychological intervention; Set (abstract data type); Decision analysis; Risk analysis (engineering); State (computer science); Management science; Markov chain; Monte Carlo method; Medicine; Machine learning; Engineering; Mathematics; Algorithm; Statistics","score_opus":0.3828859167539311,"score_gpt":0.4790400401415357,"score_spread":0.0961541233876046,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2132708603","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.26051185,0.0017964843,0.7267696,0.0068135024,0.0011904507,0.00019471227,0.000027517694,0.000068440764,0.0026274107],"genre_scores_gemma":[0.9686507,0.00008853419,0.014177921,0.01620857,0.0007451371,0.000026776404,0.000010432509,0.000030019928,0.000061927174],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9953951,0.00019217037,0.0031485616,0.00036200794,0.0003441569,0.000558001],"domain_scores_gemma":[0.99727094,0.0013154554,0.0005903791,0.00035971036,0.00004326222,0.00042022901],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.024223924,0.00014987602,0.00060020987,0.00027821807,0.0001922156,0.000060002225,0.00026772535,0.0001894967,0.0033602868],"category_scores_gemma":[0.0084735025,0.0001728398,0.00011533501,0.000189761,0.000037706355,0.00064748595,0.000065225606,0.0002684778,0.005041449],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029271992,0.0013988689,0.092392154,0.0010863177,0.00030249436,0.00003724596,0.030565869,0.046027817,0.000013283499,0.47631714,0.1314017,0.22016439],"study_design_scores_gemma":[0.0008261465,0.000040492367,0.004145725,0.00053756,0.0000053568015,0.000025544277,0.00056754384,0.85184896,0.0000018134999,0.11181221,0.029742131,0.0004464864],"about_ca_topic_score_codex":0.00008076323,"about_ca_topic_score_gemma":0.000019443436,"teacher_disagreement_score":0.8058212,"about_ca_system_score_codex":0.00028950794,"about_ca_system_score_gemma":0.00007432081,"threshold_uncertainty_score":0.9998785},"labels":[],"label_agreement":null},{"id":"W2133539059","doi":"10.1177/0272989x14526642","title":"The Fallacy of Interpreting Deaths and Driving Distances","year":2014,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Traffic and Road Safety","field":"Engineering","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research","keywords":"Casual; Notice; Demography; Transport engineering; Geography; Actuarial science; Engineering; Political science; Business; Law; Sociology","score_opus":0.002295697382072704,"score_gpt":0.2491472852735786,"score_spread":0.2468515878915059,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2133539059","genre_codex":"editorial","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021180504,0.0044387574,0.21369678,0.000042980177,0.77598965,0.000104321836,0.0000075886305,0.00018360163,0.003418305],"genre_scores_gemma":[0.6672058,0.0030857828,0.0021355115,0.000014562637,0.3273912,0.000009519629,0.0000090622225,0.00008276255,0.00006575785],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99733,0.00006323516,0.00064226345,0.00026135883,0.0014066955,0.00029642152],"domain_scores_gemma":[0.989356,0.010036116,0.00012662805,0.000295791,0.000065593886,0.000119881435],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001349658,0.00024190398,0.00043765575,0.00006337197,0.00017089337,0.00007479956,0.00055867626,0.0005963719,0.000040505973],"category_scores_gemma":[0.006755043,0.0001518875,0.000101229314,0.00009733635,0.00016972613,0.00004141529,0.0002363922,0.0009363644,0.0000067363226],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000014638617,0.000002802777,0.00006705413,0.00007347982,0.000025898069,0.0000073211745,0.000112039525,0.00015829252,0.000001170355,0.00012738998,0.45810592,0.541304],"study_design_scores_gemma":[0.00019911236,0.000024039513,0.00017978117,0.0040138885,0.000022432478,0.0000029185667,0.000057800033,0.034184653,0.0000018522916,0.00065845245,0.96045816,0.00019691589],"about_ca_topic_score_codex":0.0000021514445,"about_ca_topic_score_gemma":0.000092663526,"teacher_disagreement_score":0.66508776,"about_ca_system_score_codex":0.00004761413,"about_ca_system_score_gemma":0.000056970322,"threshold_uncertainty_score":0.80869055},"labels":[],"label_agreement":null},{"id":"W2134023402","doi":"10.1177/0272989x12454579","title":"Model Transparency and Validation","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":791,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Transparency (behavior); Computer science; Face validity; External validity; Rigour; Documentation; Management science; Health care; Medicine; Psychology; Engineering; Psychometrics","score_opus":0.49166679562981086,"score_gpt":0.48999380860289565,"score_spread":0.001672987026915207,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134023402","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.33450472,0.0026508481,0.64833814,0.008311957,0.00066125457,0.000221618,0.000031637013,0.000049697654,0.005230145],"genre_scores_gemma":[0.9712109,0.00015196203,0.021663977,0.0064815124,0.00033867697,0.000024463581,0.0000072100147,0.00001841235,0.000102908234],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9967004,0.00011565579,0.0022838283,0.0003127479,0.00023274262,0.00035462668],"domain_scores_gemma":[0.99759007,0.0012353461,0.0005491584,0.00026575063,0.00002324879,0.0003364405],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0164631,0.000117692616,0.00047201384,0.00019227692,0.0001476919,0.000046478337,0.00017979914,0.00017964214,0.0014618835],"category_scores_gemma":[0.007485583,0.00013247365,0.00005758966,0.00011837681,0.000049164428,0.00053480157,0.000048805472,0.00016695278,0.00093043543],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000429458,0.00028143683,0.25992423,0.00033193553,0.00005461698,0.0000028222287,0.0063283537,0.0010125012,0.0000061027467,0.6050491,0.039547373,0.08741859],"study_design_scores_gemma":[0.0014963726,0.00006328751,0.07184599,0.0006243041,0.000014160174,0.00003426859,0.00041845217,0.6032331,0.000007254199,0.2783571,0.043221682,0.0006840382],"about_ca_topic_score_codex":0.000019941335,"about_ca_topic_score_gemma":0.000009213353,"teacher_disagreement_score":0.6367062,"about_ca_system_score_codex":0.00012376816,"about_ca_system_score_gemma":0.00005387572,"threshold_uncertainty_score":0.9998475},"labels":[],"label_agreement":null},{"id":"W2134799417","doi":"10.1177/0272989x11421529","title":"Systematic Review and Empirical Comparison of Contemporaneous EQ-5D and SF-6D Group Mean Scores","year":2011,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":69,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Comparability; EQ-5D; Statistics; Preference; Psychology; SF-36; Medicine; Econometrics; Demography; Health related quality of life; Mathematics","score_opus":0.698916839044423,"score_gpt":0.529984587193883,"score_spread":0.16893225185054006,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2134799417","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00004772639,0.99142194,0.0041054436,0.0005688434,0.00036762972,0.0027540554,0.0001193226,0.000037490005,0.0005775688],"genre_scores_gemma":[0.0010484934,0.9936222,0.0017386109,0.003102338,0.00014103587,0.00018916359,0.000030573738,0.0000707283,0.00005686617],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98310864,0.0014182818,0.013596027,0.0010259025,0.00041884938,0.00043231263],"domain_scores_gemma":[0.981037,0.009808786,0.0076626255,0.00091476546,0.000086305576,0.00049051014],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.030041829,0.00054766773,0.011618824,0.00054721144,0.00015397284,0.00007982304,0.0006921939,0.0006898343,0.0006300671],"category_scores_gemma":[0.031146722,0.00048374388,0.00044714622,0.00030303124,0.00027872645,0.00016376011,0.0003450305,0.00065402134,0.0003970031],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000005236515,0.000085368236,0.0003588131,0.9208395,0.00022241121,0.00001268977,0.00036251041,5.6608535e-8,9.158463e-10,0.008595783,0.007144682,0.062372964],"study_design_scores_gemma":[0.00035199663,0.00014449889,0.000049807484,0.7764281,0.00065973203,0.00022911208,0.00011162774,0.00042177158,4.4564668e-9,0.0028602227,0.21818635,0.00055676914],"about_ca_topic_score_codex":0.000081388935,"about_ca_topic_score_gemma":0.00008370826,"teacher_disagreement_score":0.21104166,"about_ca_system_score_codex":0.0002200122,"about_ca_system_score_gemma":0.0002305783,"threshold_uncertainty_score":0.9997614},"labels":[],"label_agreement":null},{"id":"W2135388962","doi":"10.1177/0272989x07306781","title":"Can Computerized Decision Support Help Patients Make Complex Treatment Decisions? A Randomized Controlled Trial of an Individualized Menopause Decision Aid","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Agency for Healthcare Research and Quality","keywords":"Coaching; Randomized controlled trial; Medicine; Physical therapy; Decision aids; Patient satisfaction; Intervention (counseling); Menopause; Patient education; Family medicine; Nursing; Psychology; Alternative medicine; Internal medicine","score_opus":0.14623062078939802,"score_gpt":0.45932080891479476,"score_spread":0.31309018812539674,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2135388962","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85860604,0.00032312764,0.11341915,0.0005810146,0.005578372,0.01981643,0.00017878805,0.00030914324,0.0011879366],"genre_scores_gemma":[0.807798,0.000402863,0.1871671,0.002198186,0.00042436222,0.0013179074,0.0004946394,0.0001319492,0.00006496726],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9731972,0.0071692104,0.009878165,0.0014135474,0.006639466,0.0017023778],"domain_scores_gemma":[0.89227533,0.09720731,0.003785131,0.0031074402,0.0018996284,0.0017251711],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0151616605,0.0009833089,0.00564239,0.0015843963,0.0017720078,0.00007844615,0.0023004566,0.001270466,0.0035054793],"category_scores_gemma":[0.05292404,0.00068679894,0.001002361,0.0012309528,0.0005296584,0.0002968609,0.0014793178,0.0013638657,0.00018502805],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.5883008,0.0014354258,0.0010373135,0.000025993817,0.0001389344,0.000043371903,0.0025108543,0.000013092344,0.000012641019,0.0002913738,0.00062851276,0.40556166],"study_design_scores_gemma":[0.9634147,0.0032275026,0.0032378547,0.0029076259,0.0002643918,0.000010392774,0.001145707,0.006960733,0.00001122314,0.00988652,0.008345592,0.00058771175],"about_ca_topic_score_codex":0.00032721728,"about_ca_topic_score_gemma":0.0015724428,"teacher_disagreement_score":0.40497395,"about_ca_system_score_codex":0.0011525004,"about_ca_system_score_gemma":0.001804566,"threshold_uncertainty_score":0.9995583},"labels":[],"label_agreement":null},{"id":"W2136024222","doi":"10.1177/0272989x07309644","title":"A Second-Order Simulation Model of the Cost-Effectiveness of Managing Dyspepsia in the United States","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Helicobacter pylori-related gastroenterology studies","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Guideline; Randomized controlled trial; Population; Psychological intervention; Helicobacter pylori; Cost effectiveness; Endoscopy; Quality of life (healthcare); Intensive care medicine; Internal medicine; Environmental health; Pathology; Risk analysis (engineering)","score_opus":0.043515462851865395,"score_gpt":0.36580463102823757,"score_spread":0.32228916817637215,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2136024222","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7955809,0.000047769227,0.20308036,0.00047852332,0.000102176746,0.00046909685,0.0000030876186,0.000010651684,0.00022741918],"genre_scores_gemma":[0.99784356,0.000020194446,0.00083164405,0.0012469545,0.000020403915,0.000009930079,0.0000071177224,0.000012931466,0.0000072723783],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9978089,0.0002913661,0.00060058053,0.00019393663,0.0008367238,0.00026853723],"domain_scores_gemma":[0.99505186,0.0041552614,0.00017034024,0.0003292377,0.0002425119,0.000050794013],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0036183835,0.00012915632,0.00034137024,0.00029085117,0.000059505255,0.000004519496,0.00025551888,0.0001374573,0.00005326187],"category_scores_gemma":[0.0031032446,0.00007047682,0.00007057771,0.0009193935,0.00023151,0.000041912288,0.00015314146,0.00047599737,0.0000016893896],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004361552,0.00065619085,0.53374016,0.0004634503,0.0003376405,0.00015862084,0.0062616523,0.3154747,0.0011922499,0.0007129717,0.00017636055,0.13646448],"study_design_scores_gemma":[0.0020431192,0.00013272926,0.43973148,0.0016357428,0.000059030328,0.000017994933,0.00048927724,0.5535185,0.00015178496,0.0020002972,0.00016464273,0.000055421904],"about_ca_topic_score_codex":0.000017547316,"about_ca_topic_score_gemma":0.000103484395,"teacher_disagreement_score":0.23804379,"about_ca_system_score_codex":0.000067648456,"about_ca_system_score_gemma":0.00004743667,"threshold_uncertainty_score":0.3715098},"labels":[],"label_agreement":null},{"id":"W2137387795","doi":"10.1177/0272989x06297395","title":"A Little Planning Goes a Long Way: Multilevel Allocation of HIV Prevention Resources","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"National Institute on Drug Abuse","keywords":"Resource allocation; Actuarial science; Business; Human immunodeficiency virus (HIV); Yield (engineering); Economics; Public economics; Computer science; Medicine; Computer network","score_opus":0.05039823766169159,"score_gpt":0.42572724314820304,"score_spread":0.37532900548651144,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2137387795","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.46201834,0.00044458933,0.5320108,0.00029007145,0.00004791133,0.0001962695,0.0000027995145,0.000036515845,0.004952717],"genre_scores_gemma":[0.9850971,0.000024385454,0.010251934,0.00007844697,0.00021188708,0.000010065243,0.000021563264,0.00001763347,0.004287],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.997174,0.00007716881,0.000684775,0.00025950532,0.0014757985,0.00032875175],"domain_scores_gemma":[0.9983407,0.0007592548,0.00014267713,0.00026534926,0.00019869399,0.00029328695],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002358237,0.00011101587,0.00026760108,0.00038458328,0.00009769633,0.000021106389,0.00018345853,0.0001620961,0.001685128],"category_scores_gemma":[0.006594031,0.0000870622,0.00018945713,0.00031408714,0.000116693474,0.00010528619,0.00011939161,0.00034186596,0.00008184341],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00047347182,0.00031941093,0.013502156,0.00013048954,0.000046065696,0.00015512621,0.0007028569,0.000008246505,0.0012026327,0.00018632866,0.009438886,0.97383434],"study_design_scores_gemma":[0.007372524,0.004550003,0.7606779,0.037426207,0.0001663425,0.00041750248,0.0026046305,0.13863747,0.0064736675,0.004175772,0.037006386,0.0004915992],"about_ca_topic_score_codex":0.000016050113,"about_ca_topic_score_gemma":0.000037385842,"teacher_disagreement_score":0.9733427,"about_ca_system_score_codex":0.000066022534,"about_ca_system_score_gemma":0.00008104732,"threshold_uncertainty_score":0.99922746},"labels":[],"label_agreement":null},{"id":"W2138858328","doi":"10.1177/0272989x11416870","title":"Bridging Health Technology Assessment (HTA) and Efficient Health Care Decision Making with Multicriteria Decision Analysis (MCDA)","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":168,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University","funders":"","keywords":"Multiple-criteria decision analysis; Health technology; Decision analysis; Management science; Medicine; Consistency (knowledge bases); Health care; Transparency (behavior); Actuarial science; Operations research; Computer science; Statistics; Mathematics; Engineering; Business; Economics","score_opus":0.2001476001113418,"score_gpt":0.47856238755398384,"score_spread":0.27841478744264203,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2138858328","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3206709,0.004325302,0.6682276,0.005068663,0.00054107717,0.0006923292,0.00006216544,0.00011559058,0.00029635045],"genre_scores_gemma":[0.72738963,0.0002119276,0.26545548,0.0067083524,0.0001132694,0.00005160617,0.000014600845,0.000049745726,0.0000053449635],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9901605,0.00044798246,0.005801907,0.0016170257,0.00090875686,0.0010637796],"domain_scores_gemma":[0.9923068,0.0026899925,0.0029801268,0.0011547152,0.00019141195,0.0006769021],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021499012,0.00045123047,0.0022905904,0.002574796,0.0008791835,0.00018958225,0.0006985046,0.00038589194,0.0009535524],"category_scores_gemma":[0.006124382,0.0004481374,0.00021778565,0.001716191,0.00022022954,0.00026333536,0.00047276856,0.0006430941,0.00017298029],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00026044808,0.00031222447,0.13972141,0.00037312947,0.00029961282,0.000051287083,0.0065786946,0.0015240945,4.9263093e-7,0.015505488,0.0024514357,0.8329217],"study_design_scores_gemma":[0.004703401,0.0013421277,0.46000582,0.0076889745,0.000104906896,0.00016479746,0.010497427,0.49099073,0.0000022595102,0.015024612,0.007970144,0.0015048182],"about_ca_topic_score_codex":0.0003271395,"about_ca_topic_score_gemma":0.0009686124,"teacher_disagreement_score":0.83141685,"about_ca_system_score_codex":0.001600981,"about_ca_system_score_gemma":0.0007016308,"threshold_uncertainty_score":0.9999597},"labels":[],"label_agreement":null},{"id":"W2139793607","doi":"10.1177/0272989x12454578","title":"Dynamic Transmission Modeling","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":145,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; Western University","funders":"","keywords":"Transmission (telecommunications); Population; Disease transmission; Set (abstract data type); Computer science; Communicable disease; Risk analysis (engineering); Medicine; Virology; Environmental health; Public health; Telecommunications","score_opus":0.2676586335729907,"score_gpt":0.4942500310837464,"score_spread":0.2265913975107557,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2139793607","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12701724,0.0010540086,0.8691417,0.0012266628,0.00028872633,0.00012220991,6.9398595e-7,0.00016560282,0.0009831821],"genre_scores_gemma":[0.8710244,0.00014311216,0.12738794,0.0012425984,0.00012764255,0.000012234289,6.845054e-7,0.000020266458,0.00004110522],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9973174,0.00013087997,0.00063951715,0.0002841049,0.0010736764,0.00055441813],"domain_scores_gemma":[0.9929656,0.006294167,0.00008435175,0.00027964995,0.000043278258,0.00033299497],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0033527205,0.00018592305,0.0004260962,0.00007404955,0.00020653938,0.000013854043,0.00032935248,0.00024281236,0.0015917443],"category_scores_gemma":[0.019706745,0.00011978943,0.00016187984,0.00019911928,0.000057936686,0.000106133695,0.0002495654,0.0003897102,0.0001247257],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006832047,0.00021749077,0.0025836255,0.00008342467,0.0000274357,0.00002208756,0.0004363746,0.00037568083,0.00005703124,0.012607286,0.0033470297,0.9801742],"study_design_scores_gemma":[0.00031280285,0.000029688757,0.001033078,0.00063968945,0.000026561369,0.000014765577,0.000058360863,0.62044126,0.0000035372727,0.36790094,0.009330258,0.0002090554],"about_ca_topic_score_codex":0.0000045929837,"about_ca_topic_score_gemma":0.000004502262,"teacher_disagreement_score":0.97996515,"about_ca_system_score_codex":0.00011094394,"about_ca_system_score_gemma":0.000026718697,"threshold_uncertainty_score":0.9993209},"labels":[],"label_agreement":null},{"id":"W2140349292","doi":"10.1177/0272989x0102100506","title":"Optimal Investment in a Portfolio of HIV Prevention Programs","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"HIV, Drug Use, Sexual Risk","field":"Medicine","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"National Institute on Drug Abuse","keywords":"Population; Resource allocation; Actuarial science; Investment (military); Business; Medicine; Economics; Environmental health","score_opus":0.06362461979662012,"score_gpt":0.40015603227547725,"score_spread":0.3365314124788571,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140349292","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96813864,0.00024208517,0.01903093,0.0005030575,0.0001337976,0.00054702786,8.610986e-7,0.000057742593,0.01134587],"genre_scores_gemma":[0.97560275,0.00010281373,0.022744086,0.00050093845,0.00013437259,0.000036973932,0.000020101163,0.000026256772,0.0008317325],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99678016,0.00008247189,0.00081399945,0.00034566337,0.0016271562,0.00035056323],"domain_scores_gemma":[0.9989118,0.00019395354,0.00015851636,0.00035778663,0.000080284655,0.000297649],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0011727099,0.0001578661,0.00043557017,0.000315118,0.000027419741,0.000011998463,0.00019056292,0.0002056759,0.0013393245],"category_scores_gemma":[0.001731665,0.00012597739,0.00010704961,0.0006694227,0.00011021141,0.00008064086,0.00013866257,0.0003550665,0.000089729205],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004169994,0.0009410042,0.049556315,0.000046902525,0.000022222537,0.0010272374,0.00078287936,0.000050509287,0.00010539239,0.0003256091,0.0017975842,0.94492733],"study_design_scores_gemma":[0.037480123,0.008694586,0.5276352,0.040122736,0.0005602706,0.0033239275,0.017644312,0.17111735,0.0011473114,0.024955772,0.16544195,0.0018764476],"about_ca_topic_score_codex":0.000014861688,"about_ca_topic_score_gemma":0.000043591008,"teacher_disagreement_score":0.9430509,"about_ca_system_score_codex":0.00010877759,"about_ca_system_score_gemma":0.00021359476,"threshold_uncertainty_score":0.9995736},"labels":[],"label_agreement":null},{"id":"W2140421779","doi":"10.1177/0272989x08315241","title":"The Half-Cycle Correction Explained: Two Alternative Pedagogical Approaches","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":94,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; Health Sciences Centre; Guelph General Hospital; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Computer science; Psychology","score_opus":0.641460467607799,"score_gpt":0.4960213170226308,"score_spread":0.14543915058516815,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140421779","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.42573917,0.0046693697,0.48670062,0.04096975,0.008705866,0.0009914609,0.000029068722,0.00020934765,0.03198534],"genre_scores_gemma":[0.9884363,0.00040805808,0.0036198723,0.0050983788,0.0012292485,0.00009817639,0.000007750369,0.000031257325,0.0010709936],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99500096,0.00042325392,0.002979002,0.0006197648,0.00050336047,0.00047365774],"domain_scores_gemma":[0.9908208,0.007200177,0.0011272023,0.00050917326,0.00005766019,0.00028498148],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.01664973,0.00019729229,0.0006466578,0.0001854064,0.0009987307,0.00009806035,0.0005799776,0.00017677661,0.0011210956],"category_scores_gemma":[0.023440449,0.00016898072,0.00016495379,0.00024032462,0.00024531194,0.00030216548,0.00016086258,0.00045103472,0.002312289],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003311677,0.0005848012,0.042032957,0.000087146705,0.00021725392,0.00013107434,0.014124629,0.0049937926,9.409497e-7,0.42008382,0.21398973,0.30342266],"study_design_scores_gemma":[0.0020885654,0.00017060347,0.015944898,0.0003463209,0.000007163345,0.00029781353,0.0032120151,0.5737346,0.0000051077695,0.122111075,0.28146058,0.0006212385],"about_ca_topic_score_codex":0.0002156095,"about_ca_topic_score_gemma":0.00019969,"teacher_disagreement_score":0.56874084,"about_ca_system_score_codex":0.00032216904,"about_ca_system_score_gemma":0.00014593992,"threshold_uncertainty_score":0.99979204},"labels":[],"label_agreement":null},{"id":"W2140728867","doi":"10.1177/0272989x07307272","title":"Are Patient Decision Aids the Best Way to Improve Clinical Decision Making? Report of the IPDAS Symposium","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":73,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"Agency for Healthcare Research and Quality","keywords":"Decision aids; Proposition; Medical decision making; Decision theory; Clinical decision making; Decision analysis; Health care; Optimal decision; Management science; Psychology; Medicine; Computer science; Decision tree; Political science; Alternative medicine; Family medicine; Artificial intelligence; Economics; Epistemology","score_opus":0.15187703525862073,"score_gpt":0.4980633069236943,"score_spread":0.3461862716650736,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2140728867","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.883921,0.001015685,0.08635911,0.006673567,0.015524104,0.0036882968,0.00003578718,0.00014345517,0.0026389607],"genre_scores_gemma":[0.9559962,0.0000980502,0.030518807,0.01227519,0.0007166305,0.00018300833,0.0000052412497,0.00008957912,0.00011725114],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9836022,0.002230356,0.0065203,0.0011775612,0.0052049174,0.0012646494],"domain_scores_gemma":[0.9444439,0.04411039,0.0037808584,0.0054122084,0.0014251671,0.0008275016],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.012074713,0.00048238938,0.0010000364,0.0003732281,0.0022576558,0.000043019845,0.0032945054,0.0011230782,0.0005458448],"category_scores_gemma":[0.0879663,0.00027160134,0.00054748106,0.0015857504,0.00045376035,0.00017690165,0.0045847315,0.0033831128,0.0004262133],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010500965,0.00029689274,0.20962556,0.00006994626,0.000026806902,0.00014119621,0.002219331,0.000037570546,0.000025402429,0.00037956456,0.013282234,0.7728454],"study_design_scores_gemma":[0.003995387,0.0016742591,0.54377097,0.0379625,0.00017438758,0.00024410429,0.024149524,0.0028894814,0.00010741521,0.02248029,0.361448,0.0011037047],"about_ca_topic_score_codex":0.00010698785,"about_ca_topic_score_gemma":0.0009438405,"teacher_disagreement_score":0.7717417,"about_ca_system_score_codex":0.0005984349,"about_ca_system_score_gemma":0.0007970284,"threshold_uncertainty_score":0.9999736},"labels":[],"label_agreement":null},{"id":"W2141146484","doi":"10.1177/0272989x07312709","title":"What Factors Influence Case Managers' Resource Allocation Decisions? A Systematic Review of the Literature","year":2008,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Systematic review; Resource allocation; Business; Management science; Actuarial science; MEDLINE; Risk analysis (engineering); Psychology; Computer science; Economics; Political science","score_opus":0.06536634993108341,"score_gpt":0.45302657118854295,"score_spread":0.3876602212574595,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2141146484","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00007370312,0.992326,0.00030092217,0.00022849701,0.0026528197,0.004205506,0.000024488476,0.000058414204,0.00012962018],"genre_scores_gemma":[0.00008221584,0.99770135,0.00017697539,0.0010564667,0.00020745717,0.00024287924,0.00005826348,0.0000681627,0.00040624512],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9888803,0.0036603478,0.0036886006,0.00066977943,0.002564704,0.00053628447],"domain_scores_gemma":[0.9728992,0.021826822,0.002480033,0.0019964785,0.00051315455,0.00028430906],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.003065542,0.0005958002,0.0037292312,0.0004459574,0.0009123692,0.0000652978,0.0014234224,0.001242422,0.00028359678],"category_scores_gemma":[0.05295945,0.00029969987,0.0012147375,0.002497705,0.0001517392,0.00025407012,0.00068841886,0.0023298806,0.000115095296],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004410225,0.000021300752,0.0000012852382,0.6468345,0.00005033351,0.00049719645,0.0012756924,2.3263355e-7,3.5516756e-9,0.000029016923,0.016250765,0.33503526],"study_design_scores_gemma":[0.00009307342,0.000014933492,0.000003815579,0.7416606,0.0007168603,0.0012638858,0.0009116801,0.000002968161,9.068898e-9,0.000059114864,0.2550842,0.00018891507],"about_ca_topic_score_codex":0.000015152442,"about_ca_topic_score_gemma":0.000017917842,"teacher_disagreement_score":0.33484635,"about_ca_system_score_codex":0.00037115125,"about_ca_system_score_gemma":0.0009699421,"threshold_uncertainty_score":0.9999718},"labels":[],"label_agreement":null},{"id":"W2143457017","doi":"10.1177/0272989x07300604","title":"A Review and Meta-Analysis of Prostate Cancer Utilities","year":2007,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":102,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University Health Network","funders":"","keywords":"Prostate cancer; Affect (linguistics); Quality of life (healthcare); Variance (accounting); Medicine; Actuarial science; Cancer; Computer science; Psychology; Accounting; Nursing; Economics","score_opus":0.7527892240790977,"score_gpt":0.6008747232664101,"score_spread":0.15191450081268754,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143457017","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000019601184,0.9920239,0.0022990745,0.0023772174,0.0002066832,0.0012208287,0.0012210148,0.00002159143,0.0006277671],"genre_scores_gemma":[0.000007386124,0.9842024,0.0016479007,0.013200607,0.00008493082,0.0003122017,0.000046165627,0.00004405906,0.0004543353],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98690295,0.0006079903,0.010717324,0.00092520355,0.00041893029,0.00042759985],"domain_scores_gemma":[0.9855999,0.0072673885,0.0058878376,0.0008277654,0.00011661795,0.00030051696],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.042917147,0.00043888582,0.014264777,0.0014077631,0.00009836858,0.00004783853,0.000548339,0.00047031717,0.013012942],"category_scores_gemma":[0.014710234,0.00037489054,0.002197894,0.001311443,0.0001733081,0.00015196878,0.00021960381,0.0004703547,0.00022429312],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000042639545,0.00005651624,0.00007572745,0.13452384,0.07643338,0.000006858114,0.00025936222,0.0000035146597,1.4131585e-10,0.0093793925,0.024307055,0.7549501],"study_design_scores_gemma":[0.00006796728,0.000013874507,0.000016594438,0.008631494,0.1084213,0.000005430096,0.000021492586,0.00014265721,1.1027019e-9,0.0013277439,0.88107705,0.00027441533],"about_ca_topic_score_codex":0.00023004529,"about_ca_topic_score_gemma":0.00019147535,"teacher_disagreement_score":0.85677,"about_ca_system_score_codex":0.00027630856,"about_ca_system_score_gemma":0.00039385224,"threshold_uncertainty_score":0.9998703},"labels":[],"label_agreement":null},{"id":"W2143804633","doi":"10.1177/0272989x15595365","title":"Exclusion Criteria in National Health State Valuation Studies","year":2015,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":36,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital; University of British Columbia; Simon Fraser University; Vancouver Coastal Health Research Institute; Vancouver Coastal Health","funders":"Canadian Institutes of Health Research; Health Technology Assessment international","keywords":"Valuation (finance); Time-trade-off; Data extraction; Medicine; MEDLINE; Quality-adjusted life year; Inclusion and exclusion criteria; Actuarial science; EQ-5D; Family medicine; Quality of life (healthcare); Health related quality of life; Disease; Alternative medicine; Economics; Political science; Cost effectiveness","score_opus":0.8215206556921038,"score_gpt":0.6539012632462616,"score_spread":0.16761939244584212,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143804633","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000022122647,0.97550035,0.014210623,0.0055809664,0.0024504391,0.0013215219,0.0002438762,0.00005275867,0.0006173433],"genre_scores_gemma":[0.000048418595,0.98459595,0.0073738294,0.0066327397,0.00073132827,0.00023594024,0.00014537318,0.0000730934,0.00016330573],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9835958,0.0019829257,0.011417507,0.0011273623,0.0012451265,0.000631297],"domain_scores_gemma":[0.98708737,0.0064227893,0.0052196947,0.00053650147,0.0003117541,0.00042187035],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.11617956,0.00046397632,0.004901055,0.0015150815,0.00021849386,0.000110719106,0.0006604228,0.00048291762,0.0008637517],"category_scores_gemma":[0.060976967,0.00047787096,0.0003153081,0.00072851544,0.00009939568,0.00038446783,0.0004117303,0.0006971991,0.0029810716],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000007275185,0.000057823338,0.000043506247,0.006585367,0.00006207357,0.00000788179,0.0010904396,0.000021907827,6.581009e-10,0.0054010465,0.094091214,0.8926315],"study_design_scores_gemma":[0.00046248685,0.000057653662,0.00004960474,0.026627865,0.0000070739084,0.000019022891,0.00014352582,0.0023906121,1.0518393e-9,0.108832896,0.861058,0.00035125768],"about_ca_topic_score_codex":0.0000826995,"about_ca_topic_score_gemma":0.00023557329,"teacher_disagreement_score":0.8922802,"about_ca_system_score_codex":0.0050282665,"about_ca_system_score_gemma":0.0031701303,"threshold_uncertainty_score":0.9997673},"labels":[],"label_agreement":null},{"id":"W2144528553","doi":"10.1177/0272989x11426176","title":"Test Result–Based Sampling","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient Safety and Medication Errors","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Sample size determination; Sampling (signal processing); Statistics; Ranging; Sampling design; Simple random sample; Sample (material); Computer science; Population; Mathematics; Medicine","score_opus":0.29731825932960054,"score_gpt":0.5081463113278938,"score_spread":0.21082805199829324,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2144528553","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04285598,0.00007054667,0.8885177,0.002048527,0.0025321457,0.00048158987,0.000014011812,0.00028947165,0.06319003],"genre_scores_gemma":[0.95793205,0.000015684165,0.03300075,0.008357662,0.0002932192,0.00006141473,0.000013325639,0.000022833401,0.00030303822],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99691963,0.00023193471,0.0008414254,0.00032826033,0.001183819,0.00049490115],"domain_scores_gemma":[0.9934557,0.0052265706,0.00030643996,0.00044168395,0.00014860065,0.00042098606],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0022965192,0.0001342905,0.00022938102,0.00016529714,0.00062584813,0.0000010680708,0.0004152762,0.00035462686,0.013445321],"category_scores_gemma":[0.026008107,0.00010406334,0.00007133876,0.000345956,0.00010677537,0.00007933193,0.00014758373,0.0008457825,0.0018957198],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009035865,0.00088693283,0.3201658,0.00025418543,0.000030615665,0.0001702491,0.008484338,0.000050756116,0.0001700168,0.007806882,0.04155228,0.61952436],"study_design_scores_gemma":[0.004315248,0.0002878169,0.49955735,0.0076860995,0.00004256331,0.0000073051933,0.0014095991,0.012073526,0.000056322082,0.013675814,0.46035603,0.00053232],"about_ca_topic_score_codex":0.000019968078,"about_ca_topic_score_gemma":0.000037303307,"teacher_disagreement_score":0.9150761,"about_ca_system_score_codex":0.000082666615,"about_ca_system_score_gemma":0.00049363804,"threshold_uncertainty_score":0.9988814},"labels":[],"label_agreement":null},{"id":"W2145135401","doi":"10.1177/0272989x07307319","title":"Do Patient Decision Aids Meet Effectiveness Criteria of the International Patient Decision Aid Standards Collaboration? A Systematic Review and Meta-analysis","year":2007,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":272,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Algonquin College; University of Ottawa","funders":"","keywords":"Decision aids; Decision analysis; Medicine; Meta-analysis; Multiple-criteria decision analysis; MEDLINE; Management science; Risk analysis (engineering); Operations research; Engineering; Alternative medicine","score_opus":0.24418946367039823,"score_gpt":0.5520368353204408,"score_spread":0.30784737165004256,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2145135401","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000095224685,0.9735645,0.011655083,0.00023711471,0.002029737,0.010928956,0.0012023709,0.00005881843,0.00022818215],"genre_scores_gemma":[0.0035392917,0.9796517,0.011453804,0.0020927137,0.00006846786,0.0029522246,0.00011131631,0.0001150306,0.000015470587],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9709186,0.010676769,0.0083130635,0.0013136363,0.008094805,0.0006831362],"domain_scores_gemma":[0.9487303,0.037284166,0.0056664236,0.0037288717,0.004084798,0.00050545717],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012514364,0.00095094746,0.009215784,0.0013170156,0.0010764151,0.0001084254,0.0027049999,0.0011810806,0.003039914],"category_scores_gemma":[0.05780108,0.0005275549,0.0024878478,0.004304948,0.00023682335,0.00025939444,0.0032703776,0.0016979213,0.000040020095],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023381492,0.0002102785,0.000084920925,0.52571243,0.018544095,0.00002254396,0.0013182852,0.000010747721,7.723009e-8,0.0006090381,0.0037053425,0.44954842],"study_design_scores_gemma":[0.0006395769,0.00022670922,0.000013863217,0.5462706,0.172812,0.000024136718,0.0007125709,0.00013008984,3.0420537e-7,0.0013165298,0.2773197,0.00053390267],"about_ca_topic_score_codex":0.000034144443,"about_ca_topic_score_gemma":0.00022417241,"teacher_disagreement_score":0.44901454,"about_ca_system_score_codex":0.0017633997,"about_ca_system_score_gemma":0.0024672134,"threshold_uncertainty_score":0.9997176},"labels":[],"label_agreement":null},{"id":"W2146006549","doi":"10.1177/0272989x13511704","title":"Knowledge, Attitudes, and Self-efficacy as Predictors of Preparedness for Oncology Clinical Trials","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Ethics in Clinical Research","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Cancer Institute","keywords":"Preparedness; Self-efficacy; Clinical trial; Moderation; Medicine; Cancer; Family medicine; Psychology; Internal medicine; Social psychology","score_opus":0.5327753667992864,"score_gpt":0.6832187117492886,"score_spread":0.15044334495000222,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146006549","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9624339,0.0014979814,0.016195236,0.0035534387,0.00241359,0.0037753724,0.000009337625,0.00015066353,0.009970491],"genre_scores_gemma":[0.97254926,0.0020816508,0.022346573,0.00096029765,0.0012820443,0.00018035063,0.000010002519,0.000044665423,0.00054517604],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9919978,0.0013812431,0.003249408,0.00077645556,0.0021149863,0.00048014018],"domain_scores_gemma":[0.3811919,0.6105141,0.0011896179,0.0016105801,0.0030105582,0.002483255],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.055846673,0.00020724472,0.0017035749,0.00020593939,0.00010681401,0.000036330224,0.00050927326,0.0016617257,0.002027054],"category_scores_gemma":[0.5980261,0.00014269995,0.00043373328,0.00028698478,0.0008308654,0.000079719866,0.00064158626,0.0019058697,0.00014159946],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016572241,0.0054792557,0.02571806,0.00074128556,0.00041812885,0.000055105622,0.0006015885,8.388322e-7,0.00009604184,0.005115409,0.015364903,0.94475216],"study_design_scores_gemma":[0.059120655,0.026873471,0.42434764,0.01841395,0.0013631869,0.0003790111,0.0008255643,0.017410826,0.00024916683,0.1795235,0.27042684,0.0010662084],"about_ca_topic_score_codex":0.000016109008,"about_ca_topic_score_gemma":0.000041828967,"teacher_disagreement_score":0.94368595,"about_ca_system_score_codex":0.00013587908,"about_ca_system_score_gemma":0.0028332015,"threshold_uncertainty_score":0.9996343},"labels":[],"label_agreement":null},{"id":"W2146075572","doi":"10.1177/0272989x0102100404","title":"An Off-the-Shelf Help List","year":2001,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":93,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Off the shelf; Medicine; Computer science","score_opus":0.5076141008802234,"score_gpt":0.5445087613010589,"score_spread":0.036894660420835446,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146075572","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000018888466,0.96154416,0.026319208,0.0037047695,0.0025561126,0.0010835241,0.0001895314,0.00010394847,0.004479866],"genre_scores_gemma":[0.000051495397,0.983065,0.0024766796,0.010561492,0.0027628208,0.00021371114,0.000107536245,0.00013050507,0.0006307498],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9863408,0.0011935002,0.009669703,0.0013209966,0.00067732215,0.0007976547],"domain_scores_gemma":[0.9839018,0.009129815,0.0043581896,0.0018945934,0.00007743307,0.0006381692],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04383399,0.00058248837,0.0043255845,0.00066785293,0.0004941901,0.000341385,0.002140603,0.0010470193,0.008964957],"category_scores_gemma":[0.026839579,0.0005027841,0.0007623535,0.0007572297,0.00019428055,0.00036482714,0.00028036442,0.0011608097,0.010605345],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000031969312,0.000069748865,0.00005908576,0.0019502693,0.000055836863,0.000022715607,0.00009697989,0.000003257154,5.6726357e-10,0.0126938345,0.0409087,0.9441364],"study_design_scores_gemma":[0.0002228672,0.000055824094,0.000031859472,0.010698175,0.000043763794,0.000106679276,0.00007029993,0.002099659,2.0476538e-9,0.010106873,0.97606325,0.0005007747],"about_ca_topic_score_codex":0.000113064896,"about_ca_topic_score_gemma":0.00013167413,"teacher_disagreement_score":0.9436356,"about_ca_system_score_codex":0.00083810673,"about_ca_system_score_gemma":0.00066998723,"threshold_uncertainty_score":0.9997424},"labels":[],"label_agreement":null},{"id":"W2146395157","doi":"10.1177/0272989x07305322","title":"The Value of Bayes Theorem in the Interpretation of Subjective Diagnostic Findings: What Can We Learn from Agreement Studies?","year":2007,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vancouver Coastal Health","funders":"","keywords":"Bayes' theorem; Interpretation (philosophy); Value (mathematics); Agreement; Econometrics; Psychology; Mathematics; Statistics; Actuarial science; Medicine; Mathematical economics; Bayesian probability; Computer science; Economics; Philosophy; Linguistics","score_opus":0.1676988182647056,"score_gpt":0.4484082102253236,"score_spread":0.28070939196061806,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146395157","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9012521,0.015579216,0.06576777,0.015098423,0.0011589366,0.000618133,0.000031312476,0.0000075297903,0.0004865825],"genre_scores_gemma":[0.99354243,0.0028559265,0.0008098157,0.0026130043,0.00012515765,0.000028979839,0.0000038698204,0.000010954036,0.000009859822],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99483484,0.00058765605,0.0035237155,0.00032982946,0.0004408358,0.00028313103],"domain_scores_gemma":[0.9288646,0.06904782,0.0014974992,0.00042605607,0.00009602397,0.000068029505],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04422505,0.00012905042,0.000641466,0.00023042446,0.00015029848,0.00005671651,0.00055125804,0.000117106756,0.00015696454],"category_scores_gemma":[0.07490171,0.00009275559,0.00010759753,0.00029726332,0.00022482585,0.00021114573,0.00011624664,0.0002606556,0.000055189194],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030780517,0.00029678683,0.13123268,0.00026697342,0.00032490483,0.000016842141,0.1403121,0.0012222729,0.0000030288168,0.2646887,0.0029419854,0.4583859],"study_design_scores_gemma":[0.0012504912,0.00025750446,0.27010775,0.0052998248,0.000021631902,0.0000039457946,0.13504875,0.028914453,0.00003153245,0.55626905,0.0025166797,0.000278423],"about_ca_topic_score_codex":0.0003053155,"about_ca_topic_score_gemma":0.0018492322,"teacher_disagreement_score":0.4581075,"about_ca_system_score_codex":0.00032302778,"about_ca_system_score_gemma":0.00008945125,"threshold_uncertainty_score":0.98417145},"labels":[],"label_agreement":null},{"id":"W2146665164","doi":"10.1177/0272989x13497262","title":"Medical Therapy v. PCI in Stable Coronary Artery Disease","year":2013,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Coronary Interventions and Diagnostics","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Conventional PCI; Quality-adjusted life year; Population; Percutaneous coronary intervention; Coronary artery disease; Restenosis; Cost effectiveness; Cohort; Internal medicine; Life expectancy; Emergency medicine; Surgery; Stent; Myocardial infarction","score_opus":0.028745020620735404,"score_gpt":0.340868792821779,"score_spread":0.3121237722010436,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2146665164","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9776011,0.0020080893,0.011286373,0.0061824564,0.0005650825,0.00048476266,0.0000048045704,0.00007411741,0.0017931948],"genre_scores_gemma":[0.991818,0.00073397235,0.0011331656,0.0055480627,0.00024199337,0.00008301883,0.000021120799,0.000025407839,0.00039525077],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9966704,0.000083934974,0.00064697507,0.0003514245,0.0018582735,0.00038897878],"domain_scores_gemma":[0.9971482,0.0013554461,0.000057866917,0.00041697628,0.00010061574,0.0009208931],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0006474096,0.00017294104,0.00033612354,0.00020753284,0.00006577445,0.00003750388,0.00025316712,0.00019815637,0.07411543],"category_scores_gemma":[0.004831528,0.00012820444,0.00016549588,0.00027139453,0.00012550222,0.00013684324,0.0001715476,0.0004827244,0.00078897894],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00031796718,0.0008572805,0.15697165,0.000025010448,0.000018786228,0.003929355,0.000037300302,0.0000035948906,0.000009419612,0.0002849012,0.022141516,0.8154032],"study_design_scores_gemma":[0.004223327,0.00038942174,0.9457885,0.0035120896,0.000018821836,0.000501295,0.00010418207,0.018491708,0.000004286426,0.0074027167,0.019349707,0.00021395959],"about_ca_topic_score_codex":0.00003979583,"about_ca_topic_score_gemma":0.0000477156,"teacher_disagreement_score":0.81518924,"about_ca_system_score_codex":0.000079096615,"about_ca_system_score_gemma":0.0003674247,"threshold_uncertainty_score":0.99998903},"labels":[],"label_agreement":null},{"id":"W2149049271","doi":"10.1177/0272989x07312474","title":"Health Technology Assessment in the Cost-Disutility Plane","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; University of Toronto","funders":"","keywords":"Inefficiency; Net present value; Cost–benefit analysis; Cost effectiveness; Value of information; Computer science; Risk analysis (engineering); Economics; Operations management; Microeconomics; Business","score_opus":0.43480952321486205,"score_gpt":0.5296234635130782,"score_spread":0.09481394029821616,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2149049271","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.47003832,0.0032199053,0.15048128,0.36705723,0.0011884279,0.0015862692,0.00009221316,0.000119961405,0.00621637],"genre_scores_gemma":[0.9602304,0.00027180606,0.0065376805,0.032625094,0.00016648999,0.00010945212,0.000012584814,0.0000138711675,0.00003262424],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99412817,0.0005384928,0.0038553083,0.000545922,0.0004174519,0.0005146414],"domain_scores_gemma":[0.9949136,0.0031783595,0.0010429892,0.00066622277,0.000028706045,0.00017015178],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.035004646,0.00014966502,0.00086884294,0.00046510174,0.0003354456,0.000031087202,0.00070397306,0.00024192594,0.0011304417],"category_scores_gemma":[0.010707486,0.0001355245,0.00007630961,0.0005244105,0.00017631835,0.00014226895,0.00012118351,0.00068797683,0.0008306413],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000033598484,0.00048269096,0.5712756,0.00013983084,0.000025449363,0.00009840711,0.0026565292,0.000107837775,1.5489225e-7,0.24254055,0.10068383,0.08195554],"study_design_scores_gemma":[0.0014139124,0.00015915623,0.7126926,0.00042129456,0.0000013774999,0.0002227568,0.0019461171,0.018079262,2.1752489e-7,0.08099378,0.18376401,0.00030546394],"about_ca_topic_score_codex":0.00023889178,"about_ca_topic_score_gemma":0.00045807543,"teacher_disagreement_score":0.49019206,"about_ca_system_score_codex":0.00060368446,"about_ca_system_score_gemma":0.000501856,"threshold_uncertainty_score":0.9999473},"labels":[],"label_agreement":null},{"id":"W2149779663","doi":"10.1177/0272989x10386800","title":"Predicting EQ-5D Utility Scores from the Seattle Angina Questionnaire in Coronary Artery Disease","year":2010,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Sciences Centre; University of Toronto; University of Alberta; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research","keywords":"Statistics; Linear regression; Heteroscedasticity; Medicine; Cohort; Coronary artery disease; Regression analysis; Simple linear regression; Unstable angina; Regression; Physical therapy; Mathematics; Coronary heart disease; Cardiology","score_opus":0.2553805329291292,"score_gpt":0.41139738887091826,"score_spread":0.15601685594178905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2149779663","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.95788205,0.0017704297,0.016434563,0.020742325,0.0021353937,0.00042011932,0.00016354444,0.000070693895,0.00038086443],"genre_scores_gemma":[0.98799783,0.00004058088,0.0013574068,0.009630127,0.0008429746,0.00005959247,0.000022865488,0.00002489396,0.000023744657],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9955201,0.00036240192,0.0026680988,0.0006556344,0.00037672248,0.00041701974],"domain_scores_gemma":[0.99338096,0.00463414,0.0007370907,0.00081320346,0.000046748995,0.00038787103],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.016597228,0.00018574775,0.0005611572,0.00016792293,0.0002942262,0.000110747016,0.00060152315,0.00022747874,0.002269198],"category_scores_gemma":[0.03913152,0.00017293119,0.000121339566,0.00022620456,0.0001956161,0.0003674822,0.00019783554,0.00072353525,0.0008418729],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003912192,0.00007501595,0.9761266,0.000030446728,0.000012045052,0.000025931742,0.0006579765,0.000029742854,6.7773516e-7,0.004991556,0.0037554496,0.014255433],"study_design_scores_gemma":[0.00044953747,0.000012101795,0.87756205,0.00067600864,0.0000038212115,0.0000066896914,0.00014752227,0.064167164,1.7134492e-7,0.05236502,0.004454916,0.00015500118],"about_ca_topic_score_codex":0.00056024373,"about_ca_topic_score_gemma":0.0021000744,"teacher_disagreement_score":0.09856455,"about_ca_system_score_codex":0.00016783486,"about_ca_system_score_gemma":0.00026927877,"threshold_uncertainty_score":0.9999361},"labels":[],"label_agreement":null},{"id":"W2151481259","doi":"10.1177/0272989x09342752","title":"Increasing the Detection and Response to Adherence Problems with Cardiovascular Medication in Primary Care through Computerized Drug Management Systems: A Randomized Controlled Trial","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Medication Adherence and Compliance","field":"Medicine","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; College of Family Physicians of Canada; McGill University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Drug; Randomized controlled trial; Pharmacotherapy; Lipid profile; Intervention (counseling); Intensive care medicine; Emergency medicine; Internal medicine; Physical therapy; Pharmacology; Cholesterol","score_opus":0.01763316223466533,"score_gpt":0.29213108891049083,"score_spread":0.2744979266758255,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2151481259","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.56656253,0.004744115,0.41435716,0.0016452209,0.00036550104,0.010819268,4.962929e-7,0.0000797328,0.00142598],"genre_scores_gemma":[0.9895952,0.00042284274,0.0067036115,0.0017588673,0.00015045884,0.0012788144,0.00000457923,0.00001316929,0.000072468865],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9944746,0.0015965984,0.00097718,0.00050936785,0.0021746908,0.00026758364],"domain_scores_gemma":[0.99673367,0.0020837372,0.0002488301,0.00057274196,0.00015990998,0.00020111345],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008522589,0.00022379888,0.0014093943,0.00019582112,0.0001546204,0.00007344422,0.00024615714,0.000117034484,0.000038428214],"category_scores_gemma":[0.002284107,0.00012215067,0.00019206975,0.0005117775,0.00014835685,0.00009402769,0.000073447845,0.00033894114,0.000021428277],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.78432626,0.00006902715,0.000020591382,0.00010930799,0.0001385387,0.00008559006,0.0010154886,0.000107121254,0.000055946657,0.000060111593,0.000043801334,0.21396822],"study_design_scores_gemma":[0.9743258,0.0006837389,0.0032431374,0.008566133,0.00036417236,0.00023333778,0.0009862431,0.008666696,0.0000050567664,0.00029499902,0.0024325857,0.00019814698],"about_ca_topic_score_codex":0.00004525082,"about_ca_topic_score_gemma":0.0000068146755,"teacher_disagreement_score":0.42303267,"about_ca_system_score_codex":0.00020189483,"about_ca_system_score_gemma":0.00021312055,"threshold_uncertainty_score":0.49811596},"labels":[],"label_agreement":null},{"id":"W2152033706","doi":"10.1177/0272989x12451339","title":"Looking at Patients’ Choices through the Lens of Expected Utility","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Expected utility hypothesis; Medical decision making; Outcome (game theory); Actuarial science; Medical Expenditure Panel Survey; Perception; Psychology; Exploratory research; Management science; Medicine; Economics; Health care; Family medicine; Health insurance; Microeconomics; Sociology","score_opus":0.34594823469583896,"score_gpt":0.4572533501583475,"score_spread":0.11130511546250854,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2152033706","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94517285,0.0037984797,0.029707095,0.0049414514,0.0019783224,0.0005026051,0.00007192203,0.0000495234,0.013777779],"genre_scores_gemma":[0.9902802,0.000071710754,0.0024649105,0.006591511,0.00044845013,0.00002731395,0.000010456068,0.000021189278,0.0000842445],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9948091,0.00035910142,0.0035157143,0.0003715669,0.00045584058,0.00048867246],"domain_scores_gemma":[0.9923264,0.0049784756,0.0018173738,0.0006378228,0.00008140408,0.00015856195],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.014271248,0.00016217836,0.0007201545,0.00011554841,0.00030555925,0.00002938896,0.0005014824,0.00020183942,0.0050390037],"category_scores_gemma":[0.022314187,0.00013843036,0.00015550895,0.00024210184,0.00016838103,0.0005531245,0.00027188088,0.00023956162,0.0009789434],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000044038912,0.00028085648,0.91812927,0.00013741506,0.000056463523,4.6289153e-7,0.0054018823,0.000015533147,8.4652163e-7,0.04219501,0.024091896,0.009646356],"study_design_scores_gemma":[0.0010601548,0.000055589844,0.69516975,0.00043276415,0.000010613465,0.0000037444865,0.0007105641,0.0027110393,0.000009200076,0.016820068,0.2827031,0.0003134113],"about_ca_topic_score_codex":0.00015664121,"about_ca_topic_score_gemma":0.000073855925,"teacher_disagreement_score":0.2586112,"about_ca_system_score_codex":0.0002821862,"about_ca_system_score_gemma":0.000055989327,"threshold_uncertainty_score":0.9997989},"labels":[],"label_agreement":null},{"id":"W2155209767","doi":"10.1177/0272989x08318468","title":"``Not Everyone Who Needs One Is Going to Get One'': The Influence of Medical Brokering on Patient Candidacy for Total Joint Arthroplasty","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Total Knee Arthroplasty Outcomes","field":"Medicine","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Candidacy; Medicine; Joint arthroplasty; Health care; Promotion (chess); Negotiation; Orthopedic surgery; Nursing; Medical emergency; Family medicine; Arthroplasty; Surgery","score_opus":0.03201727682480866,"score_gpt":0.3025960665410089,"score_spread":0.27057878971620025,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155209767","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97145426,0.00016786194,0.019368699,0.0072461804,0.00048434435,0.0009290068,0.000023565957,0.00007757765,0.00024852954],"genre_scores_gemma":[0.98476046,0.00011565092,0.004163183,0.010279125,0.0003194438,0.00006432216,0.0000037643958,0.000057671292,0.00023635213],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9919502,0.00008981284,0.0013012596,0.00054550165,0.005418462,0.0006947745],"domain_scores_gemma":[0.99570763,0.0022454672,0.00027021547,0.0007166961,0.0002376696,0.000822338],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0011235733,0.00034954483,0.00094136177,0.00038412496,0.00034014438,0.000029644516,0.00034379348,0.00032392435,0.0009890994],"category_scores_gemma":[0.01217862,0.00024610583,0.00030926152,0.0006343682,0.0003914337,0.00012921699,0.0005667611,0.0007290643,0.00018417611],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0066594747,0.0020862634,0.043026615,0.0005793657,0.00059470645,0.0010189879,0.018308058,0.0032921913,0.0041895574,0.0013513493,0.037690815,0.88120264],"study_design_scores_gemma":[0.013511492,0.0074080136,0.8875287,0.024727209,0.00024105783,0.002206712,0.0010246112,0.017500926,0.009876124,0.0003539877,0.034344412,0.0012767296],"about_ca_topic_score_codex":0.000069683294,"about_ca_topic_score_gemma":0.000060215658,"teacher_disagreement_score":0.8799259,"about_ca_system_score_codex":0.00017996483,"about_ca_system_score_gemma":0.00046975928,"threshold_uncertainty_score":0.9999991},"labels":[],"label_agreement":null},{"id":"W2155395793","doi":"10.1177/0272989x04273142","title":"Optimal Statistical Decisions for Hospital Report Cards","year":2005,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"Canadian Institutes of Health Research; Institute for Clinical Evaluative Sciences","keywords":"False positive paradox; Statistical significance; False positives and false negatives; Medicine; Statistics; Quality (philosophy); Relative value; Actuarial science; Operations management; Mathematics; Economics","score_opus":0.09029463823620998,"score_gpt":0.5133631414625724,"score_spread":0.42306850322636247,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2155395793","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.24661432,0.0001186512,0.7414118,0.003966156,0.004400996,0.0012877567,0.00016002447,0.00020554126,0.0018347717],"genre_scores_gemma":[0.73979,0.00003473135,0.25551787,0.0020043056,0.0016425286,0.00048401207,0.00007909181,0.000055161792,0.00039227173],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99383324,0.00035738095,0.0019154618,0.0007117187,0.002217971,0.0009642231],"domain_scores_gemma":[0.9817875,0.015767906,0.00041894164,0.00067068764,0.0005796693,0.00077531533],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0028931252,0.00025223944,0.00052040856,0.0002174707,0.0013034279,0.000018662087,0.0003960852,0.000614198,0.0067903004],"category_scores_gemma":[0.04799006,0.00021178294,0.00015618402,0.00028138206,0.00012893364,0.00018900471,0.00030722973,0.00117457,0.00089645776],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023991606,0.00007690751,0.097469635,0.00005733367,0.000024701923,0.0002458347,0.0007618849,0.00018779193,0.0000024476185,0.005850353,0.14984605,0.7452371],"study_design_scores_gemma":[0.0033492402,0.00055558677,0.25481275,0.0026688823,0.00005742411,0.00015143225,0.0012952646,0.035838887,0.0000051360885,0.009186758,0.69140387,0.00067480747],"about_ca_topic_score_codex":0.000040065705,"about_ca_topic_score_gemma":0.00014104988,"teacher_disagreement_score":0.7445623,"about_ca_system_score_codex":0.000520195,"about_ca_system_score_gemma":0.0010592353,"threshold_uncertainty_score":0.9999967},"labels":[],"label_agreement":null},{"id":"W2157595559","doi":"10.1177/0272989x0202200411","title":"Primary Prevention Drug Therapy: Can It Meet Patients’ Requirements for Reduced Risk?","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Drug; Intensive care medicine; Primary prevention; Risk analysis (engineering); Pharmacology; Internal medicine; Disease","score_opus":0.3970579477170205,"score_gpt":0.45924310028730797,"score_spread":0.06218515257028745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2157595559","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88153535,0.0026288899,0.066916384,0.03368197,0.0033333576,0.002762749,0.00036132685,0.0001302771,0.008649714],"genre_scores_gemma":[0.9569615,0.00081921514,0.01573713,0.023889681,0.0007270492,0.00031610866,0.00007973888,0.00007287681,0.0013966952],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9940779,0.00036669066,0.003911859,0.0007033465,0.00046019795,0.0004800087],"domain_scores_gemma":[0.99475807,0.0022019194,0.0021095767,0.00053971127,0.00011748328,0.0002732552],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.016404524,0.00021420376,0.00079441484,0.0003036415,0.00035914124,0.000095927324,0.0004654883,0.00020936102,0.0033618368],"category_scores_gemma":[0.016508752,0.00023756661,0.0002187448,0.0002067126,0.000057326048,0.000349975,0.000091699556,0.00020111076,0.00061575614],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00012790372,0.0008244043,0.06263429,0.00022745274,0.00014836757,0.000002141001,0.003034376,0.000054943604,0.0000020935097,0.0077271406,0.62933165,0.29588524],"study_design_scores_gemma":[0.013032957,0.0007439716,0.106806695,0.0020354246,0.00003443795,0.000005871205,0.0005632958,0.049505983,0.00001307413,0.20255901,0.6232589,0.0014404086],"about_ca_topic_score_codex":0.00006676042,"about_ca_topic_score_gemma":0.00009688809,"teacher_disagreement_score":0.29444483,"about_ca_system_score_codex":0.0005907568,"about_ca_system_score_gemma":0.00008502869,"threshold_uncertainty_score":0.99754924},"labels":[],"label_agreement":null},{"id":"W2158062010","doi":"10.1177/0272989x02238301","title":"Willingness to Pay for What? A Note on Alternative Definitions of Health Care Program Benefits for Contingent Valuation Studies","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Institute for Advanced Research; McMaster University; St. Joseph's Hospital; St. Joseph’s Healthcare Hamilton; University of Calgary","funders":"","keywords":"Contingent valuation; Willingness to pay; Actuarial science; Valuation (finance); Health care; Economics; Cost–benefit analysis; Value (mathematics); Health economics; Public economics; Microeconomics; Statistics; Accounting; Mathematics","score_opus":0.41473583205410913,"score_gpt":0.38604142505608113,"score_spread":0.028694406998028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2158062010","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39981043,0.011972925,0.5779106,0.0047199097,0.0018321144,0.003142155,0.00029072064,0.000040754978,0.00028038808],"genre_scores_gemma":[0.9655579,0.002441021,0.029988509,0.0011862767,0.00012249225,0.0006330821,0.00003495595,0.000020464708,0.000015348889],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9984225,0.00001957968,0.0008498204,0.00036431782,0.00012679773,0.00021695714],"domain_scores_gemma":[0.99851453,0.0007481405,0.00043420892,0.00015635818,0.00006848714,0.00007828801],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012556262,0.00011773537,0.00040777648,0.00012843598,0.00018097252,0.000038918864,0.00013370765,0.00007111972,0.00009794313],"category_scores_gemma":[0.001553856,0.000108144035,0.00012352123,0.00009175677,0.000032726515,0.00014419806,0.00004610551,0.00006494058,0.00006013379],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003147834,0.0001390283,0.001883414,0.000058454218,0.000035264762,1.3365917e-7,0.003084076,0.0028938667,3.2128628e-7,0.032994024,0.00045560344,0.9584243],"study_design_scores_gemma":[0.010397727,0.008104147,0.07976622,0.0062088203,0.00008717432,0.0000042834095,0.00937514,0.39374068,0.0002375717,0.43674082,0.053951953,0.0013854721],"about_ca_topic_score_codex":0.0000065493377,"about_ca_topic_score_gemma":0.000025866877,"teacher_disagreement_score":0.9570389,"about_ca_system_score_codex":0.00027765887,"about_ca_system_score_gemma":0.000012636105,"threshold_uncertainty_score":0.44099858},"labels":[],"label_agreement":null},{"id":"W2158298440","doi":"10.1177/0272989x15585114","title":"The ONCOTYROL Prostate Cancer Outcome and Policy Model","year":2015,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto General Hospital; University of Toronto","funders":"Erasmus+; Erasmus Medisch Centrum","keywords":"Outcome (game theory); Prostate cancer; Cancer; Prostate; Medicine; Oncology; Internal medicine; Economics","score_opus":0.06896233147531967,"score_gpt":0.42756798301733,"score_spread":0.35860565154201035,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2158298440","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8677273,0.017990123,0.009361822,0.097431235,0.00084928895,0.0012523794,0.000035146175,0.00014564984,0.0052071004],"genre_scores_gemma":[0.9897778,0.0028735285,0.0015220918,0.004776784,0.0002576893,0.00013305564,0.0000021151861,0.000020978989,0.0006359577],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9981288,0.000027085434,0.0003369639,0.00025156848,0.0009567655,0.00029886165],"domain_scores_gemma":[0.99872226,0.000332272,0.0000690237,0.00025335382,0.000112744,0.0005103289],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00063414976,0.00013452848,0.00025016358,0.00006342156,0.00013832508,0.000051391387,0.000105304396,0.00008321865,0.0000375597],"category_scores_gemma":[0.001528697,0.000069880814,0.000053080767,0.00016289832,0.00012403104,0.00005145793,0.0001466427,0.00020469209,0.000020683598],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028568495,0.000065687476,0.077204496,0.000014851411,0.00003553218,0.000117970114,0.00030262905,0.00015109764,0.0000033763902,0.0012251554,0.008249234,0.9123443],"study_design_scores_gemma":[0.022648552,0.0018725679,0.13717881,0.004013899,0.0004344234,0.0004708952,0.000709142,0.45218456,0.00019721405,0.084006384,0.2954424,0.0008411393],"about_ca_topic_score_codex":0.00006280506,"about_ca_topic_score_gemma":0.00007533111,"teacher_disagreement_score":0.91150314,"about_ca_system_score_codex":0.000274166,"about_ca_system_score_gemma":0.0006611788,"threshold_uncertainty_score":0.2849657},"labels":[],"label_agreement":null},{"id":"W2158460222","doi":"10.1177/0272989x09341755","title":"The Relative Ability of Different Propensity Score Methods to Balance Measured Covariates Between Treated and Untreated Subjects in Observational Studies","year":2009,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":456,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Propensity score matching; Covariate; Statistics; Observational study; Inverse probability weighting; Weighting; Medicine; Mathematics","score_opus":0.4399813510196111,"score_gpt":0.5151726562683904,"score_spread":0.07519130524877926,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2158460222","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7554493,0.00016824409,0.24316281,0.0005098166,0.000028291845,0.00056280603,0.0000029697321,0.00006567833,0.00005006706],"genre_scores_gemma":[0.8561302,0.00006013981,0.14367136,0.00007800737,0.000018329423,0.000023359758,0.0000016265535,0.000008071373,0.000008924541],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9972337,0.00071466144,0.00076866354,0.000324212,0.0007173566,0.00024142506],"domain_scores_gemma":[0.97994345,0.019013537,0.00024061816,0.00031296772,0.00039647118,0.00009297119],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003171956,0.00019242802,0.0006584102,0.00008431691,0.00014236126,0.000014558373,0.00023361399,0.00015535303,0.000010949272],"category_scores_gemma":[0.060828973,0.00010899657,0.000043676704,0.00044508578,0.0001962963,0.0001081311,0.0001788584,0.0003462847,5.832123e-7],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00055043097,0.00020412156,0.51702327,0.000095796066,0.00013078097,0.000026381485,0.002262875,0.0000049782184,0.0043193307,0.07273739,0.00018060514,0.40246406],"study_design_scores_gemma":[0.00023666982,0.000121885496,0.48173454,0.0007924183,0.000017842185,7.8300275e-7,0.00003355814,0.00019830793,0.0022428038,0.51454157,0.000006351394,0.000073275405],"about_ca_topic_score_codex":0.000007983872,"about_ca_topic_score_gemma":0.0000727966,"teacher_disagreement_score":0.44180417,"about_ca_system_score_codex":0.00016392393,"about_ca_system_score_gemma":0.000049301027,"threshold_uncertainty_score":0.94708204},"labels":[],"label_agreement":null},{"id":"W2158784546","doi":"10.1177/0272989x03261566","title":"A Comparison of Narrative and Table Formats for Presenting Hypothetical Health States to Patients with Gastrointestinal or Pulmonary Disease","year":2004,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; McMaster University Medical Centre","funders":"","keywords":"Table (database); Presentation (obstetrics); Medicine; Narrative; Disease; Test (biology); Pulmonary disease; Health care; Family medicine; Surgery; Computer science; Pathology; Internal medicine; Data mining; Linguistics","score_opus":0.23508284086492112,"score_gpt":0.4712507837423802,"score_spread":0.2361679428774591,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2158784546","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6755295,0.00018154367,0.314214,0.0088713905,0.00008034843,0.00089470285,0.0001671098,0.000017096856,0.000044312812],"genre_scores_gemma":[0.91641724,0.0000057667207,0.080447674,0.0029359374,0.000058586895,0.00007374054,0.000023988183,0.000020813226,0.000016259273],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963398,0.00008404138,0.0024779015,0.0004045878,0.00028933547,0.00040432336],"domain_scores_gemma":[0.99610734,0.0019662401,0.0010440834,0.00021019265,0.000111283756,0.0005608489],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004616905,0.00014474892,0.00077569374,0.0001922945,0.00022957478,0.000046026948,0.00017752912,0.00004181949,0.00013475656],"category_scores_gemma":[0.011221998,0.00012683395,0.000042502976,0.0001707274,0.00008933853,0.00022663297,0.000096179814,0.00012558886,0.000023498584],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017461961,0.0010286176,0.9502527,0.0012659837,0.000032721484,0.000007740983,0.008380975,0.0038625903,3.2690713e-7,0.014903025,0.0045856335,0.013933471],"study_design_scores_gemma":[0.006803164,0.0026840884,0.71783483,0.007030622,0.000016504895,0.00003534718,0.006449244,0.16305584,0.0000030638223,0.090069816,0.005269292,0.0007481684],"about_ca_topic_score_codex":0.0000783061,"about_ca_topic_score_gemma":0.00008678831,"teacher_disagreement_score":0.24088773,"about_ca_system_score_codex":0.00020263251,"about_ca_system_score_gemma":0.00032077025,"threshold_uncertainty_score":0.9971069},"labels":[],"label_agreement":null},{"id":"W2162160016","doi":"10.1177/0272989x11426483","title":"Hic Sunt Dracones","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Psychology","score_opus":0.1363412555539356,"score_gpt":0.32751723684840156,"score_spread":0.19117598129446597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2162160016","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.104281664,0.0021402747,0.44869772,0.007941853,0.0032520876,0.0003626036,0.000023164894,0.00019068037,0.43310997],"genre_scores_gemma":[0.9898555,0.00015860164,0.0051746857,0.004243818,0.00017418143,0.000013690842,0.0000010523551,0.000014203171,0.0003642581],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9986584,0.00001736689,0.000601294,0.00030085683,0.000119106924,0.00030292882],"domain_scores_gemma":[0.99922603,0.0001441264,0.00013534055,0.00029351196,0.000015842352,0.00018516497],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0013197233,0.000094476374,0.00024381345,0.00020356337,0.000089449604,0.000023771741,0.00031622607,0.0001254463,0.00874794],"category_scores_gemma":[0.0011971287,0.000092721886,0.00007446238,0.00018007868,0.00004544532,0.00009488262,0.0001638385,0.00017536877,0.0028778163],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000013482915,0.000049209608,0.0041419845,0.000017320139,0.000010543302,0.000046623205,0.0003846542,7.395932e-7,1.4738649e-7,0.73857117,0.002870156,0.25389397],"study_design_scores_gemma":[0.00035931953,0.00006784979,0.045991413,0.00014154396,0.000002551617,0.0000092222945,0.000051956187,0.0021841007,0.000005863612,0.4811758,0.4698009,0.00020946458],"about_ca_topic_score_codex":0.00051052804,"about_ca_topic_score_gemma":0.00010509274,"teacher_disagreement_score":0.88557386,"about_ca_system_score_codex":0.000046695546,"about_ca_system_score_gemma":0.00002224389,"threshold_uncertainty_score":0.9978986},"labels":[],"label_agreement":null},{"id":"W2163103872","doi":"10.1177/0272989x07312712","title":"Do Patients' Communication Behaviors Provide Insight into Their Preferences for Participation in Decision Making?","year":2008,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Active listening; Patient participation; Psychology; Preference; Decision aids; Deference; Decision quality; Coding (social sciences); Social psychology; Applied psychology; MEDLINE; Medicine; Patient satisfaction; Nursing; Alternative medicine","score_opus":0.22766620629373663,"score_gpt":0.4864941687792341,"score_spread":0.2588279624854975,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2163103872","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9735736,0.001214692,0.019216089,0.00065908226,0.0009675553,0.0034093726,0.000024614665,0.00016910856,0.0007658864],"genre_scores_gemma":[0.9614986,0.00039925348,0.034110602,0.0014144343,0.00009734755,0.0022877986,0.00011754519,0.000057736303,0.000016663773],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9927481,0.0017775295,0.0024558723,0.0006748808,0.00158203,0.0007616058],"domain_scores_gemma":[0.9827866,0.013368348,0.000899242,0.001718996,0.00092367793,0.00030316098],"candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.0017292887,0.0003397594,0.0005888387,0.00053492206,0.0019286199,0.00003204687,0.0014727263,0.00068665214,0.00060124364],"category_scores_gemma":[0.016676225,0.0002762337,0.00013143636,0.00079721684,0.00024255266,0.0005159723,0.0009644844,0.0013518025,0.00014162957],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000537424,0.00039166064,0.5129773,0.00012210873,0.0000067884607,0.0000026642972,0.028886313,0.000033981578,0.000004325015,0.00049507926,0.0015989731,0.4549434],"study_design_scores_gemma":[0.0053277025,0.0005678658,0.88457173,0.011948179,0.000031544336,0.0000034254097,0.005020134,0.010876651,0.000021759777,0.04959379,0.031342443,0.000694796],"about_ca_topic_score_codex":0.00018530687,"about_ca_topic_score_gemma":0.0022293818,"teacher_disagreement_score":0.45424858,"about_ca_system_score_codex":0.00067474635,"about_ca_system_score_gemma":0.0007780922,"threshold_uncertainty_score":0.999969},"labels":[],"label_agreement":null},{"id":"W2167571049","doi":"10.1177/0272989x11434601","title":"The Impact of Explicit Values Clarification Exercises in a Patient Decision Aid Emerges After the Decision Is Actually Made","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":114,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Decision aids; Decision analysis; Management science; Psychology; Computer science; Medicine; Economics; Mathematics; Alternative medicine; Statistics","score_opus":0.13361205902914516,"score_gpt":0.47341729943167826,"score_spread":0.3398052404025331,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2167571049","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97427416,0.0113455765,0.008599748,0.0021171418,0.0015168701,0.0017908047,0.000037266724,0.000060836137,0.00025760016],"genre_scores_gemma":[0.9884934,0.0021966575,0.0067485697,0.0015143731,0.00021448455,0.00073665305,0.000009942261,0.000055196466,0.000030728363],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99167395,0.0015085483,0.0026280144,0.00047997292,0.0026498924,0.0010596444],"domain_scores_gemma":[0.9764912,0.019242937,0.0009659737,0.0023255388,0.00055354554,0.00042081997],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0039005624,0.00036194417,0.00055138377,0.00033911923,0.0013572408,0.000032589913,0.0016043639,0.0006244864,0.0017657782],"category_scores_gemma":[0.012251857,0.00018789871,0.0003036331,0.0009462094,0.00021015029,0.00043467822,0.0011682701,0.0016837007,0.00027801545],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024653263,0.00026968058,0.22229944,0.000046946156,0.000024155415,0.0000035018547,0.035411272,0.00004538471,0.000034348286,0.0002618813,0.017036945,0.7221011],"study_design_scores_gemma":[0.002163045,0.00040025156,0.90712595,0.0073427567,0.000046448426,0.000014449621,0.023636932,0.006785226,0.000079560064,0.029933572,0.021972913,0.0004989016],"about_ca_topic_score_codex":0.0003497327,"about_ca_topic_score_gemma":0.00022290525,"teacher_disagreement_score":0.7216022,"about_ca_system_score_codex":0.0005697789,"about_ca_system_score_gemma":0.00054236036,"threshold_uncertainty_score":0.99994284},"labels":[],"label_agreement":null},{"id":"W2171038270","doi":"10.1177/0272989x14535984","title":"Predictive Modeling of Implantation Outcome in an In Vitro Fertilization Setting","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Ovarian function and disorders","field":"Medicine","cited_by":92,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"","keywords":"In vitro fertilisation; Outcome (game theory); Human fertilization; Medicine; Computer science; Intensive care medicine; Biology; Pregnancy; Mathematics; Mathematical economics","score_opus":0.03068180435003499,"score_gpt":0.34883476275386266,"score_spread":0.3181529584038277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2171038270","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5565734,0.000013203216,0.44229463,0.00013526084,0.000117639465,0.00011312802,7.9800543e-7,0.000020620537,0.0007313522],"genre_scores_gemma":[0.9930193,0.0000042210286,0.0061832354,0.00069534633,0.000049780203,0.000006352911,0.000025609357,0.0000123573445,0.0000037908317],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9982095,0.00010846251,0.00064253446,0.00023052788,0.00064379897,0.00016520098],"domain_scores_gemma":[0.9992993,0.00029023524,0.00008583593,0.00015084358,0.0000703086,0.00010351863],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011925704,0.00008866239,0.00025679643,0.00038036078,0.000025164944,0.0000092113605,0.00006972763,0.00014583678,0.00013865135],"category_scores_gemma":[0.0033263771,0.00007697137,0.00003777275,0.00034375762,0.000021938677,0.00012727728,0.000030508572,0.00022150914,0.000006143487],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034936415,0.0003457395,0.3678895,0.00010650557,0.000007897608,0.000035125617,0.0011768743,0.02892721,0.0031003256,0.00048075817,0.000020478472,0.59441596],"study_design_scores_gemma":[0.002295324,0.00010995315,0.07991364,0.00040888655,0.000009421572,0.000010834715,0.0004028876,0.9148222,0.00008513443,0.0018675879,0.000012576374,0.00006152499],"about_ca_topic_score_codex":0.000022420958,"about_ca_topic_score_gemma":0.000060542767,"teacher_disagreement_score":0.885895,"about_ca_system_score_codex":0.000059020553,"about_ca_system_score_gemma":0.000060205126,"threshold_uncertainty_score":0.39822245},"labels":[],"label_agreement":null},{"id":"W2172241123","doi":"10.1177/0272989x04271040","title":"A Bayesian Approach to Net Health Benefits: An Illustration and Application to Modeling HIV Prevention","year":2004,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Institute of Mental Health; Agency for Healthcare Research and Quality","keywords":"Bayes' theorem; Quality-adjusted life year; Human immunodeficiency virus (HIV); Bayesian probability; Statistics; Sampling (signal processing); Cost–benefit analysis; Bayesian inference; Econometrics; Monte Carlo method; Medicine; Cost effectiveness; Actuarial science; Computer science; Mathematics; Economics; Filter (signal processing); Family medicine","score_opus":0.28559403579679826,"score_gpt":0.45349318421533275,"score_spread":0.16789914841853448,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2172241123","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08586233,0.00037350162,0.8882218,0.023749808,0.00013605994,0.0010464389,0.000029207027,0.00006156642,0.0005192967],"genre_scores_gemma":[0.8232185,0.000039154107,0.15969908,0.016444024,0.000325165,0.00017239062,0.000052948257,0.000028147331,0.000020547688],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9955464,0.00015655985,0.0027822712,0.0008159874,0.00031702768,0.00038173527],"domain_scores_gemma":[0.9979822,0.00020247017,0.00060209184,0.00045076982,0.000056276032,0.0007061657],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.013992449,0.00016973384,0.0005909806,0.000410919,0.00029380588,0.00012685377,0.00027217047,0.00018258189,0.00005798633],"category_scores_gemma":[0.0027258818,0.0002063369,0.00005055116,0.00034933104,0.00001993654,0.0004295712,0.00008501262,0.00017991093,0.00031795062],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000059926217,0.00034493042,0.0011377521,0.00019872811,0.000018580884,5.4856804e-7,0.0049095764,0.49568668,0.0000019557456,0.27970657,0.002145647,0.21578912],"study_design_scores_gemma":[0.0007521027,0.0002095468,0.0025316472,0.00047292287,0.0000026514053,0.000012378787,0.00075394736,0.9185608,5.052084e-7,0.07329093,0.003130985,0.00028157004],"about_ca_topic_score_codex":0.00028563192,"about_ca_topic_score_gemma":0.000335436,"teacher_disagreement_score":0.73735625,"about_ca_system_score_codex":0.0005581764,"about_ca_system_score_gemma":0.00021359893,"threshold_uncertainty_score":0.8414175},"labels":[],"label_agreement":null},{"id":"W2290249185","doi":"10.1177/0272989x16635130","title":"Expected Value of Sample Information with Imperfect Implementation","year":2016,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Advanced Statistical Process Monitoring","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"","keywords":"Value of information; Sample (material); Value (mathematics); Computer science; Imperfect; Statistics; Econometrics; Actuarial science; Mathematics; Economics; Artificial intelligence","score_opus":0.049879439113442656,"score_gpt":0.4413930489005117,"score_spread":0.391513609787069,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2290249185","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001126292,0.000045222903,0.9615369,0.035134092,0.00094325846,0.00035034333,0.0002564678,0.00006989011,0.0005375707],"genre_scores_gemma":[0.31420937,0.00008520564,0.34494278,0.3307621,0.008826051,0.00024790748,0.00048458044,0.00022039546,0.00022164955],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98584735,0.00033578463,0.0022564859,0.00068041286,0.010257563,0.0006223911],"domain_scores_gemma":[0.9704969,0.025988342,0.0014158951,0.0008644598,0.0010554936,0.0001789175],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0028420894,0.00039831406,0.0008396516,0.00081548066,0.00018881078,0.00022209951,0.0014787454,0.000642621,0.004563742],"category_scores_gemma":[0.03629213,0.00022626885,0.00015389896,0.0010008168,0.00026401004,0.001095699,0.00039238876,0.0011106946,0.00025020805],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008385105,0.0000050660888,0.00071138586,0.00003247214,0.00001908835,0.000058902944,0.00013579345,0.000021609612,0.0000048561715,0.0003943496,0.23499152,0.7635411],"study_design_scores_gemma":[0.0015703241,0.00040041996,0.0013517336,0.0018409336,0.000052156043,0.000050699127,0.00039273815,0.0015029666,0.00018867651,0.17386432,0.81814784,0.0006372156],"about_ca_topic_score_codex":0.00003029104,"about_ca_topic_score_gemma":0.000012811965,"teacher_disagreement_score":0.76290387,"about_ca_system_score_codex":0.00019146135,"about_ca_system_score_gemma":0.00041487502,"threshold_uncertainty_score":0.99634624},"labels":[],"label_agreement":null},{"id":"W2311411644","doi":"10.1177/0272989x16638312","title":"Analyzing Health-Related Quality of Life in the EVOLVE Trial","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Parathyroid Disorders and Treatments","field":"Medicine","cited_by":46,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"Amgen","keywords":"Cinacalcet; Medicine; Quality of life (healthcare); Placebo; Randomized controlled trial; Clinical trial; Health related quality of life; Hemodialysis; Secondary hyperparathyroidism; Physical therapy; Internal medicine; Alternative medicine","score_opus":0.08450794248199468,"score_gpt":0.4245980359837341,"score_spread":0.3400900935017394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2311411644","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9814306,0.0039339443,0.0077723116,0.005369873,0.00017886634,0.00039182176,0.0000030539693,0.000014801407,0.00090471597],"genre_scores_gemma":[0.99843055,0.00017192705,0.00045075928,0.0008310343,0.00005777845,0.00001233641,0.000002357562,0.000007267224,0.000035991536],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99724156,0.0003190199,0.0009966503,0.00021741942,0.0010011718,0.00022417115],"domain_scores_gemma":[0.99743456,0.0018000345,0.00024458364,0.00031718027,0.000039953484,0.00016366935],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028764643,0.00010062224,0.00051056885,0.00012377472,0.000050049035,0.0000059184968,0.00016103487,0.00011676625,0.0002773017],"category_scores_gemma":[0.007599208,0.000044717664,0.00013724701,0.00035635472,0.00008248742,0.000038665563,0.0000436498,0.00016198732,0.00004166287],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.03349486,0.001926039,0.19937555,0.000029941562,0.00015919747,0.0001944055,0.0009745779,0.0000026785497,0.00010666409,0.002209301,0.013617447,0.74790937],"study_design_scores_gemma":[0.24971807,0.002009357,0.7266808,0.0018517647,0.00006512733,0.000037621638,0.00052082125,0.0006417729,0.000012933416,0.016532516,0.0016734014,0.00025577634],"about_ca_topic_score_codex":0.00012906276,"about_ca_topic_score_gemma":0.00009319682,"teacher_disagreement_score":0.74765354,"about_ca_system_score_codex":0.00005969721,"about_ca_system_score_gemma":0.0002758058,"threshold_uncertainty_score":0.90975106},"labels":[],"label_agreement":null},{"id":"W2329193078","doi":"10.1177/0272989x15626397","title":"Design Features of Explicit Values Clarification Methods","year":2016,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":124,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; The Quebec Population Health Research Network; Cancer Care Ontario; Queen's University","funders":"","keywords":"Computer science; Publication; Taxonomy (biology); Data extraction; Management science; CINAHL; Selection (genetic algorithm); Multiple-criteria decision analysis; Information retrieval; MEDLINE; Data science; Operations research; Artificial intelligence; Mathematics","score_opus":0.5677519364902025,"score_gpt":0.625225247757608,"score_spread":0.057473311267405514,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2329193078","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[1.1727623e-7,0.64919186,0.34636757,0.00030314893,0.0013907838,0.0018533097,0.000034259345,0.0001045403,0.0007544272],"genre_scores_gemma":[0.000011207506,0.77934355,0.21847893,0.00046533026,0.0003604588,0.0009181774,0.000036787573,0.000095178555,0.00029037189],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9758826,0.017093498,0.0036828269,0.00078611507,0.0018546754,0.00070031104],"domain_scores_gemma":[0.94180524,0.051778544,0.0027867963,0.0026008624,0.0006444211,0.00038413485],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0070814155,0.0005211174,0.0024771392,0.0006467294,0.00072427816,0.000010539924,0.0023258482,0.0023334671,0.0020775606],"category_scores_gemma":[0.02879787,0.00033260926,0.000468881,0.0007652896,0.0001646346,0.00012960596,0.0010238129,0.0028719173,0.0005506508],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029574334,0.00003233093,0.0000047111243,0.008328651,0.000045798348,0.000002850609,0.0006915069,1.7480261e-7,6.388104e-7,0.0019170685,0.011534851,0.97741187],"study_design_scores_gemma":[0.00026758533,0.000055432854,0.000022253196,0.16681416,0.00013131485,0.00000803844,0.00023871839,0.000050457755,9.759616e-7,0.009560454,0.8225787,0.00027187058],"about_ca_topic_score_codex":0.00002179984,"about_ca_topic_score_gemma":0.0000042152437,"teacher_disagreement_score":0.97713995,"about_ca_system_score_codex":0.0005301208,"about_ca_system_score_gemma":0.0025361152,"threshold_uncertainty_score":0.9999126},"labels":[],"label_agreement":null},{"id":"W2336378046","doi":"10.1177/0272989x16643940","title":"Linked Sensitivity Analysis, Calibration, and Uncertainty Analysis Using a System Dynamics Model for Stroke Comparative Effectiveness Research","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan","funders":"National Center for Research Resources; Duke-NUS Medical School; U.S. Department of Veterans Affairs","keywords":"Robustness (evolution); Sensitivity (control systems); Uncertainty analysis; Calibration; Computer science; Stroke (engine); Statistics; Simulation; Mathematics; Engineering","score_opus":0.49337474827372163,"score_gpt":0.5373113945109718,"score_spread":0.04393664623725013,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2336378046","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39195263,0.00013092684,0.60582954,0.000901335,0.00009260164,0.0004931362,0.00053234934,0.000027131813,0.00004034027],"genre_scores_gemma":[0.98497725,0.000016068518,0.014387932,0.00030208688,0.00013230287,0.00008317433,0.000040545285,0.000018976669,0.00004166598],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9933962,0.0018656298,0.0027524,0.00095125765,0.0005193175,0.00051517994],"domain_scores_gemma":[0.97540385,0.022303712,0.0010296325,0.0005451873,0.00038881897,0.00032877972],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06144138,0.00021121104,0.0018295444,0.0016298358,0.00052997016,0.00014329459,0.00022647528,0.00030807187,0.00004842402],"category_scores_gemma":[0.012155472,0.00018774549,0.00033144877,0.0014258104,0.0002132745,0.00035188068,0.00015422543,0.00022778266,0.000020482492],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005319383,0.00019449057,0.19117144,0.00073680497,0.0047932947,0.0000129903165,0.002206271,0.56387866,0.00003057042,0.23126905,0.0005130728,0.004661418],"study_design_scores_gemma":[0.00076604844,0.00003533801,0.013299252,0.00033929438,0.00020610317,0.000002816676,0.0005647534,0.977508,0.0000016224095,0.007031813,0.00003198941,0.00021294014],"about_ca_topic_score_codex":0.0004224095,"about_ca_topic_score_gemma":0.0026686029,"teacher_disagreement_score":0.5930246,"about_ca_system_score_codex":0.0016604374,"about_ca_system_score_gemma":0.00027855436,"threshold_uncertainty_score":0.9961656},"labels":[],"label_agreement":null},{"id":"W2337644055","doi":"10.1177/0272989x16634085","title":"Effects of Design Features of Explicit Values Clarification Methods","year":2016,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":61,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; The Quebec Population Health Research Network","funders":"","keywords":"Regret; CINAHL; Congruence (geometry); Data extraction; Computer science; Context (archaeology); MEDLINE; Psychology; Management science; Social psychology; Psychological intervention; Machine learning","score_opus":0.40922831923926334,"score_gpt":0.5985183872905243,"score_spread":0.18929006805126097,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2337644055","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[8.334795e-7,0.7018486,0.29407966,0.00011204515,0.0011567551,0.002460335,0.00002365654,0.000058783746,0.0002593158],"genre_scores_gemma":[0.00006295532,0.8277514,0.17071512,0.00020099622,0.00021177424,0.0008475592,0.000018461793,0.00007917077,0.00011258103],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.976497,0.01711991,0.003528266,0.0006175666,0.0016980283,0.0005392473],"domain_scores_gemma":[0.90611684,0.08787402,0.0032635033,0.0018458209,0.0006214412,0.0002783476],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0054583773,0.00045009115,0.0027173012,0.0006405633,0.00037658008,0.000004579273,0.0019108318,0.0019641982,0.00048690362],"category_scores_gemma":[0.044314414,0.00028873287,0.0004996601,0.00073417276,0.00018325153,0.00009640632,0.00083617785,0.0020599968,0.000119005104],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003863931,0.000053236006,0.0000059903255,0.037022553,0.000058615537,0.0000025347458,0.0006836968,1.3487882e-7,0.0000054495194,0.0017175707,0.0038685335,0.956543],"study_design_scores_gemma":[0.00054004294,0.00013286075,0.00010791413,0.36209583,0.00029697895,0.0000052042333,0.00016968619,0.000040861436,0.000027546315,0.01001797,0.62626517,0.00029994684],"about_ca_topic_score_codex":0.000025001036,"about_ca_topic_score_gemma":0.0000022536306,"teacher_disagreement_score":0.9562431,"about_ca_system_score_codex":0.00032919494,"about_ca_system_score_gemma":0.0018916152,"threshold_uncertainty_score":0.9999565},"labels":[],"label_agreement":null},{"id":"W2338219775","doi":"10.1177/0272989x16644427","title":"Dissemination of Clinical Practice Guidelines","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":45,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; McMaster University","funders":"Centers for Disease Control and Prevention; Queensland Health; Canadian Diabetes Association","keywords":"Agency (philosophy); Guideline; Health care; Psychological intervention; Medicine; Target audience; Psychology; Medical education; Nursing; Business; Advertising; Sociology; Political science","score_opus":0.3661973846826643,"score_gpt":0.6576081472958724,"score_spread":0.2914107626132081,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2338219775","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1747928,0.00023718885,0.7100753,0.108690016,0.0017823464,0.00045063015,0.00001036607,0.00009516935,0.0038661507],"genre_scores_gemma":[0.7657922,0.00051835267,0.22449315,0.007952634,0.0009941043,0.0000074581976,0.0000055250634,0.00002355829,0.00021302895],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9926288,0.00024822107,0.0041105812,0.00040257387,0.0023957973,0.0002140368],"domain_scores_gemma":[0.94030005,0.054482583,0.0012268564,0.00062078773,0.0029930314,0.00037668648],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012442268,0.00013480308,0.00052602944,0.00012909327,0.00004183362,0.000011019623,0.00019853456,0.00026019267,0.0034046664],"category_scores_gemma":[0.7567164,0.000077885255,0.00022459315,0.00028899708,0.00015381089,0.00035736855,0.00017494759,0.00027470218,0.00017854181],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00087674585,0.00024462387,0.0076505654,0.000015615013,0.000036498284,0.000082996994,0.000013771957,3.2609066e-7,0.000416803,0.00047665785,0.062347017,0.9278384],"study_design_scores_gemma":[0.009534591,0.0020384092,0.17535865,0.013846734,0.00076524855,0.00094842183,0.00079206325,0.005965899,0.00073684077,0.007014312,0.782488,0.0005108511],"about_ca_topic_score_codex":0.0000089089845,"about_ca_topic_score_gemma":0.000008249888,"teacher_disagreement_score":0.9273275,"about_ca_system_score_codex":0.00004357617,"about_ca_system_score_gemma":0.00039048516,"threshold_uncertainty_score":0.9975064},"labels":[],"label_agreement":null},{"id":"W2339677920","doi":"10.1177/0272989x16636113","title":"Extent and Predictors of Decision Regret about Health Care Decisions","year":2016,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":292,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; Université Laval; University of Ottawa","funders":"","keywords":"Regret; MEDLINE; Health care; Psychology; Scale (ratio); Actuarial science; Medicine; Applied psychology; Computer science; Business","score_opus":0.2587543775411541,"score_gpt":0.5321878643264757,"score_spread":0.2734334867853216,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2339677920","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00005370556,0.9815213,0.0086671375,0.00049918634,0.0041826963,0.0036960472,0.00047297406,0.00017397813,0.0007329707],"genre_scores_gemma":[0.00055576384,0.9842921,0.012628519,0.00091874256,0.00059310807,0.00059597334,0.00013754236,0.00018372334,0.00009454277],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9830336,0.0040739467,0.006448545,0.0014036187,0.0037307362,0.0013095454],"domain_scores_gemma":[0.9502745,0.04047803,0.0036257806,0.0032758228,0.00093181763,0.0014140009],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0029598451,0.00087584456,0.004073851,0.0011277521,0.0016122509,0.000026924545,0.0023323465,0.002041341,0.0014700802],"category_scores_gemma":[0.031575322,0.0005645464,0.00066308654,0.0009996206,0.00044106704,0.00018279375,0.0031494258,0.0030686464,0.00038959476],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000065959706,0.00004290976,0.00030185236,0.009655109,0.000050694536,0.000018848892,0.001979223,1.3687082e-7,1.1193611e-8,0.0005408895,0.013353721,0.9739906],"study_design_scores_gemma":[0.0006599014,0.00016146264,0.00023636485,0.33345324,0.00007867334,0.000024963138,0.00087385444,0.000012703434,1.2018865e-8,0.0008846584,0.66331464,0.000299523],"about_ca_topic_score_codex":0.00007331227,"about_ca_topic_score_gemma":0.00021143883,"teacher_disagreement_score":0.9736911,"about_ca_system_score_codex":0.0013858981,"about_ca_system_score_gemma":0.006418146,"threshold_uncertainty_score":0.9996875},"labels":[],"label_agreement":null},{"id":"W2378256387","doi":"","title":"Should patients prescribed long-term low-dose aspirin receive proton pump inhibitors? Systematic review and meta-analysis","year":2014,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Antiplatelet Therapy and Cardiovascular Diseases","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medicine; Aspirin; Internal medicine; Clopidogrel; Observational study; Randomized controlled trial; Cohort study; Proton-pump inhibitor; Relative risk; Adverse effect; Cohort; Meta-analysis; Medical prescription; Intensive care medicine; Confidence interval; Pharmacology","score_opus":0.0390957876871832,"score_gpt":0.3338881069804447,"score_spread":0.2947923192932615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2378256387","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.70917994,0.21916306,0.052453723,0.0017141913,0.00035110587,0.01611826,0.00003449366,0.0003252706,0.00065992726],"genre_scores_gemma":[0.99059314,0.002832098,0.00025324314,0.005537197,0.00009424174,0.0004778997,0.000051091163,0.00004061445,0.00012048974],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.99507844,0.00062378525,0.0010697953,0.00061353826,0.0022965982,0.00031783085],"domain_scores_gemma":[0.99766403,0.00045723064,0.0002971547,0.00087759004,0.00024748445,0.00045649474],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0025526604,0.00033346209,0.0029481591,0.00022648319,0.00011616049,0.00005905512,0.00024137218,0.00021671766,0.001361004],"category_scores_gemma":[0.0039726514,0.00020254502,0.0018606463,0.0005901102,0.00010642579,0.0001192614,0.00015578358,0.00032539698,0.000059880735],"study_design_candidate":"meta_analysis","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010545314,0.0018614703,0.03865859,0.5244265,0.31493154,0.0008791815,0.00028204155,0.000018447692,0.000020474754,0.00009259453,0.0012803327,0.11649427],"study_design_scores_gemma":[0.003343911,0.00043302402,0.026966942,0.047418315,0.9197211,0.00010418248,0.000014993451,0.0007442518,0.0001128712,0.00018960454,0.00035472907,0.0005960791],"about_ca_topic_score_codex":0.000002305878,"about_ca_topic_score_gemma":0.0000032424034,"teacher_disagreement_score":0.60478956,"about_ca_system_score_codex":0.000039339033,"about_ca_system_score_gemma":0.000045873305,"threshold_uncertainty_score":0.9995519},"labels":[],"label_agreement":null},{"id":"W2400092153","doi":"10.1177/0272989x16646732","title":"The Role of Decision Models in Health Care Policy","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; Lawson Health Research Institute; Western University","funders":"Canadian Institutes of Health Research","keywords":"Conceptualization; Focus group; Medicaid; Decision aids; Decision model; Coding (social sciences); Health services research; Health care; Decision analysis; Medicine; Psychology; Computer science; Public health; Nursing; Alternative medicine; Artificial intelligence; Statistics; Political science; Business","score_opus":0.018019875677882803,"score_gpt":0.3560188477862532,"score_spread":0.3379989721083704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2400092153","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6716407,0.011455344,0.29842687,0.0071545662,0.000600771,0.0005807094,0.000006708024,0.000107373555,0.010026955],"genre_scores_gemma":[0.99650687,0.00047732066,0.0024779167,0.00032619972,0.00014828374,0.000010819834,4.1979348e-7,0.000012620827,0.00003953787],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9978442,0.000068951325,0.00051908457,0.00022283375,0.0010831337,0.00026176145],"domain_scores_gemma":[0.9979464,0.0013599149,0.000114067036,0.0003002317,0.00010111555,0.00017828358],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009324387,0.000092093884,0.0002613023,0.00021948178,0.00010170502,0.000009522793,0.00015967808,0.000114241506,0.00006391756],"category_scores_gemma":[0.0032077536,0.000045336794,0.00008615293,0.00043364652,0.00007697203,0.00005319254,0.00011069995,0.00017929793,0.000009676393],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020719569,0.000016771957,0.00077992195,0.000009491521,0.0000034259008,0.000009430332,0.00023691027,0.000027498805,0.0000514218,0.0009996704,0.00013852492,0.995655],"study_design_scores_gemma":[0.019562213,0.012901147,0.100901246,0.042128365,0.00004655169,0.00077822857,0.0050185206,0.112025425,0.0039773863,0.5396014,0.16216706,0.00089245813],"about_ca_topic_score_codex":0.00023776123,"about_ca_topic_score_gemma":0.00056184054,"teacher_disagreement_score":0.99476254,"about_ca_system_score_codex":0.0003212933,"about_ca_system_score_gemma":0.0005390743,"threshold_uncertainty_score":0.38402125},"labels":[],"label_agreement":null},{"id":"W2403992605","doi":"10.1177/0272989x16650181","title":"Underestimation of Variance of Predicted Health Utilities Derived from Multiattribute Utility Instruments","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Canadian Centre for Applied Research in Cancer Control; Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Statistics; Imputation (statistics); Bayesian probability; Econometrics; Standard error; Standard deviation; Population; Prior probability; Bayes' theorem; Mathematics; Missing data; Medicine","score_opus":0.3380654660943516,"score_gpt":0.45565146847164334,"score_spread":0.11758600237729172,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2403992605","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.47525278,0.00046760344,0.5152514,0.0070658782,0.00047886535,0.00029162,0.0010075395,0.000027832777,0.00015647478],"genre_scores_gemma":[0.9753361,0.00014508479,0.022573827,0.0017731988,0.000088303444,0.000017453383,0.0000311916,0.000015615218,0.000019224213],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9932317,0.00042439933,0.0051084147,0.00048613377,0.0004255913,0.0003237475],"domain_scores_gemma":[0.9915855,0.004447511,0.0030467915,0.0005699398,0.00012018029,0.0002300935],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011299516,0.00015555014,0.0010879337,0.00022453738,0.000098633194,0.000014790548,0.00039327826,0.00020637536,0.0024595612],"category_scores_gemma":[0.026525086,0.00014391735,0.000110281006,0.00019216714,0.00019212645,0.00033952144,0.00013212425,0.00012487381,0.000120718054],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033014856,0.0005232969,0.6033686,0.00084203266,0.00023816191,0.0000020044777,0.002633728,0.00004315769,0.000056756227,0.04254456,0.0085345255,0.34088305],"study_design_scores_gemma":[0.0028238103,0.00015578589,0.8068836,0.00325848,0.0000067189335,0.0000022142062,0.0002946092,0.050566357,0.00008347425,0.1324874,0.0031929656,0.00024462756],"about_ca_topic_score_codex":0.0008364448,"about_ca_topic_score_gemma":0.00015655342,"teacher_disagreement_score":0.5000833,"about_ca_system_score_codex":0.00031358656,"about_ca_system_score_gemma":0.00037987667,"threshold_uncertainty_score":0.9984523},"labels":[],"label_agreement":null},{"id":"W2415915394","doi":"10.1177/0272989x16651885","title":"Handling Regional Variation in Health State Preferences within a Country","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Hospital for Sick Children; McMaster University; Public Health Ontario; University of Toronto; University of Alberta","funders":"","keywords":"Estimator; Jurisdiction; Regional variation; Valuation (finance); Health care; Population; Estimation; Variation (astronomy); Computer science; Statistics; Econometrics; Medicine; Mathematics; Economics; Business; Political science; Environmental health; Economic growth","score_opus":0.33589835849134747,"score_gpt":0.4612256250656939,"score_spread":0.12532726657434645,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2415915394","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3592675,0.0015112374,0.5757286,0.06027995,0.0014547927,0.0005703843,0.0000781284,0.00006477308,0.0010446489],"genre_scores_gemma":[0.969476,0.00037142567,0.012670602,0.016840199,0.00030339835,0.00004817695,0.000006274261,0.000024166919,0.00025975777],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9937457,0.00039926026,0.00447197,0.0005717754,0.00040077578,0.00041052492],"domain_scores_gemma":[0.99270135,0.0047854353,0.0019040701,0.00031590753,0.000044357712,0.00024886458],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.03171364,0.00014549759,0.00069910975,0.00038855622,0.00014377171,0.00006407957,0.00031590814,0.0001588475,0.001288259],"category_scores_gemma":[0.019649355,0.00012648897,0.00005215831,0.0002660486,0.00006736494,0.00039951692,0.00006870604,0.0002013831,0.0008324608],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003685649,0.0004693706,0.26240915,0.00078856805,0.000107179156,0.000024171031,0.016006062,0.0014515483,0.000008751643,0.47249395,0.037362944,0.20850973],"study_design_scores_gemma":[0.003557803,0.0002141494,0.22647667,0.0073699118,0.000002281337,0.000025085248,0.00042264618,0.043980327,0.0000010599048,0.6386651,0.078647844,0.0006371776],"about_ca_topic_score_codex":0.0010155881,"about_ca_topic_score_gemma":0.0023082863,"teacher_disagreement_score":0.6102085,"about_ca_system_score_codex":0.000668853,"about_ca_system_score_gemma":0.00053891545,"threshold_uncertainty_score":0.9999455},"labels":[],"label_agreement":null},{"id":"W2416639700","doi":"10.1177/0272989x16662242","title":"Determinants of Change in the Cost-effectiveness Threshold","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Endowment; Publishing; Statement (logic); Independence (probability theory); Political science; Library science; Public relations; Accounting; Management; Economics; Law; Computer science","score_opus":0.5897847189267643,"score_gpt":0.5298265753517177,"score_spread":0.059958143575046585,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2416639700","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96979946,0.0009067945,0.01927431,0.007286715,0.0006919492,0.0009743015,0.00004832923,0.00001476725,0.00100336],"genre_scores_gemma":[0.9935602,0.00010681936,0.00037657452,0.0055108955,0.0002058198,0.00021360847,7.657506e-7,0.000013970449,0.000011330887],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99613696,0.0004206897,0.0024800685,0.000359151,0.00029706274,0.00030609206],"domain_scores_gemma":[0.9905395,0.008029756,0.0008451889,0.00046129004,0.00003283037,0.0000914489],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03490819,0.00011665762,0.0006787037,0.00025990838,0.00006576512,0.000020154015,0.00056090555,0.00016532135,0.0008688947],"category_scores_gemma":[0.014338497,0.000079084864,0.00008940225,0.0002318558,0.00009912396,0.00025394393,0.00009342655,0.00013195419,0.0005411208],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000047775695,0.00010477056,0.79961485,0.00013476473,0.000007127881,0.00001604975,0.0009964539,0.0000028040763,0.0000020682642,0.03212012,0.0011275519,0.1658257],"study_design_scores_gemma":[0.0013668,0.0000839687,0.9317231,0.0022838037,0.0000022518466,0.000015718228,0.00015379401,0.004047371,0.000009267314,0.05024102,0.009884864,0.00018803623],"about_ca_topic_score_codex":0.00011359937,"about_ca_topic_score_gemma":0.00027771812,"teacher_disagreement_score":0.16563766,"about_ca_system_score_codex":0.00018404641,"about_ca_system_score_gemma":0.000062213905,"threshold_uncertainty_score":0.99396414},"labels":[],"label_agreement":null},{"id":"W2433178472","doi":"10.1177/0272989x16654444","title":"Using Survival Analysis to Improve Estimates of Life Year Gains in Policy Evaluations","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"National Institute for Health and Care Research","keywords":"Life expectancy; Population; Medicine; Survival analysis; Cohort; Quality-adjusted life year; Demography; Years of potential life lost; Relative survival; Gerontology; Cost effectiveness; Surgery; Environmental health; Cancer registry; Internal medicine","score_opus":0.45798793401025295,"score_gpt":0.5515188494908664,"score_spread":0.09353091548061343,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2433178472","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4191405,0.00015544762,0.56764364,0.011524672,0.00034307217,0.00028549114,0.000124003,0.000018839191,0.00076431],"genre_scores_gemma":[0.96376556,0.000026122127,0.03314249,0.0027061854,0.00027276296,0.000023756687,0.000003388396,0.000021426955,0.00003830129],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99462223,0.00025933367,0.0037934133,0.0005176672,0.00042159416,0.00038574994],"domain_scores_gemma":[0.9929042,0.004972492,0.0010876036,0.00055083283,0.000111145084,0.00037375998],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.020468973,0.00014649561,0.0010765948,0.0017544748,0.00007989464,0.000032680942,0.0003797122,0.00017310499,0.0018344397],"category_scores_gemma":[0.096937515,0.0001373459,0.00017759624,0.0013633977,0.000070153044,0.00020893135,0.00015131927,0.00011025387,0.0005937595],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010397956,0.00027616654,0.8111701,0.00013286732,0.0004764552,0.000004754646,0.0015381892,0.0082636,0.00012150642,0.12772688,0.002102166,0.048083343],"study_design_scores_gemma":[0.0029415642,0.00014743255,0.5667726,0.001106035,0.00008309215,0.0000019050207,0.00066957274,0.31789902,0.000021039958,0.107224524,0.002426259,0.0007069964],"about_ca_topic_score_codex":0.0007693374,"about_ca_topic_score_gemma":0.00054961746,"teacher_disagreement_score":0.54462504,"about_ca_system_score_codex":0.00058180385,"about_ca_system_score_gemma":0.0005399407,"threshold_uncertainty_score":0.99907804},"labels":[],"label_agreement":null},{"id":"W2441281755","doi":"10.1177/0272989x16653397","title":"Why Do Health Economists Promote Technology Adoption Rather Than the Search for Efficiency? A Proposal for a Change in Our Approach to Economic Evaluation in Health Care","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vancouver Coastal Health Research Institute; University of British Columbia; Vancouver Coastal Health","funders":"","keywords":"Disinvestment; Health technology; Scarcity; Redress; Population health; Health economics; Health care; Economics; Argument (complex analysis); Value (mathematics); Public economics; Process (computing); Population; Emerging technologies; Actuarial science; Business; Risk analysis (engineering); Medicine; Computer science; Microeconomics; Political science; Economic growth","score_opus":0.31954636437603084,"score_gpt":0.50318503793997,"score_spread":0.1836386735639392,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2441281755","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21076532,0.002921883,0.33398873,0.4353846,0.0009875026,0.015447363,0.00028957764,0.0000617362,0.00015327804],"genre_scores_gemma":[0.95757025,0.00008728705,0.024406467,0.012587375,0.00048104263,0.0047687916,0.000024596666,0.00005595109,0.000018210558],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9935228,0.0005745861,0.0037108685,0.0010602933,0.00026068982,0.00087071204],"domain_scores_gemma":[0.9967633,0.0011414966,0.0011511885,0.00059410644,0.00010119449,0.0002487239],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.053300083,0.00023444972,0.0009986621,0.0011169488,0.00025906003,0.000079821315,0.00062338513,0.00028996027,0.00006386724],"category_scores_gemma":[0.0049607395,0.0001913305,0.00012358456,0.0003649178,0.000058375677,0.00026447425,0.00012410348,0.0002266025,0.00014124172],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005171036,0.0005234402,0.022376454,0.0015538781,0.000040880455,7.840966e-7,0.027032577,0.0020183688,0.0000018515678,0.18331344,0.012906799,0.74971443],"study_design_scores_gemma":[0.020685816,0.0033240374,0.041418646,0.007130961,0.000011972324,0.000039814302,0.023327839,0.6858361,0.000013279031,0.11511312,0.10133815,0.0017602872],"about_ca_topic_score_codex":0.0005580923,"about_ca_topic_score_gemma":0.0036523188,"teacher_disagreement_score":0.74795413,"about_ca_system_score_codex":0.004101154,"about_ca_system_score_gemma":0.0017557482,"threshold_uncertainty_score":0.99972194},"labels":[],"label_agreement":null},{"id":"W2463735421","doi":"10.1177/0272989x16655346","title":"Quantitative Framework for Retrospective Assessment of Interim Decisions in Clinical Trials","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Interim; Management science; Neglect; Clinical trial; Interim analysis; Set (abstract data type); Dimension (graph theory); Outcome (game theory); Probabilistic logic; Focus (optics); Computer science; Epistemology; Psychology; Medicine; Political science; Artificial intelligence; Economics; Mathematics; Law","score_opus":0.8337389509473252,"score_gpt":0.7541194232979128,"score_spread":0.07961952764941238,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2463735421","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.026657969,0.000028821136,0.96814704,0.0009834196,0.00240813,0.0011616374,0.00012336853,0.00004248669,0.00044710052],"genre_scores_gemma":[0.2725462,0.00012633686,0.72670186,0.00019781836,0.0002873598,0.00009335938,2.9899584e-7,0.000033778317,0.000013005449],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9827325,0.0065632476,0.0073135253,0.0008744065,0.002026045,0.0004903182],"domain_scores_gemma":[0.09947795,0.89707834,0.0017753255,0.0008309261,0.00048904534,0.00034844494],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10251212,0.0002672101,0.002819442,0.00026359968,0.00006130564,0.000027093147,0.000769648,0.0007779209,0.0019498039],"category_scores_gemma":[0.9722494,0.00015157404,0.00078078976,0.0004794307,0.00053475663,0.00008209972,0.00039633366,0.0007626604,0.000017824612],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007442065,0.00034572027,0.0033393197,0.000014217928,0.00005757794,0.000014866611,0.00002691746,1.673559e-7,0.000035547313,0.56010354,0.0009773131,0.43434063],"study_design_scores_gemma":[0.0034333253,0.0011178879,0.02069575,0.006726179,0.000070851806,0.0000022131924,0.000064742155,0.001482004,0.000045879166,0.96600586,0.00016631707,0.00018899766],"about_ca_topic_score_codex":0.0000012703431,"about_ca_topic_score_gemma":0.000013011111,"teacher_disagreement_score":0.8905151,"about_ca_system_score_codex":0.00018119384,"about_ca_system_score_gemma":0.00030985038,"threshold_uncertainty_score":0.9989625},"labels":[],"label_agreement":null},{"id":"W2481965157","doi":"10.1177/0272989x16660711","title":"Using Simulation to Model and Validate Invasive Breast Cancer Progression in Women in the Study and Control Groups of the Canadian National Breast Screening Studies I and II","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Toronto Metropolitan University; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; Health Canada","keywords":"Medicine; Breast cancer; Mammography; Cancer; Breast cancer screening; Gerontology; Internal medicine","score_opus":0.13096581636416382,"score_gpt":0.4384319024398323,"score_spread":0.3074660860756685,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2481965157","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99387586,0.00038550657,0.0013603589,0.0037596703,0.000027470594,0.0005589606,0.000017945706,0.0000029542732,0.000011290975],"genre_scores_gemma":[0.9981619,0.000030399682,0.0004166512,0.0013037478,0.0000458212,0.000033449003,1.6129985e-7,0.0000061076053,0.0000017739088],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9981206,0.00012298964,0.00030881882,0.0002340679,0.0009712552,0.00024226937],"domain_scores_gemma":[0.9988855,0.0006023468,0.000082166014,0.000097088996,0.00017217755,0.00016070297],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002025778,0.00010444559,0.000253587,0.0001733603,0.00020867876,0.000022537723,0.00009799064,0.000062030114,0.000009333792],"category_scores_gemma":[0.0014024067,0.000049472597,0.000014226572,0.0002764684,0.00012559484,0.00012725437,0.00015563422,0.0001392349,7.24999e-8],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007028759,0.000036632962,0.7006512,0.000024725694,0.000030452573,0.000037876787,0.006735536,0.004967941,0.00017468311,0.00004360742,0.000015534062,0.28657892],"study_design_scores_gemma":[0.0025614002,0.00011326141,0.91115403,0.0050674984,0.000023998426,0.000098054006,0.00299756,0.076582976,0.000002213356,0.0013151229,0.0000054823267,0.00007841377],"about_ca_topic_score_codex":0.0031273903,"about_ca_topic_score_gemma":0.065840945,"teacher_disagreement_score":0.2865005,"about_ca_system_score_codex":0.0002875564,"about_ca_system_score_gemma":0.0003161448,"threshold_uncertainty_score":0.951205},"labels":[],"label_agreement":null},{"id":"W2514144324","doi":"10.1177/0272989x16670616","title":"Using Discrete Choice Experiments with Duration to Model EQ-5D-5L Health State Preferences","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Duration (music); EQ-5D; Discrete choice; Econometrics; Medicine; Statistics; Computer science; Mathematics; Health related quality of life","score_opus":0.5593668936041863,"score_gpt":0.5074657157812797,"score_spread":0.0519011778229066,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2514144324","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.31620124,0.00025793666,0.668763,0.013548904,0.00028790248,0.0004521271,0.00006795772,0.000041159787,0.00037980435],"genre_scores_gemma":[0.90884787,0.00006623496,0.07842773,0.012090531,0.0002223551,0.00006656088,0.000005332163,0.0000388085,0.00023456216],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99465513,0.0002441784,0.003300647,0.0007411514,0.00050504226,0.0005538444],"domain_scores_gemma":[0.99606705,0.0013780468,0.0014432156,0.00048803742,0.0000721737,0.00055146357],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010078351,0.00021756257,0.00078519376,0.0003258119,0.00029041647,0.0001090273,0.0003929819,0.00011771978,0.00065309356],"category_scores_gemma":[0.006303702,0.00017007453,0.00006221787,0.00023215453,0.00006652957,0.00060239475,0.0001325115,0.00013976014,0.00061339355],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011137238,0.00077432726,0.22064117,0.000946399,0.0004278723,0.000027656948,0.022649348,0.035707276,0.0002259645,0.08197484,0.05797663,0.5775348],"study_design_scores_gemma":[0.0060678776,0.0011212401,0.043296218,0.00970307,0.000014643038,0.000034673627,0.0011648025,0.7771866,0.00007755075,0.12029018,0.039006628,0.0020365291],"about_ca_topic_score_codex":0.00029736882,"about_ca_topic_score_gemma":0.000383902,"teacher_disagreement_score":0.7414793,"about_ca_system_score_codex":0.00074511877,"about_ca_system_score_gemma":0.00043736756,"threshold_uncertainty_score":0.78841376},"labels":[],"label_agreement":null},{"id":"W2525874109","doi":"10.1177/0272989x16670638","title":"Understanding the Effects of Competition for Constrained Colonoscopy Services with the Introduction of Population-level Colorectal Cancer Screening","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Colonoscopy; Colorectal cancer; Medicine; Population; Cancer; Environmental health; Internal medicine","score_opus":0.03754278899380787,"score_gpt":0.318556062930486,"score_spread":0.28101327393667813,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2525874109","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5200205,0.00007852585,0.47664908,0.0025404554,0.00024040461,0.00042024683,0.000008137284,0.000018992321,0.000023708973],"genre_scores_gemma":[0.99677294,0.000023641926,0.0025900977,0.00014336802,0.0003680239,0.00006906099,0.000006963749,0.000013350534,0.000012532049],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9986012,0.000071060844,0.00029573328,0.00020029998,0.0006814532,0.00015027112],"domain_scores_gemma":[0.9967375,0.0026305304,0.0002612901,0.00014348283,0.00017501802,0.00005217758],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00063634693,0.00010619517,0.00025848567,0.00008769478,0.00018690925,0.000010848009,0.00010649974,0.000090897236,0.00008328066],"category_scores_gemma":[0.00077147887,0.000046264253,0.000076148666,0.00028453156,0.00023660487,0.00006506853,0.00003378271,0.000118502874,3.2097478e-7],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08057066,0.00021526383,0.04658201,0.0016966372,0.00067662675,0.000014569633,0.001795556,0.00060014013,0.051889326,0.012773242,0.0013077469,0.8018782],"study_design_scores_gemma":[0.03401762,0.03470521,0.7020931,0.041599624,0.0019647565,0.0005488308,0.0061597056,0.066486314,0.093102954,0.015673371,0.0026039497,0.0010445308],"about_ca_topic_score_codex":0.000086909386,"about_ca_topic_score_gemma":0.0005097499,"teacher_disagreement_score":0.8008337,"about_ca_system_score_codex":0.000121013094,"about_ca_system_score_gemma":0.00007472765,"threshold_uncertainty_score":0.18866014},"labels":[],"label_agreement":null},{"id":"W2526000330","doi":"","title":"Patients' outcomes and preferences for the induction and maintenance of anaesthesia for day surgery.","year":2002,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics","funders":"","keywords":"Medicine; Anesthesia; Regional anaesthesia","score_opus":0.040634499102576986,"score_gpt":0.30430405541188726,"score_spread":0.26366955630931027,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2526000330","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97455,0.0010887029,0.018728634,0.0043708556,0.00027387156,0.00072471524,0.0000060109064,0.0000192259,0.00023794748],"genre_scores_gemma":[0.9965159,0.0005755024,0.0018716078,0.0007524268,0.00006605716,0.000045876834,0.0000026243183,0.000010722389,0.00015926888],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99872756,0.00003780192,0.0003680128,0.00020495849,0.0004871903,0.00017447409],"domain_scores_gemma":[0.99090904,0.0086249625,0.00010305914,0.00015097354,0.00010421817,0.00010773905],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010740921,0.000109907356,0.0005532138,0.00007412836,0.00009429655,0.000017286411,0.00006145113,0.00011148822,0.00007248337],"category_scores_gemma":[0.00691537,0.000056109417,0.00019846366,0.00009307729,0.00015368388,0.000047741654,0.00002682923,0.00008767848,7.827727e-7],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011981329,0.000038614977,0.25879443,0.000043448148,0.000028851067,0.00000329684,0.000096391115,2.5076727e-7,0.0000012204954,0.0009398391,0.0010095465,0.7389243],"study_design_scores_gemma":[0.0015506047,0.00022414181,0.9303679,0.00038365673,0.000117547956,0.000031276842,0.00012996593,0.0016878378,0.000009510083,0.0010748368,0.064316995,0.00010576649],"about_ca_topic_score_codex":0.000003793263,"about_ca_topic_score_gemma":0.0000037058214,"teacher_disagreement_score":0.7388185,"about_ca_system_score_codex":0.000010218723,"about_ca_system_score_gemma":0.000015845995,"threshold_uncertainty_score":0.8278844},"labels":[],"label_agreement":null},{"id":"W2547171731","doi":"10.1177/0272989x16675338","title":"Toward Optimal Decision Making among Vulnerable Patients Referred for Cardiac Surgery: A Qualitative Analysis of Patient and Provider Perspectives","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Focus group; Autonomy; Medicine; Informed consent; Intensive care unit; Comprehension; Psychological intervention; Population; Qualitative research; Family medicine; Nursing; Intensive care medicine; Alternative medicine","score_opus":0.17077824414577805,"score_gpt":0.48298343375879954,"score_spread":0.31220518961302146,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2547171731","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9365815,0.0016946559,0.05753754,0.0008703161,0.000952515,0.0018471087,0.00028605893,0.000083792285,0.0001465303],"genre_scores_gemma":[0.9745891,0.00020444758,0.024126701,0.00022989162,0.00007667281,0.0006597586,0.000037211583,0.000051229687,0.000024948866],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9910454,0.00329298,0.002202975,0.0008556044,0.0018510318,0.0007520143],"domain_scores_gemma":[0.93250847,0.06328089,0.0012678324,0.0009442278,0.0016392836,0.0003592941],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0034714858,0.00033223236,0.0013027226,0.00096631347,0.0009380617,0.000023574627,0.0004695984,0.00046200617,0.00047672517],"category_scores_gemma":[0.05642887,0.00022757522,0.0004326518,0.0011230847,0.00032447278,0.00038801524,0.00071920746,0.000608342,0.00001187692],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018872085,0.00026240153,0.2982173,0.00034005046,0.00065107807,0.0000037873895,0.1499997,0.000035873443,0.000013127088,0.0010523256,0.004644053,0.5428931],"study_design_scores_gemma":[0.005423013,0.0012292803,0.5543872,0.02625671,0.001329378,0.0000012821322,0.37350738,0.014687423,0.000042740743,0.014135282,0.0073753637,0.001624928],"about_ca_topic_score_codex":0.00012508706,"about_ca_topic_score_gemma":0.00010105636,"teacher_disagreement_score":0.54126817,"about_ca_system_score_codex":0.0005377615,"about_ca_system_score_gemma":0.0004925679,"threshold_uncertainty_score":0.95151925},"labels":[],"label_agreement":null},{"id":"W2569472942","doi":"10.1177/0272989x16686559","title":"An Overview of R in Health Decision Sciences","year":2017,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Statistical Methods in Epidemiology","field":"Mathematics","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"National Cancer Institute","keywords":"Management science; Medicine; Computer science; Engineering","score_opus":0.8151315280256409,"score_gpt":0.7247683365535923,"score_spread":0.09036319147204863,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2569472942","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000023029672,0.6293886,0.36929464,0.000033325075,0.00062496023,0.00036200212,0.000015747655,0.000019718787,0.000258746],"genre_scores_gemma":[0.0000025640206,0.53303105,0.4667186,0.000108991706,0.00009040441,0.000017074331,0.0000015816312,0.000026685126,0.000003009189],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9890283,0.0032626102,0.0040163444,0.00097518426,0.0020510375,0.00066657003],"domain_scores_gemma":[0.89244527,0.10344961,0.0022680135,0.001413262,0.00007197139,0.00035187154],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.037929326,0.00044026374,0.005217974,0.000654162,0.00020800935,0.00004257251,0.0026706797,0.0008283145,0.00095175894],"category_scores_gemma":[0.3191987,0.0002906171,0.00041589516,0.00055865827,0.0007724642,0.00009700457,0.00061215955,0.0010654819,0.00003585364],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000006130184,0.00008313797,0.000011227922,0.007794289,0.000008546488,0.000046754343,0.000015278167,4.5397405e-7,2.7595575e-9,0.11559935,0.00065875426,0.87577605],"study_design_scores_gemma":[0.00011594006,0.00015412539,0.000027461543,0.14879556,0.00003437701,0.000054456166,0.000004417128,0.0006622438,9.337907e-9,0.54840076,0.30158663,0.00016402193],"about_ca_topic_score_codex":0.000027677117,"about_ca_topic_score_gemma":0.000068692425,"teacher_disagreement_score":0.8756121,"about_ca_system_score_codex":0.00018968689,"about_ca_system_score_gemma":0.0012073377,"threshold_uncertainty_score":0.9999615},"labels":[],"label_agreement":null},{"id":"W2575365675","doi":"10.1177/0272989x16683928","title":"Estimating State Transitions for Opioid Use Disorders","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Opioid Use Disorder Treatment","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AIDS Vancouver; Simon Fraser University; University of British Columbia","funders":"National Institute on Drug Abuse","keywords":"Heroin; Detoxification (alternative medicine); Hazard ratio; Opioid use disorder; Medicine; Methadone; Proportional hazards model; Prison; Opioid; Medical prescription; Internal medicine; Demography; Psychiatry; Psychology; Pharmacology; Drug","score_opus":0.023586691782234565,"score_gpt":0.3425481686181257,"score_spread":0.31896147683589116,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2575365675","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2962814,0.00009057645,0.6982777,0.004165156,0.00037977967,0.0005412375,0.000039514292,0.00010818183,0.000116436524],"genre_scores_gemma":[0.8296061,0.00008145994,0.1688945,0.00086296705,0.00012405423,0.00013129848,0.00001829704,0.000045503155,0.0002358166],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9979868,0.000030385825,0.00043798503,0.0003804723,0.00078536145,0.00037901284],"domain_scores_gemma":[0.9976701,0.0015620647,0.00006859465,0.0003321991,0.000086226544,0.00028079734],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003207435,0.0001811945,0.00030143547,0.00013845417,0.00015157042,0.000031846826,0.000119030905,0.00010612683,0.0006731982],"category_scores_gemma":[0.003111865,0.00010930688,0.00018100541,0.00016051061,0.0000911443,0.00014927237,0.000042552987,0.00011029035,0.000082533166],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002608456,0.0002554331,0.004020957,0.0000542295,0.00004998608,0.000036850244,0.00037871406,0.00002276523,0.00018981565,0.0002745853,0.0017467682,0.99270904],"study_design_scores_gemma":[0.066224985,0.0053304117,0.27852702,0.03554053,0.0013602829,0.00045394493,0.00090702676,0.20155151,0.001074879,0.25407574,0.15226343,0.0026902494],"about_ca_topic_score_codex":0.0000083083405,"about_ca_topic_score_gemma":0.00003473929,"teacher_disagreement_score":0.9900188,"about_ca_system_score_codex":0.00009367719,"about_ca_system_score_gemma":0.0001721325,"threshold_uncertainty_score":0.7371048},"labels":[],"label_agreement":null},{"id":"W2588254687","doi":"10.1177/0272989x17691954","title":"Experienced Probabilities Increase Understanding of Diagnostic Test Results in Younger and Older Adults","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":39,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Test (biology); Diagnostic test; Psychology; Medicine; Gerontology; Pediatrics","score_opus":0.20146806354432098,"score_gpt":0.4574371396016214,"score_spread":0.25596907605730046,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2588254687","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9896795,0.0010265358,0.00074478646,0.0033004312,0.0006372869,0.0013769558,0.000045021425,0.000048007307,0.0031414519],"genre_scores_gemma":[0.9972539,0.00024142105,0.0017513271,0.00040894613,0.00006551937,0.0002264644,0.0000068041563,0.000017392065,0.00002825507],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99653417,0.0005389452,0.0012627752,0.00039202988,0.0008564582,0.00041561126],"domain_scores_gemma":[0.95891714,0.038686223,0.000667784,0.0012684378,0.00019528046,0.0002651346],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0014342177,0.00014834545,0.00035137832,0.00016957324,0.0010879393,0.000024959503,0.000688895,0.00028372655,0.00021019304],"category_scores_gemma":[0.23497337,0.00012352504,0.000033248823,0.00012592059,0.0004661183,0.0003251727,0.000894741,0.0006822615,0.000013151077],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011702052,0.0002829347,0.8146823,0.0011221601,0.000007982414,0.00006540688,0.08226531,0.0000040446776,0.000012244461,0.007827926,0.0009310525,0.09162844],"study_design_scores_gemma":[0.0064560226,0.00020609041,0.8582935,0.037986077,0.000009325944,0.000006154008,0.0690289,0.004467679,0.000007691489,0.022805696,0.00043403014,0.00029884436],"about_ca_topic_score_codex":0.00078136916,"about_ca_topic_score_gemma":0.0009929339,"teacher_disagreement_score":0.23353915,"about_ca_system_score_codex":0.00028511192,"about_ca_system_score_gemma":0.00037662985,"threshold_uncertainty_score":0.83676594},"labels":[],"label_agreement":null},{"id":"W2604229464","doi":"10.1177/0272989x17700879","title":"Comparative Effectiveness of Tacrolimus-Based Steroid Sparing versus Steroid Maintenance Regimens in Kidney Transplantation: Results from Discrete Event Simulation","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Renal Transplantation Outcomes and Treatments","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Medicine; Regimen; Tacrolimus; Kidney transplantation; Adverse effect; Population; Internal medicine; Transplantation","score_opus":0.054981863078901884,"score_gpt":0.4029302360836915,"score_spread":0.3479483730047896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2604229464","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8777941,0.000044198034,0.1199752,0.00022105945,0.00047156753,0.0005294882,0.000183708,0.000021541679,0.0007591542],"genre_scores_gemma":[0.99689734,0.000048678616,0.0026248128,0.00009888191,0.000050250044,0.000017458824,0.00022966239,0.000016072501,0.000016845976],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974403,0.00019376252,0.00071076176,0.00044087812,0.0009714045,0.00024289411],"domain_scores_gemma":[0.9956974,0.0031026881,0.0003744014,0.0004985318,0.00010100053,0.00022598579],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007260643,0.00021279705,0.0006295016,0.00014036005,0.00014407659,0.00004083744,0.00021930758,0.00014010488,0.00006345319],"category_scores_gemma":[0.0009585137,0.00016223581,0.0001391622,0.000103638056,0.00011928918,0.00016868227,0.00003085799,0.00023397905,0.000013993025],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.3416222,0.0015072582,0.45904854,0.002566559,0.0010103962,0.005575265,0.0042812005,0.120336756,0.0032155842,0.00033361054,0.00003960919,0.060462993],"study_design_scores_gemma":[0.041008174,0.00048955757,0.863702,0.02326124,0.00015540414,0.000008893152,0.00004662433,0.0680728,0.0028142557,0.00022359303,0.000042578715,0.00017489439],"about_ca_topic_score_codex":0.0001287565,"about_ca_topic_score_gemma":0.00013645185,"teacher_disagreement_score":0.40465343,"about_ca_system_score_codex":0.00012360018,"about_ca_system_score_gemma":0.00013270177,"threshold_uncertainty_score":0.6615785},"labels":[],"label_agreement":null},{"id":"W2608521041","doi":"10.1177/0272989x17696999","title":"An Approach to Reconciling Competing Ethical Principles in Aggregating Heterogeneous Health Preferences","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"National Center for Advancing Translational Sciences; Social Sciences and Humanities Research Council of Canada; National Institutes of Health","keywords":"Preference; Social choice theory; Set (abstract data type); Social preferences; Psychological intervention; Aggregation problem; Quality (philosophy); Quality of life (healthcare); Psychology; Actuarial science; Management science; Social psychology; Economics; Computer science; Microeconomics","score_opus":0.5507894761827204,"score_gpt":0.5235710952411564,"score_spread":0.027218380941564035,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2608521041","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8589275,0.00058624014,0.11424834,0.01965942,0.0007706091,0.00067988463,0.000029414203,0.00007550946,0.005023094],"genre_scores_gemma":[0.89819264,0.00004216444,0.08503504,0.01620592,0.0004088764,0.00006214393,0.0000075475787,0.000031070966,0.00001457076],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99215,0.0007295442,0.0048838356,0.0010486705,0.0004802514,0.0007076887],"domain_scores_gemma":[0.99306995,0.0024688581,0.0026532346,0.0010891195,0.00006122351,0.0006576418],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.055230208,0.00023169286,0.00121363,0.00037334004,0.0009102965,0.0004347784,0.0013028046,0.00042997627,0.00026064666],"category_scores_gemma":[0.056608915,0.0002617953,0.00009906817,0.00013846191,0.00011491422,0.00035047106,0.0003211214,0.00090504636,0.00033088002],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001233967,0.00049864035,0.5174841,0.0009143568,0.0000599467,0.000035117177,0.019528985,0.021050567,0.0000027680474,0.082524955,0.0009956144,0.35678154],"study_design_scores_gemma":[0.0016432097,0.00032009993,0.17952898,0.0050883605,0.0000030582953,0.00006216519,0.0026820505,0.76954645,0.0000041954004,0.031103695,0.009058757,0.0009589826],"about_ca_topic_score_codex":0.00062158605,"about_ca_topic_score_gemma":0.00096374314,"teacher_disagreement_score":0.7484959,"about_ca_system_score_codex":0.00053789146,"about_ca_system_score_gemma":0.00036625282,"threshold_uncertainty_score":0.99998343},"labels":[],"label_agreement":null},{"id":"W2619539940","doi":"10.1177/0272989x17708564","title":"The Effect of Individual Movements and Interventions on the Spread of Influenza in Long-Term Care Facilities","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Infection Control and Ventilation","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"York University; University of Toronto; Sunnybrook Health Science Centre","funders":"Compute Canada","keywords":"Medicine; Psychological intervention; Isolation (microbiology); Infection control; Repeated measures design; Long-term care; Vaccination; Intervention (counseling); Emergency medicine; Analysis of variance; Outbreak; Demography; Intensive care medicine; Internal medicine; Nursing; Immunology; Virology; Statistics","score_opus":0.039827498520783336,"score_gpt":0.38861993972272335,"score_spread":0.34879244120194003,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2619539940","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9967558,0.0003891738,0.0015348698,0.00014029967,0.00020970407,0.0003162364,0.000007770283,0.000004641234,0.0006415254],"genre_scores_gemma":[0.9997973,0.000023750315,0.000008731339,0.000063565785,0.00003523192,0.00001634104,0.0000022254503,0.0000038481917,0.000048991667],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9987254,0.00009818069,0.0003637311,0.00010179041,0.0006150443,0.000095812335],"domain_scores_gemma":[0.9982903,0.0011365982,0.0001944229,0.00028823712,0.000057137673,0.000033322645],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012484465,0.00007148325,0.00019139027,0.00007619252,0.00021000438,0.00003567343,0.00017047429,0.00006898513,0.00012877183],"category_scores_gemma":[0.0039626816,0.000034691137,0.000106586915,0.000038314283,0.0001922553,0.000046926314,0.00013572325,0.00017876236,0.0000027158387],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00032010797,0.000022761697,0.55205667,0.000150653,0.000032417214,0.000005246142,0.0003652184,0.0000030710994,0.00001691204,0.0002257963,0.000028579738,0.44677258],"study_design_scores_gemma":[0.0016454932,0.00052995223,0.99337834,0.0033152145,0.000031592393,0.000003370232,0.000154775,0.00023174558,0.0003489347,0.00019342687,0.00013766422,0.000029486242],"about_ca_topic_score_codex":0.000023053559,"about_ca_topic_score_gemma":0.00013717399,"teacher_disagreement_score":0.4467431,"about_ca_system_score_codex":0.000017783486,"about_ca_system_score_gemma":0.000018749444,"threshold_uncertainty_score":0.47439864},"labels":[],"label_agreement":null},{"id":"W2625531829","doi":"10.1177/0272989x17711933","title":"The Importance of Considering Differences in Study Design in Network Meta-analysis: An Application Using Anti-Tumor Necrosis Factor Drugs for Ulcerative Colitis","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Golimumab; Randomized controlled trial; Adalimumab; Infliximab; Adaptive design; Medicine; Computer science; Clinical study design; Clinical trial; Internal medicine; Tumor necrosis factor alpha","score_opus":0.7447632777831494,"score_gpt":0.5500645395220185,"score_spread":0.19469873826113093,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2625531829","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5322383,0.000338571,0.4655301,0.00012868668,0.00008970599,0.0016373231,0.000009591053,0.0000023944326,0.00002533776],"genre_scores_gemma":[0.9770542,0.000012180096,0.022435255,0.000107906606,0.000047517286,0.00030281872,0.0000011278473,0.000011705979,0.000027292213],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.97398925,0.008817237,0.009332472,0.0013090154,0.0060945274,0.0004574808],"domain_scores_gemma":[0.95860213,0.029506449,0.00720303,0.003790204,0.0007264727,0.00017171624],"candidate_categories":["metaresearch","scholarly_communication"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.13231157,0.00033630198,0.005777971,0.00043260076,0.0007515682,0.0014194086,0.0031019987,0.00010855844,0.00085276394],"category_scores_gemma":[0.06966344,0.00014956188,0.0017352648,0.0015056446,0.00021330701,0.0004026383,0.00031823403,0.00023302865,0.00001102968],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001346996,0.00027026833,0.9330445,0.000019801113,0.008214985,0.000023932611,0.0023625852,0.0066619106,0.000013577053,0.0012045967,0.00019769445,0.04785146],"study_design_scores_gemma":[0.00039654825,0.000068097885,0.44184294,0.000055136057,0.006226253,0.0000017551235,0.0021004856,0.5271484,0.0000051817924,0.021893397,0.000076798984,0.00018501424],"about_ca_topic_score_codex":0.00009526233,"about_ca_topic_score_gemma":0.006129894,"teacher_disagreement_score":0.5204865,"about_ca_system_score_codex":0.000054863656,"about_ca_system_score_gemma":0.00014841887,"threshold_uncertainty_score":0.9996172},"labels":[],"label_agreement":null},{"id":"W2716137842","doi":"10.1177/0272989x17713456","title":"Taking Costs and Diagnostic Test Accuracy into Account When Designing Prevalence Studies: An Application to Childhood Tuberculosis Prevalence","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Tuberculosis; Test (biology); Tuberculin; Sample size determination; Statistics; Bayesian probability; Mathematics; Pathology","score_opus":0.05205744070992874,"score_gpt":0.42404309185694383,"score_spread":0.3719856511470151,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2716137842","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9354877,0.006197996,0.046261497,0.00956027,0.00024290875,0.0017640061,0.000010598207,0.00013083138,0.00034423167],"genre_scores_gemma":[0.9667384,0.008481127,0.020680252,0.0032242483,0.00049409084,0.00032104147,0.000005133909,0.000039379185,0.000016334312],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.995752,0.00025088224,0.00076982746,0.0010604566,0.0014610048,0.0007058258],"domain_scores_gemma":[0.9712485,0.025261212,0.0004368478,0.0015297878,0.00041315856,0.0011104787],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0038239828,0.00032006222,0.0006751524,0.00022256459,0.0009690504,0.00018942234,0.0009094632,0.00026731816,0.00021805275],"category_scores_gemma":[0.4439321,0.00024886106,0.00008640719,0.00015918765,0.00038034012,0.0005618953,0.0009909299,0.00064133754,0.00012092963],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009766261,0.00018608519,0.21415399,0.00038899833,0.000054522243,0.00011101733,0.0010171551,0.000008931049,0.0007795987,0.00006159562,0.0010815866,0.78205884],"study_design_scores_gemma":[0.0011304601,0.0008740888,0.9773825,0.009297551,0.00016340817,0.00028503907,0.00030296532,0.005539034,0.00019724137,0.003362592,0.0011016744,0.00036344264],"about_ca_topic_score_codex":0.00014531447,"about_ca_topic_score_gemma":0.00014197068,"teacher_disagreement_score":0.7816954,"about_ca_system_score_codex":0.00029052616,"about_ca_system_score_gemma":0.00017255139,"threshold_uncertainty_score":0.99999636},"labels":[],"label_agreement":null},{"id":"W2735790837","doi":"10.1177/0272989x17715636","title":"Adding Events to a Markov Model Using DICE Simulation","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare Operations and Scheduling Optimization","field":"Health Professions","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre","funders":"","keywords":"Dice; Computer science; Markov chain; Event (particle physics); Markov model; Flexibility (engineering); Software; Markov process; Transition (genetics); Discrete event simulation; Macro; Machine learning; Programming language; Simulation; Statistics; Mathematics","score_opus":0.19539266286184886,"score_gpt":0.5598629597718839,"score_spread":0.3644702969100351,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2735790837","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.27461177,0.000011822556,0.7214298,0.0012666964,0.00074519357,0.00042447084,0.000006611667,0.000047729656,0.0014559124],"genre_scores_gemma":[0.77296877,0.000009350086,0.22433499,0.002102059,0.00033536795,0.000025745512,0.0000054378784,0.000025468733,0.00019283396],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99715996,0.00026014933,0.00076117454,0.00036803194,0.0009940334,0.00045665362],"domain_scores_gemma":[0.9973068,0.0010293369,0.00028322055,0.0006357652,0.00034314435,0.00040171217],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0025925266,0.0001397559,0.00025911172,0.00016168556,0.0040784627,0.00006582701,0.00042510114,0.00033920744,0.0008355136],"category_scores_gemma":[0.015526149,0.000121683384,0.000053108557,0.0001420722,0.000023738212,0.00031550947,0.00037643313,0.00054562005,0.00017797039],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009929404,0.000031834166,0.010053646,0.000044418193,0.0000067825185,0.000010967947,0.00096266106,0.7057361,0.000053387917,0.00048843265,0.00046735682,0.2820451],"study_design_scores_gemma":[0.00041306295,0.00001819021,0.0043997345,0.0019899106,0.000008218008,0.0000010270112,0.00010573358,0.9910701,7.4837783e-7,0.0009457427,0.0009189177,0.00012861167],"about_ca_topic_score_codex":0.00007710666,"about_ca_topic_score_gemma":0.00012931936,"teacher_disagreement_score":0.498357,"about_ca_system_score_codex":0.00023758091,"about_ca_system_score_gemma":0.00058958406,"threshold_uncertainty_score":0.9972181},"labels":[],"label_agreement":null},{"id":"W2749887430","doi":"10.1177/0272989x17725748","title":"Conducting EQ-5D Valuation Studies in Resource-Constrained Countries: The Potential Use of Shrinkage Estimators to Reduce Sample Size","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Canadian Centre for Applied Research in Cancer Control; Institute for Clinical Evaluative Sciences; Hospital for Sick Children; Public Health Ontario; University of Toronto; SickKids Foundation; Sunnybrook Health Science Centre","funders":"","keywords":"Estimator; Shrinkage estimator; Mean squared error; Statistics; Standard error; Shrinkage; Mathematics; Econometrics; Sample size determination; Valuation (finance); Efficient estimator; Economics","score_opus":0.6963428736042208,"score_gpt":0.5128158556935778,"score_spread":0.18352701791064296,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2749887430","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9182126,0.00050832116,0.04979187,0.028996384,0.0013160233,0.00078388525,0.000104945466,0.000028934608,0.0002570538],"genre_scores_gemma":[0.9688909,0.00007148888,0.025312692,0.0052601066,0.00032325453,0.000052363594,0.0000032869748,0.000024847614,0.00006108468],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9941657,0.00042222635,0.0039143604,0.00056995434,0.00051559194,0.00041215014],"domain_scores_gemma":[0.9694431,0.025965197,0.0029928759,0.0011852451,0.00019279681,0.00022077479],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.034035083,0.00019152366,0.0009807043,0.00027233802,0.0006554595,0.00022454541,0.00073980243,0.00018398053,0.000550228],"category_scores_gemma":[0.3547785,0.00017907786,0.00011490826,0.00017148303,0.00042799627,0.0005049908,0.00033957977,0.0002959675,0.00016811206],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011789219,0.0007504379,0.36629066,0.0021986424,0.0010075012,0.0001660921,0.067084275,0.035752214,0.00014611111,0.23165444,0.13079116,0.16297956],"study_design_scores_gemma":[0.007872745,0.00049088727,0.33248287,0.010221995,0.00008732349,0.000100726145,0.026719518,0.23739962,0.000095271935,0.28722942,0.09534958,0.0019500465],"about_ca_topic_score_codex":0.00051101256,"about_ca_topic_score_gemma":0.00021599016,"teacher_disagreement_score":0.3207434,"about_ca_system_score_codex":0.00034406933,"about_ca_system_score_gemma":0.00022684995,"threshold_uncertainty_score":0.99466413},"labels":[],"label_agreement":null},{"id":"W2756603975","doi":"10.1177/0272989x17732990","title":"A Systematic Review and Meta-analysis of Childhood Health Utilities","year":2017,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":73,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children","funders":"University of Warwick; Economic and Social Research Council; National Institute for Health and Care Research; Advantage West Midlands","keywords":"EconLit; PsycINFO; Respondent; CINAHL; Medicine; Systematic review; Health economics; Economic evaluation; Valuation (finance); MEDLINE; Cost–utility analysis; Population; Meta-analysis; Actuarial science; Cochrane Library; Public health; Psychological intervention; Cost effectiveness; Environmental health; Nursing; Economics; Finance","score_opus":0.7321214391074573,"score_gpt":0.5889026381639694,"score_spread":0.14321880094348793,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2756603975","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[2.1165658e-7,0.9881821,0.0031482785,0.004158179,0.00018083953,0.0028781635,0.0009760835,0.000025139776,0.00045095198],"genre_scores_gemma":[0.000015674628,0.9876227,0.0012938605,0.010175678,0.00006232203,0.0004363553,0.00005998735,0.00004876344,0.00028469812],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9790195,0.0018461283,0.017091583,0.0010201936,0.000566834,0.0004557523],"domain_scores_gemma":[0.9701248,0.008816068,0.018442962,0.002108732,0.000089678586,0.00041776392],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07156291,0.0005356305,0.032693822,0.0013398723,0.00028288979,0.00012435208,0.001182563,0.00044497347,0.0032088528],"category_scores_gemma":[0.06347053,0.00043735714,0.0035330378,0.0005615896,0.00016404671,0.00020120032,0.00031492603,0.0004950745,0.0003753755],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[3.7031862e-7,0.00004215823,0.0000020725308,0.9152216,0.057736687,0.0000019726292,0.00016793651,3.004131e-7,4.91672e-12,0.0071530263,0.0030163752,0.016657552],"study_design_scores_gemma":[0.00012973625,0.00004881434,0.00002048535,0.336124,0.315258,0.000033840348,0.000060901326,0.00019862156,1.8040835e-10,0.0029624586,0.34463978,0.00052338175],"about_ca_topic_score_codex":0.00012241681,"about_ca_topic_score_gemma":0.000083950516,"teacher_disagreement_score":0.57909757,"about_ca_system_score_codex":0.0003083499,"about_ca_system_score_gemma":0.00069611316,"threshold_uncertainty_score":0.99980783},"labels":[],"label_agreement":null},{"id":"W2767104228","doi":"10.1177/0272989x17738754","title":"How Should Discrete Choice Experiments with Duration Choice Sets Be Presented for the Valuation of Health States?","year":2017,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Health and Medical Research Council; Medical Research Council; EuroQol Research Foundation; University of Technology Sydney","keywords":"Valuation (finance); Logit; Discrete choice; Mixed logit; Econometrics; Dimension (graph theory); Population; Consistency (knowledge bases); EQ-5D; Actuarial science; Multiple choice; Logistic regression; Statistics; Medicine; Computer science; Mathematics; Economics; Significant difference; Artificial intelligence; Environmental health","score_opus":0.5874022463750577,"score_gpt":0.543286751780675,"score_spread":0.04411549459438269,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2767104228","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10656534,0.0014066525,0.67628473,0.21274294,0.00081629003,0.0018184597,0.00017054759,0.000029711848,0.00016532917],"genre_scores_gemma":[0.98490965,0.00012834903,0.0074663134,0.0065936837,0.0003175276,0.00026367637,0.00006018704,0.00003673356,0.00022386103],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9955004,0.00030242527,0.0026955665,0.00052431226,0.00059340737,0.0003838988],"domain_scores_gemma":[0.9884768,0.0057728477,0.0044481945,0.0009676602,0.00016188996,0.00017256994],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02045074,0.00018479532,0.0007605595,0.00014946768,0.0009760475,0.00032601823,0.0007021565,0.00014609548,0.00013982887],"category_scores_gemma":[0.037266955,0.00014698085,0.00009880161,0.00009388393,0.00016000704,0.00073213544,0.00012425739,0.0001911058,0.000020753645],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009672259,0.00091484515,0.36275625,0.0025533857,0.0010817738,0.0000049672367,0.015491138,0.0045168977,0.000033595097,0.08819846,0.16465333,0.35882813],"study_design_scores_gemma":[0.0045977514,0.0005158388,0.31713957,0.0013606785,0.000025904297,0.0000055498185,0.0017397827,0.53087705,0.00003676269,0.018544106,0.12467195,0.00048503545],"about_ca_topic_score_codex":0.00077686744,"about_ca_topic_score_gemma":0.00061345723,"teacher_disagreement_score":0.8783443,"about_ca_system_score_codex":0.00026910764,"about_ca_system_score_gemma":0.00023723538,"threshold_uncertainty_score":0.97084254},"labels":[],"label_agreement":null},{"id":"W2793717112","doi":"10.1177/0272989x12455462","title":"Modeling Using Discrete Event Simulation","year":2012,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":337,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Context (archaeology); Discrete event simulation; Relevance (law); Process (computing); Management science; Health care; Event (particle physics); Task (project management); Data science; Representation (politics); Process modeling; Risk analysis (engineering); Operations research; Process management; Systems engineering; Medicine; Simulation; Work in process; Engineering; Operations management","score_opus":0.5191837500212184,"score_gpt":0.5402170787702313,"score_spread":0.02103332874901287,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2793717112","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29471397,0.0013134959,0.70065445,0.0014388725,0.0010475989,0.00019516575,0.000012461565,0.000039159728,0.0005848429],"genre_scores_gemma":[0.9709588,0.000022387641,0.023109613,0.0047774115,0.0010466103,0.000012217381,0.0000067297215,0.00003314317,0.000033067518],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9950111,0.0002164607,0.003459603,0.0003843148,0.00038503093,0.000543521],"domain_scores_gemma":[0.99662375,0.0017735712,0.00077215163,0.00039806086,0.000046553374,0.00038592512],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.021942656,0.00016400544,0.00062445947,0.00028270445,0.00025152485,0.00006415324,0.0002641741,0.00022228544,0.002010871],"category_scores_gemma":[0.015348657,0.00018451584,0.00013874333,0.00019936012,0.000033116954,0.0006965877,0.000129377,0.0002378485,0.0013996855],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000035262623,0.0001265473,0.037365425,0.00013117267,0.00004674759,0.000002885517,0.0017633789,0.8963388,0.0000022066847,0.046541315,0.0010736269,0.016572595],"study_design_scores_gemma":[0.00030893565,0.000012045492,0.0016213607,0.00032156523,0.0000044480967,0.0000059432086,0.00017265593,0.9763188,3.4738125e-7,0.016346399,0.004683248,0.00020430445],"about_ca_topic_score_codex":0.00009036573,"about_ca_topic_score_gemma":0.000012136932,"teacher_disagreement_score":0.67754483,"about_ca_system_score_codex":0.00047893848,"about_ca_system_score_gemma":0.000081487895,"threshold_uncertainty_score":0.99937785},"labels":[],"label_agreement":null},{"id":"W2794157321","doi":"10.1177/0272989x17711927","title":"The University of Wisconsin Breast Cancer Epidemiology Simulation Model: An Update","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":54,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"National Cancer Institute","keywords":"Breast cancer; Epidemiology; Medicine; Computer science; Cancer; Internal medicine","score_opus":0.13016724415674102,"score_gpt":0.4439719947517315,"score_spread":0.31380475059499047,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2794157321","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4341694,0.00047859512,0.55835944,0.0060030175,0.00029780678,0.00012513313,0.0000112662565,0.000036686983,0.0005186374],"genre_scores_gemma":[0.9908415,0.00011848048,0.0068483613,0.0019087257,0.00023628432,3.2996485e-7,0.00000236089,0.000006382153,0.000037594167],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9986874,0.00011704403,0.00027458995,0.00021353392,0.00048750822,0.00021995704],"domain_scores_gemma":[0.9984057,0.0007684495,0.0001238319,0.00027671084,0.0002484607,0.00017686027],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014526483,0.0000765009,0.00023395913,0.00003928094,0.00023193228,0.000004588294,0.0002045737,0.00012759825,0.0008354628],"category_scores_gemma":[0.0011585681,0.000051181298,0.00006011066,0.00014217371,0.0004396861,0.0000974245,0.00009669534,0.00017516402,0.000029603112],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015385803,0.000026084928,0.025058178,0.0000090385065,0.000026519563,0.00003471408,0.00017999094,0.014642937,0.000039799743,0.0018196495,0.0051202923,0.95150423],"study_design_scores_gemma":[0.00047721944,0.00011120529,0.035230152,0.000518838,0.000036972844,0.00003304738,0.00013714205,0.95043904,0.000006388111,0.006531879,0.0064168544,0.00006127216],"about_ca_topic_score_codex":0.00013783143,"about_ca_topic_score_gemma":0.00051608175,"teacher_disagreement_score":0.95144296,"about_ca_system_score_codex":0.000069410526,"about_ca_system_score_gemma":0.00022599055,"threshold_uncertainty_score":0.91477317},"labels":[],"label_agreement":null},{"id":"W2794982991","doi":"10.1177/0272989x18765185","title":"Individual Value Clarification Methods Based on Conjoint Analysis: A Systematic Review of Common Practice in Task Design, Statistical Analysis, and Presentation of Results","year":2018,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; The Quebec Population Health Research Network","funders":"National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Institutes of Health","keywords":"Conjoint analysis; PsycINFO; Preference; Applied psychology; Task (project management); Decision aids; Set (abstract data type); Psychology; Scopus; Systematic review; MEDLINE; Preference elicitation; Medicine; Computer science; Alternative medicine; Statistics; Engineering","score_opus":0.3675892562518126,"score_gpt":0.6074671021230458,"score_spread":0.23987784587123318,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2794982991","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[8.1101143e-7,0.6534868,0.3413926,0.00022249442,0.0001131593,0.0044050324,0.00026647787,0.000018910754,0.00009375693],"genre_scores_gemma":[0.00075319014,0.79841065,0.19790448,0.00086967973,0.000029149778,0.0009404258,0.0010549134,0.000034463206,0.000003018998],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.92719847,0.06003497,0.0087386975,0.00084877864,0.0027764968,0.000402564],"domain_scores_gemma":[0.8530836,0.13497274,0.00829137,0.0023034893,0.0010767909,0.0002720337],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032880645,0.00045284803,0.0058708894,0.0025435998,0.00023725544,0.000019047571,0.0010755549,0.0009929708,0.00013186612],"category_scores_gemma":[0.18717161,0.0003402166,0.00051266205,0.005723182,0.00025960925,0.00013104867,0.00042008224,0.0019419324,0.000015799525],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002150696,0.0002992952,0.000088791385,0.83501524,0.0017375125,0.000008500879,0.0012290976,0.000045833793,4.0851447e-8,0.00060821226,0.00036072443,0.1603917],"study_design_scores_gemma":[0.00069544296,0.00026514122,0.00045870678,0.92969966,0.034121763,0.000002820354,0.0005761158,0.029156841,2.1755928e-7,0.0003580211,0.0043818494,0.00028341747],"about_ca_topic_score_codex":0.00026414325,"about_ca_topic_score_gemma":0.00010401503,"teacher_disagreement_score":0.16010828,"about_ca_system_score_codex":0.00037164113,"about_ca_system_score_gemma":0.0016634986,"threshold_uncertainty_score":0.999905},"labels":[],"label_agreement":null},{"id":"W2795262817","doi":"10.1177/0272989x18754513","title":"Microsimulation Modeling for Health Decision Sciences Using R: A Tutorial","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"demographic modeling and climate adaptation","field":"Decision Sciences","cited_by":180,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"National Institute of Allergy and Infectious Diseases; National Cancer Institute; National Institutes of Health; University of Minnesota","keywords":"Microsimulation; Computer science; Field (mathematics); Vectorization (mathematics); Decision model; Software; Markov model; Code (set theory); Markov chain; Machine learning; Operations research; Programming language; Engineering; Set (abstract data type); Mathematics","score_opus":0.30036390565179083,"score_gpt":0.5423550754257039,"score_spread":0.2419911697739131,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2795262817","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3538059,0.00013842517,0.6425299,0.0003911049,0.0026950869,0.0002407292,0.0000069297134,0.000058309215,0.00013362276],"genre_scores_gemma":[0.75516427,0.00001845477,0.24252532,0.0011205947,0.0011318789,0.000007387734,0.0000026464797,0.000019309635,0.000010134765],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9908242,0.00020013699,0.0018668799,0.0010998101,0.005349817,0.0006591151],"domain_scores_gemma":[0.99166745,0.0058890497,0.0005198416,0.00056583487,0.001020021,0.00033782533],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.019839456,0.00023993067,0.00050838926,0.0008444173,0.0017540323,0.000673238,0.0011206635,0.00025225582,0.00038875494],"category_scores_gemma":[0.023019059,0.00017553878,0.00026012372,0.0018000506,0.00034216492,0.0005313166,0.00025129662,0.00020330609,0.000092764916],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004084747,0.00005830046,0.00045257367,0.0000056216118,0.0000068740214,0.000002032878,0.0004466533,0.08067965,0.00015858423,0.0026802444,0.0020246617,0.91307634],"study_design_scores_gemma":[0.0007042786,0.00025950896,0.00008118025,0.00041425912,0.0000074222376,0.000013340673,0.00020575838,0.8156701,0.000010504726,0.17987113,0.00259913,0.0001634261],"about_ca_topic_score_codex":0.0000403676,"about_ca_topic_score_gemma":0.00011358882,"teacher_disagreement_score":0.9129129,"about_ca_system_score_codex":0.000112939364,"about_ca_system_score_gemma":0.0005893122,"threshold_uncertainty_score":0.9995456},"labels":[],"label_agreement":null},{"id":"W2798441388","doi":"10.1177/0272989x19837983","title":"Estimating the Expected Value of Sample Information across Different Sample Sizes Using Moment Matching and Nonlinear Regression","year":2019,"lang":"en","type":"preprint","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"Engineering and Physical Sciences Research Council; Boehringer Ingelheim","keywords":"Sample size determination; Sample (material); Matching (statistics); Computer science; Moment (physics); Bayesian probability; Econometrics; Statistics; Mathematics; Artificial intelligence","score_opus":0.24019276851531318,"score_gpt":0.4735494492719988,"score_spread":0.23335668075668564,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2798441388","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5177055,0.0005202009,0.4781279,0.0014241347,0.001224664,0.00053015875,0.0004181979,0.000025422029,0.000023842358],"genre_scores_gemma":[0.7703145,0.000101614845,0.22681619,0.0021026551,0.00043836003,0.00004760043,0.00013632364,0.000038528786,0.0000042350757],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9921861,0.00040543263,0.005670672,0.0006061229,0.0006666132,0.00046504673],"domain_scores_gemma":[0.98303753,0.010653761,0.005081312,0.0009081065,0.00012573159,0.00019353502],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.015246057,0.00036415007,0.0015231178,0.0002844663,0.00041743682,0.00028655006,0.0006390369,0.000497979,0.00035145684],"category_scores_gemma":[0.02547449,0.00029891066,0.0001933659,0.00015062091,0.00012624689,0.00039776,0.0014361679,0.0007821489,0.000047307767],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040822502,0.00054115115,0.16649748,0.0123554645,0.0006835575,0.0000062260297,0.0857939,0.43819168,0.000016768368,0.038855527,0.0030882112,0.2535618],"study_design_scores_gemma":[0.00062220666,0.000034068977,0.011731341,0.004740835,0.000012880591,0.000006220657,0.0014597401,0.91656804,0.0000038369003,0.063519396,0.0009795206,0.0003219148],"about_ca_topic_score_codex":0.0014957299,"about_ca_topic_score_gemma":0.000053198164,"teacher_disagreement_score":0.47837636,"about_ca_system_score_codex":0.0005115378,"about_ca_system_score_gemma":0.00021070475,"threshold_uncertainty_score":0.9999463},"labels":[],"label_agreement":null},{"id":"W2811232920","doi":"10.1177/0272989x18776637","title":"Estimation of a Preference-Based Summary Score for the Patient-Reported Outcomes Measurement Information System: The PROMIS <sup>®</sup> -Preference (PROPr) Scoring System","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":160,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"National Institutes of Health; National Cancer Institute; Social Sciences and Humanities Research Council of Canada; National Center for Advancing Translational Sciences; National Institute of Arthritis and Musculoskeletal and Skin Diseases; Riksbankens Jubileumsfond","keywords":"Patient-Reported Outcomes Measurement Information System; Preference; Quality of life (healthcare); Medicine; Applied psychology; Physical therapy; Psychology; Statistics; Psychometrics; Clinical psychology; Computerized adaptive testing; Mathematics","score_opus":0.4962130764892222,"score_gpt":0.4115047991206986,"score_spread":0.0847082773685236,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2811232920","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.24227144,0.00012312867,0.7539767,0.0002486494,0.0011062952,0.0014384729,0.000011718025,0.00011060956,0.0007129926],"genre_scores_gemma":[0.96440214,0.0000026564983,0.035108503,0.00017201717,0.00011500944,0.0001760967,0.000003108422,0.000013269215,0.0000071864038],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98797584,0.0007443507,0.0027510158,0.0005412316,0.0075272885,0.0004602872],"domain_scores_gemma":[0.9512224,0.042038865,0.0022747053,0.0016750629,0.0026037479,0.00018520639],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.027672037,0.00028888395,0.00064665644,0.0005532342,0.0008138057,0.0004066258,0.0018731271,0.00020715094,0.0000797924],"category_scores_gemma":[0.26256475,0.00012858881,0.00024317353,0.0020525542,0.00028656996,0.00046839242,0.00039005812,0.00030969715,0.000030850264],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000246273,0.00002632223,0.010259819,0.00015028493,0.000046136014,0.000002053033,0.0007493252,0.0046922145,0.0000062870704,0.0007295905,0.00097295974,0.9821187],"study_design_scores_gemma":[0.0010109851,0.00031627584,0.033711586,0.0034614622,0.00007808475,0.000022088585,0.0044419714,0.95405316,0.00015453086,0.0019530174,0.0006077415,0.0001891124],"about_ca_topic_score_codex":0.000048284543,"about_ca_topic_score_gemma":0.000009583979,"teacher_disagreement_score":0.9819296,"about_ca_system_score_codex":0.00023712085,"about_ca_system_score_gemma":0.000438075,"threshold_uncertainty_score":0.9590631},"labels":[],"label_agreement":null},{"id":"W2888442599","doi":"10.1177/0272989x18791177","title":"Picture This: Presenting Longitudinal Patient-Reported Outcome Research Study Results to Patients","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cancer survivorship and care","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"National Cancer Institute; Patient-Centered Outcomes Research Institute","keywords":"CLARITY; Medicine; Cognition; Odds; Population; Clinical trial; Odds ratio; Psychology; Clinical psychology; Internal medicine; Psychiatry","score_opus":0.13046107035880378,"score_gpt":0.4542214859802211,"score_spread":0.32376041562141733,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2888442599","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9871716,0.00003953184,0.00084385293,0.00067365146,0.0020933354,0.0011779318,0.000008791575,0.00010552289,0.007885798],"genre_scores_gemma":[0.9969228,0.0000023783589,0.00052406493,0.0009761001,0.00096305757,0.000051387615,0.000013068617,0.00004080419,0.00050636264],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9911222,0.00030112304,0.0012270923,0.00090599543,0.005711716,0.00073189824],"domain_scores_gemma":[0.9956656,0.0009389737,0.00017427096,0.0009942654,0.0015373565,0.0006895358],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0033520423,0.00023759871,0.0004868945,0.0004856827,0.0004141302,0.00009526518,0.00040680438,0.00023481598,0.0014697937],"category_scores_gemma":[0.022026125,0.00017455842,0.00013110833,0.0013002269,0.00015779307,0.00009121944,0.0008124888,0.0009320525,0.00035372045],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017890233,0.00062386174,0.7800694,0.00002222488,0.00005462366,0.0006375534,0.005730763,7.0716186e-7,0.0000073541637,0.000006840212,0.0155543005,0.19550335],"study_design_scores_gemma":[0.0037828425,0.0030719931,0.9603572,0.0011066318,0.00007114398,0.0000393957,0.0021319787,0.00020903855,0.000053856227,0.00017491633,0.028730514,0.00027049554],"about_ca_topic_score_codex":0.00032610726,"about_ca_topic_score_gemma":0.0014678563,"teacher_disagreement_score":0.19523287,"about_ca_system_score_codex":0.00019919885,"about_ca_system_score_gemma":0.00017755898,"threshold_uncertainty_score":0.999443},"labels":[],"label_agreement":null},{"id":"W2888498061","doi":"10.1177/0272989x18792284","title":"Economically Efficient Hepatitis C Virus Treatment Prioritization Improves Health Outcomes","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"National Institute on Aging; Ivey Business School, Western University; Natural Sciences and Engineering Research Council of Canada; U.S. Department of Veterans Affairs","keywords":"Medicine; Prioritization; Population; Quality-adjusted life year; Liver disease; Disease burden; Hepatitis C virus; Economic evaluation; Intensive care medicine; Cost effectiveness; Environmental health; Internal medicine; Immunology; Virus; Risk analysis (engineering); Business; Pathology","score_opus":0.20609497081285383,"score_gpt":0.4688944917542109,"score_spread":0.26279952094135706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2888498061","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7463798,0.001296553,0.20975974,0.035297498,0.0032022456,0.001100316,0.00019741502,0.00017547574,0.0025909476],"genre_scores_gemma":[0.9507164,0.00049042294,0.018145364,0.029163409,0.0010175566,0.00009237026,0.00002966268,0.000056285353,0.00028853118],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9934131,0.00027351337,0.004577077,0.00082727836,0.00030239866,0.0006066498],"domain_scores_gemma":[0.9950912,0.001991909,0.0017075216,0.00063374353,0.000084757405,0.0004908793],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.010584319,0.0002750746,0.0012306133,0.00037777936,0.0004336881,0.00014234417,0.0003886575,0.00022728369,0.0026566314],"category_scores_gemma":[0.008497135,0.0002820302,0.00019605385,0.00020716211,0.00016471176,0.00019987408,0.00011777472,0.00015545933,0.0047289226],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013898472,0.0012064715,0.3314299,0.00026177283,0.00029750023,0.000017796374,0.0052096792,0.00038052606,0.000010277329,0.16763772,0.048622567,0.4447868],"study_design_scores_gemma":[0.0038940718,0.0015918498,0.5720925,0.00080848986,0.000014529888,0.00002829839,0.0002780832,0.19014959,0.00002037546,0.04294324,0.18709289,0.0010861099],"about_ca_topic_score_codex":0.0017992542,"about_ca_topic_score_gemma":0.0021101965,"teacher_disagreement_score":0.44370067,"about_ca_system_score_codex":0.0016040779,"about_ca_system_score_gemma":0.0004556553,"threshold_uncertainty_score":0.99996316},"labels":[],"label_agreement":null},{"id":"W2891506816","doi":"10.1177/0272989x18793401","title":"Optimal Information Collection Policies in a Markov Decision Process Framework","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Hepatitis C virus research","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"National Institute on Aging; Natural Sciences and Engineering Research Council of Canada","keywords":"Value of information; Markov decision process; Dynamic programming; Data collection; Computer science; Medicine; Actuarial science; Markov process; Statistics; Mathematics; Economics; Artificial intelligence","score_opus":0.024881749135475726,"score_gpt":0.41453006703821865,"score_spread":0.3896483179027429,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2891506816","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.74903464,0.00004304949,0.24271047,0.0008467845,0.00036979115,0.00043602518,0.0000024477874,0.00007506502,0.006481733],"genre_scores_gemma":[0.95259047,0.000100058016,0.04514206,0.0016490933,0.00039874128,0.00004658765,0.000006670049,0.0000202896,0.000046007422],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99557745,0.00008829146,0.0007725493,0.0002776081,0.0027828503,0.0005012351],"domain_scores_gemma":[0.99755746,0.0011949079,0.00010433095,0.0003497335,0.00045588118,0.0003376847],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0016872919,0.0001686368,0.0003479301,0.0009239697,0.00019423172,0.000101785125,0.0002742135,0.00044342544,0.0031485374],"category_scores_gemma":[0.026261862,0.0001372024,0.00007136986,0.0018365241,0.00021326554,0.00043464295,0.00019237884,0.0007602263,0.0006351863],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014215296,0.000103769664,0.03850679,0.000063495936,0.000010507879,0.00008477895,0.0015194723,0.00002094735,0.000014686179,0.00015298779,0.005620697,0.9524803],"study_design_scores_gemma":[0.007842534,0.0030015067,0.5148191,0.01960953,0.000046738893,0.0004615434,0.0014586232,0.36149973,0.00078949524,0.017800434,0.07181933,0.0008514734],"about_ca_topic_score_codex":0.00006703603,"about_ca_topic_score_gemma":0.0001586227,"teacher_disagreement_score":0.95162886,"about_ca_system_score_codex":0.0002936754,"about_ca_system_score_gemma":0.00051083224,"threshold_uncertainty_score":0.99776274},"labels":[],"label_agreement":null},{"id":"W2893001049","doi":"10.1177/0272989x18798833","title":"Future Directions for Cost-effectiveness Analyses in Health and Medicine","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":91,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto General Hospital; McMaster University; University of Toronto","funders":"","keywords":"Transparency (behavior); Outcomes research; Health economics; Activity-based costing; Flourishing; Set (abstract data type); Observational study; Computer science; Data science; Actuarial science; Management science; Public health; Medicine; Marketing; Psychology; Economics; Business; Alternative medicine","score_opus":0.5388850673953846,"score_gpt":0.5859453135568248,"score_spread":0.04706024616144022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2893001049","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18485071,0.0333997,0.6331628,0.13742316,0.0053777862,0.0032552185,0.00013020668,0.00012583584,0.0022745687],"genre_scores_gemma":[0.95048183,0.0016524913,0.013636692,0.030143928,0.003543573,0.00038110506,0.000025525771,0.000046210458,0.00008866411],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9958602,0.00040691596,0.0026606037,0.0005363207,0.00018233099,0.00035359638],"domain_scores_gemma":[0.9926531,0.0058443174,0.0008133561,0.0002791859,0.00007606631,0.00033402664],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.023878638,0.00014272319,0.00096097594,0.0005414041,0.000307102,0.000031462154,0.00018582231,0.00016923132,0.000679389],"category_scores_gemma":[0.024213078,0.00014147186,0.00006115391,0.0004097918,0.00017842809,0.00017029555,0.000056226334,0.00017750463,0.000109266926],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00045987763,0.0003170444,0.087158546,0.001673576,0.00014869221,0.000007895649,0.008975491,0.000061778745,0.0000049337887,0.23675884,0.07802527,0.5864081],"study_design_scores_gemma":[0.003115447,0.00060169265,0.12054231,0.0024687708,0.00000573502,0.000022086188,0.0020566822,0.035034843,0.0000013245963,0.11675543,0.71905303,0.00034265182],"about_ca_topic_score_codex":0.00034295986,"about_ca_topic_score_gemma":0.00073069334,"teacher_disagreement_score":0.7656311,"about_ca_system_score_codex":0.0003694945,"about_ca_system_score_gemma":0.00017470594,"threshold_uncertainty_score":0.9840064},"labels":[],"label_agreement":null},{"id":"W2900241106","doi":"10.1177/0272989x18802797","title":"Parallel Valuation: A Direct Comparison of EQ-5D-3L and EQ-5D-5L Societal Value Sets","year":2018,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"University of Illinois at Urbana-Champaign; EuroQol Research Foundation","keywords":"EQ-5D; Ceiling effect; Statistics; Mathematics; Linear regression; Index (typography); Valuation (finance); Econometrics; Demography; Medicine; Health related quality of life; Economics; Computer science","score_opus":0.4696954993722411,"score_gpt":0.4779208813658217,"score_spread":0.008225381993580638,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2900241106","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.75033605,0.005382363,0.21318223,0.015368884,0.0022976683,0.000911056,0.00011925294,0.0001098456,0.012292659],"genre_scores_gemma":[0.96658343,0.000116641924,0.027417568,0.0049880194,0.0006558142,0.00003665581,0.00001589119,0.000035187903,0.00015080794],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9931857,0.0004581696,0.0044767563,0.00075962115,0.000638667,0.00048111146],"domain_scores_gemma":[0.9930021,0.003985026,0.0019332488,0.0005704295,0.00015829824,0.00035090075],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.022816956,0.00024841054,0.0014157491,0.00029597495,0.00034620138,0.000091758346,0.000449527,0.0003663545,0.002157553],"category_scores_gemma":[0.015891423,0.0002734367,0.00018316231,0.00031210863,0.00037171436,0.0003139413,0.00023786833,0.0003106801,0.0010683482],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033410755,0.0009621003,0.35973072,0.0009989379,0.0004981748,0.000015293404,0.01963776,0.0008791478,0.000018684617,0.1939937,0.23943773,0.18349364],"study_design_scores_gemma":[0.003548937,0.00066285365,0.15963687,0.0014461055,0.00004573649,0.000032594668,0.0016756279,0.6035507,0.000018952673,0.15938666,0.06910246,0.00089251105],"about_ca_topic_score_codex":0.00012583955,"about_ca_topic_score_gemma":0.00009463931,"teacher_disagreement_score":0.60267156,"about_ca_system_score_codex":0.00024799746,"about_ca_system_score_gemma":0.00023827261,"threshold_uncertainty_score":0.9999718},"labels":[],"label_agreement":null},{"id":"W2912496264","doi":"10.1177/0272989x18818165","title":"Prescriber Variation in Relation to Prescribing Trends within the Preferred Drugs Initiative in Ireland (2012–2015): An Interrupted Time-Series Study Using Latent Curve Models","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Pharmaceutical Economics and Policy","field":"Economics, Econometrics and Finance","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Queen's University; Queen's University Belfast","keywords":"Medicine; Interrupted Time Series Analysis; Pharmacy; Pharmacoepidemiology; Drug class; Population; Medical prescription; Drug; Pharmaceutical Benefits Scheme; Emergency medicine; Family medicine; Pharmacology; Environmental health","score_opus":0.1263360287822225,"score_gpt":0.35236520173614216,"score_spread":0.22602917295391967,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2912496264","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9871124,0.00011465606,0.007160763,0.0006932584,0.00047067858,0.0006312406,0.000029244513,0.00002210817,0.0037656617],"genre_scores_gemma":[0.9981194,0.000011143202,0.00090350513,0.00069709076,0.00007469922,0.0000355851,0.000010256682,0.000029360464,0.00011892587],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9975662,0.0001786183,0.0011996126,0.00055495097,0.00017210175,0.0003284841],"domain_scores_gemma":[0.9988048,0.00032687833,0.00031699514,0.0003699168,0.000032006195,0.00014939599],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029490208,0.00019530513,0.00046199272,0.00062917644,0.00007404307,0.00013815242,0.00037350148,0.00017502141,0.0008349871],"category_scores_gemma":[0.000492747,0.0001720731,0.000057265417,0.0005502476,0.000030672283,0.0011065409,0.00024631387,0.0004675427,0.0001767253],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0027235108,0.0021061937,0.3122873,0.000049853108,0.00023459017,0.000047744685,0.1447694,0.41189057,0.00011747114,0.06993416,0.0007098068,0.05512938],"study_design_scores_gemma":[0.0013203896,0.00015061199,0.14798062,0.00016169247,0.000006169598,0.0000023412163,0.00024511217,0.81635845,0.0000022879508,0.033337172,0.00021963002,0.0002155064],"about_ca_topic_score_codex":0.00029226622,"about_ca_topic_score_gemma":0.00042308253,"teacher_disagreement_score":0.40446788,"about_ca_system_score_codex":0.00030033218,"about_ca_system_score_gemma":0.000048892045,"threshold_uncertainty_score":0.91425234},"labels":[],"label_agreement":null},{"id":"W2913713220","doi":"10.1177/0272989x18823757","title":"Physician and Nonphysician Estimates of Positive Predictive Value in Diagnostic v. Mass Screening Mammography: An Examination of Bayesian Reasoning","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Medicine; Test (biology); Mammography; Asymptomatic; Predictive value; Overdiagnosis; Bayesian probability; Gynecology; Statistics; Surgery; Internal medicine; Breast cancer; Cancer; Mathematics","score_opus":0.017540071810358136,"score_gpt":0.3230832364258559,"score_spread":0.30554316461549774,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2913713220","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8616302,0.0003418098,0.13579899,0.00005660668,0.00006731197,0.00036431992,0.0000106200405,0.00002037248,0.001709778],"genre_scores_gemma":[0.96593213,0.000035054356,0.033689044,0.00022728168,0.000068150046,0.000008528749,0.000017361013,0.00002011734,0.0000023310463],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99740475,0.00012423139,0.00057615107,0.00039810786,0.0012173518,0.0002794215],"domain_scores_gemma":[0.99746615,0.0016683454,0.00027236232,0.0002453426,0.00017650904,0.00017129246],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001071378,0.00018216793,0.0005665575,0.00033511728,0.000043814016,0.000014991674,0.00015419214,0.00015238159,0.000043769593],"category_scores_gemma":[0.0024730253,0.00016363688,0.0000866004,0.00059853937,0.00016089164,0.00029704327,0.00008267345,0.00029956235,0.000001550628],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005504518,0.00014061338,0.32906666,0.00013218571,0.000055639484,0.00012682761,0.0010227372,0.00094902486,0.0021365127,0.0012935853,0.000007769306,0.664518],"study_design_scores_gemma":[0.0011608148,0.0010600358,0.8166311,0.01234585,0.00007594814,0.00001912233,0.0012584813,0.16377288,0.0014608356,0.0020731525,0.0000036979604,0.00013809322],"about_ca_topic_score_codex":0.00014952575,"about_ca_topic_score_gemma":0.000018666935,"teacher_disagreement_score":0.6643799,"about_ca_system_score_codex":0.000052469404,"about_ca_system_score_gemma":0.000074815835,"threshold_uncertainty_score":0.6672919},"labels":[],"label_agreement":null},{"id":"W2914484459","doi":"10.1177/0272989x18821368","title":"Reducing Uncertainty in EQ-5D Value Sets: The Role of Spatial Correlation","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Agency for Healthcare Research and Quality; U.S. Department of Health and Human Services","keywords":"Statistics; Gaussian; Standard deviation; Correlation; Mathematics; Bayesian probability; Mean squared error; Econometrics; Spatial correlation; Valuation (finance); Confidence interval","score_opus":0.1982100475136389,"score_gpt":0.4083940910139684,"score_spread":0.2101840435003295,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2914484459","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9493668,0.0013270844,0.035115708,0.00612392,0.0015128946,0.000657766,0.000031419528,0.00002401784,0.0058403956],"genre_scores_gemma":[0.99571383,0.000038113034,0.0015001434,0.0024647277,0.00017867763,0.000019456313,0.000009602583,0.000018154517,0.00005728299],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9951235,0.00036441674,0.0034214905,0.00043129706,0.0003610249,0.0002982939],"domain_scores_gemma":[0.99348927,0.0044764145,0.001378413,0.0005093263,0.000044891305,0.00010171447],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.021975558,0.00012879487,0.00067025924,0.00028042577,0.00008861875,0.000034770987,0.0004115564,0.00021875827,0.0022612584],"category_scores_gemma":[0.0141399605,0.00011963599,0.00010296898,0.00026058152,0.000064835855,0.0001988791,0.00011632154,0.00034618,0.0011788205],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001384887,0.00017725895,0.5701108,0.00018758701,0.00004786375,0.0000038688286,0.0052042077,0.08376114,0.000011696206,0.17985936,0.0031384323,0.15735933],"study_design_scores_gemma":[0.00077847467,0.00005857086,0.10177511,0.0007798344,0.000002976002,0.0000054875322,0.0009933998,0.8001826,0.0000028027607,0.084208906,0.011037578,0.00017427634],"about_ca_topic_score_codex":0.0020523705,"about_ca_topic_score_gemma":0.00028592796,"teacher_disagreement_score":0.7164214,"about_ca_system_score_codex":0.00035369035,"about_ca_system_score_gemma":0.00018141916,"threshold_uncertainty_score":0.99959886},"labels":[],"label_agreement":null},{"id":"W2915000621","doi":"10.1177/0272989x19835239","title":"Combining Multiple Treatment Comparisons with Personalized Patient Preferences: A Randomized Trial of an Interactive Platform for Statin Treatment Selection","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"London School of Economics and Political Science","keywords":"Randomized controlled trial; Medicine; Confidence interval; Clinical endpoint; Decision aids; Physical therapy; Population; Ranking (information retrieval); Intervention (counseling); Scale (ratio); Internal medicine; Computer science; Alternative medicine; Artificial intelligence; Psychiatry","score_opus":0.28377383212436685,"score_gpt":0.45441307375078666,"score_spread":0.1706392416264198,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2915000621","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8963941,0.0001101284,0.09767627,0.00037248837,0.00049373624,0.0043402403,0.00014558648,0.00003225992,0.00043520276],"genre_scores_gemma":[0.965622,0.000032376556,0.032949492,0.00034515627,0.00009142732,0.00078188034,0.00008031672,0.00002591185,0.00007142028],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.99531007,0.00039423798,0.003153078,0.00055152597,0.00028706563,0.00030402114],"domain_scores_gemma":[0.9854023,0.011718705,0.0023148465,0.00027054182,0.00010552782,0.00018812076],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0058653327,0.00023992214,0.0019543834,0.00027615394,0.00014964763,0.000055840428,0.00015052984,0.00014483834,0.00074981706],"category_scores_gemma":[0.0046517416,0.00018872408,0.00023940831,0.00013195015,0.000085976106,0.00030298356,0.000022736296,0.00010585331,0.00007125904],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.87274945,0.003425502,0.008558116,0.0001711488,0.00090980524,0.0000020930777,0.026316188,0.0018535769,0.0000032413468,0.013383094,0.0004589952,0.072168775],"study_design_scores_gemma":[0.6388865,0.008070753,0.00031762474,0.0004827297,0.0000408833,0.0000046707773,0.00425433,0.33986428,0.00001960625,0.004266285,0.0035385469,0.00025374867],"about_ca_topic_score_codex":0.0006649913,"about_ca_topic_score_gemma":0.0006162023,"teacher_disagreement_score":0.3380107,"about_ca_system_score_codex":0.0008722969,"about_ca_system_score_gemma":0.0003057737,"threshold_uncertainty_score":0.8209971},"labels":[],"label_agreement":null},{"id":"W2919824149","doi":"10.1177/0272989x19832882","title":"Aiding Risk Information learning through Simulated Experience (ARISE): A Comparison of the Communication of Screening Test Information in Explicit and Simulated Experience Formats","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University; Sinai Health System; Lunenfeld-Tanenbaum Research Institute; University of Toronto","funders":"","keywords":"Risk communication; Test (biology); Computer science; Psychology; Machine learning; Information retrieval; Medicine; Risk analysis (engineering)","score_opus":0.11368225923922429,"score_gpt":0.453634357758752,"score_spread":0.3399520985195277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2919824149","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.980673,0.00034916305,0.016850287,0.00018679799,0.00020257714,0.0012810689,0.000008451413,0.00006252094,0.00038608955],"genre_scores_gemma":[0.9952789,0.000163031,0.004079208,0.00037753294,0.000007751493,0.000046210003,0.000030490019,0.000012166542,0.0000047138237],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99444634,0.0010531668,0.0026883818,0.00017085011,0.0012610005,0.00038028648],"domain_scores_gemma":[0.98827857,0.0077016773,0.002376477,0.0009815908,0.0005641852,0.000097515476],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0012491808,0.00019222498,0.0004909553,0.00028095467,0.0015753366,0.000020332463,0.0009108508,0.00038045208,0.00015379736],"category_scores_gemma":[0.015946073,0.0001533399,0.00006098802,0.0010349164,0.00016784109,0.0026265641,0.0018835542,0.0015612979,0.000018465518],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000121864796,0.000030043997,0.6202075,0.000191468,0.00000402125,6.0871734e-8,0.3270158,0.018128147,0.000032452204,0.0001671945,0.000012450311,0.03408902],"study_design_scores_gemma":[0.0017560228,0.000104944666,0.10741818,0.0057262466,0.000007856278,0.0000011209559,0.14470737,0.73824054,0.00013259523,0.0004041698,0.0013341353,0.00016682161],"about_ca_topic_score_codex":0.0013727073,"about_ca_topic_score_gemma":0.00010380405,"teacher_disagreement_score":0.7201124,"about_ca_system_score_codex":0.00015238563,"about_ca_system_score_gemma":0.00017093559,"threshold_uncertainty_score":0.99972445},"labels":[],"label_agreement":null},{"id":"W2935956843","doi":"10.1177/0272989x19841592","title":"Latent Class Models Reveal Poor Agreement between Discrete-Choice and Time Tradeoff Preferences","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; Impact; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Hospital for Sick Children; Public Health Ontario","funders":"Canadian Institutes of Health Research; EuroQol Research Foundation","keywords":"Latent class model; Anxiety; Time-trade-off; Psychology; Correlation; Multivariate statistics; Econometrics; Discrete choice; EQ-5D; Statistics; Medicine; Mathematics; Quality of life (healthcare); Psychiatry; Health related quality of life","score_opus":0.11032512056057094,"score_gpt":0.2653933828694856,"score_spread":0.15506826230891468,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2935956843","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9736552,0.00049254985,0.0117487395,0.00071118266,0.00021765062,0.00024035577,0.00004129268,0.000017904695,0.012875127],"genre_scores_gemma":[0.99695003,0.00012812058,0.0013639122,0.00040017188,0.00011615082,0.000010026125,0.000021067917,0.000015297033,0.0009952048],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9984008,0.000024466784,0.0006735934,0.00048545914,0.00016759505,0.00024808632],"domain_scores_gemma":[0.99906176,0.00034376653,0.00021795099,0.00021710426,0.000004882015,0.00015456046],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.000995403,0.00014607032,0.00038281688,0.000098874334,0.00007236114,0.00005942095,0.0002310318,0.00016387411,0.0043337373],"category_scores_gemma":[0.00012996826,0.00013871124,0.00007132074,0.000063502834,0.0000560172,0.00028216085,0.00017238889,0.00018299841,0.0023031905],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000017325177,0.000049374234,0.8858201,0.000026056125,0.00005576171,0.0000019534089,0.0002027954,0.0007304655,0.0000108747745,0.011187401,0.0011936761,0.1007042],"study_design_scores_gemma":[0.0009489048,0.00014909961,0.77229315,0.00017933708,0.000012746923,0.0000028209308,0.00003143816,0.10895226,0.0000060531474,0.11267495,0.004442326,0.00030692227],"about_ca_topic_score_codex":0.00003141708,"about_ca_topic_score_gemma":0.0000050950207,"teacher_disagreement_score":0.11352698,"about_ca_system_score_codex":0.0001075402,"about_ca_system_score_gemma":0.0000110819565,"threshold_uncertainty_score":0.99847364},"labels":[],"label_agreement":null},{"id":"W2936510582","doi":"10.1177/0272989x19837969","title":"Toward a Centralized, Systematic Approach to the Identification, Appraisal, and Use of Health State Utility Values for Reimbursement Decision Making: Introducing the Health Utility Book (HUB)","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":23,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Ottawa; Public Health Ontario; Sunnybrook Health Science Centre; University of Toronto; McMaster University; Health Sciences Centre; Impact","funders":"","keywords":"Reimbursement; Health care; Identification (biology); Quality (philosophy); Actuarial science; Checklist; Medicine; Risk analysis (engineering); Operations management; Management science; Business; Economics; Psychology","score_opus":0.3403244859399411,"score_gpt":0.46948516787381134,"score_spread":0.12916068193387026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2936510582","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.057065934,0.008249496,0.88404423,0.04201001,0.0015170107,0.0068095694,0.00020244603,0.000044643282,0.000056691035],"genre_scores_gemma":[0.94722354,0.0004525455,0.03438934,0.017256139,0.00017347018,0.00034622522,0.00002345564,0.000045243887,0.0000900399],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98592585,0.0019916538,0.009227179,0.0012348011,0.00086251844,0.000757971],"domain_scores_gemma":[0.97819376,0.0146895135,0.004727235,0.0018222188,0.00022968584,0.000337575],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.105909064,0.00029986983,0.0018395411,0.00028808677,0.00060027983,0.0002967628,0.00085304916,0.00013660666,0.00018674404],"category_scores_gemma":[0.046668366,0.0002209622,0.0002148661,0.00039723702,0.00015176446,0.00039132408,0.00031703006,0.000320088,0.00012298844],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012461212,0.0013379046,0.050737187,0.062297907,0.00060788513,0.0000011153232,0.06008432,0.0047531556,0.000001529734,0.08410341,0.4907639,0.24406557],"study_design_scores_gemma":[0.0030999773,0.0005984422,0.13189301,0.019275244,0.00005199915,0.00003934432,0.009458674,0.64250654,0.0000024981452,0.053565543,0.13856952,0.00093923684],"about_ca_topic_score_codex":0.00029111945,"about_ca_topic_score_gemma":0.00006371456,"teacher_disagreement_score":0.8901576,"about_ca_system_score_codex":0.00049598375,"about_ca_system_score_gemma":0.00049707777,"threshold_uncertainty_score":0.96136194},"labels":[],"label_agreement":null},{"id":"W2941632791","doi":"10.1177/0272989x19851049","title":"Manipulating the 5 Dimensions of the EuroQol Instrument: The Effects on Self-Reporting Actual Health and Valuing Hypothetical Health States","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"EuroQol Research Foundation","keywords":"Anxiety; EQ-5D; Context (archaeology); Valuation (finance); Psychology; Population; Mental health; Medicine; Psychiatry; Disease; Environmental health; Economics","score_opus":0.21052394097115604,"score_gpt":0.43098098267127777,"score_spread":0.22045704170012173,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2941632791","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9205037,0.0013668144,0.0025067648,0.07296235,0.0010983204,0.0012068821,0.000010209046,0.000036405054,0.0003085895],"genre_scores_gemma":[0.9566371,0.0002099299,0.00163921,0.041273437,0.00015833658,0.000021694079,0.000002180079,0.000029686877,0.000028456492],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.992397,0.0011056677,0.004910472,0.00050870917,0.0005660238,0.00051209563],"domain_scores_gemma":[0.9784866,0.015226695,0.0052047796,0.0007828869,0.000038917322,0.0002601045],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.034889948,0.00019375187,0.0008540229,0.000116100935,0.00076838647,0.000077338525,0.0004663043,0.00010143883,0.000094950134],"category_scores_gemma":[0.029996011,0.00010600779,0.00013841402,0.00024966884,0.000115929724,0.00010565771,0.0003240312,0.00055648235,0.000109427616],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011438504,0.00039638675,0.38058737,0.002490655,0.00030717495,0.0000071867694,0.03762051,0.0016177382,0.000009139266,0.20384951,0.017684804,0.35531515],"study_design_scores_gemma":[0.004420801,0.001992843,0.43544564,0.013771262,0.000035802303,0.0002584122,0.012038171,0.37153083,0.00002113289,0.10438377,0.0550212,0.0010801426],"about_ca_topic_score_codex":0.0002186998,"about_ca_topic_score_gemma":0.000045974502,"teacher_disagreement_score":0.36991307,"about_ca_system_score_codex":0.00028682244,"about_ca_system_score_gemma":0.000350043,"threshold_uncertainty_score":0.9937839},"labels":[],"label_agreement":null},{"id":"W2946857827","doi":"10.1177/0272989x19851346","title":"Decisional Conflict Scale Findings among Patients and Surrogates Making Health Decisions: Part II of an Anniversary Review","year":2019,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Ethics and Legal Issues in Pediatric Healthcare","field":"Medicine","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Centre intégré universitaire de santé et de services sociaux de la Capitale-Nationale; Ottawa Hospital; Centres Intégré Universitaires de Santé et de Services Sociaux; Centre Intégré Universitaire de Santé et de Services Sociaux du Centre-Sud-de-l'Île-de-Montréal; Université Laval; Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean","funders":"Institute of Health Services and Policy Research; Canadian Institutes of Health Research","keywords":"Scale (ratio); Psychology; Operations research; Management science; Medicine; Economics; Geography; Engineering","score_opus":0.12416427439300685,"score_gpt":0.47494299596039397,"score_spread":0.3507787215673871,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2946857827","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0021255128,0.99311423,0.00019915769,0.0004327498,0.0011635179,0.0023486887,0.00017897539,0.000057490088,0.00037966354],"genre_scores_gemma":[0.00079293316,0.99018645,0.004864705,0.0030662753,0.0004183553,0.000038500737,0.00039775315,0.00011977741,0.00011522582],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98909473,0.0007879213,0.0035851896,0.0014049856,0.004302614,0.0008245848],"domain_scores_gemma":[0.98750937,0.007828405,0.0014655931,0.0012114821,0.0007468694,0.0012382922],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0070103174,0.00077978836,0.0051774248,0.00074489013,0.00043202456,0.000051389736,0.00081834436,0.001388147,0.0011025133],"category_scores_gemma":[0.016406627,0.00057168823,0.00075377343,0.001082753,0.00041735815,0.00025424466,0.0010941401,0.0026350627,0.000053421914],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004870793,0.0002101523,0.0011519507,0.033827808,0.00006197822,0.00008479081,0.00017877274,1.5206871e-7,3.1470693e-9,0.000076042685,0.006812586,0.95754707],"study_design_scores_gemma":[0.0006080829,0.000663141,0.00072276447,0.37953052,0.00042865335,0.0001038617,0.000027661692,0.00006699398,2.141996e-8,0.00012202173,0.6174203,0.00030596656],"about_ca_topic_score_codex":0.000040494608,"about_ca_topic_score_gemma":0.00002936225,"teacher_disagreement_score":0.9572411,"about_ca_system_score_codex":0.00029864875,"about_ca_system_score_gemma":0.0023225842,"threshold_uncertainty_score":0.99990827},"labels":[],"label_agreement":null},{"id":"W2947651119","doi":"10.1177/0272989x19851345","title":"Decisional Conflict Scale Use over 20 Years: The Anniversary Review","year":2019,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":175,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Centre intégré universitaire de santé et de services sociaux de la Capitale-Nationale; Ottawa Hospital; Centres Intégré Universitaires de Santé et de Services Sociaux; Centre Intégré Universitaire de Santé et de Services Sociaux du Centre-Sud-de-l'Île-de-Montréal; Université Laval; Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean","funders":"Institute of Health Services and Policy Research; Canadian Institutes of Health Research","keywords":"Scale (ratio); Geography; Cartography","score_opus":0.5006534926653604,"score_gpt":0.5603488189049478,"score_spread":0.0596953262395874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2947651119","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000007362586,0.988891,0.00043393602,0.0010397886,0.0035115283,0.0046979226,0.0002143962,0.00012625825,0.0010778216],"genre_scores_gemma":[0.0000059767353,0.97903776,0.0017129931,0.016460065,0.0005967851,0.0004915969,0.000252839,0.00013599219,0.0013059912],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98587996,0.0051384987,0.0033933052,0.0010003974,0.003639352,0.0009485178],"domain_scores_gemma":[0.95395005,0.039481092,0.0017579782,0.0037755654,0.00049125147,0.0005440874],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.0031734155,0.0006444245,0.0027682213,0.00034163255,0.00078763475,0.00005285541,0.0035038947,0.0014789173,0.011741761],"category_scores_gemma":[0.015627643,0.00040056193,0.00092594256,0.0010833223,0.0002679486,0.00029808556,0.002947727,0.0052431296,0.0076307277],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000015659301,0.000026745183,0.000095192714,0.013698772,0.00004530359,0.000027062002,0.00024989454,1.9691889e-7,2.8861349e-9,0.00012920947,0.20163721,0.7840748],"study_design_scores_gemma":[0.00027425966,0.000024266179,0.00018176233,0.3094254,0.0002558955,0.000021511123,0.00007491155,0.00006283171,6.9969097e-10,0.00011783349,0.6892953,0.00026605645],"about_ca_topic_score_codex":0.00006188475,"about_ca_topic_score_gemma":0.00005440501,"teacher_disagreement_score":0.7838087,"about_ca_system_score_codex":0.00066231674,"about_ca_system_score_gemma":0.0036843012,"threshold_uncertainty_score":0.9998446},"labels":[],"label_agreement":null},{"id":"W2950093054","doi":"10.1177/0272989x18824098","title":"Development and Validation of the Evaluation Platform in COPD (EPIC): A Population-Based Outcomes Model of COPD for Canada","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Vancouver Coastal Health; Statistics Canada; University of British Columbia","funders":"Canadian Institutes of Health Research; University of Ottawa","keywords":"COPD; Medicine; Population; Comorbidity; Internal validity; EPIC; External validity; Intensive care medicine; Environmental health; Internal medicine; Statistics","score_opus":0.05696532586295661,"score_gpt":0.3671592186248106,"score_spread":0.310193892761854,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2950093054","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99482304,0.0000852293,0.0031397669,0.00044713606,0.00013502924,0.0012387868,0.000018029421,0.000004008449,0.000108982764],"genre_scores_gemma":[0.99466836,0.0000013284896,0.0050181635,0.00016355065,0.000010650783,0.0000449627,0.000042518124,0.000011674967,0.00003877657],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99680763,0.000043351716,0.00056911795,0.0002199623,0.002186863,0.0001730994],"domain_scores_gemma":[0.9983875,0.00083334924,0.00016838078,0.00024542172,0.00026912946,0.00009617372],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011889,0.00010156979,0.00030684445,0.00013240584,0.00004413412,0.000005250387,0.00012979086,0.00008615977,0.00022653156],"category_scores_gemma":[0.002440992,0.00006991662,0.00005242027,0.00020184483,0.00003918912,0.000053653115,0.000076573895,0.000125866,6.8490186e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041245003,0.000103399,0.7941286,0.00050086976,0.000027519965,0.000002251869,0.00015707742,0.009514528,0.00015075518,0.00045944288,0.000109954744,0.19443315],"study_design_scores_gemma":[0.002053385,0.000013819527,0.527558,0.00054819934,0.000019647641,9.409153e-7,0.000058882382,0.4683492,0.00039898304,0.00090109947,0.000050005496,0.000047833742],"about_ca_topic_score_codex":0.0032881384,"about_ca_topic_score_gemma":0.019682063,"teacher_disagreement_score":0.45883468,"about_ca_system_score_codex":0.0006702851,"about_ca_system_score_gemma":0.0029783302,"threshold_uncertainty_score":0.9982062},"labels":[],"label_agreement":null},{"id":"W2966994690","doi":"10.1177/0272989x19868193","title":"Are Patient Decision Aids Used in Clinical Practice after Rigorous Evaluation? A Survey of Trial Authors","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":121,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Decision aids; Psychological intervention; Medicine; Descriptive statistics; Government (linguistics); Family medicine; Health care; Alternative medicine; Nursing; Psychology","score_opus":0.3912548302796354,"score_gpt":0.5840144983006573,"score_spread":0.1927596680210219,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2966994690","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9839799,0.0011151187,0.0020617892,0.00073515734,0.007008127,0.004453023,0.00003806483,0.000053467185,0.00055539183],"genre_scores_gemma":[0.97978497,0.00006406002,0.016723791,0.002625925,0.0001578657,0.0005426851,0.00003060002,0.00004882367,0.000021312451],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9687664,0.01854383,0.0058469125,0.00086237234,0.005284636,0.00069585216],"domain_scores_gemma":[0.9191713,0.072826415,0.0031085017,0.002070737,0.0023605933,0.00046246266],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.027880816,0.00028256967,0.0010373389,0.0004853259,0.0002554485,0.000019084397,0.00094083016,0.0010604409,0.00408671],"category_scores_gemma":[0.2766576,0.00023468534,0.0001913311,0.0011767616,0.00013055738,0.0003460219,0.0010211028,0.0025423178,0.0007207157],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.028163673,0.00052680576,0.6592915,0.000069100206,0.00001767698,0.000020789741,0.0020224534,0.000035889392,3.5576755e-7,0.000048234302,0.0012368753,0.30856666],"study_design_scores_gemma":[0.03595964,0.0010873211,0.9306147,0.005704424,0.000040722014,0.0000034953835,0.0041061854,0.011874218,5.4305673e-7,0.00203393,0.008278708,0.0002961162],"about_ca_topic_score_codex":0.00070526334,"about_ca_topic_score_gemma":0.0029222725,"teacher_disagreement_score":0.30827054,"about_ca_system_score_codex":0.0005444786,"about_ca_system_score_gemma":0.0019897663,"threshold_uncertainty_score":0.99975884},"labels":[],"label_agreement":null},{"id":"W2967387702","doi":"10.1177/0272989x19866296","title":"Implementation Research on Shared Decision Making in Primary Care: Inventory of Intracluster Correlation Coefficients","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université Laval","funders":"Fonds de Recherche du Québec - Santé; Canada Research Chairs","keywords":"Interquartile range; Statistics; Random effects model; Medicine; Cluster (spacecraft); Meta-analysis; Computer science; Mathematics; Internal medicine","score_opus":0.25745492534839015,"score_gpt":0.5440534075821251,"score_spread":0.2865984822337349,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2967387702","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97672,0.00084085006,0.010786896,0.00044950316,0.002494789,0.0030382108,0.00002676025,0.0000662348,0.0055767647],"genre_scores_gemma":[0.99413615,0.000055663033,0.003936091,0.001371074,0.000103295766,0.0001827756,0.00012677386,0.000046704816,0.000041442185],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9903222,0.0024994016,0.0020718286,0.0006279571,0.0037260395,0.00075260055],"domain_scores_gemma":[0.9871178,0.010038056,0.0006232962,0.0011468157,0.00088082714,0.00019320463],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0038794295,0.000207973,0.00046513832,0.0011726643,0.00051881414,0.000021026844,0.0009157105,0.0005425086,0.002783121],"category_scores_gemma":[0.0034624436,0.00019181531,0.00008336811,0.0012437089,0.00010619302,0.00026962973,0.0010604506,0.0021143435,0.0007181068],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011108335,0.00015232693,0.470744,0.0006634503,0.0000073626193,0.000009548918,0.016734613,0.000609809,0.000037135895,0.00091815746,0.002842095,0.50617063],"study_design_scores_gemma":[0.007936571,0.0008039514,0.86011916,0.038980506,0.000013600045,0.0000034829025,0.04133761,0.037920777,0.000017785094,0.0053262874,0.0070707933,0.0004694924],"about_ca_topic_score_codex":0.00013030847,"about_ca_topic_score_gemma":0.00050796266,"teacher_disagreement_score":0.5057012,"about_ca_system_score_codex":0.0020031019,"about_ca_system_score_gemma":0.0010350206,"threshold_uncertainty_score":0.9981285},"labels":[],"label_agreement":null},{"id":"W2970783423","doi":"10.1177/0272989x19856617","title":"Exclusion Criteria as Measurements I: Identifying Invalid Responses","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; National Institutes of Health; National Center for Advancing Translational Sciences; Riksbankens Jubileumsfond","keywords":"Valuation (finance); Sample (material); Psychology; Actuarial science; Econometrics; Medicine; Social psychology; Economics","score_opus":0.575652382298607,"score_gpt":0.5373958065628411,"score_spread":0.03825657573576591,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2970783423","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93264645,0.0016932779,0.034833267,0.011126512,0.004242642,0.00072487106,0.00003051517,0.00011303615,0.014589444],"genre_scores_gemma":[0.9700003,0.000080991915,0.0073637064,0.020897945,0.00039812204,0.000034013454,0.00001022281,0.00004779621,0.0011668904],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9930906,0.0006307467,0.004106145,0.00082139723,0.0008287204,0.0005223579],"domain_scores_gemma":[0.9941412,0.0032693064,0.0013492934,0.00078085536,0.000104040235,0.0003552997],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.03932392,0.00022777086,0.0009395726,0.0005507794,0.0002777802,0.00021841504,0.00067376817,0.00029661958,0.023659129],"category_scores_gemma":[0.039946865,0.00026051508,0.0001691978,0.00030118233,0.000057385278,0.00058298843,0.00035778744,0.00031324755,0.028289784],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012493785,0.0007658907,0.5330244,0.0018060764,0.00034303847,0.00014046664,0.008329864,0.00018913616,0.000781836,0.1006687,0.25969082,0.09301044],"study_design_scores_gemma":[0.0070363902,0.0005658244,0.23105264,0.008388776,0.000028996557,0.0001845875,0.0023102807,0.03168023,0.00015653996,0.35583594,0.36053512,0.002224648],"about_ca_topic_score_codex":0.00013107285,"about_ca_topic_score_gemma":0.000035751626,"teacher_disagreement_score":0.3019717,"about_ca_system_score_codex":0.00050353823,"about_ca_system_score_gemma":0.00023537292,"threshold_uncertainty_score":0.9999847},"labels":[],"label_agreement":null},{"id":"W2971133796","doi":"10.1177/0272989x19862542","title":"Exclusion Criteria as Measurements II: Effects on Utility Functions","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":25,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Social Sciences and Humanities Research Council of Canada; National Institutes of Health; National Center for Advancing Translational Sciences; Riksbankens Jubileumsfond","keywords":"Computer science; Risk analysis (engineering); Actuarial science; Medicine; Economics","score_opus":0.3640739332785422,"score_gpt":0.4811623043652489,"score_spread":0.11708837108670672,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2971133796","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.91988206,0.00059051573,0.024138724,0.012777137,0.006122605,0.0009852578,0.000036261263,0.00012032244,0.035347123],"genre_scores_gemma":[0.97386354,0.000019737261,0.0016053386,0.022905545,0.0004935861,0.000051858948,0.000015229623,0.000033875596,0.0010113072],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99463135,0.00044192342,0.0029049693,0.00084932015,0.00069799466,0.00047443155],"domain_scores_gemma":[0.99449813,0.0033164723,0.00089333276,0.00084286055,0.000082727965,0.00036647808],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.021998439,0.00023433355,0.0008798148,0.00037269742,0.0004485393,0.00008493851,0.00044843907,0.00031241565,0.019821651],"category_scores_gemma":[0.025926944,0.00025344122,0.0001797632,0.0002697264,0.000045193905,0.00030083692,0.00026833604,0.0003683297,0.027762802],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007031843,0.0013718633,0.21073063,0.0013220882,0.00025618132,0.000036136487,0.0025862951,0.00020430113,0.00009673833,0.045067,0.46601143,0.27161413],"study_design_scores_gemma":[0.0057654595,0.0014806557,0.25517696,0.0046437113,0.000024143024,0.000035825957,0.00042469864,0.04688185,0.000056717337,0.13026227,0.553923,0.0013247079],"about_ca_topic_score_codex":0.00008200413,"about_ca_topic_score_gemma":0.000026415451,"teacher_disagreement_score":0.27028942,"about_ca_system_score_codex":0.0004648077,"about_ca_system_score_gemma":0.00015651736,"threshold_uncertainty_score":0.9999918},"labels":[],"label_agreement":null},{"id":"W2976393939","doi":"10.1177/0272989x19873625","title":"Primary Care Providers’ Preferences and Concerns Regarding Specific Visual Displays for Returning Hemoglobin A1c Test Results to Patients","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Mobile Health and mHealth Applications","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"Agency for Healthcare Research and Quality","keywords":"Primary care; Test (biology); Medicine; Hemoglobin; Psychology; Family medicine; Internal medicine","score_opus":0.04446923732404293,"score_gpt":0.4232392161828472,"score_spread":0.37876997885880426,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2976393939","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9603336,0.000618638,0.013172362,0.001861331,0.0019687747,0.011702738,0.00039086418,0.0001739884,0.009777693],"genre_scores_gemma":[0.9831322,0.00025053244,0.008564725,0.0042962674,0.0006418485,0.002488824,0.0002751894,0.000047266712,0.0003031527],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9956865,0.00020555405,0.0013275559,0.0008676923,0.0010170236,0.0008956997],"domain_scores_gemma":[0.990338,0.007492384,0.00042933147,0.00043800366,0.00042171445,0.00088051765],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001603457,0.00024251828,0.0005103115,0.00017511386,0.0010340394,0.000038304403,0.00033564688,0.00039583122,0.00021407903],"category_scores_gemma":[0.0055535687,0.00019552912,0.000062521045,0.00032022857,0.00005819383,0.0001473688,0.0003620735,0.00075240777,0.00022732331],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011236422,0.00008509561,0.22885293,0.0016130182,0.0000059409053,0.0000026679868,0.002958493,0.0000051627144,0.000050496645,0.0005789176,0.033142015,0.7315816],"study_design_scores_gemma":[0.008719744,0.0011756459,0.2890838,0.009833363,0.00002379989,0.0000027235444,0.0036841244,0.0012398368,0.000014504184,0.0010055426,0.6847023,0.0005146564],"about_ca_topic_score_codex":0.000035722143,"about_ca_topic_score_gemma":0.00013743447,"teacher_disagreement_score":0.73106694,"about_ca_system_score_codex":0.00040799516,"about_ca_system_score_gemma":0.0007661149,"threshold_uncertainty_score":0.7973446},"labels":[],"label_agreement":null},{"id":"W2980705747","doi":"10.1177/0272989x19879095","title":"Probabilistic Graphical Modeling for Estimating Risk of Coronary Artery Disease: Applications of a Flexible Machine-Learning Method","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Artificial Intelligence in Healthcare","field":"Health Professions","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo; Public Health Ontario; University of Toronto; University of Calgary","funders":"","keywords":"Machine learning; Bayesian network; Artificial intelligence; Coronary artery disease; Missing data; CAD; Computer science; Receiver operating characteristic; Probabilistic logic; Logistic regression; Risk assessment; Artificial neural network; Support vector machine; Medicine; Internal medicine; Engineering","score_opus":0.10613431735221995,"score_gpt":0.5019966610811613,"score_spread":0.3958623437289413,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2980705747","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18191294,0.00034398338,0.81473905,0.00017186346,0.00046230754,0.0021125586,0.00005592192,0.000080141144,0.000121239056],"genre_scores_gemma":[0.7651165,0.000022593353,0.23395744,0.00011604773,0.00015696621,0.00052739447,0.000022046805,0.00003570379,0.00004530204],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99536014,0.0008254893,0.0017826977,0.00050877006,0.0010244782,0.00049843435],"domain_scores_gemma":[0.98270935,0.01514435,0.0007050024,0.0005324092,0.0005789434,0.0003299672],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0044917394,0.00018349593,0.00056379865,0.00023925198,0.0005397042,0.000005782838,0.0004301548,0.00030989095,0.0009877101],"category_scores_gemma":[0.013718015,0.00015508078,0.00019929677,0.00048305353,0.00010149009,0.000085294945,0.00026137664,0.0011210336,0.000054427783],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00084882183,0.00019526151,0.22649564,0.002423411,0.00005676602,0.0000059599747,0.0016932916,0.1405438,0.000040799892,0.017646136,0.000055719385,0.6099944],"study_design_scores_gemma":[0.00022477962,0.00011047274,0.0014328398,0.002295471,0.000054350774,0.0000023150674,0.0008700375,0.91057605,0.0000032277524,0.08414912,0.0001597619,0.00012159706],"about_ca_topic_score_codex":0.00016400029,"about_ca_topic_score_gemma":0.000064228545,"teacher_disagreement_score":0.7700322,"about_ca_system_score_codex":0.00008232872,"about_ca_system_score_gemma":0.0004837415,"threshold_uncertainty_score":0.9999255},"labels":[],"label_agreement":null},{"id":"W2984574415","doi":"10.1177/0272989x19883631","title":"Using an Online, Modified Delphi Approach to Engage Patients and Caregivers in Determining the Patient-Centeredness of Duchenne Muscular Dystrophy Care Considerations","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"Economic and Social Research Council; National Institutes of Health; Patient-Centered Outcomes Research Institute","keywords":"Operationalization; Duchenne muscular dystrophy; Medicine; Guideline; Delphi method; Health care; Family medicine","score_opus":0.17600927948378278,"score_gpt":0.42835953039000746,"score_spread":0.25235025090622465,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2984574415","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99227715,0.000028596465,0.005957602,0.000073418865,0.00018305732,0.0008215511,0.00001936446,0.000023158673,0.00061613537],"genre_scores_gemma":[0.982728,0.000006265353,0.017074104,0.00012562345,0.000026741676,0.000014990529,0.000008346723,0.000014724518,0.0000011930854],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.996694,0.0006163533,0.00041773645,0.00036366095,0.0015758754,0.00033235006],"domain_scores_gemma":[0.9980533,0.0009507703,0.0001068281,0.00029846837,0.00038916082,0.00020148842],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011319903,0.000118582226,0.00023207488,0.00024550033,0.00033884254,0.00008733362,0.00040143883,0.00016459421,0.000040409217],"category_scores_gemma":[0.0046887984,0.00009352585,0.000048224505,0.00040746108,0.00024489898,0.00019456969,0.0002999327,0.00036465327,0.0000012158167],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002230981,0.001377396,0.14450137,0.00015620474,0.00003679064,0.000078038494,0.30218324,0.006192955,0.0003030196,0.004124143,0.00019160197,0.5406321],"study_design_scores_gemma":[0.008472476,0.0015114237,0.17841077,0.0038782838,0.00009153344,0.00001495828,0.6064454,0.19201088,0.00016594911,0.005377881,0.0019332223,0.0016872793],"about_ca_topic_score_codex":0.0006229529,"about_ca_topic_score_gemma":0.0006985557,"teacher_disagreement_score":0.53894484,"about_ca_system_score_codex":0.0001671823,"about_ca_system_score_gemma":0.00021319684,"threshold_uncertainty_score":0.56132686},"labels":[],"label_agreement":null},{"id":"W2984840138","doi":"10.1177/0272989x19884541","title":"SURE Test Accuracy for Decisional Conflict Screening among Parents Making Decisions for Their Child","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Ethics and Legal Issues in Pediatric Healthcare","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Université Laval; Ottawa Hospital; Western University","funders":"Canadian Institutes of Health Research","keywords":"Confidence interval; Medicine; Interquartile range; Test (biology); Statistics; Analysis of variance; Internal medicine; Mathematics","score_opus":0.07942466816868657,"score_gpt":0.4291954191464154,"score_spread":0.3497707509777288,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2984840138","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.35620892,0.0031291381,0.6246626,0.005492445,0.0029180718,0.004484347,0.00034854445,0.00024028939,0.002515685],"genre_scores_gemma":[0.93651646,0.00032634783,0.05585206,0.004936203,0.0017217975,0.00016983792,0.00013727666,0.00011629866,0.00022370869],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9938194,0.00009670597,0.0014961896,0.0010635656,0.0025232988,0.001000832],"domain_scores_gemma":[0.9391626,0.057643585,0.00044412978,0.000974703,0.0009848633,0.0007901342],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004302078,0.00047016243,0.0010251204,0.00040898818,0.0006546406,0.00015185657,0.0007678521,0.0009276288,0.0009866139],"category_scores_gemma":[0.09905502,0.0003482493,0.00060814625,0.00056270947,0.00015819058,0.00023248706,0.00039604484,0.001468297,0.00006791732],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001150058,0.00028232535,0.1372488,0.00028091503,0.000093825336,0.00008154105,0.0005970189,0.0000872137,0.000019428982,0.0017989008,0.007858591,0.8505014],"study_design_scores_gemma":[0.011934746,0.0018753154,0.11533083,0.022664145,0.0002509514,0.00034293524,0.0012962555,0.121054485,0.000063106636,0.012017783,0.7120075,0.0011619045],"about_ca_topic_score_codex":0.000015956228,"about_ca_topic_score_gemma":0.000042301162,"teacher_disagreement_score":0.8493395,"about_ca_system_score_codex":0.00011985573,"about_ca_system_score_gemma":0.00051476544,"threshold_uncertainty_score":0.9999266},"labels":[],"label_agreement":null},{"id":"W2994984937","doi":"10.1177/0272989x19889356","title":"Development and Calibration of a Dynamic HIV Transmission Model for 6 US Cities","year":2019,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AIDS Vancouver; Simon Fraser University","funders":"National Institute on Drug Abuse","keywords":"Atlanta; Replicate; Calibration; Robustness (evolution); Computer science; Face validity; Statistics; Medicine; Mathematics","score_opus":0.032153682435968604,"score_gpt":0.36666736034765113,"score_spread":0.3345136779116825,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2994984937","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39451274,0.00013228826,0.60439354,0.0002819737,0.000015045462,0.00022770083,0.000004424035,0.000010271251,0.00042199678],"genre_scores_gemma":[0.9045099,0.000045687597,0.09072541,0.00010439669,0.000012679956,0.000021069869,0.00002269787,0.000011066388,0.004547105],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99865264,0.00001791467,0.00035428934,0.00016619073,0.0006523145,0.00015666694],"domain_scores_gemma":[0.99940974,0.00020548233,0.000041389765,0.00010771795,0.00007168712,0.0001639944],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004470697,0.00007125966,0.00020143932,0.00014514585,0.00004987289,0.000011649079,0.000068667985,0.00009030273,0.0006202727],"category_scores_gemma":[0.00043814135,0.000051796917,0.00007597457,0.00007928927,0.000043672626,0.00006822792,0.00003878155,0.00011441676,0.0000068926747],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040565367,0.00008781542,0.0006955222,0.00035250903,0.000023014527,0.000005364401,0.00062526274,0.00003539509,0.002836004,0.0005580484,0.002192241,0.99218315],"study_design_scores_gemma":[0.0012088668,0.00033228804,0.0008193105,0.0019822347,0.000010671255,0.000012312697,0.00011466896,0.9905761,0.00040113062,0.0010583398,0.0034300047,0.0000541022],"about_ca_topic_score_codex":0.0000012808897,"about_ca_topic_score_gemma":0.000008840795,"teacher_disagreement_score":0.9921291,"about_ca_system_score_codex":0.00003297697,"about_ca_system_score_gemma":0.00019229575,"threshold_uncertainty_score":0.6791551},"labels":[],"label_agreement":null},{"id":"W2999942009","doi":"10.1177/0272989x19896567","title":"Exploring the Benefits of Transformations in Health Utility Mapping","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University Health Network; Toronto Public Health; Toronto General Hospital; University of Toronto","funders":"","keywords":"Skewness; Econometrics; Statistics; Estimator; Computer science; Kurtosis; Transformation (genetics); Sample size determination; Mathematics","score_opus":0.7041942650901617,"score_gpt":0.4628553782038871,"score_spread":0.24133888688627458,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2999942009","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.52632797,0.0025600605,0.2605523,0.20758387,0.0005441551,0.00064702367,0.00008154034,0.000043963213,0.0016591314],"genre_scores_gemma":[0.9779812,0.00042066202,0.003565906,0.017815761,0.00015230237,0.000045003482,0.0000039912875,0.000012008365,0.0000031367952],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9940809,0.00028865025,0.0047170124,0.000320618,0.00028646615,0.00030636456],"domain_scores_gemma":[0.9959128,0.0025535428,0.0009927881,0.00028819643,0.000030703555,0.0002220026],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.020964522,0.00010837723,0.00069815776,0.00020340829,0.00015657138,0.000027408998,0.00040597183,0.00006682283,0.0006456795],"category_scores_gemma":[0.01338239,0.000104600695,0.000099474295,0.0005249841,0.000057247933,0.0003977085,0.000068844245,0.0003060422,0.0002404934],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007980944,0.0002207916,0.081105225,0.0019360998,0.00005916527,0.0000034667598,0.06688778,0.0058010924,8.1824686e-7,0.2742855,0.011113693,0.55850655],"study_design_scores_gemma":[0.0017922177,0.00014334332,0.5688949,0.0028161881,0.0000028161191,0.00000747026,0.0073818294,0.29882264,0.0000039134343,0.031792957,0.08791394,0.00042777113],"about_ca_topic_score_codex":0.0001799453,"about_ca_topic_score_gemma":0.00013667691,"teacher_disagreement_score":0.5580788,"about_ca_system_score_codex":0.0001460956,"about_ca_system_score_gemma":0.00016287474,"threshold_uncertainty_score":0.9949283},"labels":[],"label_agreement":null},{"id":"W3003342603","doi":"10.1177/0272989x19893973","title":"A Multidimensional Array Representation of State-Transition Model Dynamics","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"demographic modeling and climate adaptation","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"Graduate School of Public Health, University of Pittsburgh; National Cancer Institute; Erasmus Universitair Medisch Centrum Rotterdam; National Institutes of Health; Hospital for Sick Children; National Institute of Allergy and Infectious Diseases; University of Pittsburgh; University of Toronto; University of Minnesota","keywords":"TRACE (psycholinguistics); Occupancy; Computer science; Representation (politics); Cohort; Transition (genetics); Econometrics; Statistics; Mathematics; Engineering; Chemistry","score_opus":0.11931891840557238,"score_gpt":0.40860248214655187,"score_spread":0.2892835637409795,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3003342603","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.33380547,0.000025870297,0.66356325,0.0018654764,0.00013667067,0.00009177019,0.000019211997,0.000039212617,0.0004530309],"genre_scores_gemma":[0.93565106,0.000031866068,0.06281398,0.0013949205,0.000045429108,0.000004974967,0.000020380341,0.000014299809,0.000023059783],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99272615,0.00020324052,0.0013195005,0.0006089575,0.0049175825,0.00022455177],"domain_scores_gemma":[0.9962469,0.0021894984,0.00036899044,0.0003622528,0.0005472455,0.0002851543],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0024088253,0.00014994341,0.00039215194,0.00025852447,0.00012146546,0.000062476676,0.00047148197,0.00014512493,0.0005426415],"category_scores_gemma":[0.0096260095,0.000115189745,0.00020927447,0.001227281,0.00012124416,0.00029317802,0.00008920985,0.00025457802,0.00009756717],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00048057584,0.000100722966,0.0009903297,0.000012494245,0.000018019067,0.000023667219,0.0023128076,0.43741712,0.0024848,0.0026861618,0.0015396026,0.5519337],"study_design_scores_gemma":[0.00048013634,0.00006224464,0.0005901007,0.00010900208,0.000010623424,0.0000047008484,0.0006008264,0.927928,0.00018764229,0.06986234,0.00005507867,0.00010933968],"about_ca_topic_score_codex":0.000010825771,"about_ca_topic_score_gemma":0.000027456272,"teacher_disagreement_score":0.6018456,"about_ca_system_score_codex":0.000026826381,"about_ca_system_score_gemma":0.00012425888,"threshold_uncertainty_score":0.99871635},"labels":[],"label_agreement":null},{"id":"W3016378832","doi":"10.1177/0272989x20912402","title":"Calculating the Expected Value of Sample Information in Practice: Considerations from 3 Case Studies","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":44,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"National Center for Advancing Translational Sciences; NIH Clinical Center; National Cancer Institute; Medical Research Council; National Institute for Health and Care Research; Canadian Institutes of Health Research; Norges Forskningsråd; National Institute of Allergy and Infectious Diseases; Stanford University","keywords":"Monte Carlo method; Sample size determination; Computer science; Range (aeronautics); Sample (material); Computation; Observational study; Mathematical optimization; Econometrics; Statistics; Data mining; Algorithm; Mathematics","score_opus":0.42392029892803645,"score_gpt":0.5099003153606245,"score_spread":0.08598001643258807,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3016378832","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6662821,0.0024310776,0.21600421,0.11293389,0.0006199033,0.00072735024,0.00025506763,0.000049213286,0.0006972072],"genre_scores_gemma":[0.92735624,0.000054804375,0.037516583,0.03484655,0.00017223979,0.000032365573,0.00001226653,0.000008276256,6.63628e-7],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99485725,0.00045899424,0.0039855144,0.00024259034,0.00027145704,0.00018416955],"domain_scores_gemma":[0.96245027,0.034976278,0.0019609493,0.00031244787,0.00016123343,0.00013883789],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009719587,0.00010941543,0.00060487085,0.00014583503,0.00020737943,0.00006455335,0.00015080214,0.00011592203,0.00063016283],"category_scores_gemma":[0.28549534,0.0001012722,0.00006544118,0.00028840514,0.000084372456,0.0007306066,0.00012182401,0.00026579507,0.00019408029],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017912366,0.00023786385,0.060654055,0.00047672153,0.00052988867,0.0003895251,0.20252264,0.02548894,0.00000389588,0.6300204,0.050752766,0.028744232],"study_design_scores_gemma":[0.0023669146,0.00011668056,0.007985048,0.00081639725,0.00003155003,0.00019004675,0.06830341,0.7573602,0.000005978273,0.13936004,0.023022318,0.00044137274],"about_ca_topic_score_codex":0.0011356017,"about_ca_topic_score_gemma":0.00024185769,"teacher_disagreement_score":0.7318713,"about_ca_system_score_codex":0.00016987378,"about_ca_system_score_gemma":0.00015602383,"threshold_uncertainty_score":0.72052324},"labels":[],"label_agreement":null},{"id":"W3019206888","doi":"10.1177/0272989x20915241","title":"Framework for Drug Formulary Decision Using Multiple-Criteria Decision Analysis","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Formulary; Multiple-criteria decision analysis; Computer science; Decision analysis; Portfolio; Set (abstract data type); Preference; Risk analysis (engineering); Management science; Operations research; Actuarial science; Medicine; Mathematics; Economics; Statistics; Family medicine","score_opus":0.4249918625366504,"score_gpt":0.5139184468295667,"score_spread":0.08892658429291628,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3019206888","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2426571,0.0012550217,0.7459351,0.008216809,0.0009673172,0.0006654893,0.00014633067,0.00008351853,0.00007332648],"genre_scores_gemma":[0.53912437,0.000063307256,0.43916982,0.02076221,0.00075045263,0.000041178708,0.000028328797,0.00005236398,0.000007933588],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99061424,0.000241229,0.0062826257,0.0013558229,0.0007537813,0.0007523134],"domain_scores_gemma":[0.9724109,0.023611527,0.0020083135,0.0009032127,0.00017686242,0.00088915974],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.018634418,0.0003875381,0.001991886,0.00084898825,0.00052846444,0.00027006224,0.0009771066,0.00054204295,0.0033637304],"category_scores_gemma":[0.1153344,0.0004272283,0.0007473018,0.0014032725,0.00008235666,0.0006081269,0.00035248642,0.0004951079,0.0008505684],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002994768,0.0007488311,0.19185387,0.0011141982,0.0019106655,0.00010634571,0.009410839,0.02315578,0.000030869618,0.12862575,0.14885534,0.49119273],"study_design_scores_gemma":[0.0012253252,0.00006580503,0.0038241118,0.0006848605,0.000070287104,0.0000050059225,0.00028282305,0.8004293,0.000002963683,0.16591617,0.02706393,0.0004294038],"about_ca_topic_score_codex":0.00007026787,"about_ca_topic_score_gemma":0.0000898713,"teacher_disagreement_score":0.77727354,"about_ca_system_score_codex":0.00045682708,"about_ca_system_score_gemma":0.00019208399,"threshold_uncertainty_score":0.9999274},"labels":[],"label_agreement":null},{"id":"W3026079137","doi":"10.1177/0272989x20911870","title":"20th Anniversary Ottawa Decision Support Framework: Part 3 Overview of Systematic Reviews and Updated Framework","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":380,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université de Sherbrooke; Western University; Ottawa Hospital; Université Laval; University of Ottawa","funders":"","keywords":"Psychological intervention; Decision aids; Decision quality; Decision support system; Quality (philosophy); Systematic review; Clinical decision support system; Information needs; Scale (ratio); Psychology; MEDLINE; Medicine; Applied psychology; Computer science; Nursing; Patient satisfaction; Alternative medicine; Data mining; Political science","score_opus":0.37610958092090724,"score_gpt":0.4976828657435156,"score_spread":0.12157328482260837,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3026079137","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03288069,0.24228014,0.6591511,0.033291414,0.00779519,0.018898001,0.0002080232,0.00081447535,0.0046809986],"genre_scores_gemma":[0.71571434,0.023827778,0.21372633,0.045267045,0.0006507375,0.00052974257,0.00007503105,0.00012824418,0.00008077119],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9898114,0.003188897,0.0037755754,0.00066689344,0.0019329223,0.00062429934],"domain_scores_gemma":[0.97487324,0.020232849,0.0017657793,0.0017245159,0.00047977004,0.0009238491],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0033520937,0.00034674958,0.0016291465,0.00018212399,0.0006194272,0.000021376096,0.0012231853,0.00093502906,0.006762533],"category_scores_gemma":[0.07628648,0.0002660643,0.0001984724,0.001125219,0.00016718436,0.00022876565,0.0012665841,0.0022897464,0.0007373478],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008671288,0.000335794,0.05806285,0.14496513,0.00017148617,0.00019263766,0.033328556,0.00001876242,0.000011832063,0.030661097,0.3399581,0.39142665],"study_design_scores_gemma":[0.0021874704,0.0006632792,0.0074356995,0.5916724,0.00026692092,0.000027694728,0.008612429,0.01232188,0.0000042472475,0.048638992,0.32722366,0.0009452794],"about_ca_topic_score_codex":0.000019753488,"about_ca_topic_score_gemma":0.00001601014,"teacher_disagreement_score":0.6828336,"about_ca_system_score_codex":0.0001376095,"about_ca_system_score_gemma":0.00047208345,"threshold_uncertainty_score":0.99997914},"labels":[],"label_agreement":null},{"id":"W3026084796","doi":"10.1177/0272989x20904955","title":"What Helps People Make Values-Congruent Medical Decisions? Eleven Strategies Tested across 6 Studies","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Psychology; Congruence (geometry); Harm; Decision quality; Outcome (game theory); Quality (philosophy); Social psychology; Actuarial science; Medicine; Patient satisfaction; Mathematics; Economics","score_opus":0.3472612559997576,"score_gpt":0.5336396425117766,"score_spread":0.18637838651201905,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3026084796","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7633396,0.042147223,0.026634771,0.14787684,0.01382896,0.0036382752,0.000093353876,0.0013450561,0.0010959449],"genre_scores_gemma":[0.95606923,0.006556178,0.005383797,0.030338848,0.00095521106,0.00043190084,0.00004909797,0.0000959481,0.000119811775],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9863473,0.002562865,0.0029623895,0.0010931108,0.0055487235,0.0014856223],"domain_scores_gemma":[0.956836,0.037228487,0.00083720696,0.0017401478,0.0013954988,0.0019626934],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0033980838,0.00052016694,0.001230111,0.0001661189,0.0024100845,0.00018641278,0.002449951,0.0009945316,0.0033075023],"category_scores_gemma":[0.120900325,0.00041149108,0.0002445013,0.0012390565,0.0004337782,0.000808274,0.0036303238,0.0031973198,0.001437808],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00080431794,0.000294506,0.030161442,0.0007557187,0.00021844247,0.00040456062,0.12721682,0.00009165113,0.000026084936,0.002835568,0.043835625,0.7933553],"study_design_scores_gemma":[0.007295247,0.000865886,0.04018645,0.040373635,0.00009603396,0.00008030015,0.76659673,0.033552185,0.00001915579,0.038533363,0.07093163,0.0014693873],"about_ca_topic_score_codex":0.00010540288,"about_ca_topic_score_gemma":0.00263047,"teacher_disagreement_score":0.79188585,"about_ca_system_score_codex":0.00044961518,"about_ca_system_score_gemma":0.0025841512,"threshold_uncertainty_score":0.9998337},"labels":[],"label_agreement":null},{"id":"W3028037905","doi":"10.1177/0272989x20915452","title":"The Theoretical Relationship between Sample Size and Expected Predictive Precision for EQ-5D Valuation Studies: A Mathematical Exploration and Simulation Study","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Centre for Applied Research in Cancer Control; Public Health Ontario; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"","keywords":"Homoscedasticity; Sample size determination; Heteroscedasticity; Ordinary least squares; Mathematics; Mean squared error; Statistics; Valuation (finance); Econometrics; Economics","score_opus":0.6966471250686277,"score_gpt":0.519689248180606,"score_spread":0.17695787688802167,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028037905","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29356834,0.0005770393,0.6848945,0.018874047,0.00013653607,0.0018158546,0.000049953378,0.000050418443,0.000033336095],"genre_scores_gemma":[0.98292655,0.00006487306,0.014947163,0.0012598058,0.00044619368,0.00030964203,0.000011350763,0.000028314218,0.0000061400415],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99465686,0.0007612389,0.0031403592,0.0006687691,0.0004959183,0.0002768383],"domain_scores_gemma":[0.8231134,0.17487794,0.001114455,0.00034730742,0.00022560792,0.0003212886],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.021468725,0.00019602252,0.0007495088,0.000107910804,0.0007441235,0.0001811333,0.00019852343,0.00019575535,0.00009246784],"category_scores_gemma":[0.40554938,0.00016662285,0.00006766758,0.0002520895,0.00023498041,0.0004794912,0.0001668902,0.0002483558,0.000051622083],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010866312,0.0003677747,0.26705888,0.00057851104,0.0003953191,0.0000026120315,0.08070972,0.004323703,0.0000010641479,0.58906186,0.002578306,0.053835634],"study_design_scores_gemma":[0.001023445,0.0003746628,0.045576427,0.00014745256,0.000025950814,6.412271e-7,0.005876473,0.35704866,2.0883846e-7,0.589548,0.00025285027,0.00012523105],"about_ca_topic_score_codex":0.000003300816,"about_ca_topic_score_gemma":0.000008149938,"teacher_disagreement_score":0.6893582,"about_ca_system_score_codex":0.00017217049,"about_ca_system_score_gemma":0.00007153225,"threshold_uncertainty_score":0.74406743},"labels":[],"label_agreement":null},{"id":"W3032948830","doi":"10.1177/0272989x20925368","title":"Decision Regret among Informal Caregivers Making Housing Decisions for Older Adults with Cognitive Impairment: A Cross-sectional Analysis","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Geriatric Care and Nursing Homes","field":"Health Professions","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Regret; Psychology; Preference; Gerontology; Cognition; Cross-sectional study; Multilevel model; Medicine; Psychiatry","score_opus":0.036090157266998275,"score_gpt":0.4036313623881015,"score_spread":0.3675412051211032,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3032948830","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.625275,0.00011530071,0.37193453,0.00018285465,0.00081295054,0.0010300948,0.0000827489,0.00015260982,0.0004138675],"genre_scores_gemma":[0.9848235,0.000035242287,0.012678629,0.0012856534,0.000763251,0.00016095444,0.00013351816,0.00006861011,0.000050587183],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9935576,0.0002104445,0.0015106539,0.001030441,0.0026120592,0.0010788324],"domain_scores_gemma":[0.9884825,0.008621466,0.00069272023,0.00039516747,0.0010759752,0.0007322126],"candidate_categories":["metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0016262761,0.0004467131,0.00087201427,0.0007345434,0.0020579572,0.00015690722,0.0005406643,0.0006869007,0.0023140665],"category_scores_gemma":[0.007745996,0.00034248206,0.00057441165,0.0024631917,0.00025312693,0.00048437912,0.00042709784,0.0011112409,0.000087455934],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013051912,0.00012360878,0.6438737,0.00015062961,0.0005169158,0.00019972675,0.019109363,0.0007238299,0.0000010233058,0.0000593963,0.005548426,0.31664145],"study_design_scores_gemma":[0.009430025,0.0005958407,0.93240255,0.0064856275,0.0006312298,0.0000164847,0.0163394,0.03179456,0.0000033663266,0.0002383511,0.0014727506,0.00058983534],"about_ca_topic_score_codex":0.000060656206,"about_ca_topic_score_gemma":0.00055707776,"teacher_disagreement_score":0.3595485,"about_ca_system_score_codex":0.0002772978,"about_ca_system_score_gemma":0.0006509845,"threshold_uncertainty_score":0.9999027},"labels":[],"label_agreement":null},{"id":"W3033258331","doi":"10.1177/0272989x20924019","title":"Using Both Time Tradeoff and Discrete Choice Experiments in Valuing the EQ-5D: Impact of Model Misspecification on Value Sets","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; Public Health Ontario; SickKids Foundation; University of Toronto; McMaster University; Institute for Clinical Evaluative Sciences; Impact","funders":"EuroQol Research Foundation","keywords":"Homoscedasticity; Heteroscedasticity; Econometrics; Tobit model; Mathematics; Valuation (finance); Linear model; Statistics; Computer science; Economics; Accounting","score_opus":0.6346022759749333,"score_gpt":0.5158828980732656,"score_spread":0.11871937790166776,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3033258331","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9166965,0.00095038244,0.0729032,0.008069243,0.00011151273,0.00045838582,0.000051416187,0.000019494439,0.00073988637],"genre_scores_gemma":[0.9891537,0.000047379966,0.006651144,0.003948774,0.00014456574,0.000011941522,0.0000054057077,0.000027177914,0.000009937306],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9960746,0.00030127814,0.0024931172,0.000506099,0.00033430016,0.00029061685],"domain_scores_gemma":[0.9961992,0.0022304908,0.0009817788,0.0003371586,0.000020450585,0.00023093258],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0079381075,0.0001723421,0.00068103307,0.0001898545,0.00012814565,0.00005939212,0.00034013911,0.00015219332,0.00035516027],"category_scores_gemma":[0.007966839,0.00015228566,0.00011692745,0.00024548746,0.00007781788,0.00023983432,0.00008982384,0.00026553156,0.000098471806],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011243699,0.0009855436,0.17459461,0.00097675,0.00050833885,0.00003099422,0.06355471,0.5790114,0.001738057,0.08884827,0.024557551,0.06406945],"study_design_scores_gemma":[0.00066174724,0.00007388233,0.026014721,0.00036456497,0.000003739065,0.0000026048726,0.00017587116,0.9663933,0.000008246836,0.0059185103,0.00023391044,0.00014892561],"about_ca_topic_score_codex":0.0001533386,"about_ca_topic_score_gemma":0.0000046045393,"teacher_disagreement_score":0.3873819,"about_ca_system_score_codex":0.0003460761,"about_ca_system_score_gemma":0.00013561467,"threshold_uncertainty_score":0.9537626},"labels":[],"label_agreement":null},{"id":"W3033624969","doi":"10.1177/0272989x20924645","title":"20th Anniversary Update of the Ottawa Decision Support Framework: Part 2 Subanalysis of a Systematic Review of Patient Decision Aids","year":2020,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":78,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"University of Ottawa","keywords":"Medicine; Decision aids; Confidence interval; Relative risk; Clinical trial; Randomized controlled trial; MEDLINE; Health care; Meta-analysis; Physical therapy; Alternative medicine; Internal medicine","score_opus":0.16002458941674766,"score_gpt":0.47306795577202454,"score_spread":0.31304336635527685,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3033624969","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000026504054,0.9801794,0.010892824,0.00036696577,0.0017058811,0.0062620565,0.00031616286,0.000044882283,0.00020526817],"genre_scores_gemma":[0.00075968774,0.9747784,0.021705534,0.001970878,0.000095960386,0.00046072013,0.00010333575,0.00011064232,0.00001483354],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9708648,0.00800246,0.013209714,0.0010051926,0.006188423,0.0007294347],"domain_scores_gemma":[0.9419057,0.03705022,0.012987599,0.005657134,0.0018052406,0.00059411343],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.00571056,0.000802531,0.008603475,0.000743131,0.00049756956,0.000010723506,0.004653834,0.0015694452,0.0035888916],"category_scores_gemma":[0.09377032,0.00048339405,0.0024327007,0.0038546328,0.0003473078,0.00015438412,0.003611796,0.0032880318,0.00020547166],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003820427,0.00008286806,0.00010896523,0.5060262,0.00016848287,0.000012350302,0.0004273709,0.0000012687311,1.437128e-8,0.00029435466,0.005867013,0.48697287],"study_design_scores_gemma":[0.00023638998,0.00015363301,0.000008743315,0.7301052,0.0020546173,0.000010678253,0.00046647532,0.000113401584,2.0935025e-7,0.0010865452,0.26551685,0.00024723247],"about_ca_topic_score_codex":0.00004035662,"about_ca_topic_score_gemma":0.000028667844,"teacher_disagreement_score":0.48672566,"about_ca_system_score_codex":0.0005713097,"about_ca_system_score_gemma":0.0036278425,"threshold_uncertainty_score":0.99976176},"labels":[],"label_agreement":null},{"id":"W3038915840","doi":"10.1177/0272989x20937257","title":"A Systematic Review of Methods Used for Confounding Adjustment in Observational Economic Evaluations in Cardiology Conducted between 2013 and 2017","year":2020,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; Programs for Assessment of Technology in Health Research Institute; McMaster University; Centre Hospitalier de l’Université de Montréal; Impact; McGill University; Université de Montréal; Université Laval","funders":"Canadian Institutes of Health Research","keywords":"Observational study; Confounding; Propensity score matching; Medicine; Systematic review; Cochrane Library; MEDLINE; Matching (statistics); Randomized controlled trial; Internal medicine; Pathology","score_opus":0.7375894583404606,"score_gpt":0.6331474688234388,"score_spread":0.10444198951702188,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3038915840","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000030489136,0.95461905,0.03555795,0.002630804,0.00068588526,0.0058632484,0.00053630385,0.000013749433,0.00006250462],"genre_scores_gemma":[0.000030824398,0.9761473,0.01992496,0.0016302512,0.00031444003,0.0016414624,0.0002494806,0.000048587666,0.000012682938],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9813589,0.00310748,0.013948749,0.0009370944,0.00024218315,0.00040560134],"domain_scores_gemma":[0.9588474,0.03406207,0.0061847107,0.00058976025,0.00008882969,0.00022722913],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.067955,0.00040521906,0.009934577,0.00089221017,0.00007104223,0.000036372283,0.00061117124,0.0005900218,0.00030003366],"category_scores_gemma":[0.07132279,0.00042148933,0.00048747248,0.0003648885,0.00010345539,0.00021035621,0.00018377531,0.0004822357,0.00015844498],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004443999,0.000022358448,0.00043671043,0.9284409,0.00033785266,0.000003009846,0.000175079,0.000019386742,1.3308477e-8,0.016501622,0.0046082884,0.049450282],"study_design_scores_gemma":[0.0010699946,0.0000723118,0.0004934241,0.79852164,0.00041226982,0.000019573376,0.00008632546,0.0070082378,9.786534e-9,0.0111083565,0.18069737,0.00051047135],"about_ca_topic_score_codex":0.00008649368,"about_ca_topic_score_gemma":0.00007199363,"teacher_disagreement_score":0.17608908,"about_ca_system_score_codex":0.0014122743,"about_ca_system_score_gemma":0.001262926,"threshold_uncertainty_score":0.9998237},"labels":[],"label_agreement":null},{"id":"W3042390659","doi":"10.1177/0272989x20936209","title":"20th Anniversary Update of the Ottawa Decision Support Framework Part 1: A Systematic Review of the Decisional Needs of People Making Health or Social Decisions","year":2020,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":154,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; Western University; University of Ottawa","funders":"","keywords":"Inclusion (mineral); Psychology; Information needs; Decision quality; Decision aids; Systematic review; Quality (philosophy); Needs assessment; Applied psychology; Social psychology; MEDLINE; Medicine; Nursing; Alternative medicine; Patient satisfaction; Computer science; Sociology","score_opus":0.22945663983648734,"score_gpt":0.499710372327355,"score_spread":0.27025373249086765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3042390659","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000012937598,0.97057277,0.01113474,0.00481283,0.0034384844,0.009067894,0.00069993455,0.000059112434,0.00020129728],"genre_scores_gemma":[0.0012040843,0.97895956,0.010338072,0.008345681,0.00031839884,0.0005531683,0.00006611786,0.00016186373,0.000053052598],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.96643925,0.011656537,0.012786939,0.0008661305,0.0071786926,0.0010724673],"domain_scores_gemma":[0.91442955,0.0631541,0.015358667,0.004861786,0.0016186516,0.00057726476],"candidate_categories":["metaresearch","metaepi_narrow","sts","open_science","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.008707007,0.0009380824,0.008320584,0.00065560004,0.0018838923,0.000020052763,0.0075476235,0.0016229665,0.0031687787],"category_scores_gemma":[0.12089552,0.00048116877,0.0025039068,0.006237371,0.000535088,0.00016863461,0.0061062565,0.004843299,0.00013567523],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000104773215,0.00013730653,0.000115656825,0.5329814,0.00016378959,0.000007024862,0.0029792632,0.0000012809438,1.5827508e-8,0.0018552886,0.043488707,0.4181655],"study_design_scores_gemma":[0.0003561658,0.00009780629,0.000077853416,0.7316453,0.00068801775,0.000036431113,0.0016944583,0.00006964607,3.482782e-8,0.0016864617,0.26337737,0.0002704642],"about_ca_topic_score_codex":0.000038090046,"about_ca_topic_score_gemma":0.000132655,"teacher_disagreement_score":0.41789502,"about_ca_system_score_codex":0.00090576493,"about_ca_system_score_gemma":0.014528699,"threshold_uncertainty_score":0.999764},"labels":[],"label_agreement":null},{"id":"W3043284836","doi":"10.1177/0272989x20938056","title":"The Effect of Information Formats and Incidental Affect on Prior and Posterior Probability Judgments","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Metropolitan University; St. Michael's Hospital","funders":"Natural Sciences and Engineering Research Council of Canada; Ontario Ministry of Research, Innovation and Science; Canada Research Chairs","keywords":"Computer science; Bayesian probability; Test (biology); Table (database); Bayesian inference; Task (project management); Posterior probability; Inference; Probabilistic logic; Psychology; Artificial intelligence; Machine learning; Natural language processing; Information retrieval; Statistics; Data mining; Mathematics","score_opus":0.04839277711880483,"score_gpt":0.37641689393958977,"score_spread":0.3280241168207849,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3043284836","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99630004,0.00007177826,0.0015701678,0.0009690963,0.00035914042,0.00046053904,0.000011148716,0.00002397594,0.00023410238],"genre_scores_gemma":[0.99849737,0.000021872207,0.0007797968,0.0006444735,0.00003592207,0.000010306116,9.949148e-7,0.0000067718684,0.000002509555],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99590194,0.00028590788,0.0011283943,0.00036459585,0.0020913132,0.000227864],"domain_scores_gemma":[0.9879865,0.010818681,0.000428015,0.00038801873,0.00009362021,0.00028514984],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006296173,0.00018447729,0.00043095363,0.00012875051,0.00032166063,0.00047805696,0.00063322193,0.00012422293,0.00009770114],"category_scores_gemma":[0.03220035,0.00009611679,0.00010135607,0.00031348754,0.00024044784,0.00062428735,0.00071637443,0.00026666335,0.00008632362],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006730668,0.000009665932,0.016188286,0.000012226192,0.0000039819674,0.0000060384687,0.00024658945,0.0000081466205,0.000023368724,0.000031433043,0.00047925906,0.9823179],"study_design_scores_gemma":[0.012955771,0.016846113,0.663988,0.0031812282,0.00016978131,0.00065899524,0.0021172313,0.11052173,0.0025477158,0.15274324,0.03269305,0.00157718],"about_ca_topic_score_codex":0.0000030596061,"about_ca_topic_score_gemma":0.000009220947,"teacher_disagreement_score":0.9807408,"about_ca_system_score_codex":0.000039029917,"about_ca_system_score_gemma":0.000043220083,"threshold_uncertainty_score":0.97595185},"labels":[],"label_agreement":null},{"id":"W3043915800","doi":"10.1177/0272989x20940673","title":"Implausible States: Prevalence of EQ-5D-5L States in the General Population and Its Effect on Health State Valuation","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Valuation (finance); Population; EQ-5D; Medicine; Actuarial science; Econometrics; Mathematics; Economics; Disease; Environmental health; Accounting","score_opus":0.3310405290731772,"score_gpt":0.46515360280704693,"score_spread":0.13411307373386971,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3043915800","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94837654,0.0018795031,0.01867035,0.029748404,0.00021433238,0.00090348424,0.0001370106,0.000021013533,0.000049336522],"genre_scores_gemma":[0.9797481,0.0007677483,0.0012226911,0.018026492,0.0001230985,0.000042680036,0.000038352657,0.000018544251,0.000012330561],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9945678,0.001035624,0.0030652422,0.0005096559,0.00049199763,0.00032969844],"domain_scores_gemma":[0.9939114,0.004229438,0.001380895,0.00024495416,0.000041225667,0.0001921074],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.026550092,0.00016308248,0.0007121939,0.00022590124,0.00013273871,0.000055333912,0.00028782332,0.00009504462,0.0002333955],"category_scores_gemma":[0.010341176,0.00014346902,0.000057984253,0.0003191619,0.000034234352,0.00025214624,0.00006895771,0.00026030137,0.00012882266],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007807524,0.0003768453,0.5648191,0.007537573,0.00009585692,0.00001964577,0.03707681,0.03821543,0.000013086567,0.055449124,0.021170327,0.27444544],"study_design_scores_gemma":[0.0017939726,0.0013282966,0.40866235,0.0011862644,0.000006412301,0.000006341823,0.00033716692,0.53377295,0.0000076042384,0.050590705,0.0020234261,0.0002845176],"about_ca_topic_score_codex":0.00035485713,"about_ca_topic_score_gemma":0.00012838587,"teacher_disagreement_score":0.49555752,"about_ca_system_score_codex":0.00021559373,"about_ca_system_score_gemma":0.000107353626,"threshold_uncertainty_score":0.99799514},"labels":[],"label_agreement":null},{"id":"W3043995085","doi":"10.1177/0272989x20940999","title":"Use of Enhanced Data Visualization to Improve Patient Judgments about Hypertension Control","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Blood Pressure and Hypertension Studies","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; University of Toronto","funders":"National Institute of Diabetes and Digestive and Kidney Diseases; Agency for Healthcare Research and Quality","keywords":"Smoothing; Visualization; Blood pressure; Data visualization; Medicine; Computer science; Data mining; Internal medicine; Computer vision","score_opus":0.10625747278756788,"score_gpt":0.35268532183207907,"score_spread":0.2464278490445112,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3043995085","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5205889,0.0024580446,0.46370533,0.009915718,0.0011400401,0.0015740236,0.00012576512,0.00016594557,0.0003262613],"genre_scores_gemma":[0.94687504,0.00011585421,0.0050913566,0.04765902,0.0001876114,0.0000090220965,0.000028695109,0.00002299203,0.000010436254],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9970396,0.000065854416,0.00066631124,0.0005348807,0.0014682551,0.00022509159],"domain_scores_gemma":[0.9978751,0.0006367013,0.00015075473,0.00058357103,0.00035601004,0.00039784572],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00027995245,0.00016463635,0.00069786335,0.00008847826,0.00007483398,0.000021088475,0.00022261874,0.00014017287,0.00025714614],"category_scores_gemma":[0.0060511623,0.0001177293,0.00006926802,0.00029526657,0.000048281876,0.00013492405,0.00057893957,0.0001774292,0.000058879803],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016536139,0.00030521367,0.0011815891,0.00010067521,0.00026253026,0.00020335014,0.00043880395,0.0000260833,0.056098517,0.00004039967,0.07934704,0.8603422],"study_design_scores_gemma":[0.02174496,0.0075153513,0.037806787,0.011484085,0.0041521676,0.00015801146,0.0008227626,0.24296162,0.044922303,0.00019825417,0.62676847,0.0014652138],"about_ca_topic_score_codex":0.000011197618,"about_ca_topic_score_gemma":0.000003591205,"teacher_disagreement_score":0.85887694,"about_ca_system_score_codex":0.000008157891,"about_ca_system_score_gemma":0.00006951555,"threshold_uncertainty_score":0.72442436},"labels":[],"label_agreement":null},{"id":"W3044981225","doi":"10.1177/0272989x20940660","title":"What Matters to Patients and Families: A Content and Process Framework for Clarifying Preferences, Concerns, and Values","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; Jewish General Hospital; Université Laval","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research","keywords":"Psychology; Content analysis; Sociocultural evolution; Social psychology; Medicine; Sociology","score_opus":0.3791259558971669,"score_gpt":0.49493426248568184,"score_spread":0.11580830658851493,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3044981225","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9119346,0.0051165204,0.03831641,0.04082025,0.0008912042,0.0027560976,0.000035203844,0.000097009324,0.000032710333],"genre_scores_gemma":[0.9189586,0.0015332813,0.021614334,0.057327278,0.00013638898,0.000388315,0.000011282341,0.000025845127,0.000004644414],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9974078,0.00033548728,0.0007245536,0.0004960488,0.0006598749,0.00037625225],"domain_scores_gemma":[0.99339455,0.005193317,0.0002348889,0.00024724117,0.00030398622,0.00062599086],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0004081949,0.00017574686,0.00036651595,0.000076496435,0.00071283465,0.00007744611,0.00029255875,0.00033598937,0.00006248304],"category_scores_gemma":[0.008340661,0.00014278576,0.000021924978,0.00016464657,0.00013238462,0.00029043,0.0005958541,0.0007386035,0.000008626454],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005943936,0.000037625745,0.3413148,0.0021246085,0.000024612047,0.0000021254205,0.08483921,0.0000022572938,0.0000049892487,0.0012198538,0.0034917444,0.5663438],"study_design_scores_gemma":[0.0121662095,0.0027030187,0.28547335,0.07492666,0.00014842792,0.000005267943,0.43667048,0.0261183,0.000023678536,0.11313494,0.046908777,0.0017208741],"about_ca_topic_score_codex":0.000040668572,"about_ca_topic_score_gemma":0.000041850406,"teacher_disagreement_score":0.56462294,"about_ca_system_score_codex":0.000052714324,"about_ca_system_score_gemma":0.00014516053,"threshold_uncertainty_score":0.99851525},"labels":[],"label_agreement":null},{"id":"W3047388302","doi":"10.1177/0272989x20945308","title":"Dissemination Science to Advance the Use of Simulation Modeling: Our Obligation Moving Forward","year":2020,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AIDS Vancouver; Simon Fraser University","funders":"National Institute on Drug Abuse","keywords":"Obligation; Computer science; Management science; Political science; Economics; Law","score_opus":0.45949069402512926,"score_gpt":0.6851882684662697,"score_spread":0.22569757444114041,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3047388302","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031664253,0.000010286643,0.70820475,0.00828236,0.2787512,0.0013533892,0.000099757905,0.00007113479,0.000060725215],"genre_scores_gemma":[0.6750863,0.000052095034,0.028668026,0.008986256,0.28661487,0.00027494688,0.000060471342,0.00008434199,0.0001726767],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98635614,0.0008028787,0.0024289913,0.00088227756,0.008661799,0.0008679163],"domain_scores_gemma":[0.9574429,0.036767077,0.0014139552,0.0008342304,0.0027944143,0.00074743753],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.010343194,0.00029097308,0.0005536302,0.00061246264,0.0016190378,0.000107063366,0.0014702657,0.00062572764,0.00014059723],"category_scores_gemma":[0.38179854,0.00021427353,0.000093586554,0.002659237,0.0001387818,0.00088154536,0.00093562296,0.0017642255,0.00016280993],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000119513374,0.000013585866,0.00011165622,0.0002213104,0.0000046511154,0.0000029908504,0.006161403,0.07577984,0.00007815104,0.00018999462,0.7827177,0.13459921],"study_design_scores_gemma":[0.00022395152,0.00005793405,0.00013373562,0.002643467,0.000012965303,1.5698156e-7,0.00062562805,0.5544332,0.0000046133073,0.0005404302,0.44116822,0.00015564683],"about_ca_topic_score_codex":0.000067951696,"about_ca_topic_score_gemma":0.00011008055,"teacher_disagreement_score":0.6795367,"about_ca_system_score_codex":0.0008025488,"about_ca_system_score_gemma":0.0045594615,"threshold_uncertainty_score":0.9996807},"labels":[],"label_agreement":null},{"id":"W3080367820","doi":"10.1177/0272989x20944875","title":"Optimal Allocation of Research Funds under a Budget Constraint","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Budget constraint; Sample (material); Portfolio; Stylized fact; Population; Constraint (computer-aided design); Investment (military); Computer science; Sample size determination; Point (geometry); Economics; Microeconomics; Actuarial science; Finance; Statistics","score_opus":0.5850875478991512,"score_gpt":0.5434805942254187,"score_spread":0.04160695367373246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3080367820","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2537511,0.0011990891,0.6067275,0.12899493,0.00054246775,0.00055491104,0.00007113967,0.000067631445,0.008091236],"genre_scores_gemma":[0.9671367,0.00006050494,0.0184933,0.013789507,0.00041003473,0.000026444433,0.000010043595,0.000022740503,0.000050765862],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9943748,0.0004411043,0.003562709,0.0005601626,0.0006477387,0.0004134503],"domain_scores_gemma":[0.99317724,0.00494987,0.0008726478,0.00038312317,0.00018550272,0.00043161475],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.030306919,0.0001262162,0.00072582735,0.00033278926,0.00015367426,0.0000576143,0.0005017347,0.00024393294,0.0044090506],"category_scores_gemma":[0.036318347,0.0001454077,0.00010969874,0.00051637826,0.00028875214,0.00019534834,0.00018972489,0.00045732319,0.0017378249],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000183362,0.00025976333,0.011716211,0.0006726442,0.0001179595,0.000021120572,0.004924335,0.0040341704,0.000032852007,0.8459668,0.104665674,0.0274051],"study_design_scores_gemma":[0.0045054555,0.0009879702,0.03819914,0.0023183278,0.00001551421,0.000048665213,0.010671045,0.5889784,0.00005854897,0.20323808,0.14992486,0.0010539538],"about_ca_topic_score_codex":0.00006268573,"about_ca_topic_score_gemma":0.00001700554,"teacher_disagreement_score":0.7133856,"about_ca_system_score_codex":0.00023600926,"about_ca_system_score_gemma":0.00036969763,"threshold_uncertainty_score":0.9990394},"labels":[],"label_agreement":null},{"id":"W3089680180","doi":"10.1177/0272989x20951782","title":"A Scoring Algorithm for Deriving Utility Values from the Neuro-QoL for Patients with Multiple Sclerosis","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Biogen","keywords":"Population; Quality of life (healthcare); Item response theory; Construct validity; Multiple sclerosis; Medicine; Psychometrics; Clinical psychology; Psychiatry","score_opus":0.39054031836612574,"score_gpt":0.4051338473780875,"score_spread":0.014593529011961759,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3089680180","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.1924193,0.00031526512,0.7911481,0.013479941,0.00057617493,0.0013599809,0.00061986776,0.000053970143,0.000027347687],"genre_scores_gemma":[0.81674135,0.00002154216,0.15281446,0.029211067,0.00082959863,0.0002785175,0.000047714973,0.000051387284,0.000004365751],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9959993,0.00016134726,0.0023242815,0.0007653366,0.0003282418,0.00042154375],"domain_scores_gemma":[0.97935724,0.01879737,0.0010093703,0.00041779046,0.0001227328,0.00029551797],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0070590232,0.00019930823,0.00073222804,0.000064239975,0.0004816051,0.00014461899,0.0005480253,0.0001429831,0.00027104886],"category_scores_gemma":[0.053740505,0.00017103698,0.00016803316,0.00017402138,0.00008014937,0.0002811979,0.00014841955,0.0002067504,0.00009821705],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00028642645,0.00020885588,0.49045217,0.00019371434,0.00012968454,0.0000010719799,0.0026645628,0.00014596422,0.0000011769595,0.0009577552,0.035496946,0.46946168],"study_design_scores_gemma":[0.0031532813,0.00022276386,0.22803886,0.00058503443,0.000015956628,3.2005184e-7,0.0003561479,0.73347,0.0000028258078,0.009114169,0.02474113,0.00029948863],"about_ca_topic_score_codex":0.00008649016,"about_ca_topic_score_gemma":0.000053090887,"teacher_disagreement_score":0.73332405,"about_ca_system_score_codex":0.00012180826,"about_ca_system_score_gemma":0.00010345626,"threshold_uncertainty_score":0.95423025},"labels":[],"label_agreement":null},{"id":"W3089914228","doi":"10.1177/0272989x20951778","title":"Incorporating Mortality in Health Utility Measures","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Riksbankens Jubileumsfond","keywords":"Expected utility hypothesis; Corollary; Quality-adjusted life year; Outcome (game theory); Subjective expected utility; Quality (philosophy); Function (biology); Scale (ratio); State space; Health Utilities Index; Index (typography); State (computer science); Computer science; Actuarial science; Medicine; Econometrics; Risk analysis (engineering); Mathematics; Statistics; Mathematical economics; Economics; Health related quality of life; Geography","score_opus":0.5748531287029683,"score_gpt":0.5067643710572967,"score_spread":0.06808875764567168,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3089914228","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.40723422,0.0031891067,0.4314102,0.15135121,0.00094709673,0.00091202674,0.00009518549,0.00016352681,0.0046974174],"genre_scores_gemma":[0.9336646,0.000052450196,0.008448618,0.057431653,0.00034130848,0.000024495635,0.000007874344,0.00002133388,0.0000076638735],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9912391,0.0007146325,0.0062873326,0.0007791777,0.0005097746,0.00046998574],"domain_scores_gemma":[0.99490565,0.002066179,0.001927215,0.0004429763,0.000045502402,0.0006125004],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.044893507,0.00018222297,0.0011653528,0.00019757843,0.00017072119,0.00007103199,0.00047625887,0.00019884772,0.0016301112],"category_scores_gemma":[0.05398778,0.00021432448,0.00011581347,0.00048916345,0.0000872816,0.0002945819,0.00016938169,0.0004932549,0.00096885103],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00004668717,0.00013896608,0.8566963,0.00053374394,0.000030150699,0.000024551025,0.0024777171,0.0003470839,5.65817e-7,0.060075168,0.018893141,0.06073594],"study_design_scores_gemma":[0.0016191372,0.00014950188,0.5199305,0.0011115989,0.0000024043695,0.0000071281256,0.00096514507,0.27933487,0.0000011682887,0.14703383,0.04929874,0.00054599677],"about_ca_topic_score_codex":0.0006834507,"about_ca_topic_score_gemma":0.0006996778,"teacher_disagreement_score":0.52643037,"about_ca_system_score_codex":0.0003710366,"about_ca_system_score_gemma":0.00044848016,"threshold_uncertainty_score":0.999809},"labels":[],"label_agreement":null},{"id":"W3093581638","doi":"10.1177/0272989x20963038","title":"Extent and Predictors of Decision Regret among Informal Caregivers Making Decisions for a Loved One: A Systematic Review","year":2020,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Canadian Apheresis Group","funders":"","keywords":"Regret; Decision quality; Decision aids; Psychology; Odds ratio; Quality of life (healthcare); MEDLINE; Odds; Applied psychology; Medicine; Nursing; Patient satisfaction; Computer science; Alternative medicine; Logistic regression","score_opus":0.2425309727595998,"score_gpt":0.4814729607629243,"score_spread":0.23894198800332447,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3093581638","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000018046314,0.96120095,0.019715752,0.0002554665,0.0014990639,0.016630255,0.00027220015,0.0001518471,0.00025644444],"genre_scores_gemma":[0.00043582093,0.9804548,0.014267974,0.0010876696,0.00023683565,0.0031652113,0.00014662322,0.00016553629,0.000039523486],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9832101,0.0027698048,0.008295458,0.0011476022,0.0035386852,0.0010383427],"domain_scores_gemma":[0.9414984,0.048083406,0.005711013,0.0026344862,0.0011178604,0.000954857],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.004134042,0.0009357769,0.0070618214,0.00078596105,0.0010198277,0.000044102526,0.0025317525,0.0016280866,0.00035850023],"category_scores_gemma":[0.116294466,0.0006972465,0.0011329533,0.0014093412,0.0003272899,0.00032358462,0.0027385761,0.0028152794,0.00009787817],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006687446,0.000030188068,0.000057935587,0.5385944,0.00014261209,0.000019074312,0.0011430826,3.7578155e-7,6.3044343e-9,0.00017899764,0.0025243978,0.457242],"study_design_scores_gemma":[0.00072221545,0.00016254393,0.000031484626,0.762411,0.0017221046,0.00002986306,0.00092306786,0.0005429794,8.067624e-9,0.00026452017,0.23278807,0.0004021589],"about_ca_topic_score_codex":0.000023433982,"about_ca_topic_score_gemma":0.00009350668,"teacher_disagreement_score":0.45683986,"about_ca_system_score_codex":0.00070833194,"about_ca_system_score_gemma":0.0027459732,"threshold_uncertainty_score":0.999668},"labels":[],"label_agreement":null},{"id":"W3094530411","doi":"10.1177/0272989x20954391","title":"Implementing Interventions with Varying Marginal Cost-Effectiveness: An Application in Precision Medicine","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Wellcome Trust","keywords":"Precision medicine; Marginal cost; Computer science; Cost–benefit analysis; Quality-adjusted life year; Cost effectiveness; Medicine; Risk analysis (engineering); Economics","score_opus":0.44742690829855647,"score_gpt":0.5261188130482587,"score_spread":0.07869190474970222,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094530411","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12947112,0.00063297217,0.8527336,0.014215822,0.0001896992,0.0012389311,0.000023433657,0.00006949105,0.001424912],"genre_scores_gemma":[0.98204434,0.000040285944,0.011464486,0.0056826016,0.0003963581,0.0002838739,0.00004649664,0.00003542146,0.0000061634587],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99414366,0.0005481473,0.0035899458,0.0008150105,0.00041657334,0.00048662932],"domain_scores_gemma":[0.9949629,0.0029516418,0.0012476821,0.00038823884,0.000073189694,0.0003763509],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02726795,0.00019945561,0.0008814202,0.00039582208,0.00022456172,0.00006338421,0.00046912584,0.00015084159,0.0016825256],"category_scores_gemma":[0.011292433,0.00019929046,0.00009153986,0.0005935884,0.000080718615,0.000562073,0.00017297971,0.00037051315,0.00037508985],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008271274,0.00054417347,0.4075184,0.001852437,0.000108507506,0.00005558574,0.0057049445,0.0028208129,0.000050044502,0.089272596,0.0036335525,0.48761183],"study_design_scores_gemma":[0.009384299,0.0014485389,0.2676524,0.010520757,0.000033218817,0.00005207526,0.0023492803,0.59561545,0.000011957096,0.0591552,0.052718736,0.0010580696],"about_ca_topic_score_codex":0.00025640277,"about_ca_topic_score_gemma":0.00034384095,"teacher_disagreement_score":0.8525732,"about_ca_system_score_codex":0.00036116256,"about_ca_system_score_gemma":0.00009486038,"threshold_uncertainty_score":0.9992301},"labels":[],"label_agreement":null},{"id":"W3105015934","doi":"10.1177/0272989x20961091","title":"Exploring the Cost Effectiveness of Shared Decision Making for Choosing between Disease-Modifying Drugs for Relapsing-Remitting Multiple Sclerosis in the Netherlands: A State Transition Model","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Genentech","keywords":"Relapsing remitting; Multiple sclerosis; Disease; Medicine; State (computer science); Intensive care medicine; Psychology; Computer science; Psychiatry; Internal medicine","score_opus":0.3143077038812684,"score_gpt":0.4024328861154659,"score_spread":0.08812518223419752,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3105015934","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.48133782,0.00024149298,0.5144632,0.0012591372,0.00008082393,0.002486717,0.00008482088,0.000037994818,0.000008042264],"genre_scores_gemma":[0.97501194,0.00025139566,0.022517033,0.00056313496,0.00033659165,0.001203583,0.00003791387,0.000077869416,5.0396255e-7],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.994748,0.00049031124,0.0011097534,0.0007485603,0.0021659832,0.0007374148],"domain_scores_gemma":[0.9657866,0.032868423,0.00026995927,0.00045712263,0.00032660505,0.0002912962],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007239116,0.00034732482,0.00083639,0.0002386787,0.00065285625,0.00012802306,0.000604908,0.00011799775,0.000008853864],"category_scores_gemma":[0.025828987,0.00021559089,0.00042128874,0.00069862494,0.00018111087,0.00034307604,0.00024498056,0.00067656423,0.0000016170712],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010538954,0.00014245888,0.009956832,0.0014528654,0.00011771795,0.000011821706,0.016534677,0.026621887,0.002555212,0.000021020882,0.00017173192,0.9318748],"study_design_scores_gemma":[0.0074535958,0.00021789283,0.121364675,0.024267163,0.00014388349,0.0000017605838,0.0013788112,0.84280413,0.00030682725,0.0017447989,0.00007625853,0.000240205],"about_ca_topic_score_codex":0.000012698737,"about_ca_topic_score_gemma":0.000012718481,"teacher_disagreement_score":0.9316346,"about_ca_system_score_codex":0.00021782272,"about_ca_system_score_gemma":0.00017878707,"threshold_uncertainty_score":0.9823769},"labels":[],"label_agreement":null},{"id":"W3112353248","doi":"10.1177/0272989x20978208","title":"What Works in Implementing Patient Decision Aids in Routine Clinical Settings? A Rapid Realist Review and Update from the International Patient Decision Aid Standards Collaboration","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":231,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; University of British Columbia; Ottawa Hospital; Western University","funders":"Cardiff University; Healthwise; Massachusetts General Hospital","keywords":"Decision aids; Medicine; Decision support system; Medical emergency; Management science; Computer science; Alternative medicine; Engineering; Data mining; Pathology","score_opus":0.13064345016930967,"score_gpt":0.49129230006873426,"score_spread":0.3606488498994246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3112353248","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94029874,0.021566898,0.011054253,0.019014105,0.004418757,0.0031736556,0.0002640138,0.00005036487,0.00015918903],"genre_scores_gemma":[0.75193644,0.07459147,0.019142693,0.15274197,0.0005740729,0.00027507634,0.00068430183,0.000051555522,0.0000023945702],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98904365,0.0017263079,0.0042993925,0.0009301864,0.0032332307,0.0007672363],"domain_scores_gemma":[0.98878145,0.008749073,0.0010245697,0.00047308175,0.0003917522,0.00058004475],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011359699,0.0003304641,0.0008009525,0.00015772133,0.00057011744,0.00016673826,0.0005763362,0.0003545252,0.003108246],"category_scores_gemma":[0.011642716,0.00022999548,0.00010908445,0.0009209227,0.00008074436,0.00064263766,0.0013638791,0.0014995729,0.00005818322],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013398721,0.00011766659,0.1046289,0.0002019508,0.000011996835,0.000044468637,0.0028246883,0.000010183205,0.0000010564943,0.00006928049,0.02406314,0.8666868],"study_design_scores_gemma":[0.01150931,0.0014095565,0.015185669,0.20687184,0.000077431505,0.000004076605,0.021624226,0.014529508,0.000007491433,0.0046746666,0.7234825,0.00062376657],"about_ca_topic_score_codex":0.00014725472,"about_ca_topic_score_gemma":0.002387471,"teacher_disagreement_score":0.86606306,"about_ca_system_score_codex":0.0006289647,"about_ca_system_score_gemma":0.00063320424,"threshold_uncertainty_score":0.99780303},"labels":[],"label_agreement":null},{"id":"W3112360353","doi":"10.1177/0272989x20979693","title":"Feasibility of Rapidly Developing and Widely Disseminating Patient Decision Aids to Respond to Urgent Decisional Needs due to the COVID-19 Pandemic","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Toronto Metropolitan University; University of Toronto; Centre intégré de santé et de services sociaux de Chaudière-Appalaches; Royal Ottawa Mental Health Centre; Ottawa Hospital; Champlain Regional College; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Dissemination; Stakeholder; Decision aids; Pandemic; Information Dissemination; Needs assessment; Public relations; Residence; Business; Psychology; Coronavirus disease 2019 (COVID-19); Medicine; Medical education; Nursing; Political science; Computer science; Sociology; Alternative medicine; World Wide Web","score_opus":0.3316780828239632,"score_gpt":0.503798233724647,"score_spread":0.17212015090068383,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3112360353","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.64767265,0.00063772994,0.30428115,0.0441207,0.0006872019,0.0024095566,0.000047942252,0.00008748989,0.000055548066],"genre_scores_gemma":[0.80695224,0.00004230783,0.10514597,0.087346144,0.00014810824,0.00030249555,0.000009570008,0.000042668875,0.000010516792],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99017715,0.0021175959,0.00285691,0.00092935585,0.0030953004,0.00082367734],"domain_scores_gemma":[0.95984644,0.033915922,0.0005763589,0.001399926,0.00083927007,0.0034221078],"candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.0049196836,0.0003681929,0.00080965675,0.00056794414,0.0015097848,0.000039430593,0.0015769004,0.00037881063,0.0007918262],"category_scores_gemma":[0.21809237,0.00026547592,0.0001286166,0.0025917075,0.000106826905,0.0001380696,0.003946952,0.0011731301,0.00018073707],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025566565,0.000052376432,0.0948915,0.0002015453,0.000015400814,0.000024367753,0.06460533,0.0006341727,0.00013094538,0.00073682744,0.019435441,0.8167154],"study_design_scores_gemma":[0.005292583,0.00268738,0.19493355,0.01826946,0.00007567218,0.000115374096,0.11081869,0.012285319,0.00007757059,0.015768388,0.63788354,0.0017924779],"about_ca_topic_score_codex":0.00012431643,"about_ca_topic_score_gemma":0.00042293413,"teacher_disagreement_score":0.8149229,"about_ca_system_score_codex":0.00092086993,"about_ca_system_score_gemma":0.0021503898,"threshold_uncertainty_score":0.99997973},"labels":[],"label_agreement":null},{"id":"W3115341605","doi":"10.1177/0272989x20977885","title":"Observer Ratings of Shared Decision Making Do Not Match Patient Reports: An Observational Study in 5 Family Medicine Practices","year":2020,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université Laval","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Observational study; Observer (physics); Clinical Practice; Health professionals; Family medicine; Health care; Physical therapy; Internal medicine","score_opus":0.6461170982668331,"score_gpt":0.5565483977875814,"score_spread":0.0895687004792517,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3115341605","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98327124,0.00092854083,0.0034861881,0.007815526,0.0012763456,0.0025825466,0.000021001411,0.00013584993,0.00048276538],"genre_scores_gemma":[0.95145446,0.00005234158,0.035422757,0.012348586,0.0002935357,0.000310555,0.000050663573,0.0000615211,0.000005553274],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98616016,0.0025979243,0.005006554,0.00107192,0.004515963,0.0006474811],"domain_scores_gemma":[0.97684836,0.015289841,0.0044665677,0.0015980054,0.0012551977,0.0005420456],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004140776,0.00037127812,0.0009945158,0.00034098953,0.00066123373,0.000035064662,0.0011883007,0.0004680388,0.002410675],"category_scores_gemma":[0.08748339,0.00030575035,0.00009574241,0.0014455811,0.00013991125,0.00083201216,0.0013901627,0.001924742,0.000048204973],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014035522,0.0005493352,0.7987385,0.00033373473,0.000026753673,0.0002900084,0.05673428,0.00018699924,0.00016162494,0.00026842873,0.004424095,0.1368827],"study_design_scores_gemma":[0.0030091226,0.0014611479,0.8814781,0.01186923,0.000041385247,0.000012740467,0.0739059,0.01923375,0.0000033031895,0.0036367201,0.004939754,0.00040888728],"about_ca_topic_score_codex":0.0005778706,"about_ca_topic_score_gemma":0.0006587001,"teacher_disagreement_score":0.13647382,"about_ca_system_score_codex":0.00035163705,"about_ca_system_score_gemma":0.0010543679,"threshold_uncertainty_score":0.99993944},"labels":[],"label_agreement":null},{"id":"W3122845035","doi":"10.1177/0272989x20985836","title":"Efficacy of Advance Care Planning Videos for Patients: A Randomized Controlled Trial in Cancer, Heart, and Kidney Failure Outpatient Settings","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Palliative Care and End-of-Life Issues","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Covenant Health; University of Alberta; University of Calgary","funders":"","keywords":"Advance care planning; Conversation; Psychological intervention; Medicine; Randomized controlled trial; Health care; Intervention (counseling); Quality of life (healthcare); Odds; Family medicine; Nursing; Psychology; Palliative care; Logistic regression; Internal medicine","score_opus":0.03892593167789706,"score_gpt":0.4285294931363553,"score_spread":0.38960356145845826,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3122845035","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9780129,0.010627395,0.007030516,0.0007936583,0.00048701215,0.00291767,0.00003444554,0.000014725271,0.00008163351],"genre_scores_gemma":[0.9764985,0.00048492654,0.021176571,0.0012258686,0.00018434162,0.0003426905,0.000046204972,0.000020294012,0.000020603642],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.99734527,0.00020980326,0.0010835669,0.00032026705,0.0008212785,0.00021979562],"domain_scores_gemma":[0.9915859,0.0072253407,0.00033420417,0.00016577078,0.00048916636,0.00019963295],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0008270758,0.00016509333,0.0019030808,0.00014545313,0.00003775001,0.000008513769,0.000073939314,0.00016167406,0.000091909584],"category_scores_gemma":[0.07110599,0.00011761809,0.00030387766,0.0001860689,0.00010908299,0.00004718996,0.00007192323,0.00021270468,3.1348364e-7],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.94361717,0.0003116672,0.011793799,0.0004251875,0.00012329599,0.000031606098,0.0035699073,0.00002941833,0.000106527805,0.000032278593,0.0011853704,0.038773753],"study_design_scores_gemma":[0.9696182,0.00054405193,0.0004393479,0.016942298,0.00017142095,0.0000018652231,0.001113172,0.0013858381,0.00034065175,0.00009801081,0.0092332605,0.00011190923],"about_ca_topic_score_codex":0.000017332122,"about_ca_topic_score_gemma":0.000009861447,"teacher_disagreement_score":0.07027891,"about_ca_system_score_codex":0.00006473697,"about_ca_system_score_gemma":0.0003478476,"threshold_uncertainty_score":0.93671846},"labels":[],"label_agreement":null},{"id":"W3135117411","doi":"10.1177/0272989x20984134","title":"User Involvement in the Design and Development of Patient Decision Aids and Other Personal Health Tools: A Systematic Review","year":2021,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":141,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Regent Park Community Health Centre; Women's College Hospital; McMaster University; University of Toronto; Université Laval","funders":"Canadian Institutes of Health Research; Patient-Centered Outcomes Research Institute","keywords":"Usability; Decision aids; Context (archaeology); User-centered design; Computer science; Knowledge management; Process management; Human–computer interaction; Medicine; Engineering","score_opus":0.42237291448278635,"score_gpt":0.5130292441222223,"score_spread":0.09065632963943593,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3135117411","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00014820762,0.9764961,0.012767136,0.00006878554,0.00041446582,0.010042159,0.000023744606,0.000011705396,0.000027725147],"genre_scores_gemma":[0.000028596178,0.96070594,0.0244314,0.012905166,0.00004556682,0.0017631832,0.00006636339,0.000042383275,0.000011386263],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.98447,0.006324013,0.005162444,0.0006918456,0.0026515434,0.0007001433],"domain_scores_gemma":[0.9849542,0.012219532,0.0017587803,0.00053540745,0.000111256806,0.00042081965],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.017131155,0.0005098988,0.0034793688,0.00029189722,0.0005743076,0.000026114041,0.0004489784,0.0004034846,0.00047956145],"category_scores_gemma":[0.0020904322,0.0002769368,0.00017561711,0.0005343014,0.00007307932,0.00006224546,0.0005767076,0.0011020615,0.00003659677],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000010086408,0.000080011596,0.00003834278,0.4563155,0.00002335936,0.000010813708,0.0030898598,4.100847e-8,1.1758664e-9,0.00007221223,0.00056011527,0.5397997],"study_design_scores_gemma":[0.00030108358,0.00018610613,0.000004273384,0.66387266,0.00009809271,0.000013383709,0.0018252126,0.00002133339,1.06319025e-8,0.000036513135,0.33348837,0.00015295765],"about_ca_topic_score_codex":0.000028272487,"about_ca_topic_score_gemma":0.00011893656,"teacher_disagreement_score":0.53964674,"about_ca_system_score_codex":0.00056810764,"about_ca_system_score_gemma":0.0028173109,"threshold_uncertainty_score":0.9999683},"labels":[],"label_agreement":null},{"id":"W3139390944","doi":"10.1177/0272989x21997330","title":"Guidance and/or Decision Coaching with Patient Decision Aids: Scoping Reviews to Inform the International Patient Decision Aid Standards (IPDAS)","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":56,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Cochrane; Queen's University; Université de Sherbrooke; Western University; Ottawa Hospital","funders":"Novo Nordisk Fonden","keywords":"Decision aids; Coaching; Decision analysis; Decision support system; Medicine; Management science; MEDLINE; Psychology; Computer science; Alternative medicine; Political science; Engineering; Psychotherapist","score_opus":0.14308785271339008,"score_gpt":0.4771929174194201,"score_spread":0.33410506470603,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3139390944","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.43946883,0.017413266,0.5222137,0.0063545867,0.0064335666,0.005346045,0.00013640773,0.0002373878,0.0023961973],"genre_scores_gemma":[0.5336718,0.0049074637,0.4329462,0.026861042,0.000400312,0.00095534866,0.000052062103,0.00012823638,0.00007753431],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98603827,0.0013800908,0.0038090858,0.0011847443,0.0065805255,0.0010072876],"domain_scores_gemma":[0.9729511,0.020226,0.00123232,0.0024523437,0.0022275979,0.00091063534],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005074101,0.00057070947,0.0010156939,0.0004625166,0.0030006808,0.00022291497,0.001753971,0.00048532157,0.0025357974],"category_scores_gemma":[0.07181463,0.0003376549,0.00018865879,0.0015745778,0.00016913276,0.0005947675,0.0041544703,0.001993077,0.00025757132],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0024588283,0.000095784504,0.0086126495,0.00017266325,0.000022514894,0.000081992366,0.0067508626,0.00019979938,0.00000926075,0.00044923314,0.013108339,0.9680381],"study_design_scores_gemma":[0.0036897266,0.0010899632,0.0032240439,0.124554306,0.00004138268,0.00013619955,0.009802469,0.005633001,0.000044376193,0.0044214744,0.8466521,0.0007109593],"about_ca_topic_score_codex":0.00006417197,"about_ca_topic_score_gemma":0.0019701621,"teacher_disagreement_score":0.9673271,"about_ca_system_score_codex":0.0013089522,"about_ca_system_score_gemma":0.002352594,"threshold_uncertainty_score":0.99990755},"labels":[],"label_agreement":null},{"id":"W3140466070","doi":"10.1177/0272989x21998980","title":"Layperson Views about the Design and Evaluation of Decision Aids: A Public Deliberation","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of Guelph","funders":"Patient-Centered Outcomes Research Institute","keywords":"Decision aids; Deliberation; Medicine; Facilitator; Layperson; Decision analysis; Trustworthiness; Family medicine; Psychology; Alternative medicine; Social psychology; Political science","score_opus":0.526994354579252,"score_gpt":0.532461009416268,"score_spread":0.005466654837015916,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3140466070","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.43905237,0.031091262,0.51335883,0.009595931,0.0018674191,0.0028734298,0.000009207471,0.000098247794,0.002053276],"genre_scores_gemma":[0.9377005,0.00078473135,0.058092166,0.0028939222,0.00013393034,0.00032183615,0.000019211851,0.000025224248,0.000028478076],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98818827,0.006683348,0.0014219831,0.00042011487,0.0029071236,0.00037913743],"domain_scores_gemma":[0.9816429,0.014267323,0.0005601233,0.001105787,0.0021925126,0.0002313389],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.009227037,0.00016489901,0.00035124423,0.00016727812,0.0011137032,0.000040424602,0.00049695716,0.00035993298,0.0022641062],"category_scores_gemma":[0.036551103,0.000112517424,0.00007523107,0.00070514734,0.0001257059,0.00024872704,0.0005219154,0.0008218319,0.00007793211],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007289756,0.00004874387,0.002557515,0.00006297087,0.0000122114325,0.0000032900807,0.0040544462,0.00009279369,0.00007183798,0.0016082808,0.00307783,0.98833716],"study_design_scores_gemma":[0.008276095,0.00045588208,0.078997724,0.014362501,0.000272268,0.000084653744,0.026053391,0.70197284,0.00018155875,0.085645415,0.082983136,0.0007145521],"about_ca_topic_score_codex":0.000022062284,"about_ca_topic_score_gemma":0.0002233035,"teacher_disagreement_score":0.9876226,"about_ca_system_score_codex":0.00024624378,"about_ca_system_score_gemma":0.0018053097,"threshold_uncertainty_score":0.998648},"labels":[],"label_agreement":null},{"id":"W3142319574","doi":"10.1177/0272989x211004532","title":"Using Published Health Utilities in Cost-Utility Analyses: Discrepancies and Issues in Cardiovascular Disease","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Medicine; Credibility; Quality-adjusted life year; Cost–utility analysis; Logistic regression; Health economics; Value (mathematics); Public health; Actuarial science; Risk analysis (engineering); Cost effectiveness; Statistics; Economics; Pathology","score_opus":0.5519049230256677,"score_gpt":0.5375153133700927,"score_spread":0.014389609655575009,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3142319574","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7750574,0.14119837,0.047057364,0.03465345,0.00065360166,0.0006951982,0.00012525357,0.000044119362,0.00051529263],"genre_scores_gemma":[0.9885379,0.00071532646,0.006428102,0.004055439,0.00014759437,0.000038434115,0.000018717645,0.00002006319,0.000038399292],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9933581,0.0010101079,0.0037537855,0.00090794207,0.00045338486,0.0005166911],"domain_scores_gemma":[0.99630296,0.0018446,0.0005581938,0.0007857605,0.00006781836,0.00044063575],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03118014,0.00019103405,0.0015469423,0.00052048493,0.00015286228,0.00020483285,0.00023936489,0.00014848358,0.0009102288],"category_scores_gemma":[0.04429608,0.00021973702,0.00020276045,0.00061574875,0.00013734482,0.000619448,0.0002348578,0.0003270808,0.0000356322],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001209232,0.00053698855,0.9066089,0.0014182287,0.00022971044,0.00033883558,0.005449297,0.0042431904,5.8987666e-7,0.036130406,0.0041656205,0.040757306],"study_design_scores_gemma":[0.0015630057,0.000025456027,0.58348745,0.0017939542,0.000011926674,0.000020687474,0.0049959114,0.30331248,0.000001105159,0.062535174,0.04180268,0.00045018498],"about_ca_topic_score_codex":0.0034896787,"about_ca_topic_score_gemma":0.0016669137,"teacher_disagreement_score":0.32312146,"about_ca_system_score_codex":0.00058224966,"about_ca_system_score_gemma":0.00068705063,"threshold_uncertainty_score":0.99760395},"labels":[],"label_agreement":null},{"id":"W3157855096","doi":"10.1177/0272989x211007842","title":"Representing Tuberculosis Transmission with Complex Contagion: An Agent-Based Simulation Modeling Approach","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Center for Advancing Translational Sciences; U.S. National Library of Medicine; National Institute of Allergy and Infectious Diseases; National Institutes of Health","keywords":"Transmission (telecommunications); Tuberculosis; Incidence (geometry); Epidemic model; Econometrics; Population; Computer science; Medicine; Outbreak; Statistics; Demography; Virology; Mathematics; Environmental health; Telecommunications; Pathology","score_opus":0.1142381757232625,"score_gpt":0.41478631647593756,"score_spread":0.30054814075267505,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3157855096","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29343796,0.00023968604,0.7022835,0.002324299,0.00004397812,0.00022365616,0.0000016682949,0.00009136199,0.0013538885],"genre_scores_gemma":[0.90049195,0.00004543592,0.095778584,0.0031793255,0.00023958142,0.000018794517,0.00018123521,0.00003801122,0.000027099693],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99554497,0.0005119258,0.00070785015,0.0008072017,0.0018496179,0.00057840534],"domain_scores_gemma":[0.99694157,0.0012135127,0.00008367006,0.00060840475,0.00039182027,0.00076104776],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017386264,0.00022976742,0.0006179753,0.00019543698,0.00026782998,0.00005520578,0.00018544824,0.00029824415,0.0014664099],"category_scores_gemma":[0.0034036606,0.00016433612,0.00018426789,0.0005168789,0.00009847629,0.00014952403,0.000085402164,0.000631993,0.000019886875],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018282579,0.0007045231,0.014170822,0.00019076193,0.00011249225,0.0011273915,0.00016540935,0.43653047,0.0022044736,0.0001313529,0.0005269147,0.54230714],"study_design_scores_gemma":[0.0023172286,0.0002387913,0.0051011597,0.0008184288,0.00007094308,0.00021367818,0.00018140949,0.98865163,0.00010153994,0.00021817727,0.0019059029,0.00018108115],"about_ca_topic_score_codex":0.000039890994,"about_ca_topic_score_gemma":0.000017025484,"teacher_disagreement_score":0.607054,"about_ca_system_score_codex":0.00009695248,"about_ca_system_score_gemma":0.0003157219,"threshold_uncertainty_score":0.9994464},"labels":[],"label_agreement":null},{"id":"W3160824471","doi":"10.1177/0272989x211011606","title":"One Size Does Not Fit All: Financial Incentives Needed to Change Physical Exercise Levels for Different Groups","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Overweight; Incentive; Percentile; Sample (material); Physical activity; Population; Payment; Quarter (Canadian coin); Intervention (counseling); Psychology; Medicine; Demography; Obesity; Environmental health; Physical therapy; Business; Economics; Statistics; Finance","score_opus":0.19547689848189587,"score_gpt":0.30545594351259897,"score_spread":0.1099790450307031,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3160824471","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94716406,0.00022952582,0.04759455,0.003337327,0.00088864274,0.0004141653,0.00017441812,0.000024700215,0.00017263196],"genre_scores_gemma":[0.9901776,0.00015604842,0.0037878042,0.004789669,0.0006190871,0.0001896154,0.000018474708,0.000028637722,0.0002330697],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99825,0.000018759612,0.0006037478,0.0006021849,0.00017086262,0.000354438],"domain_scores_gemma":[0.9987029,0.0006208994,0.00017193121,0.00027728843,0.00001842943,0.00020858228],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005086673,0.000186039,0.0005700198,0.00008381366,0.0001245583,0.0000712086,0.00025032402,0.00015210951,0.001804967],"category_scores_gemma":[0.0016346787,0.00017491086,0.00020357322,0.00010880772,0.00004712057,0.00018570782,0.00028738557,0.00015218387,0.00031127775],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00042383358,0.001511716,0.016455758,0.00012123292,0.000096190044,0.000027325761,0.004237062,0.00006743188,0.00079287123,0.044060387,0.002914513,0.92929167],"study_design_scores_gemma":[0.0028456394,0.0002145232,0.81767064,0.00062455685,0.000041030948,0.0000025733827,0.00032969657,0.014575554,0.0032265356,0.144191,0.015500903,0.000777343],"about_ca_topic_score_codex":0.0000120504965,"about_ca_topic_score_gemma":0.00003427682,"teacher_disagreement_score":0.92851436,"about_ca_system_score_codex":0.00017547145,"about_ca_system_score_gemma":0.00001711346,"threshold_uncertainty_score":0.99910754},"labels":[],"label_agreement":null},{"id":"W3160848562","doi":"10.1177/0272989x211073098","title":"Calculating Expected Value of Sample Information Adjusting for Imperfect Implementation","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Imperfect; Monte Carlo method; Matching (statistics); Sampling (signal processing); Sample (material); Moment (physics); Value of information; Expected value","score_opus":0.5071110705585236,"score_gpt":0.5631016622128868,"score_spread":0.055990591654363264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3160848562","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.30908278,0.00007735764,0.68926036,0.0001500468,0.00039127158,0.0006540408,0.000061128165,0.0000079752945,0.00031503473],"genre_scores_gemma":[0.9253876,0.0000014609969,0.07376205,0.00053012255,0.00007703094,0.00014931634,0.000059881513,0.000009193767,0.00002331074],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9732031,0.0037704068,0.010146006,0.0005292824,0.012039252,0.00031193075],"domain_scores_gemma":[0.9599207,0.031523,0.005927958,0.0014386448,0.0010303792,0.00015930239],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10915827,0.00018665673,0.0015590428,0.0005963155,0.000547328,0.00027144194,0.0013086797,0.000062278996,0.03353796],"category_scores_gemma":[0.18428288,0.000111148394,0.0010238228,0.0016834092,0.000028413415,0.0003993677,0.00050757104,0.00019517547,0.00006476399],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00002787702,0.00002099649,0.011730238,0.00003876931,0.000048175123,0.0000011953249,0.0013010049,0.00326604,0.000060597627,0.004601929,0.013636845,0.96526635],"study_design_scores_gemma":[0.0012358087,0.00024622303,0.012823553,0.00014364728,0.0001704228,0.000030551408,0.01437941,0.85780305,0.000075536416,0.028316509,0.084477454,0.00029783463],"about_ca_topic_score_codex":0.000038455375,"about_ca_topic_score_gemma":0.000017477942,"teacher_disagreement_score":0.9649685,"about_ca_system_score_codex":0.00009539274,"about_ca_system_score_gemma":0.00015303605,"threshold_uncertainty_score":0.96734554},"labels":[],"label_agreement":null},{"id":"W3160970860","doi":"10.1177/0272989x211010738","title":"The Multifocal Approach to Sharing in Shared Decision Making: A Critical Appraisal of the MAPPIN’SDM","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Hospital for Sick Children; Ottawa Hospital; Western University; SickKids Foundation; Dalhousie University; University of Toronto; University of Ottawa","funders":"","keywords":"Critical appraisal; Decision aids; Decision theory; Decision analysis; MEDLINE; Reliability (semiconductor)","score_opus":0.22239076927180668,"score_gpt":0.5031467805379212,"score_spread":0.2807560112661145,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3160970860","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.44296426,0.009418599,0.43017283,0.06451191,0.020584542,0.0065934444,0.00013414674,0.00045272688,0.025167538],"genre_scores_gemma":[0.9488973,0.000040436596,0.047543287,0.002799728,0.00021742206,0.0003909725,0.000007221482,0.000049508453,0.000054144293],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9912399,0.0019723405,0.0022961616,0.00082401105,0.0026753973,0.0009921766],"domain_scores_gemma":[0.94871706,0.04674335,0.00040101536,0.0028448254,0.0008461131,0.00044765763],"candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0030262081,0.000279692,0.0005648833,0.00020577764,0.0015759475,0.00006430391,0.0026722408,0.00056431786,0.00043496696],"category_scores_gemma":[0.13186575,0.00017715429,0.00020340996,0.0015749191,0.000288779,0.00014506365,0.0047397884,0.00245599,0.00012423932],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014767863,0.0010254427,0.18112029,0.000962383,0.000050771458,0.00009673876,0.031980924,0.00060335983,0.00014201306,0.05341771,0.022512939,0.7066106],"study_design_scores_gemma":[0.005024152,0.00017003214,0.49560708,0.03779159,0.0000521965,0.00009530788,0.04516717,0.26399928,0.000038749713,0.04651272,0.10452194,0.0010197912],"about_ca_topic_score_codex":0.000059762915,"about_ca_topic_score_gemma":0.0005804412,"teacher_disagreement_score":0.70559084,"about_ca_system_score_codex":0.0003736092,"about_ca_system_score_gemma":0.0010093933,"threshold_uncertainty_score":0.9998454},"labels":[],"label_agreement":null},{"id":"W3163248277","doi":"10.1177/0272989x211011120","title":"Are We Improving? Update and Critical Appraisal of the Reporting of Decision Process and Quality Measures in Trials Evaluating Patient Decision Aids","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"","keywords":"Decision aids; Medicine; Interpretability; Concordance; Reliability (semiconductor); Systematic review; MEDLINE; Critical appraisal; Randomized controlled trial; Alternative medicine; Computer science; Machine learning","score_opus":0.49892779776983465,"score_gpt":0.607310710316577,"score_spread":0.10838291254674237,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3163248277","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97050107,0.008443201,0.016796824,0.001741793,0.001551026,0.00086201215,0.00002404352,0.000027172804,0.000052854917],"genre_scores_gemma":[0.96268517,0.0002823662,0.036343273,0.00048788136,0.00007016806,0.000098116325,0.0000029794382,0.00002815787,0.0000018824603],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.979451,0.0068496745,0.0090889735,0.0007380435,0.0033612037,0.00051112176],"domain_scores_gemma":[0.7980827,0.17751351,0.016383171,0.0028750524,0.004469923,0.00067563675],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02308885,0.00025169554,0.0014339113,0.00023230886,0.00056519633,0.000026244907,0.00050958386,0.0005006206,0.00014141061],"category_scores_gemma":[0.7213465,0.00017151667,0.00015804161,0.0008100781,0.00033246263,0.00019530476,0.00169549,0.0013257771,0.0000014934761],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005111954,0.0001066527,0.23950201,0.0009105064,0.000009323728,0.000015217193,0.0043346547,0.000029316168,0.00025534432,0.0004782565,0.000060315826,0.7537872],"study_design_scores_gemma":[0.010186774,0.00070759654,0.6174429,0.10770646,0.00021250115,0.000112545764,0.09957878,0.035711654,0.001347578,0.124515526,0.001474242,0.001003435],"about_ca_topic_score_codex":0.00010662054,"about_ca_topic_score_gemma":0.0005680757,"teacher_disagreement_score":0.7527838,"about_ca_system_score_codex":0.00015017981,"about_ca_system_score_gemma":0.0014854175,"threshold_uncertainty_score":0.80021805},"labels":[],"label_agreement":null},{"id":"W3165044452","doi":"10.1177/0272989x211011100","title":"Do Personal Stories Make Patient Decision Aids More Effective? An Update from the International Patient Decision Aids Standards","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":67,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Narrative; Decision aids; Context (archaeology); Construct (python library); Psychology; Empirical evidence; Narrative review; Social psychology; Medicine; Computer science; Epistemology; History; Psychotherapist; Alternative medicine; Linguistics","score_opus":0.07712263461990963,"score_gpt":0.4451288075704485,"score_spread":0.36800617295053883,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3165044452","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9253615,0.0054588737,0.05155227,0.0035124584,0.011018164,0.0014282283,0.0009531385,0.00019715307,0.0005182016],"genre_scores_gemma":[0.93739206,0.0004593231,0.050965797,0.009514075,0.00076968013,0.0004344917,0.00034844692,0.00009525483,0.000020873595],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9844398,0.0029738222,0.0022790225,0.0013218727,0.008063621,0.0009218514],"domain_scores_gemma":[0.97068524,0.02265953,0.0008064908,0.002264445,0.002818786,0.00076547835],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0026722276,0.0005439431,0.0007588056,0.00027588027,0.002735298,0.00020370683,0.0019512905,0.0007365464,0.008179968],"category_scores_gemma":[0.034481216,0.0003861401,0.00030968848,0.0008647709,0.0003032532,0.00045716442,0.0032177812,0.0026087807,0.00024853722],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013700017,0.00024013192,0.04881666,0.000026865895,0.000065847555,0.00012268177,0.019462168,0.000056230965,0.00002551852,0.00078833656,0.012206411,0.91681916],"study_design_scores_gemma":[0.0052815583,0.001233617,0.06552179,0.01206282,0.000095832016,0.000049649996,0.07043763,0.011223177,0.0000946216,0.03011679,0.80292964,0.0009528943],"about_ca_topic_score_codex":0.00028754742,"about_ca_topic_score_gemma":0.0012413628,"teacher_disagreement_score":0.91586626,"about_ca_system_score_codex":0.0017822754,"about_ca_system_score_gemma":0.0020498848,"threshold_uncertainty_score":0.99985904},"labels":[],"label_agreement":null},{"id":"W3165218162","doi":"10.1177/0272989x211011101","title":"Addressing Health Literacy in Patient Decision Aids: An Update from the International Patient Decision Aid Standards","year":2021,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Health Literacy and Information Accessibility","field":"Health Professions","cited_by":91,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"National Institute on Aging","keywords":"Decision aids; Health literacy; Decision analysis; Medicine; Decision support system; Literacy; Clinical decision support system; Health care; Medical education; Psychology; Computer science; Alternative medicine; Artificial intelligence; Economic growth; Pathology; Economics","score_opus":0.14523596099373273,"score_gpt":0.5675497928479848,"score_spread":0.4223138318542521,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3165218162","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0039966847,0.87300026,0.107045256,0.0005113294,0.010571185,0.0027169227,0.0016215097,0.00012811781,0.00040870762],"genre_scores_gemma":[0.0005579195,0.90095603,0.07217558,0.02294316,0.0010349347,0.00037729298,0.001836889,0.00010859659,0.000009580473],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.97306216,0.0056268713,0.011147445,0.0015843159,0.00721189,0.0013673081],"domain_scores_gemma":[0.96905935,0.022219129,0.0040856157,0.002171056,0.0013698088,0.00109506],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.016870862,0.00088011567,0.002994239,0.00077417796,0.0017637411,0.00052594166,0.002337919,0.0013655184,0.014895558],"category_scores_gemma":[0.026877085,0.0005387896,0.00061359984,0.0015414223,0.00012984946,0.0019025445,0.002397945,0.0052361935,0.0004044086],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00025275734,0.00021505066,0.00049546483,0.001176479,0.000028429233,0.00008618401,0.0035988528,0.000009107276,5.1672777e-9,0.00012444521,0.021231253,0.97278196],"study_design_scores_gemma":[0.00096931093,0.00013399245,0.00012708636,0.2083566,0.000027690718,0.000016592123,0.0006834931,0.002973438,3.4788822e-8,0.0014518189,0.7848837,0.00037625627],"about_ca_topic_score_codex":0.00027244163,"about_ca_topic_score_gemma":0.0005531942,"teacher_disagreement_score":0.97240573,"about_ca_system_score_codex":0.0033776225,"about_ca_system_score_gemma":0.0089068515,"threshold_uncertainty_score":0.9999309},"labels":[],"label_agreement":null},{"id":"W3171715214","doi":"","title":"FORECASTING DEMAND FOR PERSONAL PROTECTIVE EQUIPMENT FOR ONTARIO'S ACUTE CARE HOSPITALS DURING THE COVID-19 PANDEMIC","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Disaster Response and Management","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Personal protective equipment; Pandemic; Coronavirus disease 2019 (COVID-19); 2019-20 coronavirus outbreak; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Medical emergency; Medicine; Acute care; Emergency medicine; Health care; Virology; Economics; Outbreak; Economic growth; Infectious disease (medical specialty); Internal medicine; Disease","score_opus":0.11620476589432249,"score_gpt":0.45169723508717574,"score_spread":0.33549246919285325,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3171715214","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5933946,0.00029335497,0.39956295,0.0018477642,0.0007490284,0.0034200186,0.00004638821,0.000059734622,0.0006261926],"genre_scores_gemma":[0.98019516,0.00001881425,0.0043797125,0.0073953546,0.00043620364,0.0037999395,0.000030712745,0.0000429873,0.0037011176],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9970548,0.0003042305,0.00060890656,0.0005236004,0.0008286594,0.0006798026],"domain_scores_gemma":[0.99492913,0.0040601124,0.0002200366,0.00025051445,0.0002441414,0.00029604393],"candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002205323,0.00020000458,0.00033236167,0.00008036551,0.0019310802,0.000047355905,0.00029643235,0.00020394968,0.0010469112],"category_scores_gemma":[0.00576748,0.00013489164,0.00021283902,0.00014481464,0.00005720306,0.00008537254,0.00059736904,0.00048968836,0.000013808323],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007149269,0.00018260024,0.040102165,0.0036133553,0.00085811026,0.00068019266,0.3590865,0.00010086107,0.00021804549,0.0014096118,0.024887046,0.56171227],"study_design_scores_gemma":[0.015883,0.0008347236,0.016183782,0.005533459,0.00052492827,0.00016720219,0.153093,0.011961458,0.000048549566,0.019570258,0.77523917,0.00096048944],"about_ca_topic_score_codex":0.00017955476,"about_ca_topic_score_gemma":0.012337495,"teacher_disagreement_score":0.7503521,"about_ca_system_score_codex":0.0011167583,"about_ca_system_score_gemma":0.0013648638,"threshold_uncertainty_score":0.99986625},"labels":[],"label_agreement":null},{"id":"W3173177335","doi":"10.1177/0272989x211020317","title":"A Systematic Review and Meta-Analysis of Patient Decision Aids for Socially Disadvantaged Populations: Update from the International Patient Decision Aid Standards (IPDAS)","year":2021,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"National Cancer Institute; National Institute on Aging","keywords":"Decision aids; Disadvantaged; Psychological intervention; Meta-analysis; Psychology; Medicine; Systematic review; MEDLINE; Alternative medicine; Nursing; Political science","score_opus":0.3128538356211123,"score_gpt":0.5467821474587048,"score_spread":0.2339283118375925,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3173177335","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00002473947,0.94239163,0.041047435,0.0011164577,0.0014978964,0.008681078,0.0051589883,0.000053033877,0.000028734321],"genre_scores_gemma":[0.00021370836,0.9337646,0.055316705,0.0047376165,0.000091540074,0.00395965,0.0017952971,0.00010592358,0.000014992461],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9774867,0.005983048,0.008906434,0.0012632748,0.0057449928,0.00061549933],"domain_scores_gemma":[0.93306637,0.055129122,0.005861898,0.0030095447,0.0025393956,0.00039365035],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0060842438,0.0008292424,0.00959153,0.00070275884,0.0013631608,0.0000973314,0.0023144544,0.00091051235,0.0038383116],"category_scores_gemma":[0.08307918,0.00047979262,0.0036254793,0.0019371797,0.00016577676,0.00022448359,0.0025585145,0.0016982346,0.000028964309],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00011393571,0.00017372335,0.0000624189,0.13554896,0.034683134,0.0000128984875,0.0016026145,0.000014162167,1.400943e-8,0.0020671785,0.008532909,0.817188],"study_design_scores_gemma":[0.0005226844,0.00009652203,0.000007337119,0.35044622,0.28055963,0.0000048759325,0.0012742926,0.00040388288,2.2818812e-8,0.0046085822,0.36163858,0.00043739687],"about_ca_topic_score_codex":0.000091130205,"about_ca_topic_score_gemma":0.00086635136,"teacher_disagreement_score":0.81675065,"about_ca_system_score_codex":0.0010828285,"about_ca_system_score_gemma":0.002415671,"threshold_uncertainty_score":0.99993694},"labels":[],"label_agreement":null},{"id":"W3175840800","doi":"10.1177/0272989x211021397","title":"Providing Balanced Information about Options in Patient Decision Aids: An Update from the International Patient Decision Aid Standards","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":74,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa; University of Toronto","funders":"","keywords":"Decision aids; Decision analysis; Decision support system; Medicine; Computer science; Management science; Medical emergency; Alternative medicine; Data mining; Economics","score_opus":0.07628924294912272,"score_gpt":0.4314560621396858,"score_spread":0.3551668191905631,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3175840800","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8826024,0.0017385357,0.10154263,0.0042372956,0.0065805013,0.0015623957,0.00056060206,0.0001827467,0.0009929222],"genre_scores_gemma":[0.901161,0.0009073253,0.085616946,0.010976156,0.00025799003,0.00041762786,0.000614539,0.00004272795,0.0000056566123],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9890093,0.0016596944,0.0030597232,0.0007016785,0.0048236265,0.00074593775],"domain_scores_gemma":[0.98510045,0.00981379,0.0009370562,0.0018401479,0.0018478804,0.00046070653],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0023409028,0.00035720062,0.00051756937,0.00039053918,0.0014692126,0.00019525421,0.0015452206,0.0005486151,0.003924484],"category_scores_gemma":[0.031183211,0.00027026655,0.0001539524,0.0009947019,0.000110392204,0.0014724218,0.0019778823,0.0019446057,0.0003291444],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000798112,0.00017073267,0.023202587,0.000025802987,0.00001943147,0.00004054525,0.010904302,0.0006809707,0.000024592613,0.0014863797,0.008196455,0.9544501],"study_design_scores_gemma":[0.0065861363,0.00040000654,0.037493814,0.015920708,0.000032738222,0.000025821733,0.034969985,0.06377505,0.000083628576,0.03543909,0.804524,0.0007489975],"about_ca_topic_score_codex":0.00030884004,"about_ca_topic_score_gemma":0.001992727,"teacher_disagreement_score":0.9537011,"about_ca_system_score_codex":0.0016022439,"about_ca_system_score_gemma":0.002119953,"threshold_uncertainty_score":0.99997497},"labels":[],"label_agreement":null},{"id":"W3176558350","doi":"10.1177/0272989x211014163","title":"Systematic Development of Patient Decision Aids: An Update from the IPDAS Collaboration","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":196,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Université Laval; Centre hospitalier universitaire de Québec","funders":"National Cancer Institute; Novo Nordisk Fonden; Canadian Institutes of Health Research; Patient-Centered Outcomes Research Institute; Cancer Prevention and Research Institute of Texas","keywords":"Decision aids; Medicine; Management science; Computer science; Alternative medicine; Pathology; Engineering","score_opus":0.11320291037273196,"score_gpt":0.44884749727225587,"score_spread":0.3356445868995239,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3176558350","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9422325,0.001135058,0.051997386,0.00020158286,0.0028706251,0.0012499669,0.000035373338,0.000035789788,0.00024172843],"genre_scores_gemma":[0.93478066,0.000091025,0.057935085,0.006481345,0.00017377349,0.00020224407,0.00028546134,0.000023838877,0.000026582218],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9927621,0.00150913,0.002420836,0.00041253844,0.0024673054,0.00042811388],"domain_scores_gemma":[0.99200386,0.0057483413,0.00069582625,0.0006851892,0.00050028757,0.00036651903],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0030462404,0.00018951163,0.0005059664,0.00007709122,0.0010056357,0.000023144657,0.00039192836,0.00024843277,0.0028179921],"category_scores_gemma":[0.0029591455,0.000116375704,0.000063563624,0.000536561,0.00004131097,0.00013639047,0.00038313356,0.00044149344,0.00039276486],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009435261,0.0010717819,0.03661944,0.010515756,0.00013600105,0.00011764499,0.05280678,0.000063473555,0.00021059145,0.0038111024,0.008295333,0.8854086],"study_design_scores_gemma":[0.009835185,0.0016511336,0.022779204,0.6049076,0.0003095626,0.000015968784,0.18242608,0.020293497,0.0032787025,0.03923117,0.11369793,0.0015739716],"about_ca_topic_score_codex":0.00003310492,"about_ca_topic_score_gemma":0.0011092046,"teacher_disagreement_score":0.8838346,"about_ca_system_score_codex":0.00028634226,"about_ca_system_score_gemma":0.0014883068,"threshold_uncertainty_score":0.99809355},"labels":[],"label_agreement":null},{"id":"W3184405065","doi":"10.1177/0272989x211026292","title":"Simulating Study Data to Support Expected Value of Sample Information Calculations: A Tutorial","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Medical Research Council; Norges Forskningsråd; University of Bristol; Department of Health and Social Care; Strong; Canadian Institutes of Health Research; National Institute for Health and Care Research; University Hospitals Bristol NHS Foundation Trust","keywords":"Bridging (networking); Missing data; Probabilistic logic; Sample (material); Key (lock); Value (mathematics); Outcome (game theory); Sample size determination","score_opus":0.4848813593112004,"score_gpt":0.5247604619037006,"score_spread":0.039879102592500215,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3184405065","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.49070337,0.000098980454,0.5021594,0.002994309,0.0019771191,0.00082836545,0.0005183945,0.000055867265,0.0006641729],"genre_scores_gemma":[0.943758,0.0000048633383,0.05055456,0.004818909,0.00053114456,0.00003264772,0.00026575613,0.000017162827,0.000016976257],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99301153,0.00034637869,0.005245249,0.0005090224,0.00059218117,0.0002956318],"domain_scores_gemma":[0.9905333,0.006295968,0.0014234153,0.0012448599,0.00023546301,0.0002670272],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.017621351,0.00014420536,0.00080251205,0.00036634106,0.00017970905,0.0001115063,0.00060237147,0.0001532311,0.0026166697],"category_scores_gemma":[0.14045297,0.00017426883,0.00007271553,0.00060010183,0.000026248592,0.0008994031,0.00057206326,0.00018790587,0.0006263959],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003200852,0.001941599,0.4999926,0.00070386496,0.00048682716,0.000057052213,0.03750965,0.03500961,0.000009010399,0.14292364,0.09696988,0.18407619],"study_design_scores_gemma":[0.005094733,0.0005256022,0.1398207,0.00092052954,0.000038563932,0.000020581452,0.010623801,0.633539,0.0000067731553,0.02466247,0.18387587,0.0008713286],"about_ca_topic_score_codex":0.0003837762,"about_ca_topic_score_gemma":0.00021490903,"teacher_disagreement_score":0.5985294,"about_ca_system_score_codex":0.00023300105,"about_ca_system_score_gemma":0.00046590166,"threshold_uncertainty_score":0.99829507},"labels":[],"label_agreement":null},{"id":"W3192788213","doi":"10.1177/0272989x211025525","title":"Testing for a Sweet Spot in Randomized Trials","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Canada Research Chairs; BrightFocus Foundation","keywords":"Sweet spot; Randomized controlled trial; Risk stratification; Clinical trial; MEDLINE; Blind spot","score_opus":0.39334255561661463,"score_gpt":0.5333925520726709,"score_spread":0.14004999645605626,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3192788213","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02951256,0.00018830768,0.966955,0.00033937843,0.00020563297,0.0008351199,0.0000032744117,0.00019139906,0.0017692986],"genre_scores_gemma":[0.27623054,0.000014708679,0.7229354,0.00037485856,0.00014305311,0.00022217153,0.0000022597762,0.000025775147,0.00005123989],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9962864,0.0007333465,0.0015627196,0.00033497557,0.00076958747,0.00031296338],"domain_scores_gemma":[0.8261805,0.17254946,0.00043976287,0.00039844235,0.0002994029,0.00013239202],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.018058911,0.00016280904,0.0015911025,0.00014842469,0.000056190205,0.000044682372,0.00023611916,0.00021161136,0.0006444852],"category_scores_gemma":[0.66738874,0.0001209333,0.00027064703,0.0004136437,0.00007608331,0.00008584556,0.00016733719,0.0002787021,0.0000074119525],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005480124,0.00014055216,0.00008185926,0.00009878188,0.000028781727,0.00031467743,0.00016075342,0.000008318636,0.0013594718,0.11487852,0.002269208,0.87517893],"study_design_scores_gemma":[0.032331876,0.000031075404,0.00001842405,0.0020029491,0.00003262839,0.00003197517,0.000047940932,0.007113601,0.0014170326,0.9563614,0.000469894,0.0001412089],"about_ca_topic_score_codex":0.0000022673391,"about_ca_topic_score_gemma":0.000022925757,"teacher_disagreement_score":0.8750377,"about_ca_system_score_codex":0.00006821318,"about_ca_system_score_gemma":0.0001964019,"threshold_uncertainty_score":0.70566607},"labels":[],"label_agreement":null},{"id":"W3194460113","doi":"10.1177/0272989x211035681","title":"The International Patient Decision Aid Standards (IPDAS) Collaboration: Evidence Update 2.0","year":2021,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":183,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Cancer Institute","keywords":"MEDLINE; Decision aids; Medicine; Computer science; Alternative medicine; Political science","score_opus":0.10136996779831609,"score_gpt":0.49393526365466905,"score_spread":0.39256529585635297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3194460113","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003055806,0.025632698,0.009067763,0.014312707,0.94661367,0.0015506472,0.00073376775,0.00019912743,0.0015840378],"genre_scores_gemma":[0.008813169,0.046784867,0.025435533,0.007969651,0.9062967,0.0019156038,0.001385612,0.00031692444,0.0010819574],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9736798,0.0031764924,0.003802851,0.0012982236,0.016987706,0.0010549579],"domain_scores_gemma":[0.89181995,0.089940414,0.0023583358,0.003555976,0.011600077,0.00072522927],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.006532542,0.0006213063,0.00093491876,0.00032460794,0.00410608,0.00041733295,0.004327247,0.0025353245,0.0063693877],"category_scores_gemma":[0.27471304,0.00045147326,0.00027871103,0.0011243828,0.0002986049,0.00048688374,0.0045255204,0.006305774,0.00062646467],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007602911,0.000038074577,0.00019699393,0.00011237665,0.000044561293,0.000063613414,0.0011481497,0.000013964351,5.495698e-7,0.00029090402,0.78034604,0.21698451],"study_design_scores_gemma":[0.0011611085,0.00012758512,0.00003103171,0.023677733,0.000035666613,0.0000030916867,0.0035309829,0.0012880674,0.0000016723718,0.0031846846,0.9665614,0.00039701542],"about_ca_topic_score_codex":0.000071473085,"about_ca_topic_score_gemma":0.002346425,"teacher_disagreement_score":0.2681805,"about_ca_system_score_codex":0.0037036298,"about_ca_system_score_gemma":0.01866313,"threshold_uncertainty_score":0.9997937},"labels":[],"label_agreement":null},{"id":"W3195294569","doi":"10.1177/0272989x221103163","title":"An Introductory Tutorial on Cohort State-Transition Models in R Using a Cost-Effectiveness Analysis Example","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":35,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; University of Toronto; University of Ottawa; Hospital for Sick Children","funders":"National Institute of Allergy and Infectious Diseases; National Cancer Institute; National Institute on Drug Abuse","keywords":"Computer science; Probabilistic logic; Cohort; Decision model; Decision analysis; Statistical model; Outcome (game theory); Transition (genetics); Constant (computer programming); Econometrics; Operations research; Statistics; Machine learning; Artificial intelligence; Mathematics; Mathematical economics","score_opus":0.3463333329277015,"score_gpt":0.4768139777248217,"score_spread":0.13048064479712018,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3195294569","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.67381316,0.000120621255,0.32352677,0.00049311767,0.0011445289,0.0006506688,0.00012376517,0.00003183447,0.00009554197],"genre_scores_gemma":[0.99440175,0.000014194768,0.001998727,0.002789309,0.00045446015,0.00022565124,0.00007897123,0.000031583062,0.000005323483],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9933408,0.0017706233,0.002884825,0.000948232,0.0006313619,0.0004241714],"domain_scores_gemma":[0.9958049,0.0024892718,0.00080734765,0.00061760173,0.00005222967,0.00022864225],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.043982636,0.00018988928,0.0011064586,0.0013372222,0.00035577887,0.00007779161,0.00039336167,0.00012788732,0.001737685],"category_scores_gemma":[0.0027507523,0.00024523324,0.00016490856,0.0010773068,0.000045835775,0.00045840745,0.000097655684,0.0004954221,0.00006785977],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029169655,0.00041195238,0.030311866,0.000047974598,0.00014240801,0.000021708382,0.0017908737,0.9507719,0.000005669227,0.011653074,0.00035844717,0.0041923975],"study_design_scores_gemma":[0.0011151448,0.00014385022,0.047151484,0.00008470584,0.000025803785,0.0000049758014,0.00041686938,0.909712,0.0000012978567,0.03930297,0.0017474844,0.00029337566],"about_ca_topic_score_codex":0.0022250635,"about_ca_topic_score_gemma":0.00031269525,"teacher_disagreement_score":0.32152805,"about_ca_system_score_codex":0.002097774,"about_ca_system_score_gemma":0.00024851764,"threshold_uncertainty_score":1},"labels":[],"label_agreement":null},{"id":"W3196833216","doi":"10.1177/0272989x211039743","title":"The Impact of 4 Risk Communication Interventions on Cancer Screening Preferences and Knowledge","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; University of Toronto","funders":"National Cancer Institute; University of Missouri","keywords":"Cancer screening; Psychological intervention; Risk communication; Risk assessment; Public health; Medical screening; MEDLINE","score_opus":0.3693068698823472,"score_gpt":0.5753126543999911,"score_spread":0.20600578451764395,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3196833216","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9072776,0.07238777,0.010417026,0.0030299728,0.00078953523,0.0007815122,0.00009907712,0.00008605012,0.0051315012],"genre_scores_gemma":[0.9890914,0.0073753763,0.0031458295,0.00010761877,0.00005474832,0.00013728847,0.000013488573,0.000013395578,0.0000608488],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99503076,0.0030846214,0.00091725634,0.00020895936,0.00049719587,0.00026121255],"domain_scores_gemma":[0.98022383,0.0175282,0.00055046956,0.0009681519,0.00057363656,0.0001556989],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0015512274,0.00011250534,0.00024121847,0.00008736405,0.0017112771,0.000020139003,0.0005866939,0.00017882518,0.0008175575],"category_scores_gemma":[0.013438923,0.000070665345,0.00014201857,0.00035259165,0.00018608666,0.000077782,0.0008646989,0.0011218052,0.00002301784],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00010419607,0.000083044695,0.1355511,0.000095363845,0.000046220597,7.1961125e-7,0.0049171634,0.00004260434,0.000003012937,0.0020719587,0.0027235784,0.85436106],"study_design_scores_gemma":[0.0013475075,0.0001818029,0.91416323,0.02582733,0.000046170248,0.0000029749142,0.0128455255,0.015253713,0.0000112089565,0.025705654,0.00442489,0.00018996278],"about_ca_topic_score_codex":0.0007839452,"about_ca_topic_score_gemma":0.0048035854,"teacher_disagreement_score":0.8541711,"about_ca_system_score_codex":0.00013140372,"about_ca_system_score_gemma":0.0006717898,"threshold_uncertainty_score":0.9995884},"labels":[],"label_agreement":null},{"id":"W3198103700","doi":"10.1177/0272989x221121747","title":"A Tutorial on Time-Dependent Cohort State-Transition Models in R Using a Cost-Effectiveness Analysis Example","year":2022,"lang":"en","type":"preprint","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University of Ottawa; Hospital for Sick Children","funders":"National Institute of Allergy and Infectious Diseases; National Institute on Drug Abuse; National Institutes of Health; National Cancer Institute; Gordon and Betty Moore Foundation; American Diabetes Association","keywords":"Dependency (UML); Computer science; Probabilistic logic; Notation; Transition (genetics); Code (set theory); State (computer science); Residence time (fluid dynamics); Algorithm; Programming language; Artificial intelligence; Mathematics","score_opus":0.3896723100063631,"score_gpt":0.48455391700742223,"score_spread":0.09488160700105913,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3198103700","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.47918493,0.00035275865,0.5141499,0.00060899544,0.0022887571,0.0021234439,0.00065257296,0.00006653518,0.0005721086],"genre_scores_gemma":[0.9921273,0.000113610884,0.0034179864,0.0024768484,0.00058626465,0.0008054739,0.00033742504,0.0000894054,0.00004569745],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.98837745,0.0022643018,0.0057326485,0.00184019,0.0011391364,0.0006462566],"domain_scores_gemma":[0.99055964,0.005607911,0.0023135808,0.0011009726,0.00009339316,0.00032449103],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.05979331,0.00045670997,0.0027473846,0.0022973502,0.00027155245,0.00020427842,0.0007525417,0.00062707195,0.0049516917],"category_scores_gemma":[0.0058279126,0.0005889714,0.0005156174,0.00087938725,0.00005859876,0.00027733546,0.00065576023,0.0014926943,0.00034420134],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002673741,0.00031690378,0.008504221,0.00024583293,0.00046744378,0.00005209159,0.0011079272,0.9841794,6.12037e-7,0.002318122,0.00035585157,0.0021842287],"study_design_scores_gemma":[0.0012151656,0.00006632378,0.0118549485,0.0008483392,0.00008944441,0.000004019785,0.00012684848,0.901789,5.9384047e-7,0.08253472,0.00091047806,0.00056007487],"about_ca_topic_score_codex":0.0057411585,"about_ca_topic_score_gemma":0.0004690363,"teacher_disagreement_score":0.5129424,"about_ca_system_score_codex":0.0046427874,"about_ca_system_score_gemma":0.0006823989,"threshold_uncertainty_score":0.99965614},"labels":[],"label_agreement":null},{"id":"W3204947748","doi":"10.1177/0272989x211037946","title":"Clarifying Values: An Updated and Expanded Systematic Review and Meta-Analysis","year":2021,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":135,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Université Laval; Centre hospitalier universitaire de Québec","funders":"National Heart, Lung, and Blood Institute; Canadian Institutes of Health Research","keywords":"Data extraction; CINAHL; Meta-analysis; MEDLINE; Cochrane Library; Confidence interval; Decision aids; Medicine; Randomized controlled trial; Psychology; Set (abstract data type); Computer science; Alternative medicine; Surgery; Pathology; Internal medicine","score_opus":0.5658833335798786,"score_gpt":0.5826820061646499,"score_spread":0.0167986725847713,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3204947748","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[8.654117e-7,0.99244636,0.0016558331,0.0005790817,0.00019881358,0.0048090224,0.000066710665,0.000120106786,0.0001231954],"genre_scores_gemma":[0.00003639659,0.98727167,0.004621352,0.0055070715,0.00005583321,0.0017788109,0.0004751253,0.00008273582,0.00017103083],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9768267,0.0147176,0.0048621693,0.0011713759,0.0018029659,0.00061916857],"domain_scores_gemma":[0.9840271,0.009200189,0.002431681,0.0029729407,0.00059589144,0.00077220233],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0058542797,0.00069195835,0.012810653,0.00064434984,0.001148606,0.0000769208,0.0011259008,0.0013696626,0.0043702153],"category_scores_gemma":[0.012800563,0.00044410213,0.0013185571,0.0019304759,0.00013027622,0.00020376353,0.0015522023,0.0030599146,0.00007358966],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000019065033,0.00003177602,0.0000063565963,0.91565627,0.022552025,0.000033787896,0.0005723833,9.658288e-8,1.9381898e-9,0.00023128267,0.0009668706,0.05994724],"study_design_scores_gemma":[0.00014031386,0.000024365752,0.0000013418021,0.373796,0.55897653,0.000041438092,0.00069162017,0.00018553932,1.285267e-9,0.00023444435,0.06552577,0.0003826691],"about_ca_topic_score_codex":0.000034911805,"about_ca_topic_score_gemma":0.00008866849,"teacher_disagreement_score":0.5418603,"about_ca_system_score_codex":0.00022460741,"about_ca_system_score_gemma":0.0009866084,"threshold_uncertainty_score":0.99992675},"labels":[],"label_agreement":null},{"id":"W3206292861","doi":"10.1177/0272989x211050909","title":"Model-Based ROC Curve: Examining the Effect of Case Mix and Model Calibration on the ROC Plot","year":2021,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Calibration; Plot (graphics); Smoothing; Receiver operating characteristic; Statistical model; Statistical hypothesis testing","score_opus":0.11780946287119767,"score_gpt":0.38415580651221704,"score_spread":0.2663463436410194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3206292861","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92291886,0.0007352144,0.07298524,0.0027540706,0.00007440615,0.0002963817,0.000005138601,0.00002154275,0.0002091662],"genre_scores_gemma":[0.9960422,0.000047710386,0.0013268965,0.0024169718,0.00005853921,0.00005767005,0.0000073353826,0.000020759811,0.000021951162],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.998039,0.00022895567,0.0003399254,0.00030318045,0.00091596594,0.0001729413],"domain_scores_gemma":[0.9936928,0.00561257,0.00008895995,0.0004189229,0.00005872107,0.00012799124],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010741679,0.00016151376,0.00032018992,0.000054566208,0.00018019497,0.00003481914,0.00008888523,0.00011873937,0.00016492291],"category_scores_gemma":[0.0032982163,0.000074070165,0.00009395922,0.00018507047,0.00008576084,0.000034969977,0.000081784005,0.0002514648,0.000003897027],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014491971,0.00096235913,0.04113131,0.0002544242,0.00034040486,0.007611266,0.0011984403,0.10404464,0.00086673396,0.0025823924,0.0071851183,0.83237374],"study_design_scores_gemma":[0.001608338,0.00047046275,0.0014516745,0.0010278985,0.00016313653,0.00027266052,0.000074303796,0.9909339,0.003445124,0.0004397462,0.000042723637,0.0000700408],"about_ca_topic_score_codex":0.0000067970495,"about_ca_topic_score_gemma":0.000025769818,"teacher_disagreement_score":0.8868893,"about_ca_system_score_codex":0.000041996398,"about_ca_system_score_gemma":0.00013367555,"threshold_uncertainty_score":0.39485112},"labels":[],"label_agreement":null},{"id":"W3209663054","doi":"10.1177/0272989x221078789","title":"Uncertainty and the Value of Information in Risk Prediction Modeling","year":2022,"lang":"en","type":"preprint","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Bayesian probability; Econometrics; Monte Carlo method; Value of information; Context (archaeology); Sample (material); Computer science; Statistics; Geography; Economics; Mathematics; Artificial intelligence; Physics","score_opus":0.19421620862095515,"score_gpt":0.42677879443686745,"score_spread":0.2325625858159123,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3209663054","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5218244,0.002963554,0.4625109,0.0074694944,0.0018983501,0.0012081759,0.00055998896,0.00003700598,0.0015281375],"genre_scores_gemma":[0.9906459,0.0015673412,0.0045891143,0.002737633,0.00017919151,0.00018811968,0.000067885514,0.000016499132,0.000008309526],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99160326,0.0009023978,0.0063095996,0.00043365904,0.0005047982,0.00024631718],"domain_scores_gemma":[0.9916624,0.0043417388,0.0032592386,0.00057449884,0.0000592541,0.00010291103],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06094759,0.00018423151,0.0010853779,0.00059278973,0.0001990376,0.00008008238,0.00051376416,0.00035739856,0.0006938909],"category_scores_gemma":[0.036735766,0.00017509256,0.00014782217,0.000208339,0.00012449347,0.0002922708,0.000852195,0.0011674635,0.000041453604],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0001631793,0.000034724446,0.015835801,0.0004224601,0.00004648507,7.946044e-7,0.0048064664,0.8832239,8.653273e-9,0.07863452,0.0011189545,0.015712744],"study_design_scores_gemma":[0.0010688627,0.000016001517,0.004073996,0.000498454,0.000007042192,0.000002074167,0.0007767594,0.77882755,1.4011595e-8,0.21244852,0.0021743155,0.00010639936],"about_ca_topic_score_codex":0.002366074,"about_ca_topic_score_gemma":0.000111652815,"teacher_disagreement_score":0.46882153,"about_ca_system_score_codex":0.0005305747,"about_ca_system_score_gemma":0.00028586586,"threshold_uncertainty_score":0.9713782},"labels":[],"label_agreement":null},{"id":"W4205116055","doi":"10.1177/0272989x211072858","title":"Prioritizing Research in an Era of Personalized Medicine: The Potential Value of Unexplained Heterogeneity","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Institute of Human Development, Child and Youth Health; Canadian Institutes of Health Research; Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Prioritization; Value (mathematics); Key (lock); Precision medicine; Personalized medicine; MEDLINE","score_opus":0.25720169022223494,"score_gpt":0.5324951527152464,"score_spread":0.27529346249301145,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4205116055","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8908302,0.00024596887,0.107390545,0.00078325387,0.00012820784,0.00036367038,0.000006128494,0.00004719266,0.00020482196],"genre_scores_gemma":[0.9666851,0.000038307095,0.033012208,0.00011271073,0.00007073782,0.00005122513,0.0000024999933,0.000021943442,0.0000052313467],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9937968,0.0013327469,0.00086182693,0.00030319527,0.0033782823,0.00032711556],"domain_scores_gemma":[0.99385834,0.0050570583,0.00023238432,0.0005454024,0.00021414764,0.00009269436],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.010393911,0.00012420706,0.000437721,0.00025726168,0.00022198402,0.000008479947,0.00088117557,0.00009826352,0.0006263176],"category_scores_gemma":[0.007322989,0.00008599912,0.00007918322,0.00081480155,0.0005675345,0.0000899455,0.00072148105,0.0012011853,6.445851e-7],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013869015,0.0009384,0.0028500743,0.00031803787,0.00005330555,0.0006194107,0.0072306823,0.00029353,0.046698343,0.49818987,0.0011696533,0.4402518],"study_design_scores_gemma":[0.0012477329,0.00078580156,0.0018182804,0.0013842604,0.000017826396,0.00008045655,0.0037817792,0.020257372,0.001284181,0.96883637,0.00034650866,0.00015946037],"about_ca_topic_score_codex":0.000051321338,"about_ca_topic_score_gemma":0.00004220555,"teacher_disagreement_score":0.47064647,"about_ca_system_score_codex":0.00013578098,"about_ca_system_score_gemma":0.00015713728,"threshold_uncertainty_score":0.87668306},"labels":[],"label_agreement":null},{"id":"W4206334503","doi":"10.1177/0272989x211065471","title":"A Systematic Literature Review of Health Utility Values in Breast Cancer","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; University of Toronto; McMaster University","funders":"","keywords":"Breast cancer; Systematic review; MEDLINE; Alternative medicine; Medical literature; Cancer; Quality-adjusted life year","score_opus":0.23321755729818008,"score_gpt":0.48419020456524026,"score_spread":0.25097264726706014,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206334503","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014814172,0.8628105,0.008737918,0.10816382,0.0016288989,0.0021691704,0.0009018172,0.000047674923,0.0007260358],"genre_scores_gemma":[0.7974956,0.07689821,0.0025816082,0.12161707,0.0002617165,0.0009082707,0.000045530447,0.000056217676,0.00013577359],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9886638,0.0019967414,0.007813327,0.0005397396,0.0006177988,0.0003686031],"domain_scores_gemma":[0.9937886,0.0023104066,0.0030819776,0.00058200455,0.000071194125,0.00016579981],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.076387934,0.00015564833,0.0019605758,0.0004079295,0.0001896484,0.000026315554,0.00053105643,0.00008743211,0.0058913683],"category_scores_gemma":[0.012273006,0.00016442836,0.0001741082,0.0008171638,0.000046286317,0.00016886836,0.00022992896,0.00044901267,0.0000789784],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000070527,0.000618631,0.06591565,0.7354363,0.000120539684,0.000036534013,0.005484124,0.00016582607,1.6457234e-7,0.031183071,0.13625894,0.02470965],"study_design_scores_gemma":[0.0013578136,0.00011270612,0.0712545,0.8480742,0.000013678202,0.00018246313,0.0012829682,0.033218347,8.6049724e-8,0.034083534,0.009889507,0.00053022127],"about_ca_topic_score_codex":0.00021006037,"about_ca_topic_score_gemma":0.000039550265,"teacher_disagreement_score":0.7859123,"about_ca_system_score_codex":0.00081265625,"about_ca_system_score_gemma":0.00044828042,"threshold_uncertainty_score":0.996047},"labels":[],"label_agreement":null},{"id":"W4223569106","doi":"10.1177/0272989x221091567","title":"Author Response to “Optimal Sample Size Calculation for Clinical Research under a Budget Constraint”","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Sample size determination; Sample (material); Clinical research; Research design; Large sample","score_opus":0.7118625322965719,"score_gpt":0.6283684702587494,"score_spread":0.0834940620378225,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4223569106","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.36824062,0.0002164688,0.49373752,0.13400722,0.0016064286,0.0013520273,0.0005474423,0.0000606938,0.00023157826],"genre_scores_gemma":[0.85812354,0.0000072934426,0.112490155,0.02708556,0.0006966532,0.000608227,0.000021066937,0.00005546551,0.0009120095],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9887818,0.0030480232,0.0053724847,0.001059703,0.00095719163,0.0007808174],"domain_scores_gemma":[0.8505782,0.14615002,0.0011794838,0.0009535313,0.00027640772,0.00086236506],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.19989173,0.00017505931,0.0009276725,0.0006155904,0.0010149358,0.00012437241,0.00069067033,0.00025890346,0.0071174083],"category_scores_gemma":[0.34276372,0.00021587967,0.00025492263,0.00064164196,0.00019095307,0.00015202994,0.0005186367,0.0007935594,0.000809747],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0038256883,0.0005369322,0.014846054,0.00011136489,0.00011026624,0.000022199225,0.0033134394,0.006296242,0.000006395268,0.13150515,0.8018071,0.037619166],"study_design_scores_gemma":[0.0023619311,0.0008740786,0.045201745,0.00015824282,0.0000061775763,0.000024223105,0.0031143022,0.09237098,5.7686367e-7,0.11794116,0.73751944,0.000427118],"about_ca_topic_score_codex":0.00008973998,"about_ca_topic_score_gemma":0.000043902906,"teacher_disagreement_score":0.48988295,"about_ca_system_score_codex":0.0010487696,"about_ca_system_score_gemma":0.00082323316,"threshold_uncertainty_score":0.99996823},"labels":[],"label_agreement":null},{"id":"W4225700113","doi":"10.1177/0272989x221089268","title":"Developing Economic Models for Assessing the Cost-Effectiveness of Multiple Diagnostic Tests: Methods and Applications","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; University of Toronto; Public Health Ontario","funders":"","keywords":"Computer science; Statistical hypothesis testing; Measure (data warehouse); Conditional independence; Econometrics; Data mining; Machine learning; Risk analysis (engineering); Statistics; Artificial intelligence; Mathematics; Medicine","score_opus":0.4644500495612203,"score_gpt":0.5643131345755833,"score_spread":0.09986308501436297,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4225700113","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029177122,0.0019218655,0.96334577,0.002775833,0.00042769496,0.0019372447,0.00014751556,0.000024578136,0.00024236724],"genre_scores_gemma":[0.859739,0.000086085565,0.13450536,0.002181236,0.00016011893,0.003269015,0.000019640036,0.000032512064,0.0000069972843],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99573874,0.0009517262,0.0023734355,0.00050513033,0.00015630314,0.00027467217],"domain_scores_gemma":[0.9127522,0.08551587,0.0011844548,0.00039118133,0.00004996252,0.000106316395],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.046427563,0.00013664439,0.00069286925,0.00021587525,0.00068915996,0.00007555023,0.0004446648,0.00008734264,0.00014186138],"category_scores_gemma":[0.0257466,0.00014009506,0.000096971744,0.0001780397,0.00010795248,0.00026349153,0.00031183442,0.00022199447,0.00001654105],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000043994216,0.00010623796,0.030277917,0.0006725061,0.0000932883,0.0000010375759,0.0006788482,0.04097304,0.000006475519,0.7240874,0.0010241508,0.20203511],"study_design_scores_gemma":[0.00097090454,0.000039303748,0.021699263,0.00029278878,0.000011431938,0.000015679983,0.00086860853,0.527913,0.0000068185454,0.4115124,0.03643094,0.00023889197],"about_ca_topic_score_codex":0.00007292261,"about_ca_topic_score_gemma":0.00002144638,"teacher_disagreement_score":0.8305619,"about_ca_system_score_codex":0.0006990571,"about_ca_system_score_gemma":0.00033103605,"threshold_uncertainty_score":0.98245996},"labels":[],"label_agreement":null},{"id":"W4229017484","doi":"10.1177/0272989x221097106","title":"Trends in Author-Reported Cost-Effectiveness Thresholds in the United States from 1995 to 2018: Implications for Discount Rates","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Actuarial science; Economics; Quality-adjusted life year; Econometrics; Psychology; Medicine; Demography; Cost effectiveness; Operations management; Sociology","score_opus":0.4892766028035407,"score_gpt":0.5402100220425117,"score_spread":0.05093341923897099,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4229017484","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8286015,0.00041862632,0.09100391,0.07642558,0.0006680966,0.001593549,0.0010594085,0.00002946904,0.00019990194],"genre_scores_gemma":[0.9812489,0.000021148704,0.0012979357,0.013848946,0.0001254409,0.0027319824,0.00065676233,0.00002809486,0.00004075477],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9948189,0.00077144464,0.0030533506,0.00065172254,0.0002844843,0.00042006202],"domain_scores_gemma":[0.9872932,0.011132835,0.0007720307,0.0006104775,0.00004113368,0.00015032689],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03222163,0.00016837928,0.00067750603,0.001157245,0.0003451385,0.00009740168,0.0006850112,0.00011046501,0.001065919],"category_scores_gemma":[0.008712511,0.00016342357,0.000103622464,0.0014700892,0.000047583635,0.00015388329,0.00019137272,0.00035910323,0.00007117352],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00040489514,0.0007823903,0.71329254,0.00008361461,0.000089515976,0.000022212433,0.009532083,0.0346583,0.0000050747926,0.08984396,0.09996623,0.05131921],"study_design_scores_gemma":[0.0011606944,0.00009040549,0.71238637,0.0001482811,0.0000049810487,0.0000063186526,0.0017410786,0.052733578,4.9200673e-7,0.10617216,0.12532,0.00023563385],"about_ca_topic_score_codex":0.0024212382,"about_ca_topic_score_gemma":0.0019103932,"teacher_disagreement_score":0.15264748,"about_ca_system_score_codex":0.0009414072,"about_ca_system_score_gemma":0.000121618934,"threshold_uncertainty_score":0.99984723},"labels":[],"label_agreement":null},{"id":"W4235755648","doi":"10.1177/0272989x11393284","title":"Calculating Utility Decrements Associated With an Adverse Event: Marginal Tobit and CLAD Coefficients Should Be Used With Caution","year":2011,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Nuclear reactor physics and engineering","field":"Engineering","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Programs for Assessment of Technology in Health Research Institute; McMaster University; St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Tobit model; Econometrics; Event (particle physics); Statistics; Adverse effect; Medicine; Actuarial science; Economics; Mathematics; Internal medicine","score_opus":0.046465737371244306,"score_gpt":0.27336339965056133,"score_spread":0.22689766227931701,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4235755648","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.86050653,0.00001741438,0.13821042,0.0000051137754,0.00012297022,0.000118232834,0.0000062154922,0.00018258576,0.0008305454],"genre_scores_gemma":[0.99735993,0.0000035265848,0.002491951,0.00004074132,0.000029949793,0.000005109411,0.000014312385,0.000050910552,0.0000035640821],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99825394,0.000027412421,0.00025646627,0.0002648543,0.0008647009,0.00033262145],"domain_scores_gemma":[0.9993642,0.00007792419,0.0000488376,0.00020270729,0.00004825399,0.00025806902],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00041164737,0.00020099911,0.00021998762,0.000076779004,0.00010205884,0.0000287045,0.00013786792,0.00012753828,0.00020372117],"category_scores_gemma":[0.00012734189,0.0001581339,0.000028759361,0.0002478751,0.00005545213,0.00019803898,0.00005754147,0.0003098902,0.0000035585406],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0013352821,0.0014089593,0.18765694,0.00035734123,0.000910232,0.0011811221,0.0074397232,0.033069216,0.0026264372,0.0033138026,0.00043717466,0.7602638],"study_design_scores_gemma":[0.0014165319,0.00023672079,0.17044179,0.0010397221,0.000052824886,0.000019735191,0.00019132867,0.82541215,0.0000903334,0.0001262487,0.0005950881,0.00037753623],"about_ca_topic_score_codex":0.000025305619,"about_ca_topic_score_gemma":0.000089914036,"teacher_disagreement_score":0.7923429,"about_ca_system_score_codex":0.000087707434,"about_ca_system_score_gemma":0.000022309689,"threshold_uncertainty_score":0.6448513},"labels":[],"label_agreement":null},{"id":"W4240903860","doi":"10.1177/0272989x0002000314","title":"In reply","year":2000,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Medicine; Psychology","score_opus":0.1998691739961063,"score_gpt":0.5038477459426589,"score_spread":0.30397857194655264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4240903860","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7572624,0.0038206298,0.002414138,0.061595187,0.0036489917,0.0020292276,0.000011694384,0.00044939533,0.1687683],"genre_scores_gemma":[0.95089495,0.00024228661,0.0060602333,0.040911093,0.00024696445,0.00015543986,0.000004631272,0.00002477896,0.0014596222],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9961506,0.00092985376,0.0010960421,0.00031900345,0.0010032123,0.0005012801],"domain_scores_gemma":[0.9947297,0.0038788205,0.00012954051,0.0009252274,0.00009187893,0.00024482483],"candidate_categories":["insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0010765307,0.000113128895,0.00024853577,0.00016299695,0.0004295682,0.000006229069,0.00065086223,0.00027176106,0.05356771],"category_scores_gemma":[0.0052281944,0.00009715952,0.000045483503,0.0004836512,0.000061788414,0.00014551525,0.00023034967,0.0012972158,0.003923495],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019344955,0.00007342774,0.04082342,0.00007070019,0.0000030069962,0.000076811804,0.0047221314,0.000016359816,0.0000024471276,0.0015699026,0.05312793,0.8993204],"study_design_scores_gemma":[0.0018334988,0.000068847665,0.056385074,0.004058719,0.0000033012338,0.0000107807855,0.0017484566,0.006270519,0.0000014913537,0.027709538,0.9016755,0.00023425344],"about_ca_topic_score_codex":0.00022661895,"about_ca_topic_score_gemma":0.0003063677,"teacher_disagreement_score":0.8990862,"about_ca_system_score_codex":0.00021770415,"about_ca_system_score_gemma":0.00038193451,"threshold_uncertainty_score":0.99685204},"labels":[],"label_agreement":null},{"id":"W4244401570","doi":"10.1177/0272989x16682012","title":"38th Annual Meeting of the Society for Medical Decision Making","year":2016,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medical decision making; Management science; Computer science; Operations research; Medicine; Family medicine; Economics; Mathematics","score_opus":0.19778335142073666,"score_gpt":0.4563976460414733,"score_spread":0.25861429462073665,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4244401570","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15643412,0.0015901161,0.7998128,0.03692074,0.0027222985,0.00080800976,0.000272912,0.000064322034,0.0013746844],"genre_scores_gemma":[0.93438566,0.00015765637,0.049067046,0.015091503,0.0010192996,0.00007340794,0.0000024502785,0.00005683282,0.00014614139],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9919564,0.00026257298,0.005301119,0.0007551666,0.0011037505,0.00062097964],"domain_scores_gemma":[0.9729879,0.023220332,0.002435451,0.00082136807,0.00020018649,0.00033478503],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04371069,0.00024158332,0.0010592233,0.00018008963,0.00039769447,0.000047512563,0.001212617,0.0005335139,0.0024038944],"category_scores_gemma":[0.11657909,0.00016965104,0.0007930117,0.0003288411,0.00027409798,0.00031280084,0.00048670563,0.0003063203,0.0002902795],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020739181,0.0002371963,0.04194456,0.0004379363,0.00022042463,0.000005631562,0.0024505146,0.000077156605,0.000012004678,0.095679075,0.2516113,0.6071168],"study_design_scores_gemma":[0.0083731,0.00033893823,0.04756556,0.02502849,0.000059705777,0.000087233835,0.0023964222,0.07828907,0.00003772607,0.4589589,0.37745985,0.0014049885],"about_ca_topic_score_codex":0.000025501886,"about_ca_topic_score_gemma":0.00008364408,"teacher_disagreement_score":0.77795154,"about_ca_system_score_codex":0.00040695348,"about_ca_system_score_gemma":0.0004638454,"threshold_uncertainty_score":0.99850804},"labels":[],"label_agreement":null},{"id":"W4246414783","doi":"10.1177/02729890122062451","title":"Risk-Adjusted Monitoring of Binary Surgical Outcomes","year":2001,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cardiac, Anesthesia and Surgical Outcomes","field":"Medicine","cited_by":82,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Waterloo","funders":"Medical Research Council; Natural Sciences and Engineering Research Council of Canada; Medical Research Council Canada","keywords":"Medicine","score_opus":0.030931835879935296,"score_gpt":0.3506340975862551,"score_spread":0.31970226170631977,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4246414783","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98671246,0.00075984874,0.001374098,0.0010497596,0.0006118068,0.00018516181,0.0000017273458,0.00010475863,0.009200384],"genre_scores_gemma":[0.9937554,0.00080549374,0.0039725807,0.00020246579,0.00037967897,0.000007384806,0.0000029840896,0.000029023753,0.00084497826],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996189,0.00014339031,0.0007967029,0.00035284308,0.0021260427,0.0003920413],"domain_scores_gemma":[0.9939626,0.0048572347,0.00016483995,0.00045170993,0.00011961271,0.00044399442],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014113035,0.00022449186,0.001194811,0.0002397175,0.0000914505,0.000015500991,0.00020870783,0.00031178628,0.0015911371],"category_scores_gemma":[0.004705229,0.00015003113,0.0007254167,0.0005079844,0.00013725615,0.00006445593,0.00012599952,0.00049252383,0.000100523925],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005004149,0.00014299528,0.7804524,0.000020263515,0.00006718206,0.0051570637,0.00004315922,0.000010048503,0.000022790811,0.00036816875,0.00012795781,0.21308756],"study_design_scores_gemma":[0.0026236,0.00017514023,0.9305083,0.00068146567,0.00013686584,0.0011029367,0.00012914019,0.00051698374,0.000043686116,0.00009555664,0.0638187,0.00016761596],"about_ca_topic_score_codex":0.000021323734,"about_ca_topic_score_gemma":0.0000014017245,"teacher_disagreement_score":0.21291995,"about_ca_system_score_codex":0.000053826512,"about_ca_system_score_gemma":0.00008757497,"threshold_uncertainty_score":0.9993215},"labels":[],"label_agreement":null},{"id":"W4280622755","doi":"10.1177/0272989x221099493","title":"Noninferiority Margin Size and Acceptance of Trial Results: Contingent Valuation Survey of Clinician Preferences for Noninferior Mortality","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Hospitalier Universitaire Sainte-Justine; McMaster University; McGill University; Children's Hospital of Eastern Ontario; Health Sciences Centre; BC Children's Hospital; University of Toronto; Sunnybrook Health Science Centre; University of British Columbia; Public Health Ontario; Université de Montréal; Hospital for Sick Children","funders":"","keywords":"Contingent valuation; Margin (machine learning); Actuarial science; Valuation (finance); Medicine; Psychology; Statistics; Economics; Willingness to pay; Computer science; Accounting; Mathematics; Microeconomics","score_opus":0.6880218127831303,"score_gpt":0.6223625294813742,"score_spread":0.06565928330175619,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280622755","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.702294,0.000041846015,0.29387492,0.000073756426,0.0013592832,0.0013241519,0.00084057054,0.000021474856,0.0001700067],"genre_scores_gemma":[0.6676353,0.00002437998,0.33203837,0.00006119332,0.00011605018,0.000094595685,0.0000052248993,0.000014323383,0.000010510206],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9899346,0.0034319996,0.0035411913,0.0005749297,0.0022338966,0.00028336947],"domain_scores_gemma":[0.61765873,0.3786548,0.002188679,0.0006875437,0.0006208394,0.00018939501],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.054097705,0.0001818868,0.0013025338,0.00006905721,0.00015015862,0.00002344365,0.0005047542,0.00020336472,0.000888267],"category_scores_gemma":[0.72775537,0.0001531226,0.00018316568,0.00034392384,0.0003049912,0.0000510375,0.0005066586,0.0004095454,5.76282e-7],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08355233,0.0012829398,0.026565915,0.00041235483,0.00022362308,0.0000090629155,0.00044387692,0.000018155617,0.00008274823,0.0124721015,0.0009163074,0.8740206],"study_design_scores_gemma":[0.029093456,0.0017398426,0.2438665,0.0004098894,0.00015916897,0.0000014154567,0.00010329118,0.0043857857,0.00010636941,0.71977085,0.00014887897,0.00021457172],"about_ca_topic_score_codex":0.00007213908,"about_ca_topic_score_gemma":0.00016657195,"teacher_disagreement_score":0.873806,"about_ca_system_score_codex":0.000052319912,"about_ca_system_score_gemma":0.00035752915,"threshold_uncertainty_score":0.97400546},"labels":[],"label_agreement":null},{"id":"W4282927183","doi":"10.1177/0272989x221100717","title":"A Machine-Learning Approach for Estimating Subgroup- and Individual-Level Treatment Effects: An Illustration Using the 65 Trial","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Hospitalier Universitaire de Sherbrooke; Université de Sherbrooke","funders":"Health Technology Assessment Programme; Medical Research Council; National Institute for Health and Care Research","keywords":"Covariate; Overfitting; Medicine; Parametric statistics; Causal inference; Subgroup analysis; Machine learning; Confidence interval; Econometrics; Statistics; Computer science; Mathematics; Internal medicine","score_opus":0.32824647215952396,"score_gpt":0.4587967880202238,"score_spread":0.13055031586069982,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4282927183","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15554358,0.00006693715,0.8427238,0.000030923075,0.00014555377,0.0012692031,0.000013263738,0.00015702515,0.00004971512],"genre_scores_gemma":[0.43807986,0.000002921958,0.5612696,0.000071009694,0.00014732113,0.00037006263,0.000022237573,0.000027587868,0.000009397392],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99739796,0.000464995,0.0004946829,0.00037794316,0.0009771659,0.00028728048],"domain_scores_gemma":[0.9944075,0.0049033347,0.00028200768,0.00026470018,0.000041836585,0.00010063767],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028934279,0.00021262452,0.0003338603,0.00011054287,0.00097236066,0.00011100252,0.00029525033,0.00010708278,0.000062900224],"category_scores_gemma":[0.006136669,0.0001436153,0.000079569996,0.00020215662,0.00007998808,0.00016356332,0.00028616763,0.0004450534,1.7761768e-7],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019315052,0.00041069798,0.00011362563,0.00008826212,0.000057966015,0.000022360864,0.0023819911,0.004556196,0.00021236957,0.016899472,0.00005181935,0.97327375],"study_design_scores_gemma":[0.0055840947,0.0014140109,0.000015451642,0.00010317876,0.00008886874,0.00007273138,0.00047800795,0.8749583,0.00006722887,0.116898075,0.0001404486,0.00017964753],"about_ca_topic_score_codex":0.00001530222,"about_ca_topic_score_gemma":0.000007689817,"teacher_disagreement_score":0.9730941,"about_ca_system_score_codex":0.00018770686,"about_ca_system_score_gemma":0.00007900228,"threshold_uncertainty_score":0.74787104},"labels":[],"label_agreement":null},{"id":"W4294578992","doi":"10.1177/0272989x221121343","title":"The Fallacy of a Single Diagnosis","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; Canada Research Chairs; Alfred P. Sloan Foundation","keywords":"Fallacy; Medicine; Single use; Intensive care medicine; Psychology; Computer science; Epistemology; Philosophy; Engineering","score_opus":0.03809602560556307,"score_gpt":0.37673629229054345,"score_spread":0.3386402666849804,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4294578992","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9754458,0.0019432119,0.00446939,0.0100604985,0.0016192505,0.00030304017,0.00000886242,0.000069241956,0.006080717],"genre_scores_gemma":[0.996163,0.00018957446,0.001134569,0.0020589603,0.00018671335,0.00006508448,0.0000033231327,0.000014217922,0.00018452929],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9971217,0.000111575115,0.00052469195,0.00021278417,0.0018032216,0.00022602107],"domain_scores_gemma":[0.9353791,0.06366225,0.00015550251,0.0004611834,0.000089246125,0.00025273766],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017007779,0.00008431654,0.000270232,0.000056741545,0.00027867052,0.000014852237,0.00027731297,0.00005985902,0.0033529138],"category_scores_gemma":[0.20136365,0.00005221428,0.00016979755,0.0003378679,0.00014676989,0.00001245724,0.00039923686,0.0003736773,0.000027952066],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00026387646,0.00047569748,0.030811315,0.000005200767,0.00002363615,0.00013261862,0.00007736686,0.000019096775,0.000021122032,0.00079876423,0.03141137,0.93595994],"study_design_scores_gemma":[0.00783268,0.004820041,0.20062512,0.008714167,0.00031679097,0.0007924282,0.0011591137,0.006391895,0.00034452748,0.029402953,0.7390939,0.000506415],"about_ca_topic_score_codex":0.000013365852,"about_ca_topic_score_gemma":0.000005589637,"teacher_disagreement_score":0.93545353,"about_ca_system_score_codex":0.000061892504,"about_ca_system_score_gemma":0.0001601389,"threshold_uncertainty_score":0.9975582},"labels":[],"label_agreement":null},{"id":"W4311021019","doi":"10.1177/0272989x221140803","title":"The Naturalness Bias Influences Drug and Vaccine Decisions across Cultures","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Queen's University","funders":"","keywords":"Naturalness; Context (archaeology); Medicine; Psychology; Biology","score_opus":0.10140468805260669,"score_gpt":0.3868197453188923,"score_spread":0.2854150572662856,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311021019","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9833561,0.0009255202,0.00031566055,0.013361863,0.0015821875,0.00011396728,0.00001236803,0.000047058395,0.00028527033],"genre_scores_gemma":[0.9907604,0.0003157273,0.00019407213,0.0082627395,0.00010024236,0.000030061954,9.811981e-7,0.0000073591614,0.0003284093],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99694914,0.00027869246,0.00033734113,0.000436156,0.0016575597,0.0003410969],"domain_scores_gemma":[0.99455595,0.0049364795,0.000098158285,0.00023071564,0.000027730699,0.00015096524],"candidate_categories":["sts"],"consensus_categories":[],"category_scores_codex":[0.0015007857,0.00012548057,0.0001464785,0.000043854987,0.0027424854,0.00011750531,0.0008289327,0.00005088229,0.00047504148],"category_scores_gemma":[0.0056681125,0.000068438785,0.0000618879,0.00036391293,0.00022599878,0.000098337165,0.0010207362,0.00057892693,0.000035358433],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00033934155,0.00012117751,0.0036624768,0.000002672584,0.0000074112254,0.0002809064,0.00076186366,0.00020279772,0.0013354102,0.0055625644,0.051331248,0.9363921],"study_design_scores_gemma":[0.0025076831,0.00027910265,0.2572442,0.00039176986,0.000016688928,0.0017075464,0.0026292512,0.0049000517,0.0010935721,0.16718853,0.56141543,0.00062616315],"about_ca_topic_score_codex":0.0000057957864,"about_ca_topic_score_gemma":0.000025397068,"teacher_disagreement_score":0.935766,"about_ca_system_score_codex":0.000030092982,"about_ca_system_score_gemma":0.000029806313,"threshold_uncertainty_score":0.9985558},"labels":[],"label_agreement":null},{"id":"W4312129922","doi":"10.1177/0272989x221145048","title":"A Comparison of Ordered Categorical versus Discrete Choices within a Stated Preference Survey of Whole-Blood Donors","year":2022,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Institute for Health and Care Research","keywords":"Categorical variable; Preference; Blood donor; Discrete choice; Donation; Value (mathematics); Econometrics; Social psychology; Actuarial science; Psychology; Statistics; Economics; Mathematics; Medicine","score_opus":0.25853949112003316,"score_gpt":0.3382689019496059,"score_spread":0.07972941082957274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4312129922","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98397374,0.000516189,0.013795275,0.000049186445,0.0005150469,0.00014790763,0.0002011762,0.000011384744,0.00079007924],"genre_scores_gemma":[0.9987603,0.000048542384,0.000962421,0.000029140443,0.000014022842,0.00002134721,0.000097083466,0.000015142072,0.000052005253],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9977866,0.00011066963,0.0012919385,0.00036944187,0.00025335705,0.00018799526],"domain_scores_gemma":[0.9980724,0.0007054719,0.0008398967,0.00027314384,0.000019296218,0.000089833804],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0021018845,0.00012348218,0.00055695965,0.00016332307,0.000111527814,0.000015329219,0.00041450936,0.00008313206,0.0028564143],"category_scores_gemma":[0.001150109,0.00013133358,0.000072953946,0.0003270921,0.00010823447,0.000106153464,0.00033230608,0.000271965,0.000043777894],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049271213,0.00034599882,0.9794294,0.000021062835,0.00007926588,0.0000026262646,0.00072781526,0.0069232606,0.00002172851,0.004149137,0.00023421805,0.007572769],"study_design_scores_gemma":[0.0030856472,0.00063676597,0.9017608,0.000040204934,0.000020268162,0.0000024143255,0.00041953666,0.08782909,0.00006474239,0.0052432553,0.0006565509,0.0002407214],"about_ca_topic_score_codex":0.0011156623,"about_ca_topic_score_gemma":0.00046348787,"teacher_disagreement_score":0.080905825,"about_ca_system_score_codex":0.000106493884,"about_ca_system_score_gemma":0.000047990143,"threshold_uncertainty_score":0.9980551},"labels":[],"label_agreement":null},{"id":"W4313236587","doi":"10.1177/0272989x221146506","title":"Evaluating the Impact and Potential Impact of Machine Learning on Medical Decision Making","year":2022,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Medical decision making; Computer science; Economic impact analysis; Impact assessment; Machine learning; Artificial intelligence; Management science; Risk analysis (engineering); Medicine; Engineering; Medical emergency; Political science","score_opus":0.12859560589951846,"score_gpt":0.5519003418460754,"score_spread":0.42330473594655693,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4313236587","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5448557,0.0040380424,0.015158395,0.0009766988,0.43314528,0.0012513195,0.00008036587,0.00012753178,0.00036661825],"genre_scores_gemma":[0.7728911,0.001209813,0.0010128048,0.00022251307,0.22427449,0.000042053143,0.00017753219,0.00011028426,0.000059415408],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9846983,0.0009372418,0.002043264,0.000973313,0.010577263,0.00077060086],"domain_scores_gemma":[0.9621023,0.033967867,0.0012317452,0.001035442,0.00080397003,0.0008586746],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.011743911,0.00058469205,0.0012310421,0.0006213525,0.00084773643,0.00011454259,0.0007718236,0.0014901946,0.018086376],"category_scores_gemma":[0.13878037,0.00034838583,0.00070798636,0.0007261673,0.00031288184,0.00009712109,0.00072048645,0.00601931,0.000043517764],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0032152787,0.00019213904,0.0014617183,0.00009959736,0.00012825485,0.00021781644,0.00055280875,0.0016775528,0.000017035147,0.0000058977857,0.15284066,0.83959126],"study_design_scores_gemma":[0.002219326,0.026868127,0.009014579,0.037795678,0.0009858848,0.0012459534,0.0017366471,0.84549266,0.0000331004,0.0112768635,0.061688874,0.0016422872],"about_ca_topic_score_codex":0.00080700155,"about_ca_topic_score_gemma":0.000054496195,"teacher_disagreement_score":0.84381515,"about_ca_system_score_codex":0.00090327143,"about_ca_system_score_gemma":0.004289002,"threshold_uncertainty_score":0.9998968},"labels":[],"label_agreement":null},{"id":"W4323036834","doi":"10.1177/0272989x231159381","title":"Efficient Designs for Valuation Studies That Use Time Tradeoff (TTO) Tasks to Map Latent Utilities from Discrete Choice Experiments to the Interval Scale: Selection of Health States for TTO Tasks","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children","funders":"EuroQol Research Foundation","keywords":"Valuation (finance); Econometrics; Statistics; Time-trade-off; Scale (ratio); Computer science; Mathematics; Economics; Psychology","score_opus":0.6200249976242158,"score_gpt":0.5267580317199398,"score_spread":0.09326696590427597,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4323036834","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.52790743,0.0007598342,0.43814448,0.02861735,0.0011750963,0.0025677024,0.0007639969,0.00006001411,0.0000040929394],"genre_scores_gemma":[0.96688616,0.00006781839,0.022129204,0.008744883,0.00031506486,0.0012307412,0.0001722393,0.000060681738,0.00039318702],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99453926,0.0004782899,0.0032242148,0.00070763397,0.0004920209,0.00055855775],"domain_scores_gemma":[0.9877709,0.01030543,0.0011243547,0.00038404253,0.00016846048,0.0002467686],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.021962741,0.00023301312,0.0010397014,0.0004183963,0.00044963358,0.00011286685,0.00038328013,0.00012977804,0.00023859444],"category_scores_gemma":[0.014544769,0.00020786535,0.00019261576,0.00035607605,0.00006556394,0.00015959478,0.00016529072,0.00012861633,0.0004101483],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0014219988,0.0006274699,0.049333587,0.0017215227,0.0010516883,0.0000010900865,0.10992047,0.08989859,0.00009425285,0.0046063294,0.6267469,0.1145761],"study_design_scores_gemma":[0.002009835,0.000979894,0.05541303,0.0020581956,0.000033116306,0.0000012830594,0.008782673,0.8725157,0.00009691238,0.015234186,0.042354904,0.0005203017],"about_ca_topic_score_codex":0.00056495075,"about_ca_topic_score_gemma":0.00033423968,"teacher_disagreement_score":0.7826171,"about_ca_system_score_codex":0.0006993113,"about_ca_system_score_gemma":0.00015634428,"threshold_uncertainty_score":0.9937561},"labels":[],"label_agreement":null},{"id":"W4361002668","doi":"10.1177/0272989x231162069","title":"General-Purpose Methods for Simulating Survival Data for Expected Value of Sample Information Calculations","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Statistical Methods and Inference","field":"Mathematics","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada; Norges Forskningsråd; Medical Research Council; NordForsk; National Institute for Health and Care Research","keywords":"Quantile; Probabilistic logic; Sampling (signal processing); Computer science; Sample size determination; Sample (material); Statistics; Parametric statistics; Survival function; Mathematical optimization; Mathematics; Econometrics; Algorithm; Survival analysis","score_opus":0.32096188952517984,"score_gpt":0.5644152090591844,"score_spread":0.24345331953400456,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4361002668","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0060871094,0.000010776003,0.99184746,0.00011215704,0.0004994301,0.00063627085,0.00059433084,0.00010186394,0.00011061625],"genre_scores_gemma":[0.011844056,0.000004254506,0.9874961,0.00010674766,0.00014596623,0.000074069576,0.00029636777,0.000020905009,0.000011584222],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99756414,0.0002550338,0.0010027611,0.00025329762,0.0006327819,0.0002920157],"domain_scores_gemma":[0.9223311,0.076331824,0.00027289477,0.0006151471,0.00032820614,0.00012080684],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0056405114,0.00013379025,0.0004150176,0.00017398449,0.00017602702,0.0000435023,0.0004950063,0.00015565673,0.00019495834],"category_scores_gemma":[0.26339802,0.00010923518,0.000093849696,0.0004995052,0.000055236833,0.00019397415,0.0003620932,0.00011805032,0.000004193794],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000044110613,0.00002080277,0.000060661678,0.000117564676,0.00001743685,3.1500628e-7,0.00014573021,0.0002935429,0.00007016098,0.2562616,0.0013922912,0.7415758],"study_design_scores_gemma":[0.00037993264,0.000036802212,0.00028969647,0.00013382961,0.00002508087,4.9775184e-7,0.00006539207,0.57669914,0.000043150365,0.41720062,0.0050494056,0.00007645136],"about_ca_topic_score_codex":0.00001528187,"about_ca_topic_score_gemma":0.0000053635977,"teacher_disagreement_score":0.74149936,"about_ca_system_score_codex":0.000024680585,"about_ca_system_score_gemma":0.00010096864,"threshold_uncertainty_score":0.7428067},"labels":[],"label_agreement":null},{"id":"W4377011008","doi":"10.1177/0272989x231171912","title":"Comparison of Preferences and Data Quality between Discrete Choice Experiments Conducted in Online and Face-to-Face Respondents","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Economic and Environmental Valuation","field":"Economics, Econometrics and Finance","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children","funders":"EuroQol Research Foundation; Bristol-Myers Squibb","keywords":"Respondent; Multinomial logistic regression; Comparability; EQ-5D; Statistics; Computer science; Psychology; Medicine; Mathematics; Health related quality of life","score_opus":0.5425458878616087,"score_gpt":0.4799996421422702,"score_spread":0.06254624571933853,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4377011008","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9954153,0.00046269444,0.003111692,0.00031949376,0.000105618106,0.00015935497,0.00030555515,0.000011936368,0.000108406006],"genre_scores_gemma":[0.9984712,0.00015928096,0.0010636176,0.00006477711,0.000027388662,0.000004658237,0.00015085735,0.000008625773,0.000049578284],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9983215,0.000052527157,0.0008571384,0.00047080184,0.00013749236,0.00016054185],"domain_scores_gemma":[0.9987143,0.00061872986,0.00023124342,0.00032037531,0.0000040801565,0.000111217654],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016540462,0.00009729024,0.00042767607,0.00019596334,0.000044381668,0.000027321106,0.0003369418,0.00010313658,0.00019996653],"category_scores_gemma":[0.0013246272,0.000099758654,0.000014516666,0.00019743408,0.00007680585,0.00019670169,0.00070815196,0.00013025952,0.000060985996],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000020678408,0.00003121622,0.97033393,0.000012070481,0.000010266648,5.027823e-7,0.0005671618,0.00004549694,0.000027922604,0.0001678869,0.00014510077,0.028637774],"study_design_scores_gemma":[0.0005814271,0.000043472777,0.9815189,0.00011144155,0.0000025593822,1.8060861e-7,0.0006774114,0.013769246,0.000024194977,0.0024918637,0.00067282334,0.0001064531],"about_ca_topic_score_codex":0.00014310668,"about_ca_topic_score_gemma":0.00011426526,"teacher_disagreement_score":0.028531322,"about_ca_system_score_codex":0.000038372735,"about_ca_system_score_gemma":0.00000875773,"threshold_uncertainty_score":0.406804},"labels":[],"label_agreement":null},{"id":"W4378515301","doi":"10.1177/0272989x231173699","title":"Bayesian Models with Spatial Correlation Improve the Precision of EQ-5D-5L Value Sets","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; Institute for Clinical Evaluative Sciences; Impact; Saint Mary's University; McMaster University","funders":"","keywords":"Bayesian probability; Econometrics; Correlation; Statistics; Bayesian inference; Spatial correlation; Valuation (finance); Mean squared error; Mathematics; Economics","score_opus":0.23585982120301197,"score_gpt":0.41043182593888117,"score_spread":0.1745720047358692,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4378515301","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11752653,0.00028544277,0.86863256,0.009011239,0.0012103806,0.000730938,0.00008113773,0.000082283164,0.0024395224],"genre_scores_gemma":[0.9899615,0.00007577517,0.006950521,0.0024361482,0.00028918716,0.000059086517,0.000025419045,0.00004127806,0.00016107371],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99458784,0.0003463639,0.0033776804,0.0005921409,0.000686032,0.00040995618],"domain_scores_gemma":[0.9918506,0.005491063,0.0016622967,0.00069469213,0.00010116368,0.00020018949],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021585045,0.00019653824,0.00076921756,0.00038770807,0.0002659108,0.000067980975,0.00053740625,0.0002733758,0.00069025776],"category_scores_gemma":[0.010092762,0.0001586411,0.0001301347,0.0005550082,0.00012722584,0.0003728583,0.00018150039,0.00034763417,0.00084922754],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065847096,0.00041639095,0.03492876,0.00057940645,0.00025386523,0.00003961046,0.009037953,0.18830092,0.000011117828,0.3028329,0.05785021,0.4050904],"study_design_scores_gemma":[0.00070269266,0.00014701208,0.010724319,0.00046950407,0.000007204349,0.000006967384,0.00027796574,0.8495256,0.0000028251386,0.13566424,0.0023000822,0.00017162012],"about_ca_topic_score_codex":0.0003297213,"about_ca_topic_score_gemma":0.0001567532,"teacher_disagreement_score":0.872435,"about_ca_system_score_codex":0.00020841854,"about_ca_system_score_gemma":0.00022532204,"threshold_uncertainty_score":0.9999287},"labels":[],"label_agreement":null},{"id":"W4379197079","doi":"10.1177/0272989x231171166","title":"Closed-Form Solution of the Unit Normal Loss Integral in 2 Dimensions, with Application in Value-of-Information Analysis","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"Institute of Circulatory and Respiratory Health; Faculty of Pharmaceutical Sciences, University of British Columbia; British Columbia Lung Association; Genome British Columbia; Canadian Lung Association; AstraZeneca Canada; Reseau canadien de recherche respiratoire; Canadian Thoracic Society; Canadian Institutes of Health Research; Genome Canada; GlaxoSmithKline; AstraZeneca","keywords":"Computer science; Computation; Closed-form expression; Dimension (graph theory); Mathematical optimization; Fraction (chemistry); Monte Carlo method; Scalar (mathematics); Algorithm; Applied mathematics; Mathematics; Statistics","score_opus":0.12599576749446828,"score_gpt":0.41230982065439925,"score_spread":0.28631405315993097,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4379197079","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8940826,0.00007065612,0.101693325,0.003189214,0.0001435956,0.00038181094,0.000033284683,0.000016511185,0.00038900416],"genre_scores_gemma":[0.99704736,0.00004906984,0.0020853097,0.00069421215,0.000026168853,0.00004175757,0.000037353922,0.0000074471486,0.000011291501],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99559283,0.00017670858,0.0033753007,0.00022400453,0.00038290722,0.00024822727],"domain_scores_gemma":[0.9967811,0.0011642189,0.0015092212,0.0004095843,0.000068679634,0.000067170324],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014874602,0.00010592368,0.00062400714,0.0012920306,0.00006988078,0.000015535581,0.00034852684,0.00016349064,0.00009568053],"category_scores_gemma":[0.0045204978,0.0000900797,0.00010939604,0.0027491932,0.00009221527,0.00041014928,0.00011700786,0.00023680426,0.00014388074],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007509778,0.000069393995,0.9020257,0.00013557324,0.00004712742,0.0000010840373,0.0018363135,0.0139016835,0.0000017923052,0.0596824,0.00037762691,0.021846218],"study_design_scores_gemma":[0.00053210364,0.000025274352,0.5998109,0.0002871948,0.000009350333,0.000001068787,0.00049682433,0.3869623,0.0000039625834,0.010973271,0.00081283005,0.000084901476],"about_ca_topic_score_codex":0.0010254309,"about_ca_topic_score_gemma":0.0018712662,"teacher_disagreement_score":0.3730606,"about_ca_system_score_codex":0.00024935836,"about_ca_system_score_gemma":0.00015653773,"threshold_uncertainty_score":0.54117846},"labels":[],"label_agreement":null},{"id":"W4381598380","doi":"10.1177/0272989x231178317","title":"Value-of-Information Analysis for External Validation of Risk Prediction Models","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute; Canadian Institutes of Health Research; National Institutes of Health; Memorial Sloan-Kettering Cancer Center","keywords":"False positive paradox; Population; Computation; Value of information; Econometrics; Computer science; Medicine; Algorithm; Mathematics; Machine learning; Artificial intelligence","score_opus":0.2832958129662357,"score_gpt":0.4518197556432132,"score_spread":0.1685239426769775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4381598380","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29466802,0.00009647592,0.70332956,0.00045727118,0.00040494397,0.00026545857,0.0005045275,0.000035024743,0.00023872775],"genre_scores_gemma":[0.971637,0.00014058794,0.02752589,0.00034023565,0.00014539751,0.000050803843,0.00013421844,0.0000112760235,0.000014600892],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9944356,0.00016483244,0.0045184246,0.0002561291,0.0004119448,0.0002130461],"domain_scores_gemma":[0.992816,0.0037247292,0.002858764,0.00033676342,0.00015500688,0.00010878448],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.025006235,0.00010331166,0.0007320993,0.0012444415,0.00011117842,0.000031687618,0.00025842793,0.00018396061,0.00034716135],"category_scores_gemma":[0.018737422,0.00011806482,0.00027253677,0.00093685114,0.000042009222,0.0007435462,0.000059208392,0.000109090724,0.00020156323],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018229983,0.00011216674,0.16531348,0.00059885636,0.00051781326,5.456739e-7,0.004473778,0.53007036,0.0000045401093,0.22989976,0.0077311513,0.061095264],"study_design_scores_gemma":[0.0004721871,0.000038685735,0.04972132,0.00016156872,0.000042346015,4.7974737e-7,0.0002096249,0.7756929,0.000013327786,0.17298044,0.00058941724,0.00007771079],"about_ca_topic_score_codex":0.00011753909,"about_ca_topic_score_gemma":0.000009638361,"teacher_disagreement_score":0.676969,"about_ca_system_score_codex":0.00013676842,"about_ca_system_score_gemma":0.00008746973,"threshold_uncertainty_score":0.9895282},"labels":[],"label_agreement":null},{"id":"W4383709926","doi":"10.1177/0272989x231182436","title":"Shared Decision Making for Elective Surgical Procedures in Older Adults with and without Cognitive Insufficiencies","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Agency for Healthcare Research and Quality","keywords":"Regret; Cognition; Montreal Cognitive Assessment; Medicine; Scale (ratio); Cognitive test; Physical therapy; Cognitive impairment; Psychiatry","score_opus":0.11132517919390728,"score_gpt":0.46149652197940094,"score_spread":0.35017134278549367,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4383709926","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9833005,0.0011474907,0.010259274,0.00080136885,0.00032485859,0.0034774286,0.00004581774,0.00025161274,0.00039166948],"genre_scores_gemma":[0.9907278,0.00014518182,0.0065152803,0.0009084963,0.0001174406,0.0014656482,0.000030835126,0.00005733384,0.000032024764],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9952751,0.0005659989,0.0010935747,0.0007850357,0.0014303977,0.0008499244],"domain_scores_gemma":[0.98085755,0.017345969,0.0003778642,0.00041165182,0.0007301403,0.0002768508],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0014809876,0.00028103884,0.000533965,0.00060896046,0.0010449067,0.000037033817,0.00047234664,0.00044068517,0.0001942522],"category_scores_gemma":[0.021351643,0.00021016532,0.000056856712,0.0014264125,0.00021213948,0.00023395423,0.0006250598,0.0010873424,0.000056588015],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.012020512,0.00028273388,0.4936596,0.0011783938,0.000031105086,0.00010759251,0.054432467,0.00004448998,0.0000027085464,0.0008235212,0.0017820834,0.4356348],"study_design_scores_gemma":[0.012413611,0.00070570584,0.79799014,0.06981048,0.000028106107,0.000050281287,0.05592677,0.056216784,0.0000039951856,0.0054873326,0.0007890232,0.00057776127],"about_ca_topic_score_codex":0.000039200208,"about_ca_topic_score_gemma":0.0014220187,"teacher_disagreement_score":0.43505704,"about_ca_system_score_codex":0.00022875503,"about_ca_system_score_gemma":0.00094951893,"threshold_uncertainty_score":0.9868919},"labels":[],"label_agreement":null},{"id":"W4385064319","doi":"10.1177/0272989x231188027","title":"Constrained Optimization for Decision Making in Health Care Using Python: A Tutorial","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Childhood Cancer Survivors' Quality of Life","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Queen's University; University of Toronto","funders":"","keywords":"Python (programming language); Computer science; Solver; Optimization problem; Mathematical optimization; Software; Programming language; Algorithm; Mathematics","score_opus":0.061405061495195035,"score_gpt":0.4364930278491112,"score_spread":0.37508796635391617,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385064319","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17602311,0.00077396684,0.8145085,0.0027871146,0.0039613787,0.001421353,0.000032605774,0.00033638833,0.00015560925],"genre_scores_gemma":[0.756583,0.000081370476,0.23783687,0.003430357,0.0018770819,0.000039177485,0.000060169492,0.00008246392,0.000009508655],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99457186,0.00021770615,0.0015252188,0.0007259773,0.0022545233,0.0007047267],"domain_scores_gemma":[0.99451494,0.00408531,0.00033676162,0.00046327867,0.00023094201,0.00036876492],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.003942226,0.00028239394,0.0008443297,0.00069050695,0.00026380416,0.000072184484,0.00026578366,0.00035754745,0.00046092912],"category_scores_gemma":[0.017162522,0.00026759412,0.00022437527,0.001346819,0.00010936127,0.00013230955,0.00021426714,0.00045697446,0.00001983018],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0028162538,0.00014879864,0.0030511403,0.00088321,0.00004634276,0.00027424243,0.006024535,0.02950101,0.000027506747,0.00048991886,0.006302563,0.9504345],"study_design_scores_gemma":[0.0127162095,0.0008794353,0.005296395,0.037507903,0.00007214789,0.00024886845,0.00512065,0.92156607,0.000017397988,0.0034582668,0.012401039,0.0007156342],"about_ca_topic_score_codex":0.000056591507,"about_ca_topic_score_gemma":0.0001949487,"teacher_disagreement_score":0.94971883,"about_ca_system_score_codex":0.0006234979,"about_ca_system_score_gemma":0.0013123534,"threshold_uncertainty_score":0.99997765},"labels":[],"label_agreement":null},{"id":"W4386624723","doi":"10.1177/0272989x231197772","title":"Pricing Treatments Cost-Effectively when They Have Multiple Indications: Not Just a Simple Threshold Analysis","year":2023,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"Natural Sciences and Engineering Research Council of Canada; Canada Research Chairs","keywords":"Reimbursement; Purchasing; Actuarial science; Value (mathematics); Economics; Profit (economics); Microeconomics; Willingness to pay; Operations management; Health care; Computer science","score_opus":0.40485676902061546,"score_gpt":0.4881285417326019,"score_spread":0.08327177271198644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386624723","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.77833724,0.0004354805,0.20470239,0.0104388,0.00064160425,0.0019963314,0.00038699943,0.0003257233,0.0027354483],"genre_scores_gemma":[0.9891875,0.00015643382,0.0033724776,0.0061466987,0.00025294855,0.00043898798,0.00014246511,0.000045315723,0.00025717926],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9941546,0.00030410956,0.0034278773,0.000925173,0.00056491484,0.0006233446],"domain_scores_gemma":[0.98841166,0.008864867,0.0015070643,0.0007742852,0.00006530145,0.00037681137],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.017949361,0.00026728222,0.0012128428,0.0015477025,0.0004964616,0.00017634315,0.0005769552,0.00030101268,0.0017139578],"category_scores_gemma":[0.03073081,0.00028312937,0.0003403142,0.0008544393,0.00007342818,0.00035160864,0.00022906328,0.00031864416,0.0042642755],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009766771,0.000368805,0.87360996,0.00015419489,0.0014943638,0.000043600627,0.009031043,0.0038629714,0.0000043908985,0.015395276,0.025222324,0.07071543],"study_design_scores_gemma":[0.0020384232,0.00009136501,0.4241637,0.00026495827,0.0001455277,0.000005375441,0.0013624477,0.49860737,0.00000480657,0.04695568,0.025844535,0.00051580457],"about_ca_topic_score_codex":0.00051483716,"about_ca_topic_score_gemma":0.0009086916,"teacher_disagreement_score":0.4947444,"about_ca_system_score_codex":0.0007567789,"about_ca_system_score_gemma":0.00014852367,"threshold_uncertainty_score":0.9999621},"labels":[],"label_agreement":null},{"id":"W4386725475","doi":"10.1177/0272989x231197828","title":"Persistent Challenges to a Single Diagnosis","year":2023,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canada Research Chairs","keywords":"Intensive care medicine; Medicine; Computer science","score_opus":0.2617529304617697,"score_gpt":0.43723682466500874,"score_spread":0.17548389420323907,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386725475","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00072862866,0.008068923,0.00029656838,0.981803,0.0017979154,0.0009021664,0.000032353462,0.00034329147,0.0060271453],"genre_scores_gemma":[0.0010414685,0.0058781,0.002374111,0.9743673,0.01250749,0.00064045755,0.000032805834,0.00022107094,0.002937152],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99091005,0.00016574512,0.0007253181,0.001024463,0.0060553113,0.0011190902],"domain_scores_gemma":[0.9934669,0.0046468447,0.00011131122,0.0009027295,0.00036493986,0.0005072372],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0013189267,0.00047492038,0.0011108316,0.00093150523,0.0002037787,0.00007140499,0.0006336895,0.0012293032,0.0020872748],"category_scores_gemma":[0.028370637,0.00036011072,0.0006125694,0.00059056905,0.00018708136,0.000039328705,0.0012234566,0.0030487855,0.0046662907],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000080020145,0.000056553334,0.00006201631,0.0003226918,0.000187699,0.004811647,0.00027101694,0.0000015979235,0.0000022563656,0.0000037975374,0.66830957,0.32589117],"study_design_scores_gemma":[0.0005993971,0.00081548677,0.0006721339,0.009155735,0.00012929378,0.00009028386,0.00016087548,0.00008196479,0.0000060614575,0.00018564248,0.9877716,0.0003315594],"about_ca_topic_score_codex":0.00001557442,"about_ca_topic_score_gemma":0.00003068204,"teacher_disagreement_score":0.32555962,"about_ca_system_score_codex":0.0005309113,"about_ca_system_score_gemma":0.0002480407,"threshold_uncertainty_score":0.9998851},"labels":[],"label_agreement":null},{"id":"W4389751101","doi":"10.1177/0272989x231214791","title":"Reporting Economic Evaluations with Value of Information Analyses Using the CHEERS Value of Information (CHEERS-VOI) Reporting Guideline","year":2023,"lang":"en","type":"letter","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Institute of Health Economics; University of Ottawa; Public Health Ontario; Hospital for Sick Children","funders":"","keywords":"Guideline; Value (mathematics); Value of information; Actuarial science; Economic evaluation; Medicine; Accounting; Business; Economics; Statistics; Mathematics","score_opus":0.45071480569326694,"score_gpt":0.5347654543255534,"score_spread":0.0840506486322865,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389751101","genre_codex":"methods","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07126245,0.00044672607,0.6592222,0.26062194,0.0030395996,0.0022881522,0.00056557945,0.0001739585,0.0023793767],"genre_scores_gemma":[0.20446953,0.000323102,0.20532347,0.5789181,0.0065916874,0.00037492823,0.003439674,0.00035972762,0.00019977923],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.95269656,0.0006077371,0.04393709,0.0006221429,0.0014621619,0.00067428243],"domain_scores_gemma":[0.8979091,0.0052374755,0.09458057,0.0014160497,0.00069903384,0.00015776316],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09271658,0.00047821944,0.0026139189,0.0016966966,0.0003923017,0.00020895735,0.0008816326,0.00094796723,0.0005459257],"category_scores_gemma":[0.15789379,0.00042766755,0.0005247818,0.00084279454,0.00024947486,0.0017984472,0.00029384685,0.0012912018,0.00028735868],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007080538,0.000037101356,0.008152707,0.0023012562,0.0008340439,0.000025227308,0.004334911,0.32957214,0.000004168879,0.006005113,0.63660413,0.012058376],"study_design_scores_gemma":[0.000898328,0.0001089685,0.004199747,0.0036512108,0.00019869501,0.00010970132,0.0023220768,0.87386215,0.000013522852,0.00965692,0.1042909,0.00068779796],"about_ca_topic_score_codex":0.002136734,"about_ca_topic_score_gemma":0.00007205997,"teacher_disagreement_score":0.54429,"about_ca_system_score_codex":0.0012160492,"about_ca_system_score_gemma":0.002004381,"threshold_uncertainty_score":0.9998175},"labels":[],"label_agreement":null},{"id":"W4392854760","doi":"10.1177/0272989x241234070","title":"CAR T-cell Therapy for Diffuse Large B-cell Lymphoma in Canada: A Cost-Utility Analysis","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"CAR-T cell therapy research","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Sunnybrook Health Science Centre; University of Waterloo; Canadian Centre for Applied Research in Cancer Control; University of Guelph; Toronto General Hospital","funders":"Canadian Institutes of Health Research; Canada First Research Excellence Fund","keywords":"Medicine; Clinical trial; Chimeric antigen receptor; Quality-adjusted life year; Cost-effectiveness analysis; Cost effectiveness; Oncology; Internal medicine; Intensive care medicine; Immunotherapy; Cancer","score_opus":0.03373929135459469,"score_gpt":0.3590436528543052,"score_spread":0.3253043614997105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392854760","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9624999,0.0048601413,0.028095864,0.00085372257,0.00045361373,0.0011617928,0.00007415539,0.00006622109,0.0019345784],"genre_scores_gemma":[0.9956255,0.0006079429,0.00056101655,0.001400391,0.00020164953,0.00012807944,0.000043137683,0.000052147454,0.0013801688],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9954097,0.00014770201,0.0006798898,0.00073249877,0.0022563103,0.00077386375],"domain_scores_gemma":[0.99625355,0.0024186,0.000050538434,0.0006440312,0.00013033811,0.00050294213],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002272016,0.00025220492,0.00071856234,0.000603687,0.000089336456,0.00006838177,0.00036324852,0.0002221761,0.007333224],"category_scores_gemma":[0.00084471045,0.00019219318,0.00040141898,0.0018308705,0.00006115192,0.00005309313,0.00013361129,0.0004560384,0.000028547316],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015349255,0.0015076108,0.015651118,0.00017440016,0.00025316223,0.0016512257,0.00028899417,0.000008341953,0.0014598871,0.000033076867,0.008233175,0.96920407],"study_design_scores_gemma":[0.02817574,0.0003550556,0.034306508,0.00071646,0.00033358362,0.000045209403,0.0004993876,0.490039,0.0019343427,0.00055416237,0.44255394,0.00048659637],"about_ca_topic_score_codex":0.05540683,"about_ca_topic_score_gemma":0.38924697,"teacher_disagreement_score":0.9687175,"about_ca_system_score_codex":0.0012121042,"about_ca_system_score_gemma":0.0034533564,"threshold_uncertainty_score":0.9935742},"labels":[],"label_agreement":null},{"id":"W4394572654","doi":"10.1177/0272989x241239899","title":"Capturing Valuation Study Sampling Uncertainty in the Estimation of Health State Utility Values Using the EQ-5D-3L","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Markov chain Monte Carlo; Econometrics; Valuation (finance); Bayesian probability; Prior probability; EQ-5D; Statistics; Posterior probability; Probability distribution; Mathematics; Point estimation; Computer science; Mathematical optimization; Economics; Health related quality of life; Medicine","score_opus":0.6504843824883728,"score_gpt":0.533176660139088,"score_spread":0.11730772234928477,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4394572654","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5732303,0.0020450375,0.41645938,0.006734708,0.00059567136,0.00082363054,0.000021861819,0.000026119013,0.000063284504],"genre_scores_gemma":[0.99290955,0.00004750239,0.004777378,0.0020539418,0.00013896669,0.000039542585,0.000007836219,0.000019467121,0.0000058014048],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99151057,0.0018534642,0.0049626254,0.00057328644,0.0007153583,0.00038470284],"domain_scores_gemma":[0.98993576,0.008147321,0.0012057815,0.00057478086,0.00005860758,0.00007773451],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.112130925,0.00016846797,0.0007121299,0.00036535316,0.00035164185,0.00018186816,0.00049352774,0.000093004666,0.00018842441],"category_scores_gemma":[0.015332869,0.0001268233,0.00010980238,0.0005884861,0.000095206444,0.0003146527,0.000111390465,0.00042775628,0.000063643965],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006757739,0.0005144235,0.053135313,0.0010427892,0.00013215149,0.000010854023,0.09706712,0.34460974,0.0000010377408,0.024511239,0.0013918682,0.4775159],"study_design_scores_gemma":[0.00028603544,0.000056292683,0.05291401,0.0008343487,0.00000644978,0.000005363912,0.0058852774,0.8290207,2.1134899e-7,0.1104515,0.00043260367,0.0001072092],"about_ca_topic_score_codex":0.0018894431,"about_ca_topic_score_gemma":0.00045470052,"teacher_disagreement_score":0.48441097,"about_ca_system_score_codex":0.0005945512,"about_ca_system_score_gemma":0.000439713,"threshold_uncertainty_score":0.9929614},"labels":[],"label_agreement":null},{"id":"W4394610985","doi":"10.1177/0272989x241241328","title":"Creating a Multiply Imputed Value Set for the EQ-5D-5L in Canada: State-Level Misspecification Terms Are Needed to Characterize Parameter Uncertainty Correctly","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University; Impact; Institute for Clinical Evaluative Sciences; Canadian Centre for Applied Research in Cancer Control; Health Sciences Centre; Hospital for Sick Children; University of Toronto; SickKids Foundation; Sunnybrook Health Science Centre","funders":"Canadian Statistical Sciences Institute; Mitacs; Canadian Centre for Applied Research in Cancer Control","keywords":"Statistics; Valuation (finance); Imputation (statistics); Econometrics; Value (mathematics); Sample (material); Sample size determination; Confidence interval; Set (abstract data type); Mathematics; Computer science; Missing data; Accounting; Economics","score_opus":0.38008825439222765,"score_gpt":0.41746842486252,"score_spread":0.03738017047029235,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4394610985","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.57472575,0.00068738905,0.3618523,0.05637897,0.0030669216,0.0018983496,0.0012152946,0.00007056673,0.00010444323],"genre_scores_gemma":[0.9624007,0.00005488411,0.008610587,0.027769031,0.0004597852,0.0004318859,0.0000702865,0.000054185723,0.00014866873],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9944191,0.00024933816,0.003587199,0.0007902937,0.00038206598,0.00057202246],"domain_scores_gemma":[0.97841847,0.019823143,0.00084449776,0.00052993634,0.00008021856,0.00030374777],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.013732986,0.0002468395,0.000830085,0.00035172177,0.0002391115,0.00026041942,0.00056147145,0.00015444808,0.0003552757],"category_scores_gemma":[0.03471042,0.00022174222,0.00012844906,0.000482919,0.00003878132,0.00021612241,0.000114669325,0.00037465754,0.0002399856],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006892791,0.00020605951,0.039395902,0.0013165368,0.0003606011,0.00011556018,0.019646563,0.023663389,0.000047763297,0.014281376,0.28829524,0.61198175],"study_design_scores_gemma":[0.0006811577,0.000044573804,0.08853799,0.0013920307,0.000006965869,0.000011082654,0.0021439553,0.8322436,0.000004478342,0.004984145,0.06961126,0.00033872758],"about_ca_topic_score_codex":0.24110916,"about_ca_topic_score_gemma":0.28904468,"teacher_disagreement_score":0.8085802,"about_ca_system_score_codex":0.002155488,"about_ca_system_score_gemma":0.00094307924,"threshold_uncertainty_score":0.9734206},"labels":[],"label_agreement":null},{"id":"W4396862807","doi":"10.1177/0272989x241249182","title":"The Impact of Model Assumptions on Personalized Lung Cancer Screening Recommendations","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; Canadian Partnership Against Cancer","funders":"National Cancer Institute; Erasmus Universitair Medisch Centrum Rotterdam; Health Canada; Cancer Australia; Partenariat Canadien Contre Le Cancer; Universität Zürich; Centre Hospitalier Universitaire Vaudois; ZonMw; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; National Institutes of Health; Cancer Research UK; Teva Pharmaceutical Industries","keywords":"Lung cancer; Medicine; Lung cancer screening; Cancer; Stage (stratigraphy); Incidence (geometry); Epidemiology; Internal medicine; Oncology; Demography; Biology; Mathematics","score_opus":0.06180540773436391,"score_gpt":0.4747675484882615,"score_spread":0.4129621407538976,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396862807","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4414187,0.03658348,0.4332136,0.07553835,0.0021032263,0.0016260248,0.00028171064,0.00035598842,0.008878917],"genre_scores_gemma":[0.9931628,0.00285848,0.002914237,0.0003346326,0.00017623829,0.00011024683,0.000012551689,0.00002228303,0.00040853492],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9984423,0.000038653918,0.00031403665,0.00023656187,0.0007686481,0.00019983632],"domain_scores_gemma":[0.99801123,0.0014797195,0.000050520302,0.00023231798,0.00007325279,0.00015296771],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0004846962,0.00012307217,0.00021636779,0.00011131747,0.00019513993,0.00005160893,0.000106124455,0.000078037636,0.0019320565],"category_scores_gemma":[0.00044054515,0.00006308842,0.00029226934,0.0002551411,0.00007213478,0.000047637866,0.000047352663,0.00026778746,0.000014563442],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00024079463,0.00016392837,0.00534529,0.000048080157,0.0005398608,0.000049186616,0.00028613934,0.0031385517,0.000020031786,0.0033857017,0.09920995,0.88757247],"study_design_scores_gemma":[0.001056285,0.0001914706,0.010974863,0.008383891,0.00024324055,0.000021733316,0.000046606703,0.972146,0.00002012549,0.0011880193,0.0056333123,0.00009448931],"about_ca_topic_score_codex":0.000068403715,"about_ca_topic_score_gemma":0.000034481625,"teacher_disagreement_score":0.96900743,"about_ca_system_score_codex":0.00040945108,"about_ca_system_score_gemma":0.0003898788,"threshold_uncertainty_score":0.9989803},"labels":[],"label_agreement":null},{"id":"W4399215419","doi":"10.1177/0272989x241256121","title":"Thinking Fast, Slow, and Forever: Daniel Kahneman Obituary","year":2024,"lang":"en","type":"editorial","venue":"Medical Decision Making","topic":"Decision-Making and Behavioral Economics","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canada Research Chairs","keywords":"Obituary; Prospect theory; Psychology; Economics; Computer science; Mathematical economics; Philosophy; Microeconomics; Theology","score_opus":0.0449707019638836,"score_gpt":0.40334184378099586,"score_spread":0.35837114181711227,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399215419","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007473793,0.0027904615,0.021078488,0.0008893124,0.9573523,0.0004410831,0.00025709515,0.00035793538,0.009359569],"genre_scores_gemma":[0.018251535,0.0010948457,0.02327206,0.0016119166,0.9436273,0.000067727546,0.00011031482,0.00043031632,0.011533966],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.976024,0.00039398207,0.0035430503,0.0032127663,0.015762521,0.0010636868],"domain_scores_gemma":[0.9503024,0.044488523,0.0010761404,0.0024084745,0.00077890983,0.0009455077],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.017943349,0.0010548163,0.0021296411,0.0016453636,0.00078202406,0.004328612,0.005268728,0.003220507,0.0023295959],"category_scores_gemma":[0.0793205,0.0007510968,0.0007947191,0.0013163276,0.000565273,0.0006509015,0.0047287815,0.004120836,0.0029287057],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000053153195,0.000032832304,0.00002418543,0.000019647441,0.000020217083,0.0005773114,0.00016573347,0.0000065072363,8.0170986e-7,0.00013899662,0.52381027,0.47515035],"study_design_scores_gemma":[0.000386455,0.00011581706,0.000024841069,0.0027056502,0.00008892057,0.00007298188,0.00018798895,0.002768024,8.2774505e-7,0.3492396,0.6438375,0.0005713856],"about_ca_topic_score_codex":0.00002840158,"about_ca_topic_score_gemma":0.00017131996,"teacher_disagreement_score":0.47457895,"about_ca_system_score_codex":0.00031930025,"about_ca_system_score_gemma":0.0011675141,"threshold_uncertainty_score":0.999494},"labels":[],"label_agreement":null},{"id":"W4401179823","doi":"10.1177/0272989x241264287","title":"Accurate EVSI Estimation for Nonlinear Models Using the Gaussian Approximation Method","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Hospital for Sick Children; Public Health Ontario","funders":"Canadian Statistical Sciences Institute; Natural Sciences and Engineering Research Council of Canada","keywords":"Monte Carlo method; Gaussian; Nonlinear system; Mathematical optimization; Conditional expectation; Computer science; Mathematics; Nonparametric statistics; Spline (mechanical); Conditional probability distribution; Algorithm; Applied mathematics; Econometrics; Statistics","score_opus":0.49757559236396287,"score_gpt":0.5470503735051236,"score_spread":0.04947478114116077,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401179823","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0030166204,0.0016274124,0.9742192,0.018199049,0.0013279465,0.00081205956,0.00010430784,0.000095855736,0.00059755833],"genre_scores_gemma":[0.24164611,0.00004933287,0.750543,0.0064751282,0.0009156399,0.0001571827,0.00004685509,0.00005859213,0.000108168126],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99547577,0.00026684362,0.00302871,0.0005624088,0.00032032127,0.0003459228],"domain_scores_gemma":[0.9923391,0.0062654964,0.00075799075,0.00042153057,0.00006833969,0.00014754848],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.030975506,0.00017520443,0.0005569631,0.00031954024,0.00034683023,0.0003386801,0.00037476447,0.0002281535,0.00039423272],"category_scores_gemma":[0.01198022,0.00015117659,0.0001730106,0.0003480353,0.000050145354,0.00076063356,0.000081588136,0.00026757823,0.0002796646],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046715537,0.00006170996,0.00004849845,0.00080590305,0.000097566706,0.0000049294927,0.00273094,0.18329549,0.0000035260741,0.58746135,0.012675242,0.21276812],"study_design_scores_gemma":[0.00019716559,0.000017557419,0.00003420971,0.0004856458,0.000009118321,0.000013488473,0.0001637928,0.76068,0.000001647563,0.22887224,0.009395427,0.00012969323],"about_ca_topic_score_codex":0.000044528097,"about_ca_topic_score_gemma":0.00001557845,"teacher_disagreement_score":0.57738453,"about_ca_system_score_codex":0.0003754395,"about_ca_system_score_gemma":0.00024404202,"threshold_uncertainty_score":0.99781466},"labels":[],"label_agreement":null},{"id":"W4402710798","doi":"10.1177/0272989x241280611","title":"Incorporating Social Determinants of Health in Infectious Disease Models: A Systematic Review of Guidelines","year":2024,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"COVID-19 epidemiological studies","field":"Mathematics","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"London Health Sciences Centre; University of Toronto; University of Winnipeg; International Centre for Infectious Diseases; Public Health Ontario; York University; University of Ottawa; Impact; Public Health Agency of Canada; Western University","funders":"Johns Hopkins University; Gordon and Betty Moore Foundation","keywords":"Conceptualization; Social determinants of health; Disease; Socioeconomic status; Social epidemiology; Health equity; Infectious disease (medical specialty); Public health; Health policy; Medicine; Disease burden; Environmental health; Population; Computer science","score_opus":0.5676633535359924,"score_gpt":0.5978546136969257,"score_spread":0.03019126016093332,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402710798","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000041175317,0.9824593,0.014618659,0.00031226178,0.00022392953,0.002242289,0.000043368902,0.000060548806,0.00003552216],"genre_scores_gemma":[0.00013906778,0.9948391,0.0037285783,0.0008144639,0.00011125808,0.00029894421,0.0000054194647,0.000056270408,0.000006905267],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98619103,0.0021116654,0.009164732,0.0006190784,0.0015398846,0.00037358448],"domain_scores_gemma":[0.9796102,0.014866075,0.004486315,0.000572715,0.00029396085,0.0001707448],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.017932892,0.00054473546,0.010089233,0.00038663085,0.00008586011,0.00001307573,0.0006955717,0.0003626607,0.000035351834],"category_scores_gemma":[0.22048138,0.00031613748,0.0011475186,0.0011200064,0.00015381335,0.00005156004,0.0010076775,0.00066564,0.000011137996],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000017398753,0.00004407289,0.0000044303883,0.6202272,0.00004041545,0.00003705849,0.000022173317,6.4469197e-7,3.4454062e-10,0.0014026749,0.0012398012,0.37697977],"study_design_scores_gemma":[0.0000741653,0.000031767577,9.2395334e-7,0.82742316,0.0009810772,0.000011998335,0.000008712996,0.0024654137,9.322406e-10,0.16711642,0.0016993509,0.00018699603],"about_ca_topic_score_codex":0.000028046261,"about_ca_topic_score_gemma":0.00006016842,"teacher_disagreement_score":0.3767928,"about_ca_system_score_codex":0.00038246537,"about_ca_system_score_gemma":0.0011083338,"threshold_uncertainty_score":0.9999291},"labels":[],"label_agreement":null},{"id":"W4405792194","doi":"10.1177/0272989x241305414","title":"Microsimulation Estimates of Decision Uncertainty and Value of Information Are Biased but Consistent","year":2024,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"demographic modeling and climate adaptation","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"National Cancer Institute; National Institutes of Health; Canada Research Chairs","keywords":"Probabilistic logic; Econometrics; Value of information; Variance (accounting); Population; Bayesian probability; Statistics; Computer science; Mathematics; Economics; Mathematical economics","score_opus":0.06353643939135127,"score_gpt":0.39139299015467505,"score_spread":0.32785655076332376,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405792194","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5311537,0.0005115588,0.4675438,0.00014438636,0.0003722885,0.00010065185,0.000023509554,0.000030200541,0.00011989085],"genre_scores_gemma":[0.9774658,0.00011941052,0.022221629,0.00013956988,0.00002381493,0.000003940789,0.000011325756,0.000008419401,0.000006097143],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99449265,0.00011527819,0.0017200108,0.00035195876,0.0031373648,0.00018272469],"domain_scores_gemma":[0.98728156,0.011130339,0.00048726168,0.0003697494,0.00059422146,0.00013686759],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005119287,0.00015845691,0.00043124685,0.00083511305,0.00010598736,0.00023230146,0.0003474819,0.0001996127,0.00024187015],"category_scores_gemma":[0.019609863,0.00010898519,0.00016184775,0.0010476513,0.00021762715,0.000524834,0.0001626685,0.00017390292,0.000033962002],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00016568386,0.00003110808,0.0039267745,0.00006892154,0.000016429776,0.000010442733,0.00041083797,0.023366528,0.00038098058,0.0032709942,0.0005625936,0.9677887],"study_design_scores_gemma":[0.00039664644,0.000072477254,0.01372716,0.002052222,0.000029533052,0.000022483488,0.00067885837,0.88034934,0.0001770623,0.10167959,0.00070935127,0.000105275125],"about_ca_topic_score_codex":0.000035984594,"about_ca_topic_score_gemma":0.000017966699,"teacher_disagreement_score":0.96768343,"about_ca_system_score_codex":0.000027564562,"about_ca_system_score_gemma":0.00012165783,"threshold_uncertainty_score":0.98864836},"labels":[],"label_agreement":null},{"id":"W4406112123","doi":"10.1177/0272989x241308768","title":"A Fast Nonparametric Sampling Method for Time to Event in Individual-Level Simulation Models","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"demographic modeling and climate adaptation","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"National Cancer Institute","keywords":"Nonparametric statistics; Event (particle physics); Sampling (signal processing); Computer science; Statistics; Discrete event simulation; Econometrics; Mathematics","score_opus":0.27810593574976383,"score_gpt":0.5104915170126392,"score_spread":0.23238558126287534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406112123","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03181843,0.00009414763,0.9653853,0.0008087637,0.00036695116,0.00055901613,0.000027211741,0.0000493899,0.0008907538],"genre_scores_gemma":[0.6465542,0.000003837921,0.351571,0.0014158988,0.000040178253,0.000057949877,0.000007633834,0.000012697584,0.00033661455],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99342114,0.00028971708,0.0016012173,0.00086644496,0.0033872311,0.00043426573],"domain_scores_gemma":[0.9758666,0.02268724,0.00023092121,0.0005342019,0.00047489285,0.00020612923],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01654519,0.00020255083,0.0005124928,0.0027727312,0.00020004778,0.00030997826,0.00096475694,0.0002679028,0.0004539394],"category_scores_gemma":[0.038739707,0.00016164836,0.00021108096,0.0048562726,0.000027663478,0.00026871386,0.00032841734,0.00026433275,0.00018141346],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000073843104,0.000038390095,0.00009745123,0.000003205808,0.00000575343,0.0000013396221,0.0001420734,0.4752967,0.000009783752,0.00063831656,0.00045923554,0.5232339],"study_design_scores_gemma":[0.00053779635,0.00004115147,0.0023406541,0.00035964703,0.000013421428,0.0000010764729,0.00012981733,0.7557279,0.000004059548,0.23934527,0.0013744985,0.00012471809],"about_ca_topic_score_codex":0.000013877065,"about_ca_topic_score_gemma":0.000041687457,"teacher_disagreement_score":0.6147358,"about_ca_system_score_codex":0.00008811085,"about_ca_system_score_gemma":0.00017858499,"threshold_uncertainty_score":0.9693574},"labels":[],"label_agreement":null},{"id":"W4407139701","doi":"10.1177/0272989x251314031","title":"Development of a Microsimulation Model to Project the Future Prevalence of Childhood Cancer in Ontario, Canada","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Childhood Cancer Survivors' Quality of Life","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Microsimulation; Projections of population growth; Medicine; Population; Demography; Incidence (geometry); Cohort; Cancer; Cancer registry; Childhood cancer; Environmental health; Population growth; Internal medicine","score_opus":0.026355956264115757,"score_gpt":0.356343390752889,"score_spread":0.32998743448877327,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407139701","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9834297,0.0006391848,0.011668856,0.0029240511,0.0003642032,0.0007028924,0.000010740471,0.000008403771,0.00025197465],"genre_scores_gemma":[0.98292863,0.00002513235,0.014031325,0.002757958,0.00005071252,0.000037388832,0.0000021277094,0.0000088447405,0.00015787836],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99720144,0.000055620556,0.0008734903,0.00028220855,0.0013858182,0.00020139606],"domain_scores_gemma":[0.9989653,0.00034006912,0.00015547627,0.0003265671,0.00013403954,0.00007852694],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010113638,0.00013220392,0.00038209476,0.00016415677,0.000055053337,0.000005263847,0.00029407282,0.00010199929,0.00020996439],"category_scores_gemma":[0.0007053016,0.000092727256,0.000052534437,0.00050312374,0.000036967936,0.0000287425,0.00016839767,0.00033923972,3.7107617e-7],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011583875,0.0005425131,0.24664262,0.0014306082,0.00015657983,0.000019482526,0.0396079,0.008881989,0.0006326268,0.00034242586,0.007970961,0.6926139],"study_design_scores_gemma":[0.0013332167,0.000037556212,0.9744077,0.009554551,0.00004365186,0.000003239189,0.0006498561,0.0070609027,0.00070794346,0.0002188017,0.0058510993,0.00013149694],"about_ca_topic_score_codex":0.3652749,"about_ca_topic_score_gemma":0.9872757,"teacher_disagreement_score":0.7277651,"about_ca_system_score_codex":0.00064128067,"about_ca_system_score_gemma":0.012599006,"threshold_uncertainty_score":0.99299866},"labels":[],"label_agreement":null},{"id":"W4407674404","doi":"10.1177/0272989x251314010","title":"Expected Value of Sample Information Calculations for Risk Prediction Model Validation","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; University of British Columbia","funders":"National Cancer Institute; Canadian Institutes of Health Research; National Institutes of Health; Memorial Sloan-Kettering Cancer Center","keywords":"Sample size determination; False positive paradox; Statistics; Population; Computer science; Sample (material); Computation; Econometrics; Data mining; Mathematics; Algorithm; Medicine","score_opus":0.20777516947501984,"score_gpt":0.4581390370143673,"score_spread":0.2503638675393475,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407674404","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.08673726,0.00013656724,0.9073375,0.0025988088,0.00063211325,0.00069276727,0.0008306192,0.00005520028,0.0009791853],"genre_scores_gemma":[0.8620302,0.00006055677,0.1343816,0.0028838837,0.00012788067,0.00019553144,0.0002535989,0.00001344172,0.000053309177],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99502534,0.00014392592,0.004084122,0.00027759897,0.00025454644,0.00021444085],"domain_scores_gemma":[0.9932817,0.0043579517,0.0017010254,0.00036408985,0.00019896457,0.000096252945],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011353097,0.00011924367,0.0005254183,0.0006205304,0.00025423797,0.000052322015,0.00022565415,0.00024000989,0.0002367328],"category_scores_gemma":[0.0526811,0.000141799,0.00013946292,0.0003705081,0.00004502375,0.0006794607,0.000058515085,0.0001473516,0.00007106017],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000105522944,0.00012949346,0.04765128,0.00039868464,0.00010042583,7.8016974e-8,0.001703618,0.14669473,0.0000024668775,0.7319871,0.036683176,0.03454342],"study_design_scores_gemma":[0.00070988986,0.000025603576,0.010859356,0.00023614427,0.000011343673,3.370265e-7,0.00015669515,0.783985,0.0000071251857,0.19590022,0.008022703,0.00008557622],"about_ca_topic_score_codex":0.00017246178,"about_ca_topic_score_gemma":0.000024673025,"teacher_disagreement_score":0.77529293,"about_ca_system_score_codex":0.00031659732,"about_ca_system_score_gemma":0.00024888804,"threshold_uncertainty_score":0.95529854},"labels":[],"label_agreement":null},{"id":"W4407747386","doi":"10.1177/0272989x251318640","title":"The Effect of Patient Decision Aid Attributes on Patient Outcomes: A Network Meta-Analysis of a Systematic Review","year":2025,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cochrane; Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Psychological intervention; Medicine; Decision aids; Meta-analysis; Randomized controlled trial; Decision quality; Feeling; Confidence interval; Psychology; Patient satisfaction; Alternative medicine; Social psychology; Nursing; Internal medicine","score_opus":0.3426469521862825,"score_gpt":0.5351406734317351,"score_spread":0.1924937212454526,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407747386","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000054868196,0.98269814,0.0010273387,0.0003886996,0.0012959803,0.0141316075,0.00023860236,0.000055813238,0.00015835307],"genre_scores_gemma":[0.00019654384,0.98972833,0.0011292866,0.0018987516,0.00002835741,0.0067827706,0.00013089344,0.00006453742,0.000040549654],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9579199,0.023047647,0.012183209,0.0009365471,0.0050250934,0.0008875987],"domain_scores_gemma":[0.7583855,0.22733359,0.008170522,0.004996804,0.0008075088,0.0003060908],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.00987509,0.0009901911,0.018827522,0.0009700341,0.0010847643,0.0000205394,0.0030529483,0.0010298118,0.00067700195],"category_scores_gemma":[0.12517011,0.00044659676,0.0075318916,0.0050282744,0.00017356714,0.000055257282,0.002159007,0.002443833,0.00010411735],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00006663185,0.000052901727,0.00007795154,0.62187994,0.024458641,0.000005894323,0.00013867412,0.000032870692,3.1022523e-10,0.00020113813,0.0040796823,0.34900564],"study_design_scores_gemma":[0.00022330887,0.00040282434,0.0000033408323,0.6885128,0.20286652,0.000001133385,0.00006874442,0.00011830572,1.9123622e-8,0.00014412256,0.10743536,0.00022352654],"about_ca_topic_score_codex":0.000033937275,"about_ca_topic_score_gemma":0.000084543535,"teacher_disagreement_score":0.34878212,"about_ca_system_score_codex":0.0005864759,"about_ca_system_score_gemma":0.0011599148,"threshold_uncertainty_score":0.99985754},"labels":[],"label_agreement":null},{"id":"W4407858132","doi":"10.1177/0272989x251319342","title":"Health Utilities in People with Hepatitis C Virus Infection: A Study Using Real-World Population-Level Data","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Ontario; University of Toronto; University of Waterloo; Children's Hospital of Eastern Ontario; University Health Network","funders":"Canadian Liver Foundation","keywords":"Medicine; Propensity score matching; Hepatitis C; Cohort; Hepatitis C virus; Population; Mental health; Comorbidity; Cohort study; Demography; Environmental health; Immunology; Gerontology; Internal medicine; Psychiatry; Virus","score_opus":0.17994547393743474,"score_gpt":0.4156707202062075,"score_spread":0.23572524626877275,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407858132","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8514199,0.00037751856,0.1384358,0.0041168868,0.00079265237,0.00074870605,0.00012449459,0.000058035395,0.003925978],"genre_scores_gemma":[0.9950222,0.00054777635,0.00244569,0.0016565813,0.00007581662,0.000022990818,0.00002462881,0.000012382292,0.00019198062],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.997966,0.00008071683,0.00094835827,0.000512328,0.00018087265,0.00031173968],"domain_scores_gemma":[0.9987514,0.0002815427,0.0002154578,0.0006305248,0.000020591111,0.000100505444],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022410736,0.0001253415,0.00047579684,0.0007700779,0.00020122224,0.000070832975,0.00030910605,0.000059883867,0.00027865102],"category_scores_gemma":[0.00071543525,0.00012672598,0.00002558078,0.0010864404,0.000024259942,0.00022107661,0.00039171544,0.00022709512,0.00002238423],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000023647928,0.00014232633,0.8830223,0.000058382084,0.00001774023,0.000011055394,0.00035829074,0.00012710679,3.6789443e-9,0.035177078,0.0005328649,0.08052917],"study_design_scores_gemma":[0.0005116684,0.00006433428,0.9484844,0.00043234444,0.000003059592,0.0000014934285,0.00008519551,0.030396296,2.9014645e-8,0.011318736,0.00859196,0.000110483605],"about_ca_topic_score_codex":0.49753574,"about_ca_topic_score_gemma":0.75620407,"teacher_disagreement_score":0.25866833,"about_ca_system_score_codex":0.00040807718,"about_ca_system_score_gemma":0.00016897208,"threshold_uncertainty_score":0.5167736},"labels":[],"label_agreement":null},{"id":"W4407924818","doi":"10.1177/0272989x251319808","title":"Changing Time Representation in Microsimulation Models","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"demographic modeling and climate adaptation","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Microsimulation; Time horizon; Constraint (computer-aided design); Econometrics; Computer science; Inefficiency; Discrete event simulation; Resource (disambiguation); Economics; Simulation; Mathematical optimization; Mathematics; Microeconomics; Engineering","score_opus":0.09839049484844621,"score_gpt":0.4481347265346176,"score_spread":0.3497442316861714,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407924818","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34462434,0.00010220051,0.64793634,0.0005852171,0.00035859025,0.00012279178,0.0000011186603,0.000046854744,0.0062225224],"genre_scores_gemma":[0.99298465,0.000023739349,0.0054088994,0.00085348135,0.00004236276,0.000011652373,0.0000059893496,0.000008331962,0.0006608967],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99536645,0.0002409248,0.0010839743,0.0005973734,0.0024016178,0.00030966476],"domain_scores_gemma":[0.99535096,0.0036911669,0.0001723598,0.00047504308,0.00023221345,0.000078267],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0066535156,0.000119755,0.00028849812,0.0017969831,0.00017063583,0.00022653215,0.0005513338,0.00018611549,0.000542944],"category_scores_gemma":[0.00843168,0.000101085185,0.00011249354,0.0034142684,0.000052638483,0.0004320556,0.00020854153,0.0002170518,0.00021135849],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00007413027,0.000041118223,0.0027663405,0.00000280799,0.000003997083,0.00001488302,0.00044151716,0.09184349,0.00014504467,0.003192041,0.0008902189,0.9005844],"study_design_scores_gemma":[0.0003476553,0.0000072022904,0.0038087384,0.0002887661,0.0000034893696,0.0000018539216,0.00026971873,0.7026198,0.000018179366,0.2923396,0.00022748727,0.00006753167],"about_ca_topic_score_codex":0.000010704995,"about_ca_topic_score_gemma":0.00003262692,"teacher_disagreement_score":0.90051687,"about_ca_system_score_codex":0.000054615568,"about_ca_system_score_gemma":0.00008157502,"threshold_uncertainty_score":0.9999207},"labels":[],"label_agreement":null},{"id":"W4408672482","doi":"10.1177/0272989x251325828","title":"Immediate Death: Not So Bad If You Discount the Future but Still Worse than It Should Be","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; Public Health Ontario; University of Toronto","funders":"EuroQol Research Foundation","keywords":"Time-trade-off; Valuation (finance); Discounting; Time preference; Duration (music); Statistics; Quality-adjusted life year; Econometrics; Willingness to pay; Actuarial science; Mathematics; Demography; Quality of life (healthcare); Medicine; Economics","score_opus":0.26535177404626187,"score_gpt":0.46123908223127524,"score_spread":0.19588730818501338,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408672482","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23917596,0.0057921177,0.10329519,0.63451654,0.009151883,0.0011240307,0.00039148406,0.00012914173,0.006423657],"genre_scores_gemma":[0.84599066,0.00083942164,0.0023706695,0.14653486,0.0021959047,0.00013124992,0.000029513849,0.000047657908,0.0018600697],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.9931223,0.0003907921,0.0042237514,0.00089870417,0.0006504478,0.0007139979],"domain_scores_gemma":[0.9918691,0.0053917496,0.0012101246,0.0010976512,0.00009194715,0.0003394577],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.024203468,0.00033512682,0.0010943847,0.00042398766,0.0005795583,0.00031202773,0.0011735205,0.0004826866,0.0026307486],"category_scores_gemma":[0.016390894,0.0002797726,0.00028507772,0.00046270958,0.00020970151,0.00032912375,0.000393818,0.00080940104,0.001404505],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000254506,0.00033064577,0.06256942,0.00036969388,0.00028452292,0.000059194634,0.00574923,0.00044416974,0.0000027042252,0.4411228,0.38765168,0.10116142],"study_design_scores_gemma":[0.0010943001,0.000044154385,0.058241654,0.00076716224,0.000021952817,0.000011002464,0.00237231,0.023623148,0.0000026799823,0.042227317,0.871174,0.00042034546],"about_ca_topic_score_codex":0.00032417788,"about_ca_topic_score_gemma":0.0006073294,"teacher_disagreement_score":0.6068147,"about_ca_system_score_codex":0.0005123832,"about_ca_system_score_gemma":0.00048383093,"threshold_uncertainty_score":0.9999654},"labels":[],"label_agreement":null},{"id":"W4408673451","doi":"10.1177/0272989x251324936","title":"A Nonparametric Approach for Estimating the Effective Sample Size in Gaussian Approximation of Expected Value of Sample Information","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Forecasting Techniques and Applications","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; Hospital for Sick Children; Public Health Ontario","funders":"Canadian Statistical Sciences Institute; Natural Sciences and Engineering Research Council of Canada","keywords":"Nonparametric statistics; Sample size determination; Gaussian; Sample (material); Probability distribution; Computer science; Statistics; Mathematical optimization; Mathematics","score_opus":0.04552504678224666,"score_gpt":0.4011517871229981,"score_spread":0.3556267403407514,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408673451","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.05747041,0.000021920168,0.94014645,0.00020515437,0.00009146228,0.0012156533,0.000034036,0.00003264557,0.0007822989],"genre_scores_gemma":[0.5161951,6.699198e-7,0.48350585,0.00007282099,0.00001131337,0.0002032996,0.0000061537676,0.0000030095944,0.0000017711421],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9964465,0.00018521032,0.0014452972,0.00027316395,0.0014552375,0.0001945755],"domain_scores_gemma":[0.93772215,0.060465556,0.00073985953,0.00058592495,0.00043830913,0.00004820697],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006681625,0.00012176143,0.00038880054,0.00066746917,0.00015185808,0.000077731376,0.0008624182,0.00015019883,0.00004938669],"category_scores_gemma":[0.210415,0.000075779535,0.00013474542,0.0038340136,0.0001516918,0.00022680197,0.00023226131,0.00019756988,0.0000013940306],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008731936,0.00009930418,0.002350687,0.000056434666,0.000007687265,7.0471586e-8,0.0006074262,0.006868598,0.000023554707,0.027733302,0.00066417304,0.9615014],"study_design_scores_gemma":[0.00035801393,0.000052372496,0.007756931,0.00026061755,0.000008093052,0.0000010458702,0.00032852837,0.74499846,0.00020467641,0.24575661,0.00021886166,0.0000557839],"about_ca_topic_score_codex":0.00013205496,"about_ca_topic_score_gemma":0.000011439745,"teacher_disagreement_score":0.9614457,"about_ca_system_score_codex":0.00006290225,"about_ca_system_score_gemma":0.00011581255,"threshold_uncertainty_score":0.796236},"labels":[],"label_agreement":null},{"id":"W4408725561","doi":"10.1177/0272989x251326069","title":"People Living with Chronic Pain Experience a High Prevalence of Decision Regret in Canada: A Pan-Canadian Online Survey","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia; Institut du Savoir Montfort; University of Ottawa; Centre intégré universitaire de santé et de services sociaux de la Capitale-Nationale; Research Canada; Centres Intégré Universitaires de Santé et de Services Sociaux; Université Laval; Centre Intégré Universitaire de Santé et de Services Sociaux du Saguenay–Lac-Saint-Jean; Centre Intégré Universitaire de Santé et de Services Sociaux du Centre-Sud-de-l'Île-de-Montréal; Université de Sherbrooke","funders":"Institute of Musculoskeletal Health and Arthritis; Canadian Institutes of Health Research","keywords":"Regret; Observational study; Checklist; Scale (ratio); Population; Sample (material); Stratified sampling; Psychology; Medicine; Environmental health; Geography; Computer science; Cartography; Pathology","score_opus":0.1044273161445798,"score_gpt":0.4180405234719196,"score_spread":0.3136132073273398,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408725561","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9809393,0.0026722301,0.012994331,0.0013454126,0.00079729117,0.0010053973,0.00010064609,0.000030118294,0.00011525317],"genre_scores_gemma":[0.9928719,0.00036311464,0.004365221,0.0020642804,0.000037335234,0.00019589644,0.000019533458,0.000023612805,0.000059104088],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99245566,0.0031074458,0.0015787887,0.00055617647,0.0014863894,0.00081556756],"domain_scores_gemma":[0.9648635,0.032475762,0.00042442413,0.0012977226,0.0005100795,0.0004285409],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003404811,0.00022416888,0.0005480231,0.00048373404,0.0005009367,0.000009100622,0.0013120632,0.00028663964,0.0011499634],"category_scores_gemma":[0.045920566,0.00018578241,0.00003562611,0.0016960313,0.000104756815,0.00012695647,0.0005692271,0.001223136,0.000009443976],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.00009516291,0.000037109465,0.8745683,0.0005206738,0.000005205652,0.000019455543,0.002335133,0.00006186661,0.000001105934,0.000104083985,0.0011028392,0.12114909],"study_design_scores_gemma":[0.00063507067,0.000080983795,0.9554315,0.026654437,0.0000050146377,0.0000018288895,0.0038624103,0.012204894,7.9862826e-7,0.00026330704,0.0007004815,0.0001592351],"about_ca_topic_score_codex":0.98380655,"about_ca_topic_score_gemma":0.9998197,"teacher_disagreement_score":0.12098985,"about_ca_system_score_codex":0.003995293,"about_ca_system_score_gemma":0.028610328,"threshold_uncertainty_score":0.9998282},"labels":[],"label_agreement":null},{"id":"W4409168450","doi":"10.1177/0272989x251326908","title":"Understanding Delayed Diabetes Diagnosis: An Agent-Based Model of Health-Seeking Behavior","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Timeline; Psychological intervention; Theory of planned behavior; Medical diagnosis; Health care; Context (archaeology); Psychology; Framing (construction); Health belief model; Applied psychology; Medicine; Gerontology; Nursing; Health promotion; Public health; Computer science; Political science","score_opus":0.7130150731856079,"score_gpt":0.6610780821134697,"score_spread":0.05193699107213812,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409168450","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.31100836,0.00006156582,0.6787847,0.008208548,0.0007086827,0.0007885854,0.000039428083,0.00009355627,0.00030655516],"genre_scores_gemma":[0.9386856,0.000029138353,0.0133297,0.04737471,0.000059281254,0.00046697995,0.000007916175,0.000022128548,0.000024549809],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9929458,0.0015166136,0.002244733,0.0005033604,0.0017404348,0.0010490712],"domain_scores_gemma":[0.98964167,0.008441663,0.0006967603,0.00053243357,0.00019755449,0.0004898989],"candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00982576,0.00019360005,0.00055381615,0.00061986904,0.0013784673,0.000022190521,0.0006294472,0.00026018746,0.0010345686],"category_scores_gemma":[0.007988864,0.00017343032,0.00008764161,0.0010087228,0.00017087675,0.00021813705,0.0002516698,0.0006561156,0.000029301984],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015046763,0.00046293432,0.5772983,0.001250926,0.000039767092,0.000015232319,0.010357251,0.0028618316,0.0003760236,0.0190138,0.036185436,0.35198802],"study_design_scores_gemma":[0.004065604,0.000387711,0.035544567,0.009496833,0.00006481712,4.5614115e-7,0.00912892,0.9166567,0.00008099677,0.020634016,0.0034761422,0.00046326546],"about_ca_topic_score_codex":0.00013106842,"about_ca_topic_score_gemma":0.0006951452,"teacher_disagreement_score":0.9137948,"about_ca_system_score_codex":0.00093326875,"about_ca_system_score_gemma":0.0033740413,"threshold_uncertainty_score":0.9999216},"labels":[],"label_agreement":null},{"id":"W4409696671","doi":"10.1177/0272989x251332597","title":"Optimizing the Harms and Benefits of Cervical Screening in a Partially Vaccinated Population in Ontario, Canada: A Modeling Study","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cervical Cancer and HPV Research","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Toronto; Cancer Care Ontario","funders":"","keywords":"Cervical screening; Medicine; Population; Family medicine; Environmental health; Cervical cancer; Internal medicine","score_opus":0.054347069747631524,"score_gpt":0.3584407811936575,"score_spread":0.304093711446026,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409696671","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99004453,0.0004766978,0.007462317,0.0010463868,0.00006650119,0.00046821096,6.427577e-7,0.0000075800754,0.00042711024],"genre_scores_gemma":[0.99877095,0.00001745843,0.0007207749,0.0004233657,0.000021358914,0.000020605323,0.0000030310723,0.000007849231,0.000014600581],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99759036,0.000099647266,0.00063234684,0.00029041385,0.001114923,0.0002722935],"domain_scores_gemma":[0.9989895,0.00055907876,0.000041013922,0.00019210509,0.000105468716,0.00011281203],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001413285,0.000107259075,0.00037831912,0.00025123198,0.00006871954,0.00002175749,0.00016263885,0.00009827304,0.0006392951],"category_scores_gemma":[0.0012891687,0.00007370991,0.000031324937,0.00061999547,0.00001781992,0.00004577813,0.00025731607,0.0006979566,1.7597142e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008330498,0.00012468814,0.6620796,0.000034665598,0.000016452852,0.00007862586,0.00054990384,0.018442772,0.0000018904046,0.00008513404,0.000016060387,0.31773716],"study_design_scores_gemma":[0.0018005753,0.000056050954,0.6640649,0.00054240326,0.00001398819,0.0000030516048,0.00039142778,0.33286077,0.0000021120416,0.00021020025,0.000013219895,0.000041309315],"about_ca_topic_score_codex":0.8421994,"about_ca_topic_score_gemma":0.99600625,"teacher_disagreement_score":0.31769586,"about_ca_system_score_codex":0.0002885145,"about_ca_system_score_gemma":0.00076571095,"threshold_uncertainty_score":0.6999833},"labels":[],"label_agreement":null},{"id":"W4410850908","doi":"10.1177/0272989x251343082","title":"So You’ve Got a High AUC, Now What? An Overview of Important Considerations when Bringing Machine-Learning Models from Computer to Bedside","year":2025,"lang":"en","type":"review","venue":"Medical Decision Making","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary; Public Health Ontario; University of Toronto","funders":"","keywords":"Machine learning; Computer science; Artificial intelligence; Bridge (graph theory); Software deployment; Key (lock); Model selection; Clinical decision support system; Health care; Clinical Practice; Calibration; Medical physics; Decision support system; Medicine; Software engineering","score_opus":0.33264249588838335,"score_gpt":0.5007955916939023,"score_spread":0.16815309580551896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410850908","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00035321395,0.88813496,0.107317805,0.0010684496,0.0018987011,0.0010495684,0.000041090858,0.00008879648,0.000047406782],"genre_scores_gemma":[0.001565513,0.95383745,0.04047652,0.0025063842,0.0010633891,0.000081048165,0.00031747628,0.000066558925,0.00008568388],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.99403197,0.00042316926,0.002774841,0.0009070921,0.0013758065,0.00048714783],"domain_scores_gemma":[0.9927918,0.0046971105,0.00066909526,0.00080258545,0.0004146705,0.0006247707],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0016926379,0.00048352696,0.0023405,0.00064847205,0.00031338097,0.0002386682,0.00035777426,0.0006376266,0.001889294],"category_scores_gemma":[0.003547473,0.00039788353,0.00045600475,0.00051426265,0.00009114331,0.00032130108,0.00032255595,0.001259387,0.00007527945],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000024097684,0.00010437426,0.000044567143,0.0033442567,0.00009079561,0.000120269804,0.0012179391,0.00015958522,3.3695562e-7,0.0007218627,0.0012142054,0.9929577],"study_design_scores_gemma":[0.00019337812,0.0005272419,0.000029505523,0.4769529,0.0015432521,0.0002163826,0.0006254354,0.119281955,0.00001045614,0.04930885,0.35043916,0.0008714826],"about_ca_topic_score_codex":0.0011619792,"about_ca_topic_score_gemma":0.0007495603,"teacher_disagreement_score":0.99208623,"about_ca_system_score_codex":0.00027117957,"about_ca_system_score_gemma":0.0023970136,"threshold_uncertainty_score":0.9998473},"labels":[],"label_agreement":null},{"id":"W4411241327","doi":"10.1177/0272989x251340941","title":"Decision Frameworks for Assessing Cost-Effectiveness Given Previous Nonoptimal Decisions","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Decision analysis; Computer science; Cost effectiveness; Management science; Operations research; Risk analysis (engineering); Business; Statistics; Mathematics; Economics","score_opus":0.2988048024459036,"score_gpt":0.5268310687258189,"score_spread":0.2280262662799153,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411241327","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.065230414,0.0024536857,0.9173825,0.0068053277,0.0033866463,0.002118983,0.00008829185,0.000110020526,0.0024241738],"genre_scores_gemma":[0.768994,0.00022085397,0.21484292,0.014325843,0.0005468767,0.000794781,0.00003288559,0.000072688745,0.00016918876],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99199545,0.0005217949,0.0049820095,0.0012313989,0.0004995562,0.0007697718],"domain_scores_gemma":[0.94103324,0.0559834,0.00137136,0.00096171134,0.0002554758,0.000394819],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.035522353,0.00036309476,0.0016258304,0.0008673849,0.00072457537,0.00042463967,0.0009645823,0.0010566881,0.0010066705],"category_scores_gemma":[0.13736269,0.00039899527,0.00039751182,0.0007170767,0.00014036283,0.00063368416,0.00036964164,0.0008886817,0.0006763173],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00038926356,0.00038644215,0.024480583,0.00036983046,0.0001588018,0.00001752997,0.00031180657,0.0030295744,0.0000026520843,0.111106746,0.06506561,0.79468113],"study_design_scores_gemma":[0.0037905923,0.0001457736,0.070611045,0.010424193,0.00003909484,0.000019298988,0.00032859703,0.22548793,0.000006588802,0.48346746,0.20491624,0.00076317834],"about_ca_topic_score_codex":0.000038451522,"about_ca_topic_score_gemma":0.000045772897,"teacher_disagreement_score":0.79391795,"about_ca_system_score_codex":0.0009380762,"about_ca_system_score_gemma":0.00052310043,"threshold_uncertainty_score":0.99990654},"labels":[],"label_agreement":null},{"id":"W4411604797","doi":"10.1177/0272989x251346788","title":"Forewarning Artificial Intelligence about Cognitive Biases","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canada Research Chairs","keywords":"Cognition; Cognitive bias; Psychology; Vigilance (psychology); Cognitive psychology; Generative grammar; Artificial intelligence; Computer science","score_opus":0.7150225695664488,"score_gpt":0.6469278259500016,"score_spread":0.06809474361644718,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411604797","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3141544,0.00097437116,0.65434444,0.010105482,0.0040820576,0.00091605633,0.000021153766,0.0002510448,0.0151510015],"genre_scores_gemma":[0.9692348,0.00021091694,0.00323386,0.02618687,0.00062874943,0.00012501673,0.000015453545,0.00002714027,0.0003371659],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9927369,0.0025616651,0.0017163301,0.0006248384,0.0015103108,0.00084997766],"domain_scores_gemma":[0.9280269,0.07005209,0.00048735327,0.00041697294,0.0006162269,0.00040044685],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.009017221,0.00023196584,0.00048577125,0.00040147657,0.0016188272,0.00004815408,0.0005255815,0.0005874561,0.005619884],"category_scores_gemma":[0.23881625,0.00019671908,0.00011048983,0.0009803494,0.00024336234,0.00025830918,0.00062836026,0.0023446556,0.0014608111],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0002394742,0.00008174098,0.011819366,0.00014542376,0.00002247295,0.00008686049,0.0015514853,0.000004599199,0.000005653241,0.015504558,0.0023820482,0.96815634],"study_design_scores_gemma":[0.0016602984,0.0005491867,0.11287308,0.15375741,0.0002738207,0.000030782725,0.072598964,0.046726774,0.00030726413,0.48403496,0.12571765,0.0014698287],"about_ca_topic_score_codex":0.00021732197,"about_ca_topic_score_gemma":0.0009175247,"teacher_disagreement_score":0.9666865,"about_ca_system_score_codex":0.00022196824,"about_ca_system_score_gemma":0.001634817,"threshold_uncertainty_score":0.99995697},"labels":[],"label_agreement":null},{"id":"W4411628502","doi":"10.1177/0272989x251342596","title":"Impact of Limited Sample Size and Follow-up on Partitioned Survival and Multistate Modeling-Based Health Economic Models: A Simulation Study","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Hospital for Sick Children; Canadian Centre for Applied Research in Cancer Control; Health Sciences Centre; Sunnybrook Health Science Centre; SickKids Foundation; University of Toronto","funders":"","keywords":"Sample size determination; Censoring (clinical trials); Statistics; Population; Survival analysis; Computer science; Econometrics; Medicine; Mathematics","score_opus":0.3758509519465261,"score_gpt":0.5091020371469848,"score_spread":0.13325108520045864,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411628502","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.6764626,0.000334187,0.320187,0.0018237599,0.00029736702,0.00060841616,0.00018885142,0.000027981805,0.00006987448],"genre_scores_gemma":[0.99465364,0.00007636339,0.0034870626,0.0016415415,0.000054425702,0.00003632057,0.000017291171,0.000021282525,0.000012059766],"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9950935,0.00043550652,0.003311267,0.0006356731,0.00020634056,0.000317703],"domain_scores_gemma":[0.98636806,0.011903498,0.001019568,0.0003983833,0.00006523584,0.00024526604],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.012450026,0.00020657721,0.0010858382,0.0004370073,0.00022907779,0.00007580723,0.000160205,0.00014151452,0.0001660891],"category_scores_gemma":[0.013239127,0.00022432263,0.00011706041,0.00017263967,0.00006042373,0.00022417623,0.000078802266,0.00018412588,0.000022009926],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004678536,0.00033335164,0.06402783,0.00017201508,0.000112198075,0.0000011251714,0.00096615066,0.91547096,1.6585767e-7,0.008307653,0.00028424402,0.009856425],"study_design_scores_gemma":[0.0031939838,0.0003660606,0.03616826,0.0003809435,0.000006777109,2.868002e-7,0.00029607158,0.9223573,6.286833e-8,0.037029244,0.00005171886,0.00014925955],"about_ca_topic_score_codex":0.0019855571,"about_ca_topic_score_gemma":0.00077372266,"teacher_disagreement_score":0.31819108,"about_ca_system_score_codex":0.0005395532,"about_ca_system_score_gemma":0.00040623915,"threshold_uncertainty_score":0.9950728},"labels":[],"label_agreement":null},{"id":"W4412022710","doi":"10.1177/0272989x251346213","title":"Organ Donation Decisions: When Deviating from the Status Quo Heightens Perceived Vulnerability","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Organ Donation and Transplantation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sunnybrook Health Science Centre","funders":"","keywords":"Status quo; Organ donation; Vulnerability (computing); Donation; Opt-out; Psychology; Status quo bias; Social psychology; Affect (linguistics); Medicine; Transplantation; Political science; Business; Law; Surgery; Advertising; Computer security","score_opus":0.02629408031786699,"score_gpt":0.3418043409329664,"score_spread":0.3155102606150994,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412022710","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.68358403,0.00043202902,0.30077347,0.010895375,0.00069553085,0.00047600313,0.000025909401,0.00015036944,0.0029672775],"genre_scores_gemma":[0.97179496,0.0003363612,0.020272993,0.0071301367,0.00020633923,0.000015831685,0.000115952,0.00001884272,0.00010858697],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99671686,0.00023213014,0.00071790477,0.0004638279,0.0015281463,0.00034112087],"domain_scores_gemma":[0.9924849,0.006417447,0.000115883995,0.0004665175,0.0002889358,0.0002262896],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001079869,0.00018527175,0.0003151775,0.00016418117,0.00040958033,0.00008852256,0.00025886943,0.00022574009,0.0042482037],"category_scores_gemma":[0.015378772,0.00011812555,0.00011937617,0.0005086848,0.00011159274,0.00012946514,0.000091760914,0.0005541245,0.00010966495],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004049772,0.00018189113,0.12232543,0.000023858485,0.00008684448,0.000058279937,0.0018141975,0.000019168561,0.00093654566,0.0022092408,0.0041033654,0.8678362],"study_design_scores_gemma":[0.0026376178,0.000052271134,0.9408105,0.0019936473,0.00015005292,0.000017490822,0.0005504113,0.012588671,0.0002680364,0.03116422,0.009609395,0.00015769868],"about_ca_topic_score_codex":0.000090836926,"about_ca_topic_score_gemma":0.00035643933,"teacher_disagreement_score":0.8676785,"about_ca_system_score_codex":0.00021526439,"about_ca_system_score_gemma":0.00042692196,"threshold_uncertainty_score":0.996662},"labels":[],"label_agreement":null},{"id":"W4412393289","doi":"10.1177/0272989x251346894","title":"Process for Rapid Co-development of a Decision Aid Prototype for Population-wide Cancer Screening","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Centre Intégré de Santé et Services Sociaux de la Gaspésie; Ministère de la Santé et des Services Sociaux (Québec); Institut National d'Excellence en Santé et en Services Sociaux; Université du Québec; Corporation d’Aménagement et de Protection de la Sainte-Anne","funders":"","keywords":"Delphi method; Knowledge translation; Process (computing); Process management; Knowledge management; Best practice; Computer science; Checklist; Population; Corporate governance; Stakeholder; Business; Medicine; Psychology; Public relations; Political science; Artificial intelligence","score_opus":0.24531577281920877,"score_gpt":0.5480549138819828,"score_spread":0.302739141062774,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412393289","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09298847,0.001984184,0.8900931,0.0024055906,0.0011573253,0.010636319,0.00005149359,0.00013989997,0.00054360164],"genre_scores_gemma":[0.7544803,0.000043519343,0.2331172,0.0023599565,0.00012702735,0.009614332,0.00009373909,0.000043188324,0.000120756806],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955367,0.000270582,0.0020650597,0.0004807685,0.0010989067,0.00054802606],"domain_scores_gemma":[0.9847944,0.01259847,0.0006935319,0.00062654243,0.0010958728,0.00019119387],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0019239332,0.00020987587,0.0005358504,0.0003832684,0.0014300008,0.000010796792,0.0007626087,0.0004124403,0.00046913876],"category_scores_gemma":[0.018084304,0.00017909947,0.00011924369,0.000539156,0.00005777507,0.00013518223,0.00026386662,0.0005090153,0.000008758509],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002385631,0.00007955192,0.08730548,0.0019219461,0.000038879738,4.734635e-7,0.0025437025,0.000094985226,0.000008388402,0.0007953023,0.0044894246,0.9003362],"study_design_scores_gemma":[0.018406749,0.00061749294,0.16097093,0.087035194,0.00012593064,0.0000015792283,0.00789075,0.07612469,0.00050661917,0.06513106,0.582123,0.00106604],"about_ca_topic_score_codex":0.000046908786,"about_ca_topic_score_gemma":0.0003150225,"teacher_disagreement_score":0.8992702,"about_ca_system_score_codex":0.00026926826,"about_ca_system_score_gemma":0.002050162,"threshold_uncertainty_score":0.99987},"labels":[],"label_agreement":null},{"id":"W4414252417","doi":"10.1177/0272989x251368866","title":"Integrating Shared Decision Making and Decision Support Tools into Clinical Practice Guidelines: What Does It Take? A Qualitative Study","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Decision support system; Qualitative research; Inclusion (mineral); Clinical decision support system; Clinical Practice; Decision aids; MEDLINE; Decision analysis","score_opus":0.33853877625651563,"score_gpt":0.6476444398542482,"score_spread":0.3091056635977326,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414252417","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3101209,0.0008157289,0.657962,0.02421595,0.0039341487,0.0021905727,0.0000126956475,0.00017690935,0.0005710442],"genre_scores_gemma":[0.44706964,0.000953522,0.5135829,0.03751878,0.0006277941,0.00008509117,0.00004135122,0.00006193042,0.00005896553],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.98115414,0.0016755797,0.009422978,0.002071291,0.0049129804,0.00076300127],"domain_scores_gemma":[0.7797289,0.2089774,0.00251376,0.0020414747,0.0057808612,0.00095759693],"candidate_categories":["metaresearch","metaepi_narrow","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04152686,0.0006823874,0.0017303607,0.00074881356,0.0005708982,0.0011351003,0.00076452194,0.0007128113,0.002211627],"category_scores_gemma":[0.6809735,0.00046997258,0.00043187503,0.0014597105,0.00032588656,0.0031829302,0.0016719011,0.0021015771,0.00015174985],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005400322,0.0010349042,0.0022496213,0.00006815408,0.00024265761,0.0007403627,0.009535542,0.000007161773,0.000016504739,0.0002633292,0.035011854,0.94542956],"study_design_scores_gemma":[0.018622654,0.0038603886,0.010263014,0.032329958,0.0016950005,0.00047061473,0.7111725,0.030287595,0.000009340159,0.05284149,0.13733463,0.0011127677],"about_ca_topic_score_codex":0.000084533356,"about_ca_topic_score_gemma":0.0011045906,"teacher_disagreement_score":0.9443168,"about_ca_system_score_codex":0.00038488366,"about_ca_system_score_gemma":0.0013713244,"threshold_uncertainty_score":0.99990183},"labels":[],"label_agreement":null},{"id":"W4416222288","doi":"10.1177/0272989x251388915","title":"Vaccination Strategies against HPV Infection and Cervical Cancer in China: A Transmission Modeling Study","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Cervical Cancer and HPV Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"National Natural Science Foundation of China; Natural Science Foundation of Guangxi Zhuang Autonomous Region","keywords":"Vaccination; Cervical cancer; HPV infection; Transmission (telecommunications); Incidence (geometry); Cancer; Epidemiology; Disease burden; Disease Eradication","score_opus":0.02889525165772696,"score_gpt":0.42143614367560567,"score_spread":0.39254089201787873,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416222288","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.87179583,0.0011981651,0.11608666,0.0015370146,0.00016714759,0.00057671685,8.074903e-7,0.000054028613,0.008583644],"genre_scores_gemma":[0.99810076,0.0011432287,0.00021022833,0.00033583716,0.00009939631,0.000059764287,0.0000025724348,0.0000122277725,0.000036012003],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9977816,0.00008947517,0.00045263272,0.0003998993,0.0010121529,0.0002642534],"domain_scores_gemma":[0.99933463,0.00018328977,0.000026455058,0.00015750364,0.00012023679,0.00017787842],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00093428034,0.00014196083,0.0003353143,0.00044560918,0.00012428784,0.00007208059,0.00009233164,0.00018923235,0.0027807574],"category_scores_gemma":[0.0004144291,0.00010807853,0.00005493926,0.0007382121,0.000024616957,0.000140569,0.000112804766,0.00065063004,0.0000027849674],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006370704,0.00030849563,0.045511395,0.00016912495,0.000017106682,0.00006125851,0.00036792838,0.0014319095,0.000026363177,0.00015939517,0.00003927648,0.9512707],"study_design_scores_gemma":[0.0023491357,0.0001404304,0.57943904,0.0004135343,0.00002171573,0.0000032243895,0.00025312402,0.4148049,0.000006262097,0.0023472148,0.00015275108,0.000068670925],"about_ca_topic_score_codex":0.0004393941,"about_ca_topic_score_gemma":0.0015038289,"teacher_disagreement_score":0.95120203,"about_ca_system_score_codex":0.00019919315,"about_ca_system_score_gemma":0.0003742172,"threshold_uncertainty_score":0.99813086},"labels":[],"label_agreement":null},{"id":"W4416910737","doi":"10.1177/0272989x251395096","title":"When Do Published Cost-Effectiveness Analyses Include Societal Costs? An Empirical Analysis, 2013–2023","year":2025,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"AstraZeneca United States","keywords":"Odds; Odds ratio; Inclusion (mineral); Health care; Cost–benefit analysis; Productivity; Logistic regression; Economic impact analysis; Multivariate analysis","score_opus":0.43270332424000224,"score_gpt":0.5730518000136273,"score_spread":0.14034847577362508,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416910737","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23060909,0.0035945117,0.7230774,0.03253817,0.0019403872,0.0011956309,0.00020250787,0.00017085945,0.0066714217],"genre_scores_gemma":[0.97130764,0.00012707137,0.007227192,0.02020741,0.0004248991,0.00021959245,0.0001440426,0.000036794325,0.00030534144],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99079895,0.0014805234,0.0049984055,0.0013654493,0.00066884904,0.00068780687],"domain_scores_gemma":[0.9879775,0.008661235,0.0014372137,0.0011459978,0.00022409247,0.0005539278],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.04371259,0.00034302252,0.0021117367,0.0016834046,0.00048051827,0.00057575735,0.0011174905,0.0005762334,0.0068453955],"category_scores_gemma":[0.031676847,0.0003647131,0.0005768817,0.0021077911,0.00019766222,0.00097301847,0.00045319967,0.0006259886,0.0010340597],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019872251,0.000764216,0.5885725,0.00030593164,0.0021680202,0.000032525666,0.0010831167,0.002688293,0.000004105304,0.031003257,0.32613757,0.04704171],"study_design_scores_gemma":[0.0025988927,0.00013892895,0.48676914,0.0007813502,0.00028056075,0.000007740437,0.0010875335,0.27457955,0.000004873045,0.12908426,0.10377763,0.0008895272],"about_ca_topic_score_codex":0.00072764454,"about_ca_topic_score_gemma":0.00089594617,"teacher_disagreement_score":0.7406986,"about_ca_system_score_codex":0.001370217,"about_ca_system_score_gemma":0.00051576056,"threshold_uncertainty_score":0.9998805},"labels":[],"label_agreement":null}]}