{"meta":{"query_hash":"ded5295a5df6","filters":{"venue":"Journal of Comparative Effectiveness Research"},"cohort_total":135,"direct_labels_cover":2,"predictions_cover":135,"exported":135,"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/ded5295a5df6","api":"https://metacan.xera.ac/api/v1/cohort?venue=Journal+of+Comparative+Effectiveness+Research"},"results":[{"id":"W1877489180","doi":"10.2217/cer.15.38","title":"Systematic overview of cost–effectiveness thresholds in ten countries across four continents","year":2015,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":149,"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; Comparative effectiveness research; MEDLINE; Pathology; Alternative medicine","score_opus":0.7926023616127016,"score_gpt":0.6204472118303856,"score_spread":0.17215514978231594,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1877489180","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":[0.9749052,0.017498987,0.0025187249,0.00073893234,0.0006016724,0.003007142,0.000095821386,0.000008388166,0.00062516745],"genre_scores_gemma":[0.9990392,0.00023270998,0.0001295858,0.00007948438,0.00015370488,0.00028444073,0.0000035723508,0.000029239318,0.000048074337],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9816668,0.0111043025,0.005330349,0.00042625866,0.0007738183,0.00069847656],"domain_scores_gemma":[0.97763616,0.015635071,0.0036985967,0.0004836209,0.0022170246,0.00032951593],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.13532008,0.00028109213,0.0037814248,0.0010221502,0.00015306904,0.00013981923,0.0007381265,0.00016951845,0.00002980064],"category_scores_gemma":[0.010178123,0.000272573,0.0002649,0.0009652708,0.0003724419,0.0010210014,0.00016756167,0.00072415464,0.00025419262],"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":[0.0022650773,0.00093886914,0.8672377,0.05586817,0.00090782484,0.000054065007,0.013461934,0.006722766,0.0000938594,0.050567545,0.0018452494,0.0000369549],"study_design_scores_gemma":[0.0099017685,0.0016652127,0.9113455,0.029075125,0.000027015829,0.00010994322,0.0076130773,0.0030398709,0.0008947223,0.034090143,0.0016655389,0.00057212025],"about_ca_topic_score_codex":0.00055331143,"about_ca_topic_score_gemma":0.00016108283,"teacher_disagreement_score":0.12514195,"about_ca_system_score_codex":0.0020552557,"about_ca_system_score_gemma":0.00076274184,"threshold_uncertainty_score":0.99997264},"labels":[],"label_agreement":null},{"id":"W1918795463","doi":"10.2217/cer.15.40","title":"Performance of the disease risk score in a cohort study with policy-induced selection bias","year":2015,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Bone health and osteoporosis research","field":"Medicine","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":"Queen's University; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences; University of Toronto; Toronto Public Health; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Selection bias; Propensity score matching; Cohort; Cohort study; Selection (genetic algorithm); Disease; Demography; Internal medicine; Pathology","score_opus":0.2018569255815383,"score_gpt":0.4763414049829235,"score_spread":0.2744844794013852,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1918795463","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":[0.9962931,0.00016051915,0.000010505319,0.00022193807,0.000055902103,0.0028793784,0.0000023766029,0.000003886976,0.0003723568],"genre_scores_gemma":[0.9996237,0.000041830237,0.000018338344,0.0000074890863,0.00013994957,0.00010573008,5.7267295e-7,0.000014981655,0.000047401263],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99132824,0.00506221,0.0005080852,0.00021721087,0.0024431546,0.00044108852],"domain_scores_gemma":[0.99428993,0.0010577113,0.00032519968,0.00029841822,0.0034796614,0.0005490784],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.011333054,0.00014814961,0.00071378157,0.0009365244,0.0001752031,0.000024609762,0.00024991666,0.000054255284,0.00000688952],"category_scores_gemma":[0.0017432842,0.00008021205,0.000086881795,0.0027704576,0.00026344127,0.0002035337,0.00009541892,0.0016588651,0.0000052628075],"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":[0.027313499,0.0014582222,0.9672339,0.00023514642,0.00011840154,0.00003713794,0.0011402238,0.00074703665,0.0010807235,0.000011860527,0.000052332038,0.00057153887],"study_design_scores_gemma":[0.005500516,0.013490635,0.9712512,0.0007943958,0.000041854364,0.000056900248,0.0010534357,0.0017036665,0.0059797955,0.00003677341,0.00002322952,0.0000675807],"about_ca_topic_score_codex":0.001086218,"about_ca_topic_score_gemma":0.00036109652,"teacher_disagreement_score":0.021812983,"about_ca_system_score_codex":0.0008753417,"about_ca_system_score_gemma":0.005550972,"threshold_uncertainty_score":0.9847186},"labels":[],"label_agreement":null},{"id":"W1980477074","doi":"10.2217/cer.12.68","title":"Complex questions command complex analyses: comparative effectiveness of drug treatment strategies in atrial fibrillation","year":2012,"lang":"en","type":"editorial","venue":"Journal of Comparative Effectiveness Research","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Atrial fibrillation; Drug; Comparative effectiveness research; Internal medicine; Intensive care medicine; Cardiology; Pharmacology; Alternative medicine","score_opus":0.3658750221936212,"score_gpt":0.5605074714917893,"score_spread":0.19463244929816803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1980477074","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":[0.8511642,0.009028086,0.0023895772,0.00016166884,0.12376897,0.008917721,0.0003704268,0.000060546125,0.004138793],"genre_scores_gemma":[0.8453562,0.00053218997,0.00031958148,7.680864e-7,0.15306814,0.000018287174,0.00055992143,0.000050209033,0.0000946981],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9799811,0.013544959,0.0020492245,0.00058977073,0.0030553648,0.00077953964],"domain_scores_gemma":[0.9599326,0.032840934,0.0016923097,0.0005610727,0.004632317,0.0003407729],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.009136927,0.00081603334,0.005185589,0.0028424093,0.0003193252,0.00018666984,0.00045097555,0.0005464003,0.0001459775],"category_scores_gemma":[0.00088944036,0.0006319559,0.0013782971,0.001924418,0.0008425621,0.0007750636,0.00020833607,0.0017002167,0.00003160713],"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":[0.26352805,0.002212771,0.29400262,0.017760009,0.028743548,0.00042765014,0.018128281,0.17060673,0.059137467,0.006873348,0.13568456,0.0028949478],"study_design_scores_gemma":[0.027410531,0.009741514,0.75647664,0.00611737,0.002093018,0.000030799263,0.0039553153,0.001993276,0.0042450908,0.0015135908,0.18544315,0.000979674],"about_ca_topic_score_codex":0.0007860707,"about_ca_topic_score_gemma":0.000250855,"teacher_disagreement_score":0.46247405,"about_ca_system_score_codex":0.0023239632,"about_ca_system_score_gemma":0.0021675471,"threshold_uncertainty_score":0.99961317},"labels":[],"label_agreement":null},{"id":"W2002611374","doi":"10.2217/cer.13.93","title":"Comparative effectiveness research in antineoplastic-induced nausea and vomiting control in children","year":2014,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Nausea and vomiting management","field":"Medicine","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":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Medicine; Antiemetic; Nausea; Vomiting; Intensive care medicine; Adverse effect; Psychological intervention; Anesthesia; Psychiatry; Pharmacology","score_opus":0.2884027624500407,"score_gpt":0.5531594896994838,"score_spread":0.2647567272494431,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2002611374","genre_codex":"review","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":[0.48605004,0.50594735,0.00023665094,0.0000332359,0.00017266197,0.006694071,0.000013504998,0.000014636643,0.00083783665],"genre_scores_gemma":[0.74505174,0.2539448,0.000050871287,0.0000052381515,0.00048256788,0.0003624361,0.000014549371,0.000072210874,0.000015563073],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9583795,0.03430832,0.0020840804,0.0010966477,0.0025976202,0.0015338254],"domain_scores_gemma":[0.9571071,0.03858517,0.00090506853,0.0005755088,0.002331496,0.0004956539],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.044088587,0.0008634633,0.008583523,0.0067870994,0.00034530912,0.00024895326,0.000749002,0.0005615222,0.000014669376],"category_scores_gemma":[0.0018905492,0.0006616692,0.00056178175,0.003948489,0.0007647741,0.00033017964,0.0004835004,0.0077743675,0.000056047917],"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":[0.02941699,0.009015501,0.45407632,0.20336121,0.011928634,0.0045693563,0.0032958945,0.00049702154,0.0020394092,0.0043211104,0.000624229,0.27685434],"study_design_scores_gemma":[0.029574992,0.009174394,0.6830186,0.2506862,0.0009633154,0.0013587687,0.0005748746,0.0007831615,0.00061695685,0.00062844343,0.021330517,0.0012897613],"about_ca_topic_score_codex":0.00016104642,"about_ca_topic_score_gemma":0.00005184798,"teacher_disagreement_score":0.27556458,"about_ca_system_score_codex":0.0017412864,"about_ca_system_score_gemma":0.0016324817,"threshold_uncertainty_score":0.9995835},"labels":[],"label_agreement":null},{"id":"W2003321779","doi":"10.2217/cer.14.76","title":"EORTC QLQ-BR23 and FACT-B for the assessment of quality of life in patients with breast cancer: a literature review","year":2015,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Cancer survivorship and care","field":"Medicine","cited_by":131,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Response Biomedical (Canada); Health Sciences Centre; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Breast cancer; MEDLINE; Quality of life (healthcare); Population; Family medicine; Cancer; Gerontology; Clinical psychology; Internal medicine; Environmental health; Nursing","score_opus":0.22569149329793603,"score_gpt":0.5681167581685256,"score_spread":0.34242526487058955,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2003321779","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":[0.0073867263,0.98757595,0.00007845432,0.0001052538,0.00012351807,0.0041415617,0.0005193704,0.000001393695,0.00006776591],"genre_scores_gemma":[0.055135068,0.9443138,0.000021692314,0.000012763496,0.00011438867,0.00033249462,0.000034290653,0.000023361632,0.000012111481],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.9925616,0.004088168,0.0012110816,0.00026136104,0.0016122679,0.0002655447],"domain_scores_gemma":[0.9873976,0.004044993,0.0013706791,0.00034366094,0.006635046,0.00020800745],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008916451,0.0003127771,0.004570063,0.0004304742,0.000052486368,0.000026197957,0.00030133143,0.0001511736,0.000016270149],"category_scores_gemma":[0.0002689132,0.00015216952,0.00039020347,0.0011528991,0.00030715647,0.00012902201,0.00009219569,0.001405473,1.6908787e-7],"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":[0.0063986285,0.0009825226,0.08742684,0.51333404,0.00247929,0.000011943251,0.0010072457,0.000009740764,0.0000020371367,0.00019398023,0.00075705006,0.3873967],"study_design_scores_gemma":[0.004784128,0.0030760826,0.45452324,0.44776896,0.0009775525,0.00003464502,0.00019829566,0.000006063362,0.0000029794103,0.00003052539,0.0883671,0.0002304523],"about_ca_topic_score_codex":0.00022830228,"about_ca_topic_score_gemma":0.00020888483,"teacher_disagreement_score":0.38716626,"about_ca_system_score_codex":0.0005531751,"about_ca_system_score_gemma":0.0030990748,"threshold_uncertainty_score":0.6205293},"labels":[],"label_agreement":null},{"id":"W2007231551","doi":"10.2217/cer.13.92","title":"Comparative effectiveness of transurethral resection of bladder tumors and office fulguration for recurrent bladder tumors","year":2014,"lang":"en","type":"letter","venue":"Journal of Comparative Effectiveness Research","topic":"Bladder and Urothelial Cancer Treatments","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":"McGill University Health Centre; Université de Montréal","funders":"","keywords":"Fulguration; Medicine; Resection; Bladder tumor; Surgery; Urology; Bladder cancer; Internal medicine; Cancer","score_opus":0.13018667208462448,"score_gpt":0.4399852707932906,"score_spread":0.3097985987086661,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2007231551","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":[0.9692832,0.005031181,0.009636593,0.0043695737,0.00088859553,0.008838762,0.00031665977,0.000023725264,0.0016116786],"genre_scores_gemma":[0.99731445,0.00012166131,0.00023665224,0.00035982465,0.0012888311,0.00035990032,0.00014953275,0.000080819525,0.000088320994],"study_design_codex":"randomized_trial","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9875697,0.007415586,0.0014823895,0.00071012013,0.002137575,0.0006846283],"domain_scores_gemma":[0.9820213,0.009868073,0.0013429027,0.0004435997,0.006093242,0.00023092135],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0061624013,0.0006889191,0.003847621,0.001427657,0.00023444105,0.000054686738,0.0003544196,0.000615013,0.000056304856],"category_scores_gemma":[0.00036736712,0.0005368596,0.00070309517,0.00085538597,0.0010850633,0.0003146515,0.00006520355,0.003072578,0.0000051540837],"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":[0.49669242,0.0070964675,0.011521636,0.07298248,0.029368006,0.00071065634,0.014505279,0.0018523771,0.21559185,0.00084389054,0.14433937,0.0044955816],"study_design_scores_gemma":[0.0348975,0.09962492,0.19898438,0.024743216,0.0028522813,0.00091704255,0.0012853763,0.0010168961,0.6182525,0.0027957547,0.013346149,0.0012839438],"about_ca_topic_score_codex":0.00011161213,"about_ca_topic_score_gemma":0.000025680567,"teacher_disagreement_score":0.4617949,"about_ca_system_score_codex":0.00060225173,"about_ca_system_score_gemma":0.0013667728,"threshold_uncertainty_score":0.9997083},"labels":[],"label_agreement":null},{"id":"W2028846339","doi":"10.2217/cer.12.63","title":"The utilization of telephone follow-up in the advanced cancer population: a review of the literature","year":2012,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Cancer survivorship and care","field":"Medicine","cited_by":36,"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; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; CINAHL; MEDLINE; Population; Cochrane Library; Cancer; Clinical trial; Palliative care; Randomized controlled trial; Patient satisfaction; Family medicine; Physical therapy; Surgery; Psychological intervention; Internal medicine; Nursing","score_opus":0.24741432605784686,"score_gpt":0.5369253003219292,"score_spread":0.28951097426408234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2028846339","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":[0.0024165192,0.99370676,0.000015421962,0.00019482276,0.000732917,0.0026520449,0.00002801617,0.0000014338407,0.00025207613],"genre_scores_gemma":[0.10049092,0.89884615,0.0000030922854,0.000018771443,0.00028317637,0.00023927607,0.000013805414,0.000018666173,0.00008613571],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9884152,0.008319799,0.0010930555,0.00017203517,0.0017088605,0.0002910652],"domain_scores_gemma":[0.9915456,0.004099402,0.0012469259,0.0005621098,0.002479714,0.00006623519],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008291189,0.00026330075,0.0021859794,0.0002820937,0.0001601945,0.000028518754,0.000685994,0.00015866329,0.00002965167],"category_scores_gemma":[0.00058866345,0.00010296777,0.0008163629,0.0031982965,0.00025148463,0.00014185463,0.00009744091,0.0018063646,0.0000010834583],"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":[0.0011909428,0.000251057,0.0017629811,0.20520654,0.0005863698,0.000010683197,0.0041135103,0.000008780537,0.000027137241,0.00047173165,0.0005714738,0.7857988],"study_design_scores_gemma":[0.00082943775,0.00037123356,0.017263127,0.4550399,0.00065136433,0.00009598175,0.00046050415,0.0000020167772,0.00010578781,0.00010025528,0.5249667,0.00011368889],"about_ca_topic_score_codex":0.0001039535,"about_ca_topic_score_gemma":0.00034679324,"teacher_disagreement_score":0.7856851,"about_ca_system_score_codex":0.00043036463,"about_ca_system_score_gemma":0.000966708,"threshold_uncertainty_score":0.78478557},"labels":[],"label_agreement":null},{"id":"W2030471933","doi":"10.2217/cer.12.72","title":"Adverse events across generations of bone-modifying agents in patients with solid tumor cancers reported in Phase III randomized trials","year":2012,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Bone health and treatments","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":"Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Adverse effect; Placebo; Nausea; Internal medicine; Vomiting; Adverse Event Reporting System; Oncology; Confounding; Bisphosphonate; Osteoporosis; Pathology; Alternative medicine","score_opus":0.4541017486537403,"score_gpt":0.6110403243727319,"score_spread":0.1569385757189916,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030471933","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":[0.54272735,0.44763085,0.00008157916,0.00000544738,0.00026151535,0.009169565,0.0000728081,0.0000036041722,0.00004725381],"genre_scores_gemma":[0.7196237,0.2790518,0.00013304301,0.0000046438413,0.000103020815,0.0008825962,0.00011926585,0.000048576978,0.000033353346],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9770971,0.015207042,0.0046243565,0.00042034127,0.0017906945,0.00086048344],"domain_scores_gemma":[0.986802,0.005938557,0.004440227,0.0004110928,0.001916943,0.0004911826],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.026133884,0.0005325156,0.012638395,0.001767673,0.00015514206,0.000014219691,0.00019227594,0.00023840497,0.00003988754],"category_scores_gemma":[0.0022473875,0.00032816594,0.0009754135,0.0016214259,0.00030891038,0.00032566092,0.00008110775,0.0017719668,0.000006429392],"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":[0.76846915,0.014536752,0.051767476,0.032360934,0.0090360455,0.0015361578,0.0034148393,0.0006413923,0.000015942709,0.000016542312,0.000061726074,0.11814305],"study_design_scores_gemma":[0.86001164,0.0039494094,0.041226123,0.0860502,0.0031670933,0.00018635258,0.00021104436,0.00015338074,0.00012055076,0.00003165005,0.0044027767,0.0004898114],"about_ca_topic_score_codex":0.0005503638,"about_ca_topic_score_gemma":0.00017037809,"teacher_disagreement_score":0.17689632,"about_ca_system_score_codex":0.0027941405,"about_ca_system_score_gemma":0.0042075654,"threshold_uncertainty_score":0.99991703},"labels":[],"label_agreement":null},{"id":"W2034390498","doi":"10.2217/cer.12.37","title":"Optimizing the design of pragmatic trials: key issues remain","year":2012,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":"McGill University","funders":"","keywords":"Medicine; Intervention (counseling); Context (archaeology); Task (project management); Risk analysis (engineering); Clinical trial; Clinical study design; Management science; Research design; Nursing","score_opus":0.9072114903384465,"score_gpt":0.670455447640037,"score_spread":0.23675604269840955,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2034390498","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":[0.00007203906,0.9636545,0.029825116,0.0007175035,0.00074050727,0.004304057,0.000081372775,0.000006810229,0.00059808366],"genre_scores_gemma":[0.0013844888,0.991458,0.005130788,0.000033512253,0.0013570103,0.0004079999,0.000009161043,0.00006934871,0.00014970696],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.90559673,0.080258645,0.012261017,0.00045235452,0.0006875534,0.0007436771],"domain_scores_gemma":[0.8669191,0.115307905,0.015707085,0.00076080294,0.0010495947,0.0002555362],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4304324,0.00047344054,0.011346611,0.0017471295,0.00036947505,0.00018425,0.0013717485,0.0003694924,0.00026302444],"category_scores_gemma":[0.028571716,0.00033730944,0.0011610792,0.0010330363,0.00044307756,0.00067573105,0.00018020053,0.0018288267,0.000555936],"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":[0.001978452,0.0036981788,0.00056438945,0.36109167,0.021571148,0.000047175097,0.06554909,0.014474753,0.00001433415,0.35624722,0.054182608,0.12058098],"study_design_scores_gemma":[0.00096729764,0.0006884734,0.000102606034,0.025201848,0.00030214875,0.000068584064,0.0015046166,0.00059826043,0.000010949005,0.010772823,0.959344,0.0004383484],"about_ca_topic_score_codex":0.000080726386,"about_ca_topic_score_gemma":0.0000044527146,"teacher_disagreement_score":0.90516144,"about_ca_system_score_codex":0.0015170509,"about_ca_system_score_gemma":0.0014637038,"threshold_uncertainty_score":0.9999079},"labels":[],"label_agreement":null},{"id":"W2049305235","doi":"10.2217/cer.12.71","title":"Use of relative effectiveness information in reimbursement and pricing decisions in Europe","year":2012,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Pharmaceutical Economics and Policy","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":"McGill University","funders":"","keywords":"Reimbursement; Process (computing); Medicine; Comparative effectiveness research; Actuarial science; Public economics; Economics; Health care; Computer science; Alternative medicine; Economic growth","score_opus":0.34571217471328664,"score_gpt":0.45467960943436125,"score_spread":0.10896743472107462,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2049305235","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":[0.9930772,0.0013200952,0.0028371976,0.00007786272,0.00014159991,0.00045949031,0.000017732798,0.000001415042,0.0020673587],"genre_scores_gemma":[0.99910265,0.00060342153,0.0002102283,0.000016766448,0.000035032517,0.000018732617,0.0000014144805,0.000007533279,0.0000042347906],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99750584,0.00096301327,0.0009451497,0.00012291354,0.00011057018,0.00035252268],"domain_scores_gemma":[0.9937418,0.005194146,0.00045024892,0.0001228106,0.00035079877,0.00014018772],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0091525,0.00011915842,0.0006887649,0.0014609034,0.000053425738,0.0000529855,0.0001521679,0.0000682036,0.000014755726],"category_scores_gemma":[0.0017043377,0.00011468702,0.00005804933,0.0010197281,0.00014646916,0.002253537,0.00012301421,0.00064742705,0.000027570688],"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":[0.0012264026,0.0004868354,0.8179194,0.00013006983,0.000108595865,0.0000051952907,0.0043438156,0.0064261425,0.0008851999,0.16584791,0.000043011947,0.0025774154],"study_design_scores_gemma":[0.0015028378,0.0003115293,0.9777347,0.00036334846,0.0000033017657,0.000007638392,0.00011903185,0.0028296693,0.0028005287,0.00988289,0.004327418,0.0001171509],"about_ca_topic_score_codex":0.00014337344,"about_ca_topic_score_gemma":0.000010810992,"teacher_disagreement_score":0.15981525,"about_ca_system_score_codex":0.00037658957,"about_ca_system_score_gemma":0.0000756811,"threshold_uncertainty_score":0.4676801},"labels":[],"label_agreement":null},{"id":"W2065741177","doi":"10.2217/cer.14.41","title":"Development, characteristics and validity of the EORTC QLQ-PR25 and the FACT-P for assessment of quality of life in prostate cancer patients","year":2014,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Prostate cancer; Quality of life (healthcare); Cancer; Medical physics; Oncology; Internal medicine; Nursing","score_opus":0.7458504868248589,"score_gpt":0.6286639508228942,"score_spread":0.11718653600196471,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2065741177","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":[0.648856,0.3446209,0.0005491084,0.00025988783,0.0002917415,0.0047015073,0.0006510434,0.0000010478664,0.00006877508],"genre_scores_gemma":[0.62746215,0.37167004,0.00029576613,0.000027956874,0.00008216631,0.00041010109,0.000012606431,0.000022579687,0.000016649841],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9832767,0.0092542255,0.006536801,0.00029515795,0.0003706865,0.0002664323],"domain_scores_gemma":[0.96766335,0.019892499,0.010963021,0.00029342037,0.001087319,0.000100413556],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07990823,0.00023332221,0.00525798,0.00047480734,0.00014251668,0.000034712415,0.00044220206,0.00013109462,0.000009082983],"category_scores_gemma":[0.0067766192,0.00015520344,0.00028253862,0.00032557684,0.00057739986,0.00015883743,0.00018463981,0.00061643426,8.502459e-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":[0.0008186509,0.000707769,0.8100866,0.14276116,0.0012660521,1.6411272e-7,0.0054699583,0.000057845053,0.0000019027552,0.025206245,0.00020614419,0.0134175],"study_design_scores_gemma":[0.0037890107,0.00032934223,0.9381313,0.011328703,0.00005416307,9.487499e-7,0.00029871947,0.00029054674,0.000010342623,0.0020315524,0.043510992,0.00022442934],"about_ca_topic_score_codex":0.000248882,"about_ca_topic_score_gemma":0.000088418055,"teacher_disagreement_score":0.13143244,"about_ca_system_score_codex":0.0005736941,"about_ca_system_score_gemma":0.0017226178,"threshold_uncertainty_score":0.9474281},"labels":[],"label_agreement":null},{"id":"W2090075299","doi":"10.2217/cer.13.91","title":"The vexing problem of defining the meaning, role and measurement of values in treatment decision-making","year":2014,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":"McMaster University","funders":"","keywords":"Meaning (existential); Value (mathematics); Medicine; Process (computing); Reductionism; Management science; Epistemology; Psychology; Computer science; Psychotherapist; Economics","score_opus":0.5259805117371518,"score_gpt":0.602337979086101,"score_spread":0.07635746734894922,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2090075299","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":[0.0025069641,0.99238104,0.00015986733,0.00005099908,0.00017196992,0.003575355,0.000008894257,0.0000037274892,0.0011412072],"genre_scores_gemma":[0.31468892,0.68408316,0.00057398767,0.0000017019651,0.00006630346,0.00055094087,0.0000012678994,0.00002834372,0.000005360666],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9559055,0.03888326,0.002573035,0.0002657315,0.0018401027,0.0005323778],"domain_scores_gemma":[0.9156359,0.07684189,0.003111833,0.00071902585,0.0035946413,0.000096705066],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.022566836,0.00031451788,0.0024640088,0.000665971,0.0012743025,0.000029216086,0.0010720659,0.00021595546,0.000005809618],"category_scores_gemma":[0.0030969342,0.00015375942,0.00030874053,0.000859371,0.0004590176,0.00011623267,0.00051733997,0.0024590262,0.00000613059],"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":[0.0007838108,0.00024023937,0.006536909,0.021084838,0.0005969455,0.000003495962,0.024834616,0.00023280847,0.000014290318,0.00600456,0.00004588728,0.9396216],"study_design_scores_gemma":[0.0027085748,0.0037012731,0.0104020815,0.6189552,0.00047003548,0.000038057715,0.033705376,0.0007461057,0.00006902812,0.02323811,0.30549416,0.0004720238],"about_ca_topic_score_codex":0.00019644568,"about_ca_topic_score_gemma":0.00026954632,"teacher_disagreement_score":0.93914956,"about_ca_system_score_codex":0.0018368,"about_ca_system_score_gemma":0.0028811507,"threshold_uncertainty_score":0.99984235},"labels":[],"label_agreement":null},{"id":"W2096357246","doi":"10.2217/cer.12.24","title":"Which design for which question? An exploration toward a translation table for comparative effectiveness research","year":2012,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Clinical practice guidelines implementation","field":"Medicine","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":"McMaster University","funders":"","keywords":"Comparative effectiveness research; Table (database); Medicine; Value (mathematics); Data science; Computer science; Alternative medicine; Machine learning; Data mining; Pathology","score_opus":0.9231174789930429,"score_gpt":0.7216988848994658,"score_spread":0.2014185940935771,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096357246","genre_codex":"methods","genre_gemma":"review","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":[0.0007224856,0.47454116,0.5000116,0.0002828016,0.0007121319,0.02342434,0.00014828764,0.000033803346,0.00012334992],"genre_scores_gemma":[0.29793546,0.6418023,0.035717037,0.000015956515,0.005467321,0.016663495,0.0019288745,0.00039197228,0.000077568446],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9496731,0.041519817,0.003486145,0.0010966704,0.0028303969,0.0013938382],"domain_scores_gemma":[0.85936934,0.10715839,0.0019056296,0.0006918621,0.03016289,0.0007119038],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1110199,0.0008366415,0.0055868244,0.002535039,0.0009066839,0.00036820487,0.00073913066,0.0008178211,0.000048557173],"category_scores_gemma":[0.011916678,0.0006783643,0.0008059888,0.0043341736,0.0003775741,0.003616905,0.00011926976,0.0038523003,0.000053065996],"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":[0.1012122,0.011884658,0.00032515582,0.15473709,0.008340933,0.000033624514,0.010132613,0.0051989956,0.0025522613,0.008107329,0.0035015915,0.69397354],"study_design_scores_gemma":[0.027904173,0.06334884,0.0017313063,0.074044,0.008291084,0.00053490914,0.0062715854,0.01274153,0.0050644646,0.018941866,0.7788159,0.0023103256],"about_ca_topic_score_codex":0.000083751074,"about_ca_topic_score_gemma":0.00013801116,"teacher_disagreement_score":0.77531433,"about_ca_system_score_codex":0.002445325,"about_ca_system_score_gemma":0.006437011,"threshold_uncertainty_score":0.99956673},"labels":[],"label_agreement":null},{"id":"W2105538774","doi":"10.2217/cer.12.79","title":"Comparing radial and femoral access for coronary angiography and interventions","year":2013,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Vascular Procedures and Complications","field":"Medicine","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":"Hamilton General Hospital; McMaster University","funders":"","keywords":"Medicine; Radial artery; Percutaneous coronary intervention; Cardiac catheterization; Femoral artery; Percutaneous; Bivalirudin; Vascular access; Hemostasis; Cardiology; Internal medicine; Surgery; Radiology; Artery; Myocardial infarction","score_opus":0.5502028699496919,"score_gpt":0.5854484250669251,"score_spread":0.03524555511723315,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2105538774","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":[0.009560546,0.98424625,0.002676247,0.000047068716,0.000099141485,0.0032039261,0.00001474296,0.000007398446,0.00014467684],"genre_scores_gemma":[0.060567245,0.9378701,0.0005828346,0.0000039951756,0.0002733368,0.00060298695,0.000027983539,0.000033963497,0.00003755976],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9971926,0.0011115533,0.00066291686,0.00030795092,0.0004185991,0.0003063873],"domain_scores_gemma":[0.9959014,0.0019535078,0.00040330304,0.00021494034,0.0012375873,0.000289282],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0018786851,0.0002747088,0.0022961036,0.0011707471,0.00029497006,0.00022188085,0.0002984397,0.0001442482,0.000015316535],"category_scores_gemma":[0.00013621004,0.00019681393,0.0013285776,0.0006208823,0.00036767396,0.00026381176,0.000250483,0.00091249606,0.0000036214024],"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":[0.0012237948,0.0013804289,0.013775265,0.14883302,0.0095490245,0.000059425696,0.00036734866,0.000013558307,0.000032367723,0.0037066315,0.004884155,0.816175],"study_design_scores_gemma":[0.0089143785,0.0077669052,0.31947666,0.15247785,0.0070666084,0.010268452,0.0004121177,0.0005093419,0.00002922142,0.004315777,0.48778266,0.0009800392],"about_ca_topic_score_codex":0.000012614606,"about_ca_topic_score_gemma":0.0000049606433,"teacher_disagreement_score":0.8151949,"about_ca_system_score_codex":0.0001274259,"about_ca_system_score_gemma":0.00036719325,"threshold_uncertainty_score":0.8025839},"labels":[],"label_agreement":null},{"id":"W2109513430","doi":"10.2217/cer.15.36","title":"Improving the evidence base for better comparative effectiveness research","year":2015,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":"McGill University; Royal Victoria Hospital","funders":"","keywords":"Comparative effectiveness research; Medicine; Evidence-based medicine; Management science; Outcomes research; Quality (philosophy); MEDLINE; Engineering ethics; Alternative medicine; Data science; Epistemology; Computer science; Pathology","score_opus":0.92690975542635,"score_gpt":0.6926157293442052,"score_spread":0.23429402608214478,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2109513430","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":[0.0034947435,0.97415996,0.007929014,0.0011180073,0.0013394005,0.011162053,0.00031064847,0.000019334637,0.00046686607],"genre_scores_gemma":[0.03934012,0.94349444,0.0019801517,0.0002414004,0.0057098684,0.008440066,0.00007426229,0.00032148857,0.0003981836],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9269541,0.061040003,0.007313122,0.0014336761,0.0016005172,0.0016585903],"domain_scores_gemma":[0.81577826,0.16982608,0.0060315877,0.001314971,0.0064492403,0.0005998972],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.39279127,0.00084098696,0.009646504,0.0030661009,0.0014585403,0.000737994,0.0028455942,0.0006316316,0.000087434426],"category_scores_gemma":[0.028210994,0.0006533988,0.0013684719,0.0024282455,0.0013251551,0.0015248596,0.0006103244,0.004840343,0.0011633431],"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":[0.008439777,0.0040644733,0.0031330003,0.43393818,0.01293929,0.00017533546,0.0304868,0.0021605492,0.000060583134,0.14178735,0.206825,0.15598966],"study_design_scores_gemma":[0.0022058678,0.0028488124,0.0007736021,0.046226855,0.00019041689,0.000107247935,0.0023195038,0.001256774,0.000049808343,0.020140428,0.92292106,0.00095963874],"about_ca_topic_score_codex":0.00036786695,"about_ca_topic_score_gemma":0.00004444319,"teacher_disagreement_score":0.71609604,"about_ca_system_score_codex":0.006924687,"about_ca_system_score_gemma":0.0053268094,"threshold_uncertainty_score":0.99984145},"labels":[],"label_agreement":null},{"id":"W2113687724","doi":"10.2217/cer.15.34","title":"Impact of paliperidone palmitate versus oral atypical antipsychotics on healthcare outcomes in schizophrenia patients","year":2015,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Schizophrenia research and treatment","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":"Group for Research in Decision Analysis","funders":"","keywords":"Paliperidone Palmitate; Medicine; Schizophrenia (object-oriented programming); Medicaid; Health care; Paliperidone; Formulary; Marginal structural model; Pharmacy; Health economics; Psychiatry; Emergency medicine; Antipsychotic; Internal medicine; Family medicine; Confounding; Public health","score_opus":0.2815064902716655,"score_gpt":0.5329407486024871,"score_spread":0.25143425833082156,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2113687724","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":[0.9980047,0.0003183467,0.00002385766,0.0003051324,0.00026543072,0.0009537876,0.000026828486,0.0000056902963,0.000096233525],"genre_scores_gemma":[0.99925196,0.00005686801,0.0005394092,0.0000040547334,0.00008170161,0.000023003895,0.000011963294,0.000021039355,0.000009969757],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9937563,0.0026375127,0.00064761424,0.00026326888,0.0021133393,0.0005819434],"domain_scores_gemma":[0.99322444,0.0024065687,0.00023368452,0.00031986786,0.0030615618,0.00075389765],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026312703,0.00025560166,0.0012928259,0.0010998977,0.000067125045,0.000030697815,0.0002408795,0.00011281762,0.000020187586],"category_scores_gemma":[0.0012618,0.00016102749,0.00035118026,0.0009896106,0.00030070677,0.00016521839,0.000075720876,0.0012672996,0.00003585202],"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":[0.44053403,0.0027279377,0.55295557,0.000059437152,0.00049562525,0.0001748437,0.00028466407,0.00016817174,0.00013765469,0.00027273578,0.00013798612,0.0020513406],"study_design_scores_gemma":[0.050900377,0.03903213,0.9070597,0.0004907861,0.000025082727,0.000016457974,0.00016862057,0.00012480436,0.0012558765,0.0008144965,0.000010084291,0.000101574595],"about_ca_topic_score_codex":0.00010015391,"about_ca_topic_score_gemma":0.00008364657,"teacher_disagreement_score":0.38963366,"about_ca_system_score_codex":0.0012884332,"about_ca_system_score_gemma":0.001637712,"threshold_uncertainty_score":0.656651},"labels":[],"label_agreement":null},{"id":"W2121793416","doi":"10.2217/cer.12.69","title":"Incidence of skeletal morbidity rates over time in patients with multiple myeloma-related bone disease as reported in randomized trials employing bone-modifying agents","year":2012,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Myeloma Research and Treatments","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":"University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Multiple myeloma; Incidence (geometry); Internal medicine; Bone disease; Randomized controlled trial; Surgery; Oncology; Osteoporosis","score_opus":0.24757855998991143,"score_gpt":0.5136934388489022,"score_spread":0.26611487885899077,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2121793416","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":[0.57914853,0.41361013,0.000015075016,0.0000054923025,0.00007628938,0.0070497864,0.000040200433,0.000007122093,0.00004738511],"genre_scores_gemma":[0.8013272,0.19772989,0.000084634026,0.0000013598138,0.00007993534,0.0004823038,0.00016555043,0.0000728433,0.0000562629],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9625345,0.027973164,0.0043074037,0.0006669163,0.0034672443,0.0010507941],"domain_scores_gemma":[0.9657369,0.0268634,0.0035998083,0.00064285443,0.0022832581,0.0008737285],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03271902,0.00078677526,0.012857976,0.0035197756,0.00014130169,0.00007155595,0.00038555247,0.00035988927,0.0002031236],"category_scores_gemma":[0.021682067,0.0004834324,0.001540294,0.0023074837,0.00071472843,0.0006224678,0.00027917768,0.002687591,0.00005089076],"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":[0.40242168,0.0077864686,0.48025966,0.022050016,0.007637517,0.0057617603,0.0007088016,0.00012854248,0.000112537826,0.000013356214,0.000034856588,0.0730848],"study_design_scores_gemma":[0.23138812,0.0016253982,0.6959299,0.06732467,0.0018810779,0.000121707555,0.000028798248,0.00047560394,0.00012934109,0.000089397574,0.0005329582,0.00047307153],"about_ca_topic_score_codex":0.0004061509,"about_ca_topic_score_gemma":0.000029601862,"teacher_disagreement_score":0.22217871,"about_ca_system_score_codex":0.0016383464,"about_ca_system_score_gemma":0.0026433717,"threshold_uncertainty_score":0.99976176},"labels":[],"label_agreement":null},{"id":"W2123800537","doi":"10.2217/cer.14.5","title":"Accelerated versus standard care in hip fracture patients: does speed save lives?","year":2014,"lang":"en","type":"editorial","venue":"Journal of Comparative Effectiveness Research","topic":"Hip and Femur Fractures","field":"Medicine","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":"McMaster University; Population Health Research Institute","funders":"","keywords":"Medicine; MEDLINE; Hip fracture; Biostatistics; Pulmonary embolism; Deep vein; Epidemiology; Surgery; Internal medicine; Thrombosis; Osteoporosis","score_opus":0.06034044854847201,"score_gpt":0.45411528233701376,"score_spread":0.3937748337885417,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2123800537","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":[0.53714645,0.0034739887,0.00013087175,0.00012297808,0.45305875,0.002688147,0.00023146898,0.000027158118,0.0031202054],"genre_scores_gemma":[0.76087785,0.00024305237,0.0000439103,0.000012928711,0.23838745,0.000024298311,0.00025781733,0.000069409034,0.000083283834],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98808676,0.004834934,0.0010567656,0.0006118012,0.0045802486,0.0008295051],"domain_scores_gemma":[0.9714404,0.014198686,0.00082307874,0.0005030681,0.012555704,0.0004790636],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0038963482,0.00064968946,0.0028063862,0.0017645059,0.00022573711,0.00020849679,0.0006468931,0.0011190894,0.00023758088],"category_scores_gemma":[0.0054903915,0.0004083151,0.0004506434,0.0009458644,0.00042039293,0.0003142801,0.00016232418,0.008585302,0.00004361403],"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":[0.23249689,0.0011035526,0.011976148,0.00272898,0.0015741254,0.0008814576,0.010338085,0.000509364,0.00041194088,0.000010917505,0.7352658,0.0027027573],"study_design_scores_gemma":[0.036566526,0.027012888,0.057802875,0.006591117,0.00030764364,0.000007824385,0.0033053635,0.000030056435,0.010090954,0.00019752252,0.8573769,0.0007103571],"about_ca_topic_score_codex":0.00009535535,"about_ca_topic_score_gemma":0.0001363764,"teacher_disagreement_score":0.22373143,"about_ca_system_score_codex":0.0026566202,"about_ca_system_score_gemma":0.002796089,"threshold_uncertainty_score":0.99983686},"labels":[],"label_agreement":null},{"id":"W2143296285","doi":"10.2217/cer.11.6","title":"Clinical practice change requires more than comparative effectiveness evidence: abdominal aortic aneurysm management in the USA","year":2012,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Aortic aneurysm repair treatments","field":"Medicine","cited_by":14,"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; Reimbursement; Abdominal aortic aneurysm; Endovascular aneurysm repair; Comparative effectiveness research; Aortic aneurysm; Abdominal surgery; Clinical trial; Randomized controlled trial; Surgery; General surgery; Health care; Aneurysm; Internal medicine","score_opus":0.432308873191544,"score_gpt":0.5887352008087636,"score_spread":0.15642632761721964,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2143296285","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":[0.9874753,0.004314841,0.0006352029,0.0006527955,0.0004946697,0.0043702815,0.0000030266242,0.000014760844,0.002039163],"genre_scores_gemma":[0.9967782,0.0004998411,0.0010607471,0.00014342144,0.0008242772,0.000601508,0.0000034645116,0.000032128035,0.000056400455],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9739755,0.019987581,0.0013848796,0.0005302724,0.0030440942,0.0010777004],"domain_scores_gemma":[0.96919185,0.027148439,0.00077597226,0.00069999596,0.0017137239,0.00046999636],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.029569274,0.000467246,0.0017087854,0.0007767773,0.0003298193,0.000097565804,0.00069077616,0.00019138619,0.000047508333],"category_scores_gemma":[0.0021034856,0.0002902666,0.0004590323,0.0016078788,0.0010483599,0.0018221656,0.00027415226,0.0023156337,0.00014293235],"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":[0.025727639,0.0066142003,0.94234735,0.00084349734,0.0014769359,0.0043662507,0.012121647,0.00005035575,0.0007047864,0.00062410545,0.0004097663,0.0047134436],"study_design_scores_gemma":[0.004360094,0.005220209,0.97525674,0.0023475895,0.00055146625,0.0027462468,0.007162066,0.00029772098,0.0011539267,0.00012553518,0.00054362835,0.00023480166],"about_ca_topic_score_codex":0.0001422802,"about_ca_topic_score_gemma":0.000016709042,"teacher_disagreement_score":0.032909345,"about_ca_system_score_codex":0.00088869256,"about_ca_system_score_gemma":0.0003408382,"threshold_uncertainty_score":0.99998605},"labels":[],"label_agreement":null},{"id":"W2289680983","doi":"10.2217/cer.15.66","title":"Better research reporting to improve the utility of routine data for making better treatment decisions","year":2016,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":"Institute for Clinical Evaluative Sciences; Children's Hospital of Eastern Ontario; University of Ottawa","funders":"National Institute for Health and Care Research","keywords":"Medicine; Checklist; Comparative effectiveness research; Observational study; Transparency (behavior); Electronic health record; Family medicine; Data collection; Alternative medicine; Medical emergency; Health care; Computer science; Pathology","score_opus":0.9333516952933743,"score_gpt":0.7129547362859485,"score_spread":0.22039695900742573,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2289680983","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":[0.0055184625,0.95377475,0.021385962,0.006094093,0.0009369875,0.009253262,0.0022209184,0.000008116341,0.00080745603],"genre_scores_gemma":[0.09724722,0.88366646,0.007949241,0.00037344056,0.006236012,0.0035177735,0.00016175697,0.0002740176,0.0005740651],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9692037,0.014272007,0.013322355,0.0012109217,0.0008981848,0.0010928141],"domain_scores_gemma":[0.8779338,0.10235457,0.013941023,0.0027580594,0.0026998387,0.00031271434],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27073362,0.00043506254,0.006150951,0.0018604038,0.0007156605,0.00021545314,0.0023916797,0.00030412833,0.000154168],"category_scores_gemma":[0.061002903,0.00028110767,0.00081924634,0.0010241837,0.0004343671,0.00053095486,0.0008775965,0.0013283382,0.000269528],"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":[0.0012717284,0.001362574,0.0051598162,0.026047064,0.004868482,0.000036770103,0.0038370125,0.000027906857,0.000012527453,0.008397533,0.06978233,0.8791962],"study_design_scores_gemma":[0.0009215522,0.0011324052,0.0020808098,0.01891719,0.00006899669,0.000031863823,0.00038823762,0.00035299937,0.000017410239,0.008967546,0.96682554,0.00029545545],"about_ca_topic_score_codex":0.00014314361,"about_ca_topic_score_gemma":0.000049533428,"teacher_disagreement_score":0.8970432,"about_ca_system_score_codex":0.0022587706,"about_ca_system_score_gemma":0.002039473,"threshold_uncertainty_score":0.9999641},"labels":[],"label_agreement":null},{"id":"W2424016786","doi":"10.2217/cer-2015-0016","title":"Are drug-coated balloons cost effective for femoropopliteal occlusive disease? A comparison of bare metal stents and uncoated balloons","year":2016,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Peripheral Artery Disease Management","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":true,"ca_institutions":"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; Centre Hospitalier Universitaire de Sherbrooke","funders":"National Institutes of Health; National Institute for Health and Care Research","keywords":"Medicine; Bare metal; Balloon; Surgery; Stent; Restenosis","score_opus":0.12081910414501004,"score_gpt":0.46378408107019786,"score_spread":0.3429649769251878,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2424016786","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":[0.98610204,0.0012525746,0.005316271,0.0006633418,0.00017102066,0.0060394243,0.00025850633,0.000020726084,0.00017609009],"genre_scores_gemma":[0.99894047,0.000036665766,0.00012995719,0.000030457664,0.00011288826,0.00053222367,0.000015668653,0.000038242884,0.00016341548],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9941364,0.002533788,0.00081268424,0.00049387844,0.0013930287,0.0006302162],"domain_scores_gemma":[0.99077666,0.0038749678,0.00057109224,0.00034513045,0.0036505826,0.0007815527],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0023357507,0.00037834066,0.0016859712,0.0007202553,0.00028502138,0.000053070817,0.00031878625,0.00008224711,0.000071060975],"category_scores_gemma":[0.0009021322,0.00024129331,0.00040652265,0.0006866322,0.0007488713,0.0003715709,0.00022379594,0.00048666855,0.000011808465],"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":[0.059083853,0.004968025,0.7223419,0.0027630103,0.004717225,0.00046012618,0.0022712708,0.00028851518,0.19011919,0.000547647,0.0035600339,0.008879206],"study_design_scores_gemma":[0.0128871715,0.0027969487,0.916238,0.0031754787,0.00060483714,0.000028585002,0.0023000508,0.0011479472,0.059126284,0.00033157642,0.0010832394,0.00027986415],"about_ca_topic_score_codex":0.000017825763,"about_ca_topic_score_gemma":0.00003516897,"teacher_disagreement_score":0.19389611,"about_ca_system_score_codex":0.0007107572,"about_ca_system_score_gemma":0.00035652483,"threshold_uncertainty_score":0.9839656},"labels":[],"label_agreement":null},{"id":"W2580600991","doi":"10.2217/cer-2016-0093","title":"The comparative effectiveness of antiretroviral therapies for HIV: evidence to inform precision public health","year":2017,"lang":"en","type":"editorial","venue":"Journal of Comparative Effectiveness Research","topic":"HIV/AIDS Research and Interventions","field":"Medicine","cited_by":2,"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 British Columbia","funders":"World Health Organization","keywords":"Medicine; Comparative effectiveness research; Human immunodeficiency virus (HIV); Public health; Antiretroviral therapy; Family medicine; Intensive care medicine; Virology; Alternative medicine; Viral load; Pathology","score_opus":0.2714059329206396,"score_gpt":0.5631588489108037,"score_spread":0.2917529159901641,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2580600991","genre_codex":"editorial","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":[0.35067928,0.09072442,0.09821001,0.008936796,0.38275486,0.05866516,0.0026869567,0.00011586208,0.0072266473],"genre_scores_gemma":[0.9119485,0.0024218585,0.00043748293,0.000005825395,0.08076396,0.0010515673,0.00011445651,0.00008770259,0.0031686253],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97966236,0.011260047,0.0017176273,0.0006243612,0.00536964,0.0013659676],"domain_scores_gemma":[0.89734036,0.0779933,0.0018408427,0.001163402,0.020764142,0.0008979498],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04931839,0.0005924033,0.0034454498,0.0017795074,0.0020489353,0.0006620875,0.002162606,0.0004590709,0.000028576267],"category_scores_gemma":[0.029422466,0.00036417914,0.0012165206,0.0011737064,0.0015894327,0.0010629186,0.00054738915,0.00380726,0.000044784152],"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":[0.040282227,0.001056242,0.0011986657,0.005559762,0.002529641,0.000021468333,0.0034281262,0.000031485746,0.0037542686,0.00061419903,0.9215124,0.020011516],"study_design_scores_gemma":[0.0072412924,0.11064707,0.025957072,0.038657162,0.00012379844,0.000032651944,0.0019644392,0.00036829404,0.011538773,0.0008357618,0.8021463,0.00048736052],"about_ca_topic_score_codex":0.00013613341,"about_ca_topic_score_gemma":0.000124246,"teacher_disagreement_score":0.5612693,"about_ca_system_score_codex":0.0018709075,"about_ca_system_score_gemma":0.009022252,"threshold_uncertainty_score":0.999881},"labels":[],"label_agreement":null},{"id":"W2747351937","doi":"10.2217/cer-2017-0043","title":"Real-world adherence and economic outcomes associated with paliperidone palmitate versus oral atypical antipsychotics in schizophrenia patients with substance-related disorders using Medicaid benefits","year":2017,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Schizophrenia research and treatment","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":false,"ca_institutions":"Group for Research in Decision Analysis","funders":"","keywords":"Paliperidone Palmitate; Medicine; Antipsychotic; Schizophrenia (object-oriented programming); Medicaid; Paliperidone; Risperidone; Psychiatry; Internal medicine; Health care","score_opus":0.1177252280232934,"score_gpt":0.4255221851048258,"score_spread":0.30779695708153243,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2747351937","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":[0.99828666,0.00020612605,0.000016877102,0.0001974601,0.00015242562,0.0009788814,0.000020187577,0.000008340201,0.00013305034],"genre_scores_gemma":[0.9990093,0.00026628343,0.0006110135,0.0000015126489,0.000029409715,0.000023988217,0.00001160025,0.000029661078,0.00001720566],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964036,0.0009870129,0.00048394583,0.00037851467,0.001145943,0.00060100225],"domain_scores_gemma":[0.9960036,0.0019607528,0.00045984413,0.00036908535,0.000773145,0.00043356646],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016736982,0.0002992666,0.0011004332,0.00062064873,0.00043638473,0.0001349022,0.00029452666,0.000099533034,0.00002873343],"category_scores_gemma":[0.00036186058,0.0001888896,0.0000906466,0.00038232873,0.000989665,0.0004376075,0.000074811505,0.0011029246,0.0000061381297],"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":[0.2198215,0.0006522058,0.77585196,0.000031083568,0.0009166228,0.00013667293,0.00018922811,0.0002710166,0.000029981476,0.00031790714,0.0000064615674,0.0017753313],"study_design_scores_gemma":[0.0585917,0.005519805,0.93268555,0.001403433,0.00009759178,0.000009567,0.00010029229,0.0008486935,0.00018721522,0.00037500143,0.0000028092538,0.00017831178],"about_ca_topic_score_codex":0.00026569678,"about_ca_topic_score_gemma":0.012249261,"teacher_disagreement_score":0.1612298,"about_ca_system_score_codex":0.00079253,"about_ca_system_score_gemma":0.000740141,"threshold_uncertainty_score":0.7702694},"labels":[],"label_agreement":null},{"id":"W2770256141","doi":"10.2217/cer-2017-0011","title":"Effectiveness and safety of anti-ischemic trimetazidine in patients with stable angina pectoris and Type 2 diabetes","year":2017,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Cardiovascular Function and Risk Factors","field":"Medicine","cited_by":20,"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":"Trimetazidine; Medicine; Tolerability; Internal medicine; Cardiology; Angina; Diabetes mellitus; Type 2 diabetes; Coronary artery disease; Ischemia; Adverse effect; Unstable angina; Myocardial infarction; Endocrinology","score_opus":0.03753972083842771,"score_gpt":0.3606793636673458,"score_spread":0.3231396428289181,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2770256141","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":[0.9974182,0.0014625718,0.00008831695,0.000017273582,0.00012314894,0.0006845324,0.0000053832473,0.0000029671514,0.0001975607],"genre_scores_gemma":[0.99962324,0.00025362193,0.000044471526,8.557028e-7,0.00004347899,0.0000059176114,0.0000025351123,0.000013825929,0.000012051566],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9976012,0.0010060075,0.00026851115,0.00020729621,0.00067099585,0.00024595895],"domain_scores_gemma":[0.9967945,0.0010841108,0.00022808436,0.0002681284,0.0014793953,0.00014577115],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0033863795,0.00015557201,0.0010431152,0.0004577249,0.00019258361,0.000045783887,0.00010523528,0.00007369739,0.000010731009],"category_scores_gemma":[0.00049647625,0.00010134487,0.000099194825,0.00032478938,0.00043552113,0.00028617226,0.000088483815,0.0005418816,9.454352e-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":[0.009012812,0.00028270684,0.96876055,0.00049521006,0.00059012126,0.000015869437,0.00015837728,0.00003632987,0.019585852,0.000019083687,0.000021142905,0.0010219569],"study_design_scores_gemma":[0.008692688,0.0023499464,0.91849625,0.0009812339,0.00009164023,0.000011938623,0.00007976197,0.000028902934,0.06895054,0.000016308219,0.00020906863,0.000091713504],"about_ca_topic_score_codex":0.00005047847,"about_ca_topic_score_gemma":0.0000033998786,"teacher_disagreement_score":0.050264284,"about_ca_system_score_codex":0.00013851271,"about_ca_system_score_gemma":0.00018529434,"threshold_uncertainty_score":0.41327238},"labels":[],"label_agreement":null},{"id":"W2789990939","doi":"10.2217/cer-2017-0023","title":"Number needed to treat in indirect treatment comparison","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Thyroid Cancer Diagnosis and Treatment","field":"Medicine","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":"Ottawa Hospital","funders":"","keywords":"Medicine; Lenvatinib; Number needed to treat; Sorafenib; Oncology; Internal medicine; Odds ratio; Relative risk; Confidence interval","score_opus":0.2184234576921588,"score_gpt":0.5380514190024712,"score_spread":0.31962796131031246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2789990939","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":[0.9902861,0.0006634861,0.00010287851,0.0006639881,0.00016591331,0.001090034,0.0000045028023,0.000008665966,0.0070144176],"genre_scores_gemma":[0.99877906,0.000087126915,0.00031463063,0.000038622933,0.00041526862,0.00015616807,0.000002307625,0.000018835986,0.00018795893],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9967514,0.0012341783,0.0004911964,0.00027304143,0.0007921548,0.0004580257],"domain_scores_gemma":[0.9967669,0.0013718014,0.00013510413,0.00024398263,0.0011205586,0.00036162706],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014235162,0.00022205187,0.0011501381,0.0007046096,0.00012377041,0.00004862835,0.00014300985,0.00007505004,0.00021039176],"category_scores_gemma":[0.0001093583,0.00014844195,0.00017807729,0.0011354485,0.00019656282,0.00012516209,0.000048163063,0.00038138017,0.0002922371],"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":[0.02212224,0.007447579,0.90483606,0.0001177726,0.0007239334,0.0005698076,0.012829717,0.00021802852,0.022041209,0.00026201803,0.00812351,0.02070814],"study_design_scores_gemma":[0.006981063,0.014373157,0.62138957,0.0006187229,0.00006238456,0.00009897616,0.0011806231,0.000053443946,0.35004413,0.00012223134,0.004960617,0.00011506183],"about_ca_topic_score_codex":0.00045161945,"about_ca_topic_score_gemma":0.00028688085,"teacher_disagreement_score":0.32800293,"about_ca_system_score_codex":0.002150578,"about_ca_system_score_gemma":0.00044028283,"threshold_uncertainty_score":0.6053287},"labels":[],"label_agreement":null},{"id":"W2794986206","doi":"10.2217/cer-2017-0092","title":"Analysis of indirect treatment comparisons in national health technology assessments and requirements for industry submissions","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Scope (computer science); Health technology; Medicine; Comparative effectiveness research; Alternative medicine; Health care; Computer science; Political science","score_opus":0.8138658705718702,"score_gpt":0.6792615665071889,"score_spread":0.13460430406468127,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2794986206","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":[0.9888721,0.0011306173,0.005146954,0.0031331615,0.0001260149,0.0009456281,0.00015497894,0.000003824875,0.00048668694],"genre_scores_gemma":[0.9979698,0.00006371577,0.0016484229,0.00006695991,0.00008445891,0.000111633686,0.000010324954,0.000008546357,0.000036134257],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99474597,0.0017869305,0.00254509,0.0003012975,0.0002744222,0.00034631722],"domain_scores_gemma":[0.9925666,0.004056441,0.002101465,0.0001705457,0.0009492891,0.00015567151],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03192726,0.0001291002,0.0017554223,0.0040517133,0.00022900186,0.000032698885,0.00021724547,0.00016873793,0.000051418352],"category_scores_gemma":[0.001742424,0.00013090885,0.00012345426,0.0017891101,0.00030311567,0.0002696611,0.000051910007,0.00038917846,0.000008484097],"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":[0.0002866371,0.0005316396,0.977611,0.00013424452,0.0013536861,5.839131e-7,0.0015367749,0.0004565125,0.00006318513,0.016785538,0.0009914761,0.00024870984],"study_design_scores_gemma":[0.002598398,0.0029064494,0.9694702,0.00029504515,0.000026691732,0.0000032683085,0.0024872692,0.009374471,0.0004956692,0.009002526,0.003206567,0.00013345298],"about_ca_topic_score_codex":0.00023727107,"about_ca_topic_score_gemma":0.00035390383,"teacher_disagreement_score":0.030184835,"about_ca_system_score_codex":0.001928642,"about_ca_system_score_gemma":0.0009854165,"threshold_uncertainty_score":0.9968346},"labels":[],"label_agreement":null},{"id":"W2803925386","doi":"10.2217/cer-2018-0015","title":"Effects of calcimimetics on long-term outcomes in dialysis patients: literature review and Bayesian meta-analysis","year":2018,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Parathyroid Disorders and Treatments","field":"Medicine","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":"Dalhousie University","funders":"","keywords":"Medicine; Cinacalcet; Observational study; Randomized controlled trial; Discontinuation; Hazard ratio; Meta-analysis; Internal medicine; MEDLINE; Dialysis; Intensive care medicine; Relative risk; Parathyroidectomy; Secondary hyperparathyroidism; Confidence interval; Parathyroid hormone","score_opus":0.1963978398121837,"score_gpt":0.5228953080177915,"score_spread":0.32649746820560777,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2803925386","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":[0.006757737,0.9897135,0.00007799054,0.000025970927,0.000048861777,0.0032304055,0.00005729586,0.0000028279524,0.00008540312],"genre_scores_gemma":[0.026723968,0.97276574,0.000046033885,0.000018508714,0.000036929512,0.00022738788,0.00006183636,0.00003410372,0.000085465676],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9893326,0.006825604,0.0014966863,0.0005363982,0.0013939667,0.00041475525],"domain_scores_gemma":[0.98959184,0.0064131855,0.0012114536,0.00052628684,0.0019724783,0.00028474338],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0029028486,0.00066534773,0.0134741785,0.0031082723,0.00008205698,0.000054742137,0.00032972585,0.00029394825,0.00001943101],"category_scores_gemma":[0.0008934151,0.00036396403,0.00384639,0.003918539,0.00031080542,0.00014291944,0.00012315575,0.0013674721,0.000008744639],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012516931,0.010376591,0.05907398,0.16827333,0.74211806,0.0013002465,0.0005707949,0.0000050660356,0.0000027084827,0.000053499636,0.00075024326,0.016223758],"study_design_scores_gemma":[0.0049369656,0.0069943764,0.14286286,0.041669667,0.7992002,0.000047884554,0.0000069900657,0.0000034938782,0.00009950428,0.0000611421,0.0036108692,0.00050604594],"about_ca_topic_score_codex":0.000008078222,"about_ca_topic_score_gemma":0.000013443106,"teacher_disagreement_score":0.12660366,"about_ca_system_score_codex":0.00034258736,"about_ca_system_score_gemma":0.0003123219,"threshold_uncertainty_score":0.9998812},"labels":[],"label_agreement":null},{"id":"W2806016224","doi":"10.2217/cer-2017-0029","title":"Comparison of EQ-5D and SF-36 in untreated patients with symptoms of intermittent claudication","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Peripheral Artery Disease Management","field":"Medicine","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":"GlaxoSmithKline (Canada); University of Alberta","funders":"Universiteit Maastricht","keywords":"Medicine; EQ-5D; Valuation (finance); Intermittent claudication; Quality of life (healthcare); Physical therapy; Tariff; Quality-adjusted life year; Health related quality of life; Arterial disease; Disease; Finance; Surgery; Internal medicine; Cost effectiveness","score_opus":0.1271241282527948,"score_gpt":0.43425018516687686,"score_spread":0.30712605691408207,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2806016224","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":[0.9979191,0.00018198554,0.0004756876,0.000062690684,0.000034146473,0.00073443673,0.0000038209923,0.0000027290916,0.0005854429],"genre_scores_gemma":[0.99980575,0.000007927373,0.000104918916,0.000005930999,0.000033981378,0.000015172371,0.000004913555,0.000008574144,0.00001285849],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974332,0.0008667324,0.00052590156,0.0001607518,0.0008199134,0.0001935365],"domain_scores_gemma":[0.99698395,0.00040720464,0.0003650672,0.00016737403,0.0019532198,0.00012317624],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010416141,0.00012054711,0.0007298394,0.00059238164,0.000041616306,0.000011854933,0.00012798271,0.000038412014,0.00003741182],"category_scores_gemma":[0.000091174,0.000081883,0.000051787774,0.0005787843,0.0005731136,0.00015023327,0.00008227893,0.0002926683,0.0000032227604],"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":[0.011472231,0.0017277487,0.97109586,0.00034610723,0.00028940043,0.000008798715,0.0015142732,0.000057722056,0.007022046,0.000084079315,0.00008651989,0.006295219],"study_design_scores_gemma":[0.0048367465,0.009989363,0.9618424,0.0015161451,0.000054756656,0.0000036762422,0.0004304429,0.0010093555,0.020111384,0.000081708204,0.00006152434,0.00006247162],"about_ca_topic_score_codex":0.000029986013,"about_ca_topic_score_gemma":0.000020875712,"teacher_disagreement_score":0.013089337,"about_ca_system_score_codex":0.0002174058,"about_ca_system_score_gemma":0.00012169373,"threshold_uncertainty_score":0.33390918},"labels":[],"label_agreement":null},{"id":"W2808685910","doi":"10.2217/cer-2017-0090","title":"Examples of dramatic failures and their effectiveness in modern surgical disciplines: can we learn from our mistakes?","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Cardiac, Anesthesia and Surgical Outcomes","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":"University of Calgary","funders":"","keywords":"Subspecialty; Skepticism; Medicine; Craft; Engineering ethics; Health care; Curiosity; Epistemology; Psychology; Law; Engineering; Pathology; Social psychology","score_opus":0.11592673528596632,"score_gpt":0.4303357293267681,"score_spread":0.3144089940408018,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808685910","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":[0.9947305,0.0023309423,0.0007365283,0.0007222685,0.00009939995,0.0007418922,0.000014031742,0.000008224845,0.0006162564],"genre_scores_gemma":[0.9993184,0.000118168056,0.00011618109,0.000004712617,0.00032558068,0.000028623323,0.000004759311,0.000024603989,0.000058957576],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99340606,0.004151602,0.0006938965,0.00034625427,0.00095416466,0.00044802084],"domain_scores_gemma":[0.9842873,0.01370389,0.00024956884,0.0002820518,0.0011800522,0.00029714438],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005142687,0.00029504986,0.0022899997,0.0006389524,0.00013409418,0.00006274694,0.00023325905,0.0001567423,0.000020697802],"category_scores_gemma":[0.0006362733,0.00018440028,0.0003933166,0.000685352,0.0006570877,0.00016690351,0.00008738426,0.000851433,0.000004314648],"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":[0.037543308,0.0021949008,0.8026959,0.002366179,0.001880657,0.0015753799,0.027276019,0.00045786833,0.09977922,0.0030797573,0.000044934884,0.021105919],"study_design_scores_gemma":[0.0075409357,0.0026805485,0.92011076,0.0031341552,0.00008087998,0.00035699375,0.010846188,0.0013493204,0.049099404,0.0033243801,0.001222143,0.00025429708],"about_ca_topic_score_codex":0.00039680218,"about_ca_topic_score_gemma":0.00018262045,"teacher_disagreement_score":0.1174149,"about_ca_system_score_codex":0.00021167532,"about_ca_system_score_gemma":0.00036095566,"threshold_uncertainty_score":0.75196254},"labels":[],"label_agreement":null},{"id":"W2889227700","doi":"10.2217/cer-2018-0020","title":"Relative effectiveness of sunitinib versus everolimus in advanced pancreatic neuroendocrine tumors: an updated matching-adjusted indirect comparison","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Neuroendocrine Tumor Research Advances","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":"EVRAZ (Canada)","funders":"","keywords":"Sunitinib; Everolimus; Medicine; Hazard ratio; Internal medicine; Oncology; Neuroendocrine tumors; Urology; Confidence interval; Cancer","score_opus":0.1270138429126655,"score_gpt":0.4709140956346132,"score_spread":0.3439002527219477,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889227700","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":[0.99450487,0.00044297814,0.00027507832,0.000065778986,0.00034027614,0.0025107253,0.000019596977,0.000040499846,0.0018001712],"genre_scores_gemma":[0.9989374,0.000049416394,0.00055145944,0.0000070283563,0.00019269016,0.00010476328,0.00003378766,0.000077699464,0.00004570384],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.97934294,0.01496242,0.0014354853,0.0006962858,0.0023771694,0.0011857309],"domain_scores_gemma":[0.98233336,0.011099355,0.0008040787,0.0006685102,0.004499515,0.0005951827],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.005218398,0.00052493147,0.0028264066,0.0025409814,0.00023855512,0.000061011597,0.0007323481,0.000041691535,0.00011902036],"category_scores_gemma":[0.0023515357,0.00043945422,0.00027149948,0.0035113725,0.001423987,0.0013363843,0.00028739154,0.0028032716,0.000033886754],"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":[0.37330058,0.00642149,0.17779754,0.0022305239,0.0013726526,0.010016711,0.00300981,0.0032149572,0.4188482,0.0013914987,0.00010410779,0.0022919362],"study_design_scores_gemma":[0.018029938,0.031585887,0.5350451,0.0025865864,0.00009545232,0.0010091598,0.001487073,0.00066983124,0.40738714,0.0017107861,0.00011354121,0.00027954602],"about_ca_topic_score_codex":0.00020927176,"about_ca_topic_score_gemma":0.00017211691,"teacher_disagreement_score":0.35724753,"about_ca_system_score_codex":0.00078829174,"about_ca_system_score_gemma":0.00093149656,"threshold_uncertainty_score":0.99980575},"labels":[],"label_agreement":null},{"id":"W2889900146","doi":"10.2217/cer-2018-0035","title":"Some issues for the evaluation of noninferiority trials","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","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":"McGill University Health Centre","funders":"H2020 European Research Council","keywords":"Medicine; Confidence interval; Margin (machine learning); Adjuvant therapy; Placebo; Medical physics; Cancer; Internal medicine; Alternative medicine","score_opus":0.9613010515836794,"score_gpt":0.7999582645757043,"score_spread":0.1613427870079751,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889900146","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":[0.6369655,0.0023079114,0.35168877,0.0006761304,0.0021948684,0.0055947923,0.00006218144,0.000009866429,0.0004999484],"genre_scores_gemma":[0.913784,0.0000696138,0.08309387,0.0000054161455,0.002853018,0.00015396312,1.878253e-7,0.000015596854,0.000024308607],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.95042163,0.044645898,0.0017966288,0.00020562926,0.002614726,0.0003154732],"domain_scores_gemma":[0.4651242,0.5198521,0.0012510961,0.00037929122,0.013294192,0.00009913244],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28422365,0.0001452267,0.0017539529,0.00023102546,0.0002521859,0.000057792975,0.00055489683,0.000116571806,0.0001806135],"category_scores_gemma":[0.40860817,0.00007936214,0.00039930578,0.00045035663,0.0010739013,0.0001726127,0.00010033045,0.0005589482,0.00000883433],"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":[0.031061139,0.002489796,0.0014355235,0.0013229197,0.0042946874,0.0000055196906,0.0044987164,0.0003143568,0.058883928,0.77388406,0.014801952,0.107007384],"study_design_scores_gemma":[0.0031536804,0.0026162337,0.00962351,0.00032686646,0.0003226359,0.0000033469764,0.00026501282,0.0029278693,0.11356974,0.866661,0.00045659475,0.00007349693],"about_ca_topic_score_codex":0.000009569658,"about_ca_topic_score_gemma":0.0000039743913,"teacher_disagreement_score":0.48529744,"about_ca_system_score_codex":0.00018761348,"about_ca_system_score_gemma":0.000588053,"threshold_uncertainty_score":0.73704237},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2892182512","doi":"10.2217/cer-2018-0006","title":"M-bioscore: proposing a new statistical model for prognostic factors in metastatic breast cancer patients","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Breast Cancer Treatment Studies","field":"Biochemistry, Genetics and Molecular Biology","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 Calgary","funders":"","keywords":"Medicine; Proportional hazards model; Breast cancer; Oncology; Internal medicine; Hazard ratio; Cohort; Multivariate analysis; Cancer; Multivariate statistics; Metastatic breast cancer; Survival analysis; Confidence interval; Statistics","score_opus":0.1123983302583288,"score_gpt":0.46056779330651276,"score_spread":0.34816946304818397,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2892182512","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":[0.9623511,0.00030027874,0.036099233,0.00007623322,0.000121077675,0.0008683254,0.00015811175,0.0000020576656,0.000023572067],"genre_scores_gemma":[0.99808246,0.000024048712,0.0015187891,0.0000077626555,0.00020634731,0.00010125259,0.000019876237,0.000019380996,0.000020061527],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9980561,0.0005436051,0.0003143503,0.00026778431,0.00042957775,0.0003885465],"domain_scores_gemma":[0.99745166,0.00057511625,0.00015168871,0.00010934779,0.0015737811,0.0001383946],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0006782897,0.00018281222,0.00042585225,0.00019185581,0.00014733186,0.00003714348,0.0001848064,0.000049577116,0.000009019191],"category_scores_gemma":[0.0002958352,0.00013263144,0.00008098216,0.00025079135,0.00032587204,0.000020258038,0.00008899053,0.00020357297,0.00000161399],"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":[0.011788626,0.0017486089,0.9424167,0.00024337802,0.0008987369,0.000008321083,0.0023610725,0.0017721812,0.03296624,0.00032168176,0.0018327517,0.0036416843],"study_design_scores_gemma":[0.0043055476,0.002588076,0.94169635,0.00033944793,0.000069256486,0.000011222033,0.00020387227,0.0034461631,0.045735087,0.0013472368,0.000068062625,0.00018969217],"about_ca_topic_score_codex":0.0001033631,"about_ca_topic_score_gemma":0.00023497613,"teacher_disagreement_score":0.035731368,"about_ca_system_score_codex":0.0002474718,"about_ca_system_score_gemma":0.0007540605,"threshold_uncertainty_score":0.54085535},"labels":[],"label_agreement":null},{"id":"W2895775669","doi":"10.2217/cer-2018-0065","title":"Importance of assessing and adjusting for cross-study heterogeneity in network meta-analysis: a case study of psoriasis","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":43,"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":"Medicine; Psoriasis; Psoriasis Area and Severity Index; Covariate; Placebo; Meta-analysis; Psychological intervention; Clinical trial; Internal medicine; Physical therapy; Statistics; Alternative medicine; Psychiatry; Dermatology; Pathology","score_opus":0.9347608835346701,"score_gpt":0.7117826814298529,"score_spread":0.22297820210481722,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2895775669","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":[0.9845545,0.0019098955,0.01097531,0.000009446126,0.00007512293,0.0023937873,0.0000068274744,5.581332e-7,0.00007452013],"genre_scores_gemma":[0.99834985,0.00000223654,0.0014073702,0.0000022088873,0.00009958848,0.00010820517,2.0323718e-7,0.000008195147,0.00002217364],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.91405994,0.06852771,0.010294107,0.00085170736,0.0058603543,0.00040616098],"domain_scores_gemma":[0.93835163,0.03903906,0.008154143,0.0014129607,0.012893199,0.00014901476],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.35055038,0.00027902832,0.0098351855,0.0016211455,0.00032815695,0.0005333509,0.0010445882,0.00005733549,0.00022820395],"category_scores_gemma":[0.01714783,0.00013921528,0.0023958492,0.0058194497,0.00032075963,0.00046048296,0.00025704765,0.0004001834,0.000003454374],"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":[0.00034099317,0.0011826929,0.9390821,0.000100227946,0.045330726,0.00012798261,0.004895138,0.008307068,0.00011099393,0.000019832469,0.0000584624,0.00044378842],"study_design_scores_gemma":[0.0017488996,0.0035894546,0.9251434,0.000047169444,0.027239736,0.00010285906,0.030382618,0.010476474,0.0004209122,0.0006804928,0.00002119492,0.00014676228],"about_ca_topic_score_codex":0.00014802377,"about_ca_topic_score_gemma":0.0020543768,"teacher_disagreement_score":0.33340257,"about_ca_system_score_codex":0.00006967889,"about_ca_system_score_gemma":0.00014391808,"threshold_uncertainty_score":0.9911311},"labels":[],"label_agreement":null},{"id":"W2904705793","doi":"10.2217/cer-2018-0051","title":"Using a society database to evaluate a patient safety collaborative: the Cardiovascular Surgical Translational Study","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Health Policy Implementation Science","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":false,"ca_institutions":"McGill University","funders":"Agency for Healthcare Research and Quality; Society of Thoracic Surgeons","keywords":"Medicine; Quality management; Context (archaeology); Intervention (counseling); Patient safety; Propensity score matching; Outcome (game theory); Protocol (science); Health care; Medical emergency; Service (business); Nursing; Alternative medicine; Surgery","score_opus":0.8011968449105192,"score_gpt":0.7514784632609536,"score_spread":0.04971838164956566,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2904705793","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":[0.97564906,0.00018935172,0.016112154,0.0015860979,0.0004580259,0.0052004154,0.00009458444,0.000008493227,0.00070183154],"genre_scores_gemma":[0.9971991,0.00001476045,0.0014528085,0.00028132118,0.00077637075,0.00024052158,0.0000016825818,0.000015922637,0.000017499095],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.9526671,0.041276716,0.0012214568,0.00036311388,0.0036503016,0.0008212899],"domain_scores_gemma":[0.9733095,0.01600015,0.00040300502,0.0004761249,0.009375123,0.0004360902],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.054894514,0.00019630695,0.0007583383,0.00033705373,0.00340123,0.000052413678,0.00056704174,0.00007066249,0.00032544276],"category_scores_gemma":[0.0014338889,0.00012238222,0.0002458097,0.0031872005,0.0005769859,0.0004316584,0.00027734402,0.0014225384,0.000120856566],"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":[0.016046561,0.0016011078,0.09031408,0.00055278745,0.0033729768,0.00015583285,0.8398115,0.014243957,0.008368878,0.0020098938,0.016639112,0.0068833143],"study_design_scores_gemma":[0.018884227,0.010870929,0.4148883,0.0012559616,0.00025198003,0.000096013355,0.18883587,0.009995986,0.0026265082,0.0006712946,0.35098815,0.00063477503],"about_ca_topic_score_codex":0.00021185586,"about_ca_topic_score_gemma":0.00013107203,"teacher_disagreement_score":0.65097564,"about_ca_system_score_codex":0.00091182487,"about_ca_system_score_gemma":0.0048512877,"threshold_uncertainty_score":0.9978962},"labels":[],"label_agreement":null},{"id":"W2905207641","doi":"10.2217/cer-2018-0033","title":"Early cost–effectiveness modeling for better decisions in public research investment of personalized medicine technologies","year":2018,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Advancing Health Outcomes; St. Paul's Hospital; University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Investment (military); Risk analysis (engineering); Cost effectiveness; Personalized medicine; Process (computing); Medicine; Key (lock); Health care; Cost–benefit analysis; Comparative effectiveness research; Health technology; Investment decisions; Public health; Business; Computer science; Economics; Finance; Computer security","score_opus":0.8503316042240933,"score_gpt":0.6303314722597517,"score_spread":0.22000013196434154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2905207641","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":[0.9295493,0.003321155,0.055436812,0.0074648745,0.00031578587,0.0028853058,0.000041343068,0.000013857767,0.0009715987],"genre_scores_gemma":[0.996941,0.00018550373,0.0017782662,0.000094755516,0.0003149413,0.00062580267,0.000004396476,0.00003042992,0.00002494994],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.98595166,0.008041479,0.003662929,0.0005857001,0.0008038632,0.00095433963],"domain_scores_gemma":[0.9600911,0.033668775,0.001325399,0.00050929567,0.00420233,0.00020309004],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.16624917,0.00022999401,0.002144864,0.0041851215,0.00048070843,0.00008995826,0.0009200101,0.00024166597,0.00006061694],"category_scores_gemma":[0.029200137,0.00021507526,0.0001986869,0.0018582029,0.0016957469,0.0007619068,0.0002307826,0.0011003698,0.000081307495],"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":[0.00580373,0.001736079,0.14115846,0.0018202633,0.0010102565,0.000016385931,0.024317147,0.0027527302,0.0031364115,0.8100178,0.005753364,0.00247736],"study_design_scores_gemma":[0.013302346,0.008239018,0.1268968,0.005131803,0.00001714393,0.000024566967,0.021227563,0.049846217,0.0032117704,0.75743383,0.014074013,0.00059491437],"about_ca_topic_score_codex":0.00040329408,"about_ca_topic_score_gemma":0.00008824429,"teacher_disagreement_score":0.13704903,"about_ca_system_score_codex":0.0019115406,"about_ca_system_score_gemma":0.0007137837,"threshold_uncertainty_score":0.9789773},"labels":[],"label_agreement":null},{"id":"W2908997167","doi":"10.2217/cer-2018-0113","title":"Population-based validation of the National Comprehensive Cancer Network recommendations for baseline imaging for bladder cancer: a case for routine baseline bone scan?","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Bladder and Urothelial Cancer Treatments","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":"University of Calgary","funders":"","keywords":"Medicine; Metastasis; Cancer; Bone metastasis; Asymptomatic; Lung cancer; Radiology; Population; Lung; Oncology; Internal medicine","score_opus":0.1506586076286012,"score_gpt":0.4861626378263604,"score_spread":0.3355040301977592,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2908997167","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":[0.89286083,0.0021526194,0.08313068,0.008474715,0.0010985885,0.010038691,0.002180628,0.000012606009,0.000050623872],"genre_scores_gemma":[0.9939484,0.00003073757,0.003305923,0.00021768412,0.0008873714,0.0012854612,0.00018475889,0.000033754914,0.00010592871],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99714065,0.00089869706,0.0006730352,0.0002677569,0.0006764093,0.0003434489],"domain_scores_gemma":[0.9840612,0.0072118305,0.0005535037,0.0001605671,0.007904654,0.00010825317],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003300793,0.00019690301,0.00073603506,0.0002826662,0.00036332375,0.00003151197,0.00013093559,0.000059945134,0.00011892621],"category_scores_gemma":[0.0004603586,0.00013900496,0.0003681253,0.00054168154,0.000084886444,0.00017132069,0.000028384475,0.00027435424,7.1104176e-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":[0.05581086,0.0027441166,0.45313758,0.003985216,0.0040052976,0.000018740186,0.0011288121,0.3828529,0.02852122,0.0013395735,0.049406514,0.017049173],"study_design_scores_gemma":[0.074652985,0.0058725453,0.24819118,0.006548349,0.0018965992,0.00035739713,0.0011192563,0.46808234,0.16245417,0.0049361074,0.025093455,0.00079560705],"about_ca_topic_score_codex":0.00047431217,"about_ca_topic_score_gemma":0.00013722354,"teacher_disagreement_score":0.20494638,"about_ca_system_score_codex":0.0007832757,"about_ca_system_score_gemma":0.0014116176,"threshold_uncertainty_score":0.5668458},"labels":[],"label_agreement":null},{"id":"W2909436531","doi":"10.2217/cer-2018-0073","title":"A cost–utility analysis of dulaglutide versus insulin glargine as third-line therapy for Type 2 diabetes in Canada","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Diabetes Treatment and Management","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Eli Lilly (Canada)","funders":"","keywords":"Medicine; Insulin glargine; Dulaglutide; Type 2 diabetes; Diabetes mellitus; Internal medicine; Insulin; Exenatide; Endocrinology","score_opus":0.16141366002993843,"score_gpt":0.457361692122425,"score_spread":0.29594803209248655,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2909436531","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":[0.995166,0.0018707525,0.00004689035,0.00014950246,0.00021675018,0.0018048381,0.000015442998,0.0000027891756,0.0007270483],"genre_scores_gemma":[0.99949354,0.00020259478,0.000086043816,0.000018178833,0.000037215155,0.000051321735,0.000025137224,0.000010400745,0.00007557518],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974133,0.00069775316,0.000472781,0.0002114965,0.0008348213,0.00036987892],"domain_scores_gemma":[0.9946025,0.0032700375,0.00022632496,0.00025771905,0.0015130496,0.0001303787],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026721396,0.0001590423,0.0012180329,0.0008987249,0.00003710423,0.000014460518,0.00018622709,0.0000445351,0.00012660444],"category_scores_gemma":[0.00030828398,0.00011577021,0.00021943101,0.0019123158,0.00007911252,0.000111171044,0.000052124495,0.00033655573,0.0000056280614],"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":[0.03967309,0.0013030457,0.93119174,0.0003393936,0.009177423,0.000036918813,0.0003287312,0.0034675181,0.006445391,0.00028076512,0.0003222367,0.007433753],"study_design_scores_gemma":[0.013436121,0.0063765342,0.9311091,0.00024476845,0.00040975065,4.2953562e-7,0.00046508145,0.010738807,0.03534467,0.00014025785,0.0016323916,0.000102102465],"about_ca_topic_score_codex":0.024260057,"about_ca_topic_score_gemma":0.048290927,"teacher_disagreement_score":0.02889928,"about_ca_system_score_codex":0.0008712355,"about_ca_system_score_gemma":0.0014184725,"threshold_uncertainty_score":0.98223746},"labels":[],"label_agreement":null},{"id":"W2914902447","doi":"10.2217/cer-2018-0140","title":"Medicaid spending burden among beneficiaries with treatment-resistant depression","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Mental Health Treatment and Access","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":false,"ca_institutions":"Group for Research in Decision Analysis","funders":"H2020 European Research Council; Janssen Scientific Affairs","keywords":"Medicaid; Major depressive disorder; Medicine; Depression (economics); Treatment-resistant depression; Psychiatry; Cohort; Incidence (geometry); Emergency medicine; Health care; Internal medicine","score_opus":0.11921362968322262,"score_gpt":0.49022224188282787,"score_spread":0.37100861219960524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2914902447","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":[0.98270285,0.002177077,0.000115466784,0.000054938242,0.0004472047,0.0012177549,0.000006328788,0.0000121645,0.0132662],"genre_scores_gemma":[0.9987124,0.000038673723,0.000063724845,0.0000025743593,0.00026190004,0.000086047374,0.000006077969,0.000023412022,0.0008051948],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9955343,0.0020725047,0.0004404206,0.00033277227,0.0010450048,0.00057501986],"domain_scores_gemma":[0.99594307,0.0027261442,0.00036094323,0.0002952225,0.0004197671,0.00025485145],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0019977977,0.00024393649,0.00075882266,0.0005249548,0.00037189136,0.00009548419,0.00035536624,0.00010626843,0.0012927607],"category_scores_gemma":[0.000023030781,0.00014382532,0.000116279436,0.0005089158,0.0002216938,0.00039380978,0.000066219545,0.0005573341,0.00022269507],"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":[0.024658086,0.0013263009,0.93243873,0.00021979083,0.0011949971,0.00071333343,0.0077899364,0.00044564973,0.0057891887,0.0027030848,0.00016811055,0.022552812],"study_design_scores_gemma":[0.007317629,0.00850309,0.95940715,0.0012248867,0.00006273059,0.0000816736,0.005458344,0.00013563831,0.01610015,0.00040001867,0.001096284,0.00021239184],"about_ca_topic_score_codex":0.00056787085,"about_ca_topic_score_gemma":0.00012877844,"teacher_disagreement_score":0.026968453,"about_ca_system_score_codex":0.0005745391,"about_ca_system_score_gemma":0.00012249619,"threshold_uncertainty_score":0.9996202},"labels":[],"label_agreement":null},{"id":"W2946062710","doi":"10.2217/cer-2018-0148","title":"Is routine baseline brain imaging needed for all newly diagnosed non-small-cell lung cancer patients?","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","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":"University of Calgary","funders":"","keywords":"Medicine; Brain metastasis; Oncology; Lung cancer; Internal medicine; Stage (stratigraphy); Metastasis; Adenocarcinoma; Cohort; Cancer; Radiology","score_opus":0.05032851838770644,"score_gpt":0.4331460358121984,"score_spread":0.382817517424492,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2946062710","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":[0.9877686,0.002419055,0.0017420858,0.0034118162,0.0004437843,0.003631557,0.00005159911,0.000009504479,0.0005220231],"genre_scores_gemma":[0.9975847,0.00022807173,0.0003673215,0.0005655606,0.00042108435,0.00042369653,0.000027114404,0.000045556808,0.0003368883],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9966415,0.000813807,0.0005790731,0.00040633953,0.00095057284,0.0006086876],"domain_scores_gemma":[0.9907738,0.0052598016,0.00035217887,0.00029468624,0.00298855,0.00033097027],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029098708,0.0002930868,0.0010360524,0.0004761651,0.00013653388,0.000082127626,0.00023772527,0.000076228505,0.00025330947],"category_scores_gemma":[0.00021582568,0.00021362698,0.00037744053,0.00047663404,0.000083772255,0.00022914395,0.00008720082,0.0006182466,0.000027815138],"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":[0.005885816,0.002565101,0.93588674,0.0013201659,0.0009771524,0.00006411641,0.0024199968,0.00051678316,0.008477321,0.000026165952,0.0400933,0.0017673436],"study_design_scores_gemma":[0.05404303,0.006056843,0.7605018,0.0037974946,0.00051539525,0.000017181566,0.0010687088,0.017011648,0.14997825,0.00008868868,0.006510092,0.00041087507],"about_ca_topic_score_codex":0.00039115982,"about_ca_topic_score_gemma":0.000028086752,"teacher_disagreement_score":0.17538494,"about_ca_system_score_codex":0.0013631878,"about_ca_system_score_gemma":0.0007263323,"threshold_uncertainty_score":0.8711455},"labels":[],"label_agreement":null},{"id":"W2953676957","doi":"10.2217/cer-2019-0052","title":"Cost–effectiveness of lenalidomide plus low-dose dexamethasone for newly diagnosed multiple myeloma patients ineligible for stem cell transplantation in China","year":2019,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Myeloma Research and Treatments","field":"Medicine","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":"University of Toronto","funders":"","keywords":"Medicine; Multiple myeloma; Lenalidomide; Cost effectiveness; Quality-adjusted life year; Transplantation; Bortezomib; Dexamethasone; Autologous stem-cell transplantation; China; Internal medicine; Oncology; Risk analysis (engineering)","score_opus":0.23850686821996936,"score_gpt":0.48982209004031313,"score_spread":0.25131522182034377,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2953676957","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":[0.39750376,0.5558798,0.009854261,0.000005790199,0.00038436768,0.035399903,0.00088222534,0.000011946449,0.00007797265],"genre_scores_gemma":[0.5073461,0.4869815,0.0008068012,0.0000015328915,0.00015685748,0.0039208597,0.00059492595,0.00013236154,0.000059038233],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9873981,0.0072506666,0.0018151286,0.00075998466,0.0017221854,0.0010539547],"domain_scores_gemma":[0.9539292,0.04070357,0.0011974329,0.00048441262,0.0032529559,0.00043243932],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.008514439,0.0007623327,0.0060749366,0.002517216,0.00016640645,0.000070380986,0.00058718416,0.000474246,0.000009629278],"category_scores_gemma":[0.0010660456,0.0005594785,0.001441515,0.0014241522,0.00026110234,0.0003732093,0.000097086704,0.0014741819,0.000014836022],"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":[0.14556701,0.0064254627,0.015132162,0.40990922,0.00285035,0.0003152257,0.0012669573,0.0017926346,0.00086388126,0.000043792883,0.00010177896,0.41573152],"study_design_scores_gemma":[0.29776457,0.033846356,0.3106767,0.2807076,0.0037453931,0.00018460643,0.000501351,0.0035471192,0.04765999,0.0002739097,0.019267863,0.001824535],"about_ca_topic_score_codex":0.00013285206,"about_ca_topic_score_gemma":0.000034929435,"teacher_disagreement_score":0.413907,"about_ca_system_score_codex":0.002332754,"about_ca_system_score_gemma":0.003058845,"threshold_uncertainty_score":0.99968565},"labels":[],"label_agreement":null},{"id":"W2969452269","doi":"10.2217/cer-2019-0091","title":"Ospemifene versus local estrogen: adherence and costs in postmenopausal dyspareunia","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Estrogen and related hormone effects","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":"Duchesnay (Canada)","funders":"","keywords":"Medicine; Pharmacy; Menopause; Healthcare system; Internal medicine; Gynecology; Health care; Family medicine","score_opus":0.04406549513679099,"score_gpt":0.40015497872199624,"score_spread":0.35608948358520526,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2969452269","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":[0.9924457,0.0054694307,0.00020395621,0.000022987051,0.00023669265,0.0004405549,0.0000028049085,0.0000022337715,0.0011756136],"genre_scores_gemma":[0.9994962,0.0002508344,0.00008991958,0.0000051179354,0.000083900835,0.000013065864,0.000005848392,0.000014261767,0.00004083676],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99730587,0.0013589908,0.00027279984,0.00027014228,0.00043380223,0.00035838873],"domain_scores_gemma":[0.9984888,0.00058750395,0.00011834293,0.00018834486,0.00045146808,0.00016556094],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0017871559,0.00016705664,0.00038997395,0.00021374466,0.00006246676,0.00003653669,0.00024333107,0.00017565189,0.0000143044035],"category_scores_gemma":[0.00010442962,0.00013542529,0.000048092817,0.0003152608,0.000266155,0.000019100851,0.00015818757,0.000694386,0.00003183993],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.013299722,0.0002569991,0.048783056,0.000058538782,0.0002593108,0.00011310068,0.00013821604,0.00093410286,0.93189776,0.00065022794,0.000114365015,0.003494588],"study_design_scores_gemma":[0.013113011,0.039357495,0.16347195,0.00060361397,0.00003330163,0.0003296595,0.0010276397,0.0006912738,0.7795254,0.00020131082,0.00120086,0.00044447483],"about_ca_topic_score_codex":0.000026924106,"about_ca_topic_score_gemma":0.000011147337,"teacher_disagreement_score":0.15237236,"about_ca_system_score_codex":0.00014411648,"about_ca_system_score_gemma":0.00026874462,"threshold_uncertainty_score":0.5522483},"labels":[],"label_agreement":null},{"id":"W2982129066","doi":"10.2217/cer-2019-0043","title":"Population-based assessment of the National Comprehensive Cancer Network recommendations for baseline imaging of rectal cancer","year":2019,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Colorectal Cancer Surgical Treatments","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":"University of Calgary; University of Alberta","funders":"","keywords":"Medicine; Colorectal cancer; Cancer; Metastasis; Lung cancer; Radiology; Population; Oncology; Cohort; Internal medicine","score_opus":0.18378541269258641,"score_gpt":0.5422935991056796,"score_spread":0.35850818641309323,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2982129066","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":[0.9936266,0.00078968564,0.0012127915,0.0015225571,0.0005462755,0.0017385385,0.00012027331,0.0000033300641,0.0004399408],"genre_scores_gemma":[0.99830407,0.00004045107,0.0011108023,0.000050162565,0.0002158876,0.00020192814,0.000017385586,0.000013955439,0.000045381028],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9965307,0.0013674418,0.000555898,0.00017787788,0.0011396422,0.0002284408],"domain_scores_gemma":[0.9876339,0.0061593005,0.000555546,0.00012842739,0.0054433336,0.00007948199],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0020755162,0.00013340604,0.00072635134,0.00023706559,0.00012142752,0.000011247858,0.00016371514,0.000041576353,0.00027608086],"category_scores_gemma":[0.0001968845,0.00008818551,0.0002903351,0.0007826484,0.00012421506,0.00010994505,0.000053437332,0.0004152381,6.178209e-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":[0.0034049004,0.0004330326,0.9272919,0.00036486294,0.00050169614,0.0000012199237,0.000083978484,0.057070345,0.008201677,0.00032338893,0.0008891964,0.0014337598],"study_design_scores_gemma":[0.0040418287,0.0007386889,0.9531414,0.0017303536,0.00007996768,0.0000055404316,0.00007278136,0.02715543,0.011559873,0.00041159536,0.0009893823,0.000073184056],"about_ca_topic_score_codex":0.00029219594,"about_ca_topic_score_gemma":0.000055823686,"teacher_disagreement_score":0.029914916,"about_ca_system_score_codex":0.0011458702,"about_ca_system_score_gemma":0.0015149899,"threshold_uncertainty_score":0.35961008},"labels":[],"label_agreement":null},{"id":"W2990463671","doi":"10.2217/cer-2019-0111","title":"The lateral and medial approach in total arthroplasty for valgus knee: a meta-analysis of current literature","year":2019,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Total Knee Arthroplasty Outcomes","field":"Medicine","cited_by":14,"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; Valgus; Total knee arthroplasty; Valgus deformity; Meta-analysis; Range of motion; Cochrane Library; Osteoarthritis; Physical therapy; Arthroplasty; Randomized controlled trial; Physical medicine and rehabilitation; Surgery; Internal medicine","score_opus":0.3371849787735286,"score_gpt":0.5177686102147684,"score_spread":0.1805836314412398,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2990463671","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":[0.008550593,0.9870493,0.0002758417,0.000030657993,0.00035822068,0.0035390828,0.00013943575,0.000003311028,0.000053596224],"genre_scores_gemma":[0.052492104,0.94625944,0.00021911999,0.0000019464007,0.00027493862,0.00044384724,0.000067376815,0.00005103542,0.00019019713],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.99166363,0.0044895634,0.0014499339,0.00046012184,0.001408032,0.00052872393],"domain_scores_gemma":[0.98308414,0.013729122,0.00095516624,0.00043784725,0.0015874978,0.00020622752],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0069574304,0.0005048915,0.008502195,0.00217098,0.00013681123,0.00012613676,0.00025802234,0.00025916263,0.000018067378],"category_scores_gemma":[0.00061064144,0.000249118,0.0036203293,0.0025350447,0.0005212784,0.00019556271,0.00019375465,0.0022984908,0.000003968448],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010344639,0.0026426534,0.0032990028,0.105437845,0.5533399,0.0001748038,0.0053731175,0.00056693144,0.000029883942,0.0023619721,0.0005593546,0.3158699],"study_design_scores_gemma":[0.0137935495,0.010870075,0.043557383,0.022324214,0.5217875,0.001455463,0.0005405519,0.0021096773,0.000044494627,0.00036808325,0.38185817,0.0012908108],"about_ca_topic_score_codex":0.000003410685,"about_ca_topic_score_gemma":0.000010022942,"teacher_disagreement_score":0.3812988,"about_ca_system_score_codex":0.0002760519,"about_ca_system_score_gemma":0.00090281264,"threshold_uncertainty_score":0.9999961},"labels":[],"label_agreement":null},{"id":"W3009343637","doi":"10.2217/cer-2019-0187","title":"Clinical therapeutic efficacy of mesenchymal stem cells derived from adipose or bone marrow for knee osteoarthritis: a meta-analysis of randomized controlled trials","year":2020,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Osteoarthritis Treatment and Mechanisms","field":"Medicine","cited_by":44,"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; Osteoarthritis; WOMAC; Randomized controlled trial; Meta-analysis; Cochrane Library; Mesenchymal stem cell; Clinical trial; Internal medicine; Bone marrow; Visual analogue scale; Surgery; Pathology","score_opus":0.4233379971369119,"score_gpt":0.5279693907700566,"score_spread":0.10463139363314466,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3009343637","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":[0.0074100913,0.96493405,0.001391685,0.000036831418,0.00039377253,0.02513065,0.0006482142,0.0000110520095,0.000043634143],"genre_scores_gemma":[0.09472658,0.89957505,0.0013067774,0.000012550329,0.0005904128,0.003281735,0.00013548543,0.000116986,0.00025443945],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.92576164,0.0614666,0.009252443,0.00074432563,0.002222779,0.0005522156],"domain_scores_gemma":[0.80555844,0.18257116,0.0074314917,0.00056558807,0.003426954,0.00044636388],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaepi_broad"],"category_scores_codex":[0.05190999,0.0009656383,0.094432265,0.002344555,0.00011713939,0.00007580072,0.0005107831,0.0006371003,0.00049193634],"category_scores_gemma":[0.0068035033,0.00048354833,0.042322278,0.0018959871,0.0007615961,0.00014203785,0.00013118726,0.0015428157,0.0000104269275],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.38987947,0.0010269973,0.0000021634753,0.002533191,0.41367534,0.00018940955,0.00020332562,0.000008182233,0.0009434958,0.000059291644,0.000010095539,0.19146904],"study_design_scores_gemma":[0.3557545,0.020658651,0.000006673891,0.0056064934,0.61041665,0.000030031804,0.00014544328,0.000058847887,0.0025232658,0.00008459459,0.0044638454,0.00025103491],"about_ca_topic_score_codex":0.000011961253,"about_ca_topic_score_gemma":0.00001387561,"teacher_disagreement_score":0.1967413,"about_ca_system_score_codex":0.00021747401,"about_ca_system_score_gemma":0.0027967803,"threshold_uncertainty_score":0.99976164},"labels":[],"label_agreement":null},{"id":"W3012991619","doi":"10.2217/cer-2020-0032","title":"Real-world direct healthcare costs of treating recurrent high-grade serous ovarian cancer with cytotoxic chemotherapy","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Ovarian cancer diagnosis and treatment","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 Ottawa; McGill University; McGill University Health Centre","funders":"","keywords":"Medicine; Chemotherapy; Ovarian cancer; Internal medicine; Serous fluid; Oncology; Cytotoxic T cell; Cancer; Retrospective cohort study","score_opus":0.12655509568559534,"score_gpt":0.4576220673763797,"score_spread":0.33106697169078436,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3012991619","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":[0.987909,0.0020511467,0.00016724481,0.007328298,0.00011248306,0.0014246069,0.000040100538,0.000014854049,0.00095224526],"genre_scores_gemma":[0.9972524,0.0013977492,0.0006282472,0.000075417855,0.0004265806,0.00015444028,0.000006258337,0.00003695496,0.000021946105],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9959904,0.0015154113,0.00055319734,0.00035227326,0.0011777324,0.00041097304],"domain_scores_gemma":[0.9959173,0.001409843,0.00046756756,0.00022311388,0.0015322239,0.00044992616],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011385097,0.00026204792,0.0013619342,0.00036630465,0.00014478265,0.000035488683,0.0001758013,0.000064438005,0.00008971953],"category_scores_gemma":[0.00006224576,0.00017401476,0.00017674443,0.0013476863,0.00019333324,0.00013311306,0.000037229252,0.0007712291,0.0000037342422],"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":[0.06998552,0.0052911006,0.8009052,0.0032763972,0.0065383026,0.0012240302,0.011363477,0.003424762,0.03233237,0.003200551,0.0016822711,0.060775984],"study_design_scores_gemma":[0.013152382,0.015758883,0.7780484,0.0052598873,0.00018653355,0.000045337536,0.00049651135,0.00042686498,0.18581635,0.000097056974,0.000500253,0.00021151759],"about_ca_topic_score_codex":0.0022608945,"about_ca_topic_score_gemma":0.00049635296,"teacher_disagreement_score":0.15348399,"about_ca_system_score_codex":0.0014569166,"about_ca_system_score_gemma":0.0011450829,"threshold_uncertainty_score":0.7096116},"labels":[],"label_agreement":null},{"id":"W3014980098","doi":"10.2217/cer-2019-0179","title":"Impact of cytoreductive surgery on outcomes of metastatic appendiceal carcinoma: a real-world, population-based study","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Intraperitoneal and Appendiceal Malignancies","field":"Medicine","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 Alberta","funders":"","keywords":"Medicine; Adenocarcinoma; Hazard ratio; Proportional hazards model; Oncology; Internal medicine; Population; Stage (stratigraphy); Cancer; Survival analysis; Carcinoma; Surgery; Confidence interval","score_opus":0.19626007860773395,"score_gpt":0.48208794353406725,"score_spread":0.2858278649263333,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3014980098","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":[0.9979539,0.00009656602,0.000075181895,0.00018063972,0.00008754086,0.0012574084,0.000041541756,0.000007116139,0.00030011265],"genre_scores_gemma":[0.99973077,0.000003660055,0.00007048617,0.000009547911,0.00011849992,0.000024417495,0.000010538585,0.000022308423,0.000009761547],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99406755,0.0026100706,0.0009974813,0.0002650842,0.0016900763,0.00036971908],"domain_scores_gemma":[0.9909639,0.005978964,0.00073087093,0.00021501958,0.0018043282,0.00030693554],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026330373,0.00026969274,0.0021659334,0.0010229619,0.00011840377,0.000018169778,0.00021042708,0.000042718944,0.00009217936],"category_scores_gemma":[0.00056427316,0.00017561724,0.0006614811,0.0013853849,0.00024095444,0.00016269967,0.000047352598,0.00071099884,0.0000068766753],"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":[0.01489068,0.0028231926,0.9646211,0.0004896233,0.0013122417,0.00026897842,0.0016707391,0.001509646,0.01190001,0.00028514504,0.00006867731,0.00015993186],"study_design_scores_gemma":[0.0028487663,0.013144977,0.95873004,0.0004043885,0.00018356169,0.000014356124,0.0017294139,0.00025938,0.022467619,0.00009658047,0.0000023783596,0.00011856361],"about_ca_topic_score_codex":0.0014088878,"about_ca_topic_score_gemma":0.0000286205,"teacher_disagreement_score":0.012041913,"about_ca_system_score_codex":0.0003879244,"about_ca_system_score_gemma":0.0010916255,"threshold_uncertainty_score":0.7161463},"labels":[],"label_agreement":null},{"id":"W3081136664","doi":"10.2217/cer-2020-0095","title":"Meta-analyses of ataluren randomized controlled trials in nonsense mutation Duchenne muscular dystrophy","year":2020,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Muscle Physiology and Disorders","field":"Biochemistry, Genetics and Molecular Biology","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":"BC Children's Hospital; University of Calgary; Alberta Children's Hospital; Western University","funders":"National Institutes of Health; Ministero della Salute; Cytokinetics; Parent Project Muscular Dystrophy; Fondazione Telethon; PTC Therapeutics; Muscular Dystrophy Association; Sarepta Therapeutics; Ionis Pharmaceuticals; FibroGen; AveXis; Biogen; Bristol-Myers Squibb","keywords":"Medicine; Duchenne muscular dystrophy; Placebo; Internal medicine; Randomized controlled trial; Ambulatory; Nonsense mutation; Meta-analysis; Subgroup analysis; Physical therapy; Mutation; Genetics","score_opus":0.5116218416020909,"score_gpt":0.5738630164585028,"score_spread":0.062241174856411896,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3081136664","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":[0.0072901426,0.98666805,0.0021202455,0.000027429882,0.000104865576,0.0036429896,0.00003680525,0.0000016468608,0.0001078183],"genre_scores_gemma":[0.14596447,0.853006,0.00015226222,0.0000052994483,0.00020636845,0.0005417805,0.00008504561,0.00002676441,0.000012007466],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.91550666,0.079697154,0.003348108,0.00043728665,0.00070889393,0.0003019111],"domain_scores_gemma":[0.97641456,0.01838131,0.003255556,0.0003401307,0.0014794919,0.00012897524],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.038219914,0.0004983286,0.023184799,0.00090319046,0.00006616145,0.000033130007,0.00048583542,0.0003917408,0.00004977677],"category_scores_gemma":[0.015534536,0.00027346893,0.008219945,0.00069033814,0.000554106,0.000013573754,0.00012963531,0.00095150317,0.0000058529267],"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":[0.6861932,0.0011331141,0.0000035108387,0.012030721,0.23246665,0.00026862626,0.00027889057,0.0025485042,0.040337645,0.00026145214,0.0003698696,0.024107777],"study_design_scores_gemma":[0.69429326,0.008421006,0.00010908928,0.009831356,0.1378324,0.00022690032,0.00037874866,0.00033901358,0.025431661,0.0034043286,0.11809171,0.0016405521],"about_ca_topic_score_codex":0.000029013318,"about_ca_topic_score_gemma":0.0000059027398,"teacher_disagreement_score":0.13867433,"about_ca_system_score_codex":0.000059558442,"about_ca_system_score_gemma":0.0011485084,"threshold_uncertainty_score":0.99997175},"labels":[],"label_agreement":null},{"id":"W3084779451","doi":"10.2217/cer-2020-0155","title":"Patient and caregiver experiences at a specialized primary care center for autistic adults","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Chronic Disease Management Strategies","field":"Medicine","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":"Royal College of Physicians and Surgeons of Canada","funders":"National Center for Advancing Translational Sciences","keywords":"Medicine; Health care; Medical home; Autism; Special needs; Patient satisfaction; Population; Family medicine; Healthcare delivery; Primary care; Psychiatry; Nursing; Environmental health","score_opus":0.11677642429807876,"score_gpt":0.4300710642970473,"score_spread":0.3132946399989685,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3084779451","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":[0.9910573,0.0028297454,0.00056656974,0.0003422701,0.00014558798,0.0015724825,0.000028375325,0.000009219005,0.0034484875],"genre_scores_gemma":[0.9993231,0.000059004284,0.0001225534,0.000060153932,0.00025421966,0.00012524218,0.000020846972,0.000012438346,0.000022441052],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.997874,0.00050149346,0.00029492142,0.0002582498,0.0007889307,0.0002823942],"domain_scores_gemma":[0.9978697,0.00066713674,0.00014494805,0.0001075768,0.00095897773,0.0002516672],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003328413,0.00014988174,0.0005795343,0.00016511847,0.00017234946,0.000057007695,0.00011553468,0.000037654765,0.0000925587],"category_scores_gemma":[0.00014762378,0.00011293395,0.00013692839,0.00022906365,0.00033322047,0.0001556753,0.00017077287,0.0002485224,0.000006351223],"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":[0.21082522,0.0018588454,0.04509832,0.026425285,0.0023190742,0.0019069592,0.64461905,0.0003538041,0.010300782,0.0020469564,0.028457418,0.02578832],"study_design_scores_gemma":[0.05187946,0.029406551,0.50622785,0.0060460404,0.0004004772,0.00018740701,0.31866732,0.0017021728,0.014169798,0.00039636172,0.07019993,0.00071663805],"about_ca_topic_score_codex":0.000005714731,"about_ca_topic_score_gemma":0.000004345239,"teacher_disagreement_score":0.46112952,"about_ca_system_score_codex":0.0006020902,"about_ca_system_score_gemma":0.00032651718,"threshold_uncertainty_score":0.4605313},"labels":[],"label_agreement":null},{"id":"W3088056570","doi":"10.2217/cer-2020-0124","title":"Experiences of cancer patients with outpatient care in the USA: a population-based study","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Patient Satisfaction 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":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Medicine; Medical Expenditure Panel Survey; Logistic regression; Health care; Odds ratio; Odds; Family medicine; Cancer; Population; Outpatient clinic; Gerontology; Environmental health; Internal medicine; Health insurance","score_opus":0.2881492551541447,"score_gpt":0.5823500811200549,"score_spread":0.29420082596591024,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3088056570","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":[0.99574333,0.00020010845,0.00006162787,0.00012487355,0.00025108652,0.0035191814,0.000015624071,0.00000343074,0.000080721744],"genre_scores_gemma":[0.99894756,0.000002376975,0.000029153589,0.000053968753,0.00008039992,0.0008705771,0.0000044552125,0.000010451836,0.0000010601591],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98465204,0.0121597815,0.00082141993,0.00020800317,0.0018223261,0.00033644246],"domain_scores_gemma":[0.9905527,0.005174643,0.00068489346,0.00016906143,0.0032892239,0.00012944585],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019745168,0.00014359318,0.0005742967,0.00033063142,0.00047892448,0.000010483304,0.00036892592,0.000068783374,0.00011084933],"category_scores_gemma":[0.0004539139,0.000081736725,0.000056841713,0.0011956773,0.00010034947,0.00019970484,0.000052200212,0.0013807445,0.0000060772195],"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":[0.002523551,0.00024459316,0.7457588,0.00030560215,0.00002770024,0.0000033727915,0.24894013,0.0020288683,0.0000057898355,0.000015258658,0.00001728056,0.00012908755],"study_design_scores_gemma":[0.0024940749,0.005498256,0.8000271,0.0005294405,0.000008716741,4.731068e-8,0.19119836,0.00007875737,0.00007350051,0.0000064843034,0.00002359285,0.00006170919],"about_ca_topic_score_codex":0.0018384993,"about_ca_topic_score_gemma":0.002574916,"teacher_disagreement_score":0.057741757,"about_ca_system_score_codex":0.00060337636,"about_ca_system_score_gemma":0.000859795,"threshold_uncertainty_score":0.59987247},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3094115567","doi":"10.2217/cer-2020-0122","title":"Comparison of ofatumumab and other disease-modifying therapies for relapsing multiple sclerosis: a network meta-analysis","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Sclerosis Research Studies","field":"Medicine","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":"EVERSANA (Canada)","funders":"","keywords":"Ofatumumab; Medicine; Natalizumab; Ocrelizumab; Alemtuzumab; Multiple sclerosis; Internal medicine; Meta-analysis; Oncology; Disease; Immunology; Rituximab","score_opus":0.6942617178013222,"score_gpt":0.5352789439256699,"score_spread":0.1589827738756523,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094115567","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":[0.90111697,0.059117526,0.035375763,0.0016272401,0.00005619815,0.0024730032,0.00007158427,0.000017910686,0.0001437841],"genre_scores_gemma":[0.9957246,0.00027944148,0.003504519,0.000049567505,0.00023422863,0.00016318714,0.0000033240153,0.000030633615,0.000010470587],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9939282,0.002649344,0.00089761737,0.00040178935,0.0015736335,0.00054943125],"domain_scores_gemma":[0.9867545,0.009453798,0.0005372859,0.00026888092,0.0024774126,0.00050815404],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0048657893,0.0002799149,0.0032822844,0.000571046,0.00039009587,0.000081722785,0.00027129566,0.00007606244,0.000055302844],"category_scores_gemma":[0.0021119413,0.00019039409,0.001319512,0.0016240274,0.000656191,0.00020280606,0.00019102222,0.00081659865,0.0000019170272],"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":[0.034538694,0.0012250153,0.5973555,0.002235421,0.2632749,0.000017216642,0.014011597,0.01875313,0.065200314,0.0005289303,0.0012769776,0.0015823185],"study_design_scores_gemma":[0.008956924,0.004742072,0.82526404,0.00076801603,0.046140958,0.000004270996,0.0044995374,0.0897702,0.017572254,0.00050052727,0.0014108083,0.0003703799],"about_ca_topic_score_codex":0.00003148205,"about_ca_topic_score_gemma":0.000020220552,"teacher_disagreement_score":0.22790858,"about_ca_system_score_codex":0.0001243773,"about_ca_system_score_gemma":0.00022903646,"threshold_uncertainty_score":0.77640456},"labels":[],"label_agreement":null},{"id":"W3109637262","doi":"10.2217/cer-2020-0079","title":"Impact of socioeconomic status on presentation, treatment and outcomes of patients with pancreatic cancer","year":2020,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Global Cancer Incidence and Screening","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":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Medicine; Socioeconomic status; Pancreatic cancer; Logistic regression; Odds ratio; Epidemiology; Internal medicine; Proportional hazards model; Cancer registry; Cancer; Pancreatectomy; Surgery; Demography; Pancreas; Population; Environmental health","score_opus":0.1825159377404781,"score_gpt":0.5083818008052303,"score_spread":0.3258658630647522,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3109637262","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":[0.99862546,0.00046781072,0.00006251829,0.000108476524,0.000014606835,0.00054840115,0.000036749523,0.0000015290618,0.0001344229],"genre_scores_gemma":[0.9997339,0.00012743744,0.00006520952,0.000009595131,0.00003232401,0.00001572301,0.000002985852,0.0000071037875,0.0000056968956],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9984901,0.00042223386,0.00030662116,0.00012260312,0.00047567906,0.00018275838],"domain_scores_gemma":[0.9973798,0.00094682764,0.00029052023,0.00007993777,0.0011103539,0.00019257535],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00033044597,0.00011718044,0.00077370624,0.00015241806,0.000041623356,0.000010109257,0.00005995021,0.000027526206,0.000057559773],"category_scores_gemma":[0.00009562577,0.00006651784,0.00011825029,0.00017241207,0.00018896935,0.0001354168,0.00001812308,0.00018872069,0.000001343349],"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":[0.013497186,0.0002730323,0.97901183,0.000111066474,0.0010530272,0.000005678148,0.0019846573,0.0016615954,0.0011520631,0.000022653294,0.000056771034,0.0011704308],"study_design_scores_gemma":[0.005044497,0.019196842,0.9684307,0.00027184578,0.000095789495,0.0000015502478,0.00046640763,0.000118683005,0.006281762,0.00003242723,0.000011210964,0.000048279067],"about_ca_topic_score_codex":0.0005093443,"about_ca_topic_score_gemma":0.000015407446,"teacher_disagreement_score":0.01892381,"about_ca_system_score_codex":0.00048275033,"about_ca_system_score_gemma":0.0005745323,"threshold_uncertainty_score":0.27125186},"labels":[],"label_agreement":null},{"id":"W3121768136","doi":"10.2217/cer-2020-0187","title":"Patient-related barriers to some virtual healthcare services among cancer patients in the USA: a population-based study","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Telemedicine and Telehealth Implementation","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":"University of Alberta","funders":"","keywords":"Medicine; Health care; Logistic regression; Health Information National Trends Survey; Family medicine; Medical prescription; Odds ratio; Population; Odds; Cohort; National Health Interview Survey; Demography; Gerontology; Nursing; Health information; Environmental health; Internal medicine","score_opus":0.07384825278749714,"score_gpt":0.48197826962356877,"score_spread":0.40813001683607164,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3121768136","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":[0.99570435,0.00017791639,0.000013226884,0.0015890864,0.0002445218,0.002237001,0.0000133427675,0.0000042550964,0.000016269838],"genre_scores_gemma":[0.9990537,0.000009088674,0.00001674331,0.0005701406,0.00010639913,0.00019383949,0.000030598727,0.000013203181,0.0000062462454],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.992599,0.004299725,0.00076656986,0.00024388911,0.0016959932,0.00039478415],"domain_scores_gemma":[0.9955628,0.0014482317,0.0002741812,0.00019418092,0.0021251144,0.00039547603],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026520286,0.00015375014,0.00058624643,0.00063793175,0.00027313273,0.00004269112,0.00015894599,0.000056422447,0.00012153049],"category_scores_gemma":[0.00039602423,0.00010293322,0.000078193014,0.0015926817,0.000043745502,0.0002504679,0.00004625015,0.0008593734,0.0000058146825],"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":[0.0013823203,0.00085133046,0.9815178,0.00018867399,0.00010260509,0.00013070926,0.013007481,0.00079630647,0.00013122629,0.000045238474,0.00008909816,0.0017572069],"study_design_scores_gemma":[0.0052852724,0.0075064534,0.9670384,0.0006669784,0.00003377739,0.000005063042,0.018331572,0.00014830875,0.0007704823,0.000066037224,0.00006826584,0.00007941599],"about_ca_topic_score_codex":0.0043537747,"about_ca_topic_score_gemma":0.008619865,"teacher_disagreement_score":0.014479432,"about_ca_system_score_codex":0.0006236078,"about_ca_system_score_gemma":0.0011173931,"threshold_uncertainty_score":0.6581634},"labels":[],"label_agreement":null},{"id":"W3127294621","doi":"10.2217/cer-2020-0173","title":"Factors affecting treatment selection and overall survival for first-line EGFR-tyrosine kinase inhibitor therapy in non-small-cell lung cancer","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Lung Cancer Treatments and Mutations","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":true,"ca_institutions":"University of Waterloo","funders":"","keywords":"Medicine; Osimertinib; Lung cancer; Oncology; Internal medicine; Retrospective cohort study; Tyrosine-kinase inhibitor; Tyrosine kinase; Cohort; Cancer; Targeted therapy; Erlotinib; Epidermal growth factor receptor","score_opus":0.11775978284061421,"score_gpt":0.46844371141222735,"score_spread":0.3506839285716131,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3127294621","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":[0.9910594,0.005233133,0.0014562773,0.00023656945,0.00025841928,0.0016289677,0.000018992507,0.0000061844516,0.0001020518],"genre_scores_gemma":[0.9978713,0.0012267655,0.00017951107,0.000007597515,0.0003265926,0.00018496269,0.000016220272,0.000024101046,0.00016299385],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99812603,0.00057373056,0.0003234947,0.00027706724,0.00036809364,0.0003316033],"domain_scores_gemma":[0.9962236,0.002089983,0.00015341141,0.000097613854,0.0012733256,0.00016201052],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007950289,0.00021603359,0.0007053835,0.00034646367,0.0002465526,0.00006510709,0.000046488683,0.00007786065,0.000026590998],"category_scores_gemma":[0.00008900695,0.00014811567,0.00015565324,0.00049407425,0.000050597824,0.00016050335,0.000023585353,0.00034327726,3.7035983e-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":[0.0058966754,0.004171815,0.85177493,0.0008542677,0.0013799712,0.0001487829,0.008075744,0.0047770827,0.116474904,0.000052477455,0.000106473235,0.006286884],"study_design_scores_gemma":[0.016727818,0.005332877,0.59361595,0.0008177549,0.0001446909,0.00001897774,0.0009025212,0.0018081676,0.38015565,0.00006546509,0.00027598153,0.00013413865],"about_ca_topic_score_codex":0.00036879137,"about_ca_topic_score_gemma":0.00092063996,"teacher_disagreement_score":0.26368076,"about_ca_system_score_codex":0.0015830329,"about_ca_system_score_gemma":0.0008158888,"threshold_uncertainty_score":0.6039982},"labels":[],"label_agreement":null},{"id":"W3128163814","doi":"10.2217/cer-2020-0137","title":"Heterogeneity in renal end points of cardiovascular outcomes trials in Type 2 diabetes","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Diabetes Treatment and Management","field":"Medicine","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; Impact","funders":"","keywords":"Medicine; End stage renal disease; Renal function; Type 2 diabetes; Clinical endpoint; Internal medicine; Intensive care medicine; Clinical trial; Creatinine; Diabetes mellitus; Urology; Disease; Endocrinology","score_opus":0.20559037985098247,"score_gpt":0.4711150407476389,"score_spread":0.2655246608966564,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3128163814","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":[0.9898834,0.008480419,0.000049700782,0.00013666952,0.0001422476,0.00062051794,0.0000029996386,0.0000022710235,0.0006817716],"genre_scores_gemma":[0.9992546,0.00037235298,0.0002417077,0.000008711926,0.000049434446,0.00002491596,0.00000495395,0.000009755301,0.000033599055],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99127984,0.0063220477,0.0008146401,0.0002055554,0.001055311,0.00032259207],"domain_scores_gemma":[0.9946212,0.003770258,0.00020466991,0.0002759486,0.001026585,0.00010132759],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.016066622,0.00014694652,0.0023107722,0.0008194181,0.00002586409,0.000022038404,0.00012759457,0.000063040745,0.00005998424],"category_scores_gemma":[0.002133158,0.00010515175,0.0005658225,0.0009963177,0.00012314031,0.00013603696,0.00010565689,0.0004411383,0.0000072160997],"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":[0.001979525,0.0018687523,0.95747703,0.0004999768,0.0052219103,0.0007956814,0.0003948973,0.0013286866,0.02296924,0.0002224983,0.00010007486,0.007141736],"study_design_scores_gemma":[0.0054248455,0.0010389801,0.7642286,0.0007670454,0.00015616356,0.000005163805,0.00015060682,0.000052442778,0.22706933,0.00039537865,0.00064519583,0.00006628127],"about_ca_topic_score_codex":0.000023922461,"about_ca_topic_score_gemma":0.000032459924,"teacher_disagreement_score":0.20410009,"about_ca_system_score_codex":0.0003203854,"about_ca_system_score_gemma":0.00031748708,"threshold_uncertainty_score":0.55684024},"labels":[],"label_agreement":null},{"id":"W3129342924","doi":"10.2217/cer-2020-0272","title":"Matching-adjusted indirect comparison of palbociclib versus ribociclib and abemaciclib in hormone receptor-positive/HER2-negative advanced breast cancer","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Breast Cancer Therapies","field":"Medicine","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Fulvestrant; Palbociclib; Hazard ratio; Oncology; Internal medicine; Breast cancer; Hormone receptor; Gynecology; Cancer; Confidence interval; Metastatic breast cancer; Estrogen receptor","score_opus":0.09824933317196574,"score_gpt":0.46670897891742386,"score_spread":0.3684596457454581,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3129342924","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":[0.98722035,0.010221773,0.00041396948,0.0004665058,0.00023981798,0.0008151878,0.00009063562,0.000015048692,0.0005167365],"genre_scores_gemma":[0.997011,0.000862218,0.0017486772,0.000015175902,0.00017130014,0.000080872516,0.000009863199,0.000047221252,0.00005365565],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99374664,0.0028372246,0.00096873945,0.0004907884,0.001344164,0.00061246305],"domain_scores_gemma":[0.9903334,0.005130684,0.0006492436,0.00025287532,0.0033209731,0.0003128178],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0020008122,0.00036865124,0.002078515,0.0006881967,0.00016885863,0.000043043077,0.00023386495,0.0001746836,0.00012903492],"category_scores_gemma":[0.00034112044,0.00031938154,0.00020661045,0.0018717414,0.0007754243,0.00054663356,0.00017549253,0.0017574077,0.0000062271743],"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":[0.121939614,0.002454155,0.13943395,0.0008614711,0.002173362,0.00036614054,0.020381404,0.0027419962,0.651737,0.00046466466,0.00008775824,0.05735853],"study_design_scores_gemma":[0.016665734,0.0021694163,0.67640847,0.002477426,0.000088611516,0.00021503161,0.011059156,0.00013978707,0.28964922,0.0007310045,0.0001469726,0.00024920658],"about_ca_topic_score_codex":0.00022564737,"about_ca_topic_score_gemma":0.00024268015,"teacher_disagreement_score":0.5369745,"about_ca_system_score_codex":0.0010674953,"about_ca_system_score_gemma":0.0010051174,"threshold_uncertainty_score":0.99992585},"labels":[],"label_agreement":null},{"id":"W3129382481","doi":"10.2217/cer-2020-0267","title":"Efficacy classification of modern therapies in multiple sclerosis","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":118,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Ocrelizumab; Cladribine; Ofatumumab; Multiple sclerosis; Relapsing remitting; Internal medicine; Placebo; MEDLINE; Oncology; Intensive care medicine; Alternative medicine; Rituximab; Pathology; Immunology","score_opus":0.436733189315095,"score_gpt":0.47966743189235955,"score_spread":0.042934242577264525,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3129382481","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":[0.9921694,0.004694711,0.0011768638,0.00042957935,0.00006856816,0.0006937122,0.000006142677,0.0000057903057,0.0007552688],"genre_scores_gemma":[0.9977134,0.0014221985,0.00064922904,0.0000049666205,0.000102596416,0.000049222206,0.000003883704,0.000017431239,0.00003707418],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.9937113,0.0032653762,0.00068946637,0.00027374868,0.0016481616,0.0004119761],"domain_scores_gemma":[0.9861233,0.008677774,0.0002474658,0.00033092202,0.004453756,0.00016674974],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0043875943,0.00016471402,0.0010227319,0.0008284219,0.00012991461,0.000036191857,0.00023127807,0.00009107425,0.00004960753],"category_scores_gemma":[0.0038236245,0.00012714586,0.0002326746,0.0014256826,0.0005477614,0.00021621119,0.00014616949,0.0010022315,0.000009387128],"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":[0.004542001,0.0017074682,0.18726654,0.00035584098,0.0003794309,0.000049678747,0.0030628426,0.00039999312,0.78687114,0.0001439993,0.00011356763,0.01510751],"study_design_scores_gemma":[0.005327792,0.0006949479,0.7782831,0.0014642337,0.000014404643,0.000017538168,0.0012884127,0.0019100616,0.21054707,0.00027513644,0.0001056018,0.000071652496],"about_ca_topic_score_codex":0.000027404241,"about_ca_topic_score_gemma":0.00007664615,"teacher_disagreement_score":0.5910166,"about_ca_system_score_codex":0.0004107603,"about_ca_system_score_gemma":0.00083983486,"threshold_uncertainty_score":0.5184858},"labels":[],"label_agreement":null},{"id":"W3159940388","doi":"10.2217/cer-2020-0279","title":"Current practice of stakeholder engagement: researchers’ experiences in North America, DACH countries and China","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Stakeholder; Stakeholder engagement; China; Descriptive statistics; Medicine; Variance (accounting); Public relations; Political science; Accounting; Business","score_opus":0.7311056339849136,"score_gpt":0.5893864776314628,"score_spread":0.14171915635345078,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3159940388","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":[0.9670883,0.027069332,0.0012728095,0.0026988387,0.00021258321,0.0005239375,0.000034669094,0.0000026082732,0.0010969234],"genre_scores_gemma":[0.99676,0.0022440574,0.0006694818,0.000088975314,0.00011283592,0.000085074644,0.0000047448875,0.000011019631,0.000023823053],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9870138,0.009214788,0.002437422,0.00036807454,0.0005449819,0.00042088752],"domain_scores_gemma":[0.98357284,0.013126903,0.0016405085,0.0002695822,0.0012075899,0.00018259823],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.049063567,0.00014288045,0.0012087377,0.00079716626,0.0002089674,0.00012954253,0.00031160627,0.00005520206,0.00018367944],"category_scores_gemma":[0.01193325,0.00015045656,0.00007896718,0.001003205,0.0004973215,0.0010852196,0.00015101528,0.0009694386,0.000040394956],"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":[0.0015328556,0.0020727515,0.63705176,0.002769367,0.00055350765,0.000093402035,0.30783176,0.002198791,0.0000818086,0.04009045,0.0038249786,0.0018985502],"study_design_scores_gemma":[0.0026311763,0.0010232268,0.7611776,0.00075693143,0.000009434049,0.000034280543,0.13588774,0.001258572,0.0005977532,0.0040517934,0.09226154,0.00030995492],"about_ca_topic_score_codex":0.00026899826,"about_ca_topic_score_gemma":0.00011054464,"teacher_disagreement_score":0.171944,"about_ca_system_score_codex":0.00056793477,"about_ca_system_score_gemma":0.00079205795,"threshold_uncertainty_score":0.9963896},"labels":[],"label_agreement":null},{"id":"W3164315881","doi":"10.2217/cer-2021-0069","title":"Propensity score matching versus coarsened exact matching in observational comparative effectiveness research","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"London Health Sciences Centre; Western University","funders":"Physicians' Services Incorporated Foundation","keywords":"Propensity score matching; Observational study; Medicine; Randomized controlled trial; Matching (statistics); Comparative effectiveness research; Clinical trial; Balance (ability); Statistics; Physical therapy; Internal medicine; Mathematics; Alternative medicine; Pathology","score_opus":0.8095129096967779,"score_gpt":0.6082277517660837,"score_spread":0.20128515793069424,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3164315881","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":[0.9661387,0.00039424532,0.028811453,0.000118128686,0.00033375202,0.0018212904,0.000014281533,0.00005587193,0.0023122574],"genre_scores_gemma":[0.99059767,0.000056019464,0.008770002,0.000007142757,0.00021870434,0.00023070599,0.000013432881,0.000057473197,0.000048824753],"study_design_codex":"bench_or_experimental","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9689988,0.024358248,0.0012552764,0.00077300303,0.0034374173,0.0011772704],"domain_scores_gemma":[0.9334999,0.056171134,0.0005535347,0.0006323943,0.008841533,0.0003015337],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.029281761,0.0004999183,0.0021037946,0.0013478708,0.00072342367,0.00032717007,0.00095902395,0.00027560574,0.00008923419],"category_scores_gemma":[0.0042354017,0.00043745773,0.00028342032,0.0032025075,0.00083918584,0.001529105,0.0007536031,0.0047407947,0.00003412241],"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":[0.0341129,0.005048009,0.049567983,0.0029363255,0.0013463561,0.0042413985,0.017562144,0.0102914255,0.56370646,0.30882704,0.00076701294,0.0015929585],"study_design_scores_gemma":[0.0038412823,0.0015326318,0.2669365,0.0038912785,0.000026945614,0.0001281122,0.0037300983,0.00026303786,0.24985293,0.46924895,0.000098425655,0.00044979042],"about_ca_topic_score_codex":0.0001544799,"about_ca_topic_score_gemma":0.0003936029,"teacher_disagreement_score":0.3138535,"about_ca_system_score_codex":0.0023937686,"about_ca_system_score_gemma":0.0019755869,"threshold_uncertainty_score":0.9998077},"labels":[],"label_agreement":null},{"id":"W3166331404","doi":"10.2217/cer-2020-0251","title":"Patterns and association of vaccination among adults with a history of cancer in the USA: a population-based study","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Hepatitis B Virus Studies","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":"University of Alberta","funders":"","keywords":"Medicine; Vaccination; Hepatitis B; Logistic regression; Shingles; Demography; Odds ratio; Population; Cohort; Immunology; Internal medicine; Environmental health","score_opus":0.07739618956074105,"score_gpt":0.4322549109253295,"score_spread":0.35485872136458846,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3166331404","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":[0.99836373,0.0006378329,0.00007258699,0.000094761046,0.00004022112,0.0006818779,0.0000054403517,8.287555e-7,0.00010270071],"genre_scores_gemma":[0.99980253,0.000049949485,0.000018077759,0.0000048563315,0.000030099976,0.000058473186,0.0000014194707,0.0000045864435,0.000030027166],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9959624,0.0024548015,0.00033990567,0.00010768072,0.0010183605,0.00011685755],"domain_scores_gemma":[0.9946615,0.002353924,0.00037736143,0.00009395281,0.0024835838,0.000029642499],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0028182354,0.00007430031,0.0004935153,0.0002973503,0.00003713222,0.000006398005,0.000059174363,0.00003328987,0.000031654636],"category_scores_gemma":[0.0005444232,0.000047425696,0.000041176812,0.00040851635,0.000031612595,0.00010429606,0.000020155463,0.00036880118,1.5610897e-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":[0.0014189325,0.00076864363,0.9879991,0.00035280504,0.00021654823,0.000042504955,0.007850622,0.0002852899,0.00055819243,0.000011067805,0.000025596904,0.00047070574],"study_design_scores_gemma":[0.0035260806,0.0016885534,0.99004173,0.00096172304,0.00006187052,0.000002411697,0.0015201479,0.00039511122,0.0017470161,0.000010583031,0.0000069677353,0.000037818103],"about_ca_topic_score_codex":0.0068919645,"about_ca_topic_score_gemma":0.07317733,"teacher_disagreement_score":0.06628537,"about_ca_system_score_codex":0.0012290601,"about_ca_system_score_gemma":0.0005265695,"threshold_uncertainty_score":0.9997212},"labels":[],"label_agreement":null},{"id":"W3174892139","doi":"10.2217/cer-2020-0278","title":"Gender, socioeconomic status and emergency department visits among cancer survivors in the USA: a population-based study","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Palliative Care and End-of-Life Issues","field":"Medicine","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 Alberta","funders":"","keywords":"Medicine; Emergency department; Socioeconomic status; Logistic regression; Demography; Population; Odds ratio; Cohort; Cohort study; Environmental health; Internal medicine; Psychiatry","score_opus":0.2850653871005384,"score_gpt":0.5466509751092106,"score_spread":0.26158558800867215,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3174892139","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":[0.99360865,0.0052773925,0.000091293536,0.000078112615,0.0001261468,0.0006773943,0.00000864354,0.0000016196503,0.00013072966],"genre_scores_gemma":[0.99910295,0.00066353084,0.000048218113,0.000008986865,0.00008343347,0.00006621828,0.0000048276024,0.000008680236,0.000013152027],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99536717,0.0030473701,0.00046175986,0.00019222674,0.00066022016,0.0002712696],"domain_scores_gemma":[0.99650705,0.0020350274,0.00019956591,0.0001525069,0.0009954359,0.000110439556],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002662857,0.00013051108,0.0006131161,0.00028634293,0.00010899695,0.000022129916,0.00010617855,0.000043640022,0.00013180314],"category_scores_gemma":[0.00022381384,0.00008874955,0.000103180144,0.00042016635,0.000073327494,0.00012564626,0.000039310205,0.00049358245,0.0000024872115],"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":[0.000687095,0.0006965451,0.9904886,0.00015139669,0.00028455385,0.00010357694,0.005708687,0.0008960647,0.00063822436,0.00004105622,0.00002755321,0.00027664856],"study_design_scores_gemma":[0.0021228981,0.00080314593,0.986658,0.00029902987,0.000052787203,0.0000029057185,0.0064952346,0.00021920494,0.0030182027,0.00014688971,0.00010790947,0.00007383225],"about_ca_topic_score_codex":0.0020310893,"about_ca_topic_score_gemma":0.010284895,"teacher_disagreement_score":0.008253806,"about_ca_system_score_codex":0.0003663068,"about_ca_system_score_gemma":0.0005126728,"threshold_uncertainty_score":0.5739213},"labels":[],"label_agreement":null},{"id":"W3186137785","doi":"10.2217/cer-2021-0059","title":"Review and estimation of disutility for joint health states of severe and nonsevere hypoglycemic events in diabetes","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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 Regina; Health Sciences Centre; Sunnybrook Health Science Centre; Toronto General Hospital; University Health Network; University of Toronto; Public Health Ontario","funders":"","keywords":"Medicine; Estimator; Hypoglycemia; Estimation; Diabetes mellitus; Construct (python library); Statistics; Econometrics; Endocrinology; Mathematics; Economics; Computer science","score_opus":0.5231676622634022,"score_gpt":0.5650792269013128,"score_spread":0.04191156463791068,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3186137785","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":[0.9213973,0.07239749,0.0017047878,0.0032917482,0.00004966744,0.0009987132,0.00013769457,0.0000011388516,0.000021425813],"genre_scores_gemma":[0.99302214,0.005409614,0.0013243444,0.00015280243,0.000017277884,0.000048414062,0.000010299787,0.000007490112,0.000007631953],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99403405,0.0027353184,0.002632489,0.00022961042,0.0001369174,0.00023160964],"domain_scores_gemma":[0.99116236,0.0060663414,0.0018905312,0.00015972211,0.00060873874,0.00011232374],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03657222,0.00010251445,0.0017244836,0.0003434639,0.00007498325,0.000011034166,0.00009505228,0.00004794743,0.0000174814],"category_scores_gemma":[0.004261962,0.000107690445,0.000081196056,0.00030357053,0.00013709442,0.00031143302,0.000060084738,0.00024685933,0.0000020212974],"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":[0.0007260103,0.0010206993,0.8758733,0.09553087,0.0005222415,0.0000014399515,0.006389908,0.004325853,0.00022523142,0.010331061,0.0023528,0.0027005428],"study_design_scores_gemma":[0.0023240051,0.0008932575,0.9269851,0.007532252,0.000013053088,0.000013097936,0.0010510986,0.009146494,0.00061009015,0.0510294,0.00025624436,0.0001459274],"about_ca_topic_score_codex":0.00010252802,"about_ca_topic_score_gemma":0.00004423022,"teacher_disagreement_score":0.08799862,"about_ca_system_score_codex":0.0003472785,"about_ca_system_score_gemma":0.0004605169,"threshold_uncertainty_score":0.99205166},"labels":[],"label_agreement":null},{"id":"W3188666308","doi":"10.2217/cer-2021-0150","title":"Economic burden of rivaroxaban and warfarin among nonvalvular atrial fibrillation patients with obesity and polypharmacy","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","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":"Group for Research in Decision Analysis","funders":"Janssen Scientific Affairs","keywords":"Medicine; Rivaroxaban; Polypharmacy; Atrial fibrillation; Warfarin; Obesity; Internal medicine; Cardiology; Intensive care medicine","score_opus":0.08725138394218204,"score_gpt":0.4219668102787172,"score_spread":0.33471542633653517,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3188666308","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":[0.9981001,0.0006706866,0.0002768431,0.00012946021,0.0000883901,0.000532401,0.000003686202,0.000003354554,0.0001950833],"genre_scores_gemma":[0.99907964,0.0001382479,0.00018777786,0.0000012563268,0.00052143424,4.046195e-7,0.000005967107,0.000008930067,0.00005632108],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99809843,0.00075669755,0.00033906297,0.00016773886,0.00048488454,0.00015316955],"domain_scores_gemma":[0.9976309,0.0010805946,0.0002452039,0.00010602626,0.00081098126,0.00012631631],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011811582,0.000112681344,0.0005961888,0.00028383415,0.00010426614,0.00004140805,0.000047099464,0.00004420072,0.00003741504],"category_scores_gemma":[0.00014545147,0.00008481151,0.00011071851,0.00021794348,0.00021133237,0.00029088903,0.000100183555,0.000227873,0.0000019812433],"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":[0.0036135735,0.000011162293,0.99114347,0.0001648882,0.00042421222,0.000016662927,0.00042963543,0.001493803,0.0013048465,0.00022512925,0.000015763126,0.0011568629],"study_design_scores_gemma":[0.006266951,0.0008375039,0.98450524,0.00022574185,0.00012079474,0.00001356831,0.00021539783,0.001180159,0.0060480386,0.00016162876,0.0003477127,0.00007725091],"about_ca_topic_score_codex":0.00006398271,"about_ca_topic_score_gemma":0.000007155134,"teacher_disagreement_score":0.0066382084,"about_ca_system_score_codex":0.00011152435,"about_ca_system_score_gemma":0.00018227712,"threshold_uncertainty_score":0.3458513},"labels":[],"label_agreement":null},{"id":"W3199921489","doi":"10.2217/cer-2021-0113","title":"Outcomes of patients with nonmetastatic gastric adenocarcinoma according to perioperative treatment strategy: a real-world, population-based study","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Gastric Cancer Management and Outcomes","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":"University of Alberta","funders":"","keywords":"Medicine; Perioperative; Hazard ratio; Internal medicine; Proportional hazards model; Gastrectomy; Cancer; Population; Radiation therapy; Surveillance, Epidemiology, and End Results; Oncology; Surgery; Cancer registry; Confidence interval","score_opus":0.10723860011433634,"score_gpt":0.4420697015525942,"score_spread":0.33483110143825784,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3199921489","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":[0.99642444,0.000078650846,0.0008769823,0.00011037045,0.00007356073,0.002110453,0.000013651802,0.000006420667,0.0003054451],"genre_scores_gemma":[0.9988118,0.000004381309,0.0007530304,0.000012688366,0.000045925062,0.00009029973,0.000018141905,0.00002167227,0.0002420899],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9957758,0.0014636937,0.00061732804,0.00031521314,0.0014418178,0.00038611834],"domain_scores_gemma":[0.99511737,0.0015450233,0.0003025604,0.00029803274,0.0025068058,0.0002301941],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010976346,0.000264664,0.001362496,0.0012205817,0.0001707767,0.0000718851,0.00015014526,0.000024868028,0.0000858077],"category_scores_gemma":[0.00016359259,0.00016986336,0.00018925538,0.002096587,0.00005862448,0.0001930489,0.0000694945,0.00032121484,0.000005162798],"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":[0.004649134,0.0027360055,0.98434716,0.00012081108,0.0011584419,0.00020858421,0.0010541722,0.0041943514,0.00016518093,0.0000465932,0.000009348952,0.0013102044],"study_design_scores_gemma":[0.010205702,0.013516739,0.9678325,0.00023605113,0.0003122338,0.000003747978,0.0064200796,0.00010263079,0.0012272835,0.000010248028,0.000010422884,0.00012233945],"about_ca_topic_score_codex":0.00039907967,"about_ca_topic_score_gemma":0.0004351099,"teacher_disagreement_score":0.016514651,"about_ca_system_score_codex":0.0010234106,"about_ca_system_score_gemma":0.0007865561,"threshold_uncertainty_score":0.6926827},"labels":[],"label_agreement":null},{"id":"W3202268091","doi":"10.2217/cer-2021-0203","title":"Letter in reply","year":2021,"lang":"en","type":"letter","venue":"Journal of Comparative Effectiveness Research","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine","score_opus":0.6833183359875381,"score_gpt":0.5728567734507829,"score_spread":0.11046156253675521,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3202268091","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":[0.048805706,0.010822691,0.0011350244,0.93019736,0.0019382294,0.0012905694,0.000096509415,0.000010425905,0.0057034916],"genre_scores_gemma":[0.1074788,0.00041769593,0.001039911,0.8720798,0.015998283,0.0003266342,0.00009332463,0.00015545335,0.0024100945],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9778986,0.013571578,0.006108766,0.000783171,0.0007240975,0.0009137609],"domain_scores_gemma":[0.97605246,0.01791649,0.003925855,0.000731272,0.0012117869,0.00016215231],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.08153039,0.00037831895,0.0037416634,0.0027324948,0.00017982881,0.00026139602,0.0008989087,0.0006216171,0.0005680815],"category_scores_gemma":[0.0062587494,0.00042563764,0.0004598732,0.0009977076,0.00028536437,0.0006579776,0.00017176673,0.006392697,0.0010084908],"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":[0.00009356285,0.00013016065,0.0072881407,0.001159424,0.00025435607,0.00048556534,0.0012339115,0.00021928028,0.000008473695,0.001784174,0.987323,0.000019988855],"study_design_scores_gemma":[0.0012695548,0.00032688896,0.016975928,0.0016413734,0.0000064313185,0.00007746352,0.00031713207,0.00016505811,0.00007531546,0.012793959,0.9659671,0.00038377926],"about_ca_topic_score_codex":0.00036419832,"about_ca_topic_score_gemma":0.00004546093,"teacher_disagreement_score":0.07527164,"about_ca_system_score_codex":0.0029128222,"about_ca_system_score_gemma":0.0011097874,"threshold_uncertainty_score":0.9998195},"labels":[],"label_agreement":null},{"id":"W3203002280","doi":"10.2217/cer-2021-0013","title":"Prognostic and predictive value of microsatellite instability status among patients with colorectal cancer","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Genetic factors in colorectal cancer","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":"University of Alberta","funders":"","keywords":"Medicine; Microsatellite instability; Internal medicine; Colorectal cancer; Chemotherapy; Oncology; Stage (stratigraphy); Proportional hazards model; Predictive value; Multivariate analysis; Cancer; Adenocarcinoma; Microsatellite","score_opus":0.04044524166403691,"score_gpt":0.38173063036785504,"score_spread":0.34128538870381814,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3203002280","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":[0.99564266,0.0023055999,0.00021700919,0.000021412116,0.00017651782,0.0012461784,0.00007255509,0.0000042619745,0.00031380862],"genre_scores_gemma":[0.99923253,0.00017810831,0.00041947528,0.000003876791,0.0000603006,0.00006840681,0.0000087237295,0.000018877174,0.000009679754],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.995917,0.0016854992,0.00045030183,0.0003119144,0.0012324728,0.0004027751],"domain_scores_gemma":[0.99067163,0.0024058558,0.00028607887,0.00016254293,0.006185911,0.00028796968],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0012632395,0.00019160747,0.0008415212,0.0002502758,0.00010068414,0.000030153495,0.00010288374,0.00009550311,0.000057152527],"category_scores_gemma":[0.0007580005,0.00013575554,0.000092288596,0.0007931814,0.0010954284,0.00016814,0.00012152015,0.00082164764,6.863359e-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":[0.011719572,0.0007900015,0.94886047,0.00051489746,0.0005167608,0.00004296809,0.0018202357,0.0004898912,0.03430679,0.000016685923,0.000016043328,0.0009056717],"study_design_scores_gemma":[0.0030387652,0.0054360237,0.78127,0.0009148941,0.0001017542,0.000021591144,0.00037549052,0.00015588295,0.20844007,0.00011066685,0.0000456415,0.00008921419],"about_ca_topic_score_codex":0.00007268604,"about_ca_topic_score_gemma":0.000062825486,"teacher_disagreement_score":0.17413327,"about_ca_system_score_codex":0.00058071065,"about_ca_system_score_gemma":0.0016557538,"threshold_uncertainty_score":0.55359507},"labels":[],"label_agreement":null},{"id":"W3206442947","doi":"10.2217/cer-2021-0121","title":"Real-world evidence: a practical toolbox for collecting health state utilities","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Université de Montréal","funders":"","keywords":"Toolbox; Credibility; Reimbursement; Medicine; Data science; Evidence-based medicine; Health care; Computer science; Alternative medicine","score_opus":0.8581222248983025,"score_gpt":0.6763503743707396,"score_spread":0.18177185052756284,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3206442947","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":[0.8557505,0.015042201,0.064886674,0.053860433,0.0015267209,0.0033264135,0.0002105837,0.000031614378,0.005364844],"genre_scores_gemma":[0.9855496,0.00062498695,0.010596373,0.0005750074,0.0006217375,0.0002321444,0.000006232312,0.000035619203,0.0017582639],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9855452,0.008728109,0.0040271166,0.0004939604,0.00045524584,0.00075034256],"domain_scores_gemma":[0.94320214,0.051037047,0.0027589723,0.0003458615,0.002300625,0.0003553828],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10857645,0.00019632616,0.0019763396,0.00093406934,0.0006826944,0.0002899311,0.000281809,0.00007653975,0.00015853113],"category_scores_gemma":[0.026780501,0.00021751145,0.00027855675,0.0010664436,0.0002067595,0.0011086803,0.00008129388,0.00085536897,0.00012215646],"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":[0.0056263367,0.0020761865,0.23704733,0.010652114,0.0020007442,0.00010825936,0.041125458,0.0040696347,0.00040491484,0.48603788,0.20973358,0.0011175647],"study_design_scores_gemma":[0.008544251,0.0063476833,0.5014164,0.008135447,0.000035356392,0.00027941912,0.038375825,0.013208451,0.0038750065,0.2341684,0.18434209,0.0012717033],"about_ca_topic_score_codex":0.00061439205,"about_ca_topic_score_gemma":0.0003854033,"teacher_disagreement_score":0.26436904,"about_ca_system_score_codex":0.0028850564,"about_ca_system_score_gemma":0.004839119,"threshold_uncertainty_score":0.98141736},"labels":[],"label_agreement":null},{"id":"W3209536742","doi":"10.2217/cer-2021-0018","title":"Meta-analyses of deflazacort versus prednisone/prednisolone in patients with nonsense mutation Duchenne muscular dystrophy","year":2021,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Muscle Physiology and Disorders","field":"Biochemistry, Genetics and Molecular Biology","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":"London Health Sciences Centre; Western University","funders":"National Institute for Health and Care Research; PTC Therapeutics","keywords":"Deflazacort; Medicine; Prednisone; Nonsense mutation; Duchenne muscular dystrophy; Prednisolone; Muscular dystrophy; Internal medicine; Placebo; Pediatrics; Physical therapy; Mutation; Genetics; Pathology","score_opus":0.35404409198600484,"score_gpt":0.5138161215238461,"score_spread":0.1597720295378413,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3209536742","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":[0.2793907,0.71935314,0.000400091,0.000002611524,0.000086681786,0.0006793549,0.00003398993,0.0000011539242,0.00005227064],"genre_scores_gemma":[0.5380819,0.46115416,0.00025988542,0.000002104128,0.000081010825,0.00011606148,0.00025789734,0.00003791777,0.000009062615],"study_design_codex":"meta_analysis","study_design_gemma":"not_applicable","domain_scores_codex":[0.9913456,0.006258987,0.00088545104,0.0004716689,0.00069382647,0.00034447527],"domain_scores_gemma":[0.9954694,0.0010544363,0.000873174,0.00045440122,0.00204295,0.00010562176],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0018166032,0.00040067363,0.0030876107,0.0005626178,0.000079944155,0.000023030987,0.00038709634,0.00031932336,0.000024424155],"category_scores_gemma":[0.0003949305,0.000269424,0.0010339275,0.0007753973,0.00038956734,0.000018827233,0.00015210643,0.0007638673,0.0000028020133],"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":[0.17795748,0.023494696,0.0050262255,0.06454468,0.44919363,0.0013738847,0.0017025033,0.033528246,0.040673446,0.00043081478,0.0014257255,0.2006487],"study_design_scores_gemma":[0.12854017,0.17191747,0.13472597,0.054645307,0.14932483,0.00047904716,0.002474462,0.00016181705,0.08663354,0.00088365836,0.26099938,0.009214339],"about_ca_topic_score_codex":0.000024053448,"about_ca_topic_score_gemma":0.000023444261,"teacher_disagreement_score":0.2998688,"about_ca_system_score_codex":0.000096321455,"about_ca_system_score_gemma":0.0009646756,"threshold_uncertainty_score":0.9999758},"labels":[],"label_agreement":null},{"id":"W3211800240","doi":"10.2217/cer-2021-0221","title":"Palbociclib versus abemaciclib in HR+/HER2- advanced breast cancer: an indirect comparison of patient-reported end points","year":2021,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Breast Cancer Therapies","field":"Medicine","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Fulvestrant; Nausea; Breast cancer; Quality of life (healthcare); Internal medicine; Oncology; Palbociclib; Cancer; Metastatic breast cancer; Estrogen receptor","score_opus":0.13336599440221616,"score_gpt":0.5031046785495277,"score_spread":0.3697386841473115,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3211800240","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":[0.99289787,0.0050487677,0.00020702074,0.00025803663,0.00045959288,0.0006855835,0.0000330168,0.000014932761,0.00039515144],"genre_scores_gemma":[0.9986938,0.00016882205,0.0008228176,0.000014436327,0.00015347132,0.00007309835,0.000013435863,0.000041629497,0.000018450697],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99316955,0.0028237423,0.0011924157,0.00045805454,0.0017434215,0.00061284564],"domain_scores_gemma":[0.9923293,0.0022208982,0.000838847,0.00045130213,0.0038049675,0.00035470873],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0019679747,0.00031131622,0.0018712889,0.0007411913,0.00012207434,0.000033037206,0.00026843548,0.00014017698,0.0002968155],"category_scores_gemma":[0.0003321359,0.00026908919,0.00023239397,0.0018237956,0.00048603542,0.0005822902,0.000116465155,0.0014401878,0.00000477839],"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":[0.09122621,0.004560722,0.47725222,0.00061468146,0.0016330197,0.0011097838,0.008230328,0.0051268814,0.2541439,0.00015728392,0.000035324258,0.15590966],"study_design_scores_gemma":[0.012139636,0.0030490153,0.60772955,0.0016362686,0.000071451934,0.00034588636,0.0058679515,0.00020969917,0.36806256,0.00029166156,0.00039767913,0.00019865869],"about_ca_topic_score_codex":0.000110590525,"about_ca_topic_score_gemma":0.00025276837,"teacher_disagreement_score":0.15571101,"about_ca_system_score_codex":0.0010096803,"about_ca_system_score_gemma":0.0015327646,"threshold_uncertainty_score":0.99997616},"labels":[],"label_agreement":null},{"id":"W4206181287","doi":"10.2217/cer-2021-0139","title":"SPECT/CT versus planar imaging to determine treatment strategy for non-small-cell lung cancer: a cost–effectiveness analysis","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Lung Cancer Diagnosis and Treatment","field":"Medicine","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":"London Health Sciences Centre; St. Joseph's Hospital; Western University","funders":"GE Healthcare","keywords":"Scintigraphy; Lung cancer; Lung; Lung volumes; Pneumonectomy; Quality-adjusted life year; Predictive value of tests","score_opus":0.14473501541538358,"score_gpt":0.4838716371160943,"score_spread":0.33913662170071074,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206181287","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":[0.98798454,0.002803788,0.0027010345,0.00027418687,0.00041796983,0.005120199,0.0001716277,0.00001299361,0.0005136496],"genre_scores_gemma":[0.9945598,0.00018633746,0.0001715396,0.000018848656,0.00026431034,0.0046345554,0.000034857792,0.000040956274,0.00008877647],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9953421,0.0017504921,0.0005209027,0.0005717288,0.0010578671,0.00075695605],"domain_scores_gemma":[0.9929287,0.004966098,0.00029068606,0.0003727923,0.0009895004,0.00045226185],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0028189432,0.00037131156,0.0015671578,0.0014002356,0.00063567556,0.00009502543,0.0002934955,0.0000179511,0.00015191897],"category_scores_gemma":[0.000057398873,0.00029013384,0.0007088377,0.0019609523,0.00008726951,0.0001290398,0.00011632798,0.0005316539,0.0000046082814],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.08615815,0.0074423407,0.6953136,0.0014071477,0.020927958,0.0044293976,0.002106525,0.14603801,0.013946236,0.0001570719,0.001528042,0.020545565],"study_design_scores_gemma":[0.055238087,0.040107973,0.6397113,0.0010004814,0.0081270905,0.0003481405,0.0030759152,0.021715121,0.22540146,0.00009307859,0.004441726,0.0007395935],"about_ca_topic_score_codex":0.00090185605,"about_ca_topic_score_gemma":0.00028631894,"teacher_disagreement_score":0.21145523,"about_ca_system_score_codex":0.008197388,"about_ca_system_score_gemma":0.0014078085,"threshold_uncertainty_score":0.99995506},"labels":[],"label_agreement":null},{"id":"W4206611799","doi":"10.2217/cer-2021-0258","title":"Early mobilization in enhanced recovery after surgery pathways: current evidence and recent advancements","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Enhanced Recovery After Surgery","field":"Medicine","cited_by":298,"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 Calgary","funders":"Canadian Institutes of Health Research","keywords":"Prehabilitation; Medicine; Mobilization; Perioperative; Intensive care medicine; Physical medicine and rehabilitation; Adverse effect; Physical therapy; Surgery; Internal medicine","score_opus":0.18433600964542252,"score_gpt":0.43885297524285716,"score_spread":0.25451696559743464,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4206611799","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":[0.9794026,0.016487427,0.002223363,0.00007123215,0.00082636194,0.00091023656,0.000006373104,0.000007250595,0.00006515141],"genre_scores_gemma":[0.9932548,0.006029916,0.00007889513,0.000019066969,0.000109704866,0.00045590661,0.0000038746443,0.00002009383,0.000027782946],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99292946,0.004071552,0.0007426971,0.00036557854,0.0014577504,0.00043296715],"domain_scores_gemma":[0.9922876,0.00588849,0.00032633042,0.00023300716,0.001082988,0.00018163573],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.007485966,0.00019105739,0.00085246866,0.0012511679,0.00015190273,0.000045236222,0.00013673848,0.0000446339,0.00016205691],"category_scores_gemma":[0.0010031404,0.00017070945,0.00014882418,0.0014416457,0.00014325595,0.00075342105,0.00022814167,0.0013374102,0.0000064641945],"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":[0.066410266,0.0029722487,0.6609496,0.0014118903,0.0002867617,0.00073779264,0.0036236949,0.0036606777,0.11302229,0.000029543819,0.0003855899,0.14650965],"study_design_scores_gemma":[0.001575237,0.0030182616,0.93640345,0.004019791,0.000024647838,0.00010112883,0.00053740846,0.00012628776,0.051652513,0.00079732813,0.001521878,0.00022206747],"about_ca_topic_score_codex":0.00000875083,"about_ca_topic_score_gemma":0.0000062314225,"teacher_disagreement_score":0.27545384,"about_ca_system_score_codex":0.001129307,"about_ca_system_score_gemma":0.0007267113,"threshold_uncertainty_score":0.6961329},"labels":[],"label_agreement":null},{"id":"W4210548176","doi":"10.2217/cer-2021-0133","title":"A real-world, population-based study for the outcomes of patients with metastatic colorectal cancer to the liver with distant lymph node metastases treated with metastasectomy","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Hepatocellular Carcinoma Treatment and Prognosis","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":"University of Alberta","funders":"","keywords":"Metastasectomy; Medicine; Hazard ratio; Colorectal cancer; Lymph node; Propensity score matching; Internal medicine; Oncology; Cancer; Surgery; Confidence interval","score_opus":0.12725708432665986,"score_gpt":0.3983784441271667,"score_spread":0.2711213598005069,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4210548176","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":[0.9912141,0.0003384529,0.0010001806,0.00023256792,0.000028089622,0.0068140943,0.00035317638,0.000006889622,0.000012435015],"genre_scores_gemma":[0.99662936,0.0000038167263,0.00045143097,0.000017736724,0.000020388941,0.0027545462,0.000062302315,0.000035525445,0.000024872399],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99407816,0.0030022147,0.00041861096,0.00028051788,0.0018742911,0.0003461804],"domain_scores_gemma":[0.99183637,0.005511683,0.00039217345,0.00029070588,0.0018218039,0.00014727708],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002517203,0.00028022897,0.0010385856,0.00051106076,0.00063718786,0.000048231737,0.0002713372,0.000011346455,0.00009259832],"category_scores_gemma":[0.000074088784,0.000108962595,0.00017851595,0.0016757012,0.00015307871,0.00010353365,0.000059793623,0.0003497448,6.3649634e-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":[0.051080853,0.0027773571,0.9323078,0.00009368199,0.0054410207,0.00011897674,0.0010274785,0.006678756,0.000060573213,0.000029590221,0.000028050945,0.00035589215],"study_design_scores_gemma":[0.012275486,0.016111294,0.96286315,0.0001189934,0.0022757566,0.000007612848,0.0017012146,0.0014667433,0.0029422033,0.0000042604966,0.00009924995,0.00013402726],"about_ca_topic_score_codex":0.002713848,"about_ca_topic_score_gemma":0.0017620976,"teacher_disagreement_score":0.038805366,"about_ca_system_score_codex":0.0006370279,"about_ca_system_score_gemma":0.0006956552,"threshold_uncertainty_score":0.49007982},"labels":[],"label_agreement":null},{"id":"W4212770753","doi":"10.2217/cer-2021-0261","title":"Systematic literature review of evidence in amyloid light-chain amyloidosis","year":2022,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Amyloidosis: Diagnosis, Treatment, Outcomes","field":"Biochemistry, Genetics and Molecular Biology","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Systematic review; Randomized controlled trial; Observational study; Amyloidosis; MEDLINE; Intensive care medicine; Quality of life (healthcare); Pathology; Nursing","score_opus":0.22678034929922558,"score_gpt":0.49261936796091677,"score_spread":0.2658390186616912,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4212770753","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":[0.00045313785,0.99485964,0.00003172523,0.00007103532,0.00028249892,0.0041014897,0.000049050355,0.0000040213636,0.00014742311],"genre_scores_gemma":[0.002241748,0.9956915,0.00006855573,0.000026130665,0.00018199162,0.0015513541,0.000052102907,0.000071297734,0.00011533204],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9746157,0.019893337,0.0024383347,0.0007024646,0.0017575525,0.00059263123],"domain_scores_gemma":[0.98787504,0.007235661,0.002342537,0.0009706323,0.0013649841,0.00021113163],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.012826962,0.00072628254,0.0058980472,0.0013786108,0.00014729428,0.000076710145,0.0015575517,0.00038899036,0.00008413259],"category_scores_gemma":[0.0048122825,0.00046703644,0.001666909,0.0025345741,0.00021419843,0.00004083296,0.0006385536,0.0018846161,0.000014979778],"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":[0.00027314932,0.0006601106,0.00021635978,0.9842287,0.0018110733,0.00024014669,0.00021648187,0.000013591848,0.000464196,0.000049474977,0.0014319388,0.010394778],"study_design_scores_gemma":[0.000519049,0.0018024001,0.00010746262,0.8534076,0.0006453835,0.0002979372,0.00010063211,0.0000010907694,0.0015965964,0.00001783711,0.14113306,0.00037095867],"about_ca_topic_score_codex":0.000008856547,"about_ca_topic_score_gemma":0.0000081298185,"teacher_disagreement_score":0.13970111,"about_ca_system_score_codex":0.0008411978,"about_ca_system_score_gemma":0.0013111561,"threshold_uncertainty_score":0.99977815},"labels":[],"label_agreement":null},{"id":"W4212996296","doi":"10.2217/cer-2021-0163","title":"Patterns of emergency department visits preceding colorectal cancer diagnosis: a population-based study","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","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 Calgary; University of Alberta","funders":"","keywords":"Medicine; Emergency department; Context (archaeology); Odds ratio; Colorectal cancer; Comorbidity; Logistic regression; Population; Internal medicine; Charlson comorbidity index; Ambulatory; Cancer; Pediatrics","score_opus":0.11538822911581846,"score_gpt":0.4640978761601192,"score_spread":0.34870964704430074,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4212996296","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":[0.9965995,0.0007826867,0.00044318833,0.000062365994,0.00057558,0.0014359318,0.000030455063,0.000010816007,0.00005950137],"genre_scores_gemma":[0.99818844,0.000022443777,0.000024102716,0.0000032698156,0.00017526286,0.0015486872,0.0000062340687,0.000015713646,0.000015822337],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99499476,0.0023836277,0.00053336495,0.00022627333,0.0016077167,0.0002542504],"domain_scores_gemma":[0.99714315,0.0010515227,0.00036943573,0.00012488561,0.0011767772,0.00013422155],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029082561,0.00014026831,0.0006260345,0.00055195176,0.00054346723,0.000024543806,0.0001618655,0.000030859963,0.00043088428],"category_scores_gemma":[0.0001816605,0.000120543766,0.00019545281,0.0009543253,0.000014497667,0.00010318298,0.00010819681,0.0007599796,8.7968385e-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":[0.019129422,0.001650613,0.9624977,0.0001435527,0.00036832003,0.000040634517,0.0009637379,0.010214847,0.0031206515,0.0000052255377,0.000099202145,0.0017661038],"study_design_scores_gemma":[0.002714381,0.02499012,0.9369172,0.00024076604,0.00009385398,0.000016142083,0.0015387627,0.0011448028,0.032130376,0.000030436657,0.00008982581,0.000093364164],"about_ca_topic_score_codex":0.0008070495,"about_ca_topic_score_gemma":0.0004086685,"teacher_disagreement_score":0.029009722,"about_ca_system_score_codex":0.0010614442,"about_ca_system_score_gemma":0.00038949496,"threshold_uncertainty_score":0.4915632},"labels":[],"label_agreement":null},{"id":"W4221007314","doi":"10.2217/cer-2022-0043","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 6","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Real world evidence; Medicine; Reimbursement; Health technology; Comparative effectiveness research; Health economics; Health care; Alternative medicine; Public health; Economic growth; Nursing; Pathology","score_opus":0.7989013653416447,"score_gpt":0.6486979193858494,"score_spread":0.15020344595579527,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4221007314","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":[0.96039647,0.003612536,0.0064121825,0.022948325,0.00089139945,0.0038581188,0.00011202338,0.000012889448,0.0017560724],"genre_scores_gemma":[0.9936138,0.0002358107,0.0045462623,0.00019121925,0.0000951776,0.0008376131,0.0000066883085,0.000022625825,0.00045082596],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98890597,0.004774955,0.004693758,0.0004771296,0.00051235384,0.0006358634],"domain_scores_gemma":[0.9859572,0.009529845,0.00336443,0.00033583745,0.0006291482,0.00018352571],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.099838056,0.00016720893,0.0017115027,0.0031839993,0.00049864076,0.000047758716,0.0006105111,0.000060171133,0.00008169986],"category_scores_gemma":[0.003527029,0.00020096966,0.00010523166,0.0019881604,0.00010556548,0.0004939849,0.00025985716,0.0009189725,0.000025701314],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0015084299,0.0008547203,0.5782697,0.0019819648,0.00041003525,0.000008744437,0.014961957,0.014552401,0.00023106668,0.37178543,0.013906318,0.0015291933],"study_design_scores_gemma":[0.009003266,0.010735609,0.6720561,0.0076577337,0.00001651795,0.00003892617,0.04909061,0.005483627,0.000668607,0.09431937,0.14996064,0.0009690074],"about_ca_topic_score_codex":0.00078663806,"about_ca_topic_score_gemma":0.00049534615,"teacher_disagreement_score":0.27746606,"about_ca_system_score_codex":0.0055813254,"about_ca_system_score_gemma":0.0019551998,"threshold_uncertainty_score":0.99823606},"labels":[],"label_agreement":null},{"id":"W4225010633","doi":"10.2217/cer-2021-0205","title":"Sex-based differences in the outcomes of patients with lung carcinoids","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Neuroendocrine Tumor Research Advances","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":"University of Alberta","funders":"","keywords":"Medicine; Proportional hazards model; Internal medicine; Hazard ratio; Lung cancer; Epidemiology; Lung; Incidence (geometry); Population; Multivariate analysis; Carcinoid tumors; Gastroenterology; Oncology; Confidence interval","score_opus":0.08334786066464446,"score_gpt":0.43232584226208215,"score_spread":0.3489779815974377,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4225010633","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":[0.9976508,0.0002446914,0.00007460108,0.00038959878,0.000047396574,0.001268844,0.000014903144,0.0000029841137,0.00030616554],"genre_scores_gemma":[0.9996924,0.0000044786016,0.000056718312,0.000024034325,0.000021339514,0.00014704285,0.0000033160575,0.000011976598,0.000038733084],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9903051,0.005510519,0.00040200702,0.00018472938,0.0032031962,0.00039443278],"domain_scores_gemma":[0.99275166,0.0056023267,0.00022389347,0.00026172033,0.0010558645,0.0001045155],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00304039,0.0001477547,0.0008111287,0.00083883724,0.00020504174,0.000023324088,0.0005276064,0.0000064597416,0.000075907235],"category_scores_gemma":[0.0003780765,0.00007762103,0.00011807535,0.0012254484,0.00040303264,0.00013540742,0.00012872358,0.0015495521,8.3491346e-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":[0.005982873,0.0015446577,0.9894735,0.00016773566,0.00012438724,0.0005947586,0.0006046757,0.0006293783,0.0005793391,0.00008754609,0.00008789886,0.0001232201],"study_design_scores_gemma":[0.0049451073,0.007854161,0.97910017,0.00009166039,0.000021178274,0.000055450764,0.0017800278,0.00023512116,0.005625559,0.000120057884,0.00010469468,0.00006683789],"about_ca_topic_score_codex":0.000044974666,"about_ca_topic_score_gemma":0.0000138324485,"teacher_disagreement_score":0.010373386,"about_ca_system_score_codex":0.00028540049,"about_ca_system_score_gemma":0.00055319,"threshold_uncertainty_score":0.6732119},"labels":[],"label_agreement":null},{"id":"W4229044080","doi":"10.2217/cer-2021-0314","title":"Economic burden of drug overdose deaths before and during the COVID-19 pandemic in the USA","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Opioid Use Disorder Treatment","field":"Medicine","cited_by":10,"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; Pandemic; Coronavirus disease 2019 (COVID-19); Drug overdose; Quarter (Canadian coin); Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); 2019-20 coronavirus outbreak; Emergency medicine; Medical emergency; Environmental health; Poison control; Virology; Infectious disease (medical specialty); Disease; Outbreak; Geography; Internal medicine","score_opus":0.07913024373929181,"score_gpt":0.4399704730732,"score_spread":0.36084022933390814,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4229044080","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":[0.99611217,0.0015344312,0.000013090275,0.0010292798,0.00005734293,0.0010271958,0.000014283803,0.0000023559408,0.00020985675],"genre_scores_gemma":[0.9995266,0.00015185807,0.0000100975985,0.000031979635,0.00007872339,0.00012185202,0.0000018936157,0.000009811007,0.00006714775],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99505764,0.003432396,0.00036129483,0.0001668846,0.0007335901,0.0002481804],"domain_scores_gemma":[0.99621016,0.0031030548,0.00019596907,0.00023862593,0.0001465663,0.0001056271],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0053457073,0.00012428884,0.00048636788,0.00036416127,0.00036323647,0.000026720183,0.00032727828,0.00002394815,0.00010272825],"category_scores_gemma":[0.0001689203,0.000068543515,0.00011067002,0.00035494126,0.0002812774,0.00009330481,0.0002064529,0.0009688978,0.000003344706],"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":[0.0030846493,0.00031524707,0.9495882,0.000253343,0.00022739741,0.00014364524,0.03507507,0.008791722,0.0017259559,0.00045847977,0.000114248476,0.00022204776],"study_design_scores_gemma":[0.0044316985,0.0013289765,0.97460663,0.00008626012,0.000040352388,0.0006416311,0.01582423,0.0003872186,0.0005935304,0.0010122674,0.0009891563,0.000058058045],"about_ca_topic_score_codex":0.0009688176,"about_ca_topic_score_gemma":0.00038279325,"teacher_disagreement_score":0.025018426,"about_ca_system_score_codex":0.0012096897,"about_ca_system_score_gemma":0.0007611739,"threshold_uncertainty_score":0.42094326},"labels":[],"label_agreement":null},{"id":"W4229451358","doi":"10.2217/cer-2021-0278","title":"Channeling effects in the prescription of new therapies: the case of emicizumab for hemophilia A","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Hemophilia Treatment and Research","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":false,"ca_institutions":"Roche (Canada)","funders":"F. Hoffmann-La Roche","keywords":"Medicine; Medical prescription; Health insurance; Health care; Pediatrics; Internal medicine; Pharmacology","score_opus":0.1627061552598147,"score_gpt":0.46460012567932396,"score_spread":0.30189397041950927,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4229451358","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":[0.99156207,0.0035499686,0.0006825655,0.001083525,0.000114176175,0.0028521952,0.000007208824,0.0000020297032,0.00014623908],"genre_scores_gemma":[0.9993082,0.00006484801,0.00010593272,0.000012892749,0.00017232391,0.00022628433,0.0000024898777,0.000012796292,0.000094181276],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9940734,0.0039438335,0.00047279324,0.00015519532,0.0010380951,0.00031668658],"domain_scores_gemma":[0.9905355,0.007988975,0.00024131102,0.00027214646,0.00088629033,0.000075761905],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009357636,0.0001295923,0.00063694717,0.0005579021,0.0003085556,0.000017980348,0.000374705,0.00004410384,0.00002826068],"category_scores_gemma":[0.0005299762,0.00006508387,0.00025708624,0.0010400463,0.00026348815,0.00009143948,0.00010442079,0.0009931519,7.001825e-7],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.113750964,0.008831084,0.054235004,0.00787422,0.004248478,0.0065128333,0.23894048,0.010919829,0.4814745,0.011964246,0.0060983393,0.055150025],"study_design_scores_gemma":[0.05690793,0.08102197,0.24576001,0.0043569976,0.00064198324,0.012069574,0.11795925,0.019181494,0.40373904,0.05432313,0.003469401,0.00056921126],"about_ca_topic_score_codex":0.00016279043,"about_ca_topic_score_gemma":0.000018452036,"teacher_disagreement_score":0.19152501,"about_ca_system_score_codex":0.0002716916,"about_ca_system_score_gemma":0.0006053351,"threshold_uncertainty_score":0.43148065},"labels":[],"label_agreement":null},{"id":"W4280554311","doi":"10.2217/cer-2022-0040","title":"Indirect treatment comparison of anifrolumab efficacy versus belimumab in adults with systemic lupus erythematosus","year":2022,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Systemic Lupus Erythematosus Research","field":"Medicine","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":"EVERSANA (Canada)","funders":"National Institute for Health and Care Research; AstraZeneca","keywords":"Medicine; Belimumab; Lupus erythematosus; Monoclonal; Immunology; Internal medicine; Dermatology; Monoclonal antibody; Antibody; B-cell activating factor","score_opus":0.21922028441166244,"score_gpt":0.48646745943190944,"score_spread":0.267247175020247,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4280554311","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":[0.24520391,0.74654555,0.000013272513,0.000005473415,0.0005126998,0.0065102517,0.00003764852,0.000017861154,0.0011533293],"genre_scores_gemma":[0.35060072,0.64818096,0.000054079428,2.7250204e-7,0.0001623031,0.0007314655,0.000043443244,0.00012593117,0.00010082611],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9773105,0.012819878,0.003090893,0.00091078767,0.0046927663,0.0011751872],"domain_scores_gemma":[0.97210926,0.022040026,0.0024124992,0.0011144765,0.00182129,0.00050247344],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0076024462,0.0009951423,0.01146686,0.0046318974,0.00025674948,0.00007160196,0.0010844906,0.00039164722,0.00017345228],"category_scores_gemma":[0.00084389554,0.0006469175,0.0012497002,0.0041823993,0.0006096663,0.0001939825,0.0003512572,0.0037716753,0.00004450014],"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":true,"study_design_scores_codex":[0.19259575,0.01696336,0.050826237,0.37374428,0.016695313,0.008105159,0.027785309,0.0007013751,0.000050332284,0.00018291366,0.00047772803,0.31187224],"study_design_scores_gemma":[0.09748426,0.0809094,0.0009946158,0.5557957,0.0028689015,0.12641744,0.01953748,0.0002763954,0.00038286456,0.000017072967,0.11393973,0.0013760966],"about_ca_topic_score_codex":0.0001366384,"about_ca_topic_score_gemma":0.00010400579,"teacher_disagreement_score":0.31049615,"about_ca_system_score_codex":0.008673818,"about_ca_system_score_gemma":0.006239344,"threshold_uncertainty_score":0.9995982},"labels":[],"label_agreement":null},{"id":"W4283806096","doi":"10.2217/cer-2021-0239","title":"Impact of perioperative chemotherapy on survival outcomes among patients with metastatic colorectal cancer to the liver","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Hepatocellular Carcinoma Treatment and Prognosis","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":"University of Alberta","funders":"","keywords":"Medicine; Colorectal cancer; Chemotherapy; Internal medicine; Oncology; Perioperative; Oxaliplatin; Proportional hazards model; Adjuvant; Cancer; Adjuvant chemotherapy; Surgery; Breast cancer","score_opus":0.14820642662481062,"score_gpt":0.4268300897564154,"score_spread":0.2786236631316048,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283806096","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":[0.99722207,0.00022425929,0.000055445325,0.00016674912,0.00008093054,0.0019500859,0.000053811535,0.000003295304,0.0002433383],"genre_scores_gemma":[0.999496,0.000016271415,0.000042864787,0.000016686194,0.0000402133,0.00030341596,0.000006589311,0.000021162103,0.000056838668],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9948161,0.00285971,0.00029184675,0.00018831546,0.0015541612,0.00028985948],"domain_scores_gemma":[0.9962925,0.0017135211,0.00020066393,0.00015951307,0.001464503,0.00016934413],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0018822905,0.00020109105,0.0008466712,0.00038166664,0.00031886224,0.000028752253,0.0002152237,0.000020426727,0.00051231845],"category_scores_gemma":[0.00008464498,0.00009699865,0.0002959416,0.00081889477,0.000207929,0.000100794336,0.000072347815,0.0006860691,0.000004879023],"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":[0.0163226,0.0014732761,0.97189045,0.000030838968,0.0019920287,0.00007526752,0.0050276765,0.0013706535,0.00068591424,0.000029383275,0.0000995225,0.0010023802],"study_design_scores_gemma":[0.005140683,0.023048047,0.9638712,0.0001063477,0.0001028115,0.00000796701,0.00069677003,0.00038671095,0.006453314,0.000004023351,0.00008942621,0.00009273458],"about_ca_topic_score_codex":0.000440107,"about_ca_topic_score_gemma":0.000053654065,"teacher_disagreement_score":0.021574771,"about_ca_system_score_codex":0.00081484485,"about_ca_system_score_gemma":0.0005737108,"threshold_uncertainty_score":0.5609528},"labels":[],"label_agreement":null},{"id":"W4288010068","doi":"10.2217/cer-2022-0041","title":"A plain language summary of the likelihood of symptom relief for patients taking fesoterodine for overactive bladder","year":2022,"lang":"en","type":"letter","venue":"Journal of Comparative Effectiveness Research","topic":"Bladder and Urothelial Cancer Treatments","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":"Pfizer (Canada); Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre; University of Alberta","funders":"","keywords":"Overactive bladder; Medicine; Urination; Accidental; Intensive care medicine; Urology; Alternative medicine; Urinary system; Internal medicine; Pathology","score_opus":0.0675549787563054,"score_gpt":0.4167074654336932,"score_spread":0.3491524866773878,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4288010068","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":[0.95044243,0.0043312837,0.0010829542,0.021396508,0.0018691897,0.015596861,0.0034285153,0.000014442038,0.0018378107],"genre_scores_gemma":[0.993774,0.0000672365,0.0002972713,0.002661887,0.0016163064,0.0007643833,0.00021718729,0.00010589145,0.00049585593],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9944313,0.0019786377,0.000835623,0.0003652203,0.0018626951,0.0005265295],"domain_scores_gemma":[0.9875705,0.007356707,0.001544052,0.00043307204,0.0030073542,0.000088281755],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.002113318,0.00035756727,0.0016617774,0.0006276603,0.00024303724,0.000022334734,0.0005225624,0.0002794287,0.00011389481],"category_scores_gemma":[0.0008033642,0.00023478136,0.0008699782,0.00052023807,0.00027283057,0.00013482998,0.00019898136,0.0023047614,7.5169845e-7],"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":[0.09222543,0.006384834,0.029348549,0.011162103,0.014620975,0.000221861,0.012548197,0.00017116465,0.02110054,0.00008782406,0.8070715,0.005057034],"study_design_scores_gemma":[0.07587979,0.092430666,0.29816198,0.01610601,0.0042250077,0.0001878684,0.0036381315,0.00028603812,0.23124734,0.00288179,0.27363676,0.0013186187],"about_ca_topic_score_codex":0.00011914264,"about_ca_topic_score_gemma":0.000007664637,"teacher_disagreement_score":0.53343475,"about_ca_system_score_codex":0.0008434035,"about_ca_system_score_gemma":0.0012207358,"threshold_uncertainty_score":0.99999696},"labels":[],"label_agreement":null},{"id":"W4292583104","doi":"10.2217/cer-2022-0074","title":"Clinical manifestations and healthcare utilization before diagnosis of transthyretin amyloidosis","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Amyloidosis: Diagnosis, Treatment, Outcomes","field":"Biochemistry, Genetics and Molecular Biology","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 Calgary","funders":"","keywords":"Medicine; Transthyretin; Amyloidosis; Comorbidity; Demographics; Pediatrics; Internal medicine","score_opus":0.16123731951039613,"score_gpt":0.4917790906481361,"score_spread":0.33054177113773997,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4292583104","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":[0.99427074,0.0036223822,0.0006205823,0.00058849715,0.00015290469,0.0005870807,0.00008119321,0.0000031992624,0.00007341218],"genre_scores_gemma":[0.9975677,0.001708562,0.00025970823,0.00003299314,0.00009502681,0.00025520934,0.00003721072,0.000019142737,0.000024425986],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9939511,0.0042200764,0.00063436915,0.00028871137,0.0006487641,0.00025699896],"domain_scores_gemma":[0.9971102,0.0012842338,0.0003699826,0.00021094465,0.00087092846,0.0001537418],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031763955,0.00015555433,0.00051758345,0.0002877227,0.00034021214,0.000020719488,0.00027118463,0.00008536378,0.000048801863],"category_scores_gemma":[0.00032533504,0.00013882178,0.00021536936,0.00041495767,0.0003416195,0.000017222477,0.00017802809,0.00048200914,0.000001226979],"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":[0.0011295712,0.0009231315,0.98213434,0.00013635641,0.00038624185,0.000024840514,0.00083784154,0.00040239218,0.005636255,0.000543339,0.00067576725,0.0071699456],"study_design_scores_gemma":[0.0020636609,0.0074078375,0.9355983,0.00010401648,0.00006140414,0.000068134905,0.0019398805,0.000061477665,0.049153637,0.00044092225,0.0029695802,0.00013114634],"about_ca_topic_score_codex":0.000047092482,"about_ca_topic_score_gemma":0.00006652684,"teacher_disagreement_score":0.046536013,"about_ca_system_score_codex":0.00011538764,"about_ca_system_score_gemma":0.0003102827,"threshold_uncertainty_score":0.5660988},"labels":[],"label_agreement":null},{"id":"W4297457303","doi":"10.2217/cer-2021-0274","title":"Developing a cross-validation tool for evaluating economic evidence in rapid literature reviews","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","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":"McGill University; Public Health Ontario","funders":"","keywords":"Credibility; Medicine; Economic evaluation; Quality (philosophy); Systematic review; Management science; Software deployment; Evidence-based medicine; Economic model; Data science; Risk analysis (engineering); MEDLINE; Computer science; Economics; Alternative medicine; Macroeconomics; Pathology","score_opus":0.9502567644813614,"score_gpt":0.7321332976015382,"score_spread":0.2181234668798232,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4297457303","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":[0.94474095,0.016432043,0.034846116,0.00028910724,0.00041593434,0.0030867476,0.000011524918,8.740594e-7,0.00017672799],"genre_scores_gemma":[0.99274105,0.00019988496,0.005710313,0.00002437116,0.00015462456,0.00071976666,0.0000032763783,0.000009713766,0.0004369988],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.8122546,0.16636868,0.011726974,0.0010146681,0.008092532,0.0005425495],"domain_scores_gemma":[0.9054046,0.078987956,0.007147266,0.0013934639,0.0069463295,0.000120330114],"candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.59821975,0.00023362406,0.0036171442,0.0015772887,0.00061025436,0.0015813013,0.0020714516,0.000050789575,0.002114521],"category_scores_gemma":[0.07769713,0.0001311321,0.0012914219,0.0027303395,0.000086114924,0.0011564848,0.0003356662,0.0009022272,0.00020170037],"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":[0.0073534483,0.0006875163,0.4046917,0.0035704363,0.0013511876,0.000101144986,0.027996907,0.24461621,0.027457858,0.023333106,0.040809777,0.2180307],"study_design_scores_gemma":[0.005148796,0.004878019,0.5764697,0.007773507,0.00020432037,0.00034240467,0.0038952897,0.12234655,0.008233849,0.09085569,0.17873593,0.0011159178],"about_ca_topic_score_codex":0.0000032688074,"about_ca_topic_score_gemma":0.0000047035096,"teacher_disagreement_score":0.5205226,"about_ca_system_score_codex":0.0011817785,"about_ca_system_score_gemma":0.00090695807,"threshold_uncertainty_score":0.99945515},"labels":[],"label_agreement":null},{"id":"W4308668157","doi":"10.2217/cer-2022-0160","title":"Healthcare and economic burden of anticholinergic use in adults with overactive bladder: a systematic literature review","year":2022,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Urinary Bladder and Prostate Research","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":"EVERSANA (Canada)","funders":"","keywords":"Anticholinergic; Medicine; Overactive bladder; Intensive care medicine; Health care; Anticholinergic agents; Systematic review; Indirect costs; MEDLINE; Internal medicine; Alternative medicine","score_opus":0.15711080105775044,"score_gpt":0.4854165608149944,"score_spread":0.32830575975724396,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4308668157","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":[0.004821687,0.9875322,0.0000032538824,0.00008160394,0.000060670613,0.0072447495,0.0001081994,0.000004113311,0.00014353871],"genre_scores_gemma":[0.011733715,0.9874513,0.00004408518,0.0000069160465,0.00012878074,0.00044383822,0.00006998459,0.000057104233,0.000064283005],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9832914,0.012469991,0.001657836,0.0005296798,0.0015070882,0.0005439887],"domain_scores_gemma":[0.98641336,0.010098628,0.001201941,0.00051410805,0.0014244865,0.00034748425],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.007672658,0.0005078292,0.006620742,0.0021067546,0.00012701469,0.0000883698,0.00038290612,0.00019768107,0.000069757996],"category_scores_gemma":[0.00060542737,0.00031766496,0.00047595968,0.0017766168,0.0002981526,0.00046013683,0.0002512859,0.0042662937,0.0000047095814],"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":[0.004112503,0.00032209407,0.00018397966,0.9814718,0.0013851059,0.0020477416,0.0015979516,0.00001758051,0.0000016189263,0.000070658876,0.00011143306,0.008677579],"study_design_scores_gemma":[0.002373361,0.007117189,0.00036842772,0.9554823,0.0006731117,0.005145237,0.00054853316,0.000047614547,0.0000058359715,0.000021075599,0.027961718,0.00025561822],"about_ca_topic_score_codex":0.00014185694,"about_ca_topic_score_gemma":0.000024541145,"teacher_disagreement_score":0.027850285,"about_ca_system_score_codex":0.0010779591,"about_ca_system_score_gemma":0.0029167214,"threshold_uncertainty_score":0.9999275},"labels":[],"label_agreement":null},{"id":"W4309827664","doi":"10.2217/cer-2022-0146","title":"Systematic review of outcomes and patient heterogeneity in relapsed or refractory diffuse large B-cell lymphoma","year":2022,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Lymphoma Diagnosis and Treatment","field":"Medicine","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":"Vancouver Coastal Health","funders":"","keywords":"Medicine; Refractory (planetary science); Clinical trial; Internal medicine; Lymphoma; Oncology; Diffuse large B-cell lymphoma","score_opus":0.1483897253639382,"score_gpt":0.4705964074388625,"score_spread":0.3222066820749243,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309827664","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":[0.045879573,0.94838095,0.0000016921445,0.000014535885,0.00012106722,0.0054077553,0.00005560576,0.000004245611,0.00013459656],"genre_scores_gemma":[0.032183155,0.966737,0.000023591327,0.000025962092,0.000020186038,0.0009241397,0.000016348133,0.000039646613,0.000030021767],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9844272,0.010640598,0.0023071598,0.00040601162,0.0017850415,0.00043400162],"domain_scores_gemma":[0.98519796,0.01144401,0.001865899,0.00057690137,0.00065271166,0.00026251562],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.005658727,0.00050896377,0.008540761,0.0010648927,0.0001203356,0.000026448673,0.0003166249,0.00016726872,0.000074147145],"category_scores_gemma":[0.0014588152,0.00027545373,0.0010089772,0.0009879409,0.00014332417,0.00009109769,0.0003076506,0.0008044659,0.000008781545],"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":[0.00065457926,0.0138460575,0.00083147053,0.9644575,0.0015902141,0.0011935958,0.0002735728,1.86381e-7,0.0000014561319,0.000019936771,0.00009890384,0.01703249],"study_design_scores_gemma":[0.012612276,0.0059521887,0.0030651249,0.928801,0.0023790828,0.0006430223,0.00022850087,0.0000030896358,0.00006739049,0.00000677736,0.045982677,0.00025889522],"about_ca_topic_score_codex":0.000020523714,"about_ca_topic_score_gemma":0.000007979699,"teacher_disagreement_score":0.045883775,"about_ca_system_score_codex":0.0021538713,"about_ca_system_score_gemma":0.0015623479,"threshold_uncertainty_score":0.9999698},"labels":[],"label_agreement":null},{"id":"W4310361476","doi":"10.2217/cer-2022-0147","title":"Real-world weight changes in people with HIV-1 at risk of weight gain (female, Black or Hispanic) switching from integrase strand transfer inhibitors","year":2022,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"HIV-related health complications and treatments","field":"Medicine","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":"Group for Research in Decision Analysis","funders":"Janssen Scientific Affairs","keywords":"Medicine; Integrase; Integrase inhibitor; Weight gain; Human immunodeficiency virus (HIV); Antiretroviral therapy; Internal medicine; Demography; Virology; Body weight; Viral load","score_opus":0.05781207410141871,"score_gpt":0.3916310975316421,"score_spread":0.3338190234302234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4310361476","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":[0.9956235,0.00057446834,0.00079110614,0.00056354894,0.00007978176,0.0014466168,0.00012633251,0.000009273811,0.0007853592],"genre_scores_gemma":[0.99844456,0.0006541405,0.00022270632,0.000009012087,0.00010874515,0.00017902006,0.00003783706,0.00003406029,0.00030989677],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9939831,0.0032420463,0.00071247015,0.0003630836,0.0012463528,0.00045298337],"domain_scores_gemma":[0.99492276,0.0034600103,0.0003088717,0.00032817986,0.0006865171,0.00029367782],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0026442187,0.000257876,0.0011786465,0.0012910997,0.0004084693,0.000017358252,0.00024041637,0.000069642156,0.00061423716],"category_scores_gemma":[0.00006997066,0.00016753889,0.00014222287,0.0018903445,0.000178383,0.00013768814,0.00009328593,0.0018867041,0.000014262319],"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":[0.0421112,0.0028419509,0.91826314,0.00042950112,0.0012762576,0.0005531901,0.017010445,0.002117123,0.013537011,0.00025198967,0.00045780366,0.0011504149],"study_design_scores_gemma":[0.0139640495,0.007829162,0.941004,0.0020871747,0.00030524257,0.00009128156,0.00319188,0.002007025,0.02809363,0.00022484601,0.0009745147,0.00022722098],"about_ca_topic_score_codex":0.0017537092,"about_ca_topic_score_gemma":0.00836145,"teacher_disagreement_score":0.02814715,"about_ca_system_score_codex":0.0015905218,"about_ca_system_score_gemma":0.0007837845,"threshold_uncertainty_score":0.8196896},"labels":[],"label_agreement":null},{"id":"W4311380649","doi":"10.2217/cer-2022-0192","title":"Letter in reply: indirect treatment comparison of anifrolumab efficacy versus belimumab in adults with systemic lupus erythematosus","year":2022,"lang":"en","type":"letter","venue":"Journal of Comparative Effectiveness Research","topic":"Systemic Lupus Erythematosus Research","field":"Medicine","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":"EVERSANA (Canada)","funders":"National Institute for Health and Care Research; AstraZeneca","keywords":"Belimumab; Medicine; Lupus erythematosus; Dermatology; Immunology; Internal medicine; Antibody","score_opus":0.10524607360021869,"score_gpt":0.4155190755214279,"score_spread":0.3102730019212092,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311380649","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":[0.97660184,0.0064906133,0.00001514406,0.008656914,0.00083634513,0.0060663857,0.000043316104,0.000021866179,0.001267571],"genre_scores_gemma":[0.9968845,0.0002875401,0.00007280488,0.00053684146,0.00090161164,0.0007415641,0.00008578102,0.00017008686,0.00031922906],"study_design_codex":"observational","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9773957,0.011634131,0.002723665,0.0010894138,0.00565521,0.0015018673],"domain_scores_gemma":[0.98001057,0.014695132,0.0017214217,0.001250191,0.0020543858,0.0002682796],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.007043479,0.00090160646,0.00598721,0.005707132,0.00018163752,0.00007912953,0.0010355664,0.0005669618,0.00017613803],"category_scores_gemma":[0.0006684412,0.00065891445,0.00060502405,0.0032554748,0.0006609618,0.00023907791,0.00032514814,0.008655736,0.000029473495],"study_design_candidate":"randomized_trial","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":[0.26424906,0.0058571557,0.5367858,0.02003217,0.0039459276,0.031136425,0.029947245,0.0022180218,0.00082320557,0.0000149164725,0.104308166,0.0006818958],"study_design_scores_gemma":[0.34837025,0.15887798,0.05439066,0.17512625,0.0010987085,0.13175689,0.029979603,0.0021085762,0.013564233,0.00008337211,0.08164644,0.0029970375],"about_ca_topic_score_codex":0.0006116399,"about_ca_topic_score_gemma":0.00030304785,"teacher_disagreement_score":0.48239514,"about_ca_system_score_codex":0.009178311,"about_ca_system_score_gemma":0.0029449898,"threshold_uncertainty_score":0.9995862},"labels":[],"label_agreement":null},{"id":"W4361208984","doi":"10.57264/cer-2022-0097","title":"Treatment goals for rheumatoid arthritis: patient engagement and goal collection","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Rheumatoid Arthritis Research and Therapies","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":"McGill University","funders":"","keywords":"Medicine; Data collection; Ranking (information retrieval); Rheumatoid arthritis; Physical therapy; Identification (biology); Prioritization; Goal setting; Process management; Computer science; Psychology; Social psychology; Internal medicine; Artificial intelligence","score_opus":0.11336434016930408,"score_gpt":0.4292838292109666,"score_spread":0.3159194890416625,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4361208984","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":[0.9866586,0.009743939,0.0005341757,0.0004337979,0.00016633849,0.0020548396,0.000015071548,0.000025082229,0.00036816645],"genre_scores_gemma":[0.95563424,0.0434711,0.00024174644,0.0000048571264,0.00008502967,0.0004438906,0.000009156733,0.000020326963,0.000089654655],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9962439,0.0016747983,0.00044834276,0.00023622048,0.00089582294,0.0005009282],"domain_scores_gemma":[0.9948878,0.0031434128,0.0001346174,0.00014648985,0.001393541,0.0002941182],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003254874,0.000173766,0.0007279694,0.0008723075,0.00051516114,0.00010307134,0.00008218455,0.0000612653,0.000033530654],"category_scores_gemma":[0.0004704847,0.00012835063,0.00017936042,0.0007333908,0.00026852128,0.00019108495,0.00006638921,0.00038989214,0.000028601968],"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":[0.062397365,0.0049717114,0.07753389,0.0010410015,0.0056380727,0.0013043345,0.034293246,0.00066094816,0.22233887,0.0076971957,0.022944044,0.5591793],"study_design_scores_gemma":[0.050629947,0.10104334,0.37812775,0.005272702,0.000024440573,0.0019819955,0.0168391,0.0053416113,0.2711429,0.009623162,0.15932085,0.0006521885],"about_ca_topic_score_codex":0.000028284754,"about_ca_topic_score_gemma":0.000016336862,"teacher_disagreement_score":0.5585271,"about_ca_system_score_codex":0.00038750155,"about_ca_system_score_gemma":0.00044943442,"threshold_uncertainty_score":0.5233987},"labels":[],"label_agreement":null},{"id":"W4367299362","doi":"10.57264/cer-2023-0009","title":"Letter in Reply: indirect comparison of quality-of-life scores between patients with advanced breast cancer receiving palbociclib and abemaciclib in combination with fulvestrant","year":2023,"lang":"en","type":"letter","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Breast Cancer Therapies","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":"EVERSANA (Canada)","funders":"Pfizer","keywords":"Palbociclib; Medicine; Fulvestrant; Oncology; Breast cancer; Quality of life (healthcare); Internal medicine; Cancer; Gynecology; Metastatic breast cancer; Estrogen receptor","score_opus":0.12516054244642327,"score_gpt":0.46097057455155016,"score_spread":0.3358100321051269,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4367299362","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":[0.9816983,0.0009975946,0.00018426393,0.015180357,0.00009048636,0.0017116623,0.000108132415,0.000010644251,0.000018612507],"genre_scores_gemma":[0.99804187,0.0000827792,0.000268589,0.0010259787,0.00030179357,0.00013077544,0.000052303567,0.00008326011,0.000012670728],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9917998,0.0032938346,0.0015239781,0.00052209117,0.00227577,0.0005845483],"domain_scores_gemma":[0.98994297,0.0056283143,0.0017313792,0.00033446655,0.0022258654,0.00013699714],"candidate_categories":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0031210457,0.0004463019,0.0033151342,0.0019454446,0.00007150761,0.0000281827,0.00029351318,0.0002715703,0.000012021929],"category_scores_gemma":[0.00026172318,0.00031889544,0.00011639442,0.00176992,0.00084326847,0.00042064616,0.000086329805,0.003734494,7.269762e-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":[0.020422526,0.00048114997,0.9652222,0.003232631,0.0007251785,0.00007249375,0.0023872608,0.0007311589,0.0007581877,0.0000038572425,0.00051331194,0.0054500415],"study_design_scores_gemma":[0.013928944,0.0030875346,0.96339774,0.015567221,0.00007486981,0.000009885292,0.00034562222,0.000022743867,0.002806656,0.0001348894,0.00033565486,0.00028825516],"about_ca_topic_score_codex":0.0004220865,"about_ca_topic_score_gemma":0.00028732416,"teacher_disagreement_score":0.016343603,"about_ca_system_score_codex":0.0007152284,"about_ca_system_score_gemma":0.0008096162,"threshold_uncertainty_score":0.9999263},"labels":[],"label_agreement":null},{"id":"W4379093489","doi":"10.57264/cer-2023-0016","title":"Comparative efficacy of therapies for relapsing multiple sclerosis: a systematic review and network meta-analysis","year":2023,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Sclerosis Research Studies","field":"Medicine","cited_by":52,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Ofatumumab; Alemtuzumab; Ocrelizumab; Natalizumab; Internal medicine; Placebo; Multiple sclerosis; Lipoarabinomannan; Oncology; Rituximab; Disease; Tuberculosis; Immunology; Alternative medicine; Pathology; Lymphoma","score_opus":0.7093580791148476,"score_gpt":0.5626411172863962,"score_spread":0.1467169618284514,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4379093489","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":[0.000089159046,0.9815004,0.00084777933,0.000073166004,0.00007291204,0.017214065,0.00014387337,0.000020461643,0.000038195692],"genre_scores_gemma":[0.005109358,0.99071777,0.00083486753,0.000009408754,0.00017939035,0.0029237764,0.000033921347,0.00008293991,0.00010855796],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9769212,0.015588132,0.003422416,0.00074560125,0.0024278003,0.00089486974],"domain_scores_gemma":[0.9192491,0.07076028,0.0029167277,0.00083415455,0.0058666486,0.00037306597],"candidate_categories":["metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.025009504,0.0008701694,0.03329588,0.0023557374,0.00046803112,0.00011758226,0.0007227767,0.00027462057,0.000030974774],"category_scores_gemma":[0.006839192,0.00050177734,0.0070160213,0.0057165227,0.0009739356,0.00020832353,0.00041580648,0.0019668753,0.000014470563],"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":[0.00045377016,0.00028223632,0.000013437902,0.6203665,0.3769443,0.000009801751,0.0005751818,0.000069139736,0.000012546786,0.000035257963,0.0005302657,0.00070751493],"study_design_scores_gemma":[0.0016455554,0.0014998956,0.0005581412,0.47579813,0.5170714,0.00003251599,0.00037975027,0.00017950528,0.000019924,0.000045148838,0.0024385299,0.00033148387],"about_ca_topic_score_codex":0.000021778842,"about_ca_topic_score_gemma":0.000025005073,"teacher_disagreement_score":0.14456838,"about_ca_system_score_codex":0.00048288025,"about_ca_system_score_gemma":0.00083780574,"threshold_uncertainty_score":0.9997434},"labels":[],"label_agreement":null},{"id":"W4385483571","doi":"10.57264/cer-2023-0066","title":"Cost–effectiveness of ropeginterferon alfa-2b-njft for the treatment of polycythemia vera","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Myeloproliferative Neoplasms: Diagnosis and Treatment","field":"Medicine","cited_by":5,"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":"Sierra Oncology; Incyte; Celgene","keywords":"Medicine; Polycythemia vera; Quality-adjusted life year; Myelofibrosis; Population; Internal medicine; Cost effectiveness; Cohort","score_opus":0.3315524533615838,"score_gpt":0.5218676901699087,"score_spread":0.19031523680832496,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385483571","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":[0.9903619,0.002237307,0.0022029325,0.0003204591,0.00029637633,0.0041957838,0.00006269225,0.00001071613,0.00031181253],"genre_scores_gemma":[0.9974534,0.0012466813,0.000118706834,0.000004068132,0.00020238965,0.0007948035,0.00001752532,0.000032018186,0.00013038574],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9956924,0.0018825192,0.00066644774,0.00029168985,0.0010000584,0.0004669355],"domain_scores_gemma":[0.97714245,0.019543588,0.00038784716,0.00037536657,0.0023747387,0.00017601454],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0039956523,0.00029509122,0.0014935698,0.0006242161,0.00021345647,0.000027674383,0.00025220128,0.0001003278,0.000043050124],"category_scores_gemma":[0.0003446391,0.00016223382,0.0005718215,0.0009534046,0.00045455343,0.0001252437,0.000055858425,0.00033036922,0.000016739716],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.19168347,0.01883154,0.32036227,0.00574309,0.018768007,0.00047281897,0.0154301375,0.010109389,0.34189886,0.0081708655,0.003217793,0.06531176],"study_design_scores_gemma":[0.009271469,0.021250807,0.3539984,0.0013400192,0.00026191107,0.000052534644,0.00074964226,0.0008739931,0.60879457,0.00046504676,0.002833494,0.0001080983],"about_ca_topic_score_codex":0.00006810619,"about_ca_topic_score_gemma":0.000012997602,"teacher_disagreement_score":0.2668957,"about_ca_system_score_codex":0.00079059973,"about_ca_system_score_gemma":0.00056674535,"threshold_uncertainty_score":0.6615703},"labels":[],"label_agreement":null},{"id":"W4385684181","doi":"10.57264/cer-2023-0061","title":"Real-world impact of patient-reported outcome measurement on overall survival, healthcare use and treatment discontinuation in cancer patients","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Cancer survivorship and care","field":"Medicine","cited_by":3,"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":"Medicine; Hazard ratio; Discontinuation; Emergency department; Proportional hazards model; Retrospective cohort study; Cohort; Internal medicine; Population; Breast cancer; Colorectal cancer; Propensity score matching; Cohort study; Emergency medicine; Cancer; Lung cancer; Confidence interval","score_opus":0.36952984768956515,"score_gpt":0.517947666750521,"score_spread":0.1484178190609558,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385684181","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":[0.9981514,0.00018870637,0.0000031533632,0.0001590487,0.00025502782,0.001112523,0.00003054294,0.0000070475376,0.00009253652],"genre_scores_gemma":[0.99938107,0.0004037903,0.0000027713347,0.000004520152,0.00005694366,0.000083976855,0.000013764119,0.000016928181,0.00003625949],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9954028,0.0017152977,0.00070168596,0.00023233067,0.0016192588,0.00032860483],"domain_scores_gemma":[0.9956935,0.0010478582,0.0003930818,0.00018823262,0.0024777073,0.00019963674],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002203597,0.00019033367,0.0008940155,0.0010764834,0.000067828194,0.000032615317,0.00006306006,0.00005300319,0.000014045733],"category_scores_gemma":[0.00027327242,0.00012733169,0.00020007792,0.0010184583,0.00008777294,0.00019940326,0.000037097063,0.00039143543,0.0000015622496],"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":[0.007258824,0.0009211565,0.9816534,0.00011405752,0.00037299132,0.000050851002,0.0032390193,0.00055631733,0.0016357196,0.000044128938,0.00006291822,0.0040906537],"study_design_scores_gemma":[0.004297315,0.008899248,0.9815412,0.0011300027,0.000035716894,0.0000014440724,0.00041476413,0.00009821724,0.0033948012,0.00004642928,0.00004965892,0.000091208574],"about_ca_topic_score_codex":0.022290993,"about_ca_topic_score_gemma":0.015575811,"teacher_disagreement_score":0.007978092,"about_ca_system_score_codex":0.0030322566,"about_ca_system_score_gemma":0.000622489,"threshold_uncertainty_score":0.9842197},"labels":[],"label_agreement":null},{"id":"W4385851857","doi":"10.57264/cer-2023-0005","title":"QDOT MICRO™ versus THERMOCOOL <sup>®</sup> SMARTTOUCH™ and THERMOCOOL SMARTTOUCH <sup>®</sup> Surround Flow in radiofrequency ablation of paroxysmal atrial fibrillation","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Atrial Fibrillation Management and Outcomes","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":"EVERSANA (Canada)","funders":"Biosense Webster","keywords":"Medicine; Atrial fibrillation; Ablation; Fluoroscopy; Internal medicine; Cardiology; Catheter ablation; Paroxysmal atrial fibrillation; Nuclear medicine; Surgery","score_opus":0.16607501389731916,"score_gpt":0.43216515214763945,"score_spread":0.2660901382503203,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385851857","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":[0.99456185,0.0011802305,0.00038900497,0.00048964226,0.00026978538,0.0021788792,0.00001558192,0.000033491313,0.0008815082],"genre_scores_gemma":[0.9971846,0.00048246118,0.000280178,0.0000044704543,0.0017437662,0.000010257707,0.000046740013,0.00004300817,0.00020455387],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.993177,0.0030156022,0.001206299,0.0004488194,0.0015289366,0.0006233355],"domain_scores_gemma":[0.9908773,0.006842738,0.0005056942,0.00034277447,0.0012088884,0.0002226131],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0064439643,0.00036423007,0.001374401,0.0017642635,0.00024825623,0.00010947923,0.00023831813,0.00023514297,0.00014552522],"category_scores_gemma":[0.0011714839,0.0003013745,0.00045862535,0.0022487857,0.00036975672,0.0006460238,0.00015104031,0.000771801,0.000048801583],"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":[0.04786118,0.00008248825,0.67379373,0.00088722596,0.001552918,0.00019662814,0.0066314954,0.22055738,0.022019634,0.00097361853,0.0005575712,0.024886126],"study_design_scores_gemma":[0.022204472,0.003253927,0.8414776,0.0008035016,0.00025120587,0.0000571287,0.0022083845,0.119243585,0.006367287,0.0012252087,0.002499224,0.00040850672],"about_ca_topic_score_codex":0.00011013074,"about_ca_topic_score_gemma":0.000011542741,"teacher_disagreement_score":0.16768382,"about_ca_system_score_codex":0.0005114314,"about_ca_system_score_gemma":0.0004500967,"threshold_uncertainty_score":0.99994385},"labels":[],"label_agreement":null},{"id":"W4386046576","doi":"10.57264/cer-2022-0175","title":"Cost–consequence analysis of ofatumumab for the treatment of relapsing-remitting multiple sclerosis in Canada","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Sclerosis Research Studies","field":"Medicine","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":"Novartis (Canada); EVERSANA (Canada); Queen's University; Memorial University of Newfoundland; University of British Columbia","funders":"","keywords":"Medicine; Ofatumumab; Natalizumab; Relapsing remitting; Ocrelizumab; Multiple sclerosis; Physical therapy; Internal medicine; Disease; Immunology; Rituximab","score_opus":0.4634725382332816,"score_gpt":0.48936304131241415,"score_spread":0.025890503079132543,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386046576","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":[0.99534225,0.0017533731,0.0005650282,0.00028297302,0.00009318096,0.0018368812,0.000072811396,0.0000035456721,0.000049924805],"genre_scores_gemma":[0.99730086,0.002180283,0.00018895237,0.0000028277548,0.000046119334,0.00023287162,0.0000065624968,0.000013273629,0.00002826946],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9954292,0.0014510993,0.00084455434,0.00023624056,0.0015186063,0.0005202725],"domain_scores_gemma":[0.9648265,0.031593207,0.00041759002,0.0003209018,0.0027038339,0.00013794402],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004962598,0.00017303387,0.0014770393,0.0015063233,0.00018772561,0.000015028615,0.00029011682,0.0000473981,0.000016096197],"category_scores_gemma":[0.0030529168,0.00010662292,0.0004033857,0.004510446,0.0004822927,0.00008723283,0.00007391768,0.00040211697,8.989317e-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":[0.008309062,0.00084457395,0.74685025,0.0005070837,0.0075350776,0.000069948204,0.00528255,0.04572453,0.16533507,0.000028354038,0.00038922072,0.019124286],"study_design_scores_gemma":[0.0040435004,0.001524107,0.8674364,0.00083938194,0.00029714883,0.0000038119124,0.0043158424,0.0378332,0.08348725,0.000019890591,0.00013113861,0.000068316396],"about_ca_topic_score_codex":0.19932182,"about_ca_topic_score_gemma":0.2660212,"teacher_disagreement_score":0.12058617,"about_ca_system_score_codex":0.0021721905,"about_ca_system_score_gemma":0.0023666406,"threshold_uncertainty_score":0.80600995},"labels":[],"label_agreement":null},{"id":"W4386439153","doi":"10.57264/cer-2023-0075","title":"Cost-utility of real-time continuous glucose monitoring versus self-monitoring of blood glucose in people with insulin-treated Type 2 diabetes in Canada","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Diabetes Management and Research","field":"Medicine","cited_by":17,"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; Glycemic; Glycated hemoglobin; Hypoglycemia; Type 2 diabetes; Diabetes mellitus; Continuous glucose monitoring; Type 1 diabetes; Quality-adjusted life year; Internal medicine; Blood Glucose Self-Monitoring; Insulin; Cost effectiveness; Endocrinology","score_opus":0.07002837368940106,"score_gpt":0.3832770973235926,"score_spread":0.31324872363419154,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386439153","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":[0.9972177,0.0008072259,6.460196e-7,0.000027518836,0.00019979816,0.0011281654,0.0000063029793,0.000012259309,0.0006003487],"genre_scores_gemma":[0.9992157,0.00052054005,0.00006968028,2.5483192e-7,0.000087899665,0.00004059058,0.0000054920074,0.000023484234,0.000036345275],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9951612,0.0015529022,0.00068865373,0.0002660788,0.0016430632,0.0006881155],"domain_scores_gemma":[0.99302024,0.0041950676,0.0002861641,0.0002836302,0.0020107757,0.000204127],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0044593504,0.0002075739,0.0012608252,0.0013033719,0.00006114536,0.000022714004,0.00032200437,0.000075432305,0.000020676693],"category_scores_gemma":[0.00059193664,0.00016136827,0.00008015092,0.003995682,0.00016543738,0.00019139094,0.00017544247,0.00090590364,0.0000054086895],"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":[0.008227251,0.00076945464,0.9331585,0.0008041846,0.00061848195,0.00031080042,0.0006243668,0.0005436099,0.05344945,0.000003644793,0.00002343339,0.0014668279],"study_design_scores_gemma":[0.006311823,0.0022981518,0.7256083,0.0014948756,0.00005951033,0.0000011436658,0.0013385137,0.0011075943,0.26167032,0.000009879885,0.0000110918845,0.00008881511],"about_ca_topic_score_codex":0.047949072,"about_ca_topic_score_gemma":0.01771643,"teacher_disagreement_score":0.20822088,"about_ca_system_score_codex":0.0006968417,"about_ca_system_score_gemma":0.0018754806,"threshold_uncertainty_score":0.9886184},"labels":[],"label_agreement":null},{"id":"W4386757294","doi":"10.57264/cer-2023-0141","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 13","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Hospital Admissions and Outcomes","field":"Medicine","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":"Medicine; Reimbursement; Real world evidence; Agency (philosophy); Health technology; Alternative medicine; Real world data; Family medicine; Health care; Data science; Economic growth; Pathology; Internal medicine","score_opus":0.3944640706343192,"score_gpt":0.5920552082771873,"score_spread":0.19759113764286806,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386757294","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":[0.9907526,0.00043545963,0.0006162088,0.0051149842,0.00023501743,0.0018701556,0.0000052348887,0.000017812594,0.0009525549],"genre_scores_gemma":[0.99547607,0.00047119948,0.0023201238,0.000020362813,0.00006227358,0.00016529132,0.0000043156074,0.000018305274,0.0014620682],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963334,0.00093371404,0.00095247605,0.00027287667,0.00095127395,0.00055626535],"domain_scores_gemma":[0.9940113,0.0038175338,0.0003824601,0.00022976716,0.0012796173,0.0002793322],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008816666,0.00015870588,0.0010359504,0.0024504883,0.0001814773,0.000023510505,0.0002515194,0.00007426783,0.000016908261],"category_scores_gemma":[0.0015234967,0.00012075139,0.00010052703,0.0035585393,0.000111979774,0.00020826324,0.00017094032,0.00078960985,0.000013536601],"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":[0.0034555483,0.00060108636,0.94520575,0.0015934737,0.0003608889,0.00013982704,0.004111529,0.00051115925,0.013755888,0.006766509,0.004420801,0.019077523],"study_design_scores_gemma":[0.0029107807,0.0057341317,0.9583412,0.012344356,0.000017928192,0.000017936769,0.0049701533,0.00031387623,0.00876455,0.0017077262,0.004737714,0.00013965242],"about_ca_topic_score_codex":0.00013718565,"about_ca_topic_score_gemma":0.00018538392,"teacher_disagreement_score":0.01893787,"about_ca_system_score_codex":0.0010314408,"about_ca_system_score_gemma":0.0018560318,"threshold_uncertainty_score":0.49240988},"labels":[],"label_agreement":null},{"id":"W4388730279","doi":"10.57264/cer-2023-0054","title":"Gastrointestinal manifestations in pediatric and adult patients with Rett syndrome: an analysis of US claims and physician survey data","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Genetics and Neurodevelopmental Disorders","field":"Biochemistry, Genetics and Molecular Biology","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":"Group for Research in Decision Analysis","funders":"Eunice Kennedy Shriver National Institute of Child Health and Human Development","keywords":"Medicine; Rett syndrome; Medical costs; Pediatrics; Health insurance; Medical care; Young adult; Health care; Internal medicine; Family medicine","score_opus":0.06856805424622395,"score_gpt":0.3855690202668885,"score_spread":0.3170009660206645,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388730279","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":[0.99951047,0.000062200714,0.00013582477,0.0000065913255,0.000009362141,0.00017380352,0.000085994194,8.997914e-7,0.000014879747],"genre_scores_gemma":[0.99937713,0.00015644738,0.0001523371,0.000003341318,0.000004994117,0.000004311071,0.00029124596,0.0000074889763,0.0000027115786],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9981943,0.000867715,0.00021200399,0.00023964497,0.0003186979,0.00016768218],"domain_scores_gemma":[0.9987056,0.00025241548,0.0001222495,0.00017680846,0.00066688395,0.000076027325],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0011493957,0.000091626534,0.0002581764,0.0005073534,0.00006077131,0.000022710177,0.00019154786,0.000020916408,8.627019e-7],"category_scores_gemma":[0.00011509305,0.00007539637,0.000019477948,0.0010435225,0.00013151647,0.000019028588,0.00016459773,0.00015720427,3.2022274e-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":[0.0005223326,0.00017822698,0.9927054,0.00003919928,0.00011357585,0.00001886748,0.00008160478,0.00097488135,0.0052070837,0.0000036043255,0.000018639019,0.00013657183],"study_design_scores_gemma":[0.0008443617,0.0017221671,0.99540484,0.00003223155,0.000046816174,0.0000133122085,0.00018498038,0.001180432,0.00047526552,0.000022517568,0.0000015780397,0.0000714792],"about_ca_topic_score_codex":0.00006012962,"about_ca_topic_score_gemma":0.00074242876,"teacher_disagreement_score":0.0047318186,"about_ca_system_score_codex":0.00001177924,"about_ca_system_score_gemma":0.00009885577,"threshold_uncertainty_score":0.30745748},"labels":[],"label_agreement":null},{"id":"W4389803456","doi":"10.57264/cer-2023-0142","title":"Real-world use of inotuzumab ozogamicin is associated with lower health care costs than blinatumomab in patients with acute lymphoblastic leukemia in the first relapsed/refractory setting","year":2023,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Acute Lymphoblastic Leukemia research","field":"Medicine","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":"Pfizer (Canada)","funders":"Pfizer Pharmaceuticals; Pfizer","keywords":"Medicine; Blinatumomab; Refractory (planetary science); Lymphoblastic Leukemia; Retrospective cohort study; Internal medicine; Emergency medicine; Pediatrics; Leukemia","score_opus":0.041280349020756044,"score_gpt":0.37272352712147,"score_spread":0.33144317810071394,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389803456","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":[0.9969078,0.000089119654,0.000040046525,0.00034508097,0.00007846568,0.0021981145,0.0000572686,0.000018005796,0.0002660993],"genre_scores_gemma":[0.99948364,0.0000716402,0.000062121035,0.00006160355,0.000053961194,0.00007370393,0.000055909237,0.00006415118,0.000073271876],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9900847,0.003705452,0.0010104999,0.00049139815,0.0036120198,0.00109594],"domain_scores_gemma":[0.9849948,0.0103795715,0.00074687495,0.0005263137,0.0030923765,0.00026000565],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0051766867,0.0003925149,0.0014991802,0.0028121788,0.0002644342,0.00008589401,0.0005745637,0.00016703641,0.000012750526],"category_scores_gemma":[0.0010173033,0.00023639912,0.00014305476,0.0063818255,0.00060057646,0.00050834805,0.00020123828,0.0023246878,0.0000124910775],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.040544994,0.0012123145,0.9341351,0.0005978409,0.0008283295,0.0023357433,0.014466853,0.0029441374,0.00018310624,0.000047284,0.002225663,0.00047863976],"study_design_scores_gemma":[0.02199758,0.011718315,0.953966,0.0066589275,0.000064989516,0.000047801434,0.0022454532,0.0021751442,0.00081713864,0.000015680514,0.00008705264,0.00020588332],"about_ca_topic_score_codex":0.0017048967,"about_ca_topic_score_gemma":0.0032511742,"teacher_disagreement_score":0.019830937,"about_ca_system_score_codex":0.02126202,"about_ca_system_score_gemma":0.0085025,"threshold_uncertainty_score":0.999977},"labels":[],"label_agreement":null},{"id":"W4390743236","doi":"10.57264/cer-2023-0124","title":"Smart connected insulin dose monitoring technologies versus standard of care: a Canadian cost–effectiveness analysis","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Diabetes Management and Research","field":"Medicine","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":"McMaster University; Impact","funders":"","keywords":"Medicine; Glycated hemoglobin; Glycemic; Life expectancy; Diabetes management; Insulin; Marginal cost; Psychological intervention; Quality-adjusted life year; Type 1 diabetes; Diabetes mellitus; Insulin pen; Type 2 diabetes; Intensive care medicine; Cost effectiveness; Internal medicine; Environmental health; Endocrinology; Risk analysis (engineering); Population; Nursing","score_opus":0.1274245942080996,"score_gpt":0.4715588744915871,"score_spread":0.3441342802834875,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390743236","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":[0.98203063,0.012098701,0.00057480216,0.00023575252,0.00061976275,0.0015926848,0.000037451664,0.000055773933,0.0027544275],"genre_scores_gemma":[0.99923587,0.00030744312,0.00014120514,7.999303e-7,0.00012342066,0.00010082427,0.000012636016,0.00002743401,0.000050376944],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9945791,0.001999924,0.0005366137,0.00038740164,0.0017759829,0.00072102103],"domain_scores_gemma":[0.98995733,0.0053843786,0.00013023193,0.00038164036,0.0038199828,0.00032640863],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005886716,0.00025696878,0.0014159025,0.005619991,0.00021972683,0.00017482394,0.00043839143,0.00018632413,0.000054992597],"category_scores_gemma":[0.0011170995,0.00019423418,0.00046876256,0.006555988,0.0005977782,0.00029683727,0.0001827397,0.001562962,0.000020501753],"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":[0.15133084,0.0012488066,0.47443974,0.016476985,0.052392155,0.005135374,0.006811891,0.0073938738,0.05282962,0.008217968,0.00084732805,0.22287542],"study_design_scores_gemma":[0.011192189,0.01957036,0.6941322,0.0061040367,0.0019855746,0.000017165441,0.010442692,0.0021925697,0.24329759,0.0007438122,0.009843241,0.00047858234],"about_ca_topic_score_codex":0.0013360622,"about_ca_topic_score_gemma":0.0015988704,"teacher_disagreement_score":0.22239684,"about_ca_system_score_codex":0.0019587877,"about_ca_system_score_gemma":0.0018933074,"threshold_uncertainty_score":0.792064},"labels":[],"label_agreement":null},{"id":"W4390743877","doi":"10.57264/cer-2023-0147","title":"Quantitative bias analysis for external control arms using real-world data in clinical trials: a primer for clinical researchers","year":2024,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Causal Inference Techniques","field":"Mathematics","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":"Public Health Ontario; University of Toronto; McMaster University; Impact","funders":"","keywords":"Robustness (evolution); Missing data; External validity; Medicine; Confounding; Sample size determination; Population; Clinical trial; Data collection; Computer science; Statistics; Machine learning","score_opus":0.9834323930954177,"score_gpt":0.8240222415215248,"score_spread":0.15941015157389293,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390743877","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":[0.0008893286,0.5295447,0.45985883,0.000031165233,0.00035519924,0.008582532,0.00067911396,0.000029043544,0.000030088835],"genre_scores_gemma":[0.0019415319,0.88001734,0.11549736,0.0000043694267,0.0011703563,0.0010524122,0.00007208915,0.00016833216,0.00007618614],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9050604,0.080735974,0.009428098,0.0015136952,0.0020953047,0.0011665077],"domain_scores_gemma":[0.59380805,0.3951927,0.0050100097,0.0014303263,0.00411909,0.00043977902],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.33425945,0.0008316274,0.016129646,0.0056879367,0.00018756163,0.00037240362,0.0027693245,0.0006265301,0.000016289465],"category_scores_gemma":[0.12180561,0.0005502163,0.0047719446,0.005079334,0.0014451537,0.00065140665,0.0007213948,0.006224374,0.000008687379],"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":[0.04622059,0.0063511305,0.0064348853,0.07784056,0.08507364,0.0005796313,0.0008958925,0.00024999375,0.0000436797,0.07763619,0.0063892147,0.6922846],"study_design_scores_gemma":[0.015715634,0.0225216,0.0018879095,0.17294413,0.060109936,0.000080349986,0.0015218693,0.067155525,0.000081045604,0.44818282,0.20681317,0.002986006],"about_ca_topic_score_codex":0.00010301141,"about_ca_topic_score_gemma":0.00043772394,"teacher_disagreement_score":0.68929857,"about_ca_system_score_codex":0.0014905385,"about_ca_system_score_gemma":0.0044742306,"threshold_uncertainty_score":0.99969494},"labels":[],"label_agreement":null},{"id":"W4391779405","doi":"10.57264/cer-2023-0159","title":"Study Protocol for the Social Interventions for Support During Treatment for Endometrial Cancer and Recurrence (SISTER) study: a community engaged national randomized trial","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Cancer survivorship and care","field":"Medicine","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":"Michael Smith Health Research BC","funders":"National Cancer Institute","keywords":"Medicine; Psychological intervention; Social support; Randomized controlled trial; Social isolation; Endometrial cancer; Intervention (counseling); Qualitative research; Peer support; Family medicine; Gynecology; Cancer; Internal medicine; Nursing; Social psychology; Psychology; Psychiatry","score_opus":0.5696956726535437,"score_gpt":0.6349042265350672,"score_spread":0.06520855388152347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391779405","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":[0.6582116,0.00018533399,0.0022032135,0.00013474895,0.00044833202,0.3386732,0.000115450915,0.000007349085,0.00002079427],"genre_scores_gemma":[0.5846109,0.0000037766274,0.000013639416,0.0000010456636,0.00071563915,0.41457593,0.0000036752726,0.0000105091385,0.000064883076],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98845553,0.009730189,0.000643484,0.00020278526,0.0007267308,0.0002412754],"domain_scores_gemma":[0.95071447,0.04671346,0.00020309338,0.000104963314,0.0021891224,0.000074913856],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.032516245,0.0001795177,0.0011228011,0.00050339225,0.0017120974,0.00018528731,0.00016942358,0.00004582894,0.000021484639],"category_scores_gemma":[0.0028818692,0.00009998891,0.0009675532,0.00043649215,0.00021014592,0.00012961286,0.00005925543,0.0010248848,2.3845914e-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":[0.951715,0.0036788592,0.0005306416,0.0012558486,0.003451892,0.000004973317,0.035510816,0.000015470792,0.0001360963,0.00007136317,0.00018097935,0.0034480682],"study_design_scores_gemma":[0.92477924,0.040187236,0.008377877,0.00078925904,0.0007117248,0.000011531798,0.022771517,0.0002425061,0.0006443209,0.00042761175,0.00094975193,0.00010739642],"about_ca_topic_score_codex":0.00017350046,"about_ca_topic_score_gemma":0.0006592474,"teacher_disagreement_score":0.07590272,"about_ca_system_score_codex":0.0008910146,"about_ca_system_score_gemma":0.001038792,"threshold_uncertainty_score":0.99958754},"labels":[],"label_agreement":null},{"id":"W4393871536","doi":"10.57264/cer-2023-0178","title":"The impact of willingness-to-pay threshold on price reduction recommendations for oncology drugs: a review of assessments conducted by the Canadian Agency for Drugs and Technologies in Health","year":2024,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Arthur B. McDonald-Canadian Astroparticle Physics Research Institute; Wilfrid Laurier University; Canadian Institute for Health Information","funders":"","keywords":"Medicine; Reimbursement; Internal medicine; Willingness to pay; Agency (philosophy); Oncology; Health care","score_opus":0.659831116409612,"score_gpt":0.6654963885408027,"score_spread":0.005665272131190724,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4393871536","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":[0.0006093803,0.96752083,0.00027124924,0.021720799,0.00042659382,0.008356116,0.0009082886,0.0000046176297,0.00018212946],"genre_scores_gemma":[0.011396719,0.9864465,0.00026568922,0.00006694497,0.00008796146,0.001603151,0.00004646216,0.000031034942,0.000055508863],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.990194,0.003926283,0.0047636307,0.0004062681,0.00021723473,0.00049255975],"domain_scores_gemma":[0.97835326,0.01545515,0.00470688,0.00039180482,0.0009382992,0.00015461477],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08046347,0.00026943776,0.0034626287,0.0013033072,0.00044693614,0.000079245874,0.00063851767,0.00021387047,0.000009194266],"category_scores_gemma":[0.0045538605,0.0001626708,0.00043982896,0.0012407737,0.0002832031,0.0001848941,0.00006737346,0.0010999283,0.0000070836522],"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":true,"study_design_scores_codex":[0.0005896171,0.0007687265,0.00022457527,0.3222046,0.003312038,0.0000010723636,0.008078473,0.00018594116,0.000002626045,0.05498814,0.24581714,0.36382708],"study_design_scores_gemma":[0.00071654044,0.0029044684,0.00020161079,0.084106684,0.00006553976,0.000015088116,0.0021847126,0.00022825744,0.00000438573,0.012597367,0.8967496,0.00022575552],"about_ca_topic_score_codex":0.005010289,"about_ca_topic_score_gemma":0.0011652756,"teacher_disagreement_score":0.65093243,"about_ca_system_score_codex":0.0054968027,"about_ca_system_score_gemma":0.0061613414,"threshold_uncertainty_score":0.9994728},"labels":[],"label_agreement":null},{"id":"W4401428163","doi":"10.57264/cer-2023-0171","title":"Patient-reported preferences for subcutaneous or intravenous administration of parenteral drug treatments in adults with immune disorders: a systematic review and meta-analysis","year":2024,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Pharmaceutical studies and practices","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":"University of Toronto","funders":"Grifols; Argenx; Sanofi; Alnylam Pharmaceuticals; UCB Pharma; Alexion Pharmaceuticals; Pfizer; Amgen","keywords":"Medicine; Meta-analysis; Drug; Intensive care medicine; Intravenous drug; Immune system; MEDLINE; Internal medicine; Pharmacology; Immunology; Human immunodeficiency virus (HIV)","score_opus":0.3327499208702503,"score_gpt":0.5498842533511791,"score_spread":0.21713433248092878,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401428163","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":[0.0030278452,0.9882559,0.000028769451,0.00007915877,0.000024950084,0.008466368,0.000075030555,0.000003189846,0.00003878932],"genre_scores_gemma":[0.20008402,0.7986924,0.000043469736,0.000003053431,0.000011374581,0.0010889153,0.00002540971,0.000015172268,0.000036187703],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9932465,0.0035010143,0.0018214646,0.00034779106,0.00081500335,0.00026825312],"domain_scores_gemma":[0.9912312,0.0058588632,0.0016075695,0.0002178881,0.00095294014,0.00013156516],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002793387,0.00039894565,0.009303096,0.0006450862,0.00007241348,0.000052429048,0.00014559992,0.000078350524,0.000033749613],"category_scores_gemma":[0.00052650884,0.00017232714,0.0011607499,0.0014124501,0.0001652265,0.00013548878,0.00005897157,0.0007487844,0.00000111546],"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":[0.005849239,0.0007788089,0.0000182529,0.7620169,0.22645646,0.0003266362,0.0006320953,0.0000016140078,3.058104e-7,0.0000047726808,0.000008757198,0.0039061443],"study_design_scores_gemma":[0.00093651685,0.010167883,0.00006621725,0.22132361,0.76508236,0.00077914976,0.0005345758,0.000020205447,0.000008101404,0.000039619008,0.00087494584,0.0001668287],"about_ca_topic_score_codex":0.00005098354,"about_ca_topic_score_gemma":0.00017857988,"teacher_disagreement_score":0.5406933,"about_ca_system_score_codex":0.00020855296,"about_ca_system_score_gemma":0.00055930106,"threshold_uncertainty_score":0.7027297},"labels":[],"label_agreement":null},{"id":"W4401517584","doi":"10.57264/cer-2024-0089","title":"A matching-adjusted indirect comparison of centanafadine versus lisdexamfetamine, methylphenidate and atomoxetine in adults with attention-deficit/hyperactivity disorder: long-term safety and efficacy","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Attention Deficit Hyperactivity Disorder","field":"Medicine","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":"Group for Research in Decision Analysis","funders":"Otsuka Pharmaceutical; Servier; Sunovion; Allergan; Purdue University","keywords":"Atomoxetine; Methylphenidate; Medicine; Attention deficit hyperactivity disorder; Term (time); Psychiatry; Atomoxetine hydrochloride","score_opus":0.08488134867198587,"score_gpt":0.43873423418700014,"score_spread":0.35385288551501426,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401517584","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":[0.99060565,0.0059362166,0.0016507225,0.0002472662,0.00013943139,0.001227605,0.00001913535,0.00002120018,0.00015275371],"genre_scores_gemma":[0.999055,0.000571219,0.00019509062,0.000001540352,0.000058383313,0.000021258462,0.000022885271,0.000041300722,0.000033313005],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99528944,0.0018835174,0.0007941011,0.00048866664,0.0011240401,0.00042025393],"domain_scores_gemma":[0.9926087,0.005723903,0.000340427,0.00023525198,0.00086573197,0.00022595371],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0026081593,0.00033568047,0.0015964119,0.0012434886,0.00015110454,0.00008904695,0.00013905216,0.00013154875,0.000042832788],"category_scores_gemma":[0.00029409057,0.00025366075,0.00014393548,0.0014927946,0.0004825755,0.0005814055,0.00013951716,0.0016134683,0.0000052936102],"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":[0.04357699,0.001991503,0.92781276,0.0021211577,0.00091178465,0.00031146186,0.0018574079,0.00032954194,0.009887354,0.00017401774,0.0000023813502,0.011023622],"study_design_scores_gemma":[0.014908423,0.0036100117,0.96993583,0.004742009,0.00023812894,0.00028000897,0.0034836999,0.0018671028,0.00068009016,0.000014292397,0.000024334451,0.00021606761],"about_ca_topic_score_codex":0.00017898517,"about_ca_topic_score_gemma":0.000550911,"teacher_disagreement_score":0.042123053,"about_ca_system_score_codex":0.00026108624,"about_ca_system_score_gemma":0.00026460356,"threshold_uncertainty_score":0.99999154},"labels":[],"label_agreement":null},{"id":"W4401518102","doi":"10.57264/cer-2024-0061","title":"The maze of real-world evidence frameworks: from a desert to a jungle! An environmental scan and comparison across regulatory and health technology assessment agencies","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Jungle; Medicine; Desert (philosophy); Real world evidence; Environmental resource management; Environmental planning; Geography; Archaeology; Internal medicine","score_opus":0.5807403555675743,"score_gpt":0.6226783898007007,"score_spread":0.041938034233126364,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401518102","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":[0.95334494,0.032019895,0.0033100739,0.010135293,0.00023318027,0.0008267199,0.000070563554,0.000010155941,0.00004915467],"genre_scores_gemma":[0.99646914,0.0013192595,0.0018679269,0.00010573142,0.000122041485,0.000056588364,0.0000020536959,0.000017158376,0.000040105562],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9940832,0.0024628507,0.0022425016,0.00043842383,0.00034220386,0.0004307815],"domain_scores_gemma":[0.98995286,0.008284068,0.0009927591,0.00035004472,0.00014543367,0.00027485078],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.042909086,0.00016570605,0.0011982223,0.00065725483,0.0004268856,0.00025208358,0.00038681884,0.00013011644,0.000021688453],"category_scores_gemma":[0.00070698874,0.00014576125,0.00006834028,0.00054186623,0.0006418712,0.00056440796,0.00021698294,0.0010134552,0.000018716906],"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":[0.00045968784,0.00031689228,0.8677408,0.001012542,0.0005945507,0.000010348314,0.040045965,0.0012151147,0.0010173662,0.07813918,0.0023774887,0.0070700604],"study_design_scores_gemma":[0.00035408846,0.0013008231,0.948872,0.0016660857,0.0000056802505,0.0000083051245,0.019754626,0.0077984147,0.00032102154,0.017359752,0.0024078337,0.00015133667],"about_ca_topic_score_codex":0.0009649515,"about_ca_topic_score_gemma":0.0010931112,"teacher_disagreement_score":0.081131235,"about_ca_system_score_codex":0.0017815757,"about_ca_system_score_gemma":0.0005380318,"threshold_uncertainty_score":0.9855265},"labels":[],"label_agreement":null},{"id":"W4401696619","doi":"10.57264/cer-2023-0161","title":"Comparative efficacy of diroximel fumarate, ozanimod and interferon beta-1a for relapsing multiple sclerosis using matching-adjusted indirect comparisons","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Sclerosis Research Studies","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":"McGill University; Ottawa Hospital; Montreal Neurological Institute and Hospital; University of Ottawa","funders":"","keywords":"Medicine; Confidence interval; Population; Internal medicine; Multiple sclerosis; Gastroenterology; Dermatology; Immunology","score_opus":0.4373288476947015,"score_gpt":0.4854581205204904,"score_spread":0.04812927282578894,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401696619","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":[0.96944034,0.009644662,0.017369233,0.0001514333,0.00032497421,0.0026718758,0.00007628976,0.000035565205,0.00028560657],"genre_scores_gemma":[0.9950126,0.00041785356,0.0041018776,0.000004580857,0.00026127664,0.00009169213,0.000012810174,0.000058283866,0.000039061815],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.99304605,0.0027641312,0.0011981835,0.0006140387,0.0015714719,0.00080611533],"domain_scores_gemma":[0.978668,0.017315,0.00042054945,0.00033116125,0.0028701194,0.00039521657],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0060310955,0.00046256545,0.0025546553,0.001739577,0.00053163216,0.00021917651,0.0003495323,0.00017874381,0.000027060954],"category_scores_gemma":[0.0008333085,0.00035490084,0.0004771248,0.0014864386,0.0011666873,0.0005460876,0.00036963317,0.0018093152,0.0000073390092],"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":[0.026257295,0.0016971829,0.018408772,0.0036371953,0.0043878774,0.000054838965,0.014938628,0.002453346,0.9226105,0.00031763184,0.0015042477,0.003732436],"study_design_scores_gemma":[0.012167724,0.009367142,0.5702271,0.01697725,0.00053833635,0.00014272008,0.00654062,0.0745241,0.30824175,0.0002325683,0.000536047,0.0005046323],"about_ca_topic_score_codex":0.00014918881,"about_ca_topic_score_gemma":0.000058600424,"teacher_disagreement_score":0.6143688,"about_ca_system_score_codex":0.00075965177,"about_ca_system_score_gemma":0.00064452074,"threshold_uncertainty_score":0.99989027},"labels":[],"label_agreement":null},{"id":"W4402206567","doi":"10.57264/cer-2024-0010","title":"One-year budget impact of InTandem™: a novel neurorehabilitation system for individuals with chronic stroke walking impairment","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Stroke Rehabilitation and Recovery","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":"EVERSANA (Canada)","funders":"","keywords":"Neurorehabilitation; Medicine; Stroke (engine); Physical medicine and rehabilitation; Chronic stroke; Physical therapy; Rehabilitation; Gerontology","score_opus":0.06186490151134039,"score_gpt":0.4369953450139674,"score_spread":0.37513044350262703,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402206567","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":[0.9788281,0.0013007174,0.017120354,0.00010761092,0.00024379666,0.0021325182,0.00006767755,0.000018552613,0.00018070595],"genre_scores_gemma":[0.9973729,0.000022188071,0.0021328318,0.000001467754,0.00028432344,0.000122337,0.000006196358,0.000028531944,0.00002918423],"study_design_codex":"bench_or_experimental","study_design_gemma":"observational","domain_scores_codex":[0.9966379,0.0009458958,0.00062251114,0.00026332744,0.0011884207,0.00034195455],"domain_scores_gemma":[0.9896589,0.0077175437,0.0002390234,0.00018525364,0.0020163842,0.00018291481],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004783587,0.00018547587,0.00089284213,0.0011869452,0.000096035044,0.00007452541,0.00013975169,0.00008973271,0.00002133494],"category_scores_gemma":[0.00038918824,0.00012012351,0.000530738,0.00075792806,0.00026947356,0.0002829091,0.000037092475,0.00065683445,0.0000044934573],"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":[0.026529688,0.0029915366,0.38723025,0.023062672,0.0074126343,0.00004683834,0.01012286,0.010752034,0.51828563,0.004694181,0.00048476714,0.008386874],"study_design_scores_gemma":[0.00705205,0.04257469,0.9059913,0.008426083,0.00020211955,0.00015181075,0.002476146,0.0028554588,0.029754814,0.00015318533,0.00021399804,0.00014831577],"about_ca_topic_score_codex":0.00002234386,"about_ca_topic_score_gemma":0.0000031924972,"teacher_disagreement_score":0.51876104,"about_ca_system_score_codex":0.0020473606,"about_ca_system_score_gemma":0.0018160284,"threshold_uncertainty_score":0.5353777},"labels":[],"label_agreement":null},{"id":"W4402428455","doi":"10.57264/cer-2024-0111","title":"A systematic review of health-related quality of life outcomes in patients with advanced breast cancer treated with palbociclib","year":2024,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Advanced Breast Cancer Therapies","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Palbociclib; Quality of life (healthcare); Clinical trial; Metastatic breast cancer; Breast cancer; Internal medicine; Randomized controlled trial; Oncology; Cancer; Physical therapy","score_opus":0.13816752987078193,"score_gpt":0.5181216320709223,"score_spread":0.37995410220014036,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4402428455","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":[0.023658324,0.9679128,0.00003393424,0.00008586799,0.000060044968,0.008013125,0.00020369173,0.000010792141,0.000021406266],"genre_scores_gemma":[0.06708774,0.93201876,0.000071686074,0.000013253139,0.000020484891,0.0006369369,0.000026722813,0.00009069682,0.000033724384],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9829451,0.009520236,0.004203045,0.00045999882,0.0023657517,0.00050591293],"domain_scores_gemma":[0.98669004,0.0033400345,0.004894536,0.0005849228,0.004222817,0.0002676518],"candidate_categories":["metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.0071310857,0.00066223356,0.015615568,0.0012701357,0.00005356151,0.000014453085,0.00040203257,0.00016460041,0.000035984838],"category_scores_gemma":[0.0005246291,0.0003279745,0.0006976433,0.0030404022,0.0005002197,0.00022391453,0.000078984165,0.0018339896,0.0000041036174],"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":[0.005264298,0.00093063316,0.014974251,0.95626676,0.006108862,0.000025510659,0.00027029798,0.000022445745,0.0000014902749,0.00002807143,0.0000057318302,0.01610164],"study_design_scores_gemma":[0.0051803566,0.002909251,0.0640571,0.9256127,0.0016368459,0.00006768248,0.00013405681,0.0000012891708,0.000005130011,0.000015019166,0.00013363035,0.00024694487],"about_ca_topic_score_codex":0.00018290684,"about_ca_topic_score_gemma":0.000027588687,"teacher_disagreement_score":0.049082845,"about_ca_system_score_codex":0.001730908,"about_ca_system_score_gemma":0.0039840196,"threshold_uncertainty_score":0.9999172},"labels":[],"label_agreement":null},{"id":"W4403038201","doi":"10.57264/cer-2024-0179","title":"Access in all areas? A round up of developments in market access and HTA: part 5","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medicine; Agency (philosophy); Regulatory agency; Expanded access; Real world evidence; Perspective (graphical); Drug approval; Public administration; Pharmacology; Oncology; Drug; Internal medicine; Political science; Social science","score_opus":0.7758501098985618,"score_gpt":0.6353775750908001,"score_spread":0.14047253480776178,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403038201","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":[0.98655295,0.0063745254,0.000517219,0.0015709939,0.0005486168,0.00079400593,0.000031633594,0.0000043513433,0.0036056996],"genre_scores_gemma":[0.9988252,0.00062963157,0.00009769344,0.00008124574,0.00010922171,0.00009309573,0.0000027788278,0.00001655152,0.00014458416],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9927391,0.003075123,0.003066027,0.0003952709,0.0003023808,0.0004221144],"domain_scores_gemma":[0.9906238,0.007850334,0.00087596476,0.00018274346,0.00031204108,0.00015515987],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.053654764,0.00016589623,0.0013722226,0.0022076403,0.000062536674,0.00034945642,0.0005756018,0.000110261426,0.00014633949],"category_scores_gemma":[0.0020827977,0.00017396762,0.00008084422,0.0010214099,0.0001645985,0.001911093,0.0002326543,0.00070272794,0.000038927654],"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":[0.000767816,0.00032872663,0.9632197,0.0027026131,0.00031191585,0.000054169188,0.005039047,0.00066950015,0.000022436978,0.019982284,0.006260597,0.00064120925],"study_design_scores_gemma":[0.0015732573,0.0001867488,0.9567868,0.0021211836,0.0000034441307,0.000021379048,0.0006035142,0.002744991,0.00015851484,0.024142623,0.011473152,0.00018438553],"about_ca_topic_score_codex":0.000532371,"about_ca_topic_score_gemma":0.00050024147,"teacher_disagreement_score":0.051571965,"about_ca_system_score_codex":0.0012986262,"about_ca_system_score_gemma":0.0006461334,"threshold_uncertainty_score":0.97446156},"labels":[],"label_agreement":null},{"id":"W4403139133","doi":"10.57264/cer-2024-0064","title":"Use of transportability methods for real-world evidence generation: a review of current applications","year":2024,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Context (archaeology); Medicine; Weighting; Data science; Medical physics; Computer science; Geography","score_opus":0.9525799695547456,"score_gpt":0.7544126675334183,"score_spread":0.19816730202132726,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403139133","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":[0.000029887658,0.9064379,0.08526436,0.00033533,0.00050190487,0.0068073473,0.00055303855,0.0000057938455,0.00006446763],"genre_scores_gemma":[0.00011152589,0.98705566,0.009898724,0.000022553126,0.0005424768,0.0022206502,0.000050190694,0.00004273656,0.000055498076],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9760545,0.011704948,0.010773649,0.000687188,0.00040336789,0.000376355],"domain_scores_gemma":[0.95089227,0.037274953,0.008138099,0.0008166606,0.0026663262,0.00021166533],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.12553419,0.00038031017,0.007662899,0.0016828557,0.00013309576,0.00006622643,0.00073689816,0.00017554621,0.00009186827],"category_scores_gemma":[0.0074615646,0.0003440621,0.001515027,0.0018262543,0.00033830907,0.0006105845,0.00007668346,0.0010725912,0.0000474098],"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":[0.00006550793,0.00037651774,0.00016850945,0.8381876,0.0008624016,5.414435e-7,0.000351636,0.000041601423,0.0000030968897,0.05887531,0.006168086,0.09489922],"study_design_scores_gemma":[0.00015464841,0.00027372208,0.00014252616,0.18382686,0.00035391754,0.0000052146956,0.000019118403,0.00022825207,0.000016752272,0.0035981238,0.81116205,0.00021878202],"about_ca_topic_score_codex":0.0001029877,"about_ca_topic_score_gemma":0.00003993278,"teacher_disagreement_score":0.804994,"about_ca_system_score_codex":0.0014934223,"about_ca_system_score_gemma":0.0026785007,"threshold_uncertainty_score":0.9999011},"labels":[],"label_agreement":null},{"id":"W4404247164","doi":"10.57264/cer-2024-0131","title":"A novel injection device to administer repository corticotropin injection for inflammatory disease treatment: findings from a market research study","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"Mallinckrodt Pharmaceuticals","keywords":"Medicine; Disease; Intensive care medicine; Pharmacology; Internal medicine; Anesthesia","score_opus":0.5685693505189394,"score_gpt":0.5743089682957716,"score_spread":0.005739617776832184,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404247164","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":[0.98095876,0.0022368294,0.009396736,0.0014087894,0.001248878,0.004121144,0.00016148703,0.000026640817,0.00044071337],"genre_scores_gemma":[0.99649894,0.000020757307,0.00037222545,0.000064401676,0.0010775358,0.0012295834,0.000008366567,0.00004555731,0.0006826557],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9904983,0.004754421,0.0028151018,0.00075146806,0.00059120986,0.0005894788],"domain_scores_gemma":[0.98168576,0.015526834,0.00060751406,0.00042739933,0.0012609455,0.00049154233],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.031544797,0.00024916738,0.0011577819,0.0022968994,0.0006576181,0.0005578728,0.00036661373,0.00012110001,0.00008187235],"category_scores_gemma":[0.0042967903,0.00025271162,0.00025545212,0.0010344932,0.0001474846,0.000863795,0.00010482798,0.00071998814,0.00026653192],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.031514503,0.009442055,0.7690189,0.0052608736,0.0054434533,0.00039986803,0.09422856,0.008298473,0.004414639,0.024819778,0.0452071,0.0019517784],"study_design_scores_gemma":[0.00437924,0.0151798,0.91681266,0.0019002294,0.00006640767,0.000065266824,0.013076356,0.014234642,0.00061382237,0.008848774,0.024295436,0.0005273607],"about_ca_topic_score_codex":0.0008594597,"about_ca_topic_score_gemma":0.00022509013,"teacher_disagreement_score":0.14779375,"about_ca_system_score_codex":0.0042328634,"about_ca_system_score_gemma":0.001236003,"threshold_uncertainty_score":0.9999925},"labels":[],"label_agreement":null},{"id":"W4404388316","doi":"10.57264/cer-2024-0118","title":"Managing the challenges of paying for gene therapy: strategies for market action and policy reform in the United States","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Biotechnology and Related Fields","field":"Medicine","cited_by":26,"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":"LEO Pharma; Otsuka Pharmaceutical; Novo Nordisk; CVS Health; Regeneron Pharmaceuticals; Bayer HealthCare; Sanofi; National Pharmaceutical Council; Centene Corporation; Mallinckrodt Pharmaceuticals; Commonwealth Fund; Alnylam Pharmaceuticals; Kaiser Permanente; GlaxoSmithKline; National Organization for Rare Disorders; Arnold Ventures; AstraZeneca; Patrick and Catherine Weldon Donaghue Medical Research Foundation","keywords":"Medicine; Payment; Finance; Business","score_opus":0.19749568484851615,"score_gpt":0.4900226394215382,"score_spread":0.292526954573022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404388316","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":[0.92977285,0.020475522,0.0044081733,0.043569095,0.00007539575,0.0011073662,0.0000039321826,0.00000720538,0.000580442],"genre_scores_gemma":[0.98063666,0.019081187,0.000062608895,0.0000331523,0.000095652635,0.00006762972,0.0000015811072,0.000007033441,0.000014513401],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99877685,0.0005892196,0.00019306665,0.00009445003,0.00017881773,0.00016761893],"domain_scores_gemma":[0.9960913,0.0033848372,0.00006632268,0.000094180614,0.0003440045,0.000019359799],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004511805,0.000077660174,0.00025042647,0.0006448304,0.00013723478,0.00003560024,0.00012197047,0.00023801967,0.000001603394],"category_scores_gemma":[0.00007049033,0.00003433126,0.00006950122,0.0005216596,0.00026595953,0.000105008665,0.000019770543,0.001375623,1.11915746e-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":[0.03880387,0.00081857544,0.0007855516,0.0070309527,0.003924967,0.00011699367,0.060371988,0.0013559908,0.081145346,0.6194157,0.0012202176,0.18500987],"study_design_scores_gemma":[0.008823717,0.014335755,0.13711347,0.0046091294,0.0001929271,0.0007508407,0.26031798,0.044938363,0.1299529,0.37934053,0.019336957,0.00028745117],"about_ca_topic_score_codex":0.00006826046,"about_ca_topic_score_gemma":0.000018549472,"teacher_disagreement_score":0.24007516,"about_ca_system_score_codex":0.000102849386,"about_ca_system_score_gemma":0.00017944376,"threshold_uncertainty_score":0.5976474},"labels":[],"label_agreement":null},{"id":"W4404765290","doi":"10.57264/cer-2024-0212","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 17","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Public Health Ontario; University of Toronto","funders":"","keywords":"Reimbursement; Medicine; Nice; Excellence; Real world evidence; Position statement; Health care; Evidence-based medicine; Health technology; Family medicine; Alternative medicine; Economic growth; Pathology","score_opus":0.7948179803753248,"score_gpt":0.6600241547357037,"score_spread":0.13479382563962106,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404765290","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":[0.90508497,0.016549831,0.024355924,0.04244199,0.0017542527,0.005234713,0.00012203355,0.00003648951,0.004419793],"genre_scores_gemma":[0.99211067,0.0006845889,0.00561418,0.00012595739,0.0001805584,0.00040235717,0.0000052136384,0.00002720096,0.00084929814],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.9915146,0.002427272,0.0045378422,0.00051659386,0.00038404172,0.000619606],"domain_scores_gemma":[0.9850745,0.012012086,0.0017463941,0.00030695554,0.0006600066,0.00020002913],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08154639,0.00018760396,0.0016768728,0.0035233398,0.00021444478,0.000116152005,0.00046224598,0.00010767698,0.000042716834],"category_scores_gemma":[0.0035538217,0.00020375155,0.0001182729,0.0019653647,0.00012669338,0.0007500637,0.000120327626,0.0008041992,0.00007497205],"study_design_candidate":"theoretical_or_conceptual","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":[0.0009388632,0.0005313893,0.26173317,0.0069189915,0.000731106,0.000018827792,0.01643386,0.0043194564,0.000320471,0.6696922,0.034605674,0.0037559476],"study_design_scores_gemma":[0.006051019,0.007972158,0.50181186,0.053679887,0.000029627621,0.000045383247,0.02194232,0.014097324,0.0013195773,0.15660352,0.23523581,0.00121154],"about_ca_topic_score_codex":0.00055294664,"about_ca_topic_score_gemma":0.0006566173,"teacher_disagreement_score":0.5130887,"about_ca_system_score_codex":0.003892481,"about_ca_system_score_gemma":0.0020700255,"threshold_uncertainty_score":0.9999314},"labels":[],"label_agreement":null},{"id":"W4405127888","doi":"10.57264/cer-2024-0133","title":"Use of individualized starting dose and niraparib hematologic adverse event management costs in ovarian cancer","year":2024,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"PARP inhibition in cancer therapy","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":"Centre Hospitalier de l’Université de Montréal; University of British Columbia","funders":"","keywords":"Medicine; Adverse effect; Neutropenia; Unit cost; Anemia; Internal medicine; Chemotherapy","score_opus":0.22567769074610858,"score_gpt":0.5142588447011285,"score_spread":0.28858115395501993,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405127888","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":[0.9925807,0.0044212854,0.00087641977,0.00038517307,0.00018361783,0.0009881093,0.000014680307,0.000008025882,0.0005419631],"genre_scores_gemma":[0.99820626,0.0008786656,0.0006547637,0.000026292864,0.000060272283,0.000109807734,0.0000020457703,0.0000144826645,0.00004742261],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963291,0.0015934963,0.00064271165,0.0002339965,0.00090269645,0.00029796053],"domain_scores_gemma":[0.9970715,0.0018840756,0.00019710879,0.00014706998,0.00056021404,0.00014004864],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0039303135,0.0001505349,0.0007114852,0.0010305818,0.000060913448,0.000043341206,0.00010230004,0.00006894626,0.00007772492],"category_scores_gemma":[0.00011047657,0.00011492242,0.000116093914,0.0009811355,0.00026043525,0.00030644157,0.00008215523,0.00073465693,0.000003929887],"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":[0.046684638,0.003247375,0.52395636,0.011660884,0.005974525,0.006365695,0.019250143,0.0069940304,0.29374573,0.039316643,0.003329106,0.039474834],"study_design_scores_gemma":[0.010956767,0.001973085,0.85631603,0.016300738,0.00014007885,0.00021138566,0.0018712005,0.0020807802,0.10363765,0.0032279275,0.0030455722,0.00023880595],"about_ca_topic_score_codex":0.00011214133,"about_ca_topic_score_gemma":0.000027665123,"teacher_disagreement_score":0.3323596,"about_ca_system_score_codex":0.0007910506,"about_ca_system_score_gemma":0.0003234611,"threshold_uncertainty_score":0.46864003},"labels":[],"label_agreement":null},{"id":"W4406964103","doi":"10.57264/cer-2024-0208","title":"Budget impact analyses of hemoglobin A1c and lipid panel point-of-care testing with Afinion™ 2 in Canada and Italy","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Abbott (Canada); EVERSANA (Canada)","funders":"","keywords":"Medicine; Point-of-care testing; Dyslipidemia; Glycated hemoglobin; Indirect costs; Health care; Emergency medicine; Internal medicine; Diabetes mellitus; Type 2 diabetes","score_opus":0.5095168284086294,"score_gpt":0.552470989915553,"score_spread":0.04295416150692355,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406964103","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":[0.9922607,0.00543513,0.00059132493,0.000486538,0.000066443834,0.00044462844,0.00006425315,0.0000013323219,0.0006496362],"genre_scores_gemma":[0.9992726,0.00006926912,0.0005609158,0.000031098214,0.00003369378,0.000014131674,0.0000025870943,0.000007677418,0.000008026853],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99570227,0.0015807357,0.001994448,0.00024348612,0.00020768969,0.00027136708],"domain_scores_gemma":[0.98753494,0.009970739,0.0013800422,0.00017508965,0.0008142348,0.00012492527],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012823795,0.00014203151,0.0015044952,0.0009843695,0.00008897336,0.000036994476,0.00018295257,0.00005152218,0.000013998388],"category_scores_gemma":[0.002703652,0.0001281167,0.00005763497,0.0008409571,0.00019662685,0.00032414627,0.00007034848,0.00037562754,0.000001151179],"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":[0.00046460974,0.00006719651,0.99084437,0.0013241982,0.00022996197,0.0000050899503,0.0014887244,0.0035208766,0.00030456722,0.001321611,0.00015453686,0.00027427997],"study_design_scores_gemma":[0.0015398172,0.00074919744,0.985483,0.0012379648,0.000009127804,0.000013651001,0.005559513,0.002188128,0.0010616056,0.0019848105,0.0000608336,0.00011237521],"about_ca_topic_score_codex":0.3017068,"about_ca_topic_score_gemma":0.10710666,"teacher_disagreement_score":0.19460014,"about_ca_system_score_codex":0.0012085707,"about_ca_system_score_gemma":0.002380405,"threshold_uncertainty_score":0.9091863},"labels":[],"label_agreement":null},{"id":"W4407666347","doi":"10.57264/cer-2024-0216","title":"The economic impact of stent retriever selection for acute ischemic stroke: a cost analysis of MASTRO I from the healthcare system perspective of the United States, Canada and eight European countries","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Acute Ischemic Stroke Management","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":"","keywords":"Medicine; Modified Rankin Scale; Solitaire Cryptographic Algorithm; Stroke (engine); Revascularization; Stent; Health care; Emergency medicine; Cohort; Surgery; Demography; Ischemic stroke; Internal medicine; Ischemia","score_opus":0.03197199617681526,"score_gpt":0.39066830819294796,"score_spread":0.3586963120161327,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407666347","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":[0.9935578,0.0011270678,0.0019089254,0.00080151926,0.000094019466,0.0014761203,0.0007708885,0.0000019987385,0.00026163785],"genre_scores_gemma":[0.9995572,0.00021529461,0.00002959823,0.000013478568,0.00004232469,0.000022436356,0.000021216016,0.000010290415,0.00008818848],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9963668,0.0018867552,0.0006459971,0.00019026913,0.00065631326,0.00025387973],"domain_scores_gemma":[0.9910826,0.0047322228,0.0007773544,0.00031641248,0.0030182605,0.00007314977],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029248372,0.00016443583,0.0008587394,0.0003886494,0.00024733646,0.000028864213,0.00037002348,0.000037810925,0.0000050213157],"category_scores_gemma":[0.00021790658,0.000080139194,0.00030620807,0.0011985049,0.0004428777,0.000059774935,0.00015409317,0.00051250233,9.844573e-8],"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":false,"study_design_scores_codex":[0.038290944,0.00018982268,0.76370525,0.0010004261,0.09772202,0.000014466294,0.0058824643,0.028770287,0.024156388,0.0022975854,0.037457038,0.0005132983],"study_design_scores_gemma":[0.0036503836,0.0015293811,0.88420814,0.0011705006,0.0037328033,0.000013612343,0.023175934,0.027333317,0.052883767,0.000023243447,0.0021721944,0.000106733736],"about_ca_topic_score_codex":0.15644576,"about_ca_topic_score_gemma":0.02680184,"teacher_disagreement_score":0.12964392,"about_ca_system_score_codex":0.004122494,"about_ca_system_score_gemma":0.0023806437,"threshold_uncertainty_score":0.9997005},"labels":[],"label_agreement":null},{"id":"W4407705966","doi":"10.57264/cer-2025-0014","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 18","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Public Health Ontario; University of Toronto","funders":"","keywords":"Real world evidence; Reimbursement; Medicine; Excellence; Real world data; Comparative effectiveness research; Health technology; Health care; Outcomes research; Evidence-based medicine; MEDLINE; Family medicine; Data science; Alternative medicine; Economic growth","score_opus":0.791885116103,"score_gpt":0.6657260335566721,"score_spread":0.1261590825463279,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407705966","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":[0.91618747,0.006089263,0.028544765,0.03507048,0.0012290254,0.0051842555,0.000067351975,0.000018611769,0.0076088104],"genre_scores_gemma":[0.9911987,0.00048289038,0.0066469368,0.00022704295,0.000083331295,0.00042252179,0.0000039582133,0.000016272139,0.0009183314],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9907835,0.0029507542,0.004866005,0.00047200435,0.00031951143,0.0006082152],"domain_scores_gemma":[0.9844365,0.011455477,0.0026407489,0.00035383448,0.0009554285,0.00015797047],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.076419964,0.00018007745,0.001868534,0.0038217502,0.0002729416,0.00006541624,0.0005374812,0.000111073925,0.000032862947],"category_scores_gemma":[0.005898116,0.00020432632,0.00010445273,0.0021987788,0.0001380521,0.000538458,0.00014965235,0.0006919733,0.00003255782],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008791217,0.0004495353,0.53065753,0.0027069838,0.00038631787,0.0000030668946,0.004349249,0.0022229247,0.00016268705,0.4416656,0.014842162,0.001674825],"study_design_scores_gemma":[0.006157929,0.0031868422,0.77815574,0.021502135,0.000015248109,0.000006815084,0.013278464,0.002826921,0.0011703657,0.10949957,0.06364133,0.00055863644],"about_ca_topic_score_codex":0.0007091518,"about_ca_topic_score_gemma":0.0007758895,"teacher_disagreement_score":0.33216602,"about_ca_system_score_codex":0.003958818,"about_ca_system_score_gemma":0.0023297586,"threshold_uncertainty_score":0.9998648},"labels":[],"label_agreement":null},{"id":"W4407745750","doi":"10.57264/cer-2024-0226","title":"The role of health economic evidence in clinical practice guidelines for colorectal cancer: a comparative analysis across countries","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":"Medicine; Guideline; Colorectal cancer; MEDLINE; Family medicine; Quality-adjusted life year; Health economics; Alternative medicine; Health care; Cancer; Cost effectiveness; Public health; Nursing; Internal medicine; Economic growth; Pathology; Risk analysis (engineering)","score_opus":0.8474889239529361,"score_gpt":0.74698083932145,"score_spread":0.10050808463148608,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407745750","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":[0.002232142,0.9858191,0.001407622,0.00424012,0.0008396254,0.004387571,0.0009053509,0.0000051691554,0.00016328454],"genre_scores_gemma":[0.005975167,0.9912119,0.00094743056,0.00016091792,0.0004960371,0.0011220609,0.000017440178,0.00002249479,0.000046557452],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9552345,0.023655998,0.01881717,0.0008699746,0.0005030755,0.00091925304],"domain_scores_gemma":[0.79841006,0.17900492,0.017189035,0.0005435328,0.0046522506,0.00020019451],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2669092,0.00046213958,0.011433767,0.001597543,0.0006637498,0.00026061965,0.0014057679,0.0003464961,0.000022343273],"category_scores_gemma":[0.045808606,0.00038573245,0.0016051423,0.0017134956,0.0007425207,0.0008279351,0.00023847472,0.0014617862,0.000050957937],"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":true,"study_design_scores_codex":[0.011114904,0.0029044175,0.0735279,0.15179534,0.061581567,0.000015106858,0.03149021,0.027681012,5.4600815e-7,0.21834217,0.06469482,0.356852],"study_design_scores_gemma":[0.0009936476,0.0010070067,0.0042488645,0.0220939,0.00032838303,0.000008027621,0.006267821,0.002363083,0.0000033750161,0.0029182928,0.9594128,0.00035476004],"about_ca_topic_score_codex":0.002037572,"about_ca_topic_score_gemma":0.0023057393,"teacher_disagreement_score":0.894718,"about_ca_system_score_codex":0.0045124097,"about_ca_system_score_gemma":0.010167189,"threshold_uncertainty_score":0.99985945},"labels":[],"label_agreement":null},{"id":"W4407884630","doi":"10.57264/cer-2024-0193","title":"Matching-adjusted indirect comparisons of efficacy outcomes between etrasimod and ozanimod for moderately to severely active ulcerative colitis","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","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":"Pfizer (Canada); Western University","funders":"","keywords":"Medicine; Ulcerative colitis; Internal medicine; Clinical trial; Placebo; Confidence interval; Disease; Pathology","score_opus":0.08833058544568213,"score_gpt":0.4340713814230241,"score_spread":0.345740795977342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407884630","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":[0.97407174,0.00013248336,0.023105897,0.00008716508,0.00010283349,0.0016969795,0.00019492871,0.00000539331,0.00060259533],"genre_scores_gemma":[0.99873734,0.0000149673615,0.0008943829,0.000022640212,0.00009011387,0.000111372225,0.000021188567,0.000022010221,0.00008599113],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.996107,0.002044705,0.000575341,0.00037788946,0.0004788155,0.0004162528],"domain_scores_gemma":[0.99490404,0.002127484,0.0002862156,0.00024686812,0.002196204,0.0002392084],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019720122,0.00027373768,0.00094683276,0.0005492815,0.0002751214,0.00006650656,0.00036756755,0.00017071882,0.0000048064567],"category_scores_gemma":[0.0005386587,0.00023986326,0.00022563082,0.00041046235,0.00029046828,0.00003098176,0.00021663491,0.00045718765,0.0000018911312],"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":[0.38369924,0.0046618953,0.38038242,0.0031532242,0.015754556,0.00017671818,0.0135038365,0.020541785,0.14124866,0.01713163,0.0071784607,0.012567585],"study_design_scores_gemma":[0.0030499802,0.0012415644,0.8509383,0.00025816838,0.00007547233,0.0000038196977,0.0006813213,0.000095663774,0.1419768,0.0012495334,0.0002529976,0.00017634932],"about_ca_topic_score_codex":0.000025285783,"about_ca_topic_score_gemma":0.00009660559,"teacher_disagreement_score":0.4705559,"about_ca_system_score_codex":0.00015005357,"about_ca_system_score_gemma":0.0007257548,"threshold_uncertainty_score":0.97813404},"labels":[],"label_agreement":null},{"id":"W4408522992","doi":"10.57264/cer-2024-0225","title":"Efficacy and safety of advanced therapies for moderately to severely active ulcerative colitis in induction and maintenance: systematic literature review and Bayesian network meta-analysis","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","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":"Pfizer (Canada); Western University","funders":"","keywords":"Medicine; Tofacitinib; Internal medicine; Maintenance therapy; Ulcerative colitis; Meta-analysis; Randomized controlled trial; Clinical trial; Placebo; Alternative medicine; Chemotherapy; Disease; Rheumatoid arthritis; Pathology","score_opus":0.08189029509954747,"score_gpt":0.4343012535341651,"score_spread":0.3524109584346176,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408522992","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":[0.002763594,0.98874635,0.0015951651,0.00003185137,0.000032972097,0.006605791,0.00019649162,0.0000014304399,0.000026376498],"genre_scores_gemma":[0.020257592,0.9782525,0.00035757452,0.00001924589,0.000052356652,0.00093566003,0.0000409611,0.00002022348,0.000063912616],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9910448,0.00680823,0.0010120381,0.0005053535,0.00031934166,0.00031020565],"domain_scores_gemma":[0.9950421,0.0020253952,0.00074627314,0.00027589026,0.0017525937,0.00015774491],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0042185737,0.00042238785,0.0048501757,0.0006228186,0.00014586962,0.00007628734,0.00020579454,0.0002383937,0.000002827411],"category_scores_gemma":[0.0007731048,0.00028714465,0.00065255514,0.0011211287,0.00019825916,0.000036325644,0.00015275709,0.0005539382,1.10156556e-7],"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":[0.023898687,0.0001925561,0.000015600837,0.84482294,0.111897126,0.00006733042,0.0011168964,0.0010963015,0.00003086747,0.00069127243,0.00017104554,0.0159994],"study_design_scores_gemma":[0.0065588276,0.003941823,0.055064064,0.7062285,0.21337283,0.00028016494,0.0008414334,0.000135329,0.00023854816,0.0012183785,0.010419313,0.0017007978],"about_ca_topic_score_codex":0.0000016479447,"about_ca_topic_score_gemma":0.00003671107,"teacher_disagreement_score":0.13859443,"about_ca_system_score_codex":0.00012020206,"about_ca_system_score_gemma":0.0004408501,"threshold_uncertainty_score":0.9999581},"labels":[],"label_agreement":null},{"id":"W4409666926","doi":"10.57264/cer-2024-0210","title":"A comparison of healthcare resource utilization and costs between patients with chronic lymphocytic leukemia treated with first-line ibrutinib or acalabrutinib using two large US real-world databases","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Chronic Lymphocytic Leukemia 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":false,"ca_institutions":"Group for Research in Decision Analysis","funders":"Janssen Scientific Affairs","keywords":"Ibrutinib; Chronic lymphocytic leukemia; Medicine; Business; Internal medicine; Leukemia","score_opus":0.19096360683406274,"score_gpt":0.5109227883965701,"score_spread":0.31995918156250736,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409666926","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":[0.99326015,0.0014403784,0.0023319377,0.000058761692,0.000030216117,0.002399739,0.00009751635,0.000023002733,0.00035833014],"genre_scores_gemma":[0.9989651,0.00019300093,0.0003645401,0.000012274968,0.00013568616,0.00003973758,0.00013476014,0.00004940217,0.00010550083],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928799,0.0024276704,0.001107587,0.00058219116,0.0021056377,0.0008969885],"domain_scores_gemma":[0.98781854,0.006930376,0.00068106,0.00057804893,0.0035536522,0.00043830962],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002911821,0.00041920043,0.0019803683,0.0015757601,0.00047647738,0.000069674556,0.0003383009,0.00012425435,0.00003212701],"category_scores_gemma":[0.0006111784,0.0002686428,0.0000980383,0.0034821697,0.0008762699,0.00032903661,0.00030269637,0.0010948908,0.0000021060255],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.046870556,0.0005344444,0.9407492,0.0032170976,0.0009806334,0.00013191554,0.00044952505,0.0041389833,0.00033668152,0.00025959153,0.00015233245,0.002179027],"study_design_scores_gemma":[0.09140444,0.011475924,0.7854002,0.018803291,0.0005017947,0.00008030851,0.0005739055,0.057908207,0.03208694,0.00001391242,0.0014065207,0.0003445522],"about_ca_topic_score_codex":0.00510824,"about_ca_topic_score_gemma":0.0051043546,"teacher_disagreement_score":0.15534902,"about_ca_system_score_codex":0.05696269,"about_ca_system_score_gemma":0.037689425,"threshold_uncertainty_score":0.9999766},"labels":[],"label_agreement":null},{"id":"W4409760204","doi":"10.57264/cer-2025-0024","title":"Time-limited reimbursement and Temporary Access Process for early access to oncology treatments in Canada: a perspective based on the epcoritamab experience","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Global Affairs Canada","funders":"","keywords":"Reimbursement; Medicine; Notice; Family medicine; Health care; Negotiation; Orphan drug; Business; Economic growth; Political science; Bioinformatics","score_opus":0.7302214171912577,"score_gpt":0.6442866495813444,"score_spread":0.08593476760991325,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409760204","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":[0.086149566,0.8798561,0.0007704054,0.009516813,0.00073896995,0.019549469,0.00068916107,0.0000101673,0.0027193597],"genre_scores_gemma":[0.5231193,0.44918853,0.00024641067,0.002683465,0.00087703206,0.023144178,0.000066759254,0.00018408612,0.0004902819],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9885321,0.006232945,0.0033860465,0.000799401,0.00044271734,0.00060678116],"domain_scores_gemma":[0.963723,0.031905834,0.002617657,0.0004468275,0.001035433,0.00027122992],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.022583226,0.00043294887,0.004219128,0.0020190298,0.00034394025,0.00026791263,0.0014376902,0.00019488925,0.000055394805],"category_scores_gemma":[0.0060071955,0.00034901386,0.00025160733,0.0015025304,0.00017797417,0.0005909287,0.0002112258,0.0009647798,0.000021955717],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.02815334,0.016587548,0.18374501,0.28907746,0.01743885,0.0008036117,0.0915914,0.029799502,0.0000033377814,0.067309864,0.11314178,0.1623483],"study_design_scores_gemma":[0.018603187,0.016436312,0.1422811,0.17552945,0.00055365026,0.00006906278,0.03425231,0.012842364,0.00008555829,0.019870339,0.57533187,0.0041448213],"about_ca_topic_score_codex":0.0720942,"about_ca_topic_score_gemma":0.024383178,"teacher_disagreement_score":0.46219006,"about_ca_system_score_codex":0.014389833,"about_ca_system_score_gemma":0.013295061,"threshold_uncertainty_score":0.99989617},"labels":[],"label_agreement":null},{"id":"W4410077388","doi":"10.57264/cer-2025-0009","title":"Network connectivity, between-study heterogeneity and timepoint challenges in generalized myasthenia gravis: a feasibility assessment of indirect treatment comparisons","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Myasthenia Gravis and Thymoma","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":"Ottawa Hospital; EVERSANA (Canada)","funders":"","keywords":"Medicine; Dosing; Placebo; Clinical trial; Randomized controlled trial; Population; Clinical study design; Intensive care medicine; Internal medicine; Alternative medicine","score_opus":0.49096799424601734,"score_gpt":0.5658844709347464,"score_spread":0.07491647668872903,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410077388","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":[0.33749658,0.65700424,0.00007966036,0.000033606568,0.00009607324,0.005007565,0.00003785455,0.000007701625,0.00023675161],"genre_scores_gemma":[0.50442433,0.49490955,0.0002121876,8.5888587e-7,0.00015509404,0.00025014533,0.000012956981,0.000026689546,0.000008157445],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.97618234,0.018944884,0.0020263463,0.00083646557,0.001324489,0.0006854799],"domain_scores_gemma":[0.9885415,0.008018274,0.001232267,0.00073660247,0.0011224345,0.00034894253],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.012036798,0.00077434874,0.00987833,0.0016068991,0.000210688,0.0000626628,0.000396639,0.00033941038,0.000012677287],"category_scores_gemma":[0.0002538319,0.00052496936,0.0009085086,0.0014055724,0.00039832166,0.000104774896,0.00036574033,0.0019064163,0.0000018303334],"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":[0.0035806135,0.014211768,0.64455456,0.040120713,0.009056974,0.0006143281,0.0015615646,0.00009213677,0.000021621818,0.000660542,0.00007453079,0.28545064],"study_design_scores_gemma":[0.0110965585,0.020456849,0.90162396,0.049709998,0.0027143585,0.00025285775,0.00054214895,0.00006764949,0.00007127081,0.00067823584,0.012231934,0.0005541457],"about_ca_topic_score_codex":0.00011639219,"about_ca_topic_score_gemma":0.0003490891,"teacher_disagreement_score":0.2848965,"about_ca_system_score_codex":0.0016206994,"about_ca_system_score_gemma":0.002273545,"threshold_uncertainty_score":0.9997202},"labels":[],"label_agreement":null},{"id":"W4410237684","doi":"10.57264/cer-2025-0023","title":"Overall survival in patients with metastatic castration-sensitive prostate cancer treated with apalutamide versus abiraterone acetate: a head-to-head analysis of real-world patients in the USA","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Prostate Cancer Treatment and Research","field":"Medicine","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":"Group for Research in Decision Analysis","funders":"","keywords":"Abiraterone acetate; Medicine; Hazard ratio; Prostate cancer; Confidence interval; Proportional hazards model; Cohort; Internal medicine; Oncology; Urology; Gynecology; Cancer; Androgen deprivation therapy","score_opus":0.09190869354461881,"score_gpt":0.45211489764668883,"score_spread":0.36020620410207005,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410237684","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":[0.9963112,0.000072955525,0.00004405456,0.00019241092,0.000046228288,0.0028079657,0.00007275393,0.0000029117848,0.00044954312],"genre_scores_gemma":[0.99941593,0.00007472432,0.00007217493,0.000013164581,0.000012374929,0.00021035859,0.000059489365,0.000015157854,0.00012661722],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99380493,0.00295649,0.00064147695,0.00036359552,0.0016951535,0.00053836644],"domain_scores_gemma":[0.9941044,0.0021258853,0.00027375965,0.00026899276,0.0030812442,0.00014570652],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0021672514,0.00027782208,0.0013702348,0.0029377432,0.00011380487,0.000057035842,0.00020563867,0.000041786687,0.000016561347],"category_scores_gemma":[0.0001013178,0.00014866459,0.0001177491,0.006838413,0.00034628087,0.00025346535,0.000049625713,0.00077522337,0.0000018572705],"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":[0.12042104,0.001736934,0.86030406,0.00016670472,0.0032094256,0.00013365582,0.0026067242,0.010511022,0.00033786832,0.000054866516,0.000011823779,0.0005058966],"study_design_scores_gemma":[0.027510053,0.007649101,0.9587023,0.0011564919,0.0006100675,6.1664366e-7,0.0006432751,0.00040209474,0.0031822661,0.000015139434,0.00000770616,0.0001208921],"about_ca_topic_score_codex":0.0069377436,"about_ca_topic_score_gemma":0.050069254,"teacher_disagreement_score":0.09839825,"about_ca_system_score_codex":0.0013008107,"about_ca_system_score_gemma":0.000985778,"threshold_uncertainty_score":0.99967515},"labels":[],"label_agreement":null},{"id":"W4410311277","doi":"10.57264/cer-2025-0063","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 19","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Public Health Ontario; University of Toronto","funders":"","keywords":"Medicine; Health technology; Reimbursement; Real world evidence; Checklist; Health economics; Natural experiment; Inflation (cosmology); Health care; Actuarial science; Public health; Nursing; Economic growth; Pathology; Economics","score_opus":0.7745056326895687,"score_gpt":0.663189442031888,"score_spread":0.11131619065768072,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410311277","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":[0.9057731,0.006785748,0.031133078,0.044280842,0.0011760119,0.0052541215,0.000070619964,0.000018844705,0.0055076],"genre_scores_gemma":[0.9910032,0.00052759214,0.0067650196,0.00028339276,0.00008133521,0.0004350045,0.000004299421,0.000015946913,0.00088417187],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99069333,0.0030517417,0.004852789,0.00047542967,0.0003207353,0.0006059527],"domain_scores_gemma":[0.9835944,0.012337032,0.0026523606,0.00035488367,0.0008780435,0.000183254],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07919394,0.0001796336,0.0018573218,0.0038984877,0.00028404972,0.000063917156,0.00053854205,0.00011104021,0.000022286948],"category_scores_gemma":[0.007021311,0.000203345,0.000104344974,0.0022196975,0.00013687242,0.00052071974,0.00015012991,0.0006909334,0.000030489491],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007632374,0.00037573496,0.54068273,0.0025119882,0.00032692656,0.0000028146867,0.004422187,0.0022776967,0.00011871841,0.43573603,0.0115134865,0.0012684303],"study_design_scores_gemma":[0.005944032,0.0028896956,0.77394617,0.01802528,0.00001419341,0.000007088178,0.013672847,0.002535368,0.0008178097,0.111595936,0.070026055,0.00052554754],"about_ca_topic_score_codex":0.000932269,"about_ca_topic_score_gemma":0.0009542221,"teacher_disagreement_score":0.32414007,"about_ca_system_score_codex":0.0043670107,"about_ca_system_score_gemma":0.0026611371,"threshold_uncertainty_score":0.99945503},"labels":[],"label_agreement":null},{"id":"W4412155371","doi":"10.57264/cer-2024-0237","title":"Matching-adjusted indirect comparison between garadacimab and lanadelumab for the long-term prophylactic treatment of patients with hereditary angioedema","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Coagulation, Bradykinin, Polyphosphates, and Angioedema","field":"Medicine","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Hereditary angioedema; Prophylactic treatment; Angioedema; Term (time); Dermatology; Surgery; Internal medicine","score_opus":0.09743824719382202,"score_gpt":0.418657052657323,"score_spread":0.321218805463501,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412155371","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":[0.98922914,0.0056250873,0.0026298044,0.00022193862,0.000106215884,0.0020705936,0.000024804183,0.0000076249476,0.00008479028],"genre_scores_gemma":[0.99945754,0.00004024398,0.00015349299,0.000005489152,0.00015217329,0.00006152001,0.000024438235,0.00001624927,0.00008884505],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9976567,0.00063895836,0.000509612,0.00021818529,0.000678332,0.00029819537],"domain_scores_gemma":[0.9937278,0.004324508,0.00033195846,0.0002101255,0.0012877414,0.00011784061],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009594057,0.00021237168,0.0010778006,0.0004334118,0.00028134685,0.00004144417,0.00014578854,0.00007885059,0.000009475184],"category_scores_gemma":[0.000074752286,0.000118934695,0.00016775931,0.0005296215,0.00033289907,0.00014840592,0.00003052432,0.0002693644,7.3465236e-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":[0.008754459,0.00096756575,0.97094476,0.00068113516,0.0014998267,0.000015823922,0.0019738197,0.00010246141,0.00014844727,0.00009464891,0.000069513306,0.014747567],"study_design_scores_gemma":[0.0078124255,0.0070438716,0.97389644,0.00076242344,0.00049025053,0.000012592825,0.0005019043,0.00019183819,0.008868176,0.00022250682,0.00009959425,0.000097985416],"about_ca_topic_score_codex":0.00009806133,"about_ca_topic_score_gemma":0.000059368092,"teacher_disagreement_score":0.014649581,"about_ca_system_score_codex":0.00028537062,"about_ca_system_score_gemma":0.0004330303,"threshold_uncertainty_score":0.48500162},"labels":[],"label_agreement":null},{"id":"W4412155435","doi":"10.57264/cer-2025-0053","title":"Crossing borders: the need for empirical evidence of real-world evidence transportability in oncology","year":2025,"lang":"en","type":"editorial","venue":"Journal of Comparative Effectiveness Research","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":"Daiichi Sankyo Europe; Flatiron Health; Gilead Sciences; Pfizer","keywords":"Real world evidence; Medicine; Comparative effectiveness research; MEDLINE; Oncology; Medical physics; Internal medicine; Alternative medicine; Pathology","score_opus":0.7535354596229806,"score_gpt":0.6749517803646315,"score_spread":0.07858367925834908,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412155435","genre_codex":"editorial","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":[0.40684453,0.059038684,0.01506108,0.029856361,0.47351652,0.012934598,0.0008051303,0.000030056399,0.001913045],"genre_scores_gemma":[0.8683473,0.0024875053,0.001041319,0.00016801039,0.12671459,0.00082625664,0.000020214691,0.000056017656,0.00033878075],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9768451,0.012765856,0.007945594,0.000809918,0.00087321573,0.0007603264],"domain_scores_gemma":[0.7433851,0.24666071,0.005167526,0.0006782734,0.003938753,0.00016967862],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.18969253,0.0003579551,0.0041867755,0.0018897607,0.00042856735,0.00020117781,0.0014320221,0.0006341145,0.000054138454],"category_scores_gemma":[0.089621454,0.00032219046,0.0005667831,0.0015759679,0.0011377648,0.0008949347,0.00012310379,0.0029048128,0.000014089],"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":true,"study_design_scores_codex":[0.010842476,0.0010699633,0.12381683,0.015776036,0.00070670445,0.000012247612,0.021177044,0.003707194,0.000033986275,0.005415922,0.8170725,0.00036907796],"study_design_scores_gemma":[0.005490971,0.004927363,0.19279262,0.03221154,0.00010225942,0.0000027156773,0.0053525832,0.0026066608,0.00021850821,0.034064468,0.72138584,0.0008444798],"about_ca_topic_score_codex":0.0030468225,"about_ca_topic_score_gemma":0.0040836576,"teacher_disagreement_score":0.4615028,"about_ca_system_score_codex":0.0051360796,"about_ca_system_score_gemma":0.011699316,"threshold_uncertainty_score":0.999923},"labels":[],"label_agreement":null},{"id":"W4412590588","doi":"10.57264/cer-2025-0113","title":"R WE ready for reimbursement? A round up of developments in real-world evidence relating to health technology assessment: part 20","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Kingston Health Sciences Centre; Public Health Ontario; University of Toronto","funders":"","keywords":"Medicine; Real world evidence; Reimbursement; Agency (philosophy); Health technology; Position statement; Statement (logic); Linkage (software); Systematic review; Real world data; Food and drug administration; Family medicine; MEDLINE; Health care; Data science; Environmental health; Economic growth; Law","score_opus":0.8514725016603549,"score_gpt":0.6930565999238583,"score_spread":0.15841590173649667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412590588","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":[0.0005827259,0.98119384,0.0035471402,0.004445287,0.00096217607,0.007249548,0.0002180754,0.000011860137,0.0017893591],"genre_scores_gemma":[0.0023046567,0.988761,0.005277413,0.000083454994,0.0002770941,0.0017133154,0.00003335395,0.00005100752,0.0014987026],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.97985905,0.0076234634,0.010278981,0.00084102084,0.00050447625,0.000893018],"domain_scores_gemma":[0.96165675,0.028212884,0.008152266,0.000596982,0.00112427,0.0002568778],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.10606972,0.0004319447,0.0073048864,0.00719664,0.00030325944,0.000097523174,0.0010503874,0.0003167177,0.000030236268],"category_scores_gemma":[0.009133329,0.00045870905,0.00037991945,0.003399969,0.00016109664,0.00054141035,0.00028208096,0.0016045622,0.00006445443],"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":true,"study_design_scores_codex":[0.0009004034,0.0011671187,0.028534563,0.3229789,0.0030072455,0.000020376667,0.006487883,0.0010654989,0.0000029026205,0.40311077,0.034764413,0.19795991],"study_design_scores_gemma":[0.0012148966,0.0011854449,0.0019234655,0.17405213,0.000036743368,0.0000076636015,0.0012603055,0.00008903242,0.0000041046455,0.007599284,0.81222063,0.0004062697],"about_ca_topic_score_codex":0.00044147528,"about_ca_topic_score_gemma":0.00069046445,"teacher_disagreement_score":0.7774562,"about_ca_system_score_codex":0.007861566,"about_ca_system_score_gemma":0.007832641,"threshold_uncertainty_score":0.99978644},"labels":[],"label_agreement":null},{"id":"W4413103009","doi":"10.57264/cer-2024-0180","title":"Identification of treatment effect modifiers and prognostic factors in newly diagnosed and relapsed or refractory multiple myeloma","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Multiple Myeloma Research and Treatments","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":"EVERSANA (Canada)","funders":"","keywords":"Medicine; Internal medicine; Multiple myeloma; Oncology; Population; Progression-free survival; Refractory (planetary science); Randomized controlled trial; Overall survival","score_opus":0.08580465074146751,"score_gpt":0.439554261778722,"score_spread":0.35374961103725444,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413103009","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":[0.99590707,0.0014644252,0.00019410717,0.00005376091,0.00006382834,0.0022119798,0.000010781182,0.000005801661,0.00008826004],"genre_scores_gemma":[0.99915266,0.00048421317,0.000063457876,0.0000014657772,0.000020288015,0.00013901843,0.000006904191,0.000011727214,0.00012027713],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99595696,0.0022645087,0.0005726569,0.00029286725,0.0005747471,0.00033825537],"domain_scores_gemma":[0.98605853,0.012653226,0.00021137566,0.00020275918,0.0006506755,0.00022343884],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0021379767,0.00022087802,0.0009570255,0.0012463711,0.00012625949,0.00004454127,0.00010343671,0.00010687436,0.0000075391977],"category_scores_gemma":[0.002873271,0.00013150058,0.00010186278,0.0007703018,0.000397933,0.00020476781,0.00006159808,0.0004825346,0.0000014851104],"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":[0.017653294,0.000982404,0.8792899,0.00065665797,0.00058778346,0.00025313228,0.0012033494,0.000060182538,0.086347915,0.00003604338,0.000017006467,0.012912338],"study_design_scores_gemma":[0.014556328,0.009022584,0.86624134,0.0009930973,0.000076977856,0.000021953818,0.0005930096,0.0017522277,0.106551245,0.00008352272,0.000036284455,0.00007140604],"about_ca_topic_score_codex":0.00022804952,"about_ca_topic_score_gemma":0.0001031674,"teacher_disagreement_score":0.02020333,"about_ca_system_score_codex":0.000723801,"about_ca_system_score_gemma":0.0005858571,"threshold_uncertainty_score":0.5362438},"labels":[],"label_agreement":null},{"id":"W4413400952","doi":"10.57264/cer-2025-0041","title":"Transportability of nonlocal real-world evidence and its relevance to health technology assessment: a primer","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Public Health Ontario; University of Toronto; University of Calgary; AstraZeneca (Canada); Toronto Public Health","funders":"","keywords":"Excellence; Agency (philosophy); Health technology; Medicine; Health care; Benchmarking; Relevance (law); Health information technology; Demographics; Evidence-based practice; Public relations; Medical education; Alternative medicine; Political science; Business; Marketing; Social science; Sociology","score_opus":0.7203703134267602,"score_gpt":0.6673336547915845,"score_spread":0.053036658635175704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413400952","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":[0.0032166264,0.98459667,0.0039362484,0.0031770763,0.00033073893,0.0040803603,0.00019938848,0.000011615972,0.00045125224],"genre_scores_gemma":[0.028732486,0.96907085,0.0014225633,0.00007160047,0.00014344662,0.00036615127,0.000005928178,0.00003107734,0.0001559239],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9819906,0.0081798965,0.007825375,0.0008604697,0.00048180987,0.00066184474],"domain_scores_gemma":[0.96561134,0.026504258,0.0055848327,0.00063788664,0.0012939607,0.00036769616],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.082353994,0.000412168,0.007945935,0.004008372,0.0002472824,0.00005419323,0.00088490505,0.0003029039,0.00005280632],"category_scores_gemma":[0.0076452484,0.0004129937,0.0004310505,0.0028416466,0.0003280088,0.0004791694,0.00017897504,0.0018159819,0.000055221237],"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":[0.00069904054,0.0017505058,0.021721046,0.47955835,0.002267977,0.000026353013,0.0029671085,0.0003689699,0.000005647563,0.3669978,0.0033958727,0.12024131],"study_design_scores_gemma":[0.0017476565,0.003975751,0.034657996,0.24746479,0.00016337998,0.000049310434,0.0008652493,0.00045427433,0.000035154215,0.011756387,0.6977357,0.0010943477],"about_ca_topic_score_codex":0.00033505558,"about_ca_topic_score_gemma":0.0002506408,"teacher_disagreement_score":0.6943398,"about_ca_system_score_codex":0.0029520383,"about_ca_system_score_gemma":0.0055779223,"threshold_uncertainty_score":0.9998322},"labels":[],"label_agreement":null},{"id":"W4414477652","doi":"10.57264/cer-2025-0125","title":"The cost-effectiveness of real-time continuous glucose monitoring versus intermittently scanned continuous glucose monitoring in individuals with insulin-treated Type 2 diabetes mellitus in Canada","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","topic":"Diabetes Management and Research","field":"Medicine","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":"Continuous glucose monitoring; Glycemic; Type 2 Diabetes Mellitus; Blood Glucose Self-Monitoring; Type 2 diabetes; Microsimulation; Benzhydryl compounds; Diabetes mellitus","score_opus":0.04761061976653703,"score_gpt":0.3797619372710775,"score_spread":0.33215131750454047,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414477652","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":[0.9891603,0.0061030807,0.000007894052,0.00007883081,0.00060247246,0.0028078575,0.000007185655,0.000010705794,0.001221665],"genre_scores_gemma":[0.99846447,0.0010600002,0.00003708985,0.000001454421,0.000099460434,0.00018592516,0.0000061378305,0.00003228365,0.000113192174],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99116904,0.0047514266,0.0009320053,0.00041923192,0.0016934279,0.0010348754],"domain_scores_gemma":[0.98329824,0.013033892,0.00038161283,0.00042960298,0.0026290424,0.00022760186],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0087909475,0.00036490586,0.0016632958,0.0014409743,0.00021083423,0.00013268615,0.00068494654,0.000121430916,0.00000751574],"category_scores_gemma":[0.0011873099,0.0002452757,0.00013034228,0.0032033566,0.0004764361,0.00027959986,0.00031114937,0.001533147,0.000004601769],"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":[0.020802695,0.00060974516,0.9223481,0.0007148812,0.0012251375,0.00031125685,0.00026483877,0.00081217027,0.034184266,0.00005246969,0.000035671732,0.018638788],"study_design_scores_gemma":[0.010041349,0.0024334416,0.73720664,0.006314193,0.00007448751,0.0000020098494,0.0015654082,0.00026351144,0.24174604,0.00006481972,0.0001320524,0.00015607226],"about_ca_topic_score_codex":0.029978136,"about_ca_topic_score_gemma":0.008725975,"teacher_disagreement_score":0.20756176,"about_ca_system_score_codex":0.002707508,"about_ca_system_score_gemma":0.0025164355,"threshold_uncertainty_score":0.99999994},"labels":[],"label_agreement":null},{"id":"W4414753286","doi":"10.57264/cer-2025-0148","title":"R WE ready for reimbursement? A round-up of developments in real-world evidence relating to health technology assessment: part 21","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":true,"ca_institutions":"Public Health Ontario; University of Toronto","funders":"","keywords":"Health technology; Emulation; Frame (networking); Comparative effectiveness research; Health information technology; MEDLINE; Technology assessment; Real world evidence; Alternative medicine","score_opus":0.7496154867001127,"score_gpt":0.6542149322955111,"score_spread":0.0954005544046016,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414753286","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":[0.9064381,0.006188697,0.030684734,0.04440791,0.0012377582,0.00518957,0.0000666634,0.000019456978,0.0057671047],"genre_scores_gemma":[0.9907923,0.0005164346,0.007031185,0.00024366115,0.00008535226,0.000446162,0.0000041490107,0.00001672298,0.0008640796],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9905123,0.0030678601,0.004978643,0.00048383017,0.00033067324,0.0006267001],"domain_scores_gemma":[0.98425084,0.011537751,0.0026939479,0.00036549175,0.0009896473,0.00016232353],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07744048,0.00018493531,0.0019185753,0.0038783858,0.00027768992,0.00006535538,0.0005556865,0.00011518522,0.00002278905],"category_scores_gemma":[0.0061034868,0.00020939582,0.00010942445,0.0022663905,0.00013736781,0.0005478227,0.00015432051,0.0007173013,0.00003403113],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0008356355,0.00044176154,0.5004242,0.0026165897,0.00036384596,0.0000029640203,0.0046318644,0.002142186,0.00016416846,0.47660178,0.009971102,0.0018039038],"study_design_scores_gemma":[0.0062546814,0.0032748778,0.77992433,0.022637809,0.000015731854,0.0000072314083,0.01430715,0.0029386103,0.0012268096,0.106661715,0.062171966,0.000579069],"about_ca_topic_score_codex":0.00075378484,"about_ca_topic_score_gemma":0.0008811225,"teacher_disagreement_score":0.36994007,"about_ca_system_score_codex":0.004083026,"about_ca_system_score_gemma":0.0024102756,"threshold_uncertainty_score":0.9997401},"labels":[],"label_agreement":null},{"id":"W4414847741","doi":"10.57264/cer-2025-0029","title":"Estimating per-protocol effects in external comparator analyses using real-world data","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Statistical Methods in Clinical Trials","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":"Public Health Ontario; University of British Columbia","funders":"","keywords":"Emulation; Protocol (science); Randomized controlled trial; Clinical trial; Estimation; Research design; Comparator","score_opus":0.9569598007759129,"score_gpt":0.8145488453194007,"score_spread":0.14241095545651217,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414847741","genre_codex":"review","genre_gemma":"methods","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":[0.0003044505,0.4601847,0.41786152,0.000008119915,0.0015393363,0.11869996,0.0002178632,0.00004432052,0.0011397342],"genre_scores_gemma":[0.000061189865,0.08346219,0.9018487,0.0000038102287,0.001504925,0.012943582,0.000006091583,0.00011698508,0.00005257131],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.92521954,0.065600045,0.0045953,0.0010978556,0.002520218,0.0009670315],"domain_scores_gemma":[0.6130826,0.3809762,0.0027847704,0.0014122269,0.0014416386,0.0003025599],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05236287,0.000903824,0.011477893,0.0033605043,0.00028424044,0.00034329624,0.0034737417,0.0003683151,0.00006983357],"category_scores_gemma":[0.09448211,0.0006466719,0.0009070826,0.0041914876,0.0008793617,0.0004087211,0.0017738935,0.0052053737,0.000022693464],"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":[0.0031446698,0.0029458094,0.00062207103,0.35881606,0.0033851082,0.0014261442,0.00018465932,0.0008503123,0.00010320541,0.018571835,0.0014518873,0.6084982],"study_design_scores_gemma":[0.0058819787,0.0018189236,0.00079991,0.5612027,0.0029668235,0.00016493359,0.000086039,0.017660001,0.00020948132,0.39450148,0.013258701,0.0014490295],"about_ca_topic_score_codex":0.00015565843,"about_ca_topic_score_gemma":0.000073569914,"teacher_disagreement_score":0.6070492,"about_ca_system_score_codex":0.0015301104,"about_ca_system_score_gemma":0.0032754666,"threshold_uncertainty_score":0.99959844},"labels":[],"label_agreement":null},{"id":"W4414883284","doi":"10.57264/cer-2025-0130","title":"Blood neurofilament light chain as a predictive biomarker for functional outcome of acute ischemic stroke: a systematic review and meta-analysis","year":2025,"lang":"en","type":"review","venue":"Journal of Comparative Effectiveness Research","topic":"Acute Ischemic Stroke Management","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":"Funnel plot; Meta-analysis; Stroke (engine); Subgroup analysis; Biomarker; Odds ratio; Publication bias; Logistic regression","score_opus":0.21129556084385506,"score_gpt":0.48919316455284734,"score_spread":0.2778976037089923,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414883284","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":[0.00006693427,0.9802403,0.0038315856,0.0002443695,0.00008231893,0.013964109,0.00041864632,0.000010033551,0.0011417271],"genre_scores_gemma":[0.0013016167,0.9911153,0.0002816443,0.00005943437,0.000057480716,0.0035306492,0.00008139684,0.000048811504,0.0035237065],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9893433,0.0043813344,0.002953535,0.00073000835,0.0020912238,0.0005006192],"domain_scores_gemma":[0.9848059,0.008225083,0.0026298426,0.0007892097,0.0032697353,0.00028024067],"candidate_categories":["metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.00904115,0.000776056,0.018119004,0.0029296489,0.00012832019,0.000051819454,0.00056012376,0.00025980198,0.00010243393],"category_scores_gemma":[0.001525969,0.00046897685,0.005999614,0.0025682936,0.00026486153,0.00014100834,0.00043817182,0.0013377921,0.0000056608737],"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":[0.0005154032,0.0003112133,0.000022708711,0.54774565,0.4492207,0.0000592281,0.000027946324,0.0000011889775,0.000050701943,0.000035515877,0.001925587,0.000084138745],"study_design_scores_gemma":[0.0013148693,0.0018135974,0.00007369977,0.09024399,0.89277846,0.00030252812,0.000058769194,0.000020127343,0.00023917854,0.0000058382593,0.012909844,0.00023911608],"about_ca_topic_score_codex":0.0000056802996,"about_ca_topic_score_gemma":6.540904e-7,"teacher_disagreement_score":0.45750168,"about_ca_system_score_codex":0.00046844478,"about_ca_system_score_gemma":0.000913532,"threshold_uncertainty_score":0.9997762},"labels":[],"label_agreement":null},{"id":"W4415260542","doi":"10.57264/cer-2025-0149","title":"R WE ready for reimbursement? A round-up of developments in real-world evidence relating to health technology assessment: part 22","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Kingston Health Sciences Centre; Public Health Ontario","funders":"","keywords":"Formulary; Health technology; Clinical trial; Health information technology; Health economics; MEDLINE; Comparative effectiveness research; Health data","score_opus":0.744417899109262,"score_gpt":0.6539982115717788,"score_spread":0.0904196875374832,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415260542","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":[0.91482455,0.005765086,0.029000696,0.037628032,0.0011954993,0.0050450084,0.00006559729,0.00001916029,0.00645636],"genre_scores_gemma":[0.9909728,0.00049768947,0.006847228,0.00024350014,0.00008404356,0.00045766577,0.000004163265,0.000016793729,0.0008761284],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"observational","domain_scores_codex":[0.990583,0.0030171282,0.004959819,0.00048225877,0.00033302035,0.0006247782],"domain_scores_gemma":[0.98430103,0.011498784,0.0026979086,0.00036453112,0.0009768348,0.00016093586],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07541417,0.00018443499,0.0019069036,0.0039239586,0.0002789594,0.00006446639,0.0005597031,0.00011343431,0.000028230468],"category_scores_gemma":[0.005976022,0.00020916662,0.00010898403,0.0022714771,0.00013680378,0.0005494895,0.00015594182,0.00071896426,0.00003373601],"study_design_candidate":"observational","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":[0.0008940831,0.0004474379,0.46671665,0.002614346,0.00036586222,0.0000029571129,0.0042354674,0.0021968754,0.00017998961,0.5088335,0.011788163,0.0017246173],"study_design_scores_gemma":[0.006393571,0.0029298633,0.7787765,0.021303833,0.000015406029,0.000007063689,0.0136308065,0.0029364906,0.0011121022,0.109061114,0.063264705,0.0005684982],"about_ca_topic_score_codex":0.0007559458,"about_ca_topic_score_gemma":0.00076432363,"teacher_disagreement_score":0.39977244,"about_ca_system_score_codex":0.004131789,"about_ca_system_score_gemma":0.0023810714,"threshold_uncertainty_score":0.9996912},"labels":[],"label_agreement":null},{"id":"W4417317269","doi":"10.57264/cer-2025-0196","title":"R WE ready for reimbursement? A round-up of developments in real-world evidence relating to health technology assessment: part 23","year":2025,"lang":"en","type":"article","venue":"Journal of Comparative Effectiveness Research","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":"Kingston Health Sciences Centre; Public Health Ontario; University of Toronto","funders":"","keywords":"Transformative learning; Health technology; MEDLINE; Health information technology; Clinical trial; Evidence-based medicine; Real world evidence","score_opus":0.6211705210896415,"score_gpt":0.6597066368166193,"score_spread":0.03853611572697779,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417317269","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":[0.6083534,0.0010423193,0.37552878,0.0030450213,0.0003961417,0.0062063057,0.000015230612,0.00008716319,0.0053256294],"genre_scores_gemma":[0.92515284,0.00030389446,0.073386885,0.000014930979,0.000027834009,0.00049794454,0.0000015970924,0.000019261102,0.0005948446],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9947508,0.0020589856,0.0014162472,0.00034019788,0.00083842635,0.0005953293],"domain_scores_gemma":[0.9842002,0.012314993,0.0007916934,0.0003570694,0.0022243322,0.0001117382],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.012851226,0.00022107291,0.0011965161,0.0026489673,0.00019538426,0.000039767157,0.0006027595,0.000110928835,0.0000063957714],"category_scores_gemma":[0.0033299942,0.00019485758,0.000086463864,0.002967677,0.00019911272,0.00042281215,0.0002762152,0.0010825358,0.0000020440748],"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":[0.0026845115,0.0009473848,0.062707596,0.0028523318,0.00041481742,0.000024822371,0.0032929927,0.00042381402,0.02720253,0.8764282,0.004693152,0.018327883],"study_design_scores_gemma":[0.0026804516,0.0048424113,0.04353916,0.03715877,0.000035280424,0.000010887425,0.004945456,0.00035731294,0.1273435,0.7751019,0.0035936295,0.0003912329],"about_ca_topic_score_codex":0.00007511421,"about_ca_topic_score_gemma":0.00044127394,"teacher_disagreement_score":0.3167994,"about_ca_system_score_codex":0.002204538,"about_ca_system_score_gemma":0.0018089096,"threshold_uncertainty_score":0.79460615},"labels":[],"label_agreement":null}]}