{"meta":{"query_hash":"88f8afce41e1","filters":{"venue":"Clinical Epidemiology"},"cohort_total":77,"direct_labels_cover":5,"predictions_cover":77,"exported":77,"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/88f8afce41e1","api":"https://metacan.xera.ac/api/v1/cohort?venue=Clinical+Epidemiology"},"results":[{"id":"W1963974031","doi":"10.2147/clep.s15574","title":"Tuberculosis mortality in HIV-infected individuals: a cross-national systematic assessment","year":2011,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Simon Fraser University; Ottawa Hospital; University of British Columbia; AIDS Vancouver; University of Ottawa","funders":"","keywords":"Tuberculosis; Medicine; Mortality rate; Demography; Population; Latin Americans; Human immunodeficiency virus (HIV); Public health; Gerontology; Environmental health; Immunology; Internal medicine; Pathology","score_opus":0.37051312444403056,"score_gpt":0.5507288884534598,"score_spread":0.18021576400942924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1963974031","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97927773,0.0006163347,0.0012027506,0.0048379484,0.0004584947,0.0014482994,0.00005033935,0.00014240494,0.011965696],"genre_scores_gemma":[0.9790387,0.0006925688,0.00793364,0.011232154,0.00032502538,0.0004236843,0.00017201202,0.00003251139,0.00014968497],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9831343,0.009628355,0.0047350447,0.0010094941,0.00038864717,0.0011041376],"domain_scores_gemma":[0.972898,0.024228243,0.0007837627,0.0008649887,0.00043497424,0.00079002074],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.040480733,0.00034567597,0.0034119403,0.0003438524,0.000081756385,0.000007410523,0.0003640274,0.0009286534,0.00088758],"category_scores_gemma":[0.14378192,0.000258626,0.00069610623,0.00043710897,0.0009261376,0.0001283578,0.00024907786,0.0016099035,0.0006190446],"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":[0.00011216912,0.00097431464,0.98304945,0.0014846853,0.0005184354,0.000056021654,0.00004663558,0.000015841433,0.0000067491264,0.01017833,0.003408693,0.00014868926],"study_design_scores_gemma":[0.0019476067,0.0006077253,0.97797644,0.00088448205,0.0001504589,0.00008321552,0.000024928444,0.003982077,0.0000032795685,0.013903454,0.0002257171,0.00021060235],"about_ca_topic_score_codex":0.0005053845,"about_ca_topic_score_gemma":0.00011899207,"teacher_disagreement_score":0.10330118,"about_ca_system_score_codex":0.00028559729,"about_ca_system_score_gemma":0.00052486063,"threshold_uncertainty_score":0.9999866},"labels":[],"label_agreement":null},{"id":"W1966844935","doi":"10.2147/clep.s29470","title":"Stability of additive treatment effects in multiple treatment comparison meta-analysis: a simulation study","year":2012,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; University of Ottawa","funders":"","keywords":"Medicine; Meta-analysis; Stability (learning theory); Statistics; Internal medicine; Computer science; Mathematics; Machine learning","score_opus":0.9660057181583439,"score_gpt":0.7021483838145138,"score_spread":0.2638573343438301,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1966844935","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":[0.9697591,0.0030265935,0.023505917,0.00025507915,0.00024062616,0.0029334244,0.000059649417,0.0000060423945,0.00021353434],"genre_scores_gemma":[0.9961085,0.00001964059,0.0030149233,0.00014486462,0.000079742014,0.00041276435,0.00003087525,0.000007536893,0.00018119405],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.7412486,0.21975051,0.035202052,0.0018228755,0.0013633813,0.0006125574],"domain_scores_gemma":[0.45588368,0.52390844,0.014102974,0.005034186,0.00064654753,0.00042419357],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.29712877,0.0004685118,0.027569,0.0004365448,0.00006487703,0.000022173228,0.00065703056,0.00025041073,0.0043845912],"category_scores_gemma":[0.3819275,0.00019039163,0.013204331,0.0015873321,0.00021668337,0.00014379513,0.00010083519,0.00017804855,0.0006098389],"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":[0.00005880212,0.005305972,0.9213491,0.000005891706,0.054273564,0.0000010402844,0.0008836486,0.009388067,4.370992e-7,0.00012547604,0.00012463219,0.008483378],"study_design_scores_gemma":[0.0008452571,0.0014108958,0.80133706,0.0000011529143,0.09191245,1.5505022e-7,0.00070899253,0.101194404,0.0000053001177,0.00068289635,0.0017814824,0.00011992295],"about_ca_topic_score_codex":0.00039069811,"about_ca_topic_score_gemma":0.0019081584,"teacher_disagreement_score":0.3041579,"about_ca_system_score_codex":0.00015708647,"about_ca_system_score_gemma":0.000051017152,"threshold_uncertainty_score":0.9966782},"labels":[],"label_agreement":null},{"id":"W1968931110","doi":"10.2147/clep.s48870","title":"A sensitivity analysis of the Children's Treatment Network trial: a randomized controlled trial of integrated services versus usual care for children with special health care needs","year":2013,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Healthcare Policy and Management","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"St. Joseph’s Healthcare Hamilton","funders":"","keywords":"Psychosocial; Medicine; Randomized controlled trial; Confidence interval; Health care; Intervention (counseling); Quality of life (healthcare); Family medicine; Protocol (science); Pediatrics; Physical therapy; Nursing; Psychiatry; Alternative medicine; Internal medicine","score_opus":0.11376698316519847,"score_gpt":0.39614803330880843,"score_spread":0.28238105014361,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1968931110","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9731694,0.0017246226,0.0017042771,0.005353922,0.0018540046,0.015039511,0.00078004994,0.000024454379,0.00034979393],"genre_scores_gemma":[0.9927771,0.00082411425,0.00093960867,0.0017372591,0.0020401606,0.0010409626,0.0005627498,0.000023791943,0.000054245484],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9904723,0.0037055404,0.004595408,0.0005904834,0.00005448058,0.0005817745],"domain_scores_gemma":[0.98210996,0.012793033,0.0040014395,0.00071924145,0.00018977317,0.0001865557],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009151147,0.00032441542,0.008383177,0.00030305394,0.00015145278,0.000011104723,0.0002595681,0.00034796484,0.000069186106],"category_scores_gemma":[0.0059254086,0.00019880652,0.002537341,0.00069401646,0.00040566555,0.000047545065,0.00008410034,0.00024728925,0.000008563536],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.9199472,0.00032553088,0.042901427,0.000039580143,0.00897672,6.015197e-8,0.0009477231,0.0015179943,8.631085e-9,0.021143446,0.00029944573,0.003900882],"study_design_scores_gemma":[0.9470514,0.00367972,0.043673012,0.000023871768,0.0013147127,2.2059682e-7,0.00029117876,0.0010476115,1.2492109e-7,0.0009143593,0.0018275125,0.00017624843],"about_ca_topic_score_codex":0.04180914,"about_ca_topic_score_gemma":0.015970822,"teacher_disagreement_score":0.027104245,"about_ca_system_score_codex":0.00020611566,"about_ca_system_score_gemma":0.000259969,"threshold_uncertainty_score":0.96457154},"labels":[],"label_agreement":null},{"id":"W1973038563","doi":"10.2147/clep.s30848","title":"Validity and reproducibility of a physical activity questionnaire for older adults: questionnaire versus accelerometer for assessing physical activity in older adults","year":2012,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Physical Activity and Health","field":"Medicine","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Infection and Immunity","funders":"Universiteit van Amsterdam","keywords":"Medicine; Reproducibility; Physical activity; Accelerometer; Physical therapy; Gerontology; Physical medicine and rehabilitation; Statistics; Computer science","score_opus":0.2791106709609,"score_gpt":0.5175820421354762,"score_spread":0.23847137117457623,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1973038563","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98485523,0.00008484841,0.0076364432,0.0041265273,0.0008264604,0.0023215527,0.000052347405,0.00007755574,0.000019029394],"genre_scores_gemma":[0.9955275,0.00003600893,0.0013077767,0.00029390358,0.0023758092,0.0003845333,0.000025154874,0.000037380265,0.000011900945],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9939406,0.002228741,0.0009367261,0.0018051317,0.00017923588,0.00090954965],"domain_scores_gemma":[0.9746611,0.022481758,0.0008033624,0.0012134678,0.00028981097,0.0005505084],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.009396257,0.0003835235,0.0028152538,0.00008964297,0.00014102097,0.0000069695375,0.00014499275,0.00055528263,0.0000036482254],"category_scores_gemma":[0.07079369,0.0003160498,0.0007214569,0.00022732759,0.00079436036,0.0006452529,0.0001517286,0.0010721014,0.0000034419807],"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":[0.018980892,0.030460563,0.62321347,0.003396585,0.000103888306,0.000001293792,0.0008712204,0.0000035879173,0.00039270482,0.0010773805,0.0002631379,0.3212353],"study_design_scores_gemma":[0.008333252,0.0024027978,0.97299254,0.0010520457,0.00019468722,0.000003940462,0.00004610411,0.011448539,0.00093622773,0.0022746834,0.000044591554,0.00027056536],"about_ca_topic_score_codex":0.00028487822,"about_ca_topic_score_gemma":0.00004710858,"teacher_disagreement_score":0.3497791,"about_ca_system_score_codex":0.00019146009,"about_ca_system_score_gemma":0.0002348512,"threshold_uncertainty_score":0.99992913},"labels":[],"label_agreement":null},{"id":"W1974686295","doi":"10.2147/clep.s70853","title":"Incorporating alternative design clinical trials in network meta-analyses","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Centre for Advancing Health Outcomes; University of British Columbia","funders":"","keywords":"Medicine; Meta-analysis; Clinical trial; Computational biology; Bioinformatics; Internal medicine; Biology","score_opus":0.9952555166874054,"score_gpt":0.7905783194865695,"score_spread":0.20467719720083588,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974686295","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007446253,0.0022071346,0.97654647,0.0065853796,0.0023533478,0.0012598946,0.0000066666294,0.000014073414,0.0035807726],"genre_scores_gemma":[0.6096548,0.00029398085,0.3585813,0.02422737,0.0044150413,0.00022801243,0.000016452379,0.000038058242,0.0025449786],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.033736456,0.8301659,0.13019963,0.003094867,0.0020092323,0.00079386716],"domain_scores_gemma":[0.0063165,0.923483,0.06253609,0.005860742,0.0011432655,0.0006603723],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.9760873,0.00063153205,0.056083992,0.00038077327,0.00014490129,0.00018034397,0.003579867,0.00079117314,0.0071803248],"category_scores_gemma":[0.9858185,0.00025896984,0.02220736,0.0016957991,0.00071223837,0.00019559218,0.00047581803,0.0013901106,0.0052539944],"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":[0.0000929164,0.00019485569,0.61538416,0.000009644664,0.0071982928,0.000010916279,0.00002256005,0.03978454,0.0000011461466,0.03397816,0.2295316,0.07379119],"study_design_scores_gemma":[0.00062245014,0.0003621643,0.052278515,0.000018740935,0.005018998,0.0000044438457,0.00003491223,0.27558196,6.599856e-7,0.6171934,0.048598815,0.0002849817],"about_ca_topic_score_codex":0.00005406536,"about_ca_topic_score_gemma":0.000096554955,"teacher_disagreement_score":0.61796516,"about_ca_system_score_codex":0.000026750437,"about_ca_system_score_gemma":0.00016296541,"threshold_uncertainty_score":0.99998623},"labels":[],"label_agreement":null},{"id":"W1991403673","doi":"10.2147/clep.s44273","title":"Interpreting discordant indirect and multiple treatment comparison meta-analyses: an evaluation of direct acting antivirals for chronic hepatitis C infection","year":2013,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Hepatitis C virus research","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":"Ottawa Hospital; University of Ottawa","funders":"Merck Sharp and Dohme","keywords":"Medicine; Chronic hepatitis; Meta-analysis; Hepatitis C; Virology; Internal medicine; Virus","score_opus":0.76599077077572,"score_gpt":0.6362892942978923,"score_spread":0.1297014764778277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1991403673","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9902567,0.001570852,0.0045086746,0.000746116,0.00017338582,0.0021810806,0.000027217226,0.00004317853,0.0004927547],"genre_scores_gemma":[0.9933525,0.0013540211,0.0042068353,0.00019757054,0.00019840652,0.00051418657,0.000120130724,0.000026588712,0.000029777022],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9902916,0.006289907,0.002138681,0.00063692144,0.00022290293,0.00042001475],"domain_scores_gemma":[0.9760235,0.021938544,0.00090300717,0.00044654417,0.00045291154,0.00023550872],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.014442956,0.00023069212,0.002728109,0.0001718556,0.00009968539,0.000009098315,0.00007567596,0.00021226738,0.00024406936],"category_scores_gemma":[0.07650204,0.00015512611,0.0006168509,0.00014621926,0.000317472,0.0001934971,0.000068329515,0.0002899724,0.000017310618],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000076756434,0.00025346738,0.7230284,0.00005290479,0.0016862676,5.413826e-7,0.000080501384,0.0001619649,0.0009543945,0.0000053087656,0.00017272815,0.27352673],"study_design_scores_gemma":[0.0019142113,0.0048978827,0.4859785,0.000072039795,0.0041411472,0.000002694789,0.000030976385,0.4991868,0.0024511274,0.00031350693,0.0008753285,0.00013574323],"about_ca_topic_score_codex":0.010482319,"about_ca_topic_score_gemma":0.004791578,"teacher_disagreement_score":0.49902484,"about_ca_system_score_codex":0.0002888605,"about_ca_system_score_gemma":0.00022542386,"threshold_uncertainty_score":0.996107},"labels":[],"label_agreement":null},{"id":"W1999391077","doi":"10.2147/clep.s69660","title":"Bias in identification of the best treatment in a Bayesian network meta-analysis for binary outcome: a simulation study","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":82,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Natural Sciences and Engineering Research Council of Canada; Canadian Institutes of Health Research; McMaster University","keywords":"Rank (graph theory); Statistics; Sample size determination; Bayesian probability; Meta-analysis; Computer science; Identification (biology); Bayesian network; Outcome (game theory); Mathematics; Medicine; Biology; Internal medicine","score_opus":0.9612837799759205,"score_gpt":0.691199470085655,"score_spread":0.27008430989026555,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1999391077","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7505909,0.0006468885,0.23999545,0.0045171864,0.00040798166,0.0037186053,0.00001576044,0.0000031481468,0.00010404685],"genre_scores_gemma":[0.9960603,0.000008100878,0.0019445955,0.0005378856,0.00007273684,0.0003321388,0.000007964056,0.000006730601,0.001029557],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.79512715,0.15667014,0.04510938,0.0016653554,0.0010361931,0.0003917885],"domain_scores_gemma":[0.6268255,0.3462957,0.019978747,0.0061501213,0.0006152603,0.00013466834],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.50097686,0.00025829772,0.013652899,0.0004555674,0.000069624126,0.0000363095,0.001219829,0.00020919739,0.00044922944],"category_scores_gemma":[0.45964304,0.0000991436,0.009785567,0.0029651793,0.00013081128,0.00006716685,0.000112285656,0.00017037903,0.00010732873],"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":[0.000014446298,0.00041774943,0.7379698,0.0000045548495,0.005614016,1.7412414e-7,0.00006912064,0.25325114,1.6969074e-7,0.00055625214,0.00012570464,0.0019768653],"study_design_scores_gemma":[0.00026739112,0.00019889689,0.550824,0.0000012686704,0.021235747,7.513106e-8,0.00012452524,0.41490233,7.05972e-8,0.011467801,0.00092375604,0.000054173815],"about_ca_topic_score_codex":0.00014649982,"about_ca_topic_score_gemma":0.0026980732,"teacher_disagreement_score":0.24546935,"about_ca_system_score_codex":0.000041682637,"about_ca_system_score_gemma":0.000036106678,"threshold_uncertainty_score":0.9869006},"labels":[],"label_agreement":null},{"id":"W2000894790","doi":"10.2147/clep.s24818","title":"Characteristics of Dutch and Swiss primary care COPD patients - baseline data of the ICE COLD ERIC study","year":2011,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":29,"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":"Lunge Zürich; Universität Zürich; University of Toronto; National Science Foundation; Universiteit van Amsterdam; McGill University; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung","keywords":"Medicine; COPD; Exacerbation; Cohort; Quality of life (healthcare); Internal medicine; Obstructive lung disease; Prospective cohort study; Physical therapy; Cohort study; Disease; Intensive care medicine","score_opus":0.224250863836034,"score_gpt":0.4326623001468718,"score_spread":0.2084114363108378,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2000894790","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9959044,0.0009923445,0.000033324184,0.00032972943,0.00040550964,0.0012051769,0.00070834195,0.0000127013145,0.00040847668],"genre_scores_gemma":[0.99740463,0.00035435733,0.00071321375,0.0008765394,0.00016116614,0.00001725377,0.0003703808,0.000023261922,0.00007917208],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99444354,0.0023895982,0.0019193037,0.00066232076,0.00024426988,0.0003409903],"domain_scores_gemma":[0.9912166,0.0054686987,0.0007553947,0.0017948634,0.00048737787,0.00027707644],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003981218,0.00018274576,0.0014798412,0.00005192448,0.00004957494,0.0000010285681,0.0006129395,0.00021332242,0.00013316324],"category_scores_gemma":[0.022383487,0.00012288516,0.00014508894,0.00018659148,0.0011892833,0.000052424715,0.001340086,0.00056703354,0.000008466012],"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":[0.0008154946,0.0018948708,0.98388004,0.0007746681,0.00014172426,0.000013206992,0.00015588356,9.027628e-8,0.000016584554,0.000085667976,0.00088771776,0.011334075],"study_design_scores_gemma":[0.002428302,0.00097178324,0.9947469,0.000117265285,0.00030848794,0.00000330564,0.00025821154,0.00021304691,0.000008254075,0.00009328991,0.0007584387,0.00009269691],"about_ca_topic_score_codex":0.0002273818,"about_ca_topic_score_gemma":0.000023156332,"teacher_disagreement_score":0.01840227,"about_ca_system_score_codex":0.00006641045,"about_ca_system_score_gemma":0.00038003505,"threshold_uncertainty_score":0.9858514},"labels":[],"label_agreement":null},{"id":"W2002049303","doi":"10.2147/clep.s16526","title":"Multiple treatment comparison meta-analyses: a step forward into complexity","year":2011,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":101,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Centre for Advancing Health Outcomes; University of British Columbia; University of Ottawa","funders":"","keywords":"Meta-analysis; Context (archaeology); Medicine; Transparency (behavior); Clinical trial; Sample size determination; Comparative effectiveness research; Computer science; Management science; Data science; Alternative medicine; Pathology; Statistics","score_opus":0.9913104801979001,"score_gpt":0.7335427350719073,"score_spread":0.2577677451259929,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2002049303","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15205535,0.008134438,0.8036221,0.0068969578,0.0021079243,0.00275516,0.00008599695,0.00006551745,0.024276523],"genre_scores_gemma":[0.7685176,0.00009105989,0.22358814,0.0035245891,0.00021465906,0.00017748975,0.000031385665,0.000019573481,0.0038354704],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.79089403,0.15099567,0.052290622,0.00309678,0.0019026488,0.0008202313],"domain_scores_gemma":[0.7076791,0.25203893,0.025948206,0.011593117,0.0015588065,0.0011818362],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.3540503,0.0006706758,0.028165597,0.00030119932,0.00022926797,0.00008200861,0.0028711688,0.0004561704,0.036480904],"category_scores_gemma":[0.49000147,0.0002633568,0.018243624,0.0008341929,0.00090343924,0.00014934552,0.00037114022,0.00044418385,0.015627796],"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":[0.00007462104,0.0007775819,0.84170514,0.000013975331,0.012511957,0.000007829902,0.00040803044,0.00006386321,0.0000019570625,0.010495816,0.0989988,0.034940418],"study_design_scores_gemma":[0.0010096577,0.0011664873,0.26648524,0.00000652346,0.016502041,0.000010296261,0.0006718054,0.091904394,0.00001015693,0.18111794,0.440626,0.00048948],"about_ca_topic_score_codex":0.0010240783,"about_ca_topic_score_gemma":0.0020302632,"teacher_disagreement_score":0.6164623,"about_ca_system_score_codex":0.00007036346,"about_ca_system_score_gemma":0.000094314804,"threshold_uncertainty_score":0.9999819},"labels":[],"label_agreement":null},{"id":"W2005204638","doi":"10.2147/clep.s30621","title":"Heart disease as a risk factor for dementia","year":2013,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":234,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; McGill University; Ontario Stroke Network; Ottawa Hospital","funders":"","keywords":"Medicine; Dementia; Life expectancy; Context (archaeology); Disease; valvular heart disease; Coronary artery disease; Atrial fibrillation; Myocardial infarction; Risk factor; Intensive care medicine; Heart failure; Cardiology; Internal medicine; Environmental health; Population","score_opus":0.1797232788210484,"score_gpt":0.5135392433290075,"score_spread":0.33381596450795903,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2005204638","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96049637,0.00064480887,0.0057473453,0.029649267,0.00037075052,0.0019794432,0.00006722717,0.00007292601,0.0009718384],"genre_scores_gemma":[0.966609,0.00023245832,0.0035097105,0.026343228,0.0005221238,0.00050262484,0.00004837895,0.00002204764,0.0022104504],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99640584,0.0011794762,0.0010766317,0.00055075897,0.00012833324,0.00065893214],"domain_scores_gemma":[0.9886034,0.009591905,0.00019702036,0.00038118716,0.0002662171,0.00096026214],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0030976709,0.00016015294,0.00081980025,0.00006349285,0.00010186489,0.000008965313,0.00012099332,0.00017993135,0.0063711395],"category_scores_gemma":[0.051071443,0.000116832365,0.00052910554,0.00007283354,0.00038825333,0.0000662191,0.00009825967,0.00041169906,0.0027862352],"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":[0.00042988683,0.000359113,0.92966586,0.000042619187,0.00019906177,0.0000047497724,0.000004899168,2.8376456e-7,0.000029354904,0.00046702725,0.03303183,0.03576529],"study_design_scores_gemma":[0.002075362,0.0017327737,0.91951966,0.000024611873,0.0002327619,0.0000043263026,0.000016323984,0.0007388058,0.000011460618,0.026134545,0.04940316,0.000106212115],"about_ca_topic_score_codex":0.000076176264,"about_ca_topic_score_gemma":0.0000022110246,"teacher_disagreement_score":0.047973774,"about_ca_system_score_codex":0.000018942852,"about_ca_system_score_gemma":0.00018627077,"threshold_uncertainty_score":0.9979902},"labels":[],"label_agreement":null},{"id":"W2024530015","doi":"10.2147/clep.s66677","title":"Quality assessment of observational studies in a drug-safety systematic review, comparison of two tools: the Newcastle–Ottawa Scale and the RTI item bank","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":435,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Ottawa Hospital Research Institute","keywords":"Inter-rater reliability; Statistic; Medicine; Cohen's kappa; Scale (ratio); Percentile; Statistics; Rating scale; Mathematics","score_opus":0.9662587831656922,"score_gpt":0.7377302309802874,"score_spread":0.2285285521854048,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2024530015","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5992967,0.1619303,0.113254115,0.10604285,0.0018809496,0.011311412,0.00008757287,0.000015679077,0.0061804666],"genre_scores_gemma":[0.97865367,0.0031305796,0.010343154,0.00674478,0.00011028645,0.0002193991,0.000011277991,0.0000085825895,0.0007782976],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.39670017,0.5038176,0.093597285,0.0017922213,0.0035296346,0.0005630556],"domain_scores_gemma":[0.08742234,0.8718832,0.033255145,0.005313523,0.001992162,0.00013362113],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7397061,0.00029468437,0.01949222,0.000086977474,0.00013016287,0.00003189757,0.0018516076,0.0001236299,0.00023314821],"category_scores_gemma":[0.81453484,0.00009697976,0.0023523013,0.0007755723,0.0017237173,0.00010000433,0.00044089506,0.0004793727,0.00006673176],"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":[0.000021048254,0.00011496463,0.8535742,0.014706303,0.00042391158,8.628559e-8,0.00040143554,0.00017416269,8.765656e-7,0.108337365,0.020644518,0.001601119],"study_design_scores_gemma":[0.0014879424,0.0000980968,0.84541476,0.010292976,0.0013034855,0.000005608358,0.0016514724,0.050380208,3.5569835e-7,0.07981381,0.009326367,0.00022493069],"about_ca_topic_score_codex":0.00007656838,"about_ca_topic_score_gemma":0.00076697796,"teacher_disagreement_score":0.37935698,"about_ca_system_score_codex":0.000030609895,"about_ca_system_score_gemma":0.0000867661,"threshold_uncertainty_score":0.99506444},"labels":[],"label_agreement":null},{"id":"W2030916112","doi":"10.2147/clep.s72282","title":"Sepsis in Canadian children: a national analysis using administrative data","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Sepsis Diagnosis 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":true,"ca_institutions":"BC Children's Hospital; University of British Columbia; Alberta Children's Hospital; University of Calgary","funders":"","keywords":"Medicine; Sepsis; Pediatrics; Cohort; Population; Odds ratio; Odds; Psychological intervention; Emergency medicine; Intensive care medicine; Internal medicine; Logistic regression; Environmental health","score_opus":0.669073383099606,"score_gpt":0.5938584463520648,"score_spread":0.0752149367475412,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2030916112","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97840166,0.0003122744,0.00039402736,0.018019225,0.0001518899,0.00027685138,0.0001655208,0.000017491355,0.002261088],"genre_scores_gemma":[0.9742628,0.00009950118,0.005888172,0.018470744,0.00024993197,0.000012145661,0.0009856308,0.000008885705,0.000022229744],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99647933,0.0011562094,0.0011613066,0.0006770844,0.00010343364,0.00042262548],"domain_scores_gemma":[0.99073887,0.0077275746,0.00023090687,0.0006594828,0.00008533746,0.0005578562],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0064097713,0.00014157164,0.0012095978,0.00030678295,0.00005070677,0.0000038411586,0.00019810165,0.00026967452,0.00031916623],"category_scores_gemma":[0.032967884,0.000113201306,0.000259343,0.00046871748,0.00018971185,0.000041578834,0.00006982813,0.00029181654,0.00005991618],"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":[0.00002336567,0.00024580045,0.99178207,0.0000018291872,0.000898121,0.000010061214,0.000008358576,0.00013808557,2.0451338e-7,0.0018582153,0.00274151,0.002292372],"study_design_scores_gemma":[0.0008464074,0.0002206675,0.95084995,0.000017430524,0.0006384373,0.000012157378,0.000007834385,0.04177574,7.462518e-7,0.0009213041,0.0046196007,0.00008973408],"about_ca_topic_score_codex":0.10409457,"about_ca_topic_score_gemma":0.2759101,"teacher_disagreement_score":0.17181554,"about_ca_system_score_codex":0.00019961948,"about_ca_system_score_gemma":0.0006827903,"threshold_uncertainty_score":0.9751778},"labels":[],"label_agreement":null},{"id":"W2033335157","doi":"10.2147/clep.s68889","title":"Global prevalence and incidence of traumatic spinal cord injury","year":2014,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Spinal Cord Injury Research","field":"Medicine","cited_by":956,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Toronto Western Hospital","funders":"","keywords":"Medicine; Incidence (geometry); Systematic review; MEDLINE; Demography; Environmental health; Poison control","score_opus":0.46473361140431524,"score_gpt":0.63412993612777,"score_spread":0.16939632472345478,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2033335157","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":[0.0023931698,0.9944872,0.00049558096,0.000294381,0.0004680957,0.001286955,0.00005204157,0.000039328803,0.00048321247],"genre_scores_gemma":[0.00077277434,0.9942633,0.0032874758,0.0007991414,0.0005460568,0.00008497696,0.000022022225,0.000033260145,0.0001910199],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98737776,0.0054649916,0.0051054996,0.0010973592,0.00029110405,0.0006633117],"domain_scores_gemma":[0.9814025,0.014452872,0.0020228217,0.0012331493,0.00024431676,0.0006443257],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.015557463,0.00048673354,0.0091963075,0.00014838247,0.00004161562,0.0000039528595,0.0006225048,0.0013110547,0.00016831348],"category_scores_gemma":[0.10530931,0.00034666382,0.0011146994,0.00042240848,0.002154598,0.000040570885,0.00048378273,0.0015678572,0.00018552894],"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":[0.0016622299,0.00009836933,0.025915805,0.064604245,0.00014880594,0.000018334498,3.918653e-7,1.1954972e-8,2.8417377e-8,0.0010915531,0.0015991097,0.9048611],"study_design_scores_gemma":[0.0005639139,0.05139685,0.0848722,0.059723638,0.002496789,0.0008982416,0.0000029298722,0.000062031504,4.3558146e-8,0.0031019973,0.7964283,0.00045305057],"about_ca_topic_score_codex":0.00004569849,"about_ca_topic_score_gemma":0.0000054972384,"teacher_disagreement_score":0.90440804,"about_ca_system_score_codex":0.0001256399,"about_ca_system_score_gemma":0.00082448864,"threshold_uncertainty_score":0.99998546},"labels":[],"label_agreement":null},{"id":"W2070568765","doi":"10.2147/clep.s30816","title":"Matched case-control studies: a review of reported statistical methodology","year":2012,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":116,"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":"Confounding; Medicine; Case-control study; Statistics; Statistical analysis; Statistical significance; Meta-analysis; Sample size determination; Disease; Internal medicine; Mathematics","score_opus":0.9765472903717273,"score_gpt":0.7807672798787588,"score_spread":0.19578001049296845,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2070568765","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[1.096266e-7,0.7418706,0.24940798,0.00040030802,0.004085169,0.002901687,0.0009643173,0.00012602995,0.00024383934],"genre_scores_gemma":[9.194628e-8,0.5190393,0.4771798,0.0020070581,0.0011753492,0.00039492658,0.000022642238,0.00011487629,0.00006599121],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.7235433,0.230509,0.04097937,0.0025626188,0.00048452834,0.001921193],"domain_scores_gemma":[0.11666592,0.86123765,0.017077334,0.0032510208,0.0008209656,0.0009471245],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","research_integrity"],"category_scores_codex":[0.30210567,0.0014249585,0.053946927,0.00022300023,0.00009251901,0.000003954613,0.0011335397,0.004014986,0.0029223217],"category_scores_gemma":[0.9785821,0.0009532587,0.0050138594,0.00063443754,0.0037646135,0.000048479644,0.00070481957,0.004215289,0.00028932324],"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":[0.00005511284,0.00031683335,0.000024155841,0.22857969,0.0048459424,0.00085153687,0.0000066279376,1.3767717e-8,3.9656936e-9,0.09425359,0.0140220225,0.6570445],"study_design_scores_gemma":[0.0006988685,0.0004374469,0.000006215487,0.047420833,0.032967214,0.0032458962,0.000010739483,0.0000043424034,1.5967842e-8,0.5401315,0.37449762,0.00057931023],"about_ca_topic_score_codex":0.000016666287,"about_ca_topic_score_gemma":0.000004440566,"teacher_disagreement_score":0.6764764,"about_ca_system_score_codex":0.00018384302,"about_ca_system_score_gemma":0.00074564357,"threshold_uncertainty_score":0.99985003},"labels":[],"label_agreement":null},{"id":"W2087350855","doi":"10.2147/clep.s37357","title":"Gallbladder cancer: epidemiology and outcome","year":2014,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Cholangiocarcinoma and Gallbladder Cancer Studies","field":"Medicine","cited_by":1063,"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; Gallbladder cancer; Malignancy; Gallbladder; Cholecystectomy; Cancer; Gallstones; Dysplasia; Epidemiology; Internal medicine; Biliary tract; Carcinoma; Gastroenterology","score_opus":0.5159708001379594,"score_gpt":0.5842238537031021,"score_spread":0.06825305356514277,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2087350855","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":[0.000017641525,0.9841024,0.00020138542,0.010294214,0.003044485,0.0010631697,0.000055874374,0.00014260353,0.0010782062],"genre_scores_gemma":[0.000022175338,0.9702446,0.0014498679,0.02111566,0.0043709944,0.0004290615,0.00015514858,0.00011569743,0.0020968295],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9830356,0.006441379,0.0071672606,0.0019580137,0.00010944461,0.0012882587],"domain_scores_gemma":[0.94061595,0.054829348,0.0022896242,0.0012449304,0.0001324044,0.0008877659],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.014176963,0.0009878187,0.02122022,0.00022587045,0.00014430234,0.000004431928,0.0003734089,0.0023900222,0.00030574633],"category_scores_gemma":[0.050570972,0.00065883144,0.002899544,0.00018551263,0.001744934,0.000031489733,0.00052864366,0.0027070662,0.0004464959],"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":[0.00003627131,0.000055365945,0.23542792,0.012103725,0.0010909633,0.00002737871,0.0000062957974,1.2330669e-7,2.6618798e-9,0.0012683119,0.060963146,0.6890205],"study_design_scores_gemma":[0.000829419,0.0005365174,0.029407227,0.0044378573,0.004343098,0.0002507437,0.0000042897714,0.000017268208,2.486168e-9,0.0010303218,0.9586256,0.00051763834],"about_ca_topic_score_codex":0.00018858917,"about_ca_topic_score_gemma":0.0000631633,"teacher_disagreement_score":0.89766246,"about_ca_system_score_codex":0.00014738087,"about_ca_system_score_gemma":0.00042120976,"threshold_uncertainty_score":0.99959373},"labels":[],"label_agreement":null},{"id":"W2089714348","doi":"10.2147/clep.s53302","title":"SVR12 is higher than SVR24 in treatment-naïve hepatitis C genotype 1 patients treated with peginterferon plus ribavirin","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Hepatitis C virus research","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; University of British Columbia; McMaster University","funders":"","keywords":"Ribavirin; Genotype; Medicine; Virology; Hepatitis C; Internal medicine; Gastroenterology; Chronic hepatitis; Biology; Genetics; Gene; Virus","score_opus":0.21469649018181272,"score_gpt":0.459312472612921,"score_spread":0.2446159824311083,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2089714348","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9866376,0.00014879477,0.0003784556,0.004210197,0.0003566268,0.0009552685,0.00005882716,0.00011261209,0.0071416255],"genre_scores_gemma":[0.98486197,0.0009522525,0.0014923982,0.005031816,0.00043780415,0.00015322416,0.00037586343,0.00008829764,0.006606375],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9920032,0.0033229326,0.0020930998,0.0011725004,0.0002803056,0.0011279634],"domain_scores_gemma":[0.99104464,0.0064450456,0.00042892678,0.0011212694,0.00024882023,0.0007112879],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0027226186,0.00047892437,0.0024232895,0.00025353135,0.000076179196,0.000010568073,0.00028410915,0.0007485427,0.0014427176],"category_scores_gemma":[0.008144089,0.00033206216,0.00034870201,0.00034155895,0.00068562495,0.00008516358,0.00013958494,0.00089298544,0.0019461802],"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":[0.001683477,0.00072548044,0.9501834,0.00003345687,0.00019336355,0.0000474994,0.00009577672,0.000008264556,0.000011147836,0.00016567345,0.0056417007,0.041210763],"study_design_scores_gemma":[0.008106358,0.008114702,0.89093363,0.00013460596,0.0001031867,0.000006354729,0.0000037063348,0.0012621231,0.00002524414,0.00021557948,0.090812825,0.0002816884],"about_ca_topic_score_codex":0.007386569,"about_ca_topic_score_gemma":0.0044251475,"teacher_disagreement_score":0.085171126,"about_ca_system_score_codex":0.00059724826,"about_ca_system_score_gemma":0.00023376913,"threshold_uncertainty_score":0.99991316},"labels":[],"label_agreement":null},{"id":"W2104126672","doi":"10.2147/clep.s56554","title":"Empirical comparison of four baseline covariate adjustment methods in analysis of continuous outcomes in randomized controlled trials","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":89,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hamilton Health Sciences; St. Joseph’s Healthcare Hamilton; Population Health Research Institute; McMaster University","funders":"","keywords":"Analysis of covariance; Medicine; Randomized controlled trial; Covariate; Confidence interval; Analysis of variance; Sample size determination; Statistics; Physical therapy; Internal medicine; Mathematics","score_opus":0.8521315348875381,"score_gpt":0.7325592599955091,"score_spread":0.11957227489202904,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2104126672","genre_codex":"methods","genre_gemma":"methods","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.051714,0.00036354145,0.94062334,0.0024970304,0.0013807184,0.0029942116,0.00006951861,0.000035112243,0.0003225038],"genre_scores_gemma":[0.15166596,0.00020978981,0.8463288,0.001256951,0.00017262858,0.0002839295,0.00000819516,0.000027502521,0.000046258854],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.37318924,0.5636389,0.060239725,0.0014255533,0.0005350127,0.0009715356],"domain_scores_gemma":[0.039805926,0.945891,0.0125244185,0.0011324611,0.00036805932,0.00027813908],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6617769,0.0005069491,0.07913311,0.0009267643,0.000018157107,0.0000039010765,0.0006165086,0.0011995938,0.00055967126],"category_scores_gemma":[0.99192756,0.00030628804,0.0072673024,0.0010608713,0.0013392593,0.00002704654,0.00022699052,0.0011190381,0.000005516232],"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":[0.15473197,0.0038292047,0.6608513,0.00015064172,0.010983696,0.000004082378,0.000118628224,0.0010857653,0.000029025681,0.13143781,0.0005073868,0.036270477],"study_design_scores_gemma":[0.22518471,0.00044126672,0.06500127,0.0000856574,0.00752302,2.2293636e-7,0.000028849401,0.15169066,0.000015922225,0.54980344,0.00003170001,0.00019330502],"about_ca_topic_score_codex":0.00017040818,"about_ca_topic_score_gemma":0.0001236959,"teacher_disagreement_score":0.59585005,"about_ca_system_score_codex":0.000056240253,"about_ca_system_score_gemma":0.00012686527,"threshold_uncertainty_score":0.9999389},"labels":[],"label_agreement":null},{"id":"W2127327506","doi":"10.2147/clep.s9242","title":"Interpreting meta-analysis according to the adequacy of sample size. An example using isoniazid chemoprophylaxis for tuberculosis in purified protein derivative negative HIV-infected individuals","year":2010,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":280,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Meta-analysis; Medicine; Chemoprophylaxis; Randomized controlled trial; Sample size determination; Tuberculosis; Isoniazid; Placebo; Clinical trial; Systematic review; Internal medicine; MEDLINE; Statistics; Pathology; Alternative medicine; Biology","score_opus":0.34441605652586715,"score_gpt":0.4962971333126018,"score_spread":0.15188107678673463,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2127327506","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9505552,0.000115830204,0.023236165,0.022680204,0.00012944979,0.0029950426,0.00019605465,0.000048439004,0.000043620505],"genre_scores_gemma":[0.84711385,0.000015089084,0.1455106,0.0062633697,0.00024199054,0.00072441937,0.000070208735,0.000044626493,0.000015855947],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98643225,0.0067782965,0.0038812752,0.0014137705,0.00024855993,0.0012458155],"domain_scores_gemma":[0.8174026,0.17726496,0.0016946668,0.0019063335,0.0008462104,0.00088522217],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.02590763,0.00047572513,0.0057161045,0.0005337212,0.00019054579,0.000015533838,0.000690287,0.00078507425,0.00055918563],"category_scores_gemma":[0.45308992,0.00030366052,0.0022925204,0.0017163842,0.0008059877,0.00016741891,0.00043122703,0.0019089013,0.000012972807],"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":[0.0014302699,0.00066879246,0.93971914,0.000100123456,0.031619966,0.0000042630786,0.0010901557,0.00081533974,0.01790001,0.0014242611,0.00032582504,0.0049018427],"study_design_scores_gemma":[0.0041589527,0.005482469,0.865639,0.00008840935,0.034690987,0.000024251489,0.0021681827,0.049904782,0.008595473,0.026063414,0.002163088,0.0010210397],"about_ca_topic_score_codex":0.009120658,"about_ca_topic_score_gemma":0.0026390017,"teacher_disagreement_score":0.4271823,"about_ca_system_score_codex":0.000091062844,"about_ca_system_score_gemma":0.0002648208,"threshold_uncertainty_score":0.9999415},"labels":[],"label_agreement":null},{"id":"W2154229127","doi":"10.2147/clep.s35693","title":"Pediatric sepsis in the developing world: challenges in defining sepsis and issues in post-discharge mortality","year":2012,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":92,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"BC Children's Hospital; University of British Columbia","funders":"Canadian Institutes of Health Research; Michael Smith Health Research BC; Canadian Child Health Clinician Scientist Program","keywords":"Sepsis; Medicine; Developing country; Public health; Population; Developed country; Infectious disease (medical specialty); Pediatrics; Disease; Family medicine; Environmental health; Immunology; Economic growth; Nursing; Internal medicine","score_opus":0.5214627969714749,"score_gpt":0.5354544513736644,"score_spread":0.013991654402189524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2154229127","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8904441,0.05291361,0.0000021645537,0.055388846,0.00021655875,0.00034140266,0.000003621306,0.00001669912,0.0006729955],"genre_scores_gemma":[0.9415468,0.045718897,0.001696311,0.010541065,0.0003089123,0.00014278034,0.000018490118,0.000015156842,0.000011559697],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99471843,0.0023565674,0.0016300266,0.0005103523,0.000106323176,0.00067828724],"domain_scores_gemma":[0.98128295,0.01791325,0.00024441243,0.00037075844,0.00002934256,0.00015929283],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011931739,0.00022368706,0.0013412894,0.00032084362,0.000034511602,0.0000037728894,0.0001248474,0.00029275916,0.000058105376],"category_scores_gemma":[0.014513427,0.00014546153,0.00014969868,0.00041540424,0.00013135445,0.000089250265,0.000104467465,0.00062902976,0.000049530485],"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":[0.000033787706,0.0006285059,0.9818815,0.000059182177,0.000041018106,0.00002812673,0.0008067562,0.0000017175863,2.0469614e-7,0.0072651,0.00041883448,0.008835247],"study_design_scores_gemma":[0.0012925711,0.00016385675,0.9937686,0.0001066634,0.00008256378,0.000011076953,0.000667897,0.000037918602,0.0000010646277,0.0017977962,0.0019294692,0.00014053541],"about_ca_topic_score_codex":0.00072804966,"about_ca_topic_score_gemma":0.003473912,"teacher_disagreement_score":0.051102724,"about_ca_system_score_codex":0.000081612954,"about_ca_system_score_gemma":0.00006682684,"threshold_uncertainty_score":0.99378777},"labels":[],"label_agreement":null},{"id":"W2160865482","doi":"10.2147/clep.s59838","title":"Latitude, sunshine, and human lactase phenotype distributions may contribute to geographic patterns of modern disease: the inflammatory bowel disease model","year":2014,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Digestive system and related health","field":"Biochemistry, Genetics and Molecular Biology","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"Jewish General Hospital; McGill University","keywords":"Inflammatory bowel disease; Medicine; Latitude; Lactase; Population; Ulcerative colitis; Disease; Immunology; Environmental health; Biology; Internal medicine; Geography; Food science","score_opus":0.041822014328350504,"score_gpt":0.3750923660934541,"score_spread":0.33327035176510356,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2160865482","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9136755,0.0026460534,0.07543387,0.006459319,0.00017687806,0.0005497914,0.00094267214,0.000023760718,0.000092168695],"genre_scores_gemma":[0.9958098,0.00053127826,0.00012227948,0.0025243573,0.00022511685,0.000069786125,0.00052918267,0.00001925005,0.00016897624],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964638,0.0013972878,0.0010879679,0.00055321056,0.00008795077,0.00040983685],"domain_scores_gemma":[0.99710685,0.0009480456,0.00036373423,0.0006945816,0.00014849675,0.000738278],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031558282,0.00022092226,0.0006435908,0.000035532958,0.00019493957,0.0000062630893,0.00027680045,0.0002841985,0.000009214863],"category_scores_gemma":[0.005021296,0.00014376418,0.00033667684,0.000047642574,0.00044303612,0.000004591596,0.00022814336,0.0002741559,0.0000059178433],"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":[0.00023717836,0.00015026696,0.95559555,0.000109298424,0.00030766657,0.0000029882424,0.00000805886,0.0042318273,0.00015370731,0.037395533,0.0014316789,0.00037626407],"study_design_scores_gemma":[0.00067511346,0.0002724048,0.9725511,0.000046053472,0.000380234,0.0000010988795,0.000002788392,0.006406706,0.000006967081,0.013360256,0.0060909768,0.0002062791],"about_ca_topic_score_codex":0.00006520822,"about_ca_topic_score_gemma":0.00012835476,"teacher_disagreement_score":0.08213429,"about_ca_system_score_codex":0.000007731806,"about_ca_system_score_gemma":0.00011235302,"threshold_uncertainty_score":0.60113233},"labels":[],"label_agreement":null},{"id":"W2162824012","doi":"10.2147/clep.s44560","title":"Cognitive impairment and self-care in heart failure","year":2013,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Heart Failure Treatment and Management","field":"Medicine","cited_by":125,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McMaster University","funders":"National Center for Advancing Translational Sciences; Heart and Stroke Foundation of Canada; National Heart, Lung, and Blood Institute; National Institute on Aging; National Institutes of Health; National Center for Research Resources","keywords":"Medicine; Cognition; Heart failure; Cognitive impairment; Disease; Gerontology; Acute care; Physical therapy; Health care; Psychiatry; Internal medicine","score_opus":0.07170130591809963,"score_gpt":0.41802499834078144,"score_spread":0.3463236924226818,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2162824012","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9035957,0.000694885,0.00005069415,0.09345099,0.00012526818,0.0013511055,0.0000019758966,0.00008467625,0.0006447212],"genre_scores_gemma":[0.96759295,0.00032841385,0.016617337,0.014867094,0.00017525372,0.00018232722,0.00003336456,0.000012791962,0.00019048373],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99760646,0.0006985046,0.0008201388,0.00043500538,0.000055946646,0.00038394288],"domain_scores_gemma":[0.99617165,0.0031968486,0.00008410294,0.0001909402,0.00006871176,0.00028772288],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001411421,0.000154084,0.0008961829,0.00007532214,0.000030713414,0.0000039257234,0.000035220448,0.00024152368,0.00037813952],"category_scores_gemma":[0.0021515372,0.00011133722,0.00014209174,0.00008017785,0.0001668753,0.00004932863,0.00008954923,0.00032381192,0.0006585934],"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":[0.00006372864,0.00038681662,0.94433993,0.000055527016,0.00008241428,0.000018752207,0.00011959922,2.4801204e-7,0.0000016765321,0.0003260935,0.04766126,0.0069439304],"study_design_scores_gemma":[0.003053645,0.001789961,0.9111269,0.00009131734,0.00008403968,0.000017689401,0.0005551481,0.000119047705,0.0000026759228,0.0007174919,0.08234461,0.00009748387],"about_ca_topic_score_codex":0.00012231483,"about_ca_topic_score_gemma":0.000075722804,"teacher_disagreement_score":0.0785839,"about_ca_system_score_codex":0.00004634876,"about_ca_system_score_gemma":0.000048267146,"threshold_uncertainty_score":0.84651047},"labels":[],"label_agreement":null},{"id":"W2336126368","doi":"10.2147/clep.s104448","title":"Helping everyone do better: a call for validation studies of routinely recorded health data","year":2016,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Chronic Disease Management Strategies","field":"Medicine","cited_by":78,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Ottawa","funders":"Medical Research Council; National Institute for Health and Care Research","keywords":"Medicine; Epidemiology; Public health; Biostatistics; Family medicine; Alternative medicine; Health care; Population; Pharmacoepidemiology; Pediatrics; Gerontology; Environmental health; Nursing; Pathology","score_opus":0.6884983806733437,"score_gpt":0.6023180731285221,"score_spread":0.08618030754482153,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2336126368","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34170094,0.017052427,0.16496119,0.46592322,0.0039004735,0.0044082934,0.0006840881,0.00035021277,0.0010191611],"genre_scores_gemma":[0.8863375,0.019655785,0.045047667,0.042522687,0.0026272675,0.0002003726,0.0010313317,0.00008364857,0.0024937487],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99554276,0.0009209667,0.0023163995,0.0006997165,0.0000959522,0.00042417977],"domain_scores_gemma":[0.980467,0.017015278,0.0010269897,0.0011534217,0.00015971265,0.00017759424],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009869097,0.00015971927,0.0016236124,0.00006880487,0.000048285518,0.000002242892,0.00029097457,0.00013128854,0.000056070417],"category_scores_gemma":[0.056558575,0.00010159937,0.00020924034,0.000083408144,0.0004666398,0.00010384453,0.00037253607,0.00012815537,0.000025536383],"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":[0.0013538902,0.0004893385,0.32427403,0.0014856111,0.0013135502,0.0000064081178,0.000053544027,0.000007033529,0.00005188966,0.008628196,0.4082196,0.2541169],"study_design_scores_gemma":[0.023098381,0.006913847,0.59268266,0.0040490637,0.0012538008,0.000017976165,0.00059175893,0.0021432603,0.000033939326,0.074606635,0.29391056,0.0006981036],"about_ca_topic_score_codex":0.000038067825,"about_ca_topic_score_gemma":0.000019737978,"teacher_disagreement_score":0.54463655,"about_ca_system_score_codex":0.000096314296,"about_ca_system_score_gemma":0.0002683751,"threshold_uncertainty_score":0.9513884},"labels":[],"label_agreement":null},{"id":"W2475229149","doi":"10.2147/clep.s103098","title":"Reporting quality of stepped wedge design randomized trials: a systematic review protocol","year":2016,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hamilton Health Sciences; St. Peter's Hospital; Hamilton General Hospital; McMaster Children's Hospital; McMaster University; Queen's University; St. Joseph’s Healthcare Hamilton; Population Health Research Institute","funders":"Hamilton Health Sciences Foundation; McMaster University; Hamilton Health Sciences","keywords":"Consolidated Standards of Reporting Trials; Randomized controlled trial; Medicine; CINAHL; MEDLINE; Medical physics; Data extraction; Protocol (science); Sample size determination; Research design; Systematic review; PsycINFO; Cochrane Library; Data mining; Family medicine; Computer science; Alternative medicine; Statistics; Psychological intervention; Pathology; Nursing; Mathematics","score_opus":0.9893285022792409,"score_gpt":0.8002175333740813,"score_spread":0.18911096890515955,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2475229149","genre_codex":"review","genre_gemma":"review","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[9.617694e-9,0.56567764,0.10173166,0.00030224136,0.0005155474,0.331184,0.000027531745,0.00001171909,0.0005496518],"genre_scores_gemma":[5.6552306e-8,0.6839379,0.009077585,0.0007715139,0.0003168453,0.30216312,0.000012471699,0.000042889842,0.0036776117],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.005976839,0.61716026,0.3706787,0.0027654595,0.0028232415,0.00059551885],"domain_scores_gemma":[0.0011115607,0.5695145,0.41633335,0.010347053,0.0022914654,0.0004020393],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.99160254,0.0017008069,0.42422986,0.00065133476,0.00012101686,0.00011142603,0.0066601792,0.0019143609,0.008703055],"category_scores_gemma":[0.99894124,0.0005190247,0.10935149,0.0018245723,0.0011299253,0.00013726302,0.0006754168,0.0013262549,0.0076091583],"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":[0.0005179527,0.00008687619,0.000014042441,0.7852824,0.0036381518,0.000005196904,0.0000039815104,1.986812e-7,6.7511925e-9,0.0018276621,0.04043897,0.1681846],"study_design_scores_gemma":[0.008455108,0.000072144554,0.000001173541,0.5276489,0.018218841,0.000037158163,0.000004308481,0.00005817504,7.4973565e-9,0.012994778,0.4320824,0.00042698262],"about_ca_topic_score_codex":0.000008040959,"about_ca_topic_score_gemma":0.0000013163411,"teacher_disagreement_score":0.39164346,"about_ca_system_score_codex":0.000098959754,"about_ca_system_score_gemma":0.0014964251,"threshold_uncertainty_score":0.9997261},"labels":[],"label_agreement":null},{"id":"W2515193896","doi":"10.2147/clep.s108761","title":"Cholesterol testing among men and women with disability: the role of morbidity","year":2016,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Down syndrome and intellectual disability research","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto; St. Michael's Hospital","funders":"Canadian Cancer Society Research Institute; University of Toronto; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Medicine; Odds ratio; Logistic regression; Confidence interval; Cohort; Odds; Cohort study; Gerontology; Demography; Internal medicine","score_opus":0.20152015064942755,"score_gpt":0.43914654972502354,"score_spread":0.237626399075596,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2515193896","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.987209,0.00006146337,0.00007383156,0.0046956576,0.000022468086,0.00035683453,0.000006055822,0.000023967566,0.007550722],"genre_scores_gemma":[0.99860066,0.00007818701,0.00019884728,0.0005320507,0.00010750641,0.000053776537,9.83891e-7,0.000011519032,0.00041647974],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99666536,0.0013848097,0.00096177513,0.00043476158,0.00012456658,0.0004287343],"domain_scores_gemma":[0.9705247,0.02809023,0.00018802282,0.0007136317,0.00014364642,0.00033973833],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.007904583,0.00012520942,0.00091543724,0.000017576265,0.000052257317,0.0000028298257,0.00016583175,0.00017550167,0.00038197465],"category_scores_gemma":[0.07537906,0.000050193,0.000079139456,0.00013042771,0.005078132,0.000056443907,0.00017062582,0.00036000344,0.00004961291],"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":[0.00031082606,0.00017811765,0.9119383,0.000031521653,0.000064177424,2.9143206e-7,0.00009505404,1.8331158e-7,0.0002460471,0.00010459999,0.00023772688,0.08679314],"study_design_scores_gemma":[0.0007738602,0.0022956089,0.98915833,0.00008784161,0.00002596089,0.000027977292,0.00033319424,0.000225738,0.00007012829,0.005665944,0.0012718394,0.00006359924],"about_ca_topic_score_codex":0.00022106165,"about_ca_topic_score_gemma":0.000016008431,"teacher_disagreement_score":0.08672954,"about_ca_system_score_codex":0.000046866404,"about_ca_system_score_gemma":0.00006647077,"threshold_uncertainty_score":0.99762946},"labels":[],"label_agreement":null},{"id":"W2535390142","doi":"10.2147/clep.s110528","title":"The RECORD reporting guidelines: meeting the methodological and ethical demands of transparency in research using routinely-collected health data","year":2016,"lang":"en","type":"article","venue":"Clinical Epidemiology","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":"Children's Hospital of Eastern Ontario; University of Ottawa","funders":"Novo Nordisk Fonden; Lundbeckfonden; National Institute for Health and Care Research","keywords":"Transparency (behavior); Medicine; Data science; Environmental health; Accounting; Computer science; Business; Computer security","score_opus":0.992204283601152,"score_gpt":0.7785009185754199,"score_spread":0.21370336502573206,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2535390142","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.23616497,0.0058486355,0.06400199,0.6920176,0.0006978138,0.00081654696,0.00008041499,0.000025438894,0.0003465384],"genre_scores_gemma":[0.699064,0.022306385,0.20981847,0.066814214,0.0015690797,0.00012902268,0.000028347915,0.000065205924,0.00020523935],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.796075,0.10058596,0.09749296,0.003151158,0.00030368933,0.0023912434],"domain_scores_gemma":[0.22660631,0.7309448,0.037594948,0.0036774336,0.0006122788,0.0005642285],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.74531186,0.00017337623,0.0031215937,0.00015937373,0.00054904266,0.000016556627,0.001004339,0.0006969585,0.000037928618],"category_scores_gemma":[0.8833518,0.000099648816,0.0001744374,0.0004234709,0.0012926308,0.0000967277,0.00045450963,0.0014488136,0.000021963375],"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":[0.00009716261,0.00004098108,0.9026426,0.00010658774,0.00005542665,0.0000013243153,0.0002815366,0.00005127963,0.0000033236417,0.06374398,0.018134313,0.014841473],"study_design_scores_gemma":[0.0015002717,0.00044406942,0.55929637,0.0007609567,0.000010732213,0.000030116242,0.0010478836,0.07058607,5.888473e-7,0.31893247,0.047080893,0.0003095872],"about_ca_topic_score_codex":0.0043216734,"about_ca_topic_score_gemma":0.0020093063,"teacher_disagreement_score":0.6303588,"about_ca_system_score_codex":0.00020842007,"about_ca_system_score_gemma":0.0006263311,"threshold_uncertainty_score":0.6533106},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2565606048","doi":"10.2147/clep.s121632","title":"Antinuclear antibody prevalence in a general pediatric cohort from Mexico City: discordance between immunofluorescence and multiplex assays","year":2016,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Systemic Lupus Erythematosus Research","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":"Public Health Ontario; University of Toronto","funders":"National Institute of Environmental Health Sciences; National Institutes of Health; School of Public Health, University of Michigan; University of Michigan","keywords":"Medicine; Anti-nuclear antibody; Multiplex; Cohort; Public health; Cohort study; Immunofluorescence; Internal medicine; Antibody; Immunology; Pathology; Biology; Bioinformatics; Autoantibody","score_opus":0.13280406908141498,"score_gpt":0.4572882573169061,"score_spread":0.3244841882354911,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2565606048","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9896843,0.003034839,0.0021451719,0.0035672148,0.00036887836,0.000911474,0.00007892615,0.000077369725,0.00013181614],"genre_scores_gemma":[0.97716117,0.016135836,0.004261551,0.00054503913,0.0012609429,0.000044515076,0.00002840049,0.00004392473,0.000518594],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99196887,0.0029601606,0.0025350875,0.0013137276,0.0003074568,0.0009147266],"domain_scores_gemma":[0.97182876,0.026005499,0.00054895977,0.0010135714,0.0001277263,0.00047550738],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.010181752,0.00031142373,0.0022077006,0.00017618878,0.00006403331,0.000008068312,0.00044495435,0.00056882924,0.00016262846],"category_scores_gemma":[0.050335858,0.00019777528,0.00024047583,0.0002815506,0.0009839032,0.0001319838,0.00046370833,0.0008193006,0.00018307264],"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":[0.00030167395,0.000119197335,0.99171597,0.00017713744,0.000082245126,0.00006874269,0.000038025693,2.465162e-7,0.001966091,0.000103474806,0.0007173876,0.004709782],"study_design_scores_gemma":[0.0031923319,0.00036251725,0.9934197,0.000647871,0.00009654153,0.00015982083,0.000013918184,0.00042925353,0.00004383308,0.0010978662,0.00031030405,0.0002260573],"about_ca_topic_score_codex":0.00048332394,"about_ca_topic_score_gemma":0.000020901662,"teacher_disagreement_score":0.040154107,"about_ca_system_score_codex":0.00010222962,"about_ca_system_score_gemma":0.00019171456,"threshold_uncertainty_score":0.95766354},"labels":[],"label_agreement":null},{"id":"W2587347639","doi":"10.2147/clep.s124674","title":"Risk of gynecomastia and breast cancer associated with the use of 5-alpha reductase inhibitors for benign prostatic hyperplasia","year":2017,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Male Breast Health Studies","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Kingston General Hospital; Queen's University","funders":"National Institute of Diabetes and Digestive and Kidney Diseases; National Institutes of Health","keywords":"Medicine; Gynecomastia; Breast cancer; Odds ratio; Cohort; Internal medicine; Incidence (geometry); Urology; Gynecology; Cancer","score_opus":0.23601609745317018,"score_gpt":0.4661109742357695,"score_spread":0.23009487678259935,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2587347639","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9682646,0.0004939143,0.00007535844,0.029338583,0.00023297465,0.0009849299,0.0005735634,0.000012691855,0.000023425173],"genre_scores_gemma":[0.99538106,0.0007995088,0.002255355,0.0011065723,0.00019934234,0.00010403448,0.000016772592,0.000019982279,0.00011738311],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99694777,0.00092527457,0.0013354175,0.00035875128,0.00008919716,0.00034362267],"domain_scores_gemma":[0.9684376,0.028135303,0.0022885338,0.00057960034,0.0003809506,0.0001779854],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0046406393,0.00016624796,0.0015617706,0.000032840664,0.00023906707,0.0000030548017,0.00013099663,0.00021730835,0.000014011804],"category_scores_gemma":[0.08579306,0.00009048326,0.00015413806,0.000045545326,0.0021936095,0.000051101553,0.00010303457,0.00038197113,6.316771e-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":[0.0014110913,0.0001770349,0.9752327,0.00021490287,0.00044674517,0.0000023941316,0.000084443804,0.000010681632,0.000013093028,0.00027500626,0.009500821,0.012631056],"study_design_scores_gemma":[0.0026007106,0.0010572607,0.9928529,0.00034675893,0.00061491167,0.000032376793,0.000055416007,0.0007975954,0.000010830857,0.00030520174,0.0012466735,0.00007935389],"about_ca_topic_score_codex":0.0009113076,"about_ca_topic_score_gemma":0.00049706193,"teacher_disagreement_score":0.08115242,"about_ca_system_score_codex":0.000026041038,"about_ca_system_score_gemma":0.00022684812,"threshold_uncertainty_score":0.9219077},"labels":[],"label_agreement":null},{"id":"W2617275116","doi":"10.2147/clep.s129268","title":"Hypoglycemia: a review of definitions used in clinical trials evaluating antihyperglycemic drugs for diabetes","year":2017,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Diabetes Management and Research","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":true,"ca_institutions":"McMaster University; Impact; Precision Nanosystems (Canada); Simon Fraser University","funders":"","keywords":"Medicine; Hypoglycemia; Insulin glargine; Internal medicine; Diabetes mellitus; Type 2 diabetes; Insulin detemir; Randomized controlled trial; Clinical trial; Insulin; Endocrinology","score_opus":0.923343275527221,"score_gpt":0.7261855993875692,"score_spread":0.19715767613965174,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2617275116","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":[0.00009802971,0.98516244,0.00003389839,0.0026948273,0.0018448778,0.007565826,0.00015747124,0.00004263066,0.0023999815],"genre_scores_gemma":[0.00000891786,0.98865545,0.0028158682,0.0028671746,0.0015948258,0.0013330819,0.001069868,0.00010092609,0.0015538789],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9406547,0.033097558,0.022742767,0.0017381933,0.00042716932,0.0013395979],"domain_scores_gemma":[0.60854876,0.37471142,0.012477676,0.0029939485,0.00061043125,0.0006577706],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.26775327,0.00066782866,0.029236538,0.00049191725,0.00009751019,0.000011664454,0.0010441585,0.0022160953,0.00017300408],"category_scores_gemma":[0.7072781,0.00044802754,0.00852809,0.00036715,0.0015166161,0.00006843025,0.0005397932,0.002919625,0.00017623635],"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":[0.00013689171,0.00042545586,0.006071778,0.18093069,0.00085489615,0.0000055845812,0.0000017154593,6.649992e-8,4.560001e-8,0.00037849683,0.01728631,0.79390806],"study_design_scores_gemma":[0.003735863,0.0020400148,0.00032505373,0.24586205,0.0060894885,0.0000018526094,0.0000030040947,0.00015965546,1.7220785e-8,0.002165705,0.7393018,0.00031547647],"about_ca_topic_score_codex":0.000004662901,"about_ca_topic_score_gemma":0.0000023010991,"teacher_disagreement_score":0.7935926,"about_ca_system_score_codex":0.000121542595,"about_ca_system_score_gemma":0.0022115104,"threshold_uncertainty_score":0.99979717},"labels":[],"label_agreement":null},{"id":"W2734790762","doi":"10.2147/clep.s136565","title":"Fecal occult blood versus DNA testing: indirect comparison in a colorectal cancer screening population","year":2017,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","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":"China Scholarship Council; Deutsche Forschungsgemeinschaft","keywords":"Medicine; Colonoscopy; Colorectal cancer; Internal medicine; Gastroenterology; Gold standard (test); Fecal occult blood; Cutoff; Colorectal cancer screening; Population; Cancer","score_opus":0.4115441886125704,"score_gpt":0.5179444256739547,"score_spread":0.1064002370613843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2734790762","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.994022,0.00036131396,0.00029492678,0.0015355136,0.001712428,0.0003581373,0.000007025303,0.00013964131,0.0015689865],"genre_scores_gemma":[0.99380153,0.000051187606,0.004484796,0.00042546826,0.001056857,0.000076082295,0.00003350023,0.000023224786,0.00004735245],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996354,0.0008136182,0.0014701266,0.00069403095,0.00014763336,0.000520586],"domain_scores_gemma":[0.990856,0.0073927674,0.00092976156,0.00042375454,0.0001168553,0.0002808571],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0038756349,0.00022426706,0.0015472657,0.00012334331,0.00030333424,0.000019004514,0.00021270887,0.00057811185,0.000045256427],"category_scores_gemma":[0.06682131,0.00020120249,0.0002610772,0.00017156694,0.00033560608,0.00010018721,0.00014925342,0.0010738474,0.000018216988],"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":[0.008552543,0.00015667264,0.8987319,0.00001810394,0.00009624461,0.000023546621,0.000025301328,0.00014215741,0.00010597434,0.000029234314,0.00024825556,0.091870055],"study_design_scores_gemma":[0.0055990764,0.0041757086,0.97038287,0.00019076269,0.00018545131,0.000026814057,0.00002033813,0.018525824,0.00018173673,0.00014610968,0.00037506342,0.00019026239],"about_ca_topic_score_codex":0.0058349725,"about_ca_topic_score_gemma":0.0045967996,"teacher_disagreement_score":0.09167979,"about_ca_system_score_codex":0.000113589675,"about_ca_system_score_gemma":0.00012180263,"threshold_uncertainty_score":0.94103926},"labels":[],"label_agreement":null},{"id":"W2770879274","doi":"10.2147/clep.s148890","title":"Validation of algorithms to determine incidence of Hirschsprung disease in Ontario, Canada: a population-based study using health administrative data","year":2017,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Congenital gastrointestinal and neural anomalies","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; Children's Hospital of Eastern Ontario; University of Ottawa","funders":"Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences","keywords":"Incidence (geometry); Medicine; Confidence interval; Poisson regression; Algorithm; Population; Demography; Odds ratio; Statistics; Pediatrics; Internal medicine; Environmental health; Mathematics","score_opus":0.49203488253539773,"score_gpt":0.5127853194319227,"score_spread":0.020750436896525004,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2770879274","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9950901,0.000027512799,0.00020479104,0.003918105,0.00012898985,0.0005566209,0.000064524764,0.0000039928414,0.000005326222],"genre_scores_gemma":[0.9934253,0.0000011541923,0.005577319,0.000822957,0.000042185442,0.000006586969,0.000083634084,0.000005544527,0.00003528187],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99708754,0.0005395128,0.0016299529,0.00041378307,0.00012574521,0.00020348762],"domain_scores_gemma":[0.9951009,0.0026894794,0.0009630713,0.0008203797,0.0001158052,0.00031037084],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0021607347,0.00011835066,0.0010208748,0.00005010927,0.00006293892,0.000002292458,0.00030001547,0.00003752393,0.00002698936],"category_scores_gemma":[0.027822321,0.00009612333,0.00006217825,0.00005027162,0.00011587098,0.00006880946,0.00022259308,0.00022049295,3.958122e-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":[0.00063393364,0.00029864127,0.99614674,0.000037498212,0.000015175137,0.000056256307,0.000025185047,0.0001278469,0.000025023948,0.000006420913,0.000021074644,0.0026062168],"study_design_scores_gemma":[0.0006952046,0.0016273955,0.9894,0.00023085547,0.000046877467,0.000004957442,0.000032501786,0.007816837,0.000013674904,0.000057101428,0.000010450015,0.000064094886],"about_ca_topic_score_codex":0.9531202,"about_ca_topic_score_gemma":0.97711295,"teacher_disagreement_score":0.025661586,"about_ca_system_score_codex":0.00015807456,"about_ca_system_score_gemma":0.0024546676,"threshold_uncertainty_score":0.98036677},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2775313512","doi":"10.2147/clep.s146496","title":"Association between coffee or caffeine consumption and fecundity and fertility: a systematic review and dose–response meta-analysis","year":2017,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Coffee research and impacts","field":"Medicine","cited_by":60,"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":"Novo Nordisk Fonden; Novo Nordisk; Danish Diabetes Academy; European Food Safety Authority; Aarhus Universitet","keywords":"Fecundity; Caffeine; Fertility; Medicine; Meta-analysis; Consumption (sociology); Demography; Traditional medicine; Internal medicine; Environmental health; Population; Social science","score_opus":0.8569641366053552,"score_gpt":0.6596007806804756,"score_spread":0.19736335592487964,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2775313512","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":[0.00050643674,0.9877658,0.000068134024,0.006866505,0.000055737193,0.004300305,0.00038726482,0.00003805198,0.000011788561],"genre_scores_gemma":[0.00015000116,0.9942564,0.00019432418,0.0017277538,0.00013312072,0.000286769,0.00036051747,0.00003325755,0.0028578832],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9645553,0.02818839,0.004996153,0.0012259154,0.00031747506,0.00071672746],"domain_scores_gemma":[0.85219127,0.13953675,0.00484963,0.0017623531,0.00029748512,0.0013624869],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07561829,0.0006568744,0.030474713,0.00036186815,0.00022204997,0.00005173741,0.00023738039,0.0013070131,0.00029420733],"category_scores_gemma":[0.47537693,0.0003613117,0.0027350073,0.0002700292,0.0007063873,0.000096791926,0.00037920423,0.0014879464,0.00008499735],"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":[0.0003292292,0.000077388184,0.017266432,0.7675226,0.18095067,0.000023838857,0.000006579004,4.757058e-9,2.6083988e-8,0.00000460953,0.0030148819,0.030803783],"study_design_scores_gemma":[0.00056480645,0.0005353801,0.049710274,0.026618103,0.72227293,0.00008281406,6.463379e-7,0.000005443942,4.086523e-9,0.00006665552,0.19988213,0.00026083426],"about_ca_topic_score_codex":0.00008253833,"about_ca_topic_score_gemma":0.00015171501,"teacher_disagreement_score":0.74090445,"about_ca_system_score_codex":0.00021814331,"about_ca_system_score_gemma":0.0004888687,"threshold_uncertainty_score":0.9999895},"labels":[],"label_agreement":null},{"id":"W2789147510","doi":"10.2147/clep.s156210","title":"Self-reported perinatal depressive symptoms and postnatal symptom severity after treatment with antidepressants in pregnancy: a cross-sectional study across 12 European countries using the Edinburgh Postnatal Depression Scale","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Maternal Mental Health During Pregnancy and Postpartum","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":"Hospital for Sick Children","funders":"","keywords":"Medicine; Pregnancy; Edinburgh Postnatal Depression Scale; Depression (economics); Cross-sectional study; Confidence interval; Antidepressant; Major depressive disorder; Obstetrics; Pediatrics; Psychiatry; Depressive symptoms; Internal medicine; Mood; Anxiety","score_opus":0.07761367686057305,"score_gpt":0.4380527027434807,"score_spread":0.36043902588290766,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2789147510","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9949466,0.0020629885,0.000035655306,0.00012839632,0.00080390734,0.0016628588,0.0001306056,0.00009962956,0.00012937425],"genre_scores_gemma":[0.9977742,0.00040779493,0.0002793673,0.00063505844,0.0005727611,0.00012324654,0.000029635614,0.000044818564,0.00013314224],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9944435,0.001517789,0.0017949796,0.0011234559,0.00028614845,0.000834139],"domain_scores_gemma":[0.9963224,0.0018119001,0.0006342954,0.00067900197,0.00018249365,0.00036992878],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.002650189,0.00043068035,0.0010674011,0.00006495242,0.00048048564,0.000045101424,0.00019356153,0.00029788478,0.00008562752],"category_scores_gemma":[0.00060394173,0.00025893742,0.00014745719,0.000108324355,0.0014101268,0.00023649028,0.0004126636,0.00062725454,0.000011990226],"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":[0.0056346552,0.0007976891,0.9894873,0.00020988494,0.00030239942,0.001905087,0.0010798627,0.0000049064097,0.0000046554796,0.000001924511,0.0000031178188,0.00056849845],"study_design_scores_gemma":[0.005909502,0.003965093,0.9829735,0.0017572187,0.00015583236,0.0038843274,0.00008230871,0.00079387537,0.00010460412,0.000013283457,0.00010180196,0.00025865214],"about_ca_topic_score_codex":0.00061520474,"about_ca_topic_score_gemma":0.00029032305,"teacher_disagreement_score":0.006513818,"about_ca_system_score_codex":0.00014450935,"about_ca_system_score_gemma":0.00015060375,"threshold_uncertainty_score":0.9999863},"labels":[],"label_agreement":null},{"id":"W2791886356","doi":"10.2147/clep.s156708","title":"Critical concepts in adaptive clinical trials","year":2018,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":99,"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; University of British Columbia","funders":"Bill and Melinda Gates Foundation","keywords":"Adaptive design; Sample size determination; Clinical trial; Bayesian probability; Clinical study design; Computer science; Randomization; Research design; Medicine; Machine learning; Data mining; Artificial intelligence; Statistics; Mathematics; Pathology","score_opus":0.9828560699097272,"score_gpt":0.8344826322595544,"score_spread":0.14837343765017275,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2791886356","genre_codex":"review","genre_gemma":"review","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000012665531,0.90530485,0.0692728,0.00075070467,0.017587302,0.003616765,0.0005528015,0.00023965181,0.002673855],"genre_scores_gemma":[7.351027e-7,0.5928393,0.39564463,0.001666866,0.009115222,0.00037010422,0.00001664041,0.00015904292,0.00018748597],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.443837,0.4803655,0.067969926,0.004755162,0.000581333,0.002491081],"domain_scores_gemma":[0.011919792,0.9713263,0.011296142,0.0031597912,0.0006010366,0.0016969534],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"category_scores_codex":[0.5633891,0.0016739414,0.05982497,0.00038207436,0.00012318845,0.000033743647,0.0024515942,0.010827915,0.0025100058],"category_scores_gemma":[0.99587655,0.0011635678,0.011624728,0.0007886997,0.01010111,0.000103127706,0.0012415797,0.009490674,0.0029526805],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00023097526,0.0007964319,0.0001856826,0.0031071363,0.0005573602,0.00010015079,0.0000043144064,2.1317234e-8,6.713241e-10,0.26490897,0.053214606,0.67689437],"study_design_scores_gemma":[0.001061293,0.0011587217,0.000042366322,0.008489868,0.0020523625,0.00002182616,0.0000058217765,0.00004714408,7.1036417e-9,0.5312111,0.4552907,0.0006188042],"about_ca_topic_score_codex":0.00001455926,"about_ca_topic_score_gemma":0.000017916562,"teacher_disagreement_score":0.67627555,"about_ca_system_score_codex":0.00023914473,"about_ca_system_score_gemma":0.001591173,"threshold_uncertainty_score":0.99960077},"labels":[],"label_agreement":null},{"id":"W2796288464","doi":"10.2147/clep.s150915","title":"Apathy and depressive symptoms in older people and incident myocardial infarction, stroke, and mortality: a systematic review and meta-analysis of individual participant data","year":2018,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Cardiac Health and Mental Health","field":"Medicine","cited_by":77,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Center for Advancing Translational Sciences; National Institute of Arthritis and Musculoskeletal and Skin Diseases; National Heart, Lung, and Blood Institute; National Institute on Aging; Bayer Vital; BERLIN-CHEMIE; Biomedical Research Council; Universiteit Leiden; National Medical Research Council; Health Canada; College of Medicine and Dentistry, James Cook University; University of New South Wales; KU Leuven; School of Medicine, Indiana University; Leids Universitair Medisch Centrum; Duke Global Health Institute, Duke University; Bayer Canada; National Institutes of Health; National Institute for Health Research Southampton Biomedical Research Centre; Fondazione Salvatore Maugeri; Regione Campania; Universiteit Antwerpen; Pfizer; Università degli Studi di Firenze; Sanofi; Università degli Studi di Napoli Federico II; Chinese University of Hong Kong; Dipartimento di Medicina Sperimentale e Clinica, Università degli Studi di Firenze; University of California, San Francisco; Technische Universität München; Croucher Foundation; Regione Toscana; National University of Singapore; Radboud Universiteit; Fundação de Amparo à Pesquisa do Estado de São Paulo; Universita degli Studi di Bari Aldo Moro; National Health and Medical Research Council; Universidade de São Paulo; University of Southampton; Royal Adelaide Hospital; James Cook University; University of Oxford; Universiteit van Amsterdam; University Hospital Southampton NHS Foundation Trust; Australian Government; Australian Research Council; University of Ottawa; National Institute for Health and Care Research; Medical Research Council; Medical Research Council Canada; Teva Pharmaceutical Industries","keywords":"Apathy; Medicine; Hazard ratio; Stroke (engine); Geriatric Depression Scale; Depression (economics); Meta-analysis; Confidence interval; Proportional hazards model; PsycINFO; Myocardial infarction; Internal medicine; Physical therapy; Depressive symptoms; Disease; Psychiatry; MEDLINE; Anxiety","score_opus":0.5093893204536872,"score_gpt":0.568188406356617,"score_spread":0.058799085902929815,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2796288464","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":[0.0061287126,0.98777467,0.000015154334,0.0005947791,0.00011811514,0.0042699315,0.0010786428,0.000010130798,0.000009882954],"genre_scores_gemma":[0.0028322886,0.98952574,0.00009033012,0.0062537063,0.00010400182,0.00047827678,0.00067530235,0.00002339823,0.000016937965],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9842962,0.007562104,0.0061050905,0.0012776144,0.00023739837,0.0005215893],"domain_scores_gemma":[0.9814855,0.013656535,0.0025477465,0.0013348047,0.00010302982,0.0008724153],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.025333947,0.00046149443,0.025082007,0.0003295344,0.000058873185,0.000008483318,0.00018144365,0.0006715361,0.000026941512],"category_scores_gemma":[0.02186459,0.0002947185,0.0009908876,0.0004108042,0.0006495913,0.00007075052,0.00083354977,0.00075804954,0.0000022186493],"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":[0.00002210753,0.00011379328,0.110435314,0.78332657,0.0991289,0.000011273955,0.00012687748,4.6647397e-8,2.7569924e-9,0.000035924415,0.0006416563,0.0061575444],"study_design_scores_gemma":[0.00055682793,0.00034112658,0.14582509,0.016706485,0.83258474,0.00018027045,0.00002919366,0.000029251089,1.7137453e-9,0.000021844568,0.003539071,0.00018611058],"about_ca_topic_score_codex":0.000554532,"about_ca_topic_score_gemma":0.00047710384,"teacher_disagreement_score":0.76662004,"about_ca_system_score_codex":0.00003190185,"about_ca_system_score_gemma":0.00040179238,"threshold_uncertainty_score":0.99995047},"labels":[],"label_agreement":null},{"id":"W2800574345","doi":"10.2147/clep.s163668","title":"The impact of moderate wine consumption on the risk of developing prostate cancer","year":2018,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","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":"Université de Montréal; McGill University Health Centre","funders":"European Association of Urology","keywords":"Wine; White Wine; Medicine; Cochrane Library; Publication bias; Meta-analysis; Relative risk; Logistic regression; Prostate cancer; Internal medicine; Cancer; Confidence interval; Food science","score_opus":0.4333643660796528,"score_gpt":0.5716300129903057,"score_spread":0.1382656469106529,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2800574345","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":[0.0062193465,0.9891907,0.000029682891,0.001937179,0.0004468089,0.0017688966,0.00033701907,0.000012284684,0.00005807241],"genre_scores_gemma":[0.00058239634,0.99776256,0.00016014281,0.0003868766,0.00043998376,0.00047766202,0.000071816794,0.000033605094,0.00008496614],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9925223,0.0034909751,0.002992418,0.00049704587,0.0001029192,0.00039436738],"domain_scores_gemma":[0.9578958,0.037705857,0.0032474,0.0007808379,0.0002488482,0.00012123245],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.007611019,0.00037621992,0.0041814228,0.00004804266,0.00011823427,0.0000032227222,0.00022357386,0.00042356423,0.00006595082],"category_scores_gemma":[0.014056248,0.00013563629,0.001568704,0.000135858,0.0012228704,0.000013511917,0.00009511332,0.0007399283,0.000042841795],"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":[0.00061719865,0.00021730355,0.07132972,0.0038504794,0.004448521,0.0000042274023,0.00002734944,0.000025831747,4.059969e-8,0.0010851139,0.016100854,0.9022934],"study_design_scores_gemma":[0.002180583,0.0059626745,0.07763157,0.053696267,0.0060443366,0.000026802769,0.0000079327465,0.00039115248,0.000007883819,0.0057651433,0.84784573,0.0004399484],"about_ca_topic_score_codex":0.00030546158,"about_ca_topic_score_gemma":0.000021376793,"teacher_disagreement_score":0.90185344,"about_ca_system_score_codex":0.0002730474,"about_ca_system_score_gemma":0.0011887188,"threshold_uncertainty_score":0.99424875},"labels":[],"label_agreement":null},{"id":"W2802963693","doi":"10.2147/clep.s153000","title":"Derivation and validation of a multivariable model to predict when primary care physicians prescribe antidepressants for indications other than depression","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Pharmacovigilance and Adverse Drug Reactions","field":"Pharmacology, Toxicology and Pharmaceutics","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":"McGill University","funders":"","keywords":"Concordance; Medicine; Medical prescription; Depression (economics); Workload; Antidepressant; Statistic; Medical record; Emergency medicine; Psychiatry; Family medicine; Internal medicine; Statistics; Anxiety","score_opus":0.3380306777013947,"score_gpt":0.5400212628127203,"score_spread":0.20199058511132562,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2802963693","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8718965,0.0002495758,0.12039434,0.0014027253,0.00087483024,0.0013909785,0.00062019634,0.00009832101,0.003072518],"genre_scores_gemma":[0.98050475,0.00014770882,0.010595846,0.007438947,0.0006117765,0.00024146233,0.00021656917,0.000025865887,0.00021705772],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99721986,0.0008458088,0.00096768583,0.00052435993,0.000069254755,0.0003730528],"domain_scores_gemma":[0.9946567,0.004063337,0.0005321044,0.00028343542,0.00026329368,0.00020117423],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001803683,0.00017854493,0.0006016993,0.00010020402,0.00025483692,0.000003969981,0.00022672045,0.00044666103,0.000043889544],"category_scores_gemma":[0.00267435,0.00016597261,0.00013817022,0.000112800386,0.00045063256,0.0001834099,0.00012259104,0.0003955965,0.0000188147],"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":[0.0014044815,0.0007008081,0.8447602,0.00023146546,0.00032576625,6.344408e-7,0.0020575663,0.009397739,0.060059626,0.0012777494,0.02731435,0.05246961],"study_design_scores_gemma":[0.010635675,0.0012504542,0.40505853,0.0004279705,0.001241961,0.000010818454,0.00050164736,0.30947715,0.12346253,0.023931684,0.12301463,0.000986974],"about_ca_topic_score_codex":0.00003592176,"about_ca_topic_score_gemma":0.000015484735,"teacher_disagreement_score":0.43970168,"about_ca_system_score_codex":0.000038163922,"about_ca_system_score_gemma":0.00010922921,"threshold_uncertainty_score":0.67681664},"labels":[],"label_agreement":null},{"id":"W2806544235","doi":"10.2147/clep.s151339","title":"Development and validation of novel algorithms to identify patients with inflammatory bowel diseases in Israel: an epi-IIRN group study","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":73,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; Children's Hospital of Eastern Ontario; University of Ottawa","funders":"Leona M. and Harry B. Helmsley Charitable Trust","keywords":"Medicine; Algorithm; Inflammatory bowel disease; Ulcerative colitis; Internal medicine; Diagnosis code; Cohort; Crohn's disease; Population; Disease; Inflammatory Bowel Diseases; Predictive value; Gastroenterology; Mathematics","score_opus":0.060940398985614805,"score_gpt":0.3981053630868878,"score_spread":0.337164964101273,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2806544235","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9967667,0.000024825911,0.0021946197,0.000037194168,0.00021197076,0.00070695585,0.000029123232,0.000014171788,0.000014422983],"genre_scores_gemma":[0.9924137,0.000006269603,0.0063414117,0.0007017172,0.0002286214,0.00008053894,0.00019046103,0.00002648952,0.000010807743],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9966038,0.0008439808,0.0013479677,0.00073995936,0.00012802248,0.00033629764],"domain_scores_gemma":[0.99826574,0.00027044927,0.0003760154,0.00050668756,0.00023548945,0.0003456148],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0024386276,0.00021683463,0.00048700988,0.0000904967,0.00006494246,0.0000067096885,0.00024231517,0.0002225025,0.000007804443],"category_scores_gemma":[0.0035623182,0.00018624832,0.000060355414,0.0000798742,0.00038989028,0.00001609135,0.0002598492,0.0001223831,0.00001099045],"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":[0.0009336396,0.0014547507,0.9941757,0.00002355616,0.00006676601,0.00000781022,0.000112469286,0.00003395904,0.00030806326,0.00007546355,0.00006928839,0.00273857],"study_design_scores_gemma":[0.0026185538,0.0015588712,0.9946832,0.000025703333,0.000030147792,6.612212e-7,0.00016238599,0.0001004219,0.00013970715,0.00004699145,0.0004071905,0.0002261371],"about_ca_topic_score_codex":0.000020988295,"about_ca_topic_score_gemma":0.0003193583,"teacher_disagreement_score":0.004353033,"about_ca_system_score_codex":0.000022591521,"about_ca_system_score_gemma":0.00012918911,"threshold_uncertainty_score":0.75949866},"labels":[],"label_agreement":null},{"id":"W2808094513","doi":"10.2147/clep.s160347","title":"The impact of rheumatoid arthritis on the risk of adverse events following joint replacement: a real-world cohort study","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Total Knee Arthroplasty Outcomes","field":"Medicine","cited_by":27,"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":"NIHR Oxford Biomedical Research Centre; National Institutes of Health; Medical Research Council; National Institute for Health and Care Research; Arthritis Society","keywords":"Medicine; Rheumatoid arthritis; Cohort; Adverse effect; Cohort study; Arthritis; Internal medicine","score_opus":0.09218376995586491,"score_gpt":0.4382523417608776,"score_spread":0.34606857180501266,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2808094513","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9942266,0.00012823059,0.000032823165,0.0016964077,0.0008017491,0.0014605676,0.000015618944,0.000033885248,0.00160413],"genre_scores_gemma":[0.9970018,0.0017978296,0.00016687813,0.0004655727,0.00015844963,0.00004660167,0.0000040438526,0.00002587111,0.00033292425],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99142563,0.0042502005,0.0030615407,0.0005100362,0.00027010674,0.00048245743],"domain_scores_gemma":[0.9695858,0.02704577,0.0015662007,0.0014610122,0.00014875433,0.00019246005],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.021293906,0.00024425384,0.001959836,0.00008150601,0.00021086408,9.114473e-7,0.00024338698,0.00014894907,0.00022359322],"category_scores_gemma":[0.07002441,0.00011589421,0.0013279489,0.00024193186,0.0010152955,0.000025497115,0.00020313878,0.0006343101,0.00013958944],"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":[0.0012372617,0.0006252845,0.99103487,0.000004024896,0.0015155172,0.000006606893,0.0001279953,0.00002216646,0.0000086719,0.00045411833,0.0032201312,0.0017433614],"study_design_scores_gemma":[0.0032085746,0.007923616,0.9863651,0.00020577715,0.00010314853,0.000008132987,0.0002626837,0.000375626,0.000016459631,0.0011138512,0.00032530082,0.000091727175],"about_ca_topic_score_codex":0.001998756,"about_ca_topic_score_gemma":0.0008979982,"teacher_disagreement_score":0.048730504,"about_ca_system_score_codex":0.00007822152,"about_ca_system_score_gemma":0.00015926131,"threshold_uncertainty_score":0.93780917},"labels":[],"label_agreement":null},{"id":"W2851286693","doi":"10.2147/clep.s156772","title":"Asthma is not associated with the need for surgery in Crohn’s disease when controlling for smoking status: a population-based cohort study","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; McGill University Health Centre; University of Calgary; McGill University; Children's Hospital of Eastern Ontario; Institute for Clinical Evaluative Sciences; Research Canada; Agricultural Research Institute of Ontario; University of British Columbia","funders":"Canadian Institutes of Health Research; Government of Alberta","keywords":"Medicine; Asthma; Cohort; Crohn's disease; Disease; Internal medicine; Proportional hazards model; Inflammatory bowel disease; Cohort study; Population; Environmental health","score_opus":0.08822568545052938,"score_gpt":0.3898081600160237,"score_spread":0.3015824745654943,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2851286693","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.980582,0.0000668949,0.014374352,0.0017184375,0.00033069088,0.0026252596,0.0002678192,0.00002751463,0.000007008318],"genre_scores_gemma":[0.98644537,0.0000058773176,0.00047455597,0.010554152,0.000629223,0.0009028321,0.00086113065,0.000052459436,0.000074431795],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9948709,0.0019928948,0.0014348231,0.00084053085,0.00012453695,0.0007363265],"domain_scores_gemma":[0.98598045,0.011917647,0.0007422439,0.00067811186,0.0003948022,0.00028673466],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009932332,0.00028137767,0.00092036906,0.00006880201,0.00023062101,0.00001743158,0.00024615414,0.0003410695,0.000025436724],"category_scores_gemma":[0.053005543,0.00020506392,0.0004668729,0.000083982944,0.0003504213,0.0000068590753,0.00006120077,0.00020276714,0.0000036597037],"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":[0.0072954884,0.00032946156,0.98724055,0.000019323648,0.00027564255,0.000007217618,0.000028497145,0.00046064108,0.000014573195,0.00004506951,0.0039352295,0.00034831456],"study_design_scores_gemma":[0.0036976286,0.0003939197,0.9862088,0.00003377906,0.00015336333,1.9877929e-7,0.000031829964,0.006471356,0.0000062677505,0.0005684878,0.0021695911,0.0002647547],"about_ca_topic_score_codex":0.00011606485,"about_ca_topic_score_gemma":0.0012795201,"teacher_disagreement_score":0.04307321,"about_ca_system_score_codex":0.000058314992,"about_ca_system_score_gemma":0.00042861313,"threshold_uncertainty_score":0.9549714},"labels":[],"label_agreement":null},{"id":"W2889449103","doi":"10.2147/clep.s173449","title":"Effect of opioid-substitution therapy and mental health counseling on HIV risk among hepatitis C-infected individuals","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Hepatitis C virus research","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"AIDS Vancouver; BC Centre for Disease Control; Public Health Ontario; University of Toronto; University of British Columbia","funders":"British Columbia Centre for Disease Control; Provincial Health Services Authority","keywords":"Medicine; Hazard ratio; Internal medicine; Hepatitis C; Coinfection; Proportional hazards model; Incidence (geometry); Cohort; Population; Hepatitis C virus; Hepatitis B; Cohort study; Immunology; Human immunodeficiency virus (HIV); Confidence interval; Environmental health; Virus","score_opus":0.10968502629507274,"score_gpt":0.48273285681057465,"score_spread":0.37304783051550194,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889449103","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9947964,0.00090525456,0.0005508895,0.0018067881,0.0004306106,0.0010536757,0.000094514806,0.0000622189,0.00029966372],"genre_scores_gemma":[0.96484345,0.03100918,0.0007549226,0.0026932782,0.0004888748,0.000033597884,0.00012293772,0.000022990866,0.000030736228],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98987025,0.007241885,0.0015732107,0.00056059234,0.00023235033,0.0005216851],"domain_scores_gemma":[0.9829566,0.015402916,0.00064760086,0.00042716938,0.0001387163,0.00042701603],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.018751286,0.00021039849,0.0015975097,0.00014814401,0.00018400955,0.0000044640806,0.000117597876,0.00040019213,0.00018315038],"category_scores_gemma":[0.031624768,0.00015358075,0.00019480982,0.00018481236,0.0018625967,0.00005090969,0.000062814885,0.00079205836,0.00011648622],"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":[0.00063377625,0.000120809214,0.8721166,0.000063262385,0.00012302063,0.0000031432053,0.00011799449,0.0000010886836,0.000027970993,0.00022060587,0.0039226348,0.122649044],"study_design_scores_gemma":[0.0037777484,0.016587297,0.9655283,0.00025823538,0.00002568214,0.0000060125835,0.0000037111074,0.0023873076,0.00033343205,0.0002863972,0.010693813,0.00011205791],"about_ca_topic_score_codex":0.001338444,"about_ca_topic_score_gemma":0.00014818853,"teacher_disagreement_score":0.12253699,"about_ca_system_score_codex":0.00008680169,"about_ca_system_score_gemma":0.00020405882,"threshold_uncertainty_score":0.9765323},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2894396991","doi":"10.2147/clep.s175063","title":"Use of bone-modifying agents among breast cancer patients with bone metastasis: evidence from oncology practices in the US","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Bone health and treatments","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 Toronto; Novelis (Canada)","funders":"Amgen","keywords":"Denosumab; Medicine; Breast cancer; Bone metastasis; Zoledronic acid; Internal medicine; Oncology; Metastasis; Incidence (geometry); Cancer; Osteoporosis","score_opus":0.47271015704549096,"score_gpt":0.5491922766556715,"score_spread":0.07648211961018053,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2894396991","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9874049,0.00093030557,0.00003500918,0.010287638,0.00041177685,0.0007737574,0.00008379116,0.000016901853,0.000055923094],"genre_scores_gemma":[0.98298556,0.0013844665,0.0029771214,0.012236119,0.00021760815,0.00011201772,0.000041872354,0.000018105651,0.00002710357],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9928499,0.003784056,0.001983465,0.0006343203,0.00020960999,0.0005386812],"domain_scores_gemma":[0.972596,0.023589335,0.0025733192,0.0006381916,0.00029761312,0.0003055685],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00474301,0.00021724786,0.0016642053,0.00008649032,0.00008075344,0.000004577449,0.00015511645,0.00037692883,0.00014954236],"category_scores_gemma":[0.022593504,0.00011978466,0.00016043715,0.00024227635,0.0010258044,0.00026268204,0.00008703524,0.0006907035,0.00003931322],"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":[0.0035948462,0.0012650752,0.972393,0.000042160318,0.00023668239,0.00006887081,0.0001887349,0.00000409366,0.000001971922,0.0000112213165,0.001456551,0.020736797],"study_design_scores_gemma":[0.0030246219,0.0024630527,0.9906781,0.00050116057,0.0005216232,0.000022459042,0.000040797302,0.0003391032,0.000002809406,0.00010905643,0.0022017444,0.00009549303],"about_ca_topic_score_codex":0.029360041,"about_ca_topic_score_gemma":0.007885008,"teacher_disagreement_score":0.021475034,"about_ca_system_score_codex":0.00011357342,"about_ca_system_score_gemma":0.00038279468,"threshold_uncertainty_score":0.98563963},"labels":[],"label_agreement":null},{"id":"W2900387885","doi":"10.2147/clep.s178056","title":"Rural and urban disparities in the care of Canadian patients with inflammatory bowel disease: a population-based study","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":89,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Saskatchewan; Child and Family Research Institute; University of British Columbia; University of Alberta; University of Ottawa; McGill University Health Centre; Ottawa Hospital; CancerCare Manitoba; University of Calgary; Mount Sinai Hospital; University of Manitoba; Dalhousie University; University of Toronto; Ottawa Public Health; Children's Hospital of Eastern Ontario","funders":"Canadian Association of Gastroenterology","keywords":"Medicine; Rate ratio; Inflammatory bowel disease; Rural area; Incidence (geometry); Population; Confounding; Emergency department; Health care; Emergency medicine; Demography; Environmental health; Disease; Internal medicine","score_opus":0.022832565028453634,"score_gpt":0.3183923383919477,"score_spread":0.29555977336349404,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2900387885","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99880445,0.00018820704,0.000007792373,0.00023534853,0.00010241843,0.0005267126,0.00006999413,0.000004474108,0.000060574934],"genre_scores_gemma":[0.9977842,0.000008836528,0.00004884612,0.0016549198,0.0001749362,0.000050768216,0.00025691625,0.00001303017,0.0000075787166],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9973895,0.001285808,0.0006577918,0.00030280708,0.00007945302,0.0002846362],"domain_scores_gemma":[0.9985533,0.00040465372,0.00019082049,0.00046563186,0.00014146267,0.00024415765],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0010606837,0.00014432111,0.00027708404,0.0000967694,0.000086929045,0.0000048179368,0.00020976459,0.00014562257,0.000008340646],"category_scores_gemma":[0.0029069632,0.000099008976,0.00007061214,0.000072510826,0.00044649682,0.0000043716354,0.000045605277,0.00013273982,0.0000022765635],"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":[0.00051623874,0.00010138561,0.9985417,0.00002656581,0.000022231641,0.000010175381,0.00007763055,0.00003666524,6.430276e-7,0.00020041256,0.00031800268,0.00014834556],"study_design_scores_gemma":[0.001128555,0.00053235557,0.9973942,0.000021269418,0.000029691888,1.3909336e-7,0.00020486617,0.000055202825,5.1037364e-7,0.00007361338,0.00043844842,0.00012116616],"about_ca_topic_score_codex":0.004516081,"about_ca_topic_score_gemma":0.0950717,"teacher_disagreement_score":0.09055562,"about_ca_system_score_codex":0.000016924012,"about_ca_system_score_gemma":0.0002601112,"threshold_uncertainty_score":0.9214409},"labels":[],"label_agreement":null},{"id":"W2903526766","doi":"10.2147/clep.s184319","title":"Association between gestational diabetes mellitus and depression in parents: a retrospective cohort study","year":2018,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Gestational Diabetes Research and Management","field":"Medicine","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; McGill University Health Centre","funders":"Diabetes Canada","keywords":"Medicine; Gestational diabetes; Obstetrics; Depression (economics); Edinburgh Postnatal Depression Scale; Pregnancy; Cohort; Proportional hazards model; Hazard ratio; Cohort study; Incidence (geometry); Diabetes mellitus; Gestational age; Postpartum depression; Gestation; Pediatrics; Internal medicine; Confidence interval; Endocrinology; Depressive symptoms","score_opus":0.09609273114725701,"score_gpt":0.45700734472524523,"score_spread":0.3609146135779882,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2903526766","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99568075,0.00015915616,0.00015507547,0.0023642695,0.00015970775,0.0008758661,0.000014821783,0.000023311806,0.0005670349],"genre_scores_gemma":[0.9969297,0.00027964747,0.00075601216,0.0011053372,0.00045196604,0.000110500085,0.00011079496,0.000008991985,0.00024705435],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99643075,0.0014793008,0.00092133833,0.00050609914,0.0002626706,0.00039982275],"domain_scores_gemma":[0.9905468,0.008464761,0.00026834308,0.0001874485,0.0003142104,0.00021842268],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008953921,0.0001086192,0.0007751857,0.000118265125,0.000058665395,0.0000057037237,0.00006881554,0.00016811675,0.0000900187],"category_scores_gemma":[0.053556632,0.000088882065,0.000064000975,0.00015743801,0.00019631271,0.000060470953,0.00011355117,0.00037686867,0.00006618808],"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":[0.00005964666,0.00032141144,0.9916729,0.000017318418,0.00027531025,0.0000034274367,0.00004028154,0.0000010561606,0.000002076224,0.00013058244,0.0021533724,0.0053226585],"study_design_scores_gemma":[0.0016389715,0.0012253184,0.9895617,0.000038289574,0.00009930554,1.400554e-7,0.00006786027,0.0004548646,0.0000025064955,0.006324133,0.0005141669,0.00007278859],"about_ca_topic_score_codex":0.00007669672,"about_ca_topic_score_gemma":0.00006507587,"teacher_disagreement_score":0.04460271,"about_ca_system_score_codex":0.00019149744,"about_ca_system_score_gemma":0.00004782152,"threshold_uncertainty_score":0.9544157},"labels":[],"label_agreement":null},{"id":"W2906754414","doi":"10.2147/clep.s187381","title":"Inhospital death is a biased measure of fatal outcome from bloodstream infection","year":2019,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Royal Inland Hospital","funders":"","keywords":"Medicine; Case fatality rate; Pediatrics; Cause of death; Mortality rate; Population; Cohort study; Cohort; Emergency medicine; Internal medicine; Epidemiology; Disease","score_opus":0.46352221678043537,"score_gpt":0.5178501181490499,"score_spread":0.05432790136861454,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2906754414","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9931924,0.0004677558,0.00011495434,0.0041495273,0.0008061321,0.00042826388,0.00005840156,0.000048069567,0.0007345472],"genre_scores_gemma":[0.9914796,0.00025458715,0.0014989587,0.00625513,0.0002565994,0.000025462241,0.00012414603,0.00001850736,0.000087061184],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99671733,0.00063728506,0.0017008138,0.00053635787,0.00012302274,0.00028517246],"domain_scores_gemma":[0.99155563,0.0070323567,0.0005028196,0.0005820126,0.00010174294,0.00022543008],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0019259069,0.000194636,0.0016840262,0.00006238986,0.000019658602,0.0000017494696,0.00007638623,0.00039732252,0.00092013675],"category_scores_gemma":[0.009904357,0.00013852614,0.0006798797,0.00009297776,0.00013541372,0.000034037115,0.00005759488,0.00036515255,0.0005543912],"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":[0.0001780212,0.0011313068,0.98883015,0.000021717975,0.0004476789,0.0000129252185,0.000021392247,0.0000026022624,0.00003431199,0.00031264208,0.0015723204,0.007434949],"study_design_scores_gemma":[0.004099693,0.0025542823,0.98954904,0.000094911244,0.00037999696,0.0000073858528,0.000010215928,0.00017821342,0.00032408375,0.00076697575,0.0019249512,0.00011023387],"about_ca_topic_score_codex":0.0014963009,"about_ca_topic_score_gemma":0.000028436636,"teacher_disagreement_score":0.00797845,"about_ca_system_score_codex":0.00004527676,"about_ca_system_score_gemma":0.00007959052,"threshold_uncertainty_score":0.99999315},"labels":[],"label_agreement":null},{"id":"W2938026556","doi":"10.2147/clep.s191418","title":"<p>Epidemiology of systemic sclerosis and systemic sclerosis-associated interstitial lung disease</p>","year":2019,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Systemic Sclerosis and Related Diseases","field":"Medicine","cited_by":254,"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":"Merck KGaA","keywords":"Medicine; Interstitial lung disease; Epidemiology; Incidence (geometry); Population; Disease; Internal medicine; Demography; Lung; Environmental health","score_opus":0.12248149287740245,"score_gpt":0.37305250827074143,"score_spread":0.250571015393339,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2938026556","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.88350004,0.10875178,0.00041116707,0.0015650247,0.0033170623,0.0016253697,0.00017094315,0.00023525968,0.00042335686],"genre_scores_gemma":[0.95934755,0.035813577,0.00017266454,0.003010099,0.00070185465,0.00012200334,0.00021033357,0.00010032477,0.0005216023],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9823684,0.0075106714,0.006649626,0.0017889315,0.00033954115,0.0013428334],"domain_scores_gemma":[0.9663573,0.027028713,0.0030543148,0.0015338525,0.00047307764,0.0015527704],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.013544312,0.0007319732,0.0061003664,0.00033677733,0.00015016278,0.000008292338,0.0005411838,0.0018319311,0.00030306852],"category_scores_gemma":[0.050467532,0.0005685873,0.0014839405,0.0004357892,0.0013866856,0.00012876822,0.00041755958,0.0010057973,0.00033672276],"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":[0.0015284503,0.0005417686,0.9676312,0.0015401626,0.0012307881,0.000021711081,0.000087279106,0.0001299495,0.010552133,0.0042091613,0.010018375,0.0025090096],"study_design_scores_gemma":[0.00682494,0.0011390836,0.94432575,0.03425909,0.0016971183,0.00022575163,0.00017274206,0.010150623,0.000018209521,0.00042741225,0.00015832466,0.0006009409],"about_ca_topic_score_codex":0.00007183591,"about_ca_topic_score_gemma":0.000007925476,"teacher_disagreement_score":0.07584751,"about_ca_system_score_codex":0.00032380383,"about_ca_system_score_gemma":0.0004458819,"threshold_uncertainty_score":0.9996766},"labels":[],"label_agreement":null},{"id":"W2948485828","doi":"10.2147/clep.s198499","title":"<p>Access to palliative care for cancer patients between diagnosis and death: a national cohort study</p>","year":2019,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Palliative Care and End-of-Life Issues","field":"Medicine","cited_by":37,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Michel-Sarrazin; Centre hospitalier universitaire de Québec","funders":"Ligue Contre le Cancer","keywords":"Medicine; Palliative care; Interquartile range; Cancer; Retrospective cohort study; Cohort; Odds ratio; Place of death; Cohort study; Disease; Emergency medicine; Internal medicine; Nursing","score_opus":0.44993328031010443,"score_gpt":0.5845157339571131,"score_spread":0.13458245364700866,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2948485828","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9909685,0.0022658783,0.00024047538,0.0021580877,0.00056494575,0.0028018854,0.00043810275,0.000026811389,0.00053527654],"genre_scores_gemma":[0.99012744,0.002146494,0.0016003328,0.004161777,0.0006137939,0.0009312867,0.00026038065,0.000027684493,0.00013080564],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99649423,0.0006123732,0.0014721084,0.0007729647,0.00025248772,0.00039584492],"domain_scores_gemma":[0.97589594,0.021936037,0.00048418695,0.00026783205,0.0010827066,0.00033327244],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0021754166,0.00024400397,0.0019955179,0.0001119501,0.00006014045,0.000005428522,0.00022126717,0.0003232662,0.00009566423],"category_scores_gemma":[0.03152346,0.00019375133,0.0002582104,0.00016054718,0.0001312787,0.00006996312,0.00019826918,0.00024166437,0.00003246403],"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":[0.0002744,0.00029184474,0.9870032,0.00014042741,0.00048552454,0.0000011793571,0.00061274564,0.000015206225,0.0000024831663,0.00028941935,0.0053691063,0.005514466],"study_design_scores_gemma":[0.0027155129,0.0024332318,0.9280575,0.00036181838,0.00028709343,7.612649e-8,0.000093782095,0.00004735009,0.000019567855,0.00055769423,0.065248735,0.00017761685],"about_ca_topic_score_codex":0.00013782376,"about_ca_topic_score_gemma":0.00026826764,"teacher_disagreement_score":0.05987963,"about_ca_system_score_codex":0.00015311371,"about_ca_system_score_gemma":0.0001766831,"threshold_uncertainty_score":0.97663444},"labels":[],"label_agreement":null},{"id":"W2969639833","doi":"10.2147/clep.s206267","title":"<p>Sensitivity, specificity, positive and negative predictive values of identifying atrial fibrillation using administrative data: a systematic review and meta-analysis</p>","year":2019,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":42,"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; University of British Columbia","funders":"","keywords":"Atrial fibrillation; Medicine; Meta-analysis; Predictive value; Sensitivity (control systems); Internal medicine; Cardiology; Pooled analysis; Systematic review; MEDLINE; Biology","score_opus":0.7152435233242256,"score_gpt":0.5745098365626433,"score_spread":0.14073368676158227,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2969639833","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":[0.00004694655,0.9864989,0.0037290382,0.0006690389,0.0003133873,0.007573467,0.00097390934,0.000039347782,0.0001559956],"genre_scores_gemma":[0.0005277533,0.99377143,0.0027456335,0.00026992906,0.0011101633,0.000008435945,0.0011057811,0.000054857337,0.00040601942],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9741501,0.01603131,0.007113473,0.0018316864,0.00043100523,0.00044239406],"domain_scores_gemma":[0.9217727,0.06682649,0.009188279,0.0014720573,0.00046205652,0.00027841568],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.020632576,0.00087970996,0.031176616,0.00051602,0.00013246958,0.0000338707,0.00026505176,0.0008882481,0.00007414536],"category_scores_gemma":[0.06427413,0.00059500866,0.0055782245,0.0008899628,0.00087503495,0.00031436305,0.0010647497,0.00062657165,0.000020277144],"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":[0.00025179246,0.000020624335,0.001807421,0.59967804,0.39223802,0.00003767408,0.00007020831,0.000039638282,4.947654e-7,0.0026453685,0.00027967725,0.002931025],"study_design_scores_gemma":[0.00058768876,0.00047546122,0.0012750329,0.02882922,0.96188277,0.000089861096,0.000025507528,0.002314073,9.8252215e-8,0.0004858673,0.0035832524,0.00045114622],"about_ca_topic_score_codex":0.000024177443,"about_ca_topic_score_gemma":0.000022761205,"teacher_disagreement_score":0.5708488,"about_ca_system_score_codex":0.000110849294,"about_ca_system_score_gemma":0.0003373522,"threshold_uncertainty_score":0.9996501},"labels":[],"label_agreement":null},{"id":"W2973224614","doi":"10.2147/clep.s217969","title":"<p>Identifying pediatric diabetes cases from health administrative data: a population-based validation study in Quebec, Canada</p>","year":2019,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Diabetes and associated disorders","field":"Biochemistry, Genetics and Molecular Biology","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"McGill University; Children's Hospital of Eastern Ontario; University of Ottawa; Centre hospitalier de l'Université Laval; Centre Hospitalier Universitaire Sainte-Justine; Montreal Children's Hospital; Institut National de Santé Publique du Québec; Centre Hospitalier Universitaire de Sherbrooke; McGill University Health Centre","funders":"Ministère de la Santé; Ministère de la Santé et des Services sociaux","keywords":"Medicine; Population; Incidence (geometry); Confidence interval; Disease; Type 2 diabetes; Diabetes mellitus; Pediatrics; Environmental health; Gerontology; Internal medicine","score_opus":0.11704287954872515,"score_gpt":0.4149209764108332,"score_spread":0.29787809686210803,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2973224614","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9945837,0.0017656515,0.000054184075,0.0017623516,0.00077902654,0.0006324278,0.00036673175,0.000016961632,0.000038925682],"genre_scores_gemma":[0.98911715,0.00020679385,0.00023288693,0.0028010353,0.00037207382,0.00003478432,0.0070394473,0.000026911304,0.00016893478],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99371475,0.002636956,0.0017609473,0.0011199851,0.0001594508,0.00060792733],"domain_scores_gemma":[0.9936475,0.00427198,0.00091625284,0.0008767692,0.00007675257,0.00021072173],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0034473278,0.00026899634,0.0010115621,0.000067353256,0.000088368506,0.000021467486,0.00047797358,0.0003359201,0.00006266499],"category_scores_gemma":[0.011723536,0.00026061852,0.00013526157,0.00019467087,0.00005630646,0.000017163113,0.0002129452,0.0002729287,0.000017948287],"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":[0.00004676911,0.00037386344,0.9899439,0.000020764977,0.00008760741,0.00000452913,0.000011781418,0.00037401888,0.000103838895,0.000020335796,0.007865369,0.0011472263],"study_design_scores_gemma":[0.0014750967,0.0005895962,0.9933502,0.000022019409,0.000051696188,1.2064528e-7,0.00011013867,0.00085405953,0.000022861206,0.00013134195,0.0031092078,0.00028364555],"about_ca_topic_score_codex":0.30029044,"about_ca_topic_score_gemma":0.9171463,"teacher_disagreement_score":0.6168559,"about_ca_system_score_codex":0.0001146637,"about_ca_system_score_gemma":0.0014291642,"threshold_uncertainty_score":0.9999846},"labels":[],"label_agreement":null},{"id":"W3011290754","doi":"10.2147/clep.s245642","title":"<p>Epidemiology of Atypical Hemolytic Uremic Syndrome: A Systematic Literature Review</p>","year":2020,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Complement system in diseases","field":"Immunology and Microbiology","cited_by":122,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Incidence (geometry); Epidemiology; Population; MEDLINE; Pediatrics; Systematic review; Meta-analysis; Demography; Environmental health; Internal medicine","score_opus":0.17800049969604056,"score_gpt":0.45115739685951556,"score_spread":0.273156897163475,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3011290754","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":[0.00002100713,0.9842174,0.00024016187,0.0015650061,0.00543771,0.006984978,0.0010462331,0.0003075589,0.00017995149],"genre_scores_gemma":[0.00017996764,0.9877384,0.00046704817,0.004922396,0.0004673721,0.0014129432,0.0037224768,0.00019697387,0.00089244865],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.90057,0.071224175,0.022555498,0.0033215452,0.00015679254,0.002172034],"domain_scores_gemma":[0.8715516,0.11250482,0.011414371,0.00358905,0.00041198754,0.0005281733],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.026694253,0.002132806,0.041487206,0.0005173516,0.00020090379,0.0000071987993,0.0034090376,0.0056794737,0.001546238],"category_scores_gemma":[0.22853503,0.0015325155,0.008094139,0.0012364778,0.0023024394,0.000099056015,0.0014867267,0.004114214,0.005443708],"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":[0.0001473672,0.00043308802,0.00074363756,0.7742198,0.00419301,0.0003693852,0.000014467139,5.6157063e-7,0.0000031057941,0.0091069015,0.18887691,0.021891732],"study_design_scores_gemma":[0.00062786636,0.00067690265,0.00021295802,0.4109492,0.0057388055,0.0048035313,0.000002283336,0.0000061174032,7.412237e-8,0.0006865545,0.5755629,0.0007328004],"about_ca_topic_score_codex":0.0000042395627,"about_ca_topic_score_gemma":0.0000074169207,"teacher_disagreement_score":0.386686,"about_ca_system_score_codex":0.00031226917,"about_ca_system_score_gemma":0.0010138802,"threshold_uncertainty_score":0.99936646},"labels":[],"label_agreement":null},{"id":"W3022623176","doi":"10.2147/clep.s242097","title":"<p>Synthetic and External Controls in Clinical Trials – A Primer for Researchers</p>","year":2020,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":258,"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; University of British Columbia; Impact","funders":"","keywords":"Medicine; Critical appraisal; Control (management); Set (abstract data type); Key (lock); Computational biology; Computer science; Data science; Alternative medicine; Biology; Pathology; Artificial intelligence","score_opus":0.9492111274371475,"score_gpt":0.7575105690970199,"score_spread":0.19170055834012767,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3022623176","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000011668649,0.87404394,0.108852655,0.0030612524,0.0035742451,0.0091456175,0.0007514943,0.00014782343,0.00041133186],"genre_scores_gemma":[0.0000019493832,0.61564463,0.37731823,0.0013532853,0.004015779,0.0011987565,0.000017695978,0.00018880898,0.00026087527],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.6138697,0.31972584,0.056926277,0.005639709,0.0008442439,0.0029941925],"domain_scores_gemma":[0.041003685,0.9437801,0.01053829,0.0022368694,0.00036707008,0.0020739646],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity"],"consensus_categories":["metaresearch","metaepi_narrow","research_integrity"],"category_scores_codex":[0.4770573,0.0014844435,0.045784768,0.00034992685,0.000133862,0.00006247226,0.0020245838,0.0066491365,0.000505089],"category_scores_gemma":[0.98902917,0.00105018,0.0078935,0.00053388556,0.0038202028,0.000074441676,0.0011964813,0.006176051,0.00030124953],"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":[0.0012149722,0.00060310203,0.00014529093,0.005978483,0.00095610024,0.000083072504,0.0000070032534,8.94454e-8,5.1414264e-7,0.1445116,0.010506681,0.8359931],"study_design_scores_gemma":[0.0036494737,0.0012585713,0.00007637265,0.0055108923,0.002142259,0.000023900013,0.0000015692088,0.00025714835,5.7250887e-8,0.45852607,0.52802104,0.000532623],"about_ca_topic_score_codex":0.000005281159,"about_ca_topic_score_gemma":0.000013466765,"teacher_disagreement_score":0.8354605,"about_ca_system_score_codex":0.00019277817,"about_ca_system_score_gemma":0.0014599394,"threshold_uncertainty_score":0.9997905},"labels":[],"label_agreement":null},{"id":"W3028673208","doi":"10.2147/clep.s242080","title":"<p>Meta-Analyses Proved Inconsistent in How Missing Data Were Handled Across Their Included Primary Trials: A Methodological Survey</p>","year":2020,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","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":"Veterans Affairs Canada; Sunnybrook Health Science Centre; University of Toronto; McMaster University; Health Sciences Centre; Impact","funders":"","keywords":"Missing data; Meta-analysis; Systematic review; Medicine; MEDLINE; Statistics; Clinical trial; Data mining; Computer science; Internal medicine; Mathematics","score_opus":0.9942150291509889,"score_gpt":0.7434143475047398,"score_spread":0.25080068164624914,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028673208","genre_codex":"methods","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.20934436,0.06674291,0.4842418,0.22338364,0.0034454123,0.00771551,0.0026932503,0.00014294169,0.0022901515],"genre_scores_gemma":[0.70750964,0.0018811161,0.23166609,0.052467976,0.0020582422,0.0003567234,0.0014255275,0.0001099067,0.002524793],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.10682896,0.8078198,0.072826564,0.007503453,0.0033294298,0.0016918303],"domain_scores_gemma":[0.026883774,0.9176652,0.037907567,0.014184093,0.0017948349,0.0015645593],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.8889294,0.0011431504,0.05438826,0.00030285082,0.00033414923,0.00065080106,0.008987207,0.0012996024,0.0041627255],"category_scores_gemma":[0.97667277,0.00047003257,0.012385113,0.002544274,0.0011975608,0.00047666693,0.0044013914,0.0015531307,0.00081219064],"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":[0.0020439958,0.001304518,0.27860957,0.00040736952,0.026681341,0.00010777043,0.0007506516,0.00044958066,0.0036797584,0.001074353,0.5140764,0.17081466],"study_design_scores_gemma":[0.003676334,0.0008152379,0.4128377,0.000094095994,0.009899461,0.00004045341,0.00034983936,0.10468144,0.000029546349,0.027215011,0.4389941,0.001366755],"about_ca_topic_score_codex":0.000037365437,"about_ca_topic_score_gemma":0.0005287001,"teacher_disagreement_score":0.49816525,"about_ca_system_score_codex":0.0000742051,"about_ca_system_score_gemma":0.0005404945,"threshold_uncertainty_score":0.9999969},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3034892553","doi":"10.2147/clep.s231843","title":"<p>MultiCenter Interrupted Time Series Analysis: Incorporating Within and Between-Center Heterogeneity</p>","year":2020,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Health disparities and outcomes","field":"Social Sciences","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":"St. Michael's Hospital; McMaster University; Children's Hospital of Eastern Ontario; Impact; University of Ottawa","funders":"","keywords":"Statistics; Spurious relationship; Interrupted Time Series Analysis; Regression analysis; Statistic; Regression; Logistic regression; Statistical power; Type I and type II errors; Linear regression; Power analysis; Medicine; Econometrics; Mathematics; Algorithm","score_opus":0.1906619591927314,"score_gpt":0.4587500127519849,"score_spread":0.26808805355925347,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3034892553","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8957462,0.0005874398,0.0015083718,0.09916337,0.0006413819,0.0003766348,0.00006785891,0.00018288639,0.0017258622],"genre_scores_gemma":[0.9510335,0.0003560512,0.0036572486,0.043370835,0.0012108602,0.000016343078,0.00006829838,0.000020066487,0.0002668215],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99057657,0.005172501,0.0025015578,0.0007744321,0.00018464246,0.0007903072],"domain_scores_gemma":[0.98918396,0.008497018,0.00086936506,0.00029090696,0.00012254356,0.001036224],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008672905,0.00025716485,0.0018599413,0.000082701845,0.0004338436,0.00004581535,0.0004086528,0.00054343,0.00026508205],"category_scores_gemma":[0.029211812,0.00021898845,0.00045364475,0.00044088147,0.0013196442,0.00024046664,0.00031197513,0.000546451,0.00018688069],"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":[0.00008571764,0.00004117721,0.9876166,0.00004554995,0.00036248894,0.000007337316,0.0010184515,0.00001733936,0.000004726034,0.0068856403,0.0025439265,0.0013710719],"study_design_scores_gemma":[0.0011457434,0.0003854134,0.8980787,0.00007539372,0.00044145138,0.0000012733858,0.0005106445,0.004502716,0.0000045544402,0.0019893532,0.09241211,0.00045270254],"about_ca_topic_score_codex":0.00025207613,"about_ca_topic_score_gemma":0.0027005547,"teacher_disagreement_score":0.08986817,"about_ca_system_score_codex":0.000051908424,"about_ca_system_score_gemma":0.00015431123,"threshold_uncertainty_score":0.9789655},"labels":[],"label_agreement":null},{"id":"W3037313095","doi":"10.2147/clep.s256846","title":"<p>Methods for Measuring the Time of Transfer from Pediatric to Adult Care for Chronic Conditions Using Administrative Data: A Scoping Review</p>","year":2020,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Adolescent and Pediatric Healthcare","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University Health Centre; Montreal Children's Hospital; Children's Hospital of Eastern Ontario; Institute for Clinical Evaluative Sciences; Hospital for Sick Children","funders":"Ministère de la Santé; Ministère de la Santé et des Services sociaux","keywords":"Medicine; Adult care; Health care; Transfer (computing); Intensive care medicine; Gerontology; Computer science; Young adult","score_opus":0.6859905109983805,"score_gpt":0.6573797700767444,"score_spread":0.02861074092163607,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3037313095","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.034926504,0.14190964,0.690373,0.09215109,0.0033812006,0.024618154,0.012292683,0.00019969682,0.00014803134],"genre_scores_gemma":[0.17756432,0.093921185,0.4167435,0.239699,0.055833012,0.005149395,0.010492075,0.0004383722,0.00015914261],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9847573,0.009089073,0.0041052466,0.0010218287,0.00012730871,0.0008992206],"domain_scores_gemma":[0.92596745,0.070908956,0.000872349,0.0008591183,0.00079542055,0.00059669773],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.010334634,0.00030532185,0.0022879983,0.00005447297,0.00074930023,0.0000028741806,0.00082105125,0.0006719905,0.00019200223],"category_scores_gemma":[0.091521904,0.00022889434,0.00047380073,0.00036070865,0.0001860123,0.00009750619,0.00025747038,0.00097653526,0.000051784413],"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":[0.004805253,0.000543375,0.20432065,0.2394178,0.0010985845,0.000011239848,0.019764317,0.00036304328,0.0011821804,0.010888503,0.44714645,0.070458606],"study_design_scores_gemma":[0.03624508,0.01828444,0.08711546,0.1543288,0.015006047,0.000011121463,0.0068888813,0.1120026,0.00012111966,0.011682134,0.551831,0.006483333],"about_ca_topic_score_codex":0.0000385525,"about_ca_topic_score_gemma":0.00027800398,"teacher_disagreement_score":0.2736295,"about_ca_system_score_codex":0.0001263033,"about_ca_system_score_gemma":0.0030646098,"threshold_uncertainty_score":0.933404},"labels":[],"label_agreement":null},{"id":"W3094699016","doi":"10.2147/clep.s265287","title":"<p>Trajectory Modelling Techniques Useful to Epidemiological Research: A Comparative Narrative Review of Approaches</p>","year":2020,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Health disparities and outcomes","field":"Social Sciences","cited_by":522,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University; Université de Sherbrooke; Université de Montréal; Centre Hospitalier de l’Université de Montréal; Centre Hospitalier Universitaire de Sherbrooke; Université du Québec en Abitibi-Témiscamingue","funders":"Fonds de Recherche du Québec - Santé; Canadian Institutes of Health Research; Fondation de l’Université du Québec en Abitibi-Témiscamingue","keywords":"Trajectory; Latent class model; Terminology; Narrative review; Class (philosophy); Consistency (knowledge bases); Computer science; Latent variable; Population; Data science; Econometrics; Data mining; Medicine; Machine learning; Artificial intelligence; Mathematics","score_opus":0.8685960873080928,"score_gpt":0.6385536961698527,"score_spread":0.2300423911382401,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094699016","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":[0.0000065490603,0.95513594,0.0031863062,0.022757264,0.0007077153,0.005177034,0.00012770045,0.00020423012,0.012697272],"genre_scores_gemma":[0.000024927738,0.96482784,0.01612395,0.015166583,0.0020867463,0.0011682273,0.00013239757,0.0000613928,0.0004079349],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8873407,0.09693184,0.010071259,0.0023931896,0.00080274954,0.0024602416],"domain_scores_gemma":[0.8241008,0.16869602,0.003198771,0.001182751,0.00075391046,0.0020677613],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","sts","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.12113516,0.00095144217,0.018813856,0.00034118755,0.0007878671,0.000019293218,0.0023829874,0.0029809384,0.0003735961],"category_scores_gemma":[0.20814736,0.0006881056,0.0031167092,0.0017499501,0.00451391,0.00013270398,0.0008303035,0.004584636,0.00034199966],"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":[0.00009628946,0.00038445636,0.0002853375,0.0804178,0.00050806045,0.00003350939,0.003067471,0.000023628578,1.7991841e-8,0.2436131,0.2658733,0.40569702],"study_design_scores_gemma":[0.000085510226,0.0005202002,0.00004320582,0.05368813,0.00023519318,0.000003960655,0.0003513727,0.00009239038,3.3544985e-8,0.004641125,0.9397881,0.0005508005],"about_ca_topic_score_codex":0.0003547922,"about_ca_topic_score_gemma":0.00041508998,"teacher_disagreement_score":0.67391473,"about_ca_system_score_codex":0.0006375407,"about_ca_system_score_gemma":0.0037693789,"threshold_uncertainty_score":0.999557},"labels":[],"label_agreement":null},{"id":"W3095421842","doi":"10.2147/clep.s267584","title":"<p>A Validation of Methods for the Evaluation of Observational Studies of Screening Mammography: An Exploratory Analysis Based on Simulating Screening Cohorts</p>","year":2020,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's College Hospital; University of Toronto; Public Health Ontario","funders":"Canadian Institutes of Health Research","keywords":"Hazard ratio; Medicine; Observational study; Confidence interval; Statistics; Mammography; Breast cancer screening; Covariate; Breast cancer; Spurious relationship; Matching (statistics); Hazard; Propensity score matching; Proportional hazards model; Econometrics; Mathematics; Internal medicine; Cancer","score_opus":0.8243863299312576,"score_gpt":0.6244583277525714,"score_spread":0.19992800217868623,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3095421842","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.437354,0.0017574732,0.5578242,0.002270596,0.00009397197,0.00061765715,0.000029890394,0.000020607262,0.000031613574],"genre_scores_gemma":[0.7608526,0.00008185404,0.23631378,0.0023056876,0.00020540312,0.000060804974,0.00016531072,0.0000136029885,9.990421e-7],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9905645,0.0049485518,0.0030783445,0.0005832009,0.0005442979,0.000281096],"domain_scores_gemma":[0.94322896,0.05113786,0.0024698914,0.0005768426,0.0024023592,0.00018407742],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.044039767,0.00021134688,0.0021424573,0.0002125031,0.000117659,0.00000277699,0.00023889018,0.00024304411,0.000047599988],"category_scores_gemma":[0.11495876,0.00015648479,0.0009453508,0.0010993044,0.00052086695,0.00013175087,0.00007159148,0.00027611208,2.9331653e-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":[0.0011629391,0.000069072834,0.5230132,0.00009490061,0.0014642208,2.7338257e-7,0.00016006189,0.4216211,0.0009522728,0.0007750422,0.00012693359,0.05056],"study_design_scores_gemma":[0.0012259011,0.0013264605,0.31972098,0.00015940069,0.0034248647,2.0256789e-7,0.00035529098,0.6724048,0.00047591273,0.00068664766,0.000131867,0.0000876342],"about_ca_topic_score_codex":0.00001779742,"about_ca_topic_score_gemma":0.00001869472,"teacher_disagreement_score":0.32349858,"about_ca_system_score_codex":0.00003212401,"about_ca_system_score_gemma":0.00020565077,"threshold_uncertainty_score":0.98436224},"labels":[],"label_agreement":null},{"id":"W3116681383","doi":"10.2147/clep.s288677","title":"<p>Retraction of COVID-19 Pharmacoepidemiology Research Could Have Been Avoided by Effective Use of Reporting Guidelines</p>","year":2020,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","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":"Ottawa Hospital; University of Ottawa; Institute for Clinical Evaluative Sciences; SickKids Foundation; University of Toronto","funders":"Wellcome Trust","keywords":"Pharmacoepidemiology; Medicine; Coronavirus disease 2019 (COVID-19); Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); 2019-20 coronavirus outbreak; Internal medicine; Pediatrics; Pharmacology; Virology","score_opus":0.9713400994553343,"score_gpt":0.7444786273701581,"score_spread":0.22686147208517615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3116681383","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5864253,0.0074989046,0.26572087,0.13145085,0.0021677932,0.0046345047,0.00028930014,0.000055727207,0.0017567396],"genre_scores_gemma":[0.9246048,0.002483248,0.029852698,0.038956147,0.0013278932,0.00023725249,0.00014236593,0.00006650183,0.0023291486],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.29998085,0.5099773,0.17540884,0.0063639823,0.0062626954,0.0020063568],"domain_scores_gemma":[0.06400152,0.81839186,0.09999215,0.005532914,0.009988187,0.0020933778],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.77184165,0.0005746224,0.018656313,0.00052225793,0.0002472024,0.000049980918,0.0024534857,0.0013467098,0.003819326],"category_scores_gemma":[0.99041903,0.0003289512,0.005182189,0.0020428884,0.0018070794,0.0002983624,0.0007603153,0.0021122932,0.0007120386],"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":[0.00024092029,0.00011883332,0.2979188,0.00017723153,0.00049324095,0.00001725315,0.00010453845,0.0009689668,0.0038140055,0.0015757929,0.6846203,0.009950118],"study_design_scores_gemma":[0.0010054922,0.0009566558,0.043131396,0.000085131454,0.00051508425,0.000038486345,0.00017780704,0.103888944,0.0001742381,0.015763117,0.8339031,0.00036057047],"about_ca_topic_score_codex":0.00024356156,"about_ca_topic_score_gemma":0.00007330786,"teacher_disagreement_score":0.33817944,"about_ca_system_score_codex":0.00012123513,"about_ca_system_score_gemma":0.0005554787,"threshold_uncertainty_score":0.99994975},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"reporting","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3128504017","doi":"10.2147/clep.s292546","title":"Comparing Health Administrative and Clinical Registry Data: Trends in Incidence and Prevalence of Pediatric Inflammatory Bowel Disease in British Columbia","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; University of Alberta; BC Children's Hospital; University of British Columbia","funders":"CH.I.L.D. Foundation; BC Children's Hospital","keywords":"Medicine; Incidence (geometry); Inflammatory bowel disease; Disease; Disease registry; Epidemiology; Pediatrics; Environmental health; Internal medicine","score_opus":0.12768705028320856,"score_gpt":0.4363177961941763,"score_spread":0.30863074591096773,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3128504017","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99242276,0.0066135433,0.000043896944,0.00034313314,0.00018028013,0.00017166008,0.00017822928,0.000007751447,0.00003876053],"genre_scores_gemma":[0.98314977,0.014720476,0.0006657873,0.00074148015,0.00029073664,0.000014793924,0.00023833557,0.000014977062,0.00016362571],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9924045,0.0033625027,0.0025146368,0.0012218958,0.00009029085,0.00040617568],"domain_scores_gemma":[0.99624574,0.0014905708,0.0007426252,0.0009206584,0.00008030628,0.00052009453],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0067024822,0.00015213015,0.0008452155,0.000042122218,0.000049374827,0.000019565321,0.0003655798,0.00035810343,0.000019633726],"category_scores_gemma":[0.021665148,0.00022755186,0.000112288006,0.00012288858,0.00092310895,0.000020597994,0.0008634771,0.0005106698,9.704594e-7],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00019604125,0.00025054082,0.9944564,0.00037949768,0.000021683336,0.00043403907,0.000007756273,0.0000148792915,0.000003610151,0.000027682583,0.0018636704,0.0023442048],"study_design_scores_gemma":[0.0012949922,0.00018442438,0.99672145,0.00011916071,0.00003303328,0.000022331444,0.00003101034,0.00095747964,3.763135e-7,0.00028528707,0.00016253773,0.00018794683],"about_ca_topic_score_codex":0.0005222049,"about_ca_topic_score_gemma":0.043269105,"teacher_disagreement_score":0.0427469,"about_ca_system_score_codex":0.000019155017,"about_ca_system_score_gemma":0.00092468184,"threshold_uncertainty_score":0.9865758},"labels":[],"label_agreement":null},{"id":"W3143965524","doi":"10.2147/clep.s303560","title":"Incidence of Atrial Fibrillation in Persons with Very High Serum Levels of N-Terminal Pro-B-Type Natriuretic Peptide: The Multi-Ethnic Study of Atherosclerosis","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Atrial Fibrillation Management and Outcomes","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":"Université de Montréal","funders":"National Center for Advancing Translational Sciences; National Heart, Lung, and Blood Institute","keywords":"Atrial fibrillation; Medicine; Incidence (geometry); Internal medicine; Natriuretic peptide; Cardiology; Fibrillation; Heart failure","score_opus":0.40624930632248424,"score_gpt":0.4896038429537521,"score_spread":0.08335453663126785,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3143965524","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9966675,0.00046064917,0.00018633727,0.0015331098,0.0003292578,0.0007842624,0.000011710578,0.000010963506,0.000016230095],"genre_scores_gemma":[0.99689746,0.00022909848,0.002145564,0.00011789565,0.00036256484,0.0000013202354,0.0000065231925,0.000013943855,0.00022560859],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99525964,0.0018994901,0.0020360614,0.0003751944,0.0002135523,0.00021607577],"domain_scores_gemma":[0.98837376,0.009779737,0.0010431051,0.00046899443,0.000273025,0.00006140095],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0042091943,0.00015240535,0.0015015173,0.00009635351,0.000030279274,0.0000014586333,0.00013479599,0.0002162541,0.0001000446],"category_scores_gemma":[0.017907254,0.0000988299,0.00033503084,0.00056234084,0.00039782948,0.000053043612,0.00013771756,0.00028532572,0.000003418443],"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":[0.0010071464,0.000059128648,0.9935897,0.00010336681,0.0001820395,0.000007501693,0.00039000047,0.0015898953,0.00014239449,0.00021632193,0.00001679799,0.0026957232],"study_design_scores_gemma":[0.0045952285,0.0023820398,0.9898684,0.00017045895,0.0003462106,0.000007471837,0.0008321272,0.0014582808,0.00005633901,0.0001540228,0.000045838184,0.00008360192],"about_ca_topic_score_codex":0.00014292538,"about_ca_topic_score_gemma":0.000273857,"teacher_disagreement_score":0.013698059,"about_ca_system_score_codex":0.000029712448,"about_ca_system_score_gemma":0.0002092822,"threshold_uncertainty_score":0.9903653},"labels":[],"label_agreement":null},{"id":"W3175965832","doi":"10.2147/clep.s308852","title":"Case-Ascertainment Models to Identify Adults with Obstructive Sleep Apnea Using Health Administrative Data: Internal and External Validation","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Obstructive Sleep Apnea Research","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Health Sciences Centre; University of Toronto; Sunnybrook Health Science Centre; London Health Sciences Centre; University of Calgary; Western University; Ottawa Hospital; University of Ottawa","funders":"Ottawa Hospital Research Institute; London Health Sciences Centre","keywords":"Medicine; Logistic regression; Cart; Apnea–hypopnea index; Obstructive sleep apnea; Nomogram; Polysomnography; Statistics; Internal medicine; Apnea; Mathematics","score_opus":0.4162101704472162,"score_gpt":0.5411513899434164,"score_spread":0.12494121949620018,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3175965832","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.78453076,0.00038840278,0.21029171,0.003544795,0.00034642685,0.00066698535,0.000111110545,0.00002558768,0.00009423887],"genre_scores_gemma":[0.9140384,0.00007671585,0.08279318,0.0024130556,0.00043173958,0.00002526055,0.00014932755,0.00003407677,0.00003821338],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99309057,0.002735348,0.0016932444,0.0014817612,0.0003156279,0.0006834364],"domain_scores_gemma":[0.99239755,0.004596142,0.0005463227,0.000996447,0.00047749147,0.0009860481],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0043818294,0.0002885525,0.0013967208,0.00011505317,0.00016499597,0.000026489879,0.0002409528,0.00023890879,0.00009845524],"category_scores_gemma":[0.007488659,0.00023529641,0.00011854,0.00024295584,0.0005891982,0.00027761236,0.0007051972,0.00093764585,0.000019226585],"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":[0.010974685,0.001418597,0.6358202,0.0006141825,0.0021953494,0.017084096,0.0027648858,0.0016373993,0.0007841405,0.0061641964,0.00035979354,0.3201825],"study_design_scores_gemma":[0.013401358,0.008975232,0.39735416,0.0011119123,0.0006085361,0.059063826,0.010337348,0.4967281,0.0010203343,0.010051911,0.0003425306,0.0010047429],"about_ca_topic_score_codex":0.0005084946,"about_ca_topic_score_gemma":0.00020335756,"teacher_disagreement_score":0.4950907,"about_ca_system_score_codex":0.00027438716,"about_ca_system_score_gemma":0.0003351381,"threshold_uncertainty_score":0.9595109},"labels":[],"label_agreement":null},{"id":"W3177050212","doi":"10.2147/clep.s314802","title":"Burden of Respiratory Infection and Tuberculosis Among US States from 1990 to 2019","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Tuberculosis Research and Epidemiology","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"National Natural Science Foundation of China; Institute for Health Metrics and Evaluation; University of Washington; Bill and Melinda Gates Foundation","keywords":"Medicine; Incidence (geometry); Demography; Tuberculosis; Confidence interval; Burden of disease; Cause of death; Mortality rate; Disease burden; Disease; Internal medicine; Pathology","score_opus":0.11564902399434294,"score_gpt":0.45315037984692963,"score_spread":0.33750135585258667,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3177050212","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97633326,0.002705887,0.00038495756,0.019347416,0.00039802268,0.00028055304,0.000060211394,0.000045258606,0.00044445245],"genre_scores_gemma":[0.96032155,0.007953145,0.0030883118,0.026946783,0.0011101922,0.00004333439,0.0002315583,0.000034716035,0.0002704072],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99239665,0.0039953813,0.0020464172,0.00083552714,0.00012814476,0.00059790537],"domain_scores_gemma":[0.97775954,0.020147964,0.0003075402,0.00066920713,0.00029720124,0.0008185233],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.006292859,0.0002094774,0.0020143234,0.0001409556,0.000056443234,0.0000037215484,0.000102305756,0.00063306186,0.00038120078],"category_scores_gemma":[0.08552658,0.00017221674,0.00034935772,0.0002594774,0.00070115755,0.000059505906,0.0002883761,0.00079059094,0.00015439974],"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":[0.0003000336,0.00012092772,0.9429559,0.000045546254,0.00030153335,0.000040076848,0.000042016298,0.0001444364,0.0010155266,0.00020740763,0.04227935,0.012547247],"study_design_scores_gemma":[0.0009635911,0.0011113001,0.945047,0.00010073065,0.00011748348,0.000016305154,0.00003214689,0.001014475,0.00026662563,0.0026845403,0.04850942,0.00013634002],"about_ca_topic_score_codex":0.010434882,"about_ca_topic_score_gemma":0.00065575395,"teacher_disagreement_score":0.07923371,"about_ca_system_score_codex":0.000048734062,"about_ca_system_score_gemma":0.00016994824,"threshold_uncertainty_score":0.9961547},"labels":[],"label_agreement":null},{"id":"W3207389912","doi":"10.2147/clep.s333534","title":"Measuring Medication Adherence in a Population-Based Asthma Administrative Pharmacy Database: A Systematic Review and Meta-Analysis","year":2021,"lang":"en","type":"review","venue":"Clinical Epidemiology","topic":"Medication Adherence and Compliance","field":"Medicine","cited_by":50,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"Memorial University of Newfoundland","keywords":"Medicine; Pharmacy; Meta-analysis; Asthma; Systematic review; Population; Alternative medicine; Family medicine; MEDLINE; Internal medicine; Environmental health; Pathology","score_opus":0.7932195127474996,"score_gpt":0.6135945149813509,"score_spread":0.1796249977661487,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3207389912","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":[9.261163e-7,0.99163175,0.0016367557,0.0033453743,0.00008194521,0.0031051561,0.000081695696,0.00003319353,0.00008320608],"genre_scores_gemma":[0.0000575882,0.9829912,0.002056435,0.009846354,0.0000680487,0.002386091,0.0021202157,0.00002428618,0.000449761],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.98043096,0.010730156,0.0067982418,0.0013365066,0.00033540468,0.00036872947],"domain_scores_gemma":[0.97250867,0.02113416,0.0039851544,0.0015755388,0.00025901909,0.00053748675],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.015537442,0.0005359095,0.021987531,0.0002953305,0.000049361857,0.000008090921,0.00037610132,0.00041145633,0.0026474544],"category_scores_gemma":[0.06904065,0.0003566658,0.0028742072,0.0013064095,0.00022583199,0.0000613976,0.00009129441,0.001297445,0.00022469353],"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":[0.000008189649,0.00024528775,0.0014330522,0.9395978,0.04789311,0.00005941774,0.00000382069,5.332179e-7,7.0519017e-9,0.00035686957,0.0012775295,0.009124416],"study_design_scores_gemma":[0.00043008509,0.000108313274,0.00046510316,0.147973,0.80968964,0.00007666068,0.0000062934637,0.00030809976,1.9669141e-8,0.000048814432,0.040540725,0.0003532447],"about_ca_topic_score_codex":0.00012271099,"about_ca_topic_score_gemma":0.000055633594,"teacher_disagreement_score":0.7916248,"about_ca_system_score_codex":0.000097350814,"about_ca_system_score_gemma":0.0014332602,"threshold_uncertainty_score":0.99988854},"labels":[],"label_agreement":null},{"id":"W3208046866","doi":"10.2147/clep.s335621","title":"Positive Predictive Value of COVID-19 ICD-10 Diagnosis Codes Across Calendar Time and Clinical Setting","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":56,"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; Quarter (Canadian coin); Medical record; Coronavirus disease 2019 (COVID-19); Diagnosis code; Veterans Affairs; Emergency department; Predictive value; False positive paradox; Outpatient visits; Emergency medicine; Pediatrics; Internal medicine; Health care; Statistics; Infectious disease (medical specialty); Disease; Population","score_opus":0.41339194172535493,"score_gpt":0.6260680528223027,"score_spread":0.21267611109694778,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3208046866","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85002583,0.0017516144,0.017967993,0.11949131,0.0029365302,0.001450032,0.0033436108,0.0002525615,0.0027805208],"genre_scores_gemma":[0.6113155,0.016258603,0.01968578,0.3421458,0.0049845953,0.00043819842,0.0014352315,0.00009234242,0.0036439488],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9727503,0.019314848,0.0060862713,0.00069325813,0.00020888272,0.00094643835],"domain_scores_gemma":[0.7727588,0.22264288,0.0021102172,0.0004776466,0.0005425051,0.0014679884],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.046374064,0.00022137216,0.0020496768,0.000046928635,0.00082466693,0.0000022573804,0.000213554,0.0015239809,0.0012792593],"category_scores_gemma":[0.4030988,0.00018376671,0.00031800434,0.00018765978,0.0014173593,0.000101425256,0.00048168303,0.002443632,0.00038610582],"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":[0.00044021555,0.00019224905,0.88591033,0.00058721035,0.00013172784,0.000017284005,0.0015074929,0.00003752343,0.0000010082114,0.002529554,0.09542345,0.013221918],"study_design_scores_gemma":[0.0033048238,0.0013763787,0.8769255,0.00072672404,0.00014518907,0.000013933679,0.0016780843,0.018920489,0.000004696358,0.0071274294,0.08950995,0.00026678212],"about_ca_topic_score_codex":0.00029690194,"about_ca_topic_score_gemma":0.000092367976,"teacher_disagreement_score":0.35672474,"about_ca_system_score_codex":0.00014490531,"about_ca_system_score_gemma":0.0020163278,"threshold_uncertainty_score":0.9998578},"labels":[],"label_agreement":null},{"id":"W3216268220","doi":"10.2147/clep.s336517","title":"Incidence of Inflammatory Bowel Disease in South Asian and Chinese People: A Population-Based Cohort Study from Ontario, Canada","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","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":"Ottawa Hospital; University of Ottawa; Health Sciences Centre; Sunnybrook Health Science Centre; SickKids Foundation; University of Toronto","funders":"Hospital for Sick Children","keywords":"Medicine; Incidence (geometry); Inflammatory bowel disease; Population; Internal medicine; Ulcerative colitis; Epidemiology; Rate ratio; Demography; Cohort; Cohort study; Disease; Confidence interval; Environmental health","score_opus":0.016532511636355867,"score_gpt":0.3160693385186174,"score_spread":0.2995368268822615,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3216268220","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9986827,0.00021969406,0.00006436569,0.00023350205,0.00027173082,0.0004006856,0.00009058011,0.0000067592373,0.000029979237],"genre_scores_gemma":[0.9978623,0.000011470726,0.00020875626,0.0011892373,0.000129042,0.00005461357,0.00048324186,0.00001776918,0.000043577198],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99567217,0.0017204944,0.001419673,0.00076244626,0.00013455904,0.00029064954],"domain_scores_gemma":[0.9974494,0.0009074309,0.00040980842,0.0007287149,0.00011853867,0.00038610643],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0016263812,0.00021952251,0.00069026544,0.000033581597,0.000045902685,0.0000042564816,0.00019582514,0.00022038842,0.000101643556],"category_scores_gemma":[0.015250703,0.00021013361,0.00013569884,0.00007999931,0.0001297498,0.0000051236434,0.00019440599,0.00027592207,0.0000017033399],"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":[0.0004318795,0.00017737922,0.99778515,0.000026130983,0.00006588797,0.00028510657,0.000070471004,0.0009582281,0.000027069545,0.000013911834,0.00010336291,0.00005543064],"study_design_scores_gemma":[0.0011238153,0.000061026276,0.99750286,0.000022289041,0.000048041387,7.0564863e-7,0.00006583057,0.000652205,0.0000029353278,0.00027906054,0.00004304418,0.00019816296],"about_ca_topic_score_codex":0.50004447,"about_ca_topic_score_gemma":0.98761445,"teacher_disagreement_score":0.48756996,"about_ca_system_score_codex":0.00007906457,"about_ca_system_score_gemma":0.0028405485,"threshold_uncertainty_score":0.99304426},"labels":[],"label_agreement":null},{"id":"W4200340011","doi":"10.2147/clep.s344567","title":"Perinatal Mental Illness and Risk of Incident Autoimmune Disease: A Population-Based Propensity-Score Matched Cohort Study","year":2021,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Maternal Mental Health During Pregnancy and Postpartum","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"St. Michael's Hospital; The Scarborough Hospital; Public Health Ontario; Women's College Hospital; University of Toronto","funders":"Canadian Institutes of Health Research; Canada Research Chairs","keywords":"Medicine; Mental illness; Hazard ratio; Disease; Cohort; Population; Cohort study; Propensity score matching; Proportional hazards model; Autoimmune disease; Mental health; Psychiatry; Pediatrics; Internal medicine; Confidence interval","score_opus":0.10293718375608633,"score_gpt":0.4183105767822018,"score_spread":0.31537339302611545,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200340011","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99544376,0.0014376402,0.000040202962,0.0011910078,0.000670486,0.0011092396,0.000050332772,0.0000346869,0.000022650576],"genre_scores_gemma":[0.99768883,0.00054910826,0.00030313362,0.0009129921,0.00008357226,0.000071425944,0.00014827137,0.000015931675,0.00022671605],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99553865,0.002004689,0.0015251655,0.0005053127,0.00015575901,0.0002703991],"domain_scores_gemma":[0.9965184,0.0020676272,0.00052095,0.00046402588,0.0000940433,0.00033497025],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003292913,0.00016571097,0.001311533,0.000037474714,0.00012682956,0.000003236448,0.000077313896,0.0001367635,0.00018593078],"category_scores_gemma":[0.0059351968,0.00013164112,0.0001646817,0.00007667082,0.00019625606,0.000033571345,0.0001633953,0.00040266427,0.000010644364],"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":[0.0014080354,0.0008173996,0.9952087,0.0009553278,0.00013705864,0.00018178076,0.000067002715,0.000010153456,0.0000048869983,0.00009191931,0.00001655072,0.001101215],"study_design_scores_gemma":[0.0021054544,0.00069267326,0.9950716,0.00080226344,0.00025657448,0.000039616152,0.000045806628,0.0005203743,0.000039725546,0.00022157349,0.000083349034,0.00012101002],"about_ca_topic_score_codex":0.0007689041,"about_ca_topic_score_gemma":0.000080401835,"teacher_disagreement_score":0.0026422837,"about_ca_system_score_codex":0.00004938375,"about_ca_system_score_gemma":0.00016458308,"threshold_uncertainty_score":0.71054137},"labels":[],"label_agreement":null},{"id":"W4220681952","doi":"10.2147/clep.s352889","title":"Interpregnancy Weight Change Among Mothers of a Child with a Major Congenital Anomaly: A Danish Nationwide Cohort Study","year":2022,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Gestational Diabetes Research and Management","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Institute for Clinical Evaluative Sciences; University of Toronto; SickKids Foundation; The Scarborough Hospital; Hospital for Sick Children","funders":"Canadian Institutes of Health Research","keywords":"Danish; Cohort study; Weight change; Weight gain; Cohort; Climb; Birth weight","score_opus":0.08061341755064835,"score_gpt":0.40049130407407263,"score_spread":0.3198778865234243,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220681952","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.988172,0.0003287953,0.000535063,0.004686762,0.00016671751,0.0025256455,0.000041135943,0.000036124664,0.003507763],"genre_scores_gemma":[0.9935267,0.00006462927,0.000505519,0.0036392012,0.000114485396,0.0012446395,0.00009141612,0.000022239756,0.0007911927],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.996414,0.0014437101,0.0009836918,0.00049184513,0.00030723892,0.0003595171],"domain_scores_gemma":[0.99646246,0.0024458189,0.00035785206,0.0003773031,0.00013687699,0.0002196782],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0038829113,0.00014155138,0.00088857365,0.00016028046,0.00009932685,0.0000026371893,0.00020642165,0.000059402857,0.00069458265],"category_scores_gemma":[0.0045088585,0.00010968585,0.00017189673,0.00024258062,0.000386517,0.000056897177,0.0002856676,0.00050857157,0.0000145222475],"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":[0.000522334,0.0013169153,0.9928199,0.000045950437,0.00083954725,0.0000940952,0.0001575821,0.000013172891,0.0000010663783,0.0008029577,0.0014050832,0.001981414],"study_design_scores_gemma":[0.002658042,0.0055768574,0.98763764,0.000049164963,0.00019297944,0.000015096556,0.00051416596,0.0008810055,0.0000010559615,0.00030272498,0.0020747837,0.000096478834],"about_ca_topic_score_codex":0.0003447565,"about_ca_topic_score_gemma":0.00020709114,"teacher_disagreement_score":0.0053546843,"about_ca_system_score_codex":0.000057462494,"about_ca_system_score_gemma":0.00007627092,"threshold_uncertainty_score":0.7605193},"labels":[],"label_agreement":null},{"id":"W4286836086","doi":"10.2147/clep.s376283","title":"Identifying Pediatric Diabetes Cases from Health Administrative Data: A Population-Based Validation Study in Quebec, Canada [Corrigendum]","year":2022,"lang":"en","type":"erratum","venue":"Clinical Epidemiology","topic":"Diabetes Management and Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Medicine; Diabetes mellitus; Population; Demography; Pediatrics; Environmental health; Endocrinology","score_opus":0.4781898337116004,"score_gpt":0.5299872195237564,"score_spread":0.05179738581215598,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4286836086","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8220586,0.015257276,0.00010304414,0.03796296,0.10956384,0.009652062,0.0038957621,0.00024731096,0.0012591139],"genre_scores_gemma":[0.57162595,0.001889093,0.0007748586,0.01841083,0.014163893,0.0008265822,0.26506624,0.00026149827,0.12698103],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9829509,0.009307632,0.004015005,0.0019403414,0.00073997304,0.0010461468],"domain_scores_gemma":[0.9716808,0.02373802,0.0020096533,0.0019300954,0.00011965977,0.00052175176],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012645344,0.00048145995,0.0032664961,0.0004975757,0.00023941723,0.000032930122,0.000938849,0.00053378294,0.0013786084],"category_scores_gemma":[0.046522718,0.00043497296,0.00024334824,0.00068560074,0.00011656026,0.00012661343,0.00082738453,0.0035388975,0.000019632913],"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":[0.00005766457,0.00050031755,0.53655714,0.00021916954,0.00014728535,0.00016869164,0.000014090123,0.000017709981,8.533652e-9,0.0000055987343,0.4605823,0.0017300355],"study_design_scores_gemma":[0.0014478769,0.0010172854,0.93354934,0.00013974035,0.00035160032,2.1724476e-7,0.00036166646,0.0017884545,3.168224e-8,0.00016600995,0.060857482,0.00032028471],"about_ca_topic_score_codex":0.9146744,"about_ca_topic_score_gemma":0.9777153,"teacher_disagreement_score":0.3997248,"about_ca_system_score_codex":0.0011637465,"about_ca_system_score_gemma":0.00962642,"threshold_uncertainty_score":0.9998102},"labels":[],"label_agreement":null},{"id":"W4307558400","doi":"10.2147/clep.s370623","title":"Association Between Medication Adherence and Risk of COPD in Adult Asthma Patients: A Retrospective Cohort Study in Canada","year":2022,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Asthma and respiratory diseases","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Memorial University of Newfoundland","funders":"Memorial University of Newfoundland","keywords":"Medicine; Asthma; COPD; Hazard ratio; Internal medicine; Confidence interval; Cohort; Proportional hazards model; Propensity score matching; Cohort study; Population; Retrospective cohort study; Relative risk; Incidence (geometry); Pediatrics; Environmental health","score_opus":0.038154834504820115,"score_gpt":0.3713546514690621,"score_spread":0.333199816964242,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4307558400","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9983376,0.0001036562,0.000007637041,0.00050825166,0.00019059652,0.0006983471,0.00009247078,0.0000055203714,0.000055874618],"genre_scores_gemma":[0.9988737,0.00014253764,0.000026147107,0.00072077906,0.00005735329,0.000098593664,0.000054996643,0.0000060738166,0.00001978313],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9941382,0.0036037576,0.0014345188,0.0003699234,0.00025521108,0.00019840396],"domain_scores_gemma":[0.99269426,0.005885159,0.00094115385,0.0001947204,0.00014508406,0.00013960063],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005430967,0.00008769396,0.0010053029,0.000067713176,0.00004013876,4.594195e-7,0.0000775661,0.000097259144,0.00008407761],"category_scores_gemma":[0.04520678,0.000076349665,0.00005295319,0.00020060684,0.00008160955,0.000020560105,0.000091121125,0.00063619367,0.0000011146171],"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":[0.00013733136,0.00041576885,0.9932744,0.000009775567,0.00008793096,0.000006523649,0.000055508102,0.000012593044,8.742355e-8,0.000009972866,0.0010720479,0.004918064],"study_design_scores_gemma":[0.002522664,0.0016003477,0.9947404,0.000019090685,0.00008802387,2.3319353e-7,0.00043862892,0.00019474412,1.6517012e-7,0.00015568857,0.00018059683,0.00005940551],"about_ca_topic_score_codex":0.4540422,"about_ca_topic_score_gemma":0.50458515,"teacher_disagreement_score":0.050542958,"about_ca_system_score_codex":0.0009032765,"about_ca_system_score_gemma":0.0007073157,"threshold_uncertainty_score":0.96283585},"labels":[],"label_agreement":null},{"id":"W4308945921","doi":"10.2147/clep.s368303","title":"Implementing Multiple Imputation for Missing Data in Longitudinal Studies When Models are Not Feasible: An Example Using the Random Hot Deck Approach","year":2022,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","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; Jewish General Hospital","funders":"","keywords":"Imputation (statistics); Missing data; Longitudinal data; Statistics; Data mining; Medicine; Computer science; Mathematics","score_opus":0.9359089854971832,"score_gpt":0.6142285258336672,"score_spread":0.321680459663516,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4308945921","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.51046467,0.0011192268,0.48258168,0.0038399035,0.0006489968,0.0012132821,0.00009444371,0.000022281047,0.000015499234],"genre_scores_gemma":[0.90937823,0.00008326398,0.08752143,0.0023685715,0.0003106128,0.00018188305,0.0001236368,0.000016123693,0.00001628224],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9772592,0.013970486,0.0047248667,0.0020338432,0.00097223296,0.0010393258],"domain_scores_gemma":[0.8540591,0.13946709,0.0027550552,0.0029107104,0.00057817006,0.00022988209],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21688648,0.00023418384,0.0014194988,0.000142703,0.0013398405,0.000076283235,0.002126575,0.00011513037,0.00005805094],"category_scores_gemma":[0.12063811,0.00015011027,0.00026646294,0.0003987786,0.00038846093,0.0005840997,0.0020792861,0.0005880645,0.0000018342624],"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":[0.0011500448,0.00055519555,0.7805195,0.00003978351,0.000109172564,0.0000023455307,0.0012052217,0.17665078,0.000038582846,0.00074761186,0.005130471,0.03385128],"study_design_scores_gemma":[0.0023138472,0.00017912414,0.03763835,0.000014669454,0.000052441184,0.000004976898,0.007308568,0.78091705,0.0000018656712,0.16772242,0.0036822457,0.00016447563],"about_ca_topic_score_codex":0.0005360982,"about_ca_topic_score_gemma":0.0008213921,"teacher_disagreement_score":0.7428812,"about_ca_system_score_codex":0.00015008943,"about_ca_system_score_gemma":0.00016462374,"threshold_uncertainty_score":0.9999603},"labels":[],"label_agreement":null},{"id":"W4311313947","doi":"10.2147/clep.s389853","title":"Global Patterns and Trends in Total Burden of Hepatitis B from 1990 to 2019 and Predictions to 2030","year":2022,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Hepatitis B Virus Studies","field":"Medicine","cited_by":31,"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":"Zhengzhou University; National Natural Science Foundation of China","keywords":"Medicine; Hepatitis B; Demography; Incidence (geometry); Public health; Autoregressive integrated moving average; Burden of disease; Disease burden; Hepatitis C; Hepatitis; Mortality rate; Environmental health; Internal medicine; Population; Time series; Statistics; Pathology","score_opus":0.09861257294767324,"score_gpt":0.4317328504923207,"score_spread":0.33312027754464746,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4311313947","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97627294,0.00053188024,0.00037349985,0.020305263,0.00047675183,0.00023266641,0.0014837614,0.00002334732,0.0002998673],"genre_scores_gemma":[0.99260706,0.00059029716,0.0013735428,0.0047460836,0.00027953804,0.00007790413,0.00008419441,0.000009526431,0.0002318512],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9971803,0.0007814127,0.0011430093,0.00050108036,0.00010064411,0.00029359903],"domain_scores_gemma":[0.9963431,0.0029173046,0.00013325228,0.0002636926,0.00003261792,0.00031008365],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014350545,0.00012899419,0.0011210707,0.00011901264,0.00006267656,0.0000014733326,0.00007670093,0.00012025789,0.00037156767],"category_scores_gemma":[0.0063067963,0.000121954625,0.00009109962,0.00024589285,0.0001634391,0.000019876199,0.0005685187,0.0002867666,0.000011638864],"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":[0.00015486484,0.000081606326,0.94627124,0.000007206411,0.00007095101,0.000023103605,0.00022425987,0.0001605526,0.000005263204,0.00014145057,0.024471693,0.028387833],"study_design_scores_gemma":[0.0008769646,0.001767774,0.9882833,0.00003623066,0.000054221706,0.000022181935,0.000067304805,0.00086801447,3.840115e-7,0.0004862914,0.0074476483,0.00008970717],"about_ca_topic_score_codex":0.087083444,"about_ca_topic_score_gemma":0.01613504,"teacher_disagreement_score":0.07094841,"about_ca_system_score_codex":0.00008262317,"about_ca_system_score_gemma":0.000031625084,"threshold_uncertainty_score":0.91899574},"labels":[],"label_agreement":null},{"id":"W4391732442","doi":"10.2147/clep.s448980","title":"Risk of Arterial and Venous Thrombotic Events Among Patients with COVID-19: A Multi-National Collaboration of Regulatory Agencies from Canada, Europe, and United States","year":2024,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Heparin-Induced Thrombocytopenia and Thrombosis","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; Health Canada; University of Manitoba; Canadians Living with HIV; Manitoba Health; Ontario Stroke Network; University of British Columbia","funders":"NIHR Oxford Biomedical Research Centre; Canadian Institutes of Health Research; Health Canada; Erasmus Medisch Centrum; National Institute for Health and Care Research","keywords":"Medicine; Ambulatory; Absolute risk reduction; Coronavirus disease 2019 (COVID-19); Venous thromboembolism; Emergency medicine; Confidence interval; Cohort; Cohort study; Pediatrics; Internal medicine; Disease; Thrombosis","score_opus":0.0907891444073767,"score_gpt":0.38647026012966595,"score_spread":0.2956811157222893,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391732442","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9974675,0.00020765488,0.00022397905,0.0007945807,0.0003209037,0.0003552425,0.0005913984,0.000019307166,0.000019396155],"genre_scores_gemma":[0.99701226,0.00042633762,0.0011030717,0.00088199147,0.000067565925,0.0000075729217,0.00046682204,0.00001740244,0.000016954678],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99715346,0.001039627,0.001056254,0.00039941916,0.0001732129,0.00017804287],"domain_scores_gemma":[0.9931447,0.0055772993,0.00041764343,0.00018731048,0.00036903057,0.0003040471],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0014567991,0.00015190063,0.0010087083,0.00009422643,0.00005814097,0.000003936055,0.000060182952,0.00017656658,0.000034377972],"category_scores_gemma":[0.014177315,0.00011402933,0.000041663512,0.00027448678,0.0006432515,0.000062332685,0.000059997106,0.00023697122,0.0000011785054],"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":[0.00028321097,0.00020504631,0.9969011,0.00013176164,0.0004991955,0.000009175532,0.00032854904,0.00010602196,0.00011293454,0.00027091123,0.0009313474,0.00022076824],"study_design_scores_gemma":[0.0020537984,0.00076933106,0.9917641,0.00012691742,0.00031487684,0.000002223612,0.00011577557,0.0033590414,0.000026594225,0.00081047346,0.0005571251,0.00009973015],"about_ca_topic_score_codex":0.08764582,"about_ca_topic_score_gemma":0.04043262,"teacher_disagreement_score":0.0472132,"about_ca_system_score_codex":0.000102231796,"about_ca_system_score_gemma":0.0015185009,"threshold_uncertainty_score":0.9941267},"labels":[],"label_agreement":null},{"id":"W4391778130","doi":"10.2147/clep.s449183","title":"Variation in the Care of Children with Inflammatory Bowel Disease Within and Across Canadian Provinces: A Multi-Province Population-Based Cohort Study","year":2024,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Saskatchewan; BC Children's Hospital; University of British Columbia; University of Ottawa; Institute for Clinical Evaluative Sciences; McGill University Health Centre; CancerCare Manitoba; Ottawa Hospital; University of Manitoba; Mount Sinai Hospital; University of Calgary; University of Alberta; Agricultural Research Institute of Ontario; Dalhousie University; University of Toronto; SickKids Foundation","funders":"Canadian Institutes of Health Research; Crohn's and Colitis Canada; Pfizer; Hospital for Sick Children; Mitacs; University of Toronto; Canadian Association of Gastroenterology; Canadian Child Health Clinician Scientist Program; University of Ottawa; Dairy Farmers of Ontario; AbbVie Canada; Amgen; Bristol-Myers Squibb","keywords":"Medicine; Inflammatory bowel disease; Incidence (geometry); Emergency department; Pediatrics; Population; Cohort; Disease; Health care; Family medicine; Emergency medicine; Environmental health; Internal medicine","score_opus":0.023704690084511936,"score_gpt":0.3480058315146377,"score_spread":0.3243011414301258,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391778130","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9972542,0.00048226817,0.00028857242,0.00021557469,0.000120382,0.0014727731,0.00014133319,0.00001495285,0.000009935926],"genre_scores_gemma":[0.9984373,0.000013398087,0.00023788046,0.0007280903,0.00012349772,0.00014847105,0.0002796202,0.000021091833,0.000010639991],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9965809,0.001515549,0.00087525975,0.0006336651,0.00009818697,0.00029645188],"domain_scores_gemma":[0.99850386,0.00045523612,0.00021197897,0.000518939,0.000073126714,0.00023686505],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003434043,0.00018370542,0.00031567342,0.000053113803,0.00008968276,0.000016478101,0.00023905751,0.00020450796,0.0000030281474],"category_scores_gemma":[0.0030290748,0.00012639287,0.00008835379,0.00007179886,0.00027353675,0.0000090252715,0.00005170637,0.0002527073,0.000002410336],"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":[0.00023759947,0.00008992211,0.99613094,0.000078625046,0.00007351091,0.00006985217,0.0003170523,0.002623143,0.0000049922346,0.00013228835,0.000020666315,0.00022141184],"study_design_scores_gemma":[0.0007560046,0.00023341681,0.9939177,0.00004142054,0.00005589817,0.0000021153298,0.00023053626,0.004541417,8.698549e-7,0.000040410894,0.000028043949,0.0001521345],"about_ca_topic_score_codex":0.011504336,"about_ca_topic_score_gemma":0.26585215,"teacher_disagreement_score":0.2543478,"about_ca_system_score_codex":0.000053223794,"about_ca_system_score_gemma":0.001218982,"threshold_uncertainty_score":0.99507815},"labels":[],"label_agreement":null},{"id":"W4392059515","doi":"10.2147/clep.s438096","title":"Diagnostic Validity of Chronic Kidney Disease in Health Claims Data Over Time: Results from a Cohort of Community-Dwelling Older Adults in Germany","year":2024,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Chronic Kidney Disease and Diabetes","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":"McGill University; Jewish General Hospital","funders":"Gemeinsame Bundesausschuss","keywords":"Medicine; Kidney disease; Renal function; Confidence interval; Cohort; Internal medicine; Prospective cohort study; Cohort study; Diagnosis code; External validity; Population; Demography; Gerontology; Environmental health; Statistics","score_opus":0.13722101908150472,"score_gpt":0.44904295984754244,"score_spread":0.31182194076603775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392059515","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9770658,0.010085735,0.00007865282,0.007904412,0.00044903718,0.00082253554,0.0034741198,0.000049265447,0.00007043505],"genre_scores_gemma":[0.9834571,0.004823443,0.00014767537,0.0050814147,0.00041108366,0.000025005334,0.005985354,0.00003103397,0.00003788095],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9893424,0.0050652595,0.004073344,0.00075762515,0.00018584865,0.00057553174],"domain_scores_gemma":[0.90230304,0.09314863,0.00071845413,0.0025346358,0.000066581146,0.0012286335],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0138468435,0.00024902396,0.0021392377,0.00018551781,0.000032573127,0.0000033039373,0.00050410704,0.00033203553,0.0003612135],"category_scores_gemma":[0.12367089,0.00021515807,0.00028617796,0.00033290472,0.00059250335,0.00010326951,0.00048357502,0.0014460874,0.000075117016],"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":[0.0011072404,0.0019897935,0.94747794,0.0022712124,0.00014825932,0.00008054649,0.00019234126,0.00010232613,0.0000068986174,0.0000985981,0.042639278,0.0038855525],"study_design_scores_gemma":[0.0051831147,0.00041289345,0.9107855,0.007672952,0.00018073672,9.0240303e-7,0.000017288323,0.071077086,0.0000034711265,0.0022992543,0.0022187931,0.00014803553],"about_ca_topic_score_codex":0.0051429854,"about_ca_topic_score_gemma":0.00026095597,"teacher_disagreement_score":0.109824054,"about_ca_system_score_codex":0.00021770597,"about_ca_system_score_gemma":0.0027581528,"threshold_uncertainty_score":0.8837108},"labels":[],"label_agreement":null},{"id":"W4398216645","doi":"10.2147/clep.s456004","title":"10-Year Multimorbidity Trajectories in Older People Have Limited Benefit in Predicting Short-Term Health Outcomes in Comparison to Standard Multimorbidity Thresholds: A Population-Based Study","year":2024,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Chronic Disease Management Strategies","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":"Université Laval; Institut National de Santé Publique du Québec; McGill University Health Centre; Centre hospitalier universitaire de Québec","funders":"","keywords":"Multimorbidity; Medicine; Term (time); Population; Gerontology; Older people; Demography; Pediatrics; Environmental health","score_opus":0.22811260200563246,"score_gpt":0.5065897666034352,"score_spread":0.2784771645978028,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398216645","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9902703,0.00034905135,0.00026140647,0.004946165,0.0006840088,0.0030820556,0.00009097674,0.00020937409,0.000106654334],"genre_scores_gemma":[0.9973448,0.000034341785,0.000802588,0.00087456906,0.000195019,0.00026236085,0.0003519677,0.00004595852,0.00008841217],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9926354,0.00132176,0.0035949342,0.0012638358,0.00032913775,0.00085487706],"domain_scores_gemma":[0.9903703,0.008247672,0.00025138285,0.0006898642,0.00006521331,0.00037560664],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.00794372,0.0004049154,0.002853668,0.00062894815,0.000059460912,0.000030373178,0.00024271253,0.0003101447,0.0004837616],"category_scores_gemma":[0.010784637,0.00035815776,0.00028921833,0.0006631106,0.00010222168,0.00012592578,0.00018153843,0.0011068293,0.00003615266],"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":[0.000938845,0.0022176604,0.98404646,0.000391839,0.000088295405,0.000101321886,0.0007881585,0.0085279755,2.4018297e-7,0.00013405632,0.00056251715,0.0022026203],"study_design_scores_gemma":[0.0045271004,0.0015433497,0.93637073,0.00061022164,0.00006358947,6.6020647e-7,0.0012086277,0.055163175,2.7445404e-7,0.00015992034,0.00010843395,0.00024391533],"about_ca_topic_score_codex":0.0068170815,"about_ca_topic_score_gemma":0.09830916,"teacher_disagreement_score":0.09149207,"about_ca_system_score_codex":0.0009665489,"about_ca_system_score_gemma":0.00043050677,"threshold_uncertainty_score":0.99988705},"labels":[],"label_agreement":null},{"id":"W4409413640","doi":"10.2147/clep.s504259","title":"Extracting Cognitive Impairment Assessment Information From Unstructured Notes in Electronic Health Records Using Natural Language Processing Tools: Validation with Clinical Assessment Data","year":2025,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Machine Learning in Healthcare","field":"Computer Science","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":"U.S. National Library of Medicine; National Institutes of Health","keywords":"Health records; Medicine; Cognitive impairment; Cognition; Unstructured data; Natural language processing; Electronic health record; Data science; Information retrieval; Data mining; Computer science; Artificial intelligence; Psychiatry; Big data; Health care","score_opus":0.19230961153472326,"score_gpt":0.5624915034290198,"score_spread":0.3701818918942965,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409413640","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.30674386,0.0006093566,0.6813905,0.009477027,0.0009249192,0.0006610686,0.000023198725,0.00012263446,0.000047395602],"genre_scores_gemma":[0.7323129,0.00011669729,0.26035613,0.006329889,0.00016223094,0.000021173208,0.00069015886,0.0000090309195,0.0000017616925],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9857318,0.008201541,0.0037538141,0.0011702168,0.00027698438,0.00086568523],"domain_scores_gemma":[0.97052145,0.02577445,0.0022786744,0.0010321933,0.00021804373,0.00017519313],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.02162992,0.00028970395,0.0012203364,0.00020914167,0.00024777505,0.00013015853,0.0010822525,0.0003435559,0.000011048048],"category_scores_gemma":[0.019403975,0.00023877887,0.00010471184,0.0005845072,0.00015887772,0.0016515463,0.0007576743,0.0028808536,0.000003417604],"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":[0.00006213968,0.00007483072,0.4855453,0.000057154586,0.000035790246,0.000003156452,0.00013941828,0.00060207734,8.1065394e-7,0.0006539293,0.000032558415,0.5127928],"study_design_scores_gemma":[0.00092351646,0.00031580252,0.4707378,0.00035653185,0.000015810365,0.0000059523168,0.00018906046,0.5260031,9.74509e-7,0.0011721037,0.00014043495,0.00013888635],"about_ca_topic_score_codex":0.0024704968,"about_ca_topic_score_gemma":0.0007798228,"teacher_disagreement_score":0.52540106,"about_ca_system_score_codex":0.0006255402,"about_ca_system_score_gemma":0.0038117594,"threshold_uncertainty_score":0.99941957},"labels":[],"label_agreement":null},{"id":"W4411309459","doi":"10.2147/clep.s520168","title":"Framework for Exploration of Statistical Heterogeneity in Multi-Database Studies: A Case Study Using EXACOS-CV Studies","year":2025,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia","funders":"AstraZeneca","keywords":"Database; Medicine; Data mining; Statistics; Computational biology; Bioinformatics; Data science; Information retrieval; Computer science; Biology; Mathematics","score_opus":0.651336487342431,"score_gpt":0.6395434522200241,"score_spread":0.011793035122406925,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411309459","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.82274044,0.00677321,0.16643265,0.0009643233,0.0005468395,0.0023718923,0.00014238761,0.000025276478,0.0000029808307],"genre_scores_gemma":[0.9061089,0.00094619946,0.09146971,0.00079985295,0.00017415224,0.000415754,0.000036037458,0.00002345777,0.000025943],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9921158,0.0034673146,0.0027162894,0.0009624196,0.00015945651,0.0005787031],"domain_scores_gemma":[0.9449668,0.05295672,0.0004220293,0.00076855556,0.0006464162,0.000239516],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008758455,0.00027140975,0.0025006495,0.00025501527,0.00011621801,0.0000026345347,0.00014545707,0.00027668013,0.000014509141],"category_scores_gemma":[0.16567056,0.000224886,0.00027326783,0.00040988304,0.001285595,0.000117164476,0.00041989007,0.0007404441,0.0000042015995],"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":[0.002402154,0.0048943013,0.9663319,0.0023355128,0.0018064835,0.0023554978,0.0007973622,0.0002476001,0.00003573991,0.006296403,0.00042126066,0.012075765],"study_design_scores_gemma":[0.02413386,0.0047479677,0.69326586,0.0025924856,0.0022073376,0.00028126608,0.09607253,0.09236088,0.00007050582,0.08331693,0.0002249431,0.00072545564],"about_ca_topic_score_codex":0.00023702729,"about_ca_topic_score_gemma":0.0004678826,"teacher_disagreement_score":0.27306607,"about_ca_system_score_codex":0.0004112972,"about_ca_system_score_gemma":0.00044930837,"threshold_uncertainty_score":0.91705847},"labels":[],"label_agreement":null},{"id":"W4411611048","doi":"10.2147/clep.s526203","title":"The Hydrocephalus Association Patient-Powered Interactive Engagement Registry (HAPPIER): Design and Initial Baseline Report","year":2025,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Cerebrospinal fluid and hydrocephalus","field":"Neuroscience","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":"BC Children's Hospital","funders":"","keywords":"Medicine; Baseline (sea); Hydrocephalus; Association (psychology); Pediatrics; Surgery; Psychology","score_opus":0.12681244958124196,"score_gpt":0.43649321278742553,"score_spread":0.30968076320618354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411611048","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.72473097,0.002788358,0.13909875,0.078842185,0.017239304,0.0033072268,0.00006298846,0.00056980277,0.03336039],"genre_scores_gemma":[0.9789267,0.0013953117,0.00090325164,0.016137678,0.00034828586,0.00007740196,0.000011474085,0.000015968137,0.0021839165],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9884009,0.007654059,0.0022034815,0.0009803198,0.00019032208,0.0005709228],"domain_scores_gemma":[0.90766364,0.09008132,0.0012727558,0.00064597384,0.00014470232,0.00019158234],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.014967496,0.00022853127,0.00073941256,0.00006096315,0.00055091176,0.000038573005,0.00033915736,0.00034209847,0.0000492711],"category_scores_gemma":[0.2844446,0.00016637075,0.00021670976,0.00022232349,0.00048126542,0.00010927371,0.00035864447,0.0010247685,0.00005124152],"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":[0.0062263664,0.0018738044,0.12889604,0.0001204866,0.0006499258,0.0010937853,0.00065861887,0.0007034653,0.0074863397,0.05435247,0.38258,0.4153587],"study_design_scores_gemma":[0.0045904596,0.002637435,0.050626256,0.00024493336,0.0003446857,0.00037210068,0.0003620992,0.074426964,0.0073746503,0.11338675,0.7446352,0.0009984489],"about_ca_topic_score_codex":0.000030245184,"about_ca_topic_score_gemma":0.000008056489,"teacher_disagreement_score":0.41436023,"about_ca_system_score_codex":0.00016533842,"about_ca_system_score_gemma":0.0001971592,"threshold_uncertainty_score":0.72158283},"labels":[],"label_agreement":null},{"id":"W7117410197","doi":"10.2147/clep.s546090","title":"Validity of the Leading Causes of Death Classification for Premature Mortality in Inflammatory Bowel Disease: A Population-Based Comparison with ICD-10 Coding in Ontario, Canada","year":2025,"lang":"en","type":"article","venue":"Clinical Epidemiology","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Trillium Health Centre; McGill University; Sinai Health System; Lunenfeld-Tanenbaum Research Institute; Mount Sinai Hospital; Western University; University of Toronto; SickKids Foundation; Public Health Ontario","funders":"Hospital for Sick Children; University of Toronto; American College of Gastroenterology","keywords":"Coding (social sciences); Limiting; Inflammatory bowel disease; Residual; Health care; Healthcare system; Diagnosis code","score_opus":0.4905911918657242,"score_gpt":0.5488945518720852,"score_spread":0.05830336000636105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117410197","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9921149,0.000035314304,0.0023152113,0.0035823146,0.000520567,0.0011447102,0.000029144374,0.0000111283725,0.00024671102],"genre_scores_gemma":[0.9969448,0.000007855522,0.0003592399,0.002263826,0.00003384276,0.00015718407,0.00010045067,0.000004845195,0.00012797721],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99337935,0.002727571,0.0031972344,0.0002253409,0.0001438507,0.00032662053],"domain_scores_gemma":[0.976352,0.021472137,0.0015257187,0.00036210628,0.00016640707,0.00012162675],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0091301715,0.000109228225,0.0008403905,0.00008211538,0.0001629302,8.7115706e-7,0.00018193881,0.0003466746,0.00003585058],"category_scores_gemma":[0.028553423,0.00007448575,0.00007812967,0.00015164199,0.00010871467,0.000041790398,0.000036651203,0.0010030642,5.6989455e-7],"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":[0.0003331032,0.000053502412,0.98587584,0.0013874621,0.000008375712,1.7024216e-7,0.00011501151,0.0018768866,7.175476e-7,0.008107197,0.0021739325,0.00006780399],"study_design_scores_gemma":[0.0008776553,0.000039983486,0.96110034,0.0013494738,0.00002827935,1.8078769e-8,0.00010131298,0.034029823,0.0000013703781,0.0012941579,0.0011215128,0.000056091932],"about_ca_topic_score_codex":0.516939,"about_ca_topic_score_gemma":0.9725295,"teacher_disagreement_score":0.45559046,"about_ca_system_score_codex":0.00065388135,"about_ca_system_score_gemma":0.0056926454,"threshold_uncertainty_score":0.99994415},"labels":[],"label_agreement":null}]}