{"meta":{"query_hash":"cf12b98fc3a7","filters":{"category":"metaepi_narrow"},"cohort_total":60,"direct_labels_cover":60,"predictions_cover":60,"exported":60,"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/cf12b98fc3a7","api":"https://metacan.xera.ac/api/v1/cohort?category=metaepi_narrow"},"results":[{"id":"W1544000882","doi":"10.1111/j.1756-5391.2012.01173.x","title":"Systematic reviews of TCM trials: how does inclusion of Chinese trials affect outcome?","year":2012,"lang":"en","type":"article","venue":"Journal of Evidence-Based Medicine","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":"University of Alberta","funders":"","keywords":"Inclusion (mineral); English language; Clinical trial; Chinese language; Systematic review; MEDLINE; Randomized controlled trial; China; Alternative medicine; Medicine; Computer science; Medical education; Psychology; Pathology; Linguistics; Social psychology; Political science","score_opus":0.9254087096431709,"score_gpt":0.6418932184013989,"score_spread":0.283515491241772,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1544000882","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.38564122,0.48095945,0.071555436,0.040302485,0.009081956,0.012014627,0.000026140666,0.000008131528,0.00041056535],"genre_scores_gemma":[0.98966813,0.003777136,0.0035599447,0.00042142146,0.0016908749,0.00003660743,0.0000011957098,0.000017875831,0.00082682143],"study_design_codex":"observational","study_design_gemma":"systematic_review","domain_scores_codex":[0.6303498,0.23563015,0.10921378,0.00080768246,0.023329087,0.0006694524],"domain_scores_gemma":[0.38558534,0.3831993,0.2162828,0.0060567074,0.007872635,0.0010032163],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8492686,0.0005973679,0.037499912,0.0018333021,0.00012852259,0.00010640393,0.0027436735,0.00017772199,0.0063042007],"category_scores_gemma":[0.933216,0.0001485867,0.008555588,0.0025896144,0.00021962752,0.0010803626,0.00025539182,0.00037255188,0.000069134985],"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":[0.0018399614,0.0014401148,0.5241669,0.19458123,0.0052865897,0.000058488655,0.011027563,0.00016563926,0.06524648,0.000490934,0.14578661,0.04990954],"study_design_scores_gemma":[0.022962999,0.009835026,0.058029577,0.7365054,0.068451986,0.0005901997,0.009401921,0.004052696,0.009448092,0.011387677,0.06703042,0.0023039936],"about_ca_topic_score_codex":0.00001280025,"about_ca_topic_score_gemma":0.000006322286,"teacher_disagreement_score":0.6040269,"about_ca_system_score_codex":0.00010273201,"about_ca_system_score_gemma":0.0002551734,"threshold_uncertainty_score":0.9946042},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W1561379059","doi":"10.1371/journal.pone.0132634","title":"PEDro or Cochrane to Assess the Quality of Clinical Trials? A Meta-Epidemiological Study","year":2015,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":173,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Fondation pour la Recherche Médicale; Physiotherapy Foundation of Canada; Alberta Innovates; Killam Trusts; National Institutes of Health; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; University of Alberta; Alberta Innovates - Health Solutions; Women and Children's Health Research Institute; Children's Health Research Institute; Bill and Melinda Gates Foundation; Canadian Institutes of Health Research; National Science Foundation","keywords":"Blinding; Medicine; Meta-analysis; Systematic review; Clinical trial; MEDLINE; Physical therapy; Internal medicine","score_opus":0.9978832001452151,"score_gpt":0.8031065913405215,"score_spread":0.1947766088046936,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1561379059","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":[0.00049421,0.98433703,0.00072234916,0.0004572561,0.0002733341,0.012021552,0.00021259165,0.000010250355,0.0014714525],"genre_scores_gemma":[0.000144131,0.97631323,0.0057988935,0.00034878813,0.0007326962,0.0018200643,0.000021763888,0.000037078113,0.014783349],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.10566914,0.7451407,0.12154703,0.0033050608,0.023765158,0.00057293044],"domain_scores_gemma":[0.113553755,0.7561658,0.09788492,0.022738481,0.008476942,0.0011801195],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.93417895,0.0011043375,0.14991784,0.00038639217,0.00011986047,0.0005167568,0.007663617,0.00058061734,0.01905786],"category_scores_gemma":[0.9231139,0.0002626773,0.02818557,0.0029801282,0.00018061834,0.00010104362,0.0009175164,0.0010752594,0.0050944267],"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":[0.00019888073,0.018956227,0.0019172131,0.01861689,0.14596933,0.000027236392,0.00034400416,0.0000025674774,2.7930022e-7,0.00093058497,0.106453225,0.70658356],"study_design_scores_gemma":[0.00033265044,0.0007960787,0.00024718975,0.0019218684,0.14471927,0.000003288578,0.00038323112,0.000032026514,6.2390534e-8,0.00063489884,0.8505335,0.00039592525],"about_ca_topic_score_codex":0.000025971513,"about_ca_topic_score_gemma":0.00009582097,"teacher_disagreement_score":0.7440803,"about_ca_system_score_codex":0.000042313004,"about_ca_system_score_gemma":0.00075321546,"threshold_uncertainty_score":0.99998254},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W1974429147","doi":"10.1136/annrheumdis-2014-eular.2210","title":"THU0416 Attrition Bias in Rheumatoid Arthritis Randomised Trials with Different Modified Intention-To-Treat Approaches: A Meta-Epidemiological Study","year":2014,"lang":"en","type":"article","venue":"Annals of the Rheumatic Diseases","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Medicine; Randomized controlled trial; Population; Rheumatoid arthritis; Meta-analysis; Internal medicine; Intention-to-treat analysis; Sample size determination; Physical therapy; Imputation (statistics); Study heterogeneity; Missing data; Statistics","score_opus":0.6572351615281036,"score_gpt":0.44747438123377986,"score_spread":0.2097607802943237,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1974429147","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9457786,0.0005577875,0.04773262,0.0009078414,0.000040020634,0.004547081,0.000058239202,0.0001901363,0.00018766335],"genre_scores_gemma":[0.99482375,0.000497969,0.0024047112,0.0001434373,0.0000103230095,0.0020306432,0.0000133119365,0.000037550988,0.000038294053],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99232614,0.0042371014,0.0019797503,0.00047047794,0.0005766143,0.00040990443],"domain_scores_gemma":[0.9866925,0.010833746,0.0011123366,0.000990726,0.00017682964,0.0001938524],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0050612288,0.0005110107,0.0031926022,0.00022062649,0.00010218041,0.000051840783,0.00055886805,0.000094773,0.00009092163],"category_scores_gemma":[0.033633053,0.00024221106,0.00085681217,0.00035311352,0.00018752142,0.0001815049,0.0001820125,0.0001852763,0.0000045936545],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.049207836,0.066729434,0.09167966,0.006591593,0.020461792,0.00004156403,0.012112012,0.0061140787,0.0004670206,0.4710855,0.016228653,0.25928086],"study_design_scores_gemma":[0.015849346,0.0015384281,0.023208547,0.003472989,0.0012486029,0.000008746874,0.001630544,0.002030072,0.0023899353,0.94789886,0.000008056888,0.00071586703],"about_ca_topic_score_codex":0.00009714511,"about_ca_topic_score_gemma":0.00010068565,"teacher_disagreement_score":0.47681338,"about_ca_system_score_codex":0.000032299024,"about_ca_system_score_gemma":0.000031430467,"threshold_uncertainty_score":0.98770803},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W1998831987","doi":"10.1136/bmjopen-2014-005491","title":"Assessing bias in osteoarthritis trials included in Cochrane reviews: protocol for a meta-epidemiological study","year":2014,"lang":"en","type":"article","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital; Cochrane","funders":"National Health and Medical Research Council; University of Ottawa; Medical Research Council; Parker Institute for Cancer Immunotherapy; Canadian Institutes of Health Research; Oak Foundation","keywords":"Medicine; Epidemiology; Protocol (science); Meta-analysis; Osteoarthritis; Alternative medicine; Cochrane collaboration; Systematic review; MEDLINE; Physical therapy; Family medicine; Internal medicine; Pathology; Cochrane Library","score_opus":0.9811986829709822,"score_gpt":0.7555210505783807,"score_spread":0.22567763239260152,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1998831987","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00481998,0.00023899414,0.0073712836,0.0014617252,0.000091305825,0.979772,0.000007672538,0.000004567497,0.006232471],"genre_scores_gemma":[0.019911652,0.0000026570765,0.017623108,0.0007686071,0.00010066148,0.9602454,0.0000041733256,0.000012978575,0.0013308108],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.5092608,0.4163206,0.06432213,0.0032834536,0.0059474995,0.0008655379],"domain_scores_gemma":[0.86546403,0.103653595,0.022605328,0.0070757014,0.0009339904,0.0002673365],"candidate_categories":["metaresearch","metaepi_broad","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.85681033,0.00046905075,0.023500567,0.00044978372,0.00012747639,0.0026337018,0.0036451803,0.00013766921,0.007564977],"category_scores_gemma":[0.71849495,0.00017932673,0.0030001178,0.0017050356,0.000042490636,0.00076655846,0.0008315811,0.00020599787,0.00069424475],"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":[0.0010119154,0.003013588,0.24682085,0.0017220255,0.001440612,0.00006476354,0.0025284488,0.00029791467,0.00011163308,0.0038300287,0.18494265,0.5542156],"study_design_scores_gemma":[0.016165126,0.0017916285,0.05224356,0.002409727,0.0016283906,0.000020247067,0.0023966245,0.0067498963,0.000027508353,0.03204161,0.8835153,0.0010103728],"about_ca_topic_score_codex":0.00011214911,"about_ca_topic_score_gemma":0.001101248,"teacher_disagreement_score":0.6985727,"about_ca_system_score_codex":0.0000432807,"about_ca_system_score_gemma":0.00012164879,"threshold_uncertainty_score":0.99840164},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2001201455","doi":"10.1136/bmjopen-2013-004527","title":"Methodological characteristics and treatment effect sizes in oral health randomised controlled trials: Is there a relationship? Protocol for a meta-epidemiological study","year":2014,"lang":"en","type":"article","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research; Alberta Innovates; Killam Trusts; University of Alberta; Alberta Innovates - Health Solutions; Women and Children's Health Research Institute; Children's Health Research Institute; Fondation pour la Recherche Médicale; Alpha Omega Foundation","keywords":"Medicine; Sample size determination; Protocol (science); Randomized controlled trial; Meta-analysis; Epidemiology; Intervention (counseling); Research design; MEDLINE; Clinical trial; Placebo; Alternative medicine; Physical therapy; Statistics; Psychiatry; Internal medicine; Pathology","score_opus":0.9742725078857929,"score_gpt":0.7329682702713417,"score_spread":0.2413042376144512,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2001201455","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":"methods","domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021705756,0.00006744426,0.019669777,0.0031748277,0.00004415247,0.95470387,0.000022217528,0.0000061366964,0.0006058107],"genre_scores_gemma":[0.07789339,0.0000016577776,0.020487363,0.0005873015,0.000059824033,0.89946383,0.0000041920807,0.00000897566,0.0014934399],"study_design_codex":"observational","study_design_gemma":"randomized_trial","domain_scores_codex":[0.46782643,0.50702095,0.021367755,0.0015749827,0.0017978059,0.00041208044],"domain_scores_gemma":[0.36649677,0.61314255,0.016567161,0.0030169245,0.00046264267,0.00031395844],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.800721,0.00050417206,0.032241408,0.00017847819,0.00021224374,0.00094555895,0.0010720962,0.0001329654,0.0032543214],"category_scores_gemma":[0.722715,0.0001400415,0.0039307894,0.00039397337,0.000048895847,0.00013381692,0.00019990739,0.00012426468,0.00008213591],"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":[0.10878471,0.0023413496,0.66144127,0.00033132097,0.013127583,0.000011806459,0.002403493,0.000040796986,0.000008195908,0.0035274245,0.026404457,0.1815776],"study_design_scores_gemma":[0.47374648,0.016345412,0.37690476,0.00019175494,0.009576299,0.0000133184085,0.0009250059,0.032574493,0.0000051401726,0.027543008,0.06140215,0.0007721654],"about_ca_topic_score_codex":0.00007471153,"about_ca_topic_score_gemma":0.00012695654,"teacher_disagreement_score":0.36496177,"about_ca_system_score_codex":0.000048422877,"about_ca_system_score_gemma":0.000108761385,"threshold_uncertainty_score":0.9976568},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2018803201","doi":"10.1016/s0895-4356(99)00160-2","title":"Randomized trials versus observational studies in adolescent pregnancy prevention","year":2000,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"Adolescent Sexual and Reproductive Health","field":"Health Professions","cited_by":94,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Observational study; Randomized controlled trial; Medicine; Psychological intervention; CINAHL; PsycINFO; Pregnancy; MEDLINE; Family medicine; Psychiatry; Surgery; Internal medicine","score_opus":0.9380215926437274,"score_gpt":0.7360098352874899,"score_spread":0.20201175735623755,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2018803201","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9095352,0.021401737,0.0015257892,0.044811234,0.01838167,0.002795219,0.000006712345,0.00002970165,0.0015126801],"genre_scores_gemma":[0.78691846,0.14878437,0.013041874,0.015580771,0.020785011,0.0002140157,0.000025581652,0.00006459543,0.014585335],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.8795796,0.1023109,0.016858045,0.000401983,0.00028082414,0.0005686444],"domain_scores_gemma":[0.6621632,0.32159957,0.014226069,0.00046972866,0.0010971909,0.00044425094],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.17700537,0.00019164498,0.007367685,0.00012894135,0.00017217828,9.952893e-7,0.0002116023,0.00047991943,0.0008701438],"category_scores_gemma":[0.41809028,0.0001115708,0.0010825901,0.00017477009,0.00048607154,0.00015106848,0.00004963954,0.002277213,0.00011457706],"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":[0.47690928,0.001945216,0.3827553,0.000774741,0.00085166417,0.00003201828,0.0010067645,0.00040912116,0.000002331863,0.009457876,0.021923915,0.10393176],"study_design_scores_gemma":[0.38966817,0.0012670159,0.5075793,0.005719026,0.00025461506,0.000010177884,0.00086093205,0.00016475447,6.55908e-7,0.072980575,0.021285426,0.0002093402],"about_ca_topic_score_codex":0.000012045725,"about_ca_topic_score_gemma":0.000020224834,"teacher_disagreement_score":0.2410849,"about_ca_system_score_codex":0.00014822994,"about_ca_system_score_gemma":0.00079741643,"threshold_uncertainty_score":0.98934835},"labels":[{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2096833901","doi":"10.1371/journal.pone.0088008","title":"Quantifying Bias in Randomized Controlled Trials in Child Health: A Meta-Epidemiological Study","year":2014,"lang":"en","type":"review","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":43,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Epidemiology; Randomized controlled trial; Meta-analysis; Medicine; MEDLINE; Environmental health; Biology; Internal medicine","score_opus":0.9817470593674227,"score_gpt":0.6519590330899844,"score_spread":0.3297880262774383,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2096833901","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":[0.00014971562,0.9530569,0.00035568944,0.0008172194,0.00014734385,0.044478644,0.00002128772,0.0000115455005,0.00096170336],"genre_scores_gemma":[0.0007163549,0.9849271,0.0018639308,0.0003703247,0.00020986833,0.0107133,0.00002520355,0.000040518793,0.0011334014],"study_design_codex":"design_other","study_design_gemma":"randomized_trial","domain_scores_codex":[0.045418743,0.8011624,0.1395652,0.0030830395,0.009936431,0.0008341788],"domain_scores_gemma":[0.08946429,0.78603655,0.1148095,0.008275912,0.0009261728,0.00048757068],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.94928885,0.0016628269,0.4463966,0.0033798157,0.00013392865,0.0008348197,0.003374443,0.00052500557,0.007294501],"category_scores_gemma":[0.9237,0.00050166587,0.0609404,0.0035337722,0.000098009,0.0001198615,0.000298566,0.0013209912,0.0024052707],"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":[0.017912343,0.03241106,0.0019237332,0.05848724,0.40685248,0.00015339172,0.0015925242,0.000054724427,1.17047755e-7,0.004295456,0.007902191,0.46841472],"study_design_scores_gemma":[0.3408185,0.00057509285,0.000052453957,0.05057988,0.30759272,0.000015919122,0.00035226383,0.003479317,3.5923716e-8,0.0023793948,0.29242018,0.0017342465],"about_ca_topic_score_codex":0.00012371823,"about_ca_topic_score_gemma":0.0005874893,"teacher_disagreement_score":0.4666805,"about_ca_system_score_codex":0.00013668042,"about_ca_system_score_gemma":0.00030529944,"threshold_uncertainty_score":0.9997435},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2102685275","doi":"10.1186/2046-4053-2-24","title":"A protocol for a systematic review on the impact of unpublished studies and studies published in the gray literature in meta-analyses","year":2013,"lang":"en","type":"review","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":41,"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":"European Commission; McMaster University","keywords":"Medicine; Grey literature; Meta-analysis; Systematic review; MEDLINE; Cochrane Library; Publication bias; Gray (unit); Protocol (science); Inclusion and exclusion criteria; Gerontology; Alternative medicine; Pathology","score_opus":0.9180444448269799,"score_gpt":0.6734652623171064,"score_spread":0.24457918250987354,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102685275","genre_codex":"review","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[1.3959824e-7,0.5444773,0.000018479932,0.00031462713,0.000051476738,0.45497355,0.000032121636,0.000003616765,0.00012868067],"genre_scores_gemma":[0.0000014872909,0.49876806,0.0001676879,0.0002172592,0.00003579773,0.49994346,0.000009013341,0.00003716607,0.00082008383],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.5639909,0.3256722,0.096413024,0.0034434332,0.009424353,0.0010560548],"domain_scores_gemma":[0.52832544,0.2607465,0.16740736,0.032379437,0.010775697,0.00036557735],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.57098264,0.0033209713,0.13903429,0.0022757843,0.00023171457,0.004017989,0.008564133,0.00041524487,0.00061921973],"category_scores_gemma":[0.6502359,0.0006125668,0.032557193,0.011660253,0.0002821168,0.000995461,0.00048556284,0.001062481,0.00043019722],"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":[0.0000016910736,0.000082349616,0.0000013613802,0.88807774,0.016721485,0.000004649245,0.0014271917,7.3940157e-7,1.8298195e-8,0.0007809234,0.09211556,0.00078630296],"study_design_scores_gemma":[0.0001905577,0.00013727798,5.838609e-7,0.88765126,0.052628376,0.000081927734,0.0014966821,0.000030540927,1.9080648e-8,0.0022654333,0.055029307,0.0004880366],"about_ca_topic_score_codex":0.000018768289,"about_ca_topic_score_gemma":0.000052214593,"teacher_disagreement_score":0.10647709,"about_ca_system_score_codex":0.0003485228,"about_ca_system_score_gemma":0.0003241393,"threshold_uncertainty_score":0.9996326},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"systematic_review","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2102903042","doi":"10.1186/2046-4053-2-88","title":"How should we evaluate the risk of bias of physical therapy trials?: a psychometric and meta-epidemiological approach towards developing guidelines for the design, conduct, and reporting of RCTs in Physical Therapy (PT) area: a study protocol","year":2013,"lang":"en","type":"article","venue":"Systematic Reviews","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Alberta Innovates; Killam Trusts; Fondation pour la Recherche Médicale; University of Alberta; Canadian Institutes of Health Research; Physiotherapy Foundation of Canada; Alberta Innovates - Health Solutions; Women and Children's Health Research Institute; Children's Health Research Institute","keywords":"Medicine; Epidemiology; Randomized controlled trial; Meta-analysis; Physical therapy; Research design; Pathology","score_opus":0.9844898175982594,"score_gpt":0.6756415085022591,"score_spread":0.30884830909600025,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2102903042","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.104559354,0.035441335,0.09523479,0.0025297175,0.00006434312,0.7621102,0.000010409339,0.0000063382026,0.000043538974],"genre_scores_gemma":[0.44213796,0.0022247694,0.030571844,0.00014871014,0.00011247352,0.5246442,8.874256e-7,0.000031405514,0.0001277589],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.80639464,0.14653203,0.041282736,0.0012993976,0.004121628,0.00036958768],"domain_scores_gemma":[0.70664203,0.16660531,0.11307159,0.0068668798,0.0066412357,0.00017295868],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.62581885,0.0008596817,0.031210506,0.0004195696,0.00014422831,0.0003723549,0.0015969689,0.00010613379,0.000045684414],"category_scores_gemma":[0.6230486,0.00017403462,0.0049657742,0.0028380656,0.00024452977,0.00023104204,0.0001492538,0.00024203239,0.0000035281453],"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":[0.00084418355,0.006052303,0.033550642,0.21299121,0.057200506,0.0000028985328,0.05232253,0.0009858508,0.0033346927,0.0038051552,0.025444217,0.6034658],"study_design_scores_gemma":[0.019273367,0.010282649,0.038950767,0.016086565,0.057811346,0.00016925072,0.064479955,0.651675,0.0019880205,0.11619088,0.019738348,0.0033538863],"about_ca_topic_score_codex":0.00006179273,"about_ca_topic_score_gemma":0.0000049236596,"teacher_disagreement_score":0.6506891,"about_ca_system_score_codex":0.000019328143,"about_ca_system_score_gemma":0.000117296106,"threshold_uncertainty_score":0.9831722},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2111376665","doi":"10.1177/0306624x12447771","title":"A Meta-Analysis of Moral Reconation Therapy","year":2012,"lang":"en","type":"review","venue":"International Journal of Offender Therapy and Comparative Criminology","topic":"Psychology of Moral and Emotional Judgment","field":"Neuroscience","cited_by":76,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Saskatchewan","funders":"","keywords":"Psychology; Medicine; Psychotherapist","score_opus":0.9301869343671447,"score_gpt":0.47731096573716825,"score_spread":0.4528759686299765,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2111376665","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":[0.0044990014,0.99286723,0.00012986003,0.00050687714,0.001099209,0.00019640011,0.00008434778,0.0000066725524,0.000610405],"genre_scores_gemma":[0.042495217,0.9563383,0.00015070791,0.00063250354,0.00018697999,0.000019043096,0.000020988404,0.00001053667,0.0001457138],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9970163,0.000815378,0.0011476012,0.00031435303,0.00050650944,0.0001998567],"domain_scores_gemma":[0.99675566,0.0006872258,0.0019253137,0.00016374134,0.0003837558,0.00008430172],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00042559867,0.00035694402,0.0028952337,0.0011607964,0.00004985245,0.000027912301,0.0007305106,0.00021081963,0.0008785339],"category_scores_gemma":[0.000020502473,0.00022679193,0.0019560081,0.00023084441,0.00034807008,0.00023879582,0.00005559848,0.00042706894,0.000009801402],"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":[0.0005264998,0.00096019794,0.000021402435,0.000079559744,0.24510276,0.000038223672,0.0011195225,0.000023502686,0.0002597783,0.019104205,0.00028330507,0.73248106],"study_design_scores_gemma":[0.0013440868,0.0007326173,0.00057441235,0.00010246575,0.092770815,0.0008899009,0.00008974843,0.000019088257,0.000917912,0.0036763535,0.8984349,0.00044772678],"about_ca_topic_score_codex":0.0000054795114,"about_ca_topic_score_gemma":0.0000013216574,"teacher_disagreement_score":0.8981516,"about_ca_system_score_codex":0.00005117994,"about_ca_system_score_gemma":0.00008674535,"threshold_uncertainty_score":0.961933},"labels":[{"model":"gpt","categories":["metaepi_narrow","metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"grok","categories":["metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"opus","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2164404559","doi":"10.1136/bmjopen-2014-005297","title":"Interpreting trial results following use of different intention-to-treat approaches for preventing attrition bias: a meta-epidemiological study protocol","year":2014,"lang":"en","type":"article","venue":"BMJ Open","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"Parker Institute for Cancer Immunotherapy; Oak Foundation","keywords":"Medicine; Protocol (science); Attrition; Epidemiology; Selection bias; Publication bias; Alternative medicine; Meta-analysis; Environmental health; Pathology","score_opus":0.8468178363736624,"score_gpt":0.5865171235954936,"score_spread":0.2603007127781688,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2164404559","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.13026954,3.3986635e-7,0.28690958,0.00013637102,0.00005070685,0.58199924,0.000019154768,0.000161063,0.00045401306],"genre_scores_gemma":[0.33645946,3.5653517e-8,0.15157378,0.000029031713,0.00003961213,0.51166457,0.000011788148,0.000026458216,0.00019527705],"study_design_codex":"randomized_trial","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9957054,0.0014763727,0.0016654468,0.00056142034,0.00028128986,0.00031005082],"domain_scores_gemma":[0.9922353,0.005903223,0.0009811192,0.0006703017,0.000130256,0.00007979857],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0086612385,0.0003238663,0.001413947,0.0001134771,0.00011213154,0.0001432549,0.00056237594,0.00011561802,0.000011715375],"category_scores_gemma":[0.04870853,0.00020686019,0.00054236787,0.00014383285,0.000029693794,0.0004133851,0.00082698365,0.00014417229,0.0000013221922],"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":[0.55674386,0.042335205,0.039098985,0.010475752,0.011903017,0.000035553283,0.01893378,0.00043485875,0.010945414,0.17075813,0.018149382,0.1201861],"study_design_scores_gemma":[0.14563084,0.03836998,0.0017616124,0.0077164234,0.005774295,0.000012165874,0.0048843385,0.018242978,0.030854527,0.7425047,0.0013926477,0.0028555149],"about_ca_topic_score_codex":0.00005373886,"about_ca_topic_score_gemma":0.00007925493,"teacher_disagreement_score":0.5717466,"about_ca_system_score_codex":0.000068471534,"about_ca_system_score_gemma":0.00002005343,"threshold_uncertainty_score":0.95930463},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2287844929","doi":"10.1136/bmjopen-2015-008562","title":"What is the influence of randomisation sequence generation and allocation concealment on treatment effects of physical therapy trials? A meta-epidemiological study","year":2015,"lang":"en","type":"review","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Alberta Health; University of Alberta","funders":"Canadian Institutes of Health Research; Alberta Innovates; University of Alberta; Physiotherapy Foundation of Canada","keywords":"Medicine; Epidemiology; Meta-analysis; Alternative medicine; Sequence (biology); Public health; MEDLINE; Physical therapy; Gerontology; Internal medicine; Pathology","score_opus":0.9578471903528902,"score_gpt":0.6968038048601565,"score_spread":0.26104338549273376,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2287844929","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":[0.01744215,0.90300345,0.0001322808,0.00030056608,0.00014600997,0.07889667,0.000021918338,0.000001786729,0.000055164273],"genre_scores_gemma":[0.012001357,0.9725461,0.00018328294,0.00018781624,0.00012325375,0.014476277,0.000029949306,0.000012494718,0.00043944575],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.8752403,0.108355716,0.011689297,0.0011306702,0.0034406746,0.00014333164],"domain_scores_gemma":[0.93735325,0.040358905,0.017863233,0.0033896498,0.0009448658,0.00009009578],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.25977823,0.0006103638,0.022856496,0.00014739965,0.00008529533,0.0007136922,0.0016848724,0.00012917278,0.000114438626],"category_scores_gemma":[0.04089486,0.00016204019,0.003546937,0.00059395476,0.00011014637,0.00040198743,0.00018642258,0.0001053055,0.00006227122],"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":[0.00012626025,0.0004347092,0.000022905278,0.0007320002,0.00606437,0.0000012171951,0.0026511126,0.00023929354,0.00001898614,0.0004433747,0.0009003665,0.9883654],"study_design_scores_gemma":[0.015001053,0.012216743,0.00056637643,0.009909393,0.0917971,0.000022794557,0.0021320723,0.0059553175,0.00021778655,0.0045927423,0.8561955,0.0013931025],"about_ca_topic_score_codex":0.00008946395,"about_ca_topic_score_gemma":0.000012048749,"teacher_disagreement_score":0.9869723,"about_ca_system_score_codex":0.00009638568,"about_ca_system_score_gemma":0.00031402093,"threshold_uncertainty_score":0.9916502},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2319079108","doi":"10.1016/j.joca.2016.01.101","title":"Assessing bias in osteoarthritis trials included in cochrane reviews: A meta-epidemiological study","year":2016,"lang":"en","type":"article","venue":"Osteoarthritis and Cartilage","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; Bruyère; University of Ottawa; Cochrane","funders":"","keywords":"Blinding; Meta-analysis; Medicine; Systematic review; Publication bias; Reporting bias; MEDLINE; Study heterogeneity; Physical therapy; Sample size determination; Randomized controlled trial; Clinical trial; Psychological intervention; Epidemiology; Random effects model; Cochrane Library; Internal medicine; Statistics; Psychiatry","score_opus":0.8062375650609488,"score_gpt":0.5538883201150719,"score_spread":0.25234924494587685,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2319079108","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.92985994,0.06259112,0.000260249,0.0013309473,0.00023580818,0.003096169,0.00003916762,0.000011176807,0.002575439],"genre_scores_gemma":[0.99472517,0.002134489,0.0005589177,0.000423915,0.00010388285,0.00072428054,0.000004344964,0.000020184827,0.00130481],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.86197925,0.10453553,0.026175344,0.0025502252,0.0038801217,0.00087952596],"domain_scores_gemma":[0.96725404,0.024211073,0.005109867,0.0027313526,0.00038028386,0.00031340352],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.39478216,0.00070386496,0.017010285,0.00072311616,0.00016312282,0.0012079262,0.0010216512,0.0002276163,0.008281958],"category_scores_gemma":[0.20990098,0.00027503193,0.0028000358,0.001719583,0.00018251195,0.00095324387,0.00035841673,0.0002856595,0.00078347075],"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":[0.00007194213,0.0004230778,0.18708356,0.00007761864,0.00010265689,0.0005484025,0.0013252598,0.0000036453034,0.010663638,0.00055890606,0.004702922,0.79443836],"study_design_scores_gemma":[0.05442749,0.0093862675,0.324402,0.009667398,0.008267507,0.0007979345,0.027860902,0.00012916913,0.003651456,0.052259363,0.50279945,0.006351022],"about_ca_topic_score_codex":0.000032132237,"about_ca_topic_score_gemma":0.003131215,"teacher_disagreement_score":0.78808737,"about_ca_system_score_codex":0.000053436484,"about_ca_system_score_gemma":0.000056993773,"threshold_uncertainty_score":0.9999945},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2536368941","doi":"10.1093/ejo/cjw062","title":"Historical controls in orthodontics: need for larger meta-epidemiological studies","year":2016,"lang":"en","type":"letter","venue":"European Journal of Orthodontics","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"University of Alberta; Alberta Innovates; Killam Trusts; Alberta Innovates - Health Solutions; Women and Children's Health Research Institute; Children's Health Research Institute","keywords":"Orthodontics; Epidemiology; Medicine; Dentistry; Pathology","score_opus":0.9015952426937508,"score_gpt":0.5451367677650194,"score_spread":0.3564584749287314,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2536368941","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005886607,0.040513422,0.27153364,0.6789598,0.0044573937,0.0013638209,0.00007515552,0.000009713173,0.0024984237],"genre_scores_gemma":[0.0027227763,0.0035036467,0.2620895,0.5234507,0.028529808,0.000077957266,0.000020378553,0.00040271206,0.17920254],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.80653983,0.13975085,0.043288272,0.0018389517,0.0073164813,0.0012655925],"domain_scores_gemma":[0.891587,0.06587945,0.03267287,0.003339142,0.006224659,0.0002968845],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.4444004,0.0011375252,0.022594867,0.0013466954,0.0001421854,0.00052847614,0.0041449186,0.00040438314,0.0013604093],"category_scores_gemma":[0.343845,0.00041194982,0.015638612,0.00069491455,0.00017146266,0.00029312816,0.00031492565,0.0022165936,0.00052989786],"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":[0.00004665121,0.000034257453,0.009838042,0.000112257985,0.0056479895,0.0012902832,0.00011764107,0.000024399611,0.0000083439,0.00013436592,0.97869486,0.0040509333],"study_design_scores_gemma":[0.00087340485,0.00032271072,0.0007974132,0.00018826942,0.0067865252,0.00015973482,0.0000727794,0.00004244061,8.0959865e-7,0.001256878,0.9890705,0.0004285316],"about_ca_topic_score_codex":8.2775364e-7,"about_ca_topic_score_gemma":0.0000034662473,"teacher_disagreement_score":0.17670411,"about_ca_system_score_codex":0.00036923008,"about_ca_system_score_gemma":0.00016058239,"threshold_uncertainty_score":0.9998332},"labels":[{"model":"gemma","categories":["metaepi_broad"],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2749388036","doi":"10.1177/0022034517725049","title":"Impact of Selection Bias on Treatment Effect Size Estimates in Randomized Trials of Oral Health Interventions: A Meta-epidemiological Study","year":2017,"lang":"en","type":"review","venue":"Journal of Dental Research","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Selection bias; Meta-analysis; Epidemiology; Randomized controlled trial; Psychological intervention; Medicine; Sample size determination; Oral health; Effect modification; Environmental health; Selection (genetic algorithm); Statistics; Dentistry; Confidence interval; Internal medicine; Computer science; Mathematics; Psychiatry; Pathology","score_opus":0.9859906141623999,"score_gpt":0.8396295782838669,"score_spread":0.146361035878533,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2749388036","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":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0076722484,0.97808415,0.00088828965,0.000031580068,0.00039732805,0.012621342,0.00025895936,0.000008245147,0.000037842365],"genre_scores_gemma":[0.002802078,0.9369317,0.059576582,0.0000012082081,0.00016026704,0.00042151322,0.0000022138886,0.0000480612,0.000056319088],"study_design_codex":"design_other","study_design_gemma":"randomized_trial","domain_scores_codex":[0.7887486,0.19320612,0.015055876,0.00047628113,0.001915487,0.0005976062],"domain_scores_gemma":[0.24072504,0.7440867,0.013790346,0.0005592316,0.00055051665,0.00028820452],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.39669845,0.00063443935,0.040490534,0.0012410365,0.00009446629,0.00006680262,0.0007997012,0.00040295382,0.0005341789],"category_scores_gemma":[0.9190332,0.00025548902,0.017764425,0.000551915,0.0004685759,0.00007477305,0.00018820264,0.0017164019,0.00000618795],"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":[0.05611753,0.009903557,0.000677562,0.014773244,0.031560622,0.00009783376,0.000049734546,0.000004180853,0.0000012565476,0.0002918303,0.00037921764,0.88614345],"study_design_scores_gemma":[0.29510933,0.22264238,0.0013809913,0.13310225,0.06186797,0.00035107715,0.000110126704,0.00007055094,0.000049664348,0.28395683,0.00063206814,0.0007267734],"about_ca_topic_score_codex":0.0004981416,"about_ca_topic_score_gemma":0.000067591514,"teacher_disagreement_score":0.8854166,"about_ca_system_score_codex":0.0013414948,"about_ca_system_score_gemma":0.0010153367,"threshold_uncertainty_score":0.99998975},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2756094298","doi":"10.1136/ebmed-2017-110798","title":"What does expert opinion in guidelines mean? a meta-epidemiological study","year":2017,"lang":"en","type":"review","venue":"Evidence-Based Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":38,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Guideline; Observational study; CLARITY; Medicine; Expert opinion; Evidence-based medicine; Evidence-based practice; Family medicine; MEDLINE; Alternative medicine; Pathology; Intensive care medicine","score_opus":0.934234501038434,"score_gpt":0.7009597189922305,"score_spread":0.23327478204620344,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2756094298","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":[0.000014979963,0.9121943,0.0001927012,0.07438689,0.0059177885,0.007102703,0.000010994885,0.00014123328,0.00003841961],"genre_scores_gemma":[0.000067783214,0.9787335,0.002031612,0.00871586,0.0076115443,0.0019025069,0.0003813958,0.00012739624,0.0004284065],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9823679,0.003986571,0.008747175,0.0020974772,0.0019504447,0.0008504786],"domain_scores_gemma":[0.964281,0.026692992,0.004089397,0.0031629126,0.0011364318,0.0006372398],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.028759964,0.0015302339,0.011952052,0.00094347366,0.00020097362,0.00012695152,0.0010876268,0.0007750569,0.0021738724],"category_scores_gemma":[0.21570525,0.0007154682,0.0018428767,0.0006673709,0.00053553295,0.0014003833,0.0003349974,0.0018805627,0.00015031501],"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":[0.0004012448,0.000711383,0.00037136843,0.00715709,0.0016284203,0.00034252653,0.00023994765,0.0000037384161,0.0000012424443,0.0000085244155,0.029257316,0.9598772],"study_design_scores_gemma":[0.00312249,0.002506773,0.00013751724,0.13953836,0.011007054,0.00007603854,0.0015821398,0.0000770148,5.3034654e-7,0.00006510001,0.8413488,0.000538161],"about_ca_topic_score_codex":0.0022547774,"about_ca_topic_score_gemma":0.00059727376,"teacher_disagreement_score":0.959339,"about_ca_system_score_codex":0.00059902016,"about_ca_system_score_gemma":0.0019078997,"threshold_uncertainty_score":0.99974465},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2767159407","doi":"10.1136/bmjopen-2017-018856","title":"Association between statistical significance and time to publication among systematic reviews: a study protocol for a meta-epidemiological investigation","year":2017,"lang":"en","type":"article","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","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":"McMaster University; Impact","funders":"Japan Society for the Promotion of Science","keywords":"Medicine; Publication bias; Protocol (science); Family medicine; Clinical trial; Systematic review; Meta-analysis; Epidemiology; Publication; Peer review; Psychological intervention; MEDLINE; Medical education; Alternative medicine; Pathology; Psychiatry","score_opus":0.9313157082465333,"score_gpt":0.6675016834766426,"score_spread":0.2638140247698907,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2767159407","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0010566063,0.000007639369,0.02433292,0.004663386,0.00002228043,0.9677949,0.000057836674,0.000007191483,0.002057258],"genre_scores_gemma":[0.005251097,1.54582e-7,0.015819438,0.0003433247,0.00008147814,0.9580355,0.000014325678,0.000010532692,0.020444103],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9462198,0.036150314,0.012633798,0.0014800567,0.0031951717,0.00032087398],"domain_scores_gemma":[0.9526178,0.02080832,0.019536309,0.0047544125,0.00192688,0.0003562405],"candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.43747333,0.00032799857,0.0075542433,0.00014376514,0.0005588166,0.0060063666,0.0031075007,0.000110508634,0.0028724002],"category_scores_gemma":[0.62210315,0.00014013749,0.0006667792,0.0003478492,0.000054531243,0.0009712196,0.000613615,0.000104688406,0.002754623],"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":[0.000038969145,0.00012038511,0.43077576,0.004538728,0.0021996042,9.023231e-7,0.0005080431,0.000004691786,0.000010259132,0.0014698589,0.55792946,0.0024033396],"study_design_scores_gemma":[0.0015330492,0.0007942432,0.88735014,0.0016773328,0.008051565,0.0000019166973,0.00027303057,0.008953924,0.0000067976603,0.023167167,0.06747135,0.00071950327],"about_ca_topic_score_codex":0.00006146828,"about_ca_topic_score_gemma":0.00005321489,"teacher_disagreement_score":0.49045813,"about_ca_system_score_codex":0.00008978706,"about_ca_system_score_gemma":0.00008086727,"threshold_uncertainty_score":0.9980391},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"evaluation","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2802073505","doi":"10.1183/13993003.00376-2018","title":"The effect of anti-acid therapy on survival in idiopathic pulmonary fibrosis: a methodological review of observational studies","year":2018,"lang":"en","type":"review","venue":"European Respiratory Journal","topic":"Interstitial Lung Diseases and Idiopathic Pulmonary Fibrosis","field":"Medicine","cited_by":50,"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":"Medicine; Observational study; Pulmonary fibrosis; Idiopathic pulmonary fibrosis; Intensive care medicine; Internal medicine; Fibrosis; Lung","score_opus":0.2886020444561501,"score_gpt":0.44182245393585945,"score_spread":0.15322040947970933,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2802073505","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":[0.0011632949,0.99612594,0.000004197668,0.000081299084,0.0007622684,0.00096029026,0.000050861654,0.000015562802,0.0008363171],"genre_scores_gemma":[0.00053268723,0.9975029,0.000047892314,0.00035398305,0.0013981854,0.000030823336,0.0000140037055,0.00008287852,0.00003663695],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98222166,0.013615348,0.002453263,0.0004500392,0.00088705966,0.00037261195],"domain_scores_gemma":[0.99517214,0.0016344653,0.0019029924,0.0006512205,0.0004630026,0.00017618953],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.016559837,0.0005982317,0.0037378538,0.00025273647,0.0002052364,0.000023582777,0.00061510067,0.0001580423,0.00004274313],"category_scores_gemma":[0.0035894227,0.00029799555,0.0017592418,0.0004939828,0.00053511886,0.00007560585,0.00016707473,0.0012397954,0.00002977589],"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":[0.005455421,0.00036630136,0.0007200238,0.22558913,0.0014174369,0.0018472784,0.000051302297,7.5655095e-7,0.0002616068,0.000116018346,0.003426924,0.7607478],"study_design_scores_gemma":[0.0005292386,0.01197743,0.001455397,0.40565047,0.0007482783,0.0003594836,0.000012558815,5.33522e-7,0.000028792745,0.000007635885,0.5789949,0.00023525332],"about_ca_topic_score_codex":4.8569484e-7,"about_ca_topic_score_gemma":1.2251606e-7,"teacher_disagreement_score":0.76051253,"about_ca_system_score_codex":0.0001551596,"about_ca_system_score_gemma":0.0004540628,"threshold_uncertainty_score":0.9999472},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2886516149","doi":"10.1136/bmjebm-2018-111024.40","title":"40 How fragile is the evidence base? a meta-epidemiologic study of the fragilityindex derived from 374 randomised trials","year":2018,"lang":"en","type":"article","venue":"","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Fragility; Randomized controlled trial; Meta-analysis; Sample size determination; False positive paradox; Medicine; Statistics; Index (typography); Computer science; Mathematics; Internal medicine","score_opus":0.7901114122054491,"score_gpt":0.49293071844313197,"score_spread":0.29718069376231715,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886516149","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8660358,0.004243983,0.0056541474,0.11645231,0.0010106665,0.005142647,0.00033258466,0.000052334723,0.0010755106],"genre_scores_gemma":[0.96919054,0.00012355718,0.0009522169,0.027570277,0.0005381994,0.00066896516,0.00000437812,0.000026214357,0.00092565047],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9817071,0.009098092,0.0074064406,0.0010252264,0.00023254818,0.0005306123],"domain_scores_gemma":[0.9459751,0.04330856,0.0076649915,0.002627652,0.0002531784,0.00017047669],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12366382,0.0003702838,0.0039621857,0.00014291886,0.00057811296,0.00011979096,0.0012843542,0.00022183824,0.007694352],"category_scores_gemma":[0.10444811,0.00021587123,0.0009839339,0.0003693134,0.0003754989,0.00044167647,0.00027053105,0.00030504694,0.00061019865],"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":[0.0037143857,0.0023002706,0.49745795,0.00043406617,0.017117968,0.0000020706286,0.06029989,0.0004076193,0.00041399058,0.017189916,0.39990485,0.0007570373],"study_design_scores_gemma":[0.047869213,0.0018263098,0.6667724,0.0003565515,0.0030596233,0.0000051542693,0.04170156,0.03575674,0.001845718,0.14582163,0.052722927,0.002262168],"about_ca_topic_score_codex":0.008816908,"about_ca_topic_score_gemma":0.0014849078,"teacher_disagreement_score":0.34718192,"about_ca_system_score_codex":0.00019909164,"about_ca_system_score_gemma":0.0001784634,"threshold_uncertainty_score":0.9977835},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2889636695","doi":"10.1002/lom3.10273","title":"Methodological review and meta‐analysis of dilution assays for estimates of virus‐ and grazer‐mediated phytoplankton mortality","year":2018,"lang":"en","type":"article","venue":"Limnology and Oceanography Methods","topic":"Microbial Community Ecology and Physiology","field":"Environmental Science","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Dilution; Virus; Mortality rate; Biology; Meta-analysis; Demography; Virology; Medicine; Internal medicine","score_opus":0.187349636260773,"score_gpt":0.43220506717327944,"score_spread":0.24485543091250644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2889636695","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9419317,0.022466999,0.03428797,0.00033512726,0.00005705145,0.00046485153,0.00018377161,0.000027231756,0.000245276],"genre_scores_gemma":[0.7052842,0.011787433,0.281606,0.0011628122,0.000009486039,0.000035207788,0.00009859611,0.000008511189,0.00000777199],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99706084,0.0020498014,0.0003510225,0.00030651674,0.000038541137,0.00019326767],"domain_scores_gemma":[0.99641496,0.0029958163,0.0002563407,0.00024629835,0.000030726693,0.000055855067],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0032115667,0.0001628883,0.0012359873,0.00013753246,0.00016296258,0.0000025029913,0.00013354186,0.00025360467,0.00055059686],"category_scores_gemma":[0.00052932114,0.00012189938,0.00023480276,0.0005332879,0.0025167568,0.00006545908,0.00020254239,0.00015342694,4.554425e-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":[0.00066255964,0.0005676963,0.6547018,0.0018520181,0.09037374,0.000001696626,0.001172669,0.00001593214,0.23614289,0.0065604798,0.00086561166,0.007082851],"study_design_scores_gemma":[0.00038115354,0.0011781994,0.8000473,0.000010753792,0.14406392,0.000011087879,0.000040369003,0.000444212,0.018270984,0.034109965,0.0011588708,0.00028317902],"about_ca_topic_score_codex":0.000084812804,"about_ca_topic_score_gemma":0.00028643134,"teacher_disagreement_score":0.24731803,"about_ca_system_score_codex":0.0000027868955,"about_ca_system_score_gemma":0.0000035778917,"threshold_uncertainty_score":0.9273095},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2901826044","doi":"10.1093/ije/dyy239","title":"How often can meta-analyses of individual-level data individualize treatment? A meta-epidemiologic study","year":2018,"lang":"en","type":"review","venue":"International Journal of Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital","funders":"Canadian Institutes of Health Research; Nederlandse Organisatie voor Wetenschappelijk Onderzoek; Laura and John Arnold Foundation","keywords":"Subgroup analysis; Meta-analysis; Medicine; Cochrane Library; MEDLINE; Randomized controlled trial; Internal medicine; Biology","score_opus":0.9907924205974797,"score_gpt":0.7172303309820024,"score_spread":0.27356208961547734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2901826044","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":[0.000041195115,0.9816108,0.0059165102,0.0033241438,0.0016426125,0.0014352525,0.005821424,0.0000047385042,0.00020334538],"genre_scores_gemma":[0.00045683255,0.96510607,0.026933603,0.0011269536,0.0013429824,0.00016129792,0.0008198588,0.00006403263,0.0039883484],"study_design_codex":"meta_analysis","study_design_gemma":"not_applicable","domain_scores_codex":[0.7422636,0.18481266,0.059850182,0.0031049578,0.009160831,0.00080773904],"domain_scores_gemma":[0.63932323,0.18030466,0.15804045,0.01248094,0.009204438,0.0006462953],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.470414,0.0018159581,0.07460891,0.0037006843,0.00011589333,0.00059119717,0.02699345,0.0007299705,0.008299517],"category_scores_gemma":[0.5215519,0.0006480249,0.035088368,0.0016190264,0.0005851132,0.0006768075,0.0024054023,0.000927817,0.00026088263],"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":[0.000018570294,0.0006218968,0.0012193029,0.00024335679,0.6077243,0.00011867309,0.00020018732,0.00002277863,1.5680611e-7,0.0004393044,0.13308981,0.25630164],"study_design_scores_gemma":[0.00029067846,0.00057258445,0.00018245526,0.00019218646,0.29532042,0.00047285133,0.00020284975,0.000022102595,2.8863354e-7,0.00577676,0.6966591,0.0003077263],"about_ca_topic_score_codex":0.00020615614,"about_ca_topic_score_gemma":0.00020728225,"teacher_disagreement_score":0.5635693,"about_ca_system_score_codex":0.0002679116,"about_ca_system_score_gemma":0.001089187,"threshold_uncertainty_score":0.9995971},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2911968494","doi":"10.1136/bmjebm-2018-111057","title":"Use of modified intention-to-treat analysis in studies of direct oral anticoagulants and risk of selection bias: a systematic review","year":2019,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Atrial Fibrillation Management and Outcomes","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":"Centre Hospitalier de l’Université de Montréal","funders":"Fonds de Recherche du Québec - Santé","keywords":"Selection (genetic algorithm); Risk analysis (engineering); Selection bias; Medicine; Intensive care medicine; Psychology; Actuarial science; Computer science; Business; Artificial intelligence; Pathology","score_opus":0.5686784527537067,"score_gpt":0.5045638600117248,"score_spread":0.06411459274198184,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2911968494","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":[0.0017711039,0.9899677,0.0002542753,0.0002827995,0.000103988976,0.0075679505,0.000023436567,0.00001712889,0.000011585111],"genre_scores_gemma":[0.00520394,0.99383104,0.00031394305,0.000055806286,0.00006044392,0.000059235594,0.00005837406,0.00002461369,0.00039261614],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.993974,0.0012847815,0.0032992042,0.0004456423,0.0008087247,0.0001876828],"domain_scores_gemma":[0.9904341,0.0049069775,0.0032188185,0.00065528613,0.0006900176,0.00009481291],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0044198935,0.00043263214,0.011644305,0.0022614459,0.000019716592,0.000004655714,0.00012474672,0.0001436813,0.000033395067],"category_scores_gemma":[0.027442748,0.00025424123,0.0008559121,0.003703985,0.0001633839,0.00012616735,0.00006659931,0.00015940139,0.0000033171343],"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":[0.00015011287,0.000020657058,0.056460988,0.92883676,0.0045141103,0.0000055881364,0.0000613212,0.000055253004,0.0000049935106,0.000010958071,0.00016947444,0.009709801],"study_design_scores_gemma":[0.00065635185,0.00054553867,0.0026127666,0.8563666,0.1388536,0.000004195192,0.00006500849,0.00032466894,0.000008871146,0.0000023052323,0.0003968511,0.0001632725],"about_ca_topic_score_codex":0.0008178712,"about_ca_topic_score_gemma":0.000105957544,"teacher_disagreement_score":0.13433948,"about_ca_system_score_codex":0.00012984367,"about_ca_system_score_gemma":0.00019800136,"threshold_uncertainty_score":0.999991},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3039344762","doi":"10.1136/annrheumdis-2020-217237","title":"Population characteristics as important contextual factors in rheumatological trials: an exploratory meta-epidemiological study from an OMERACT Working Group","year":2020,"lang":"en","type":"review","venue":"Annals of the Rheumatic Diseases","topic":"Rheumatoid Arthritis Research and Therapies","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; Bruyère; University of Ottawa","funders":"Odense Universitetshospital; Oak Foundation","keywords":"Medicine; Epidemiology; Population; Meta-analysis; Physical therapy; Family medicine; Internal medicine; Gerontology; Environmental health","score_opus":0.4342632187980843,"score_gpt":0.4729137158865182,"score_spread":0.03865049708843388,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3039344762","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":[0.3189391,0.6776414,0.0000039315364,0.00012807659,0.00022011177,0.0025075416,0.00048113387,0.000074834374,0.000003875802],"genre_scores_gemma":[0.39433125,0.60407156,0.00001718313,0.00009477978,0.000082734536,0.00029254385,0.0010563931,0.00005169489,0.0000018496447],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.98109233,0.011894457,0.004445974,0.0009115578,0.0010466621,0.0006090309],"domain_scores_gemma":[0.98869425,0.006441811,0.0027705997,0.0011752613,0.00012786368,0.00079018826],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0045408625,0.0009866487,0.011924762,0.0002875377,0.00017367363,0.00013148908,0.0008617882,0.00033839475,0.0005076461],"category_scores_gemma":[0.025446588,0.0004829425,0.0025157225,0.0004818514,0.00025893733,0.00037596727,0.0002744062,0.0008191894,0.000018766155],"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":[0.00084349414,0.005061229,0.08493754,0.0022530896,0.008458403,0.00017334461,0.0013819184,9.3089824e-7,5.0540064e-7,0.0003321359,0.00015946559,0.89639795],"study_design_scores_gemma":[0.008444112,0.012350182,0.71829396,0.11042976,0.01631878,0.00015013613,0.016576653,0.00043287635,0.0000047179833,0.011532053,0.101329826,0.004136932],"about_ca_topic_score_codex":0.00068229804,"about_ca_topic_score_gemma":0.00013858969,"teacher_disagreement_score":0.892261,"about_ca_system_score_codex":0.000059398164,"about_ca_system_score_gemma":0.00034269918,"threshold_uncertainty_score":0.99976224},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3048211970","doi":"10.1016/j.jclinepi.2020.08.002","title":"Abbreviated and comprehensive literature searches led to identical or very similar effect estimates: a meta-epidemiological study","year":2020,"lang":"en","type":"review","venue":"Journal of Clinical Epidemiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","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":"Institute of Health Economics; University of Alberta","funders":"","keywords":"Interquartile range; Meta-analysis; Confidence interval; Medicine; Odds ratio; MEDLINE; Epidemiology; Statistics; Systematic review; Internal medicine; Mathematics; Biology","score_opus":0.9657825327774562,"score_gpt":0.7373159784020075,"score_spread":0.22846655437544872,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3048211970","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":[0.00059164286,0.98380446,0.006010921,0.00456506,0.0012461068,0.003678452,0.000056721175,0.000011328486,0.000035331734],"genre_scores_gemma":[0.00014478016,0.973561,0.020205524,0.004390755,0.0012765676,0.00011584032,0.000014957355,0.000054581236,0.00023602172],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.35168427,0.5268719,0.11291221,0.0033443097,0.004151956,0.001035326],"domain_scores_gemma":[0.11836913,0.81235003,0.059819687,0.0037207394,0.0031903957,0.0025500068],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity","insufficient_payload"],"category_scores_codex":[0.65717435,0.0022846272,0.18118617,0.0011588108,0.00017743459,0.0005645743,0.0049922173,0.0018434811,0.0028388405],"category_scores_gemma":[0.95434874,0.00070771464,0.043487135,0.0032226245,0.0005618383,0.00027348098,0.0013700783,0.005560776,0.0015128099],"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":[0.0007511529,0.00062727946,0.0113522485,0.0064770095,0.051405083,0.0026392918,0.00015866791,0.00007999734,9.162844e-8,0.00033451302,0.1296183,0.79655635],"study_design_scores_gemma":[0.0007525465,0.0062327664,0.002774846,0.004219295,0.03344564,0.0013661206,0.00006581525,0.00059915934,1.2755401e-8,0.002857459,0.9471851,0.00050121953],"about_ca_topic_score_codex":0.0000039070906,"about_ca_topic_score_gemma":0.000005131004,"teacher_disagreement_score":0.8175668,"about_ca_system_score_codex":0.0000732152,"about_ca_system_score_gemma":0.0006131437,"threshold_uncertainty_score":0.9995374},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3097681315","doi":"10.1186/s40814-020-00712-z","title":"More pilot trials could plan to use qualitative data: a meta-epidemiological study","year":2020,"lang":"en","type":"article","venue":"Pilot and Feasibility Studies","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","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":"St. Joseph’s Healthcare Hamilton; Population Health Research Institute; Laurentian University; McMaster University; Hamilton Health Sciences; Toronto General Hospital; University Health Network; Impact; Health Sciences Centre","funders":"","keywords":"Qualitative research; Medicine; Qualitative property; Psychological intervention; Clinical trial; Research design; Randomized controlled trial; Confidence interval; Data collection; Family medicine; Odds ratio; Clinical study design; Physical therapy; Nursing; Internal medicine","score_opus":0.9954596694666988,"score_gpt":0.7332717440502268,"score_spread":0.26218792541647207,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3097681315","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9435537,0.011328282,0.0040959227,0.030819392,0.00027016498,0.008770748,0.0009606556,0.00003912206,0.00016201768],"genre_scores_gemma":[0.99163586,0.00012943152,0.0038731808,0.0033024638,0.00013250666,0.00026766578,0.00001746659,0.000010869309,0.0006305315],"study_design_codex":"not_applicable","study_design_gemma":"qualitative","domain_scores_codex":[0.8851428,0.08963238,0.016110513,0.0039583202,0.0046406393,0.00051533856],"domain_scores_gemma":[0.7321855,0.24026987,0.009515725,0.01275678,0.003853143,0.001418994],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.35441625,0.00064344634,0.015324447,0.0001405645,0.00036389206,0.00066339405,0.0019719817,0.000046993664,0.00073599647],"category_scores_gemma":[0.68352556,0.0002393008,0.0012410311,0.0011520775,0.00029087393,0.00049625663,0.001721309,0.0002532118,0.00046205914],"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":[0.0016931281,0.0018782242,0.3874937,0.00025924196,0.031110577,0.000027729347,0.12229067,0.000031633634,0.00006568444,0.0010890366,0.45226538,0.0017949807],"study_design_scores_gemma":[0.0033799948,0.012343433,0.4426449,0.00004455004,0.021553792,0.000005762323,0.45174026,0.0012695485,0.000006812019,0.017802518,0.04787775,0.0013306914],"about_ca_topic_score_codex":0.0001075637,"about_ca_topic_score_gemma":0.00033270672,"teacher_disagreement_score":0.40438765,"about_ca_system_score_codex":0.000026375692,"about_ca_system_score_gemma":0.000035291916,"threshold_uncertainty_score":0.9992941},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3110084249","doi":"10.1016/j.joca.2020.10.004","title":"Visual Analogue Scale has higher assay sensitivity than WOMAC pain in detecting between-group differences in treatment effects: a meta-epidemiological study","year":2020,"lang":"en","type":"review","venue":"Osteoarthritis and Cartilage","topic":"Osteoarthritis Treatment and Mechanisms","field":"Medicine","cited_by":85,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; Dalhousie University; McMaster University; University of Toronto; St. Michael's Hospital","funders":"Arthritis Society","keywords":"WOMAC; Visual analogue scale; Medicine; Osteoarthritis; Meta-analysis; Randomized controlled trial; Placebo; Physical therapy; Internal medicine; Pathology","score_opus":0.08019099511676102,"score_gpt":0.32141410434270895,"score_spread":0.24122310922594792,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3110084249","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.34002736,0.6559978,0.000009322458,0.000057138037,0.00012584862,0.0035260662,0.000096338095,0.00008752849,0.00007257099],"genre_scores_gemma":[0.6172987,0.38054007,0.00008721308,0.00005505942,0.00027193342,0.001380742,0.00022792206,0.0000735585,0.00006479561],"study_design_codex":"design_other","study_design_gemma":"nonrandomized_trial","domain_scores_codex":[0.98896784,0.0069168857,0.0013213664,0.0015060549,0.00040813023,0.0008797527],"domain_scores_gemma":[0.995715,0.0030151205,0.0003726061,0.00040118728,0.000029830593,0.00046623455],"candidate_categories":["metaepi_narrow"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.0032017143,0.0013041075,0.008256466,0.00038371078,0.00018106836,0.00012674447,0.000100389436,0.00065792695,0.00004408307],"category_scores_gemma":[0.0004299374,0.00087707164,0.001139031,0.00062584225,0.00016620607,0.000118188385,0.00016121297,0.0007736042,0.000021844897],"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":[0.000070271664,0.00072622905,0.028630536,0.0012894802,0.00021928469,0.00653936,0.00029001207,2.0471104e-8,0.00012533937,0.000005871539,0.0000029858497,0.9621006],"study_design_scores_gemma":[0.07494086,0.36262017,0.041632064,0.088798314,0.080052935,0.0018533162,0.004789604,0.00003422134,0.0012024151,0.0003673017,0.3341056,0.009603172],"about_ca_topic_score_codex":0.00020590163,"about_ca_topic_score_gemma":0.007864297,"teacher_disagreement_score":0.9524974,"about_ca_system_score_codex":0.00031194984,"about_ca_system_score_gemma":0.00011578478,"threshold_uncertainty_score":0.99997103},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3134334065","doi":"10.1016/j.jebdp.2021.101544","title":"Influence of Sponsorship Bias on Treatment Effect Size Estimates in Randomized Trials of Oral Health Interventions: A Meta-epidemiological Study","year":2021,"lang":"en","type":"review","venue":"Journal of Evidence Based Dental Practice","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital; University of Alberta","funders":"Alberta Heritage Foundation for Medical Research; Alberta Innovates; Alberta Innovates - Health Solutions","keywords":"Meta-analysis; Publication bias; Randomized controlled trial; Medicine; Epidemiology; Clinical trial; Confidence interval; Sample size determination; Psychological intervention; Random effects model; Selection bias; Reporting bias; Relative risk; MEDLINE; Internal medicine; Statistics; Psychiatry; Pathology","score_opus":0.9725301236095575,"score_gpt":0.7303173162000859,"score_spread":0.2422128074094716,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3134334065","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":[0.012718346,0.9787424,0.0006284066,0.0006047137,0.00024187262,0.0070125856,0.00004090865,0.000001992175,0.000008741875],"genre_scores_gemma":[0.022990769,0.9669019,0.009312577,0.00021553233,0.0000513947,0.00037737086,0.000004171643,0.00002495205,0.000121363635],"study_design_codex":"design_other","study_design_gemma":"meta_analysis","domain_scores_codex":[0.21974352,0.6939877,0.076459765,0.0012856763,0.008109398,0.00041390222],"domain_scores_gemma":[0.05071898,0.80602294,0.13804522,0.0024666125,0.0024466456,0.00029958395],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.8537911,0.0011680009,0.09789299,0.0018291683,0.00007672135,0.0004104681,0.002381393,0.00026391938,0.0021138755],"category_scores_gemma":[0.9787639,0.00036498348,0.047945827,0.003092394,0.00021592132,0.0014030719,0.0001765974,0.00093055726,0.00009273147],"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":[0.09401018,0.021739805,0.00626162,0.049006272,0.117114395,0.0022667013,0.00063373375,0.013448556,0.0000061484684,0.00042756903,0.0023878291,0.69269717],"study_design_scores_gemma":[0.0683477,0.030797707,0.001801256,0.37842274,0.44067192,0.0018677996,0.0019621972,0.0004912569,0.000027617829,0.0006568917,0.073639825,0.0013130651],"about_ca_topic_score_codex":0.0002299755,"about_ca_topic_score_gemma":0.00005319107,"teacher_disagreement_score":0.69138414,"about_ca_system_score_codex":0.0005174481,"about_ca_system_score_gemma":0.0015358902,"threshold_uncertainty_score":0.9998802},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3142062327","doi":"10.9778/cmajo.20200140","title":"Completeness of reporting for COVID-19 case reports, January to April 2020: a meta-epidemiologic study","year":2021,"lang":"en","type":"article","venue":"CMAJ Open","topic":"COVID-19 and healthcare impacts","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"Queen's University; University of Toronto; St. Michael's Hospital","funders":"Amgen","keywords":"Coronavirus disease 2019 (COVID-19); Pandemic; Completeness (order theory); 2019-20 coronavirus outbreak; Medicine; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Coronavirus; Meta-analysis; MEDLINE; Disease; Infectious disease (medical specialty); Outbreak; Virology; Pathology; Political science; Mathematics","score_opus":0.5234770417953649,"score_gpt":0.546135499860114,"score_spread":0.022658458064749087,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3142062327","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9432207,0.0014107078,0.0037684774,0.04131962,0.00038258245,0.008769542,0.00008916399,0.00008161094,0.0009576473],"genre_scores_gemma":[0.96240497,0.000017238823,0.006517008,0.02918504,0.0001694959,0.00062277535,0.00009046778,0.000030321604,0.0009626684],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9953443,0.00037472585,0.0027008562,0.000830357,0.00028686607,0.00046291287],"domain_scores_gemma":[0.9937241,0.0016203772,0.0018083387,0.0013081381,0.0006271129,0.0009119546],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.008556453,0.00025026093,0.002133385,0.00010132447,0.00024584786,0.000045924175,0.0001635605,0.00011061513,0.0003760047],"category_scores_gemma":[0.067645565,0.0001913483,0.00036681595,0.00051159854,0.00004195537,0.00010099282,0.0005971259,0.00020413865,0.0000062689173],"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":[0.0012442818,0.0014920485,0.5867663,0.0035794554,0.0023520973,0.31918472,0.005315554,0.00024529218,0.0021885447,0.00021482239,0.072663136,0.004753764],"study_design_scores_gemma":[0.010018138,0.007630518,0.38855234,0.000616833,0.009858015,0.1473893,0.029713113,0.0002699586,0.0027766058,0.0038143825,0.3978684,0.0014924078],"about_ca_topic_score_codex":0.024940873,"about_ca_topic_score_gemma":0.0074194074,"teacher_disagreement_score":0.32520527,"about_ca_system_score_codex":0.00026556055,"about_ca_system_score_gemma":0.0029474392,"threshold_uncertainty_score":0.9815521},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3154585211","doi":"10.1016/j.jid.2021.02.287","title":"265 A meta-epidemiologic study of power and sample size in randomized controlled trials published in dermatology journals","year":2021,"lang":"en","type":"article","venue":"Journal of Investigative Dermatology","topic":"COVID-19 and healthcare impacts","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; Queen's University; Women's College Hospital; McMaster University; University of Toronto","funders":"","keywords":"Sample size determination; Medicine; Randomized controlled trial; Meta-analysis; Consolidated Standards of Reporting Trials; Sample (material); MEDLINE; Outcome (game theory); Statistics; Internal medicine; Mathematics","score_opus":0.2074584655249477,"score_gpt":0.44687235754155613,"score_spread":0.23941389201660843,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3154585211","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.89777017,0.013747001,0.00008926385,0.08640158,0.00020310441,0.0016950436,0.00000767496,0.00000665442,0.00007948696],"genre_scores_gemma":[0.98787487,0.0011956567,0.0025445668,0.008238552,0.000043795466,0.00007346137,0.0000015429238,0.0000172678,0.00001030515],"study_design_codex":"observational","study_design_gemma":"randomized_trial","domain_scores_codex":[0.968243,0.022541028,0.007884057,0.00034114503,0.0004642799,0.00052650727],"domain_scores_gemma":[0.7676263,0.21646215,0.011197522,0.00060932466,0.0030183028,0.001086384],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.02044174,0.00038390214,0.019896915,0.001333095,0.000048286063,0.00003283882,0.00017466456,0.00038474906,0.0005393107],"category_scores_gemma":[0.56202227,0.00022330304,0.0012398284,0.00079272734,0.00063436915,0.00029469896,0.000090143265,0.0012396161,8.509248e-7],"study_design_candidate":"randomized_trial","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.29163593,0.004848208,0.59702694,0.0010472294,0.026828807,0.023763722,0.024995096,0.00013361119,0.022744857,0.0027716428,0.003897017,0.00030694972],"study_design_scores_gemma":[0.7369062,0.0010932041,0.1560695,0.0005425503,0.0052458956,0.0163444,0.007340111,0.00015525903,0.00898692,0.06691496,0.00011905331,0.00028195383],"about_ca_topic_score_codex":0.00038360318,"about_ca_topic_score_gemma":0.0006981426,"teacher_disagreement_score":0.54158056,"about_ca_system_score_codex":0.000127434,"about_ca_system_score_gemma":0.0020922455,"threshold_uncertainty_score":0.99465376},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_broad"],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3158629300","doi":"10.1186/s12893-021-01278-0","title":"Bias estimation in study design: a meta-epidemiological analysis of transcatheter versus surgical aortic valve replacement","year":2021,"lang":"en","type":"article","venue":"BMC Surgery","topic":"Advanced Causal Inference Techniques","field":"Mathematics","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; Women's College Hospital; University of Toronto; Institute for Work & Health; Western University","funders":"University of Toronto; Canadian Institutes of Health Research; University of Ottawa","keywords":"Medicine; Aortic valve replacement; Propensity score matching; Stenosis; Randomized controlled trial; Aortic valve stenosis; Surgery; Aortic valve; Internal medicine; Epidemiology; Cardiology; Meta-analysis","score_opus":0.6141875518604735,"score_gpt":0.4766740758921051,"score_spread":0.13751347596836838,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3158629300","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7083968,0.00016307269,0.2907833,0.00003388581,0.000051625837,0.00038953347,0.0000056057324,0.000108183,0.00006800531],"genre_scores_gemma":[0.9282903,0.000030655618,0.07133145,0.00001545347,0.0000076283327,0.0002622665,0.000019814686,0.000015594465,0.000026814007],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9958096,0.0018705529,0.001233905,0.0004402717,0.0003620416,0.00028360277],"domain_scores_gemma":[0.973603,0.025270384,0.000347872,0.000589661,0.00012509897,0.00006397584],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0050786473,0.00024355609,0.0017164595,0.00041796037,0.000026005262,0.000013149421,0.000105338775,0.000119069635,0.0005053596],"category_scores_gemma":[0.007827071,0.00018312263,0.0015495423,0.0013205466,0.000054913013,0.0001097858,0.00004870683,0.00015667277,0.0000032745893],"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":[0.0038659219,0.011026831,0.8866314,0.00041658626,0.043199833,0.00029071333,0.0030702122,0.032831103,0.00039204126,0.012031813,0.0009442208,0.0052993004],"study_design_scores_gemma":[0.007409279,0.003181731,0.3450291,0.00045552076,0.1404983,0.000046490193,0.005217322,0.21489352,0.030614765,0.24844484,0.00021037464,0.0039987867],"about_ca_topic_score_codex":0.000026249874,"about_ca_topic_score_gemma":0.0001118249,"teacher_disagreement_score":0.5416024,"about_ca_system_score_codex":0.000094636714,"about_ca_system_score_gemma":0.000099234785,"threshold_uncertainty_score":0.93702996},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3172845662","doi":"10.1186/s12916-021-02008-2","title":"Utilization of the evidence from studies with no events in meta-analyses of adverse events: an empirical investigation","year":2021,"lang":"en","type":"article","venue":"BMC Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Medical Research Council; National Health and Medical Research Council; Qatar National Library","keywords":"Systematic review; Medicine; Meta-analysis; Adverse effect; MEDLINE; Internal medicine","score_opus":0.951712016036363,"score_gpt":0.6398872374389903,"score_spread":0.3118247785973727,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3172845662","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98434937,0.007303016,0.0066654733,0.0010467592,0.00013987078,0.00039779264,0.000006446047,0.0000015368527,0.00008973976],"genre_scores_gemma":[0.99373966,0.00011071354,0.004849961,0.00025205786,0.0000423824,0.000021823522,0.000009947655,0.000005014155,0.0009684507],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9757277,0.013291159,0.005685157,0.0006231412,0.0045744623,0.00009840299],"domain_scores_gemma":[0.9833187,0.0069116955,0.0043641934,0.0026185713,0.0027081277,0.000078731],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.037238527,0.00019219691,0.0034258827,0.00019453373,0.000037906244,0.000004260979,0.00075992785,0.000043115004,0.002704557],"category_scores_gemma":[0.14314534,0.0000610181,0.00065300585,0.0024457434,0.00017192302,0.0002729145,0.00010579211,0.00007761607,0.000034273],"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":[0.00001749747,0.000049926108,0.99279994,0.00009534943,0.0012820547,0.0000011216952,0.0021725574,0.00076455844,0.0014643184,0.000025715291,0.0011875293,0.00013941994],"study_design_scores_gemma":[0.00049631833,0.00013153425,0.96853566,0.0010076003,0.00902125,0.0000023901373,0.0072974884,0.005882036,0.0029206977,0.0043962565,0.00020221337,0.00010653165],"about_ca_topic_score_codex":0.00016324464,"about_ca_topic_score_gemma":0.0013981297,"teacher_disagreement_score":0.10590681,"about_ca_system_score_codex":0.000025266441,"about_ca_system_score_gemma":0.0001896791,"threshold_uncertainty_score":0.9982071},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3192941963","doi":"10.1111/jebm.12440","title":"Comparing and combining evidence of treatment effects in randomized and nonrandomized studies on the use of misoprostol to prevent postpartum hemorrhage","year":2021,"lang":"en","type":"review","venue":"Journal of Evidence-Based Medicine","topic":"Maternal and fetal healthcare","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; Impact","funders":"","keywords":"Misoprostol; Medicine; Randomized controlled trial; Confidence interval; Odds ratio; Meta-analysis; Placebo; MEDLINE; Relative risk; Obstetrics; Pregnancy; Internal medicine; Abortion; Alternative medicine","score_opus":0.4126879839645758,"score_gpt":0.4530451666427994,"score_spread":0.040357182678223635,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3192941963","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":[0.06696076,0.92504597,0.00002864152,0.004064195,0.00031459786,0.003576755,0.0000030604488,0.0000033912756,0.0000026280004],"genre_scores_gemma":[0.06112304,0.9376012,0.0004967728,0.00041745626,0.00015300167,0.00010538545,0.0000024145488,0.000024855202,0.00007587598],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9936243,0.0022019346,0.0028932043,0.00031655785,0.0007198243,0.00024411987],"domain_scores_gemma":[0.9640715,0.032305624,0.002402266,0.0003535662,0.0005691212,0.00029790783],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.005626046,0.00049764523,0.009934409,0.00056723866,0.000046442678,0.0000132371615,0.00014251452,0.00012817461,0.0000127400135],"category_scores_gemma":[0.019548103,0.00020372853,0.00059209426,0.00033205366,0.00039301027,0.00011607902,0.00007535798,0.0005239042,4.7287023e-7],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.2870779,0.0010979978,0.0012137223,0.39266878,0.0031533996,0.0033663078,0.004840796,0.00007961935,0.00029116674,0.00008149346,0.00020475202,0.30592406],"study_design_scores_gemma":[0.08447317,0.009027326,0.00006670343,0.898142,0.0055692657,0.0006854057,0.00025827446,0.0001972598,0.00024560103,0.000012937215,0.0011538452,0.00016821148],"about_ca_topic_score_codex":0.00019053646,"about_ca_topic_score_gemma":0.000022222697,"teacher_disagreement_score":0.5054732,"about_ca_system_score_codex":0.00029581325,"about_ca_system_score_gemma":0.0006914561,"threshold_uncertainty_score":0.98871064},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3203021509","doi":"10.1016/j.ocarto.2021.100215","title":"The state of trial registrations in the field of Orthopaedics in years 2015–2020. A meta-epidemiological study","year":2021,"lang":"en","type":"review","venue":"Osteoarthritis and Cartilage Open","topic":"Meta-analysis and systematic reviews","field":"Decision 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":false,"ca_institutions":"Western University; Hand and Upper Limb Clinic; St. Joseph's Hospital; University of Toronto; Queen's University; Canadian Institute for Military and Veteran Health Research","funders":"","keywords":"Medicine; Randomized controlled trial; Epidemiology; Population; Meta-analysis; Orthopedic surgery; Trial registration; MEDLINE; Prospective cohort study; Surgery; Internal medicine; Environmental health","score_opus":0.6498741816734732,"score_gpt":0.5627773797430697,"score_spread":0.08709680193040348,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3203021509","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":[0.0050013755,0.98761475,0.0000079633555,0.00032492005,0.0001260628,0.0053727566,0.00007533456,5.199424e-7,0.001476315],"genre_scores_gemma":[0.0055773887,0.9918593,0.00008490477,0.00008787668,0.000037206188,0.00091623317,0.000015130895,0.000009072951,0.0014128534],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9359346,0.046495583,0.013927654,0.0009824472,0.002388206,0.0002715149],"domain_scores_gemma":[0.9712659,0.02087075,0.004793054,0.0027503737,0.00026270436,0.000057205176],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1530054,0.000330561,0.010571766,0.00012676549,0.000086222964,0.00075531367,0.002692266,0.00012723124,0.00077527197],"category_scores_gemma":[0.038006097,0.00011574829,0.0026475857,0.0015494766,0.00010377037,0.00012277627,0.000526346,0.00038439818,0.000016798744],"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":[0.000062423926,0.00016018415,0.0001311093,0.00020766782,0.00020792484,0.00010276025,0.00066741987,0.000003100953,1.2515707e-7,0.00038752978,0.010595989,0.9874738],"study_design_scores_gemma":[0.001794501,0.00066603586,0.000069032925,0.00085665646,0.0017955333,0.000018167284,0.0019103719,0.000003294516,1.3079799e-7,0.0007640437,0.99198025,0.00014199196],"about_ca_topic_score_codex":0.0001105895,"about_ca_topic_score_gemma":0.0085881455,"teacher_disagreement_score":0.98733175,"about_ca_system_score_codex":0.000008438367,"about_ca_system_score_gemma":0.00022707753,"threshold_uncertainty_score":0.9700972},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4200365875","doi":"10.22541/au.163827914.48781130/v1","title":"Empirical evaluation of fundamental principles of evidence-based medicine: a meta-epidemiological study","year":2021,"lang":"en","type":"preprint","venue":"","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Grading (engineering); Meta-analysis; Medicine; Odds ratio; Systematic review; Publication bias; Statistics; MEDLINE; Econometrics; Internal medicine; Mathematics; Chemistry","score_opus":0.9923385322462839,"score_gpt":0.7148133160779869,"score_spread":0.27752521616829695,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200365875","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9561247,0.012554825,0.018170256,0.0036601818,0.0004765991,0.0050376835,0.000020100377,0.000007496889,0.0039481106],"genre_scores_gemma":[0.9885376,0.000051832772,0.009326169,0.0003002619,0.00007584161,0.00042913517,0.000025268722,0.000009879439,0.0012440034],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.6685302,0.23137328,0.05017403,0.0043128734,0.04514855,0.0004610373],"domain_scores_gemma":[0.919193,0.043889295,0.019581912,0.0077019217,0.009428367,0.00020544649],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6150869,0.0007099667,0.019026522,0.0006777918,0.00005873214,0.0001956599,0.0031903337,0.0003303995,0.12627871],"category_scores_gemma":[0.4582665,0.00022363059,0.007826623,0.0014808782,0.00032346917,0.00014363951,0.0013532883,0.00046508256,0.000094984396],"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":[0.00005161572,0.0017529619,0.9328166,0.00063720706,0.013588498,0.000008244575,0.0020134351,0.028406119,0.00014330578,0.00034441295,0.013798616,0.0064389734],"study_design_scores_gemma":[0.0011039883,0.0010475303,0.6511313,0.0010953,0.07185585,0.0000057159527,0.016106825,0.24926858,0.00016521422,0.0054520615,0.002141838,0.0006258084],"about_ca_topic_score_codex":0.00019075337,"about_ca_topic_score_gemma":0.00017016263,"teacher_disagreement_score":0.28168532,"about_ca_system_score_codex":0.00011846803,"about_ca_system_score_gemma":0.0009585096,"threshold_uncertainty_score":0.99553704},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4231686601","doi":"10.31222/osf.io/b79mk","title":"Patient-important outcomes reported in clinical studies of pharmacologic treatments for covid-19: A protocol of a meta-epidemiological study","year":2020,"lang":"en","type":"preprint","venue":"","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; McMaster University; Western University; Université de Montréal; University of Saskatchewan","funders":"","keywords":"Medicine; Epidemiology; Randomized controlled trial; Protocol (science); Pandemic; MEDLINE; Coronavirus disease 2019 (COVID-19); Clinical trial; Meta-analysis; Disease; Intensive care medicine; Family medicine; Alternative medicine; Internal medicine; Pathology","score_opus":0.9322724632072774,"score_gpt":0.6826235854805558,"score_spread":0.24964887772672162,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4231686601","genre_codex":"empirical","genre_gemma":"protocol","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.5412831,0.0012222126,0.0012358788,0.030111033,0.0008460513,0.4227562,0.0021329785,0.00012162861,0.00029096712],"genre_scores_gemma":[0.40974975,0.00018625373,0.012044384,0.014878345,0.000114807095,0.56271464,0.000100668265,0.000055674675,0.00015550829],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9635977,0.0027201774,0.03124856,0.0017847486,0.00017774681,0.00047102608],"domain_scores_gemma":[0.9614728,0.011773328,0.025094239,0.0010230195,0.00024050595,0.00039607144],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.057468195,0.0006523783,0.011139581,0.0004676315,0.000075067015,0.000016098775,0.000510406,0.0005077434,0.00045304612],"category_scores_gemma":[0.11364303,0.0005222631,0.0017419159,0.00022717136,0.00024307735,0.00010624616,0.00066299096,0.00050924235,0.000020577763],"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":[0.0003871668,0.0037075973,0.96473676,0.002602861,0.014674811,0.000015811445,0.0021958922,0.0005204486,4.14165e-7,0.0025701209,0.008539013,0.000049090333],"study_design_scores_gemma":[0.033865973,0.014245248,0.66551536,0.00029499657,0.0053867083,0.0000070795654,0.019867042,0.011527396,0.000019205389,0.2292139,0.017517554,0.0025395302],"about_ca_topic_score_codex":0.0016425204,"about_ca_topic_score_gemma":0.0003520169,"teacher_disagreement_score":0.2992214,"about_ca_system_score_codex":0.0008639705,"about_ca_system_score_gemma":0.00096295483,"threshold_uncertainty_score":0.9997229},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"reporting","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"systematic_review","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4248363397","doi":"10.21203/rs.3.rs-71534/v1","title":"Bias Estimation In Study Design: A Meta-Epidemiological Analysis of Transcatheter Versus Surgical Aortic Valve Replacement","year":2020,"lang":"en","type":"preprint","venue":"Research Square (Research Square)","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's College Hospital; University of Toronto; Western University","funders":"University of Toronto; Canadian Institutes of Health Research; University of Ottawa","keywords":"Meta-analysis; Estimation; Epidemiology; Medicine; Aortic valve replacement; Cardiology; Internal medicine; Economics; Stenosis","score_opus":0.8706539368585138,"score_gpt":0.6069626284785796,"score_spread":0.26369130837993415,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4248363397","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8910901,0.01151133,0.027230434,0.043032862,0.0006376836,0.022877958,0.0019478167,0.00020716093,0.0014646737],"genre_scores_gemma":[0.9894631,0.0009070309,0.0031251428,0.000150863,0.00022898098,0.005384768,0.0004894047,0.00010623522,0.00014450887],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.93275815,0.044592243,0.012723032,0.004035383,0.0030724283,0.0028187674],"domain_scores_gemma":[0.937174,0.05452261,0.0026388555,0.003156613,0.0014215336,0.0010863602],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.3012893,0.0008554951,0.007359541,0.008176948,0.000471332,0.00037559064,0.002171248,0.0010061658,0.0030372108],"category_scores_gemma":[0.07081569,0.00086041976,0.003296491,0.0060385824,0.000716357,0.00041654802,0.0015296707,0.0048645954,0.0010096121],"study_design_candidate":"simulation_or_modeling","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":[0.010145034,0.014740066,0.40564802,0.012112945,0.09429481,0.00019345363,0.05528642,0.34617463,0.000008373493,0.039540686,0.02008365,0.0017719114],"study_design_scores_gemma":[0.007307917,0.0086533455,0.18210615,0.0006665495,0.0022875543,0.0000021290475,0.012459744,0.74267167,0.00001686455,0.039520085,0.002557959,0.0017500132],"about_ca_topic_score_codex":0.01068636,"about_ca_topic_score_gemma":0.0012184742,"teacher_disagreement_score":0.39649707,"about_ca_system_score_codex":0.0036522294,"about_ca_system_score_gemma":0.0015618466,"threshold_uncertainty_score":0.9997682},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4293483968","doi":"10.1136/bmjopen-2021-060237","title":"Factors Associated with the Magnitude Of acUpuncture treatment effectS (FAMOUS): a meta-epidemiological study of acupuncture randomised controlled trials","year":2022,"lang":"en","type":"article","venue":"BMJ Open","topic":"Acupuncture Treatment Research Studies","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Western University; York University; Hamilton Health Sciences; University of Toronto; McMaster University; Impact","funders":"China Academy of Chinese Medical Sciences; National Natural Science Foundation of China","keywords":"Medicine; Acupuncture; Meta-analysis; Meta-regression; Randomized controlled trial; MEDLINE; Sample size determination; Internal medicine; Clinical trial; Physical therapy; Alternative medicine; Pathology; Statistics","score_opus":0.3686658038902662,"score_gpt":0.4846991083205923,"score_spread":0.11603330443032611,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4293483968","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8683627,0.0011054141,0.000009793547,0.0030420476,0.0000761755,0.1263892,0.00014585577,0.00004716672,0.00082162983],"genre_scores_gemma":[0.9567671,0.000023820152,0.000060316062,0.0003626813,0.000053450425,0.039990947,0.00015555935,0.00004127572,0.002544864],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9842198,0.011482338,0.0015165725,0.00064248516,0.0016082957,0.00053053803],"domain_scores_gemma":[0.957311,0.039205503,0.0018862289,0.0010225737,0.00039905994,0.00017567932],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.011704207,0.0013663692,0.016575314,0.00017105124,0.00043942267,0.00003766343,0.00071697356,0.00013742686,0.0009792444],"category_scores_gemma":[0.027335167,0.00024967195,0.0027744567,0.0005514708,0.00023523782,0.000070895294,0.0008077302,0.00047239146,0.0000012855905],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.42269647,0.020020725,0.119564004,0.00033165974,0.39620748,0.0006841345,0.0118781105,0.001083106,0.0033235194,0.000058836595,0.02341879,0.00073317246],"study_design_scores_gemma":[0.6197907,0.05193467,0.18700331,0.00012434041,0.12626493,0.000021803962,0.010865787,0.000042930526,0.002619791,0.00009239372,0.00071473554,0.00052465295],"about_ca_topic_score_codex":0.0003942172,"about_ca_topic_score_gemma":0.00036186518,"teacher_disagreement_score":0.26994255,"about_ca_system_score_codex":0.00041195215,"about_ca_system_score_gemma":0.00037956462,"threshold_uncertainty_score":0.9999955},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4318973081","doi":"10.1136/bmjebm-2022-112053","title":"Different meta-analysis methods can change judgements about imprecision of effect estimates: a meta-epidemiological study","year":2023,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Meta-analysis; Statistics; Random effects model; Mathematics; Medicine; Restricted maximum likelihood; Maximum likelihood; Econometrics; Internal medicine","score_opus":0.9820049695252123,"score_gpt":0.7362051535903333,"score_spread":0.24579981593487898,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4318973081","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":[0.000026479369,0.941686,0.035551738,0.0012982978,0.00088342046,0.020254027,0.00019375369,0.000051734714,0.00005459706],"genre_scores_gemma":[0.0002460588,0.96465504,0.011140316,0.0007026578,0.000682681,0.019788336,0.00023435002,0.00014404584,0.002406525],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.6700647,0.24845043,0.05377088,0.006069376,0.020360285,0.001284344],"domain_scores_gemma":[0.6266546,0.3138774,0.040614326,0.015240023,0.0026586838,0.00095495436],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.61449736,0.0036427567,0.15125279,0.006392405,0.00028243288,0.00038423185,0.008078281,0.00077824167,0.035556857],"category_scores_gemma":[0.536687,0.001118133,0.07427885,0.016860403,0.00049318926,0.00029865943,0.0011139681,0.0011487169,0.0009962902],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000035257755,0.00030827217,0.0009351403,0.013056746,0.61072904,0.000087566026,0.00019709743,0.000060264258,5.269549e-7,0.000029277599,0.011220056,0.36334074],"study_design_scores_gemma":[0.00038038517,0.0015729293,0.00061702466,0.005560864,0.8966656,0.000006883195,0.00011040448,0.0014388227,0.000001173906,0.00035358602,0.09257792,0.0007144275],"about_ca_topic_score_codex":0.00064650824,"about_ca_topic_score_gemma":0.00025075715,"teacher_disagreement_score":0.3626263,"about_ca_system_score_codex":0.0002750828,"about_ca_system_score_gemma":0.00035088262,"threshold_uncertainty_score":0.99978155},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4323536717","doi":"10.1101/2023.03.05.23286821","title":"The impact of blinding on trial results: A systematic review and meta-analysis","year":2023,"lang":"en","type":"review","venue":"medRxiv","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; Dalhousie University; McMaster University","funders":"","keywords":"Blinding; Meta-analysis; Publication bias; Medicine; Randomized controlled trial; Systematic review; MEDLINE; Odds ratio; Strictly standardized mean difference; Clinical trial; Internal medicine","score_opus":0.9072607679519703,"score_gpt":0.6315100895610637,"score_spread":0.2757506783909066,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4323536717","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":[0.0000031125228,0.9888146,0.000043855525,0.00015392822,0.00010956384,0.009787385,0.00040601313,0.000009392403,0.0006721775],"genre_scores_gemma":[0.00001846364,0.98282075,0.000035280373,0.000035370846,0.00005168383,0.0011583666,0.00002921737,0.000043308493,0.01580757],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.87164116,0.07037715,0.04652056,0.0022137216,0.0087341545,0.0005132481],"domain_scores_gemma":[0.8595928,0.0776829,0.047146555,0.014051641,0.001235917,0.00029019464],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.40791047,0.0011461528,0.07879013,0.0018251481,0.00027562928,0.001193437,0.0047813533,0.0002532954,0.0013909861],"category_scores_gemma":[0.2663695,0.00029781068,0.068661146,0.0128931515,0.000122824,0.000100267745,0.00032153557,0.0004808673,0.002618606],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003719983,0.000032170625,0.0000012894444,0.45816323,0.5294082,0.0000066151247,0.000034369816,0.0000046770624,2.700259e-9,0.00018232579,0.010258784,0.0018711191],"study_design_scores_gemma":[0.00038616787,0.0001024364,9.821096e-7,0.021692043,0.9199297,0.0000068936565,0.000021085196,0.000038228973,7.3902227e-9,0.00011766155,0.05745814,0.00024668433],"about_ca_topic_score_codex":0.000020115338,"about_ca_topic_score_gemma":0.000021413392,"teacher_disagreement_score":0.43647116,"about_ca_system_score_codex":0.0000754468,"about_ca_system_score_gemma":0.00021914754,"threshold_uncertainty_score":0.9999474},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4377010956","doi":"10.1093/fampra/cmad052","title":"The quality of paediatric asthma guidelines: evidence underpinning diagnostic test recommendations from a meta-epidemiological study","year":2023,"lang":"en","type":"article","venue":"Family Practice","topic":"Asthma and respiratory diseases","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":false,"ca_institutions":"Hospital for Sick Children; University of Toronto","funders":"U.S. Department of Health and Human Services; National Institutes of Health; Canadian Institutes of Health Research; National Institute for Health and Care Research; Canadian Thoracic Society; National Institute for Health and Care Excellence","keywords":"Medicine; Asthma; Spirometry; Epidemiology; Family medicine; Test (biology); Diagnostic test; Pediatrics; MEDLINE; Primary care; Intensive care medicine; Pathology","score_opus":0.41210880923528265,"score_gpt":0.4964183921697496,"score_spread":0.08430958293446694,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4377010956","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.85918283,0.04594523,0.0017492761,0.08799274,0.0008295893,0.002418165,0.00020829757,0.0004799361,0.0011939403],"genre_scores_gemma":[0.9859086,0.0056339023,0.0018840317,0.0055646906,0.0004492987,0.00025144056,0.000042083797,0.000024151443,0.00024176556],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.995338,0.002155445,0.0012148243,0.000426024,0.0005779564,0.00028772806],"domain_scores_gemma":[0.7961112,0.20139988,0.0007823939,0.00074362353,0.0007741653,0.0001887263],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0066764285,0.00021914359,0.00067895063,0.00008920669,0.00033874513,0.000037353755,0.00024611285,0.00008635356,0.00007176868],"category_scores_gemma":[0.46437365,0.000120750316,0.00030498617,0.0009306925,0.0001190429,0.0003205961,0.00019165722,0.0003530454,0.00021449462],"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":[0.0010370984,0.003770191,0.66877675,0.00014791617,0.0035239805,0.00039878825,0.0010983183,0.00051203085,0.0012099136,0.00031732302,0.23406507,0.08514264],"study_design_scores_gemma":[0.0007430311,0.0008585611,0.8412705,0.00010908656,0.0027779655,0.0000044503176,0.024356302,0.00031853298,0.000007932683,0.0002111304,0.12916364,0.0001788539],"about_ca_topic_score_codex":0.0006498115,"about_ca_topic_score_gemma":0.00003388775,"teacher_disagreement_score":0.4576972,"about_ca_system_score_codex":0.00004499277,"about_ca_system_score_gemma":0.00027266843,"threshold_uncertainty_score":0.5401382},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"evaluation","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4379135156","doi":"10.1016/j.mayocpiqo.2023.04.010","title":"Inadequate Reporting of Cointerventions, Other Methodological Factors, and Treatment Estimates in Cardiovascular Trials: A Meta-Epidemiological Study","year":2023,"lang":"en","type":"article","venue":"Mayo Clinic Proceedings Innovations Quality & Outcomes","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"Arts and Humanities Research Council; University of California, San Francisco; University of Bern; University of Toronto; Inselspital, Universitätsspital Bern; Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung; National Science Foundation","keywords":"Blinding; Medicine; Meta-analysis; Publication bias; Psychological intervention; Clinical trial; Odds ratio; MEDLINE; Confidence interval; Relative risk; Physical therapy; Internal medicine","score_opus":0.9823887722089711,"score_gpt":0.7142536049889211,"score_spread":0.26813516722005004,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4379135156","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99176973,0.00048444953,0.003265051,0.0018452551,0.00013754098,0.0020644285,0.000044035733,0.00004615335,0.000343367],"genre_scores_gemma":[0.9857632,0.000028745393,0.01228019,0.000114421586,0.000027358365,0.00045987085,0.000014734074,0.0000148544905,0.0012966142],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.8958231,0.027090214,0.07028135,0.0022630505,0.003983977,0.0005582916],"domain_scores_gemma":[0.59405535,0.2770793,0.11532039,0.006680341,0.006463333,0.00040126767],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.69555485,0.0005639153,0.019823296,0.0011791703,0.00018117756,0.00031567237,0.00080498535,0.00019829988,0.0013661773],"category_scores_gemma":[0.80590636,0.00022320171,0.007222652,0.0049048187,0.00017663784,0.000258474,0.0003129454,0.0002728556,0.00008845781],"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":[0.000012275148,0.00033101722,0.98253196,0.000056808338,0.006286945,0.0000032105197,0.0011225682,0.000014821734,0.000016476455,0.004423933,0.00057564647,0.004624349],"study_design_scores_gemma":[0.0006537297,0.0002295922,0.96037465,0.00002654955,0.0027110826,0.0000029122423,0.010227518,0.0003075387,0.00002384038,0.023450352,0.0017830133,0.00020921859],"about_ca_topic_score_codex":0.0004386644,"about_ca_topic_score_gemma":0.00007669386,"teacher_disagreement_score":0.30176777,"about_ca_system_score_codex":0.00006975272,"about_ca_system_score_gemma":0.00005852457,"threshold_uncertainty_score":0.9995467},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4385287363","doi":"10.1136/bmjopen-2023-073232","title":"Evaluating the impact of including non-randomised studies of interventions in meta-analysis of randomised controlled trials: a protocol for a meta-epidemiological study","year":2023,"lang":"en","type":"article","venue":"BMJ Open","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; Impact","funders":"National Science Fund for Distinguished Young Scholars; Canadian Institutes of Health Research; National Natural Science Foundation of China","keywords":"Medicine; Meta-analysis; Randomized controlled trial; Concordance; Systematic review; Psychological intervention; MEDLINE; Publication bias; Population; Epidemiology; Meta-regression; Internal medicine; Environmental health; Psychiatry","score_opus":0.9906337942003213,"score_gpt":0.8021268322848917,"score_spread":0.1885069619154296,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385287363","genre_codex":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07209261,0.00033535945,0.0034725934,0.00038318362,0.00002675472,0.9232632,0.00010014535,0.000003026766,0.0003231299],"genre_scores_gemma":[0.33525762,0.0000013344396,0.0015919514,0.00001161208,0.000009405894,0.66247326,0.0000032177938,0.0000067507226,0.00064484077],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.49953806,0.38294578,0.10585505,0.002235148,0.008728497,0.00069748226],"domain_scores_gemma":[0.37636226,0.53663,0.07500934,0.0060602077,0.005812945,0.0001252717],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.8671608,0.00047540173,0.06762571,0.0014916583,0.00014939735,0.00022385143,0.0031934313,0.00007207707,0.00594781],"category_scores_gemma":[0.7753061,0.0001263651,0.07732247,0.006045366,0.00012571561,0.00019206175,0.0009487888,0.00011793766,0.000012060439],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.031366702,0.0006140629,0.0027126458,0.0002599083,0.92701894,8.9756946e-7,0.002865176,0.03174194,0.00018052977,0.000057852412,0.002826178,0.00035519424],"study_design_scores_gemma":[0.15321234,0.0011576987,0.0036492446,0.00007701236,0.6824954,2.5788577e-7,0.006997047,0.14884518,0.00002096849,0.0033931765,0.0000073993847,0.00014426452],"about_ca_topic_score_codex":0.00024671905,"about_ca_topic_score_gemma":0.00028053293,"teacher_disagreement_score":0.26316503,"about_ca_system_score_codex":0.000034793942,"about_ca_system_score_gemma":0.00020335925,"threshold_uncertainty_score":0.9949609},"labels":[{"model":"gemma","categories":["metaepi_broad"],"domain":null,"study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4387222396","doi":"10.1007/s40279-023-01941-x","title":"Certainty of Evidence Assessment in Systematic Reviews Published by High-Impact Sports Science Journals: A Meta-epidemiological Study","year":2023,"lang":"en","type":"article","venue":"Sports Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; University of Saskatchewan","funders":"","keywords":"Certainty; Systematic review; Grading (engineering); Meta-analysis; Publication bias; Quality of evidence; Medicine; Citation; Evidence-based medicine; Sports medicine; MEDLINE; Bibliometrics; Psychology; Alternative medicine; Computer science; Library science; Political science; Pathology; Mathematics; Engineering","score_opus":0.9041380651400333,"score_gpt":0.6472378102149855,"score_spread":0.25690025492504776,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387222396","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9399861,0.031630266,0.0018551997,0.011156284,0.0012741947,0.012554677,0.000012697115,0.000037663263,0.001492888],"genre_scores_gemma":[0.9953136,0.0013644277,0.00047071849,0.00040399126,0.00008522618,0.00036579967,0.000006772526,0.000014234682,0.001975269],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.82101566,0.05519967,0.07288496,0.0045948066,0.044542216,0.0017626924],"domain_scores_gemma":[0.92449343,0.028250879,0.032926127,0.009689182,0.0036386007,0.0010018098],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.76942444,0.0007368456,0.022291487,0.0025554616,0.00018056037,0.000370945,0.003802448,0.00015020264,0.025205199],"category_scores_gemma":[0.4696264,0.00022056364,0.002551777,0.014987709,0.0005052719,0.00092559913,0.00036341007,0.0007274688,0.0001641498],"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":[0.000020124171,0.0004272578,0.7546117,0.0035396141,0.0008684395,0.00032885684,0.001598331,0.00042197967,0.00026325372,0.00031967083,0.23303339,0.0045674234],"study_design_scores_gemma":[0.00055616925,0.00052759406,0.96600616,0.011610699,0.0043717646,0.00005875682,0.006218756,0.0032703455,0.0000042598563,0.0051362785,0.00183479,0.0004044038],"about_ca_topic_score_codex":0.00029586887,"about_ca_topic_score_gemma":0.000037344307,"teacher_disagreement_score":0.299798,"about_ca_system_score_codex":0.00021408474,"about_ca_system_score_gemma":0.00038201408,"threshold_uncertainty_score":0.9922236},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4389687393","doi":"10.23736/s1973-9087.23.08310-7","title":"The influence of bias in randomized controlled trials on rehabilitation intervention effect estimates: what we have learned from meta-epidemiological studies","year":2023,"lang":"en","type":"article","venue":"European Journal of Physical and Rehabilitation Medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"Ministero della Salute","keywords":"Blinding; Rehabilitation; Randomized controlled trial; Medicine; Meta-analysis; Publication bias; Physical therapy; Physical medicine and rehabilitation; Epidemiology; Intervention (counseling); Reporting bias; Selection bias; MEDLINE; Clinical psychology; Psychiatry; Surgery; Pathology","score_opus":0.48881372665186773,"score_gpt":0.5026068912095041,"score_spread":0.01379316455763635,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389687393","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.86860156,0.013306337,0.0001751897,0.11636851,0.00032201764,0.0011658147,0.000008188365,0.000012080934,0.000040283026],"genre_scores_gemma":[0.9946722,0.0039491,0.00045137614,0.00045961828,0.0003250103,0.00008655893,0.00000588585,0.000014430861,0.000035854926],"study_design_codex":"qualitative","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9593482,0.030014083,0.00987702,0.00033090578,0.00023994176,0.00018985532],"domain_scores_gemma":[0.664882,0.32849863,0.0061327764,0.00018654726,0.00022079177,0.00007924246],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27631813,0.00022398622,0.0072612506,0.00045721544,0.00009966029,0.00003498687,0.00016638059,0.00003942669,0.000018357883],"category_scores_gemma":[0.5861516,0.00010930871,0.0011875052,0.00023229395,0.00078105956,0.00038897272,0.000034233173,0.0002679585,0.00004368811],"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":[0.2578515,0.0021464126,0.030150624,0.006569513,0.028482502,0.00003708972,0.2988199,0.12060505,0.0009844317,0.15992483,0.02009472,0.0743334],"study_design_scores_gemma":[0.2749509,0.014288981,0.12799107,0.008159671,0.0014440023,0.0000026381992,0.037401583,0.01963064,0.000011772591,0.51463103,0.0011005972,0.00038712614],"about_ca_topic_score_codex":0.000056889778,"about_ca_topic_score_gemma":0.000007919981,"teacher_disagreement_score":0.35470617,"about_ca_system_score_codex":0.00009108092,"about_ca_system_score_gemma":0.00001808662,"threshold_uncertainty_score":0.74518275},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W4390765507","doi":"10.1007/s11655-023-3617-0","title":"Factors and Their Impact on Treatment Effect of Acupuncture in Different Outcomes: A Meta-Regression of Acupuncture Randomized Controlled Trials","year":2024,"lang":"en","type":"review","venue":"Chinese Journal of Integrative Medicine","topic":"Acupuncture Treatment Research Studies","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Acupuncture; Medicine; Randomized controlled trial; Meta-analysis; Confidence interval; Strictly standardized mean difference; Physical therapy; Meta-regression; MEDLINE; Cochrane Library; Internal medicine; Alternative medicine; Pathology","score_opus":0.1108893715685012,"score_gpt":0.4947666850293225,"score_spread":0.38387731346082127,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390765507","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":[0.038671684,0.94813836,0.0000408235,0.0009111289,0.0005644429,0.011350635,0.00021972683,0.000013818818,0.000089366826],"genre_scores_gemma":[0.119371854,0.8791373,0.000011868606,0.000019055677,0.00043162005,0.0004244846,0.00011966518,0.000085020765,0.0003991417],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.98448133,0.0070435554,0.0060212626,0.00057824235,0.0014383043,0.00043732708],"domain_scores_gemma":[0.8864808,0.10589075,0.005991019,0.0005667273,0.0006677807,0.00040293732],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.007940953,0.007900817,0.17223386,0.0030219208,0.000045583933,0.000019513052,0.00032012575,0.00057940214,0.0002971027],"category_scores_gemma":[0.075487554,0.0005474354,0.039196987,0.00077919324,0.0008686429,0.00007796018,0.00010349656,0.0019400693,8.1701603e-7],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.3032271,0.0014607923,0.0029396596,0.030387307,0.57013774,0.0006612977,0.009363643,0.0000029338307,0.00017567452,0.000017281996,0.0018881458,0.079738416],"study_design_scores_gemma":[0.3964468,0.03242037,0.0006317933,0.14493908,0.42135826,0.00031210997,0.00074588857,0.000016449665,0.00035702568,0.00026652162,0.00199248,0.00051324285],"about_ca_topic_score_codex":0.000070311144,"about_ca_topic_score_gemma":0.000030395584,"teacher_disagreement_score":0.1487795,"about_ca_system_score_codex":0.001114789,"about_ca_system_score_gemma":0.0004977192,"threshold_uncertainty_score":0.99969774},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad"],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4391147924","doi":"10.2139/ssrn.4696882","title":"Hierarchy of Patient-Reported Outcome Measures in Randomized Controlled Trials for Chronic Non-Specific Low Back Pain: A Meta-Epidemiological Study – Study Protocol","year":2024,"lang":"en","type":"preprint","venue":"SSRN Electronic Journal","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"","keywords":"Epidemiology; Randomized controlled trial; Medicine; Protocol (science); Meta-analysis; Outcome (game theory); Physical therapy; Chronic pain; Hierarchy; Low back pain; Alternative medicine; Internal medicine; Pathology","score_opus":0.11050970728391607,"score_gpt":0.4062562167360266,"score_spread":0.2957465094521105,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391147924","genre_codex":"protocol","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"protocol","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2811971,0.0074660066,0.0051264954,0.0012616365,0.000637521,0.70413554,0.000009490609,0.00006379592,0.00010242589],"genre_scores_gemma":[0.6822555,0.00029458778,0.00019311173,0.000040994488,0.0004910061,0.31631693,0.000016757089,0.00007583412,0.0003152733],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9528233,0.028846005,0.012827889,0.0013928945,0.0016300326,0.002479917],"domain_scores_gemma":[0.97910064,0.013912956,0.0049490836,0.0008963412,0.00090932223,0.00023167398],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.24487555,0.0011436358,0.021195961,0.0012301417,0.00011224088,0.00008593275,0.00042232766,0.0005984028,0.00017034277],"category_scores_gemma":[0.04397351,0.00055888185,0.010240982,0.00042803504,0.00023471964,0.00005630846,0.00025849254,0.0067541827,0.000007870659],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.79975927,0.0178541,0.008818156,0.008643297,0.13608317,0.000094773044,0.0033027541,0.0015604587,0.00067282084,0.0012945106,0.00028598343,0.021630691],"study_design_scores_gemma":[0.8529336,0.026295723,0.0018386712,0.0015666847,0.0146173965,0.000048835806,0.0055136657,0.0018409858,0.000013384851,0.0944689,0.000108717184,0.00075344834],"about_ca_topic_score_codex":0.00011864199,"about_ca_topic_score_gemma":0.00044955482,"teacher_disagreement_score":0.4010584,"about_ca_system_score_codex":0.0024915196,"about_ca_system_score_gemma":0.005347149,"threshold_uncertainty_score":0.99968624},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"not_applicable","genre":"protocol","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4392604238","doi":"10.1101/2024.03.06.24303892","title":"Publication bias in pharmacogenetics of statin-associated muscle symptoms, an umbrella review with a meta-epidemiological study","year":2024,"lang":"en","type":"review","venue":"medRxiv","topic":"Lipoproteins and Cardiovascular Health","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Centre National de la Recherche Scientifique","keywords":"Funnel plot; SLCO1B1; Publication bias; Statin; Meta-analysis; Medicine; Odds ratio; Internal medicine; Pharmacogenetics; Pharmacology; Genotype; Genetics; Biology","score_opus":0.28214524288905907,"score_gpt":0.446024021149999,"score_spread":0.16387877826093994,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392604238","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":[0.00442172,0.989319,0.000008324915,0.00018713214,0.00007425309,0.005708708,0.00006928199,0.00007752977,0.00013401391],"genre_scores_gemma":[0.0012093995,0.9964302,0.00015021533,0.00037794167,0.00010784372,0.0011391884,0.0003615931,0.00009132605,0.00013228442],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99272776,0.0032315275,0.0019596557,0.00083472085,0.00085290195,0.00039343484],"domain_scores_gemma":[0.9971945,0.00023963994,0.00083188683,0.0011249154,0.00034486546,0.00026414552],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.010177835,0.00060285244,0.006521556,0.00030221918,0.000029516528,0.000018882587,0.0002902698,0.00025763173,0.00014530399],"category_scores_gemma":[0.0014082058,0.00030857197,0.0013312773,0.0016941173,0.0000633499,0.000044005486,0.00008745251,0.0008660162,0.0000339825],"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":[0.000012261172,0.0014180273,0.0020390386,0.13668312,0.00996319,0.00018209053,0.00008143046,6.010021e-7,4.953041e-7,0.000020551483,0.00065751624,0.8489417],"study_design_scores_gemma":[0.00089842506,0.0015843911,0.0039322483,0.043617785,0.09362678,0.00008617698,0.000026617332,0.000016427268,5.882396e-7,0.000006495144,0.8557896,0.00041444684],"about_ca_topic_score_codex":0.0002053888,"about_ca_topic_score_gemma":0.000073624105,"teacher_disagreement_score":0.8551321,"about_ca_system_score_codex":0.00017350467,"about_ca_system_score_gemma":0.00069929135,"threshold_uncertainty_score":0.99993664},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad","research_integrity"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4392947957","doi":"10.1371/journal.pone.0298526","title":"Do statistical heterogeneity methods impact the results of meta- analyses? A meta epidemiological study","year":2024,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Meta-analysis and systematic reviews","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":"University of Alberta","funders":"","keywords":"Meta-analysis; Random effects model; Statistics; Confidence interval; Study heterogeneity; Pooling; Medicine; Restricted maximum likelihood; Interquartile range; Mathematics; Sample size determination; Maximum likelihood; Internal medicine; Computer science; Artificial intelligence","score_opus":0.9732907961212317,"score_gpt":0.7002165803757118,"score_spread":0.27307421574551993,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392947957","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7365089,0.053556927,0.19787349,0.0028615026,0.0001657592,0.004153626,0.0014895914,0.000044869153,0.0033453312],"genre_scores_gemma":[0.8979655,0.000052873314,0.10065769,0.000101508725,0.0000600315,0.00016877118,0.000009448489,0.000012564684,0.00097165204],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.8029982,0.16356838,0.019592816,0.002622323,0.010617591,0.000600726],"domain_scores_gemma":[0.8826216,0.10571333,0.003455351,0.0067096455,0.0012013726,0.0002987238],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.39115655,0.0004932888,0.012487937,0.0003362551,0.00011756243,0.0007191339,0.0021203363,0.00009560939,0.015070885],"category_scores_gemma":[0.22802009,0.00012816428,0.0074864044,0.00235438,0.00017023536,0.0001497275,0.00033861885,0.0003621902,0.0008332206],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00021763718,0.00871711,0.014741696,0.0003257605,0.9072847,0.00011209065,0.0018351075,0.0008168663,0.0042445613,0.0057233544,0.04495695,0.011024192],"study_design_scores_gemma":[0.00069275947,0.0023834093,0.052800693,0.00007469857,0.7296615,0.000023392475,0.0011784122,0.14069305,0.0029400538,0.06205536,0.006654484,0.0008421561],"about_ca_topic_score_codex":0.00010492153,"about_ca_topic_score_gemma":0.0000514527,"teacher_disagreement_score":0.17762315,"about_ca_system_score_codex":0.000029544219,"about_ca_system_score_gemma":0.000064617685,"threshold_uncertainty_score":0.99994475},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4399706167","doi":"10.1016/j.jval.2024.03.2436","title":"PT22 Consistency Between Comparative Efficacy Estimates from Target Trial Emulations and Their Origin Target Trials: A Meta-Epidemiological Study","year":2024,"lang":"en","type":"article","venue":"Value in Health","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Cascades (Canada); University of Toronto; McMaster University","funders":"","keywords":"Consistency (knowledge bases); Epidemiology; Meta-analysis; Medicine; Computer science; Artificial intelligence; Internal medicine","score_opus":0.8587779294536831,"score_gpt":0.5553820052404167,"score_spread":0.30339592421326644,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4399706167","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.895419,0.05088916,0.008739827,0.03279048,0.0014644269,0.006842552,0.0033174318,0.00021426592,0.0003228446],"genre_scores_gemma":[0.98237634,0.00021062329,0.014163466,0.0015312701,0.0008761144,0.0005252817,0.00023298764,0.00003711964,0.00004678574],"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.980971,0.007441257,0.009615462,0.0012237977,0.00012071989,0.0006277608],"domain_scores_gemma":[0.9430918,0.0535888,0.0023258235,0.0005660908,0.000045939687,0.0003815511],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07689719,0.00045948406,0.0059252903,0.00056109397,0.00032681212,0.00018183723,0.00027372432,0.00021394233,0.00056017796],"category_scores_gemma":[0.017929804,0.00039092588,0.00042316903,0.0004511617,0.00014035654,0.00036122825,0.000081132224,0.00054106215,0.0003754278],"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":[0.0020985182,0.0030180505,0.55668366,0.0014897777,0.008946856,0.000017146514,0.07890541,0.043765772,0.0000014462406,0.27696255,0.027427321,0.0006834951],"study_design_scores_gemma":[0.0277661,0.0024872741,0.15934792,0.00044425082,0.00031553768,0.0000043453,0.007610914,0.31410107,0.0000026650148,0.4365639,0.050061826,0.0012942067],"about_ca_topic_score_codex":0.014017604,"about_ca_topic_score_gemma":0.00014861539,"teacher_disagreement_score":0.39733574,"about_ca_system_score_codex":0.0005914538,"about_ca_system_score_gemma":0.0005807551,"threshold_uncertainty_score":0.99985427},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4400747098","doi":"10.1111/dom.15771","title":"Bariatric surgery and all‐cause mortality: A methodological review of studies using a non‐surgical comparator","year":2024,"lang":"en","type":"review","venue":"Diabetes Obesity and Metabolism","topic":"Bariatric Surgery and Outcomes","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"Atara Biotherapeutics; Kowa Company; Amarin Corporation; Novo Nordisk; Merck; Novartis; Pfizer; Seqirus; AstraZeneca; National Institute of Diabetes and Digestive and Kidney Diseases; Boehringer Ingelheim","keywords":"Medicine; Surgical procedures; Surgery; General surgery; Intensive care medicine","score_opus":0.36883434458384223,"score_gpt":0.46638567469115594,"score_spread":0.0975513301073137,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400747098","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":[0.022854613,0.97496784,0.0000036128647,0.0000850945,0.00092251203,0.0010065864,0.00007413575,0.000052982406,0.000032611166],"genre_scores_gemma":[0.00013654317,0.99794024,0.00070053077,0.00041359832,0.00058146927,0.000062824765,0.00005732454,0.00004856275,0.000058893056],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9952779,0.0013001899,0.0016497994,0.00081837195,0.0004302893,0.00052347145],"domain_scores_gemma":[0.9942502,0.0041303122,0.0005954749,0.0004972758,0.00016886176,0.00035791166],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004993422,0.00077567116,0.010345721,0.0005332898,0.00009353479,0.00003601067,0.00011910316,0.0005100039,0.000034848814],"category_scores_gemma":[0.0016671817,0.00048118632,0.0014586904,0.0010297686,0.00031410722,0.000109120876,0.00038119493,0.000747418,0.00001012785],"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":[0.0000144487285,0.0002462951,0.0051226798,0.85503584,0.015580734,0.00045896583,0.00008014446,2.582239e-8,0.0000018331529,0.00069287507,0.002320412,0.12044573],"study_design_scores_gemma":[0.00031239775,0.0000290902,0.002399815,0.11048966,0.12531154,0.00031407084,0.000018249722,0.000008289128,0.0000051365373,0.00015871186,0.7603327,0.00062032335],"about_ca_topic_score_codex":0.000012616016,"about_ca_topic_score_gemma":4.100053e-7,"teacher_disagreement_score":0.7580123,"about_ca_system_score_codex":0.000034884386,"about_ca_system_score_gemma":0.00033375373,"threshold_uncertainty_score":0.99976397},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4403591944","doi":"10.1016/j.atherosclerosis.2024.118624","title":"Publication bias in pharmacogenetics of statin-associated muscle symptoms: A meta-epidemiological study","year":2024,"lang":"en","type":"review","venue":"Atherosclerosis","topic":"Lipoproteins and Cardiovascular Health","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"Association Nationale de la Recherche et de la Technologie","keywords":"Pharmacogenetics; Medicine; Statin; Epidemiology; Meta-analysis; Hydroxymethylglutaryl-CoA Reductase Inhibitors; Publication bias; Internal medicine; Pharmacology; Genetics; Genotype; Biology","score_opus":0.3292323609253614,"score_gpt":0.4284963298699932,"score_spread":0.0992639689446318,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403591944","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":[0.013112382,0.98219115,0.000011101438,0.0000875447,0.00014805369,0.004044984,0.000182017,0.00010345751,0.000119329816],"genre_scores_gemma":[0.012237863,0.985398,0.00013202614,0.00015199656,0.000117614436,0.0014673298,0.00019815456,0.00011197392,0.00018502762],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9945064,0.0016324772,0.0019376845,0.00078079815,0.000698766,0.00044385708],"domain_scores_gemma":[0.99762446,0.0003505246,0.0007243029,0.000898396,0.00021504617,0.00018728433],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0049328455,0.00061479374,0.0054194788,0.00036550994,0.000038915365,0.00003431655,0.00026340445,0.00043875046,0.00017151005],"category_scores_gemma":[0.0011257209,0.00036741624,0.0026371882,0.0014592651,0.00006915321,0.000042555926,0.00014918965,0.00074520765,0.00007515602],"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":[0.00000835024,0.0020007407,0.0015406238,0.012320447,0.015207515,0.000012883946,0.00037449607,0.0000012325121,0.000007282215,0.000031816904,0.00086314813,0.96763146],"study_design_scores_gemma":[0.0026348087,0.0019194708,0.02248095,0.013602552,0.12497164,0.000016744701,0.00020526031,0.00005645525,0.00000636995,0.000041767995,0.8331925,0.00087143184],"about_ca_topic_score_codex":0.0005224644,"about_ca_topic_score_gemma":0.000045276876,"teacher_disagreement_score":0.96676004,"about_ca_system_score_codex":0.0003293108,"about_ca_system_score_gemma":0.00040649294,"threshold_uncertainty_score":0.99987775},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4404923291","doi":"10.1186/s12916-024-03778-1","title":"Integration of non-randomized studies with randomized controlled trials in meta-analyses of clinical studies: a meta-epidemiological study on effect estimation of interventions","year":2024,"lang":"en","type":"review","venue":"BMC Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Impact","funders":"West China Hospital, Sichuan University; Sichuan Provincial Administration of Traditional Chinese Medicine; National Science Fund for Distinguished Young Scholars; Sichuan University; National Natural Science Foundation of China","keywords":"Medicine; Randomized controlled trial; Meta-analysis; Systematic review; MEDLINE; Concordance; Psychological intervention; Internal medicine; Psychiatry","score_opus":0.9831690436353597,"score_gpt":0.7817473869103255,"score_spread":0.2014216567250342,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404923291","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":[0.00044920848,0.93350893,0.030788105,0.00014165306,0.00056055177,0.034362707,0.00003249281,0.0000062544214,0.00015007246],"genre_scores_gemma":[0.01309245,0.96739185,0.005331669,0.00002449246,0.00014046664,0.012934863,0.000037657133,0.0000326177,0.0010139219],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.11776625,0.6848536,0.18629107,0.0024532552,0.00831945,0.0003163455],"domain_scores_gemma":[0.03587907,0.84003437,0.11635516,0.003933676,0.0036662559,0.00013149586],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.9116199,0.0017174792,0.44267595,0.0035870576,0.0000430413,0.000049076134,0.0014924826,0.00034291274,0.0012202694],"category_scores_gemma":[0.963791,0.0003585033,0.13212284,0.0035188468,0.0018170602,0.00010217241,0.00021424235,0.0007100983,0.000045878605],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.12182882,0.00074278447,0.000011736233,0.031246133,0.818803,0.0000078472,0.0010757729,0.00032265869,3.9481262e-7,0.00049723225,0.002194961,0.023268677],"study_design_scores_gemma":[0.28266245,0.0015426088,0.0000028182503,0.021653412,0.68877435,0.0000021296833,0.0018253593,0.0020979699,5.2374526e-7,0.001122058,0.00015756107,0.00015877954],"about_ca_topic_score_codex":0.000055693945,"about_ca_topic_score_gemma":0.00011224216,"teacher_disagreement_score":0.31055313,"about_ca_system_score_codex":0.000056023087,"about_ca_system_score_gemma":0.00026120455,"threshold_uncertainty_score":0.9998867},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4405223579","doi":"10.1093/jnci/djae318","title":"Justification, margin values, and analysis populations for oncologic noninferiority and equivalence trials: a meta-epidemiological study","year":2024,"lang":"en","type":"article","venue":"JNCI Journal of the National Cancer Institute","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"London Health Sciences Centre; Western University","funders":"National Cancer Institute; University of Texas MD Anderson Cancer Center; National Institutes of Health; Andrew Sabin Family Foundation","keywords":"Margin (machine learning); Equivalence (formal languages); Clinical endpoint; Meta-analysis; Population; Medicine; Statistics; Clinical trial; Mathematics; Computer science; Internal medicine","score_opus":0.9378516184211011,"score_gpt":0.7058903326026658,"score_spread":0.23196128581843534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4405223579","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":[0.20692088,0.008233471,0.73804027,0.02862268,0.011824166,0.003979298,0.002065611,0.00008925046,0.00022439107],"genre_scores_gemma":[0.50676787,0.00035752295,0.49063343,0.00033950646,0.0013994066,0.0002090844,0.0000021281658,0.000012832541,0.00027822892],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.9950238,0.0019754404,0.0018881058,0.00029956,0.00066420186,0.00014887868],"domain_scores_gemma":[0.948314,0.04958966,0.0010089622,0.00015083546,0.0008272994,0.00010925217],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02820563,0.00018845056,0.0015118021,0.00021188057,0.0002371892,0.0001092648,0.0002874849,0.00012366392,0.00009834002],"category_scores_gemma":[0.23235579,0.000095346426,0.0006980029,0.000645889,0.0002725653,0.00021259584,0.000101433216,0.00037312336,4.8961044e-7],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011814173,0.001257664,0.024005197,0.00075286964,0.037470937,0.000024761623,0.0009520547,0.00583053,0.00034941448,0.859831,0.035056915,0.03328726],"study_design_scores_gemma":[0.000667743,0.0001662181,0.03853773,0.000056607558,0.020113777,0.000018804196,0.000039250554,0.0040590954,0.000014529057,0.934044,0.0021628907,0.00011933185],"about_ca_topic_score_codex":0.000019628415,"about_ca_topic_score_gemma":0.000056800603,"teacher_disagreement_score":0.29984698,"about_ca_system_score_codex":0.00022346996,"about_ca_system_score_gemma":0.0003412136,"threshold_uncertainty_score":0.97755647},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4409203502","doi":"10.1016/j.drugalcdep.2025.112665","title":"Covariates in studies examining longitudinal relationships between substance use and mental health problems among youth: A meta-epidemiologic review","year":2025,"lang":"en","type":"review","venue":"Drug and Alcohol Dependence","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Hospitalier Universitaire Sainte-Justine; McMaster University; Université de Montréal; St. Joseph’s Healthcare Hamilton","funders":"Canadian Institutes of Health Research","keywords":"Mental health; Substance use; Meta-analysis; Psychology; Longitudinal study; Covariate; Clinical psychology; Psychiatry; Medicine","score_opus":0.517409130212774,"score_gpt":0.4333312985156157,"score_spread":0.08407783169715832,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409203502","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":[0.0077915746,0.98894024,0.000007712839,0.00034536148,0.00008705921,0.0026597693,0.000073417585,0.00006863113,0.00002622085],"genre_scores_gemma":[0.010043043,0.98787075,0.0005537714,0.00015763409,0.000040418825,0.0003439849,0.00035406934,0.00002908105,0.00060726947],"study_design_codex":"observational","study_design_gemma":"systematic_review","domain_scores_codex":[0.99586153,0.0009961858,0.001391504,0.0009719021,0.00029191517,0.00048698188],"domain_scores_gemma":[0.9949621,0.0036333005,0.000723925,0.00041712343,0.00007677677,0.0001867851],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.003045928,0.00071542064,0.0054723313,0.00030401492,0.00027134886,0.000042622807,0.00014379265,0.0001971891,0.000009022882],"category_scores_gemma":[0.0011919076,0.00045641503,0.0003281529,0.0006635414,0.000218541,0.00035163705,0.000104800034,0.00083564466,0.000004310232],"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":[0.000009862072,0.00004700393,0.8996386,0.05444982,0.0029136308,0.000068381756,0.0026003425,3.8855083e-7,1.9058055e-8,0.00009023936,0.00031822335,0.039863523],"study_design_scores_gemma":[0.0026446723,0.00022210593,0.18919733,0.7589585,0.03388461,0.00015490706,0.0030598876,0.000004483285,5.21929e-7,0.0002857238,0.010190347,0.0013968995],"about_ca_topic_score_codex":0.00013406352,"about_ca_topic_score_gemma":0.0012521837,"teacher_disagreement_score":0.71044123,"about_ca_system_score_codex":0.00020959569,"about_ca_system_score_gemma":0.000180311,"threshold_uncertainty_score":0.99978876},"labels":[{"model":"gemma","categories":["metaepi_broad","metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4411977700","doi":"10.1186/s12916-025-04202-y","title":"Sex and gender considerations in randomized controlled trials in critical care nephrology: a meta-epidemiologic study","year":2025,"lang":"en","type":"review","venue":"BMC Medicine","topic":"Sex and Gender in Healthcare","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":"Alberta Health Services; University of Alberta","funders":"University of Alberta; King Chulalongkorn Memorial Hospital","keywords":"Medicine; Randomized controlled trial; Nephrology; Meta-analysis; Internal medicine; MEDLINE; Intensive care medicine","score_opus":0.6544556132467321,"score_gpt":0.5824175902738793,"score_spread":0.07203802297285278,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411977700","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":[0.000049740374,0.9721523,0.0000907993,0.002546267,0.0005282955,0.023514288,0.000028495386,0.000052750023,0.0010370612],"genre_scores_gemma":[0.004955468,0.9807958,0.0005864268,0.0037281867,0.00057388627,0.00893557,0.00014816191,0.000034602726,0.0002418832],"study_design_codex":"systematic_review","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9465253,0.04347046,0.0077130366,0.0012617455,0.00044751988,0.0005819335],"domain_scores_gemma":[0.7879894,0.21002765,0.00072814326,0.0007527269,0.00024339995,0.000258676],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.058696564,0.0008731958,0.057468906,0.0015920722,0.00008804581,0.000013746556,0.00012394844,0.00083593535,0.0006278006],"category_scores_gemma":[0.3637586,0.00042656,0.0020580802,0.0005282473,0.00052948936,0.00002788085,0.00009170176,0.0012845385,0.000004513462],"study_design_candidate":"randomized_trial","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.12029653,0.0025828457,0.011548959,0.5606242,0.046194196,0.00473503,0.031042024,0.000011151412,6.3762093e-7,0.016767234,0.01033576,0.19586147],"study_design_scores_gemma":[0.78440654,0.0011189627,0.00027629136,0.014714459,0.15383998,0.00037536502,0.012515859,0.00005355151,3.665616e-8,0.0068109906,0.025289997,0.00059793773],"about_ca_topic_score_codex":0.00038331107,"about_ca_topic_score_gemma":0.0008106688,"teacher_disagreement_score":0.66411006,"about_ca_system_score_codex":0.0001678545,"about_ca_system_score_gemma":0.0019806742,"threshold_uncertainty_score":0.9998186},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaepi_broad","metaepi_narrow"],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4413340902","doi":"10.3390/curroncol32080462","title":"High Outcome-Reporting Bias in Randomized-Controlled Trials of Acupuncture for Cancer Chemotherapy-Induced Nausea and Vomiting: A Systematic Review and Meta-Epidemiological Study","year":2025,"lang":"en","type":"review","venue":"Current Oncology","topic":"Acupuncture Treatment Research Studies","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":true,"route_about_ca":false,"ca_institutions":"","funders":"Universitätsspital Zürich","keywords":"Medicine; Nausea; Acupuncture; Meta-analysis; Vomiting; Clinical trial; Reporting bias; Randomized controlled trial; MEDLINE; Publication bias; Metric (unit); Outcome (game theory); Prospective cohort study; Systematic review; Epidemiology; Physical therapy; Alternative medicine; Intensive care medicine; Internal medicine; Pathology","score_opus":0.6576657626141557,"score_gpt":0.6228960202366118,"score_spread":0.03476974237754382,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413340902","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":[0.00026697616,0.89373785,0.00001799942,0.001615342,0.00029359988,0.10394236,0.00007365516,0.00003174922,0.000020482701],"genre_scores_gemma":[0.00029239827,0.93326247,0.00019473149,0.0002563405,0.00013984893,0.06552716,0.00006227545,0.000035758774,0.00022900057],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9688949,0.014255729,0.014812248,0.0010413313,0.0004266395,0.0005691505],"domain_scores_gemma":[0.85650504,0.122908585,0.01911776,0.00067664206,0.0005895777,0.00020237738],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_narrow"],"category_scores_codex":[0.062080555,0.0016876307,0.15158913,0.0006184895,0.00008192757,0.000017412944,0.0002021927,0.0005484416,0.000056627512],"category_scores_gemma":[0.37469643,0.00047538363,0.0072570844,0.00040192067,0.00022339795,0.000032344295,0.00023278898,0.00091143046,4.3672776e-7],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0066682464,0.00070991373,0.00046389698,0.9137176,0.05041975,0.00003227592,0.00010407984,4.8621878e-8,1.1790919e-7,0.000034110322,0.00031416275,0.027535765],"study_design_scores_gemma":[0.16986828,0.0007957713,0.000025175803,0.21898746,0.6033857,0.000045043595,0.00007103305,0.00000328004,3.182262e-7,0.000057379126,0.006460439,0.00030011692],"about_ca_topic_score_codex":0.000046517045,"about_ca_topic_score_gemma":0.000068606925,"teacher_disagreement_score":0.69473016,"about_ca_system_score_codex":0.00049152353,"about_ca_system_score_gemma":0.0008301812,"threshold_uncertainty_score":0.9997698},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"reporting","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4413408738","doi":"10.1177/01410768251366880","title":"Quantifying the impact of immortal time bias: empirical evidence from meta-analyses","year":2025,"lang":"en","type":"article","venue":"Journal of the Royal Society of Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; Mental Health Research Canada; University of Ottawa","funders":"","keywords":"Interquartile range; Meta-analysis; Subgroup analysis; MEDLINE; Study heterogeneity; Medicine; Systematic review; Confidence interval; Publication bias; Random effects model; Statistics; Surgery; Internal medicine; Mathematics","score_opus":0.9106758573085312,"score_gpt":0.6310800049927433,"score_spread":0.2795958523157879,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4413408738","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8890441,0.057328742,0.011997001,0.04011418,0.00049262255,0.00039693314,0.000017561602,0.0000017983572,0.0006070748],"genre_scores_gemma":[0.9944589,0.00014173237,0.0016199031,0.0005725486,0.00021849551,0.0000014525763,4.108071e-7,0.000005744671,0.0029808036],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.97432554,0.008088527,0.010464745,0.0003303785,0.0066046887,0.00018609547],"domain_scores_gemma":[0.95955014,0.02142123,0.0145511255,0.0022945933,0.0020731497,0.00010973749],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12151389,0.00028405417,0.00609241,0.00013529987,0.0001515658,0.00009103699,0.0034807648,0.00009586285,0.009851587],"category_scores_gemma":[0.06652995,0.00006824869,0.015777117,0.0016966311,0.00050480035,0.000135135,0.00024166965,0.00044202333,0.000028076527],"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":[0.000041446274,0.000080212,0.07830391,0.00004869026,0.028274177,0.0000016672527,0.0023344879,0.009292421,0.0029787363,0.000009579023,0.8768764,0.0017582625],"study_design_scores_gemma":[0.0017127788,0.0007801814,0.5605192,0.0023890876,0.09706553,0.000036220423,0.009610614,0.31277466,0.0018221751,0.0065734745,0.0063658124,0.00035025095],"about_ca_topic_score_codex":0.0004944771,"about_ca_topic_score_gemma":0.00000961485,"teacher_disagreement_score":0.8705106,"about_ca_system_score_codex":0.000058602043,"about_ca_system_score_gemma":0.00026893517,"threshold_uncertainty_score":0.99407965},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4417358357","doi":"10.1080/08989621.2025.2600404","title":"Commercial funding of randomized controlled trials of weight-loss interventions using dietary supplements: A rapid review","year":2025,"lang":"en","type":"article","venue":"Accountability in Research","topic":"Pharmacology and Obesity Treatment","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Randomized controlled trial; Psychological intervention; Weight loss; MEDLINE; Clinical trial; Dietary supplement","score_opus":0.4723859013700117,"score_gpt":0.6036886997669948,"score_spread":0.1313027983969831,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417358357","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94195396,0.043007933,0.00007323216,0.0046273367,0.00022356205,0.008993104,0.00008419179,0.000009937408,0.0010267592],"genre_scores_gemma":[0.9838897,0.014846035,0.00025221522,0.00023700402,0.000036177047,0.0005946477,0.00005440063,0.0000060417633,0.00008377802],"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","domain_scores_codex":[0.98732126,0.0082892645,0.0031270252,0.00029219818,0.0005949405,0.00037530414],"domain_scores_gemma":[0.983124,0.014903876,0.00051822077,0.00052538386,0.0008520914,0.00007641369],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06699459,0.0001537015,0.0049938746,0.0006741619,0.00009544673,0.0000065986437,0.00021164253,0.000112880516,0.002586523],"category_scores_gemma":[0.018412255,0.00010408385,0.0015306269,0.00076400844,0.0006522695,0.00009588421,0.00019789315,0.00047235782,0.0000032726136],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.62222743,0.010393994,0.30791825,0.030645149,0.0075984243,0.000036755708,0.00086703186,0.0000051305137,0.0068173897,0.0009882096,0.0055379057,0.0069643664],"study_design_scores_gemma":[0.91403633,0.0005413575,0.037341967,0.028027501,0.0066534565,0.0000064634505,0.00068299135,0.0004138038,0.008582266,0.002871095,0.0006554518,0.00018731462],"about_ca_topic_score_codex":0.0012579589,"about_ca_topic_score_gemma":0.0001620224,"teacher_disagreement_score":0.29180893,"about_ca_system_score_codex":0.00034185633,"about_ca_system_score_gemma":0.000480482,"threshold_uncertainty_score":0.9983252},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow"],"domain":"incentives","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W6939574047","doi":"10.6084/m9.figshare.14746350","title":"Additional file 1 of Bias estimation in study design: a meta-epidemiological analysis of transcatheter versus surgical aortic valve replacement","year":2021,"lang":"en","type":"article","venue":"Open MIND","topic":"Cardiac Valve Diseases and Treatments","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Women's College Hospital; University of Toronto; Western University","funders":"","keywords":"Table (database); Chart; Aortic valve replacement; Flow chart; Valve replacement; Random error","score_opus":0.20481244366084914,"score_gpt":0.43322146448937443,"score_spread":0.2284090208285253,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W6939574047","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.89975756,0.00016378821,0.00007547609,0.0000522504,0.00002251852,0.0010558809,0.09720615,0.0000012462274,0.0016651099],"genre_scores_gemma":[0.9599971,0.0000041681146,0.0069790436,0.000008369661,0.000005577213,0.000387636,0.032393813,0.0000047043527,0.00021955583],"study_design_codex":"meta_analysis","study_design_gemma":"observational","domain_scores_codex":[0.99856955,0.00035195256,0.0004492916,0.0002832849,0.00024177507,0.000104152896],"domain_scores_gemma":[0.9968324,0.00261709,0.00013854618,0.00025010214,0.00009234988,0.00006952212],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00035036894,0.000110289904,0.0009108687,0.0001095901,0.000014517747,0.000008029963,0.00005300746,0.00004025548,0.71198577],"category_scores_gemma":[0.0008998586,0.00007737685,0.002084262,0.00049256644,0.000029243847,0.000039985134,0.000039152434,0.000047466467,0.00002745992],"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":[0.022052424,0.039602917,0.22665618,0.000086260494,0.5531789,0.00054891745,0.0019383124,0.013011402,0.000034925117,0.000015066613,0.082952626,0.059922077],"study_design_scores_gemma":[0.006132661,0.0017353112,0.9012684,0.00007656867,0.083098635,0.0000040697596,0.0006436935,0.004826518,0.00012396394,0.000015842388,0.0019335917,0.00014073178],"about_ca_topic_score_codex":0.000038856233,"about_ca_topic_score_gemma":0.000018058065,"teacher_disagreement_score":0.7119583,"about_ca_system_score_codex":0.00004369548,"about_ca_system_score_gemma":0.00012080286,"threshold_uncertainty_score":0.31553364},"labels":[{"model":"gemma","categories":["metaepi_broad","metaepi_narrow","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["insufficient_payload"],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W7117489679","doi":"10.1016/j.semarthrit.2025.152907","title":"Determining the best discriminatory physical functioning outcome measurement instrument for psoriatic arthritis trials: A meta-epidemiological study","year":2025,"lang":"en","type":"article","venue":"Seminars in Arthritis and Rheumatism","topic":"Spondyloarthritis Studies and Treatments","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Krembil Foundation; University Health Network; Ottawa Hospital; University of Ottawa","funders":"Parker Institute for Cancer Immunotherapy; Oak Foundation","keywords":"Psoriatic arthritis; Outcome (game theory); Patient-reported outcome; Arthritis; MEDLINE; Psychometrics","score_opus":0.16281216313373342,"score_gpt":0.37394785276592624,"score_spread":0.2111356896321928,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W7117489679","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8743727,0.11973844,0.00022696814,0.0013723038,0.0006948928,0.003119537,0.000033234766,0.00004139225,0.0004004875],"genre_scores_gemma":[0.95894176,0.03742121,0.0008575907,0.00030272632,0.00007121373,0.0022557909,0.00001282862,0.000021482781,0.00011536824],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9967291,0.0005267838,0.0012543111,0.00059403834,0.0004286473,0.00046714663],"domain_scores_gemma":[0.9977535,0.0014116148,0.0002604277,0.0003430334,0.00012824348,0.000103185004],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0038022269,0.0004274634,0.0022181785,0.00015552843,0.0005681948,0.000071791284,0.000102586135,0.000101270045,0.000025193276],"category_scores_gemma":[0.0033191438,0.00025423968,0.00049504417,0.00020172192,0.00019217575,0.000118823744,0.00014927208,0.00029374883,0.0000043341342],"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":[0.00014848505,0.0012472705,0.1420935,0.00006722141,0.0013082325,0.00003596766,0.001148389,0.000003054829,0.000009829358,0.0014111409,0.0003313863,0.8521955],"study_design_scores_gemma":[0.04708352,0.0076871887,0.8866857,0.021993153,0.005190059,0.00011954554,0.019587288,0.0007037446,0.00007752029,0.0079995915,0.002059714,0.00081294595],"about_ca_topic_score_codex":0.000117784824,"about_ca_topic_score_gemma":0.00047328122,"teacher_disagreement_score":0.8513826,"about_ca_system_score_codex":0.0001606742,"about_ca_system_score_gemma":0.00005793797,"threshold_uncertainty_score":0.999991},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"}]}