{"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015923152,0.9094245,0.020600228,0.027582252,0.003281126,0.014769997,0.0036806562,0.00018611533,0.004551968],"genre_scores_gemma":[0.5286565,0.28856152,0.07050012,0.026325062,0.0032435143,0.07867394,0.0027117513,0.00033166012,0.0009960318],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.20404193,0.51803386,0.21192892,0.018774731,0.04504545,0.002175119],"domain_scores_gemma":[0.11560245,0.74767303,0.09019625,0.02031721,0.025179023,0.0010321158],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5837778,0.0032627408,0.01802771,0.018407097,0.0042729974,0.014767172,0.006378443,0.006267817,0.006144984],"category_scores_gemma":[0.82465434,0.0033447582,0.014541748,0.025135128,0.008381182,0.019385848,0.008650056,0.004436077,0.0007784676],"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":[0.0027009032,0.00011666533,0.025359116,0.7685531,0.070980445,0.00033589697,0.0067268577,0.0009698062,0.00049588847,0.007162373,0.00543368,0.11116525],"study_design_scores_gemma":[0.005131751,0.0014183922,0.042080276,0.7079582,0.17876682,0.0007740536,0.003589564,0.0035249633,0.0020935666,0.025523793,0.028619893,0.0005187931],"about_ca_topic_score_codex":0.009148925,"about_ca_topic_score_gemma":0.015064589,"teacher_disagreement_score":0.9819723,"about_ca_system_score_codex":0.01910528,"about_ca_system_score_gemma":0.035429824,"threshold_uncertainty_score":0.5132761},"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":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0025780934,0.95136917,0.020833287,0.00955124,0.0054785134,0.006427243,0.00167041,0.00025818552,0.0018337592],"genre_scores_gemma":[0.22744308,0.4730799,0.1798089,0.029685592,0.009049092,0.07649571,0.0024906562,0.0006561258,0.0012909665],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.38489363,0.47613102,0.091194145,0.018242396,0.028290313,0.0012485426],"domain_scores_gemma":[0.34559986,0.5301894,0.07981639,0.024097791,0.0179605,0.0023360779],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.32634926,0.0039106905,0.03836769,0.027686872,0.001516888,0.01896381,0.006310943,0.010074567,0.0092360405],"category_scores_gemma":[0.6629876,0.004238889,0.032025058,0.024256915,0.005144932,0.012246159,0.0066036712,0.008317989,0.0015313275],"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":[0.0052369256,0.00008117652,0.005581405,0.4866207,0.42142698,0.00015429444,0.0006357697,0.00095673034,0.00021386553,0.0063588494,0.008775158,0.06395814],"study_design_scores_gemma":[0.017812662,0.0014153827,0.009555934,0.37269092,0.52030087,0.00062207057,0.0006503961,0.004487571,0.000537085,0.03699398,0.03446099,0.00047219367],"about_ca_topic_score_codex":0.0022424797,"about_ca_topic_score_gemma":0.0022987581,"teacher_disagreement_score":0.9616323,"about_ca_system_score_codex":0.007150777,"about_ca_system_score_gemma":0.009667874,"threshold_uncertainty_score":0.8307313},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2201201,0.53177387,0.21544708,0.006859262,0.007442755,0.0066567957,0.0038825967,0.0018694941,0.005948176],"genre_scores_gemma":[0.9259026,0.019297602,0.040109705,0.0042571756,0.0017780937,0.004320605,0.0012725047,0.0004284045,0.002633349],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.71683407,0.24901652,0.017323535,0.009084807,0.0054378677,0.0023032199],"domain_scores_gemma":[0.7330773,0.2265989,0.012093969,0.022740567,0.0043926258,0.0010966012],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.15971294,0.0043580392,0.015147193,0.0037484374,0.0014220421,0.006236112,0.0053462167,0.009724429,0.010320503],"category_scores_gemma":[0.26303396,0.003115857,0.05516843,0.00518877,0.0025584265,0.005347708,0.0029046931,0.008572807,0.0010380801],"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":[0.09953756,0.0003573877,0.008694963,0.025308566,0.8238805,0.0004406815,0.00028900022,0.0029145805,0.0004368626,0.0039619417,0.0021779842,0.032],"study_design_scores_gemma":[0.07975696,0.004145932,0.0066345595,0.0035629068,0.8773566,0.00039666987,0.0001588531,0.011368132,0.00052467646,0.013089339,0.002778087,0.00022737173],"about_ca_topic_score_codex":0.003990686,"about_ca_topic_score_gemma":0.0030434395,"teacher_disagreement_score":0.99564195,"about_ca_system_score_codex":0.0019539236,"about_ca_system_score_gemma":0.0031206745,"threshold_uncertainty_score":0.8446529},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000293078,0.0021978428,0.005862086,0.0005215843,0.00070727145,0.984781,0.0044291774,0.00052323093,0.00068469887],"genre_scores_gemma":[0.00031613826,0.0004153089,0.008560316,0.000116169416,0.00003159844,0.9900422,0.0003138624,0.00003757102,0.00016679728],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.79335636,0.12126401,0.055134997,0.013115033,0.013342416,0.0037871257],"domain_scores_gemma":[0.74238133,0.13411725,0.057865333,0.024077797,0.035683505,0.0058747097],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.15985464,0.01115771,0.027173933,0.015124306,0.004366661,0.011770713,0.0075557353,0.012658285,0.1066387],"category_scores_gemma":[0.30686125,0.009581524,0.02960055,0.019560054,0.008559832,0.0111901965,0.008432344,0.02088894,0.022881938],"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":[0.019273246,0.00043882112,0.00088731165,0.84144217,0.011363325,0.00039702904,0.0020133248,0.002304174,0.0017314034,0.012165494,0.059973907,0.048009798],"study_design_scores_gemma":[0.13055027,0.003589067,0.0068713115,0.47477844,0.020799754,0.0009748915,0.0014958025,0.009519718,0.0055400627,0.054570727,0.2898845,0.0014255052],"about_ca_topic_score_codex":0.0033235066,"about_ca_topic_score_gemma":0.0052416953,"teacher_disagreement_score":0.9888423,"about_ca_system_score_codex":0.016465217,"about_ca_system_score_gemma":0.035223015,"threshold_uncertainty_score":0.84540224},"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":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0005640363,0.00322192,0.009150032,0.001212458,0.0016340048,0.9780352,0.0048578633,0.00042614157,0.00089840655],"genre_scores_gemma":[0.00075296784,0.00046708694,0.00794223,0.00028443956,0.000039016133,0.990006,0.00031753295,0.00003266244,0.0001580556],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.73436505,0.15712598,0.075865924,0.011285101,0.017928123,0.0034298196],"domain_scores_gemma":[0.686529,0.18681446,0.054424837,0.030959483,0.036512896,0.0047593485],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20173547,0.006689571,0.021561213,0.01164611,0.0033381076,0.010337822,0.005941013,0.017658709,0.05366499],"category_scores_gemma":[0.42352897,0.0069154343,0.032627273,0.015105629,0.007445401,0.010457652,0.005430468,0.020011075,0.010857642],"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":[0.031955242,0.0007377188,0.0020597756,0.7996633,0.028237792,0.00047248052,0.0020912208,0.0032335145,0.0014346902,0.014582697,0.066073,0.049458563],"study_design_scores_gemma":[0.1545058,0.0050553065,0.012239666,0.4491963,0.04806022,0.0007783745,0.0015670656,0.01358191,0.0055414294,0.08280768,0.225352,0.001314361],"about_ca_topic_score_codex":0.0020284334,"about_ca_topic_score_gemma":0.0027651398,"teacher_disagreement_score":0.99331045,"about_ca_system_score_codex":0.012503481,"about_ca_system_score_gemma":0.021517139,"threshold_uncertainty_score":0.9844023},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.33261433,0.5481998,0.034624882,0.012003689,0.03519018,0.024664547,0.0034603286,0.00064586714,0.008596439],"genre_scores_gemma":[0.90561897,0.050886925,0.012567434,0.0068005864,0.0062354854,0.014771953,0.0010260371,0.00010923848,0.0019833774],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.81200635,0.15748622,0.020251544,0.005274944,0.0028028297,0.002178084],"domain_scores_gemma":[0.70182675,0.25553134,0.025671184,0.009672299,0.004007776,0.0032906537],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.098213315,0.0042029764,0.01644989,0.002148308,0.0016017769,0.0038946616,0.0026251625,0.009653111,0.010098281],"category_scores_gemma":[0.18574749,0.002445183,0.015781902,0.0023796253,0.005349136,0.005887479,0.002725186,0.009433651,0.00106984],"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":[0.84612775,0.0014297991,0.00317405,0.034553107,0.08827582,0.00011860259,0.00026340247,0.0009625003,0.000521712,0.0019220473,0.0011790254,0.021472238],"study_design_scores_gemma":[0.89075094,0.024985746,0.0026111526,0.0046197707,0.06815552,0.00009881542,0.0001538705,0.0012217731,0.000339301,0.004352837,0.0026343428,0.000075971264],"about_ca_topic_score_codex":0.0022300698,"about_ca_topic_score_gemma":0.0022073868,"teacher_disagreement_score":0.9017867,"about_ca_system_score_codex":0.0023733815,"about_ca_system_score_gemma":0.0031523127,"threshold_uncertainty_score":0},"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":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.006130174,0.940438,0.03795591,0.0021190536,0.0019447374,0.008888373,0.0014823488,0.00019421407,0.00084726705],"genre_scores_gemma":[0.33332133,0.4003554,0.18528965,0.007984627,0.0030416558,0.0657717,0.0029564346,0.0003350667,0.00094413967],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.4616793,0.40944183,0.08712559,0.013251381,0.027090672,0.0014112779],"domain_scores_gemma":[0.31093597,0.5739766,0.07886933,0.019260075,0.015947806,0.0010102844],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34036097,0.005669951,0.025653552,0.020020392,0.0015299567,0.011128466,0.005363339,0.0075594517,0.004413481],"category_scores_gemma":[0.6200874,0.003485277,0.059141044,0.0146045275,0.0034371123,0.009317963,0.006061543,0.0055970596,0.00056206423],"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":[0.0031755727,0.000086519176,0.010861536,0.43097952,0.5127032,0.00020096476,0.0004910898,0.0020172824,0.0003232377,0.0023566366,0.0021111441,0.034693345],"study_design_scores_gemma":[0.0054468536,0.0013580457,0.010424412,0.20892811,0.7509107,0.0006837572,0.00027660938,0.003011194,0.0008327457,0.010264837,0.007593318,0.00026934542],"about_ca_topic_score_codex":0.002711654,"about_ca_topic_score_gemma":0.0038275586,"teacher_disagreement_score":0.99433005,"about_ca_system_score_codex":0.0071727172,"about_ca_system_score_gemma":0.00971658,"threshold_uncertainty_score":0.81345236},"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":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":"reproducibility","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"reproducibility","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0003097001,0.00085060013,0.006112371,0.0007059421,0.00077828416,0.9820287,0.0076718605,0.00050703547,0.0010355135],"genre_scores_gemma":[0.00030367414,0.00024901467,0.011893384,0.00018182148,0.000035325436,0.98639745,0.0005484026,0.000036271715,0.00035462008],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.83207244,0.09293342,0.052360065,0.008561182,0.010814992,0.0032579482],"domain_scores_gemma":[0.6479911,0.17630118,0.044095084,0.042972002,0.0817238,0.0069168326],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.18266788,0.007809128,0.016606184,0.017619016,0.0051299697,0.007837617,0.007398313,0.013803445,0.14008932],"category_scores_gemma":[0.2992265,0.009284394,0.018966366,0.019400254,0.0070434366,0.010090238,0.0063706897,0.014677461,0.028634207],"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":[0.016322589,0.00083393644,0.0010489795,0.6758685,0.004350846,0.0011839081,0.006222287,0.0031975857,0.003808524,0.017262366,0.18028294,0.089617625],"study_design_scores_gemma":[0.06716193,0.002404804,0.005744989,0.3797173,0.0064353403,0.0008537641,0.003407267,0.0053569614,0.005952225,0.05232639,0.4692749,0.001364252],"about_ca_topic_score_codex":0.0032669103,"about_ca_topic_score_gemma":0.0061354428,"teacher_disagreement_score":0.81733215,"about_ca_system_score_codex":0.014105277,"about_ca_system_score_gemma":0.0665527,"threshold_uncertainty_score":0.96605164},"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.001863103,0.030336808,0.3020448,0.019426148,0.004879871,0.63449985,0.0035852247,0.0009192904,0.0024450566],"genre_scores_gemma":[0.0045135175,0.0041653817,0.3376359,0.002450912,0.0003051648,0.65024734,0.00033955462,0.00007362436,0.00026863377],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.15907481,0.66717416,0.13673091,0.009322339,0.02627817,0.0014196612],"domain_scores_gemma":[0.18704502,0.6071325,0.06934068,0.05581519,0.0767237,0.003942981],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7681447,0.008192635,0.020911867,0.022316914,0.0046431324,0.01612566,0.010608668,0.022420341,0.006627773],"category_scores_gemma":[0.8262535,0.0077023427,0.030500459,0.017173659,0.010028827,0.016841281,0.011334057,0.015659532,0.003375507],"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":[0.008997284,0.0007992442,0.0075340588,0.46710053,0.04764693,0.00075780763,0.0068210918,0.008259258,0.0019457376,0.06502827,0.046931785,0.33817792],"study_design_scores_gemma":[0.022439374,0.0042250506,0.00846829,0.5082804,0.06114203,0.0010184557,0.0020332036,0.031763136,0.0071940757,0.17764623,0.17442355,0.001366173],"about_ca_topic_score_codex":0.0023515788,"about_ca_topic_score_gemma":0.00336157,"teacher_disagreement_score":0.9790881,"about_ca_system_score_codex":0.012972182,"about_ca_system_score_gemma":0.054435827,"threshold_uncertainty_score":0.2859189},"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":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0050451024,0.9916971,0.00081371155,0.00036880496,0.00042259792,0.00038178157,0.00054608076,0.000041779298,0.00068314996],"genre_scores_gemma":[0.2002918,0.78915673,0.004532577,0.0020668334,0.0005342882,0.0013106919,0.0012444308,0.00007873486,0.000784011],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.98665327,0.0070603425,0.0033093146,0.0011696503,0.0015076348,0.00029981125],"domain_scores_gemma":[0.97637886,0.017296702,0.0029988901,0.0010738849,0.0019377965,0.00031382646],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.017482843,0.002575074,0.016428802,0.009863012,0.0008854513,0.0043786997,0.0017161351,0.0020275742,0.0041433442],"category_scores_gemma":[0.04541466,0.0013024762,0.037348963,0.008093194,0.0006719473,0.0022396871,0.0015463023,0.002472281,0.0003860627],"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":[0.0021928293,0.00003705094,0.0018128649,0.23540394,0.7421693,0.00014348888,0.00008752424,0.00038085296,0.0002850806,0.00021127255,0.0008893239,0.01638638],"study_design_scores_gemma":[0.000635297,0.00023047466,0.002507068,0.019275013,0.9742621,0.000110866364,0.00004579322,0.00009697836,0.00016625816,0.00028133567,0.0023672422,0.000021576545],"about_ca_topic_score_codex":0.0062565263,"about_ca_topic_score_gemma":0.014815534,"teacher_disagreement_score":0.017482843,"about_ca_system_score_codex":0.0035855426,"about_ca_system_score_gemma":0.0040900116,"threshold_uncertainty_score":0},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0010408907,0.0010347948,0.04775416,0.0009540493,0.0018322314,0.9420783,0.003314861,0.00081436307,0.0011763771],"genre_scores_gemma":[0.0014028169,0.00017918232,0.027719384,0.00026143002,0.00006183653,0.9698574,0.00025840692,0.00004161815,0.000217945],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.54440296,0.34130138,0.076611534,0.017654557,0.017225143,0.0028044123],"domain_scores_gemma":[0.60683924,0.21771897,0.050441064,0.07725621,0.043380693,0.0043638097],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36633945,0.008920852,0.015082846,0.0118787335,0.0026640175,0.00762315,0.005937747,0.019838732,0.03449022],"category_scores_gemma":[0.5241475,0.0060673426,0.022887312,0.011838195,0.0068304506,0.0071342457,0.004395207,0.017662292,0.013579774],"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":[0.12153967,0.0026936934,0.005643968,0.42069325,0.055879924,0.0016332599,0.007095133,0.014072113,0.006710199,0.07735243,0.10666287,0.18002355],"study_design_scores_gemma":[0.25204304,0.013221734,0.014405627,0.12759033,0.05286732,0.0013696677,0.0010068043,0.07255606,0.017332098,0.19788301,0.24734941,0.002374888],"about_ca_topic_score_codex":0.00096050743,"about_ca_topic_score_gemma":0.001170588,"teacher_disagreement_score":0.98491716,"about_ca_system_score_codex":0.0062880307,"about_ca_system_score_gemma":0.019142982,"threshold_uncertainty_score":0.7814163},"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015737727,0.9198148,0.047500648,0.006998532,0.0042574103,0.0031674418,0.00070140336,0.0002726943,0.0015493124],"genre_scores_gemma":[0.6447996,0.23447321,0.081069686,0.013900658,0.0065590492,0.016524483,0.0011288385,0.0005556799,0.0009887764],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.3308317,0.5555229,0.07079982,0.01606487,0.024968108,0.0018125728],"domain_scores_gemma":[0.24725217,0.6553109,0.059141695,0.026970878,0.01034998,0.00097434054],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.45749414,0.004957609,0.024495222,0.0068325954,0.0015173503,0.009728903,0.005008601,0.010862487,0.002878586],"category_scores_gemma":[0.66299635,0.004271474,0.060000505,0.008313242,0.005012969,0.012107935,0.0036267736,0.0078092725,0.00062806974],"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":[0.014915297,0.0001558367,0.015954033,0.12495721,0.76926947,0.0002375494,0.00067080895,0.0035202121,0.0004354298,0.0031851244,0.0022467626,0.064452335],"study_design_scores_gemma":[0.0139469765,0.0026865022,0.015490532,0.07194759,0.86672336,0.00045941045,0.00024107097,0.0067670564,0.0012835673,0.012716955,0.0074215154,0.0003154922],"about_ca_topic_score_codex":0.003276558,"about_ca_topic_score_gemma":0.0035138435,"teacher_disagreement_score":0.9950424,"about_ca_system_score_codex":0.0050606276,"about_ca_system_score_gemma":0.0051739053,"threshold_uncertainty_score":0.66900635},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03635264,0.9394122,0.017480638,0.00068094407,0.0011326923,0.0034016955,0.00082947465,0.00013932819,0.00057040167],"genre_scores_gemma":[0.7246042,0.21702813,0.041196913,0.0033868102,0.0011954657,0.010272186,0.0012899792,0.00028391238,0.0007424698],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.57666814,0.308152,0.07943793,0.01756638,0.015785918,0.0023896676],"domain_scores_gemma":[0.46784967,0.4589836,0.042962156,0.018499726,0.01017306,0.0015317794],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22346777,0.0071780575,0.03690466,0.013708549,0.0016334265,0.009948881,0.0051200646,0.009356797,0.004065293],"category_scores_gemma":[0.46045542,0.0064588613,0.08900198,0.01112323,0.0032612744,0.008969825,0.0071886983,0.007197494,0.00048082403],"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":[0.007142019,0.000050167368,0.0060812035,0.06648231,0.9148422,0.00011974229,0.00021622829,0.00033863413,0.00011632536,0.00023723516,0.00024287424,0.004131103],"study_design_scores_gemma":[0.004790759,0.00043147538,0.003172167,0.0074155065,0.98186594,0.00015051635,0.0001043409,0.0005568954,0.000113031565,0.0008698546,0.0004569306,0.00007253023],"about_ca_topic_score_codex":0.0053911093,"about_ca_topic_score_gemma":0.009939843,"teacher_disagreement_score":0.99282193,"about_ca_system_score_codex":0.004999755,"about_ca_system_score_gemma":0.005983931,"threshold_uncertainty_score":0.9576025},"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":"review","genre_gemma":"commentary","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0013389454,0.53360826,0.009389328,0.4164677,0.035805684,0.00044592647,0.0008540149,0.00014701855,0.0019432071],"genre_scores_gemma":[0.05356685,0.28848678,0.056558542,0.4745042,0.12077149,0.0031008685,0.0015246083,0.00021680638,0.0012697898],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.7138021,0.2082761,0.046864495,0.008292837,0.021726226,0.0010382435],"domain_scores_gemma":[0.13917352,0.781185,0.02516804,0.024480155,0.027686952,0.00230634],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31571636,0.0024892823,0.014223556,0.0067380867,0.0019613376,0.0084643075,0.007808179,0.01594719,0.010232179],"category_scores_gemma":[0.6278673,0.002665547,0.013076334,0.007061832,0.004136171,0.01737739,0.0043112664,0.022992706,0.0016907878],"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":[0.008331336,0.0004815654,0.018548576,0.1803432,0.08959745,0.0023859686,0.0015276972,0.0012861383,0.0007813798,0.029133528,0.3521517,0.31543145],"study_design_scores_gemma":[0.010908121,0.0017932254,0.024567435,0.21717292,0.09289901,0.003957992,0.0021776948,0.004804736,0.0009359845,0.14775051,0.4921061,0.0009263177],"about_ca_topic_score_codex":0.0037080843,"about_ca_topic_score_gemma":0.007016688,"teacher_disagreement_score":0.9857764,"about_ca_system_score_codex":0.004137568,"about_ca_system_score_gemma":0.008509175,"threshold_uncertainty_score":0.8438436},"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":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018494688,0.85855687,0.108771645,0.0046599796,0.0028044248,0.0035279617,0.0011086809,0.00035911085,0.0017167246],"genre_scores_gemma":[0.64948475,0.1918979,0.13026232,0.009423707,0.002338042,0.013826187,0.0014350716,0.000445777,0.00088618847],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.41291055,0.460783,0.08276537,0.016014628,0.026000556,0.0015259478],"domain_scores_gemma":[0.24845886,0.68012035,0.038896546,0.02190562,0.009942393,0.0006761646],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.42773893,0.004418932,0.018342838,0.012135994,0.0012710043,0.011325981,0.0039673112,0.0060396795,0.0023985694],"category_scores_gemma":[0.65088445,0.003276793,0.05145034,0.012460903,0.003986713,0.008142258,0.0051821577,0.0069881435,0.00033865724],"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":[0.004387647,0.000094150855,0.023587005,0.11599486,0.77849084,0.0004259996,0.0007456318,0.004926826,0.0004443965,0.005370085,0.0017611609,0.06377131],"study_design_scores_gemma":[0.005307195,0.0010996358,0.015526982,0.04889485,0.88069683,0.0008193623,0.00028911926,0.0106883235,0.0013200268,0.027910315,0.0071395165,0.00030787126],"about_ca_topic_score_codex":0.002595644,"about_ca_topic_score_gemma":0.0032475141,"teacher_disagreement_score":0.9955811,"about_ca_system_score_codex":0.004603424,"about_ca_system_score_gemma":0.0062040635,"threshold_uncertainty_score":0.7056998},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.002505329,0.9931096,0.0009176985,0.0020179916,0.0003844374,0.00016514116,0.00024420483,0.0000080222635,0.0006476319],"genre_scores_gemma":[0.16600074,0.8218055,0.004130989,0.004545182,0.001581101,0.0009429006,0.0006865091,0.000049653903,0.0002574737],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.85514176,0.09441846,0.028280536,0.0075987363,0.013389032,0.0011714653],"domain_scores_gemma":[0.45890886,0.4699189,0.04572413,0.0066956435,0.017372588,0.0013798799],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.086951666,0.0016642761,0.010420713,0.0141925,0.0010554986,0.0075931177,0.003315478,0.0047935224,0.007043593],"category_scores_gemma":[0.3649171,0.00158146,0.015374636,0.016785018,0.001972853,0.007856677,0.0025077309,0.0039122677,0.00061555015],"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":[0.00213779,0.000114615286,0.033825286,0.57405925,0.22832276,0.00034564527,0.0016941361,0.00061511865,0.00013416009,0.0031582653,0.0068427515,0.14875014],"study_design_scores_gemma":[0.0015576485,0.000609621,0.036204435,0.5032052,0.42108357,0.0008123619,0.001190456,0.0010286272,0.00025319407,0.007994121,0.025902588,0.00015815857],"about_ca_topic_score_codex":0.0059412494,"about_ca_topic_score_gemma":0.009110675,"teacher_disagreement_score":0.98957926,"about_ca_system_score_codex":0.0057800054,"about_ca_system_score_gemma":0.0062402748,"threshold_uncertainty_score":0.45984983},"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":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00089060486,0.0008364855,0.008580626,0.00047052468,0.00048279675,0.9857319,0.0023507285,0.00021970403,0.0004365596],"genre_scores_gemma":[0.00062913384,0.00015296375,0.0069511407,0.00007569106,0.00001873478,0.99189174,0.0001854158,0.000009838699,0.00008536324],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.75370055,0.14734884,0.07143454,0.009996256,0.014528178,0.002991644],"domain_scores_gemma":[0.7235413,0.14993748,0.04262769,0.04064716,0.038662877,0.0045835134],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.23763415,0.0048972075,0.010712394,0.011477964,0.0035419283,0.0050967494,0.004025607,0.010754651,0.043879565],"category_scores_gemma":[0.33976465,0.0048500495,0.018109893,0.013949683,0.0059777456,0.0075559714,0.0037786663,0.010389922,0.0094770305],"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":[0.077253245,0.0031082935,0.006166592,0.63751537,0.023840224,0.0014725785,0.007987019,0.005760844,0.004865429,0.023381999,0.082483724,0.12616475],"study_design_scores_gemma":[0.25949717,0.015025146,0.032778982,0.29346853,0.040870644,0.0015827996,0.003285394,0.025788182,0.009351021,0.07820796,0.23814565,0.0019984767],"about_ca_topic_score_codex":0.0011749157,"about_ca_topic_score_gemma":0.0026601586,"teacher_disagreement_score":0.9892876,"about_ca_system_score_codex":0.007614664,"about_ca_system_score_gemma":0.027774857,"threshold_uncertainty_score":0.9401328},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006334123,0.99806494,0.00044286996,0.00027475183,0.00010553605,0.00019461225,0.00015826573,0.000003232225,0.00012241054],"genre_scores_gemma":[0.017258849,0.9782445,0.0022557024,0.00072346086,0.00017953807,0.00088598236,0.00035707402,0.000009266897,0.00008567313],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.93850416,0.02655804,0.026417583,0.0025217216,0.0055286097,0.0004699036],"domain_scores_gemma":[0.7815145,0.1763232,0.027352277,0.0045526247,0.00973413,0.0005232738],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06683796,0.0016669781,0.00770628,0.008253897,0.000731723,0.0030097738,0.0026822048,0.0021587713,0.0025352633],"category_scores_gemma":[0.19656609,0.0012970257,0.012715929,0.009151739,0.0013261682,0.0025829584,0.0022353958,0.0015180542,0.00025978265],"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":[0.0006797076,0.000032555352,0.0047524744,0.860606,0.041147266,0.000116128555,0.00029645534,0.00025095325,0.0001981075,0.000657824,0.0013300398,0.08993259],"study_design_scores_gemma":[0.0006266778,0.00047894375,0.016534902,0.66168445,0.2890676,0.0005832982,0.00037661768,0.00022775115,0.00050939363,0.002057317,0.027770324,0.000082746126],"about_ca_topic_score_codex":0.0046321317,"about_ca_topic_score_gemma":0.008718787,"teacher_disagreement_score":0.93316203,"about_ca_system_score_codex":0.003269977,"about_ca_system_score_gemma":0.0096302545,"threshold_uncertainty_score":0.35347712},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030007564,0.8991595,0.04313177,0.009415833,0.0027701405,0.006453842,0.004609424,0.00029870024,0.00415327],"genre_scores_gemma":[0.8149791,0.1018795,0.052218217,0.011950079,0.0014459175,0.013407292,0.002937233,0.0002224658,0.0009603087],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5722841,0.29835483,0.093041904,0.013015358,0.021229027,0.0020747262],"domain_scores_gemma":[0.25880063,0.6447164,0.05275673,0.030004812,0.012643901,0.0010774959],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.29901454,0.002940666,0.01174985,0.013807865,0.00156379,0.0074028918,0.00431477,0.0046637855,0.0073329275],"category_scores_gemma":[0.6141738,0.0026376985,0.04578748,0.012579469,0.0033150732,0.007549199,0.0042923614,0.005226632,0.0009216615],"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":[0.005720903,0.00006533037,0.024148833,0.19135165,0.733145,0.00039339127,0.0005020685,0.0016056441,0.00037700808,0.0026834311,0.002666088,0.037340596],"study_design_scores_gemma":[0.003717511,0.0009780857,0.018039325,0.06694474,0.88646466,0.00042250502,0.00032204806,0.0017796482,0.0004922786,0.010867459,0.009803938,0.00016786433],"about_ca_topic_score_codex":0.0038172617,"about_ca_topic_score_gemma":0.006314867,"teacher_disagreement_score":0.98825014,"about_ca_system_score_codex":0.0045862948,"about_ca_system_score_gemma":0.00559014,"threshold_uncertainty_score":0.86443985},"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":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0035027955,0.9840013,0.009339394,0.001080606,0.0005286198,0.00033710254,0.00080926047,0.00007106086,0.0003298848],"genre_scores_gemma":[0.28609222,0.67027444,0.030939896,0.0036809375,0.0013976458,0.0037529771,0.0028285868,0.00034973133,0.0006835735],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.8950748,0.07452492,0.01740714,0.006499565,0.005861713,0.00063188217],"domain_scores_gemma":[0.70325774,0.25724646,0.017968925,0.012333344,0.008559522,0.0006340066],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.12487763,0.0035833507,0.011805563,0.0066356547,0.00071656826,0.004723509,0.0040424177,0.0027762754,0.0027932145],"category_scores_gemma":[0.28152642,0.001984845,0.035787217,0.008416833,0.0015906631,0.002326951,0.0021441795,0.0025696775,0.0003519209],"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":[0.0012270146,0.000023677041,0.009474565,0.20686096,0.74387014,0.00011318371,0.00016848282,0.0012682814,0.00027887532,0.0010066106,0.0015702558,0.03413796],"study_design_scores_gemma":[0.000504902,0.00032541118,0.0126852,0.045518074,0.927867,0.0001751779,0.00011440134,0.0009364726,0.00055964355,0.0024624884,0.008776509,0.000074753756],"about_ca_topic_score_codex":0.0063939407,"about_ca_topic_score_gemma":0.009638599,"teacher_disagreement_score":0.12487763,"about_ca_system_score_codex":0.0034664662,"about_ca_system_score_gemma":0.0043130005,"threshold_uncertainty_score":0.66042393},"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.008740398,0.8814977,0.08333945,0.014640399,0.005641197,0.0017513251,0.0015935574,0.00054112915,0.0022549157],"genre_scores_gemma":[0.5269903,0.2597743,0.16143191,0.032120794,0.00757797,0.008206721,0.0017430376,0.0010025033,0.0011524623],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.46066448,0.4706745,0.0372767,0.01613431,0.014138237,0.0011117434],"domain_scores_gemma":[0.32519642,0.6025525,0.029185308,0.035708785,0.0067340457,0.00062308076],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.35538414,0.004620461,0.021549731,0.011621831,0.0011101094,0.008979626,0.0062183416,0.008204185,0.0030904291],"category_scores_gemma":[0.64210296,0.0039378214,0.04783996,0.012094789,0.003467173,0.00962794,0.0040533254,0.0066257417,0.0009014928],"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":[0.002329902,0.000045808865,0.018138677,0.08345544,0.82120335,0.00023635713,0.00062170817,0.0032762298,0.00028749014,0.004935831,0.0069561037,0.05851313],"study_design_scores_gemma":[0.0032992626,0.00067601923,0.011065056,0.0552571,0.84000605,0.00044161742,0.00040524403,0.005452754,0.0012009057,0.05891973,0.022902159,0.00037407153],"about_ca_topic_score_codex":0.0026365872,"about_ca_topic_score_gemma":0.0033694776,"teacher_disagreement_score":0.9784503,"about_ca_system_score_codex":0.0034995768,"about_ca_system_score_gemma":0.0033870554,"threshold_uncertainty_score":0.7949261},"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0058078594,0.8283394,0.045791768,0.003719161,0.004755591,0.10579068,0.0033360445,0.0005808411,0.0018786954],"genre_scores_gemma":[0.15070623,0.21847288,0.15787537,0.008169284,0.0018476213,0.4591471,0.002182031,0.00038465252,0.0012148943],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.37677386,0.3963726,0.15953498,0.020718534,0.045105726,0.0014943678],"domain_scores_gemma":[0.3592418,0.51686525,0.081887424,0.020096533,0.020613734,0.0012952499],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.33654216,0.004568919,0.035400894,0.02557627,0.0022527387,0.00951536,0.007784355,0.010070801,0.0061234515],"category_scores_gemma":[0.5951705,0.00403466,0.036942676,0.019918203,0.007163558,0.010913661,0.0045663808,0.0046970947,0.0009691797],"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":[0.0009495462,0.000045350975,0.0018898385,0.8853156,0.07637611,0.00012962325,0.00071381713,0.0003528645,0.00014050826,0.0016847361,0.0019249099,0.03047708],"study_design_scores_gemma":[0.0077197775,0.0012123806,0.0041318196,0.7441987,0.20964512,0.00045483292,0.0007709154,0.003594737,0.000925353,0.011015952,0.015921347,0.00040907625],"about_ca_topic_score_codex":0.0032180732,"about_ca_topic_score_gemma":0.0062249578,"teacher_disagreement_score":0.96459913,"about_ca_system_score_codex":0.013065534,"about_ca_system_score_gemma":0.01951845,"threshold_uncertainty_score":0.8181616},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015446508,0.9799952,0.0022644703,0.0004282292,0.00031345995,0.0004097297,0.00068738207,0.000031186333,0.00042378783],"genre_scores_gemma":[0.56729865,0.4133268,0.010889818,0.0031917484,0.0006369197,0.0021072584,0.0018186191,0.000146261,0.00058379915],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.9533495,0.031068347,0.0090371035,0.003258367,0.0023589064,0.00092786923],"domain_scores_gemma":[0.9065724,0.080084175,0.007627961,0.0031513092,0.0019209429,0.0006432043],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.04006322,0.0023435794,0.010462247,0.0029017613,0.00067562325,0.004574848,0.0018600064,0.003048858,0.0051392354],"category_scores_gemma":[0.076674215,0.0015711592,0.030879822,0.0040065944,0.0008339774,0.0022010521,0.0028147749,0.003048077,0.00040726046],"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":[0.00718409,0.00005654878,0.019917417,0.11978612,0.83092487,0.00014817045,0.00020176485,0.00037141214,0.00023409573,0.00039171462,0.0008419498,0.019941917],"study_design_scores_gemma":[0.0021970775,0.00026912033,0.011816732,0.010417211,0.9722367,0.00008027742,0.00009299653,0.00024039757,0.0000857792,0.00049950846,0.0020314648,0.000032736858],"about_ca_topic_score_codex":0.0034847818,"about_ca_topic_score_gemma":0.009662229,"teacher_disagreement_score":0.9895378,"about_ca_system_score_codex":0.001685501,"about_ca_system_score_gemma":0.0032715346,"threshold_uncertainty_score":0.21187705},"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":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.043479796,0.8629765,0.062662676,0.0044389013,0.0030865786,0.010589498,0.007563755,0.00092008844,0.004282218],"genre_scores_gemma":[0.543724,0.19764066,0.20692855,0.012731254,0.0019059621,0.025361655,0.009284078,0.0012730781,0.0011507511],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.47367162,0.2919834,0.1668929,0.022594597,0.042952865,0.0019045412],"domain_scores_gemma":[0.21468516,0.6137548,0.07914864,0.057830498,0.0334484,0.0011325269],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3605963,0.0056548147,0.016460778,0.027629806,0.0015299392,0.0143084675,0.0070448606,0.0073681935,0.007825898],"category_scores_gemma":[0.7473789,0.0053653517,0.04366566,0.03008031,0.0028646963,0.01669905,0.008424469,0.0060118134,0.0019104932],"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":[0.008354227,0.00016893294,0.025898853,0.32372156,0.5497878,0.0011694899,0.0023217255,0.002474045,0.0024838585,0.0030026268,0.0056602075,0.07495671],"study_design_scores_gemma":[0.00516384,0.0014849678,0.034339786,0.130972,0.79451233,0.002017098,0.0008087293,0.002145492,0.0022810628,0.00846447,0.017228303,0.0005818951],"about_ca_topic_score_codex":0.0024479772,"about_ca_topic_score_gemma":0.0046967426,"teacher_disagreement_score":0.9835392,"about_ca_system_score_codex":0.005216361,"about_ca_system_score_gemma":0.007859205,"threshold_uncertainty_score":0.78849864},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.027421102,0.86132836,0.049838457,0.010091655,0.0042424714,0.03900938,0.0043420475,0.0004444628,0.0032820348],"genre_scores_gemma":[0.56270087,0.18035807,0.10542065,0.018492961,0.0017088035,0.1263173,0.003622669,0.0002940302,0.0010846977],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.6467161,0.25285885,0.07077796,0.009215303,0.018240342,0.002191407],"domain_scores_gemma":[0.3188548,0.5689544,0.059337225,0.029640079,0.021722542,0.0014910108],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.31292766,0.00368507,0.014013268,0.008890427,0.0017071854,0.008342301,0.0042901062,0.006053237,0.008012274],"category_scores_gemma":[0.535077,0.003088128,0.03934707,0.010352506,0.0028374728,0.012742452,0.004126505,0.005103778,0.0007787614],"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":[0.0048948256,0.00031876066,0.023121305,0.59423816,0.32205033,0.00019296914,0.00150743,0.00097523455,0.00023389273,0.0023451154,0.0038595432,0.04626245],"study_design_scores_gemma":[0.0079381,0.0021158434,0.014985596,0.35415286,0.59385514,0.00033603277,0.0012453506,0.0020378164,0.00071287784,0.007874913,0.014508167,0.00023719882],"about_ca_topic_score_codex":0.0042003505,"about_ca_topic_score_gemma":0.0063376212,"teacher_disagreement_score":0.99631494,"about_ca_system_score_codex":0.006589778,"about_ca_system_score_gemma":0.013296221,"threshold_uncertainty_score":0.8472825},"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":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0074367435,0.9896953,0.0014494952,0.00016985081,0.00033948253,0.00014501906,0.00029166645,0.000029318537,0.0004431694],"genre_scores_gemma":[0.3199803,0.6657963,0.007306074,0.0031781115,0.0006587213,0.0005850048,0.0012348839,0.000118342985,0.0011423299],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.98431444,0.00856783,0.0036729784,0.0019213397,0.0012189605,0.00030449408],"domain_scores_gemma":[0.9582239,0.035868116,0.002645081,0.0020027836,0.0010796571,0.00018049369],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.023761472,0.0031932238,0.016425619,0.002494819,0.0005386229,0.0036822595,0.0022876433,0.003387957,0.00459011],"category_scores_gemma":[0.044662964,0.0017270509,0.03252099,0.003458475,0.00090428,0.0022821396,0.0013428931,0.0031785606,0.0005027911],"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":[0.0063140583,0.0000739807,0.007289291,0.110703506,0.8480944,0.00013819318,0.00008809639,0.00028780804,0.0004298123,0.0001800072,0.0005186361,0.02588217],"study_design_scores_gemma":[0.0017484349,0.0004400168,0.009758625,0.005105528,0.98002267,0.00011187939,0.0000471119,0.00014808132,0.00012735979,0.00033466748,0.002124076,0.00003161728],"about_ca_topic_score_codex":0.0032859286,"about_ca_topic_score_gemma":0.008721779,"teacher_disagreement_score":0.9968068,"about_ca_system_score_codex":0.0011060906,"about_ca_system_score_gemma":0.0013104753,"threshold_uncertainty_score":0.12566417},"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":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.018441008,0.9523998,0.02205837,0.0019494775,0.0018412267,0.0012035711,0.0005962003,0.0001599246,0.0013504719],"genre_scores_gemma":[0.7095451,0.23600823,0.03786852,0.0066233943,0.0025134017,0.004555601,0.0011913261,0.00039415047,0.0013003432],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.5959351,0.31305262,0.06023706,0.0142950835,0.014588291,0.0018917908],"domain_scores_gemma":[0.28068197,0.6653673,0.026803331,0.01949228,0.0068580867,0.0007969587],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2944574,0.003335493,0.021132069,0.0049315253,0.0009900923,0.009700516,0.0045128334,0.00811897,0.0041174423],"category_scores_gemma":[0.54864204,0.003377821,0.043465134,0.005404359,0.003901636,0.0071540354,0.0039670626,0.0086869635,0.00041914417],"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":[0.015461362,0.00013033552,0.0141380355,0.09921195,0.8132609,0.00030642218,0.00052466337,0.001629515,0.00029971867,0.0029939841,0.0016056015,0.050437555],"study_design_scores_gemma":[0.008080714,0.0009955582,0.008475207,0.015552772,0.95210385,0.00037484913,0.00016176194,0.0023214105,0.00048484982,0.007640546,0.0036939625,0.00011449939],"about_ca_topic_score_codex":0.0028011696,"about_ca_topic_score_gemma":0.0030449703,"teacher_disagreement_score":0.9966645,"about_ca_system_score_codex":0.003103421,"about_ca_system_score_gemma":0.0045563644,"threshold_uncertainty_score":0.8700596},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07483619,0.9111113,0.0020619712,0.0011960812,0.0004390655,0.0006536451,0.008478526,0.00006979644,0.001153415],"genre_scores_gemma":[0.79505056,0.19031806,0.0029177812,0.001861737,0.0006871073,0.0018247173,0.0069469176,0.000089672445,0.00030335394],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.87725186,0.04425632,0.050184764,0.011092336,0.01498371,0.0022309572],"domain_scores_gemma":[0.54305506,0.29291242,0.12870115,0.014976643,0.018650167,0.0017046534],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.09545379,0.0016204865,0.006434745,0.015246945,0.00085401314,0.0056640455,0.0034236328,0.0022554155,0.0025947986],"category_scores_gemma":[0.3088078,0.0016110007,0.024727225,0.020863876,0.0013769987,0.0053313915,0.0032507572,0.0020679317,0.00032493204],"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":[0.0033349127,0.000088478846,0.39363784,0.2148505,0.35963964,0.00030044327,0.00079471833,0.00062649354,0.0001445473,0.000398103,0.002821413,0.023362847],"study_design_scores_gemma":[0.0009953789,0.0008323657,0.33071312,0.091569446,0.5639688,0.0011511573,0.0010595875,0.0009747642,0.00049509335,0.0010099306,0.0070508188,0.00017950976],"about_ca_topic_score_codex":0.007430887,"about_ca_topic_score_gemma":0.008184402,"teacher_disagreement_score":0.99356526,"about_ca_system_score_codex":0.0034161655,"about_ca_system_score_gemma":0.004234162,"threshold_uncertainty_score":0.5048139},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.18275765,0.7309811,0.062239528,0.0038404814,0.007307708,0.005602061,0.0021097653,0.0005257407,0.004636001],"genre_scores_gemma":[0.937376,0.032185297,0.019207124,0.003886299,0.0017455752,0.003543324,0.00089822087,0.00018806638,0.00097018306],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.59584343,0.31820732,0.054408792,0.018478379,0.011242805,0.001819251],"domain_scores_gemma":[0.35572967,0.57957745,0.026911616,0.029563786,0.00702381,0.0011936816],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.24553376,0.003513057,0.015223637,0.005924531,0.0014075622,0.0066857478,0.003414427,0.007221949,0.0045954906],"category_scores_gemma":[0.5028963,0.0041044625,0.051760904,0.0057038735,0.0038312515,0.0056853327,0.0032478075,0.006046287,0.0005447514],"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":[0.030760111,0.00015877529,0.022263937,0.02251687,0.91380346,0.00020841022,0.00022523002,0.00057832134,0.00029789738,0.00084395433,0.0005994923,0.0077435668],"study_design_scores_gemma":[0.02098178,0.0024584765,0.014805169,0.004096781,0.95040154,0.0002547059,0.00013984417,0.0012498152,0.00032774388,0.0035417052,0.0016493463,0.000093129835],"about_ca_topic_score_codex":0.0021201456,"about_ca_topic_score_gemma":0.0023773648,"teacher_disagreement_score":0.99648696,"about_ca_system_score_codex":0.0019787333,"about_ca_system_score_gemma":0.0020459897,"threshold_uncertainty_score":0.9303912},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020234017,0.922825,0.048004396,0.0030548149,0.0019238403,0.001905623,0.0005388303,0.00017834353,0.0013352162],"genre_scores_gemma":[0.7426174,0.1665723,0.07295031,0.0061321026,0.0020130498,0.0074952347,0.0010253686,0.0002507127,0.00094350957],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5743311,0.35355842,0.043111403,0.012283217,0.015303952,0.0014118548],"domain_scores_gemma":[0.42417705,0.5070992,0.038835097,0.019659044,0.009322677,0.0009068276],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2694727,0.0037435547,0.014914433,0.010112368,0.001172255,0.008179554,0.0036659476,0.0060864203,0.0024657617],"category_scores_gemma":[0.42866236,0.002320188,0.053836115,0.007657735,0.0023637433,0.0047767907,0.004073156,0.004660575,0.00025251682],"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":[0.00424488,0.000068678564,0.02567949,0.06419703,0.8628185,0.0002103521,0.0002885996,0.0030688387,0.00021016944,0.0019536633,0.0010554839,0.03620439],"study_design_scores_gemma":[0.0038834931,0.00071082,0.01294536,0.026789317,0.9344085,0.00039981358,0.00016648893,0.0057633747,0.000509357,0.009884291,0.0044103884,0.00012882064],"about_ca_topic_score_codex":0.0030532528,"about_ca_topic_score_gemma":0.0037618487,"teacher_disagreement_score":0.9962565,"about_ca_system_score_codex":0.0037538726,"about_ca_system_score_gemma":0.0039044877,"threshold_uncertainty_score":0.9008702},"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":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.020038746,0.9148308,0.05484042,0.0031203593,0.0009618667,0.0025942728,0.0013619807,0.00017679489,0.0020748423],"genre_scores_gemma":[0.74580103,0.17615798,0.064671606,0.0038579078,0.0012089512,0.0061395434,0.0014259471,0.00025253106,0.00048455852],"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.3318441,0.5223263,0.09971047,0.020296765,0.023741081,0.0020812943],"domain_scores_gemma":[0.07766046,0.8433947,0.046914242,0.022501366,0.008867833,0.00066132506],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4224945,0.0050057387,0.015897222,0.017202245,0.0017030436,0.010475604,0.0062694238,0.005847166,0.005429806],"category_scores_gemma":[0.779615,0.004657128,0.06649881,0.02302469,0.0054687327,0.01686984,0.0076879277,0.006555471,0.0006634507],"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":[0.004190629,0.00013463858,0.07660557,0.20175457,0.6564378,0.0006975369,0.0015858109,0.0023449543,0.00024780928,0.003627419,0.0012894708,0.051083896],"study_design_scores_gemma":[0.0021352272,0.0017098491,0.030291785,0.120970525,0.8061093,0.0017373684,0.0012572025,0.004849798,0.00076447666,0.020024262,0.009873238,0.00027699396],"about_ca_topic_score_codex":0.0034392592,"about_ca_topic_score_gemma":0.0050342954,"teacher_disagreement_score":0.57750547,"about_ca_system_score_codex":0.005237917,"about_ca_system_score_gemma":0.0056870584,"threshold_uncertainty_score":0.71216714},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.029596044,0.93554205,0.018703764,0.0020407569,0.0024682647,0.008221055,0.0013075905,0.00015556517,0.0019649111],"genre_scores_gemma":[0.66166055,0.24767186,0.059430774,0.0034618324,0.0016676177,0.023465475,0.0019177385,0.00012316281,0.00060105504],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.48129758,0.41401604,0.07632426,0.008871492,0.018222293,0.001268313],"domain_scores_gemma":[0.4258801,0.51471525,0.039146062,0.010481659,0.008878917,0.00089801475],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.21822202,0.004074464,0.02588474,0.016527867,0.00087676244,0.0084884735,0.0034595036,0.0057750256,0.0030829972],"category_scores_gemma":[0.4802741,0.0028509384,0.046313316,0.009247375,0.0032292702,0.0048768986,0.0040208986,0.004018625,0.00030911746],"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":[0.007951856,0.00013880145,0.004675395,0.16611464,0.7944062,0.00010030986,0.0002634486,0.0016033401,0.00019797547,0.0010873834,0.000406333,0.023054365],"study_design_scores_gemma":[0.013489528,0.002488114,0.006451553,0.04992105,0.9172814,0.0001401345,0.00018082547,0.0020985967,0.00052152807,0.0049681976,0.0023191394,0.00013999759],"about_ca_topic_score_codex":0.0025058386,"about_ca_topic_score_gemma":0.0027951088,"teacher_disagreement_score":0.781778,"about_ca_system_score_codex":0.0050295424,"about_ca_system_score_gemma":0.005046077,"threshold_uncertainty_score":0.96407145},"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":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015485421,0.97764957,0.0020388472,0.0012207321,0.00047628224,0.0003044296,0.0022735475,0.000043400305,0.0005076868],"genre_scores_gemma":[0.50789744,0.4700512,0.009085977,0.0027354546,0.0011207941,0.002515747,0.005972543,0.000121553196,0.00049923954],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.79612136,0.09007979,0.07469391,0.01540847,0.02189679,0.0017997017],"domain_scores_gemma":[0.57715225,0.24000934,0.14164497,0.018229526,0.020729145,0.002234763],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.20573384,0.0013585058,0.00820613,0.0181867,0.0008502262,0.0068990043,0.0026111642,0.0032023948,0.0019283291],"category_scores_gemma":[0.29811412,0.0024958656,0.029006371,0.017375592,0.001505309,0.0065910956,0.0032764222,0.0026936375,0.00045998418],"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":[0.0065400656,0.000091966074,0.17781134,0.29941678,0.4397282,0.00019358797,0.0005878034,0.0011259841,0.0005143505,0.0016682924,0.0034402418,0.0688814],"study_design_scores_gemma":[0.0017228829,0.0012535799,0.20610978,0.11587629,0.6494489,0.0011239423,0.00067489775,0.0018954349,0.0010285346,0.0023074935,0.01834854,0.00020971741],"about_ca_topic_score_codex":0.0032870737,"about_ca_topic_score_gemma":0.0069624963,"teacher_disagreement_score":0.9986415,"about_ca_system_score_codex":0.00287389,"about_ca_system_score_gemma":0.0057497993,"threshold_uncertainty_score":0.97947156},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.019507796,0.85719913,0.09454573,0.0057517714,0.0016582788,0.015596208,0.0026967586,0.00028273958,0.0027615335],"genre_scores_gemma":[0.55831313,0.19912344,0.19303606,0.005481561,0.0015986344,0.03968038,0.0019398696,0.00019443415,0.00063249195],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5545799,0.3524344,0.05352275,0.013373719,0.024806693,0.001282562],"domain_scores_gemma":[0.37305745,0.55591124,0.03653033,0.020862678,0.012376676,0.0012617174],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2955369,0.0033648852,0.014874203,0.014534951,0.0011896426,0.009511434,0.0048377775,0.005858839,0.0048375893],"category_scores_gemma":[0.6038576,0.0033906726,0.027465025,0.013032181,0.0034872552,0.0069845906,0.0038393198,0.004003591,0.0006180256],"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":[0.004561684,0.00024153107,0.030771863,0.3876622,0.48711154,0.0005224769,0.0007952172,0.0051429626,0.00059258257,0.0077371993,0.0023091887,0.07255162],"study_design_scores_gemma":[0.0057086847,0.003142127,0.025673172,0.21838799,0.68199944,0.0010744013,0.00062693004,0.015167904,0.0012776821,0.033201404,0.013365033,0.00037527658],"about_ca_topic_score_codex":0.0017489777,"about_ca_topic_score_gemma":0.0022143975,"teacher_disagreement_score":0.99663514,"about_ca_system_score_codex":0.0045039007,"about_ca_system_score_gemma":0.008205552,"threshold_uncertainty_score":0.8687284},"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":"protocol","genre_gemma":"protocol","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0027633032,0.0073816027,0.019310584,0.0013982181,0.0011881277,0.9580701,0.008274358,0.00033526172,0.0012784924],"genre_scores_gemma":[0.0033803927,0.0010712668,0.016925659,0.0003754099,0.0000545851,0.97702014,0.00089644163,0.000026737509,0.00024930143],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.8502342,0.09047709,0.04310063,0.006593305,0.007643584,0.0019511717],"domain_scores_gemma":[0.86596215,0.07381083,0.017894123,0.022444244,0.018020224,0.0018683975],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.16617206,0.005681848,0.013333878,0.00960399,0.0022091349,0.0064484966,0.0038502614,0.007164094,0.023727823],"category_scores_gemma":[0.25639054,0.0039593223,0.049919587,0.008977322,0.0026934901,0.004094141,0.005506393,0.008167389,0.0032034046],"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":[0.05953558,0.0013398465,0.009064446,0.6470866,0.16788223,0.0013784245,0.0026509615,0.008269325,0.003346952,0.014092874,0.03175579,0.053596973],"study_design_scores_gemma":[0.15169908,0.008501128,0.025246594,0.21548557,0.29217032,0.0009937782,0.0017074092,0.014284119,0.006515684,0.054079045,0.22820528,0.001112087],"about_ca_topic_score_codex":0.00214288,"about_ca_topic_score_gemma":0.003874502,"teacher_disagreement_score":0.9943181,"about_ca_system_score_codex":0.006364002,"about_ca_system_score_gemma":0.019062791,"threshold_uncertainty_score":0.8788123},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021338463,0.89422554,0.069095545,0.002251208,0.0024574155,0.007100901,0.0018184903,0.00024298643,0.0014693623],"genre_scores_gemma":[0.67093986,0.16159412,0.12297357,0.0058389814,0.0022148932,0.032387964,0.002477135,0.00032230024,0.0012512524],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.553023,0.33561054,0.07220057,0.017253563,0.02043182,0.0014804042],"domain_scores_gemma":[0.44836792,0.46555448,0.05546111,0.019445108,0.010304548,0.0008668379],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27495456,0.0040110378,0.020127146,0.013516284,0.001126451,0.0075749676,0.004265252,0.0060623786,0.0030829299],"category_scores_gemma":[0.4646929,0.0026151203,0.04905964,0.009935109,0.0023921635,0.006060819,0.004581439,0.0037026608,0.00033371372],"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":[0.004635835,0.00007099158,0.02072775,0.14700358,0.7954223,0.00019933733,0.00038632232,0.0021740438,0.00026073278,0.0019396825,0.0011055003,0.02607389],"study_design_scores_gemma":[0.005392054,0.0010104747,0.011948753,0.05559899,0.9041945,0.0004405214,0.00020769483,0.0048972038,0.0007913012,0.01014595,0.0051896656,0.000182942],"about_ca_topic_score_codex":0.0019396481,"about_ca_topic_score_gemma":0.0023521876,"teacher_disagreement_score":0.99598897,"about_ca_system_score_codex":0.0036308235,"about_ca_system_score_gemma":0.004035911,"threshold_uncertainty_score":0.8941101},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016317954,0.9727296,0.006672631,0.00080020336,0.0006207579,0.001327996,0.00090663077,0.00007642369,0.0005478065],"genre_scores_gemma":[0.68962866,0.27328402,0.023602419,0.0027230103,0.0013860445,0.006798961,0.0017922515,0.00015927538,0.0006252641],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.87152594,0.077020206,0.032448348,0.008856333,0.008941423,0.0012077439],"domain_scores_gemma":[0.7562602,0.19877212,0.029722022,0.007867597,0.0064182905,0.0009597564],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.102775425,0.003960193,0.019177293,0.00894196,0.00082466565,0.0060300743,0.003091102,0.004198546,0.0039544194],"category_scores_gemma":[0.21283299,0.0024646313,0.06587533,0.009538807,0.0018219538,0.004067805,0.0027858168,0.0037249434,0.0003758918],"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":[0.0045918045,0.00003650635,0.01346164,0.09305917,0.8817271,0.00008988079,0.000085847256,0.00044380015,0.00018955136,0.00025801314,0.0002569171,0.00579971],"study_design_scores_gemma":[0.0015431445,0.0003416139,0.0072470997,0.010073707,0.97868663,0.00010090092,0.00004783886,0.0003772361,0.00012585832,0.00064906257,0.0007764118,0.000030446387],"about_ca_topic_score_codex":0.002151804,"about_ca_topic_score_gemma":0.0029867424,"teacher_disagreement_score":0.9960398,"about_ca_system_score_codex":0.0021511794,"about_ca_system_score_gemma":0.003133684,"threshold_uncertainty_score":0.5435349},"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":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07259309,0.7599143,0.13563804,0.014048181,0.004762994,0.0060245413,0.0013071193,0.00047067902,0.0052410904],"genre_scores_gemma":[0.67572343,0.09723771,0.20566167,0.008544596,0.0018030117,0.009344206,0.00074471725,0.0003418459,0.00059885555],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.23611733,0.66761416,0.055996377,0.0096991295,0.029705776,0.0008672691],"domain_scores_gemma":[0.07915136,0.85974246,0.031956133,0.019031867,0.009654375,0.00046374576],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.52488333,0.0047944183,0.010054894,0.013212399,0.0017836511,0.013354538,0.0053790645,0.010224596,0.0024265368],"category_scores_gemma":[0.8112393,0.003440132,0.0466065,0.0103593785,0.0044246293,0.009911896,0.0052897,0.00863645,0.0003976717],"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":[0.018402154,0.00035817298,0.043099377,0.09420867,0.67202526,0.00079606264,0.0035138705,0.011045581,0.0012114411,0.009483041,0.0038987657,0.14195758],"study_design_scores_gemma":[0.016322209,0.005534968,0.03397552,0.096296035,0.7514195,0.0019067223,0.0017108571,0.023487408,0.0048072264,0.050429583,0.012790962,0.0013189776],"about_ca_topic_score_codex":0.0016326802,"about_ca_topic_score_gemma":0.002557819,"teacher_disagreement_score":0.9899451,"about_ca_system_score_codex":0.0050377194,"about_ca_system_score_gemma":0.003658023,"threshold_uncertainty_score":0.5859035},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0041060876,0.974442,0.011863859,0.00144757,0.0014019154,0.0050886455,0.0007842194,0.00014800324,0.00071768067],"genre_scores_gemma":[0.33384162,0.5626088,0.053132135,0.006472404,0.0021797612,0.03911738,0.0015076542,0.00036178215,0.00077857426],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5087068,0.3403097,0.10207275,0.016181313,0.0304803,0.0022491757],"domain_scores_gemma":[0.40500537,0.49809808,0.061215937,0.018244071,0.016219433,0.0012171996],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28436247,0.00606214,0.029588504,0.014707067,0.0018993374,0.009593459,0.0054420307,0.007467702,0.0045041186],"category_scores_gemma":[0.49657547,0.004064127,0.05436026,0.013128436,0.005688641,0.009669267,0.004743444,0.0063236495,0.0005759949],"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":[0.0032949718,0.000055478544,0.003380594,0.5568624,0.40832195,0.00012953821,0.00026954658,0.001169468,0.0001769676,0.0011112833,0.0012005712,0.02402715],"study_design_scores_gemma":[0.005555801,0.00062474544,0.0039886436,0.21770234,0.75939244,0.00022793993,0.00013672908,0.0020315123,0.0007126069,0.0056357575,0.0038229316,0.00016852403],"about_ca_topic_score_codex":0.005304501,"about_ca_topic_score_gemma":0.008023398,"teacher_disagreement_score":0.9704115,"about_ca_system_score_codex":0.00980934,"about_ca_system_score_gemma":0.013890129,"threshold_uncertainty_score":0.8825085},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.015372933,0.97251713,0.0045521483,0.0031517602,0.00041174385,0.0009671359,0.0017584934,0.000047779362,0.0012209531],"genre_scores_gemma":[0.6456736,0.32883728,0.014907386,0.0042645917,0.00065893016,0.0029092613,0.0024095979,0.000087355525,0.00025198382],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.77036715,0.14781995,0.05167039,0.010810576,0.017302176,0.0020297524],"domain_scores_gemma":[0.37370515,0.5303789,0.066870466,0.010627828,0.017194953,0.0012227212],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.15812522,0.0025587901,0.012064669,0.010856081,0.0010112259,0.008040133,0.004364097,0.004993646,0.0034787797],"category_scores_gemma":[0.47788,0.002267059,0.033550087,0.012917451,0.0028042358,0.0057236217,0.0036702328,0.0041466514,0.00029323384],"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":[0.0025196362,0.00006426034,0.052766293,0.3231842,0.5769976,0.00023563887,0.0008219713,0.0016020916,0.00012678839,0.0010557873,0.002044254,0.03858149],"study_design_scores_gemma":[0.0015580946,0.00034249554,0.034989774,0.14063177,0.8131594,0.00029649594,0.00035351785,0.0010605237,0.00026009732,0.0028038814,0.0044566463,0.000087211716],"about_ca_topic_score_codex":0.013346776,"about_ca_topic_score_gemma":0.015390922,"teacher_disagreement_score":0.99744123,"about_ca_system_score_codex":0.006077747,"about_ca_system_score_gemma":0.007092623,"threshold_uncertainty_score":0.8362561},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011878993,0.96866745,0.014050637,0.0013281718,0.0009907848,0.001570396,0.00074184674,0.00009669652,0.0006750238],"genre_scores_gemma":[0.6566767,0.2760178,0.045389272,0.0051116184,0.0026218025,0.010778326,0.0023403496,0.00031926448,0.0007448529],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.41758454,0.38532194,0.14894286,0.017215705,0.028688092,0.0022467808],"domain_scores_gemma":[0.26451632,0.5788407,0.110947505,0.03001425,0.014833561,0.0008476362],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3791918,0.0054949545,0.022186328,0.013356335,0.0016057411,0.012312471,0.004421219,0.0063828933,0.0031413557],"category_scores_gemma":[0.5903142,0.0038382078,0.056641012,0.0149877835,0.0038007281,0.009537014,0.0055533685,0.004323559,0.00043725336],"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":[0.004311949,0.000059222646,0.031520184,0.13932931,0.80524915,0.0001586586,0.0003004879,0.00086848123,0.00021187784,0.00077160867,0.00079786475,0.016421158],"study_design_scores_gemma":[0.0032658442,0.00074173306,0.020827973,0.055238366,0.9093675,0.0005021474,0.00018394724,0.0020174377,0.0006620203,0.0033861012,0.0036313864,0.00017553908],"about_ca_topic_score_codex":0.002448054,"about_ca_topic_score_gemma":0.0041812216,"teacher_disagreement_score":0.99450505,"about_ca_system_score_codex":0.004941489,"about_ca_system_score_gemma":0.00492107,"threshold_uncertainty_score":0.76556706},"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":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008823258,0.004883545,0.04720296,0.0016308131,0.0027331915,0.9349477,0.0052027423,0.0013031776,0.001213456],"genre_scores_gemma":[0.0011602184,0.00061656826,0.054788247,0.00024493109,0.0000590753,0.9424647,0.0004072095,0.000068329515,0.00019076037],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.37808928,0.42593694,0.15188168,0.0180226,0.023157645,0.0029119481],"domain_scores_gemma":[0.42448208,0.3933179,0.062103335,0.06568301,0.04983744,0.0045762463],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.41711387,0.012377353,0.026931405,0.019064613,0.004109305,0.013189422,0.009493741,0.019052958,0.032251067],"category_scores_gemma":[0.64862096,0.0099073015,0.06013873,0.022556892,0.008054997,0.010276324,0.009762772,0.02286166,0.009475463],"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":[0.029330557,0.0009065094,0.0035481348,0.70924175,0.086865894,0.0011884389,0.0044821603,0.008561665,0.0027247802,0.019827694,0.055027835,0.07829453],"study_design_scores_gemma":[0.11917233,0.006104338,0.011657629,0.3747436,0.12615697,0.0014086213,0.001665449,0.036841992,0.008750486,0.10086563,0.21056217,0.0020708404],"about_ca_topic_score_codex":0.0021833875,"about_ca_topic_score_gemma":0.0031044066,"teacher_disagreement_score":0.9876226,"about_ca_system_score_codex":0.012472298,"about_ca_system_score_gemma":0.030458145,"threshold_uncertainty_score":0.7188024},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"evaluation","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"evaluation","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.040349048,0.9382005,0.0152923325,0.0011222942,0.0008148116,0.0016214947,0.0016187273,0.0001370816,0.00084375724],"genre_scores_gemma":[0.85364,0.11862928,0.019104952,0.0022109165,0.00097328064,0.003037792,0.0018860559,0.00015796813,0.00035972946],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.51311594,0.28649786,0.14310528,0.027082074,0.027135538,0.00306335],"domain_scores_gemma":[0.20788993,0.6847789,0.06414094,0.026864307,0.014500353,0.0018256423],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.28884494,0.0061667296,0.03756827,0.023920422,0.002090979,0.013258383,0.0062117004,0.009286065,0.0041216663],"category_scores_gemma":[0.6354684,0.0072768633,0.0877481,0.016900321,0.0041078962,0.012286399,0.011003157,0.007926853,0.0004538727],"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":[0.0042372956,0.000051887695,0.021316394,0.054808125,0.91325194,0.00012983638,0.00030705947,0.00043826678,0.00010113732,0.00025979776,0.00023305202,0.0048652813],"study_design_scores_gemma":[0.0025255845,0.00056494283,0.012331576,0.011174296,0.9689242,0.00037549727,0.0002046222,0.0007742796,0.00015639319,0.001984097,0.00084442215,0.00014001531],"about_ca_topic_score_codex":0.0055780048,"about_ca_topic_score_gemma":0.006120303,"teacher_disagreement_score":0.99383324,"about_ca_system_score_codex":0.0053193304,"about_ca_system_score_gemma":0.005125978,"threshold_uncertainty_score":0.8769808},"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":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0013464926,0.9661383,0.015191668,0.013603154,0.0023669212,0.00035735883,0.00026182484,0.000074180294,0.0006600926],"genre_scores_gemma":[0.12867478,0.7631916,0.05415621,0.038517877,0.010537201,0.0035098875,0.00068108324,0.00034802253,0.00038331363],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.4396459,0.36619708,0.13187957,0.023365352,0.03622218,0.0026899665],"domain_scores_gemma":[0.05708273,0.8809866,0.029904675,0.016700879,0.014757334,0.00056770333],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4975832,0.0043045618,0.023345536,0.016711723,0.0021751048,0.017718904,0.0069426317,0.011261236,0.0045587444],"category_scores_gemma":[0.79682046,0.004190436,0.022453295,0.01520794,0.008804685,0.026259039,0.008790448,0.013006049,0.000825276],"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":[0.0018440018,0.00009315094,0.012682342,0.55251366,0.12616828,0.0004730189,0.0035475066,0.0019932871,0.00039922964,0.017113954,0.010099961,0.27307165],"study_design_scores_gemma":[0.0011299222,0.0006974183,0.008539202,0.7005482,0.1506387,0.0010477845,0.0014555969,0.0021837214,0.0012228354,0.07410577,0.057921138,0.0005097453],"about_ca_topic_score_codex":0.005435969,"about_ca_topic_score_gemma":0.0066984487,"teacher_disagreement_score":0.97665447,"about_ca_system_score_codex":0.008190667,"about_ca_system_score_gemma":0.013719797,"threshold_uncertainty_score":0.6195694},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012056513,0.9844742,0.0019186804,0.00022990735,0.0003221477,0.00033316683,0.00033447018,0.000037849954,0.0002930163],"genre_scores_gemma":[0.43332613,0.5545474,0.0070212786,0.0013757547,0.0006732862,0.0012133926,0.0010334331,0.00012005623,0.0006892077],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9650111,0.018810207,0.009352707,0.0034310843,0.0026827406,0.00071214576],"domain_scores_gemma":[0.91401327,0.074073255,0.0065915943,0.0025593885,0.002264315,0.0004982541],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.038781364,0.004138564,0.020136163,0.0040106284,0.00082343206,0.00467941,0.0029645928,0.0033525468,0.005007969],"category_scores_gemma":[0.086621016,0.0021337406,0.066799216,0.0059162634,0.0015023183,0.0027490621,0.0025427432,0.005171557,0.0004340109],"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":[0.005493706,0.000033249144,0.006030771,0.062388193,0.91756064,0.000061885316,0.000065629705,0.00032089325,0.00012597856,0.000118089534,0.00016823747,0.0076327063],"study_design_scores_gemma":[0.0012373503,0.0001734588,0.002804571,0.0032105187,0.9916951,0.00003579719,0.000029267794,0.00014329287,0.00005706818,0.00021348253,0.00038794652,0.000012190409],"about_ca_topic_score_codex":0.0059921816,"about_ca_topic_score_gemma":0.007821163,"teacher_disagreement_score":0.9958614,"about_ca_system_score_codex":0.0020153392,"about_ca_system_score_gemma":0.0032777213,"threshold_uncertainty_score":0.20509791},"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":"protocol","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"protocol","genre_consensus":"protocol","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011218678,0.019293943,0.016062507,0.000602471,0.0010297749,0.93509954,0.013625321,0.00081651605,0.002251288],"genre_scores_gemma":[0.023798626,0.0021349888,0.012956651,0.00043266627,0.00011076535,0.95788896,0.0017355623,0.00010389022,0.00083791796],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.89261127,0.057195198,0.031802032,0.007914811,0.0073972903,0.0030794467],"domain_scores_gemma":[0.91957533,0.042668253,0.020025637,0.00916768,0.0067877555,0.0017752909],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.07974789,0.007954568,0.02432784,0.0074175056,0.0031884657,0.0064005386,0.0037010482,0.007001667,0.043112885],"category_scores_gemma":[0.12714563,0.0063902945,0.038106777,0.005746009,0.0045332643,0.0042468244,0.0051153377,0.009909093,0.0043949024],"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":[0.3405632,0.001591327,0.00625427,0.3862136,0.20599835,0.00046384113,0.0010599229,0.0028781195,0.0023564762,0.0060948823,0.013690769,0.03283527],"study_design_scores_gemma":[0.6732851,0.009576448,0.017272972,0.042667743,0.20336848,0.0003216508,0.00043093358,0.0046224385,0.0022101928,0.00863122,0.03702763,0.00058524834],"about_ca_topic_score_codex":0.0029327075,"about_ca_topic_score_gemma":0.006433535,"teacher_disagreement_score":0.9756722,"about_ca_system_score_codex":0.006216636,"about_ca_system_score_gemma":0.011388161,"threshold_uncertainty_score":0.4217522},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0031317226,0.9907031,0.0028920171,0.00087652006,0.00054286275,0.0005371992,0.00078309607,0.00003380203,0.0004996185],"genre_scores_gemma":[0.26363838,0.70556736,0.0129103875,0.0065139807,0.0020239176,0.0060910536,0.0020936343,0.00015073446,0.00101054],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.9216534,0.04039764,0.02465328,0.0056376834,0.006738787,0.0009191879],"domain_scores_gemma":[0.81492037,0.14436112,0.025841683,0.00695714,0.007332887,0.00058672804],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.07509449,0.003234221,0.014811011,0.012337285,0.001212254,0.0064318376,0.0024964735,0.0030810505,0.004838774],"category_scores_gemma":[0.17705043,0.001865583,0.030773658,0.012247705,0.0014056247,0.003387347,0.0029234914,0.0018964643,0.0004991324],"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":[0.0013616813,0.000029746532,0.010154209,0.27647513,0.6879085,0.00028869393,0.00017068797,0.00060780894,0.0002830372,0.00058718614,0.0016782759,0.02045506],"study_design_scores_gemma":[0.00066313724,0.00016255729,0.0076331836,0.052965034,0.93121374,0.00024120558,0.000081650585,0.00056043715,0.00022133421,0.0016254559,0.004588398,0.00004378605],"about_ca_topic_score_codex":0.0036626712,"about_ca_topic_score_gemma":0.0060367775,"teacher_disagreement_score":0.9967658,"about_ca_system_score_codex":0.0031596557,"about_ca_system_score_gemma":0.0045548445,"threshold_uncertainty_score":0.39714235},"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":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0077699586,0.8838202,0.082606524,0.014266513,0.004679023,0.002636837,0.0016015104,0.00038215905,0.0022373127],"genre_scores_gemma":[0.50518066,0.30415305,0.15000956,0.015478463,0.0068015666,0.014916812,0.001668962,0.0008772412,0.00091364305],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.41211212,0.5115421,0.04076893,0.015350517,0.018689,0.0015373022],"domain_scores_gemma":[0.24629404,0.6945866,0.030533124,0.019738339,0.008163311,0.0006846809],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36324233,0.003999351,0.017195128,0.013471672,0.0014116028,0.011256437,0.0057895244,0.006624019,0.005096776],"category_scores_gemma":[0.6638061,0.003327116,0.058099296,0.015878776,0.004035089,0.0100772735,0.00521838,0.0071461033,0.000831424],"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":[0.002756306,0.000074971336,0.021218697,0.109864965,0.7623198,0.0003593557,0.0008442846,0.00471823,0.0002682058,0.0076593934,0.0064860727,0.083429754],"study_design_scores_gemma":[0.004995996,0.0010493206,0.014616816,0.08304467,0.8079356,0.00054251193,0.0005757509,0.010135212,0.0009075606,0.058225896,0.017555444,0.0004152837],"about_ca_topic_score_codex":0.0034753643,"about_ca_topic_score_gemma":0.0044005243,"teacher_disagreement_score":0.99600065,"about_ca_system_score_codex":0.0053309435,"about_ca_system_score_gemma":0.007016801,"threshold_uncertainty_score":0.7852356},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19711217,0.48143962,0.27005613,0.008663202,0.006207007,0.0051598367,0.016329465,0.0017756716,0.01325684],"genre_scores_gemma":[0.9605279,0.009975905,0.017671471,0.0027702628,0.0011600798,0.0023504067,0.0031809856,0.00055061525,0.001812521],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.59184194,0.33182365,0.03965371,0.023093605,0.01165155,0.0019355805],"domain_scores_gemma":[0.20108715,0.6966075,0.03998591,0.053102806,0.008077019,0.001139559],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2922475,0.0035498568,0.010885947,0.0052148565,0.0010254034,0.009133868,0.0067835487,0.006981043,0.014735965],"category_scores_gemma":[0.60529476,0.0037349872,0.038313344,0.008908444,0.0048186076,0.0078084157,0.003970312,0.008774421,0.001911259],"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":[0.026530823,0.00011522886,0.046106618,0.021821963,0.8685399,0.00042307007,0.0004741879,0.0025217934,0.00034169783,0.003156449,0.0033508064,0.026617495],"study_design_scores_gemma":[0.02008975,0.002709709,0.047043454,0.0069673103,0.887504,0.0014174333,0.00028995582,0.008535485,0.0011042325,0.016932061,0.0071320958,0.0002744838],"about_ca_topic_score_codex":0.0014231218,"about_ca_topic_score_gemma":0.0007975648,"teacher_disagreement_score":0.9964501,"about_ca_system_score_codex":0.0016619603,"about_ca_system_score_gemma":0.0013922964,"threshold_uncertainty_score":0.87278485},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0012933015,0.99633944,0.00068270916,0.00046530928,0.00018874463,0.00058047264,0.00025048776,0.000007590429,0.00019190331],"genre_scores_gemma":[0.0607488,0.9249321,0.006550255,0.0019082893,0.0003445468,0.004577169,0.0007343627,0.000031333566,0.00017321714],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.86019045,0.055821486,0.06357086,0.005488045,0.014045611,0.0008836004],"domain_scores_gemma":[0.5796017,0.3340613,0.06373059,0.0068877395,0.014893661,0.00082500326],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.10561095,0.0020707764,0.011256937,0.014250981,0.001177801,0.005153706,0.004476358,0.0038375456,0.003150479],"category_scores_gemma":[0.3272092,0.0020885656,0.019776514,0.014387123,0.002061462,0.003955005,0.0031306276,0.0020558694,0.0002991274],"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":[0.0008077614,0.00003033479,0.0039872145,0.91523165,0.04441147,0.000070623115,0.00026002293,0.00018138511,0.00015097864,0.00049032347,0.0009543645,0.033423968],"study_design_scores_gemma":[0.0009198011,0.00042225517,0.01117663,0.73338133,0.23655117,0.00037130117,0.00038653106,0.00024400293,0.00044524047,0.0014855518,0.014544659,0.00007151221],"about_ca_topic_score_codex":0.005148088,"about_ca_topic_score_gemma":0.011166858,"teacher_disagreement_score":0.89438903,"about_ca_system_score_codex":0.0064270976,"about_ca_system_score_gemma":0.012087755,"threshold_uncertainty_score":0.5585308},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.028608017,0.94139004,0.016748248,0.004297728,0.0021629573,0.0013717005,0.003453839,0.0002108397,0.0017566466],"genre_scores_gemma":[0.787738,0.17525195,0.01753287,0.007847581,0.0024356656,0.0046635442,0.0028906139,0.00020996989,0.0014298541],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.9485379,0.029161207,0.012387746,0.0045926506,0.0042632613,0.0010572418],"domain_scores_gemma":[0.83938354,0.13186064,0.015811782,0.0068131853,0.005527969,0.00060293986],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.07409009,0.0025505668,0.00939078,0.004972439,0.0008850941,0.0047840616,0.002193951,0.00310311,0.0056257234],"category_scores_gemma":[0.1305673,0.0013342362,0.037150137,0.0065810345,0.0010998521,0.002728856,0.001968239,0.0024614388,0.0003932656],"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":[0.004457353,0.000044598746,0.046586215,0.11310744,0.8104276,0.0005090673,0.00022213835,0.001166936,0.0005217412,0.00091487257,0.003157334,0.018884731],"study_design_scores_gemma":[0.0013335733,0.00026742805,0.020289952,0.012793039,0.95713305,0.00030679585,0.00008848327,0.0013997583,0.0004747186,0.0022154714,0.0036380396,0.000059654594],"about_ca_topic_score_codex":0.002950201,"about_ca_topic_score_gemma":0.0045898054,"teacher_disagreement_score":0.99744946,"about_ca_system_score_codex":0.0017685033,"about_ca_system_score_gemma":0.0033150446,"threshold_uncertainty_score":0.39183056},"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":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007116962,0.7206984,0.2509719,0.0041366196,0.0037219955,0.00933565,0.0012922605,0.0006347924,0.0020914334],"genre_scores_gemma":[0.2791634,0.18671246,0.48384386,0.005019785,0.0031201073,0.039277986,0.0015124561,0.00073820545,0.0006118444],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.11992395,0.75267434,0.08593093,0.014524214,0.026108973,0.0008376972],"domain_scores_gemma":[0.083321616,0.8276784,0.042498842,0.03177748,0.014156571,0.0005671473],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.59441227,0.007595165,0.03328933,0.03957043,0.0023319144,0.018214336,0.008183176,0.008986355,0.0054124263],"category_scores_gemma":[0.86905533,0.007350706,0.07542479,0.039855853,0.005168329,0.019378548,0.012136445,0.009238432,0.0009082945],"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":[0.0025530553,0.00013203264,0.015462685,0.27747372,0.5891928,0.000369486,0.0017075419,0.004945986,0.0004068451,0.0076215286,0.0022766725,0.09785772],"study_design_scores_gemma":[0.00324755,0.0012823911,0.008778485,0.13955443,0.79348385,0.00057869626,0.00049724156,0.014202765,0.0010633824,0.026337285,0.010503058,0.00047076578],"about_ca_topic_score_codex":0.0028153558,"about_ca_topic_score_gemma":0.0035519165,"teacher_disagreement_score":0.9924048,"about_ca_system_score_codex":0.008176958,"about_ca_system_score_gemma":0.011184864,"threshold_uncertainty_score":0.5001619},"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":"review","genre_gemma":"empirical","domain_codex":"methods","domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":"methods","prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.030023178,0.7932782,0.14433591,0.008749559,0.00401087,0.013163911,0.001725378,0.00043960876,0.0042733946],"genre_scores_gemma":[0.76350445,0.0684234,0.11885251,0.009603154,0.0022482486,0.034775276,0.0014988008,0.00025465657,0.0008396173],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.38582116,0.53888655,0.04982474,0.008688702,0.015815465,0.00096333],"domain_scores_gemma":[0.2791421,0.6478822,0.044613127,0.020415992,0.0072134756,0.00073311495],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36804473,0.0032836916,0.014442075,0.0072657415,0.0008886708,0.0067263558,0.0042719063,0.0074701584,0.004143189],"category_scores_gemma":[0.5986807,0.002335557,0.034162354,0.0073304987,0.0027302308,0.007033708,0.003051706,0.0060672853,0.0005118194],"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":[0.03729749,0.00021910928,0.020185905,0.16174252,0.6798503,0.00038412635,0.0006895188,0.005188847,0.00077615044,0.010875052,0.004522747,0.078268245],"study_design_scores_gemma":[0.032039296,0.003102367,0.011492654,0.0545486,0.82813114,0.00052928476,0.0003007086,0.014518269,0.00162498,0.04062272,0.012757381,0.00033272192],"about_ca_topic_score_codex":0.0006443718,"about_ca_topic_score_gemma":0.0008115879,"teacher_disagreement_score":0.9967163,"about_ca_system_score_codex":0.0024873635,"about_ca_system_score_gemma":0.00242627,"threshold_uncertainty_score":0.7793133},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014540433,0.99782723,0.0001909563,0.00007745226,0.000086804655,0.00004710247,0.00024823943,0.000008150938,0.00005998023],"genre_scores_gemma":[0.039683674,0.95734465,0.0013579101,0.000671732,0.00014208563,0.00017293228,0.0004669733,0.000014937174,0.00014516103],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9943997,0.0018269297,0.00219987,0.00080378004,0.0005818878,0.00018768552],"domain_scores_gemma":[0.97752017,0.016823975,0.00365042,0.0007321052,0.0010406817,0.00023262341],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.011478402,0.0020540631,0.010890472,0.005376828,0.0005020737,0.003461234,0.0019838165,0.0025627874,0.0049199234],"category_scores_gemma":[0.033718884,0.0020857798,0.021855058,0.007285721,0.0007585496,0.0022140604,0.0022639225,0.002163487,0.00039258986],"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":[0.0011069869,0.00008554033,0.022883285,0.4750574,0.44712242,0.00015906396,0.00018089556,0.0003131177,0.0003174371,0.00024819613,0.0014345865,0.051091075],"study_design_scores_gemma":[0.0004175429,0.00024301329,0.023423456,0.08764132,0.8800185,0.00018096418,0.00016356815,0.00012585272,0.0001504379,0.00044097955,0.007137134,0.000057294255],"about_ca_topic_score_codex":0.009507512,"about_ca_topic_score_gemma":0.025308894,"teacher_disagreement_score":0.9891095,"about_ca_system_score_codex":0.0014304237,"about_ca_system_score_gemma":0.004276312,"threshold_uncertainty_score":0.06070435},"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":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0136157265,0.96404594,0.015250067,0.0023946657,0.0014047492,0.001319823,0.00087339856,0.00010035537,0.0009953871],"genre_scores_gemma":[0.738074,0.1946546,0.043255676,0.009130982,0.0027130789,0.0098103015,0.0013523208,0.00024061052,0.00076841575],"study_design_codex":"meta_analysis","study_design_gemma":"systematic_review","domain_scores_codex":[0.55382687,0.34082156,0.07183878,0.015489805,0.016035236,0.0019878384],"domain_scores_gemma":[0.30286002,0.61201,0.0573642,0.018670985,0.008064168,0.0010305712],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27638158,0.0029287308,0.015259961,0.008906798,0.0012175523,0.007023036,0.0033870651,0.004776938,0.0039419285],"category_scores_gemma":[0.42517027,0.0018894821,0.05346739,0.010652134,0.0026074986,0.0061200014,0.003007488,0.0036647078,0.0003567228],"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":[0.0056775864,0.00006342297,0.050281428,0.13321348,0.7818112,0.0002775538,0.00037588374,0.0010754808,0.0002260024,0.002043163,0.0012246763,0.023730125],"study_design_scores_gemma":[0.0032026432,0.0011583806,0.026331743,0.056452163,0.8932977,0.00048083262,0.0003116492,0.0021489302,0.0005023124,0.008970038,0.0070165494,0.00012710273],"about_ca_topic_score_codex":0.0017453266,"about_ca_topic_score_gemma":0.0024224678,"teacher_disagreement_score":0.99707127,"about_ca_system_score_codex":0.0030554961,"about_ca_system_score_gemma":0.0035912849,"threshold_uncertainty_score":0.8923503},"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":"reporting","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"reporting","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.004759124,0.9738273,0.0119324215,0.0019039145,0.001257728,0.003946792,0.0012117423,0.00010107656,0.001059838],"genre_scores_gemma":[0.3246876,0.5664148,0.05468537,0.011789525,0.0018481498,0.03621179,0.0029297732,0.00027046417,0.0011624785],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5718687,0.25876173,0.12569055,0.015884982,0.025984723,0.0018093504],"domain_scores_gemma":[0.41884273,0.48163855,0.06267885,0.019254757,0.01663449,0.0009506881],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.27275917,0.002953997,0.018527037,0.008154755,0.0016656932,0.007666437,0.0036917077,0.0045274603,0.0046691876],"category_scores_gemma":[0.46814972,0.0027175273,0.027865857,0.011436196,0.003399065,0.0059678312,0.0034539911,0.0036791223,0.0005166873],"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":[0.0017144344,0.00006039441,0.009915896,0.6528687,0.2881564,0.00023302856,0.00054777804,0.00066973513,0.00024513516,0.002404852,0.0024590492,0.04072457],"study_design_scores_gemma":[0.0040540993,0.00082296703,0.015795516,0.31742373,0.6238986,0.0006040369,0.0004729572,0.0018033322,0.0011223359,0.012603743,0.021147914,0.00025064405],"about_ca_topic_score_codex":0.00502745,"about_ca_topic_score_gemma":0.0083276,"teacher_disagreement_score":0.997046,"about_ca_system_score_codex":0.007156826,"about_ca_system_score_gemma":0.011932893,"threshold_uncertainty_score":0.89681745},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0066176234,0.92181075,0.05989953,0.0021968784,0.001531181,0.0035880331,0.0025423078,0.00027816682,0.0015355117],"genre_scores_gemma":[0.5548051,0.29499057,0.10782666,0.0075626373,0.002622866,0.025711702,0.00462058,0.0006368828,0.0012228686],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.5466023,0.32941648,0.08022847,0.017746601,0.024526339,0.0014798709],"domain_scores_gemma":[0.2216016,0.7013493,0.04309535,0.024897235,0.008431614,0.000624763],"candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.34069481,0.006109349,0.024570493,0.019512208,0.001075744,0.009875962,0.006438392,0.0063927076,0.0059187934],"category_scores_gemma":[0.65579885,0.003247391,0.06299295,0.01546796,0.004546705,0.010441066,0.006663999,0.0064329524,0.00069061556],"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":[0.0016218234,0.000037888618,0.0082120085,0.22680536,0.73018795,0.00021038405,0.00032864203,0.0030416877,0.00015567344,0.0026116197,0.0012486866,0.0255383],"study_design_scores_gemma":[0.0018793094,0.0005979677,0.0074874596,0.073529854,0.87674177,0.00044560846,0.00025515232,0.003939254,0.00068242784,0.026121505,0.008093697,0.00022599722],"about_ca_topic_score_codex":0.0026520882,"about_ca_topic_score_gemma":0.0033534549,"teacher_disagreement_score":0.97542953,"about_ca_system_score_codex":0.0041103256,"about_ca_system_score_gemma":0.00457399,"threshold_uncertainty_score":0.8130407},"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":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"incentives","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":"review","domain_candidate":"incentives","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0011712805,0.9836481,0.0018068812,0.006611238,0.0014641635,0.0025944281,0.0013425892,0.000065712185,0.0012957695],"genre_scores_gemma":[0.034507684,0.93173605,0.014938472,0.005789463,0.0017549476,0.009275964,0.0012076352,0.000105534404,0.0006842766],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.70882016,0.13756905,0.09833068,0.007811281,0.045405544,0.0020632285],"domain_scores_gemma":[0.20598064,0.59450525,0.12550633,0.01098362,0.060961477,0.0020627584],"candidate_categories":["metaresearch","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22615612,0.0015877706,0.008427378,0.01935842,0.0012885389,0.008532168,0.003257671,0.005234063,0.008167794],"category_scores_gemma":[0.58770597,0.0023844116,0.008800059,0.018200016,0.0028489043,0.0096136285,0.0039582993,0.0035199297,0.0013689129],"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":[0.0013568315,0.000045209687,0.0017797671,0.76164913,0.007854154,0.000118855096,0.00045466932,0.00015467413,0.00031663457,0.002136567,0.012813805,0.21131964],"study_design_scores_gemma":[0.0017579542,0.00039396662,0.005233727,0.88916874,0.021724788,0.00037445463,0.00029259946,0.00028778528,0.00047823053,0.002665731,0.077534474,0.000087544664],"about_ca_topic_score_codex":0.0029577957,"about_ca_topic_score_gemma":0.0067056417,"teacher_disagreement_score":0.99476594,"about_ca_system_score_codex":0.010616659,"about_ca_system_score_gemma":0.028617594,"threshold_uncertainty_score":0.9542873},"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":"dataset","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00049977563,0.00044032276,0.0016555173,0.00028857082,0.00009492037,0.0028586762,0.9928048,0.00041680533,0.00094056805],"genre_scores_gemma":[0.052505426,0.0036027823,0.07610598,0.0038687391,0.00086358393,0.2318238,0.5980187,0.0032657234,0.029945362],"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","domain_scores_codex":[0.99284667,0.0023289789,0.002504446,0.00097150105,0.00091507926,0.000433271],"domain_scores_gemma":[0.791341,0.18509185,0.011623109,0.0043238024,0.0068910993,0.00072914356],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.012779776,0.0027182915,0.005098311,0.006863557,0.0010880699,0.0028902723,0.0028796552,0.0025328323,0.85100526],"category_scores_gemma":[0.15919258,0.001636222,0.0056783124,0.0094707655,0.0005811398,0.0037125722,0.0018894279,0.001681386,0.04291028],"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":[0.003988631,0.0002897191,0.005500742,0.3938678,0.0034054543,0.0002565336,0.0003729887,0.0026695929,0.0004052644,0.004888362,0.55262893,0.03172597],"study_design_scores_gemma":[0.09070505,0.0022686922,0.044496253,0.16734523,0.019922722,0.001784484,0.0011223956,0.016114054,0.0025071441,0.055981494,0.59681463,0.00093779975],"about_ca_topic_score_codex":0.0048739766,"about_ca_topic_score_gemma":0.0103675155,"teacher_disagreement_score":0.9949017,"about_ca_system_score_codex":0.0025702298,"about_ca_system_score_gemma":0.0049615414,"threshold_uncertainty_score":0.21252292},"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":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29857033,0.6606126,0.031316776,0.0030056674,0.0018752426,0.0015996598,0.0015434906,0.0002932296,0.0011830503],"genre_scores_gemma":[0.95878243,0.021657996,0.014105449,0.0021048079,0.0005222751,0.0012394029,0.001131292,0.00012486264,0.00033147287],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.87399024,0.09852011,0.014422267,0.008112062,0.0033407507,0.0016145104],"domain_scores_gemma":[0.8403716,0.13283725,0.012382389,0.0102861505,0.0029454047,0.0011772168],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.11819529,0.0037497217,0.020910768,0.0036153449,0.0012544906,0.007073642,0.0032008612,0.004834094,0.0029304035],"category_scores_gemma":[0.13013807,0.0025971346,0.041775513,0.0039767623,0.0020281824,0.0046861973,0.0035174098,0.0045203334,0.00038421262],"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":[0.022823567,0.0001269102,0.03397788,0.013343654,0.91887236,0.000064233376,0.000120202014,0.0008641222,0.00032697644,0.00027228892,0.00043831818,0.008769483],"study_design_scores_gemma":[0.013621344,0.0011934313,0.01911963,0.0016548639,0.96032333,0.000079464,0.00011228279,0.0019453733,0.00022115327,0.0009951959,0.00066436175,0.00006951587],"about_ca_topic_score_codex":0.001695256,"about_ca_topic_score_gemma":0.0024503784,"teacher_disagreement_score":0.9962503,"about_ca_system_score_codex":0.0014721362,"about_ca_system_score_gemma":0.002091353,"threshold_uncertainty_score":0.6250839},"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"}]}