{"meta":{"query_hash":"622434103f93","filters":{"venue":"Journal of Educational Evaluation for Health Professions"},"cohort_total":27,"direct_labels_cover":0,"predictions_cover":27,"exported":27,"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/622434103f93","api":"https://metacan.xera.ac/api/v1/cohort?venue=Journal+of+Educational+Evaluation+for+Health+Professions"},"results":[{"id":"W1991005418","doi":"10.3352/jeehp.2014.11.31","title":"Designing and conducting a two day orientation program for first semester undergraduate medical students","year":2014,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Orientation (vector space); Medical education; Computer science; Mathematics education; Psychology; Medicine; Mathematics","score_opus":0.16969075134538417,"score_gpt":0.5871495641180114,"score_spread":0.4174588127726273,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W1991005418","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.70872307,0.0040141954,0.072874606,0.098798744,0.020456627,0.054724686,0.0008019357,0.0034580445,0.036148135],"genre_scores_gemma":[0.50894874,0.0054587433,0.33560804,0.055005603,0.0071878345,0.04607591,0.00085764076,0.0006864827,0.040170938],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99279755,0.003601003,0.00044495595,0.00042897617,0.0016921608,0.0010353684],"domain_scores_gemma":[0.96995544,0.0037188125,0.0031375508,0.00094593875,0.0061785844,0.01606373],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.014289474,0.0005786469,0.0005236052,0.0009511952,0.0028980041,0.002360341,0.0019961512,0.0018200948,0.009147234],"category_scores_gemma":[0.026997736,0.0006120428,0.0006533699,0.00037095937,0.0011145858,0.0014633044,0.0036103383,0.0036814006,0.004411688],"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":[0.00074534,0.019060535,0.04269614,0.0017417523,0.00004839756,0.0029938396,0.026789252,0.00059593155,0.00961058,0.0011369882,0.22930636,0.665275],"study_design_scores_gemma":[0.0009851768,0.03530526,0.15471226,0.0023960937,0.00014181682,0.004457996,0.060661014,0.0017975218,0.012140045,0.003197178,0.7238138,0.00039187665],"about_ca_topic_score_codex":0.0007520771,"about_ca_topic_score_gemma":0.0025257969,"teacher_disagreement_score":0.014289474,"about_ca_system_score_codex":0.0014151603,"about_ca_system_score_gemma":0.008919882,"threshold_uncertainty_score":0.07557082},"labels":[],"label_agreement":null},{"id":"W2006382901","doi":"10.3352/jeehp.2013.10.7","title":"Continuing medical education as a national strategy to improve access to primary care in Saudi Arabia","year":2013,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Primary care; Continuing medical education; Continuing education; Test (biology); Medical education; Family medicine; Medicine; Primary health care; Program evaluation; Nursing; Political science; Environmental health","score_opus":0.07093522327175421,"score_gpt":0.5406844693645076,"score_spread":0.4697492460927534,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2006382901","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9770074,0.0042088036,0.00033538518,0.0125221675,0.0001378449,0.00015967018,0.00008036979,0.000023348917,0.005524926],"genre_scores_gemma":[0.99262154,0.002028657,0.0019743172,0.001814798,0.00006341227,0.00006247526,0.0000954682,0.0000029947366,0.0013363359],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9990847,0.00042739505,0.00006415788,0.00004042684,0.00019909932,0.00018407495],"domain_scores_gemma":[0.9966509,0.00039174146,0.0005110751,0.00009980298,0.0008822193,0.0014642692],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0020817332,0.0001255642,0.00008646715,0.00029963691,0.00078822725,0.0009710274,0.00039226364,0.0005298202,0.0019294029],"category_scores_gemma":[0.0040653856,0.000075325246,0.000120309116,0.0002855996,0.00024220328,0.0005236269,0.0010013455,0.00047680255,0.00014453992],"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":[0.00055107125,0.0026106094,0.36888465,0.0009427997,0.000054506887,0.00080789346,0.0064331903,0.00068421126,0.0056663877,0.0019356252,0.0071085165,0.6043205],"study_design_scores_gemma":[0.00024557003,0.0030233543,0.9304883,0.00081504195,0.00006548138,0.0009003025,0.0097692385,0.0011650401,0.00235984,0.00022831236,0.050908424,0.000031140942],"about_ca_topic_score_codex":0.0460269,"about_ca_topic_score_gemma":0.066062145,"teacher_disagreement_score":0.0460269,"about_ca_system_score_codex":0.0031841008,"about_ca_system_score_gemma":0.0111594675,"threshold_uncertainty_score":0.091517985},"labels":[],"label_agreement":null},{"id":"W2009174856","doi":"10.3352/jeehp.2008.5.2","title":"Proposal of the Implementation of an International Pharmacy Graduate Preliminary Examination","year":2008,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Pharmacy and Medical Practices","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Pharmacy; Accreditation; Test (biology); Medical education; Pharmacist; Medicine; Pharmacy practice; Multistate Pharmacy Jurisprudence Examination; Family medicine; Pharmaconomist","score_opus":0.5101176070322258,"score_gpt":0.6733656710429481,"score_spread":0.16324806401072234,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2009174856","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.22939123,0.0025382729,0.2987251,0.21843286,0.027266478,0.12271804,0.0015002222,0.0036462387,0.09578156],"genre_scores_gemma":[0.2425364,0.0018885498,0.61891586,0.033719223,0.0030553148,0.03767176,0.0020551858,0.00027373692,0.059884034],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.97051066,0.011189251,0.005914143,0.002187095,0.006596196,0.003602694],"domain_scores_gemma":[0.9070043,0.016284805,0.0053940723,0.0047791842,0.049758635,0.016778938],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.056998298,0.0011816088,0.00085900247,0.0038922133,0.0040413043,0.0036610446,0.0034562522,0.008702186,0.0054802154],"category_scores_gemma":[0.09224454,0.0011509473,0.002339311,0.0014320231,0.002418614,0.0032073534,0.0053157792,0.008133713,0.003278209],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001526566,0.0069949864,0.0696993,0.0033227594,0.00019195721,0.008500846,0.013181519,0.008609258,0.029712508,0.09156532,0.18075627,0.5859387],"study_design_scores_gemma":[0.0014707242,0.011345303,0.11475981,0.0029155298,0.0006223526,0.0041243825,0.017315457,0.013340906,0.024702603,0.01740783,0.79089546,0.0010996135],"about_ca_topic_score_codex":0.004521144,"about_ca_topic_score_gemma":0.005077156,"teacher_disagreement_score":0.056998298,"about_ca_system_score_codex":0.0043580453,"about_ca_system_score_gemma":0.040864307,"threshold_uncertainty_score":0.3014394},"labels":[],"label_agreement":null},{"id":"W2049217914","doi":"10.3352/jeehp.2014.11.20","title":"Reflective writing and its impact on empathy in medical education: systematic review","year":2014,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Empathy and Medical Education","field":"Medicine","cited_by":159,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Empathy; Reflective writing; CINAHL; Curriculum; Medical education; Psychology; Psychological intervention; MEDLINE; Medicine; Pedagogy; Social psychology; Psychiatry","score_opus":0.09904297616015606,"score_gpt":0.5823871396492837,"score_spread":0.4833441634891276,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2049217914","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":[0.0024026716,0.9964508,0.00014862115,0.00021574936,0.000084831474,0.0003144078,0.0001424021,0.0000071900813,0.00023323885],"genre_scores_gemma":[0.03995674,0.95771396,0.00082021137,0.00039912772,0.000097010794,0.0007564736,0.00013397017,0.000006321586,0.00011620697],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98649955,0.0051276525,0.004703133,0.00072757906,0.0026547625,0.0002872472],"domain_scores_gemma":[0.8748188,0.10089814,0.016551653,0.0012519635,0.005732973,0.0007465115],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.015453857,0.001335876,0.0071990243,0.008458454,0.0007403366,0.0032045755,0.0017362582,0.0019445571,0.0039938237],"category_scores_gemma":[0.092544734,0.00089531916,0.006177007,0.011507254,0.0014438089,0.0029471084,0.0016254645,0.0012276004,0.00023781025],"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":[0.00023691787,0.000029248713,0.0010316521,0.9572148,0.0085804,0.00006223441,0.00029462617,0.00006417958,0.00008610368,0.00012912485,0.0006678564,0.031603046],"study_design_scores_gemma":[0.00034048365,0.00034347808,0.006089868,0.9109792,0.07367313,0.00036925467,0.0005004199,0.00008082704,0.0002308872,0.00022507075,0.0071310163,0.000036364054],"about_ca_topic_score_codex":0.005151579,"about_ca_topic_score_gemma":0.016472686,"teacher_disagreement_score":0.015453857,"about_ca_system_score_codex":0.0037168814,"about_ca_system_score_gemma":0.013141447,"threshold_uncertainty_score":0.081728816},"labels":[],"label_agreement":null},{"id":"W2064169615","doi":"10.3352/jeehp.2014.11.28","title":"What steps are necessary to create written or web-based selected-response assessments?","year":2014,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","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":"Montreal General Hospital; McGill University","funders":"","keywords":"Computer science; Process (computing); Work (physics); Health care; Health professionals; Online assessment; Medical education; Formative assessment; Medicine; Psychology; Mathematics education; Engineering","score_opus":0.1018934734433987,"score_gpt":0.5380365283357339,"score_spread":0.4361430548923352,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2064169615","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.10145971,0.001998752,0.66558206,0.041653957,0.0027143985,0.14044423,0.0018162536,0.006411699,0.037918977],"genre_scores_gemma":[0.060941122,0.00046790682,0.8973238,0.0013689058,0.00015113215,0.03772642,0.0003594923,0.00023331006,0.0014278499],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.550471,0.3783982,0.034877174,0.005596416,0.027361305,0.0032959643],"domain_scores_gemma":[0.322012,0.4323053,0.03186468,0.04440266,0.16066588,0.008749507],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.35027194,0.0017517663,0.0021185633,0.0065568504,0.004037589,0.010177702,0.004747445,0.003351161,0.010867797],"category_scores_gemma":[0.5780447,0.001827321,0.0023747717,0.004434365,0.0031966039,0.0131376805,0.0048909932,0.005356166,0.008140313],"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":[0.001409401,0.0025292253,0.01463211,0.009897862,0.00019391124,0.00059081445,0.046122663,0.0021508904,0.0073791468,0.011384537,0.031340662,0.87236863],"study_design_scores_gemma":[0.003767685,0.009478336,0.12898779,0.04452303,0.00093616155,0.003280311,0.19074412,0.042789318,0.062858835,0.12337484,0.38654572,0.0027138898],"about_ca_topic_score_codex":0.0026447931,"about_ca_topic_score_gemma":0.0067940587,"teacher_disagreement_score":0.35027194,"about_ca_system_score_codex":0.0058967173,"about_ca_system_score_gemma":0.02761916,"threshold_uncertainty_score":0.8012304},"labels":[],"label_agreement":null},{"id":"W2078254629","doi":"10.3352/jeehp.2011.8.13","title":"Implementation of a multi-level evaluation strategy: a case study on a program for international medical graduates","year":2011,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Global Health Workforce Issues","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Personalization; Baseline (sea); Resource (disambiguation); Psychological intervention; Medical education; Data collection; Process (computing); Knowledge management; Intervention (counseling); Underpinning; Program evaluation; Process management; Computer science; Psychology; Medicine; Business; Nursing; Engineering; Political science","score_opus":0.5803600417563879,"score_gpt":0.693242103122204,"score_spread":0.11288206136581613,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2078254629","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9623022,0.0003334681,0.014968098,0.007940004,0.00016977938,0.005674459,0.00006989668,0.00011749759,0.008424503],"genre_scores_gemma":[0.9491105,0.0004569985,0.0397405,0.00242305,0.00008055163,0.0041594254,0.00006200033,0.00009298816,0.0038739804],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9434969,0.04629293,0.0012737534,0.0014254169,0.0028973424,0.0046136756],"domain_scores_gemma":[0.9649685,0.015804304,0.0024860862,0.0016520568,0.004643168,0.010445902],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.053729367,0.0010382052,0.0007528705,0.0015988209,0.010319028,0.0044662743,0.0039434396,0.0052182316,0.0026778155],"category_scores_gemma":[0.051993247,0.0007628418,0.0011961334,0.0013253045,0.0041767326,0.0035014132,0.006261601,0.005325788,0.00056637265],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011362848,0.047808,0.053356342,0.0035985094,0.00020270032,0.037549146,0.4927693,0.0044903555,0.011044504,0.011029552,0.011027028,0.32598826],"study_design_scores_gemma":[0.0010770719,0.039549433,0.050396547,0.0039193495,0.0002444014,0.012739314,0.7725408,0.012526479,0.012010245,0.005301726,0.089128636,0.00056604866],"about_ca_topic_score_codex":0.0058484827,"about_ca_topic_score_gemma":0.017318636,"teacher_disagreement_score":0.053729367,"about_ca_system_score_codex":0.010810706,"about_ca_system_score_gemma":0.016062446,"threshold_uncertainty_score":0.2841515},"labels":[],"label_agreement":null},{"id":"W2078326370","doi":"10.3352/jeehp.2014.11.9","title":"Students’ perception of the learning environment at Xavier University School of Medicine, Aruba: a follow-up study","year":2014,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Perception; Nationality; Psychology; Medical education; Learning environment; Academic year; Medicine; Mathematics education; Geography; Immigration","score_opus":0.0746360173861421,"score_gpt":0.4789982235541127,"score_spread":0.4043622061679706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2078326370","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9996082,0.00003008202,0.000018402541,0.000023600525,0.000004014395,0.000017875009,0.000029058227,0.0000025641677,0.00026620546],"genre_scores_gemma":[0.9989749,0.000062805666,0.0000643212,0.000028694567,0.0000048143006,0.000033953478,0.0000841493,0.0000015387648,0.0007448716],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9993118,0.0001325405,0.00004180687,0.00007074289,0.00019528437,0.00024789164],"domain_scores_gemma":[0.9968239,0.00032675988,0.0006302956,0.00013536608,0.0009792323,0.0011044744],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0019039976,0.00035524214,0.0005245041,0.0012434378,0.0020095224,0.0014220482,0.0005378577,0.0007813803,0.0019181782],"category_scores_gemma":[0.0028703639,0.0003973608,0.0007841304,0.0008218599,0.00039908144,0.00071709976,0.00090109167,0.001337764,0.00077863556],"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":[0.00032006257,0.0031143194,0.9682245,0.000051890496,0.000045512552,0.0004323103,0.011709131,0.00007894372,0.001919284,0.000026851667,0.00040863972,0.013668517],"study_design_scores_gemma":[0.0000060647762,0.0011846883,0.98879564,0.000011661968,0.000014914804,0.00006840027,0.009050346,0.000053876076,0.00031275948,0.00000494986,0.00048320278,0.000013534701],"about_ca_topic_score_codex":0.012094402,"about_ca_topic_score_gemma":0.016162757,"teacher_disagreement_score":0.012094402,"about_ca_system_score_codex":0.0011376712,"about_ca_system_score_gemma":0.0009769375,"threshold_uncertainty_score":0.02404803},"labels":[],"label_agreement":null},{"id":"W2099558626","doi":"10.3352/jeehp.2015.12.11","title":"Best-fit model of exploratory and confirmatory factor analysis of the 2010 Medical Council of Canada Qualifying Examination Part I clinical decision-making cases","year":2015,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Medical Education and Admissions","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada","funders":"","keywords":"Confirmatory factor analysis; Exploratory factor analysis; Exploratory analysis; Medical education; Clinical decision making; Factor (programming language); Psychology; Medicine; Computer science; Structural equation modeling; Family medicine; Data science; Psychometrics; Clinical psychology; Machine learning","score_opus":0.626887123476903,"score_gpt":0.5881185642960887,"score_spread":0.038768559180814344,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2099558626","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9343363,0.00036639062,0.051929917,0.0010314377,0.00012247206,0.0020659014,0.0013633949,0.00025923701,0.0085249115],"genre_scores_gemma":[0.9769031,0.00015120817,0.019170102,0.00008170615,0.000016512693,0.00090427883,0.0017857392,0.000045267934,0.0009420128],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9854423,0.007756315,0.0006369744,0.001724336,0.0034964806,0.0009436384],"domain_scores_gemma":[0.9617161,0.023712298,0.0027276203,0.0024076882,0.008454179,0.0009821428],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.027712991,0.002326715,0.0011075157,0.003417289,0.0021930283,0.0028814867,0.0014791131,0.00095044286,0.0037609036],"category_scores_gemma":[0.08902614,0.0007827705,0.0036973173,0.0025679804,0.0018283579,0.0013600353,0.0017404724,0.001776745,0.0009914067],"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":[0.00092249754,0.00093599537,0.7968297,0.00045795922,0.0020810917,0.00067951257,0.01319261,0.048929855,0.0010007534,0.015655693,0.010123352,0.10919097],"study_design_scores_gemma":[0.0003833597,0.000985025,0.49238762,0.00067855284,0.0007111746,0.0004848724,0.009318108,0.4672249,0.0010575255,0.016800525,0.009681694,0.00028663318],"about_ca_topic_score_codex":0.16494276,"about_ca_topic_score_gemma":0.13056856,"teacher_disagreement_score":0.16494276,"about_ca_system_score_codex":0.0068360977,"about_ca_system_score_gemma":0.017311342,"threshold_uncertainty_score":0.32796532},"labels":[],"label_agreement":null},{"id":"W2110847541","doi":"10.3352/jeehp.2015.12.9","title":"Challenges in shifting to an integrated curriculum in a Caribbean medical school","year":2015,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Global Health Workforce Issues","field":"Health Professions","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Curriculum; Caribbean region; Medical education; Computer science; Mathematics education; Medicine; Psychology; Pedagogy; Political science; Latin Americans","score_opus":0.33615750073251627,"score_gpt":0.6094517195581488,"score_spread":0.27329421882563254,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2110847541","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7152488,0.0058905836,0.0031522382,0.23965956,0.0031540121,0.0016932568,0.00021336424,0.00024536267,0.03074272],"genre_scores_gemma":[0.91170347,0.0048109344,0.019103143,0.050008085,0.0005842382,0.0013521749,0.00036754154,0.0001273541,0.011943086],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9765428,0.009103111,0.0016530084,0.0008513281,0.006089081,0.005760629],"domain_scores_gemma":[0.9493525,0.004794041,0.0026643253,0.0010334218,0.008844993,0.033310737],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.022591017,0.00049733603,0.00049962604,0.001202378,0.009492638,0.0076981205,0.0032707027,0.0038517523,0.011336594],"category_scores_gemma":[0.036595054,0.00042566418,0.0005308998,0.0014724671,0.002258146,0.004565997,0.013995754,0.005468491,0.001365738],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065204233,0.007836827,0.09839518,0.003525432,0.00011048134,0.0046897847,0.09140744,0.0011670408,0.00512432,0.012337715,0.100568615,0.6741853],"study_design_scores_gemma":[0.00048136342,0.0062613403,0.25953957,0.010253109,0.00012921097,0.0021508643,0.34988275,0.0018086842,0.003247114,0.006564475,0.35925522,0.0004263871],"about_ca_topic_score_codex":0.03997766,"about_ca_topic_score_gemma":0.09069881,"teacher_disagreement_score":0.03997766,"about_ca_system_score_codex":0.0112625575,"about_ca_system_score_gemma":0.04610433,"threshold_uncertainty_score":0.11947411},"labels":[],"label_agreement":null},{"id":"W2159647634","doi":"10.3352/jeehp.2015.12.10","title":"Initiating small group learning in a Caribbean medical school","year":2015,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Small group learning; Learning environment; Curriculum; Context (archaeology); Socialization; Experiential learning; Active learning (machine learning); Medical education; Problem-based learning; Cooperative learning; Psychology; Mathematics education; Pedagogy; Medicine; Computer science; Teaching method; Artificial intelligence; Geography","score_opus":0.2001838928762288,"score_gpt":0.543058781449507,"score_spread":0.34287488857327825,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2159647634","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9588031,0.0007577659,0.002133813,0.007110556,0.0002743779,0.000908027,0.00006990444,0.0001662714,0.029776176],"genre_scores_gemma":[0.9624704,0.000913391,0.008004489,0.0031660302,0.00015541862,0.00045349792,0.0001279374,0.00005403919,0.024654692],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.998602,0.0005206421,0.0000328432,0.00015477359,0.00016516434,0.00052459695],"domain_scores_gemma":[0.99314874,0.0002451981,0.00026317337,0.00011299542,0.0002472214,0.005982679],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0017283807,0.00046322355,0.00029121287,0.00093142246,0.009964736,0.0026241865,0.0016703343,0.0015353565,0.014365219],"category_scores_gemma":[0.0037612843,0.0003886018,0.00039826563,0.00061509566,0.001604368,0.0009243093,0.0067631234,0.0022028103,0.0018871498],"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":[0.0013493834,0.01632204,0.2013147,0.0009953781,0.000062193576,0.033352897,0.24364199,0.00070850283,0.014646599,0.008364998,0.04716503,0.43207628],"study_design_scores_gemma":[0.000575974,0.0110924,0.20064604,0.0015097688,0.00013979721,0.009617193,0.3778152,0.0020876098,0.0054846313,0.0066044694,0.3840439,0.00038310187],"about_ca_topic_score_codex":0.01603813,"about_ca_topic_score_gemma":0.060696628,"teacher_disagreement_score":0.01603813,"about_ca_system_score_codex":0.0031618162,"about_ca_system_score_gemma":0.011222957,"threshold_uncertainty_score":0.048056483},"labels":[],"label_agreement":null},{"id":"W2236484017","doi":"10.3352/jeehp.2015.12.52","title":"Pre-clinical versus clinical medical students’ attitudes towards the poor in the United States","year":2015,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Empathy and Medical Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"","funders":"University of Windsor","keywords":"Medical education; Psychology; Mathematics education; Medicine","score_opus":0.5124042120784384,"score_gpt":0.6739462629049746,"score_spread":0.16154205082653617,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2236484017","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9994473,0.00005696758,0.000008699306,0.00011996946,0.0000040154946,0.0000031330555,0.000008990152,4.20059e-7,0.00035049397],"genre_scores_gemma":[0.9996532,0.00007142502,0.0000131408615,0.00011218636,0.0000029118858,0.0000032478954,0.000010870588,2.6347053e-7,0.00013289087],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99930894,0.00026532472,0.00005361457,0.00004258767,0.00015653083,0.00017304685],"domain_scores_gemma":[0.99686927,0.00042900039,0.0009947062,0.000048394293,0.00029733608,0.001361355],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0009930123,0.00013034765,0.00016360692,0.00067678356,0.0007741078,0.00073826604,0.000129354,0.00045869054,0.0022778236],"category_scores_gemma":[0.0037719894,0.00015873925,0.00015793755,0.00038276578,0.0006304513,0.00045969556,0.0009780229,0.0007427899,0.00028648047],"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":[0.00014275948,0.00035968918,0.985887,0.000022210956,0.000015201227,0.000105858606,0.005322961,0.000025675508,0.0004230421,0.00005686319,0.00030931152,0.007329417],"study_design_scores_gemma":[0.000005523512,0.00034733053,0.9856954,0.000028078062,0.0000052337173,0.0001546603,0.013108424,0.00004073674,0.0001098081,0.000029700439,0.00046985733,0.000005265046],"about_ca_topic_score_codex":0.0052610044,"about_ca_topic_score_gemma":0.014087032,"teacher_disagreement_score":0.0052610044,"about_ca_system_score_codex":0.00049757666,"about_ca_system_score_gemma":0.0007031856,"threshold_uncertainty_score":0.010460734},"labels":[],"label_agreement":null},{"id":"W2262777419","doi":"10.3352/jeehp.2016.13.6","title":"Calibrating the Medical Council of Canada’s Qualifying Examination Part I using an integrated item response theory framework: a comparison of models and designs","year":2016,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada","funders":"","keywords":"Polytomous Rasch model; Item response theory; Calibration; Statistics; Set (abstract data type); Test (biology); Computer science; Type I and type II errors; Econometrics; Multiple choice; Psychology; Mathematics; Significant difference; Psychometrics; Programming language","score_opus":0.8970046108208954,"score_gpt":0.6521780930522075,"score_spread":0.2448265177686879,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2262777419","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7065201,0.0013514984,0.27838513,0.0010400796,0.00011429408,0.0054593417,0.0006991333,0.0009946662,0.0054358155],"genre_scores_gemma":[0.8078219,0.0004204489,0.18662328,0.00032957568,0.000030852585,0.002552681,0.0011272042,0.0001170558,0.0009769341],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9073586,0.06981567,0.0023311873,0.0058614644,0.013270677,0.0013624429],"domain_scores_gemma":[0.84107274,0.11875228,0.010744251,0.009187324,0.019035725,0.0012076377],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.11803806,0.0017658636,0.0013889922,0.004175979,0.0011990515,0.003445667,0.0031310783,0.0017025393,0.0018477505],"category_scores_gemma":[0.21758501,0.0010293331,0.0031993736,0.0033846274,0.0022309187,0.0017144302,0.0033647188,0.0022072752,0.0005405065],"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":[0.002079107,0.0013053063,0.5156476,0.0013603413,0.0026935488,0.00008813637,0.007503178,0.11180683,0.0013298254,0.010670555,0.0034899742,0.34202558],"study_design_scores_gemma":[0.0010099522,0.0031613363,0.40688717,0.0011755246,0.0011167179,0.0001702272,0.003890522,0.5623159,0.003354809,0.0109925335,0.005392258,0.00053315324],"about_ca_topic_score_codex":0.14374037,"about_ca_topic_score_gemma":0.10597408,"teacher_disagreement_score":0.14374037,"about_ca_system_score_codex":0.014089808,"about_ca_system_score_gemma":0.018743675,"threshold_uncertainty_score":0.6242524},"labels":[],"label_agreement":null},{"id":"W2412489802","doi":"10.3352/jeehp.2016.13.20","title":"Presidential address: launching the Korea Health Personnel Licensing Examination Institute, a government-supported special foundation from December 23, 2015","year":2016,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Pharmacy and Medical Practices","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Foundation (evidence); Government (linguistics); Medical education; Medicine; Political science","score_opus":0.3077478843118082,"score_gpt":0.5837614269185688,"score_spread":0.2760135426067606,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2412489802","genre_codex":"other","genre_gemma":"other","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"other","genre_consensus":"other","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.007298229,0.006422236,0.0014230829,0.15970041,0.120309,0.0011650936,0.005919097,0.0013197185,0.69644314],"genre_scores_gemma":[0.0144735845,0.0025165475,0.000713959,0.024334652,0.0127466945,0.00034526893,0.002075241,0.0002601114,0.94253397],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9983277,0.00017490608,0.00009808946,0.00030453963,0.00054330844,0.0005515097],"domain_scores_gemma":[0.9962425,0.00025975585,0.00031353222,0.00012442855,0.001282306,0.0017775174],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0022575874,0.0012460509,0.0005222913,0.00085731805,0.0027843136,0.0041966625,0.0007860591,0.005575548,0.1856913],"category_scores_gemma":[0.0034162933,0.00043111792,0.00062129984,0.00045679288,0.0006316749,0.0019924035,0.0022162443,0.0058088074,0.13092704],"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":[0.000029741808,0.000058930804,0.0008759729,0.00005346872,0.0000047146955,0.00014543845,0.00005251822,0.000015411391,0.0003184029,0.0013669364,0.98373127,0.013347105],"study_design_scores_gemma":[0.00001740821,0.000034874425,0.0031234086,0.00007061354,0.000006213414,0.000098473385,0.00013800812,0.000053753978,0.00020937351,0.0001391902,0.9960957,0.000013055716],"about_ca_topic_score_codex":0.0038705121,"about_ca_topic_score_gemma":0.007969474,"teacher_disagreement_score":0.1856913,"about_ca_system_score_codex":0.0014634974,"about_ca_system_score_gemma":0.0058676605,"threshold_uncertainty_score":0.62119925},"labels":[],"label_agreement":null},{"id":"W2560916963","doi":"10.3352/jeehp.2016.13.44","title":"A cost-effectiveness analysis of self-debriefing versus instructor debriefing for simulated crises in perioperative medicine in Canada","year":2016,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":31,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Ottawa Hospital; Innovation, Science and Economic Development Canada; The Wilson Centre; Health Sciences Centre; Sunnybrook Health Science Centre; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Debriefing; Psychology; Medical education; Randomized controlled trial; Medicine; Applied psychology; Surgery","score_opus":0.2423614076170188,"score_gpt":0.5514675738551226,"score_spread":0.3091061662381038,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2560916963","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98003745,0.006075651,0.0027859272,0.0010347308,0.00012763482,0.006290176,0.00088060217,0.00006446443,0.0027033077],"genre_scores_gemma":[0.9952828,0.0010763897,0.0022792309,0.00012111311,0.000020234167,0.0007891405,0.00018887289,0.0000054624356,0.00023669576],"study_design_codex":"randomized_trial","study_design_gemma":"observational","domain_scores_codex":[0.98770696,0.008301638,0.00063384825,0.0006075839,0.0019433775,0.0008066945],"domain_scores_gemma":[0.9805664,0.012026615,0.002914688,0.00040666835,0.0020055552,0.002080069],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0084375525,0.0013590889,0.0015735038,0.0017289442,0.0007823462,0.0012004777,0.0016284031,0.0014181533,0.004566684],"category_scores_gemma":[0.03767168,0.0006391756,0.0037750478,0.0011010492,0.0010853796,0.0008677147,0.0011153803,0.0019647006,0.00011396666],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.5076896,0.02691095,0.047278233,0.009978683,0.011192315,0.00023289816,0.0005340796,0.090301335,0.0024509544,0.0024665582,0.0026081381,0.29835632],"study_design_scores_gemma":[0.17749856,0.29760993,0.2118041,0.0034363368,0.023276566,0.0004194141,0.001360777,0.2727052,0.005453747,0.0018434933,0.0041120797,0.0004798002],"about_ca_topic_score_codex":0.35693678,"about_ca_topic_score_gemma":0.29116184,"teacher_disagreement_score":0.6430632,"about_ca_system_score_codex":0.03910179,"about_ca_system_score_gemma":0.028463827,"threshold_uncertainty_score":0.70971835},"labels":[],"label_agreement":null},{"id":"W2765995486","doi":"10.3352/jeehp.2017.14.25","title":"Evaluation of a course to prepare international students for the United States Medical Licensing Examination step 2 clinical skills exam","year":2017,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Global Health Workforce Issues","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Johns Hopkins University","keywords":"Medical education; Competence (human resources); United States Medical Licensing Examination; Test (biology); Medicine; Medical school; Educational measurement; Psychology; Family medicine; Curriculum; Pedagogy; Social psychology","score_opus":0.2861969031259385,"score_gpt":0.6857632889006259,"score_spread":0.3995663857746874,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2765995486","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9942352,0.00010898229,0.0005270635,0.00044877792,0.000105876956,0.0011304144,0.00018610981,0.00008162458,0.0031759893],"genre_scores_gemma":[0.99020606,0.00017877245,0.003816923,0.00030486236,0.000079341145,0.0007267981,0.00073166133,0.000013876781,0.0039416263],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9987913,0.000263041,0.00008631559,0.00007870126,0.00043760703,0.00034314883],"domain_scores_gemma":[0.9947865,0.0004798541,0.0005936587,0.0001053705,0.0014890272,0.002545611],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0029260085,0.0005863838,0.0003072116,0.00067249796,0.000604228,0.0006587514,0.0006522397,0.0005953245,0.003163184],"category_scores_gemma":[0.0076112733,0.00021965621,0.00045865274,0.00023506812,0.00018732548,0.00042198822,0.0008995132,0.0009489016,0.00076911156],"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":[0.0069027105,0.07273267,0.3690794,0.0004918547,0.00015708196,0.0015249776,0.0027600392,0.0028440424,0.014707478,0.00022137008,0.032759737,0.4958186],"study_design_scores_gemma":[0.0008658069,0.035753287,0.9314346,0.00019481103,0.00007054652,0.0007281687,0.0018789412,0.003989737,0.009752263,0.00011304437,0.015142164,0.0000766291],"about_ca_topic_score_codex":0.002449063,"about_ca_topic_score_gemma":0.009782283,"teacher_disagreement_score":0.003163184,"about_ca_system_score_codex":0.0016529592,"about_ca_system_score_gemma":0.0027291402,"threshold_uncertainty_score":0.015474379},"labels":[],"label_agreement":null},{"id":"W2782030070","doi":"10.3352/jeehp.2018.15.2","title":"Contributions of psychological needs, self-compassion, leisure-time exercise, and achievement goals to academic engagement and exhaustion in Canadian medical students","year":2018,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Healthcare professionals’ stress and burnout","field":"Health Professions","cited_by":58,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Alberta","funders":"Social Sciences and Humanities Research Council of Canada","keywords":"Psychology; Burnout; Autonomy; Competence (human resources); Emotional exhaustion; Self-determination theory; Clinical psychology; Coping (psychology); Compassion; Need for achievement; Social psychology","score_opus":0.11758942967190389,"score_gpt":0.5730412919235713,"score_spread":0.45545186225166745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2782030070","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9989531,0.000110706,0.000012208272,0.00015046801,0.0000036855997,0.000009478677,0.00008167629,0.0000010062473,0.0006777414],"genre_scores_gemma":[0.99937856,0.00010926686,0.00003878976,0.00003176122,0.0000023245063,0.000006610929,0.00009965203,7.4566344e-7,0.00033213766],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9990938,0.00007530975,0.000038510625,0.000066863664,0.0003726626,0.0003529749],"domain_scores_gemma":[0.9979686,0.00015310771,0.00040923708,0.000030514673,0.00045784216,0.0009806611],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0007041134,0.0002779284,0.00035332338,0.0009904692,0.0030725584,0.0011011395,0.00064081606,0.00046206216,0.001999595],"category_scores_gemma":[0.0032104834,0.00026474748,0.000379615,0.0011743467,0.00090233324,0.00025837467,0.0012393732,0.0008925982,0.0001401781],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000046422196,0.00012923087,0.98933977,0.00003388895,0.000016618602,0.000047872003,0.0045880335,0.00002670798,0.00015552108,0.00005955603,0.00035147226,0.005204909],"study_design_scores_gemma":[0.0000024462654,0.000028149056,0.9960977,0.000018487892,0.000005781043,0.000023331802,0.0033745526,0.00005309374,0.000028589147,0.000015141892,0.00034709228,0.000005549023],"about_ca_topic_score_codex":0.8982107,"about_ca_topic_score_gemma":0.94613427,"teacher_disagreement_score":0.101789296,"about_ca_system_score_codex":0.009299837,"about_ca_system_score_gemma":0.017563105,"threshold_uncertainty_score":0.20477748},"labels":[],"label_agreement":null},{"id":"W2886213067","doi":"10.3352/jeehp.2018.15.18","title":"The implementation and evaluation of an e-Learning training module for objective structured clinical examination raters in Canada","year":2018,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"University of Ottawa Heart Institute Foundation; University of Ottawa","keywords":"Computer science; E learning; Objective structured clinical examination; Training (meteorology); Medical education; Artificial intelligence; Medical physics; Medicine; World Wide Web; The Internet","score_opus":0.17995658875688936,"score_gpt":0.5953672734412092,"score_spread":0.41541068468431985,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2886213067","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9921434,0.0003034213,0.0020778233,0.00035894391,0.00006972874,0.0019847634,0.00021585378,0.00007388348,0.0027721955],"genre_scores_gemma":[0.97843057,0.0007360025,0.015782727,0.0003049998,0.000043443513,0.0012662602,0.00046777076,0.00004114718,0.0029271594],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.9711592,0.0126265865,0.0024423564,0.0016953467,0.010281273,0.001795317],"domain_scores_gemma":[0.883944,0.021237232,0.009136834,0.004189651,0.07292174,0.008570551],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0434464,0.0005299579,0.0004874915,0.0011854876,0.0018411699,0.0018451859,0.0018619037,0.00078186556,0.001117642],"category_scores_gemma":[0.09522446,0.00064909353,0.00048474927,0.0010523325,0.0010238078,0.0008201579,0.002072904,0.00086223055,0.0003673629],"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":[0.004594887,0.007891748,0.3720616,0.0012427748,0.00021503643,0.00047083377,0.027224254,0.0019785392,0.012791026,0.00035072447,0.0044427193,0.56673586],"study_design_scores_gemma":[0.00068396615,0.01109732,0.95387495,0.0005354257,0.0001399045,0.00025296738,0.0074412497,0.0046176678,0.008401446,0.00006209243,0.012710872,0.00018223944],"about_ca_topic_score_codex":0.4856766,"about_ca_topic_score_gemma":0.67730486,"teacher_disagreement_score":0.4856766,"about_ca_system_score_codex":0.016994819,"about_ca_system_score_gemma":0.042657908,"threshold_uncertainty_score":0.96569926},"labels":[],"label_agreement":null},{"id":"W2912914364","doi":"10.3352/jeehp.2019.16.4","title":"Cultural immersion in the education of healthcare professionals: a systematic review","year":2019,"lang":"en","type":"review","venue":"Journal of Educational Evaluation for Health Professions","topic":"Cultural Competency in Health Care","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Immersion (mathematics); Health professionals; Health care; Medical education; Psychology; Cultural competence; Pedagogy; Medicine; Political science; Mathematics","score_opus":0.3438295384040527,"score_gpt":0.6394174111540661,"score_spread":0.2955878727500134,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2912914364","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":[0.001330934,0.99738044,0.00016162838,0.00022970863,0.000104731094,0.0003599392,0.00007145072,0.0000051719303,0.00035604514],"genre_scores_gemma":[0.01446703,0.9838937,0.00056650164,0.00024490498,0.000046535046,0.00062322844,0.000060383343,0.0000029520309,0.00009471708],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.99091244,0.0040335865,0.0023821425,0.00044426802,0.0018949804,0.00033260623],"domain_scores_gemma":[0.96397513,0.027726667,0.004261521,0.00039850106,0.0030567746,0.0005814109],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008833846,0.0012219358,0.0047320593,0.008332007,0.0009991952,0.0030917795,0.0015682427,0.0019789406,0.0029977972],"category_scores_gemma":[0.04637769,0.0008501443,0.0036124003,0.010173522,0.0011682137,0.003164712,0.0020364076,0.0013292909,0.00023982517],"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":[0.00009344389,0.000030985295,0.00056939124,0.92969066,0.0017696872,0.0001127256,0.0010286728,0.000058205904,0.00013536878,0.00029378408,0.001157035,0.065060005],"study_design_scores_gemma":[0.00005679625,0.00013826988,0.0020135136,0.9717571,0.008171298,0.00032402895,0.0011979485,0.0000373575,0.00013679409,0.00018329911,0.015963905,0.000019656527],"about_ca_topic_score_codex":0.006938057,"about_ca_topic_score_gemma":0.020100076,"teacher_disagreement_score":0.008833846,"about_ca_system_score_codex":0.0044946806,"about_ca_system_score_gemma":0.020662777,"threshold_uncertainty_score":0.04671836},"labels":[],"label_agreement":null},{"id":"W2945549411","doi":"10.3352/jeehp.2019.16.14","title":"No observed effect of a student-led mock objective structured clinical examination on subsequent performance scores in medical students in Canada","year":2019,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada; McGill University; University of Ottawa","funders":"","keywords":"Objective structured clinical examination; Psychology; Medical education; Variance (accounting); Educational measurement; Analysis of variance; Cohort; Medicine; Curriculum; Internal medicine; Pedagogy","score_opus":0.0847540039834079,"score_gpt":0.5351608282847227,"score_spread":0.45040682430131473,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2945549411","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9992762,0.000036369118,0.000028486806,0.00005384781,0.000004288234,0.000009240095,0.000041811363,0.0000049441405,0.0005449276],"genre_scores_gemma":[0.9991167,0.000050632665,0.00007295854,0.000021609965,0.0000033456683,0.000006929101,0.00007019379,0.0000023179985,0.00065525225],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9984371,0.00043747204,0.00008375734,0.00018430491,0.0005306476,0.00032670243],"domain_scores_gemma":[0.99064904,0.001865682,0.0014300138,0.00035639407,0.0024537472,0.003245062],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0015487294,0.00020580598,0.00027682245,0.0003840954,0.00074522186,0.0009002909,0.00054693216,0.00042844794,0.0022114918],"category_scores_gemma":[0.015311452,0.0001740529,0.0003682918,0.0003854151,0.00053795974,0.000248592,0.0011288143,0.0006571396,0.000282477],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.003869275,0.0015122134,0.9126802,0.00011241691,0.00014057932,0.00019400184,0.002735546,0.0010506792,0.0037034627,0.00013803536,0.0014838466,0.07237977],"study_design_scores_gemma":[0.00003388713,0.0008064881,0.9968839,0.00002006494,0.000020316682,0.0000306312,0.0006536436,0.00046965873,0.0006301379,0.0000185541,0.0004197065,0.000012916385],"about_ca_topic_score_codex":0.45967332,"about_ca_topic_score_gemma":0.67593926,"teacher_disagreement_score":0.5403267,"about_ca_system_score_codex":0.004720609,"about_ca_system_score_gemma":0.008848384,"threshold_uncertainty_score":0.91399544},"labels":[],"label_agreement":null},{"id":"W2945832095","doi":"10.3352/jeehp.2019.16.13","title":"MEDTalks: a student-driven program to enhance undergraduate student understanding and interest in medical schools in Canada","year":2019,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa","funders":"","keywords":"Medical education; Curriculum; Medical school; Interest group; Undergraduate education; Undergraduate research; Small group learning; Psychology; Medicine; Pedagogy; Political science","score_opus":0.13357047338281494,"score_gpt":0.5643172881290961,"score_spread":0.4307468147462812,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2945832095","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9927347,0.000112237794,0.00043463596,0.00088710856,0.0000631796,0.00092973583,0.00021639912,0.00016443666,0.0044575743],"genre_scores_gemma":[0.9783092,0.00038854184,0.00783208,0.0009849243,0.000069661284,0.00088915846,0.00039026779,0.000027125809,0.011109093],"study_design_codex":"design_other","study_design_gemma":"observational","domain_scores_codex":[0.99881977,0.00023568365,0.000032152613,0.00009615251,0.00037768102,0.00043854112],"domain_scores_gemma":[0.9937687,0.00032298913,0.00029600676,0.000090503665,0.0005084294,0.005013321],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016055186,0.00044905915,0.00025374503,0.0006504144,0.0020165942,0.0008131883,0.0010921994,0.00041538765,0.004847346],"category_scores_gemma":[0.0034023183,0.00023657232,0.00039040972,0.0003227102,0.0007944287,0.0003705996,0.0027399892,0.0007692548,0.0003874924],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0062984773,0.038933598,0.13191645,0.0010323651,0.00011588551,0.00099794,0.030370599,0.0016005756,0.026228169,0.001431879,0.041900337,0.71917367],"study_design_scores_gemma":[0.002400192,0.024385145,0.8397231,0.00047108028,0.00012937622,0.00047024857,0.01978462,0.0026534176,0.012908788,0.0006567748,0.096166365,0.00025089874],"about_ca_topic_score_codex":0.33375502,"about_ca_topic_score_gemma":0.80171293,"teacher_disagreement_score":0.98905784,"about_ca_system_score_codex":0.010942162,"about_ca_system_score_gemma":0.049526315,"threshold_uncertainty_score":0.6636247},"labels":[],"label_agreement":null},{"id":"W3007534255","doi":"10.3352/jeehp.2020.17.5","title":"Potential of feedback during objective structured clinical examination to evoke an emotional response in medical students in Canada","year":2020,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Medical Council of Canada; University of Ottawa","funders":"","keywords":"Embarrassment; Objective structured clinical examination; Psychology; Emotional reaction; Clinical psychology; Medical education; Medicine; Social psychology; Psychiatry","score_opus":0.08914017321305964,"score_gpt":0.5361213036606929,"score_spread":0.4469811304476332,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3007534255","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99784195,0.00013140321,0.00006047266,0.00061881734,0.000016143065,0.000022494292,0.00003812392,0.0000052098258,0.0012653047],"genre_scores_gemma":[0.99883777,0.00017861309,0.00012655194,0.00022172884,0.000007841928,0.000007448993,0.00002251135,0.0000029545283,0.0005944782],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9962871,0.0010650314,0.00014784826,0.000161819,0.0015826361,0.0007555028],"domain_scores_gemma":[0.9898643,0.0022672794,0.0025603478,0.00013431333,0.0027592853,0.0024143432],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0021810408,0.00022611872,0.0002632599,0.0005695704,0.0019698818,0.0016686304,0.00047434238,0.00046674776,0.0015250355],"category_scores_gemma":[0.015666554,0.00018787353,0.00024103529,0.00055472174,0.0011806176,0.00023259033,0.0011950646,0.00091878034,0.00017219385],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009561435,0.0006640037,0.8156673,0.000334331,0.00005791938,0.0018307925,0.08516616,0.00022032254,0.008869571,0.00017641729,0.0030765713,0.082980394],"study_design_scores_gemma":[0.000026262072,0.0007251695,0.90758663,0.00018735386,0.00003250928,0.00049179787,0.0832739,0.00039837378,0.0019929586,0.00006461513,0.005144488,0.00007589028],"about_ca_topic_score_codex":0.5320092,"about_ca_topic_score_gemma":0.7429168,"teacher_disagreement_score":0.46799082,"about_ca_system_score_codex":0.010559005,"about_ca_system_score_gemma":0.015143858,"threshold_uncertainty_score":0.9414938},"labels":[],"label_agreement":null},{"id":"W3016806524","doi":"10.3352/jeehp.2020.17.12","title":"Performance of the Ebel standard-setting method for the spring 2019 Royal College of Physicians and Surgeons of Canada internal medicine certification examination consisting of multiple-choice questions","year":2020,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Psychometric Methodologies and Testing","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":true,"ca_institutions":"University of Manitoba; University of Calgary; Royal College of Physicians and Surgeons of Canada","funders":"","keywords":"Specialty; Certification; Nuclear medicine; Medicine; Statistics; Psychology; Mathematics; Family medicine; Management","score_opus":0.3982188306808459,"score_gpt":0.5391948166882257,"score_spread":0.14097598600737976,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3016806524","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97979,0.0007384244,0.008965011,0.0004659246,0.0001669128,0.0007902809,0.00055391423,0.00022617486,0.008303272],"genre_scores_gemma":[0.98195595,0.00017004936,0.014545349,0.000117836935,0.000042964428,0.0003363333,0.0005558051,0.000028454602,0.0022472723],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.96336335,0.014300113,0.0037714264,0.0025394713,0.014706663,0.0013189979],"domain_scores_gemma":[0.8994297,0.031353503,0.01591746,0.005840227,0.0444058,0.003053316],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.038954146,0.000507578,0.0005716003,0.00202663,0.0007756582,0.0010284039,0.0008865441,0.00065081683,0.0025733123],"category_scores_gemma":[0.09449459,0.00028235425,0.0007209514,0.0010080175,0.0005763513,0.00080463814,0.0017822566,0.0005698973,0.0009696],"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":[0.0020785884,0.0006494136,0.82398117,0.0004968696,0.0002402762,0.00017319208,0.002423283,0.0010348293,0.0038559635,0.00028802638,0.0038036364,0.16097477],"study_design_scores_gemma":[0.00010244842,0.0013241125,0.98165995,0.00020807458,0.000044899876,0.00033062557,0.0009703499,0.005538328,0.0060223453,0.00014078226,0.003598031,0.00006016375],"about_ca_topic_score_codex":0.010112405,"about_ca_topic_score_gemma":0.022684107,"teacher_disagreement_score":0.96104586,"about_ca_system_score_codex":0.0026419633,"about_ca_system_score_gemma":0.0046829423,"threshold_uncertainty_score":0.20601171},"labels":[],"label_agreement":null},{"id":"W4294991525","doi":"10.3352/jeehp.2022.19.26","title":"Acceptability of the 8-case objective structured clinical examination of medical students in Korea using generalizability theory: a reliability study","year":2022,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Objective structured clinical examination; Statistics; Psychology; Variance components; Medicine; Mathematics; Medical education","score_opus":0.1615378778704149,"score_gpt":0.599957094874539,"score_spread":0.43841921700412406,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4294991525","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.99717003,0.00007176429,0.0017084223,0.00007302383,0.000015212007,0.00013560524,0.00004116905,0.000010636943,0.00077420264],"genre_scores_gemma":[0.9984328,0.000027356635,0.0012807072,0.000018725585,0.000006373624,0.00009087496,0.00006806531,0.0000042553843,0.00007077766],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9783343,0.012654609,0.0026825767,0.0016000366,0.0042425487,0.0004860458],"domain_scores_gemma":[0.9241998,0.03607918,0.013520218,0.008440002,0.016420918,0.0013398485],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.035618883,0.00036442396,0.00037317435,0.0014743115,0.00050784246,0.0009338128,0.0006935053,0.0005116077,0.0006566835],"category_scores_gemma":[0.089926995,0.00031476008,0.0010117004,0.0008446589,0.0011624165,0.0011782752,0.0016289754,0.0006768523,0.00015282778],"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":[0.00025268926,0.00022096034,0.96834075,0.000118736716,0.00014757153,0.00006999583,0.005814797,0.00039573773,0.0008584018,0.00017649538,0.00016876584,0.023434995],"study_design_scores_gemma":[0.00003113487,0.0012306315,0.98649174,0.00009950587,0.00012716232,0.0002787389,0.0054205423,0.00393379,0.0010830755,0.0002646406,0.0010058774,0.000033283373],"about_ca_topic_score_codex":0.0012367978,"about_ca_topic_score_gemma":0.0018209473,"teacher_disagreement_score":0.035618883,"about_ca_system_score_codex":0.00090091606,"about_ca_system_score_gemma":0.0012180947,"threshold_uncertainty_score":0.18837291},"labels":[],"label_agreement":null},{"id":"W4383059115","doi":"10.3352/jeehp.2023.20.22","title":"Experience of introducing an electronic health records station in an objective structured clinical examination to evaluate medical students’ communication skills in Canada: a descriptive study","year":2023,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Calgary; Medical Council of Canada; University of Ottawa","funders":"University of Ottawa","keywords":"Objective structured clinical examination; Thematic analysis; Focus group; Medical education; Communication skills; Perception; Psychology; Medicine; Qualitative research","score_opus":0.11073334401227708,"score_gpt":0.5808254139298485,"score_spread":0.47009206991757146,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4383059115","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.999383,0.00005375125,0.000058545415,0.00006843836,0.0000027125523,0.000047664085,0.000044570872,0.0000031460336,0.00033810793],"genre_scores_gemma":[0.99825543,0.00024527294,0.00033728377,0.00017842778,0.0000055325845,0.00005215557,0.0000840059,0.000005653715,0.00083638896],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9974861,0.00059761776,0.00020502285,0.00019507488,0.0008636198,0.0006525752],"domain_scores_gemma":[0.9862744,0.0039926916,0.002491094,0.00026202181,0.0041859928,0.0027937482],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0030123747,0.00032251276,0.0004621106,0.0014327536,0.003132655,0.0018883097,0.00087347924,0.0007719469,0.0012456438],"category_scores_gemma":[0.014322799,0.00063460186,0.00043836556,0.0012838043,0.0013379642,0.00084513676,0.0017070351,0.001363476,0.00022707152],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00022338594,0.0014126461,0.73495704,0.0001914678,0.000027283935,0.0017355242,0.23112325,0.00011312178,0.0025060321,0.00008815303,0.0005450928,0.027076969],"study_design_scores_gemma":[0.00002138626,0.0017475503,0.7558409,0.00017041924,0.000038233415,0.0012311193,0.23439835,0.0006122646,0.0014685314,0.00002721512,0.0043573165,0.000086766064],"about_ca_topic_score_codex":0.3415388,"about_ca_topic_score_gemma":0.5355643,"teacher_disagreement_score":0.98936725,"about_ca_system_score_codex":0.010632779,"about_ca_system_score_gemma":0.018285887,"threshold_uncertainty_score":0.6791016},"labels":[],"label_agreement":null},{"id":"W4403385134","doi":"10.3352/jeehp.2024.21.28","title":"The legality and appropriateness of keeping Korean Medical Licensing Examination items confidential: a comparative analysis and review of court rulings","year":2024,"lang":"en","type":"review","venue":"Journal of Educational Evaluation for Health Professions","topic":"Medical Research and Treatments","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Principle of legality; Confidentiality; United States Medical Licensing Examination; Law; Psychology; Computer science; Medical education; Political science; Medicine; Medical school","score_opus":0.28880224413117184,"score_gpt":0.6211246266233137,"score_spread":0.33232238249214185,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4403385134","genre_codex":"empirical","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.71446973,0.15633202,0.003303329,0.016295655,0.00072079676,0.00042944754,0.0007818777,0.000033243163,0.10763381],"genre_scores_gemma":[0.9594284,0.032021318,0.0011463234,0.0051573142,0.00016977711,0.0000889434,0.0003405084,0.000025779404,0.0016216645],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9645921,0.010372532,0.0069182334,0.003816962,0.011777939,0.0025222274],"domain_scores_gemma":[0.7287083,0.21664953,0.020639969,0.005666335,0.027073529,0.0012623543],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06302637,0.00029817584,0.00071680767,0.010689123,0.0043374286,0.0059969635,0.0025822634,0.0033324617,0.0026637653],"category_scores_gemma":[0.16547884,0.00049451203,0.0011332704,0.009103377,0.005591025,0.004798939,0.0024696763,0.0033247594,0.00030839804],"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":[0.0011206339,0.0004549557,0.20914008,0.00921923,0.0014136597,0.010270763,0.04761018,0.0013705076,0.0036720908,0.32927275,0.027354898,0.35910013],"study_design_scores_gemma":[0.00025611636,0.0008593121,0.43235353,0.023937546,0.005077432,0.012450414,0.07748363,0.0028312982,0.0085094515,0.0332392,0.40221366,0.0007885197],"about_ca_topic_score_codex":0.016934708,"about_ca_topic_score_gemma":0.028678488,"teacher_disagreement_score":0.06302637,"about_ca_system_score_codex":0.0073384484,"about_ca_system_score_gemma":0.008068507,"threshold_uncertainty_score":0.3333193},"labels":[],"label_agreement":null},{"id":"W4404094486","doi":"10.3352/jeehp.2024.21.31","title":"Validation of the Blended Learning Usability Evaluation–Questionnaire (BLUE-Q) through an innovative Bayesian questionnaire validation approach","year":2024,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Health Education and Validation","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; McGill University Health Centre; University of Toronto","funders":"Canadian Institutes of Health Research; McGill University","keywords":"Usability; Likert scale; Computer science; Cronbach's alpha; CLARITY; Scale (ratio); Subject-matter expert; Knowledge management; Medical education; Psychology; Artificial intelligence; Statistics; Psychometrics; Medicine; Mathematics; Human–computer interaction; Clinical psychology; Expert system","score_opus":0.20007361076420874,"score_gpt":0.5531405136365891,"score_spread":0.3530669028723804,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4404094486","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.3366077,0.00056658214,0.5978765,0.001388648,0.00037716347,0.05115503,0.00089847785,0.0008568305,0.010273096],"genre_scores_gemma":[0.48785818,0.00030442042,0.42020243,0.0010867589,0.000109139066,0.087385446,0.0010046061,0.00026781717,0.0017812649],"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.8128926,0.12953742,0.015125196,0.006060847,0.034331493,0.0020524268],"domain_scores_gemma":[0.645056,0.22416721,0.012814434,0.023470944,0.09272143,0.0017698483],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.17639907,0.0012191603,0.0008379108,0.0032133835,0.0011053776,0.0027703147,0.0016473385,0.0011176047,0.0026556258],"category_scores_gemma":[0.28464392,0.0005541777,0.0015306146,0.0017956065,0.0021240423,0.002771299,0.0039077327,0.0018640866,0.0012250755],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0020824794,0.004295896,0.116406135,0.005072811,0.0004280056,0.00040371626,0.06051403,0.0063999575,0.033640612,0.013147956,0.01161907,0.7459894],"study_design_scores_gemma":[0.0025700561,0.025853304,0.5032915,0.009819869,0.0009207679,0.0017796784,0.051834535,0.11189655,0.06938281,0.032185525,0.18928955,0.0011759294],"about_ca_topic_score_codex":0.0010394185,"about_ca_topic_score_gemma":0.001646893,"teacher_disagreement_score":0.17639907,"about_ca_system_score_codex":0.0021841729,"about_ca_system_score_gemma":0.0056481585,"threshold_uncertainty_score":0.9328986},"labels":[],"label_agreement":null},{"id":"W4406408913","doi":"10.3352/jeehp.2025.22.5","title":"Reliability and construct validation of the Blended Learning Usability Evaluation–Questionnaire with interprofessional clinicians in Canada: a methodological study","year":2025,"lang":"en","type":"article","venue":"Journal of Educational Evaluation for Health Professions","topic":"Interprofessional Education and Collaboration","field":"Health Professions","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Juravinski Cancer Centre; The Wilson Centre; McGill University; McMaster University; University of Toronto","funders":"Health Canada; Government of Canada","keywords":"Cronbach's alpha; Usability; Construct validity; Likert scale; Credibility; Psychology; Asynchronous communication; Construct (python library); Reliability (semiconductor); Computer science; Psychometrics; Clinical psychology; Human–computer interaction; Developmental psychology","score_opus":0.15433952411410862,"score_gpt":0.5778210872841091,"score_spread":0.42348156317000046,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406408913","genre_codex":"empirical","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"metacan-v3-hybrid-931329e0061c","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.97940207,0.000770108,0.009584196,0.0003636895,0.000040921954,0.005163126,0.0004790295,0.00005011267,0.0041467366],"genre_scores_gemma":[0.9687789,0.00068287936,0.020664353,0.00022930324,0.000016645019,0.007257164,0.000642297,0.000046167075,0.00168234],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9528499,0.016208783,0.005291041,0.0035216466,0.0197318,0.0023967333],"domain_scores_gemma":[0.8357707,0.03614023,0.008557543,0.0053737345,0.11094681,0.0032109707],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06701522,0.00074208487,0.0011892125,0.00477665,0.005126698,0.002906474,0.0022375453,0.00080461666,0.0011978158],"category_scores_gemma":[0.11429892,0.0009512048,0.0013880546,0.006808618,0.0030051973,0.0012235979,0.0035030774,0.001115164,0.00034202056],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065200584,0.001062406,0.7777809,0.0017520314,0.0003277219,0.00047162775,0.075131245,0.0009605661,0.0038394553,0.0008383143,0.0013802649,0.13580345],"study_design_scores_gemma":[0.00018118508,0.0011481689,0.9528847,0.0011338188,0.00023471592,0.0002169176,0.030245434,0.003457589,0.0030916976,0.00024148672,0.0070418334,0.00012248589],"about_ca_topic_score_codex":0.67192584,"about_ca_topic_score_gemma":0.7085747,"teacher_disagreement_score":0.32807416,"about_ca_system_score_codex":0.03646896,"about_ca_system_score_gemma":0.070041955,"threshold_uncertainty_score":0.6600125},"labels":[],"label_agreement":null}]}