{"id":"W3016181295","doi":"10.1097/acm.0000000000003382","title":"Correlations Between the USMLE Step Examinations, American College of Physicians In-Training Examination, and ABIM Internal Medicine Certification Examination","year":2020,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"United States Medical Licensing Examination; Accreditation; Certification; Graduate medical education; Licensure; Medicine; Logistic regression; Family medicine; Educational measurement; Medical education; MEDLINE; Medical school; Psychology; Curriculum; Internal medicine; Management; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001646965,0.0002991394,0.0002531141,0.001076747,0.0002233791,0.0003918148,0.0003855933,0.0004793479,0.003483151],"category_scores_gemma":[0.0179843,0.0002158914,0.0002639666,0.0009757757,0.0003895595,0.0004529005,0.0005543298,0.0007783255,0.0004801055],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003157146,"about_ca_system_score_gemma":0.0007643426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004092145,"about_ca_topic_score_gemma":0.006506849,"domain_scores_codex":[0.9989023,0.0003645353,0.0001304243,0.0001566881,0.0003220083,0.0001240721],"domain_scores_gemma":[0.9881637,0.003982601,0.004747197,0.0003802477,0.001391861,0.00133446],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001203774,0.00001795896,0.9995006,0.000001719519,0.000009034234,0.000006913114,0.000006841813,0.00002454268,0.000009703093,0.000004299718,0.00005640498,0.0003500789],"study_design_scores_gemma":[0.000001533963,0.00003236948,0.9996525,0.000002686588,0.000003915962,0.00005987567,0.00002640839,0.0001411156,0.00001930087,0.000006998109,0.00005217818,0.000001011857],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9986136,0.0001101637,0.0001052142,0.00009647026,0.000007595278,0.00001002844,0.000417583,0.000006941759,0.0006323782],"genre_scores_gemma":[0.9988864,0.00004974037,0.000186833,0.00002788423,0.00001188649,0.00001319754,0.0005797588,0.000002139452,0.0002420614],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004092145,"threshold_uncertainty_score":0.01165229,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08821989476984082,"score_gpt":0.333380640570177,"score_spread":0.2451607458003362,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}