{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003839945,0.0001298125,0.0003633571,0.0002087382,0.0002966058,0.000005011169,0.0004890192,0.00009729608,0.0001016668],"category_scores_gemma":[0.003201564,0.000102875,0.00001851096,0.001306164,0.002407577,0.0002482759,0.00006171161,0.0005208058,0.000004339906],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001015331,"about_ca_system_score_gemma":0.00007375263,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001357768,"about_ca_topic_score_gemma":0.0003105871,"domain_scores_codex":[0.9973203,0.0006104003,0.0005471731,0.000273812,0.001013336,0.0002350107],"domain_scores_gemma":[0.9975656,0.001512324,0.0004260147,0.0001343147,0.0002370442,0.0001246524],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00001617395,0.00001480926,0.2284285,0.00005885391,0.00005198236,0.000005073464,0.5088009,0.00001314861,0.0008243096,0.0395285,0.01802674,0.204231],"study_design_scores_gemma":[0.0007396205,0.0002210253,0.618077,0.0003051494,0.0001149248,0.00000111222,0.3635103,0.0006125246,0.00002804305,0.0005944482,0.0156959,0.0001000007],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8473415,0.0006785385,0.01025641,0.120787,0.000394446,0.0007940097,0.00004428083,0.00007470752,0.01962908],"genre_scores_gemma":[0.9958006,0.0006345651,0.00005442519,0.001815113,0.001006301,0.00001313585,0.00005525476,0.000009415321,0.000611204],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3896485,"threshold_uncertainty_score":0.8870818,"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."}}