{"id":"W2781975634","doi":"10.1007/s40670-017-0518-x","title":"Using the NBME® Comprehensive Subject Examinations to Assess Readiness for USMLE® Step 1 and Step 2 CK: a Comparison of US/Canadian and International Medical School Students","year":2018,"lang":"en","type":"article","venue":"Medical Science Educator","topic":"Innovations in Medical Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":false,"ca_institutions":"","funders":"","keywords":"United States Medical Licensing Examination; Logistic regression; Medical education; Variance (accounting); Subject (documents); Medical school; Medicine; Psychology; Family medicine; Computer science; Internal medicine; Library 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.002623671,0.0001473803,0.0003060384,0.0005263904,0.0005281604,0.00008168838,0.0007308653,0.0001498214,0.0008450242],"category_scores_gemma":[0.01713971,0.0001056015,0.0000270232,0.001176946,0.002633859,0.0001749739,0.0002091547,0.0003429385,0.000009515666],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003782041,"about_ca_system_score_gemma":0.00584808,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002900671,"about_ca_topic_score_gemma":0.003106579,"domain_scores_codex":[0.9962792,0.00006949112,0.0005220646,0.000437986,0.002265342,0.000425848],"domain_scores_gemma":[0.996325,0.0003985312,0.0001377448,0.000313212,0.001392226,0.001433341],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00009231646,0.0009842974,0.8396364,0.0001473754,0.0001073097,0.000006151254,0.004607937,6.254069e-7,0.001675359,0.003196166,0.04313559,0.1064104],"study_design_scores_gemma":[0.001739279,0.0005264816,0.8725207,0.0006872158,0.0001122261,0.0001949612,0.01418582,0.02620908,0.0006111227,0.00009531576,0.0828394,0.0002783496],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9704929,0.00005171,0.004506366,0.02135071,0.002524392,0.0007957692,0.00001500837,0.00001362409,0.0002495005],"genre_scores_gemma":[0.9825553,0.000008497183,0.01071895,0.005415725,0.001038307,0.00008193069,0.00001508966,0.00001304,0.0001531591],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1061321,"threshold_uncertainty_score":0.9997879,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1228355728255379,"score_gpt":0.4935361474009314,"score_spread":0.3707005745753935,"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."}}