{"id":"W3183832440","doi":"10.1097/acm.0000000000004227","title":"Association Between USMLE Step 1 Scores and In-Training Examination Performance: A Meta-Analysis","year":2021,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Heart, Lung, and Blood Institute","keywords":"Observational study; Association (psychology); United States Medical Licensing Examination; MEDLINE; Educational measurement; Quality (philosophy); Graduate medical education","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01687367,0.002431842,0.01349723,0.003883654,0.000674347,0.003106671,0.002136511,0.002308829,0.002677426],"category_scores_gemma":[0.03516938,0.001446525,0.04871097,0.005234788,0.0006148326,0.00143544,0.001334124,0.002019165,0.0003299284],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001902103,"about_ca_system_score_gemma":0.001988407,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007405786,"about_ca_topic_score_gemma":0.01227377,"domain_scores_codex":[0.9908928,0.004698275,0.002083247,0.00115924,0.0008187285,0.0003477443],"domain_scores_gemma":[0.9730525,0.01928838,0.004242567,0.001246676,0.001770509,0.0003994212],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.006022716,0.00006783514,0.01733976,0.05309036,0.9177203,0.0001059221,0.00004569754,0.0006115929,0.0001654966,0.00005375595,0.0003428856,0.004433554],"study_design_scores_gemma":[0.0009032742,0.0002703767,0.009873564,0.003037663,0.9850336,0.00006000539,0.00002106755,0.0003005356,0.00009564577,0.00007516349,0.0003106605,0.00001849344],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.05495943,0.9408247,0.001554512,0.0003199122,0.0002603997,0.0004254945,0.001118221,0.00006986362,0.0004674286],"genre_scores_gemma":[0.6926525,0.2978377,0.004389376,0.0009613596,0.0003682675,0.001379184,0.001689684,0.00004969301,0.0006720835],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9831263,"threshold_uncertainty_score":0.08923757,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1754051824360542,"score_gpt":0.3628093346019531,"score_spread":0.1874041521658989,"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."}}