{"id":"W3155877991","doi":"10.2196/25903","title":"The United States Medical Licensing Exam Step 2 Clinical Skills Examination: Potential Alternatives During and After the COVID-19 Pandemic","year":2021,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Accreditation; Medical education; Coronavirus disease 2019 (COVID-19); United States Medical Licensing Examination; Graduate medical education; Pandemic; Matching (statistics); Medical school; Psychology; Medicine; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.003612076,0.0002177321,0.0002906353,0.000174112,0.0006064475,0.0001142369,0.0002851969,0.0004030782,0.0015091],"category_scores_gemma":[0.03109087,0.0001352547,0.00008893228,0.0009374197,0.00125227,0.000132245,0.0001818965,0.001352055,0.00002319475],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002955126,"about_ca_system_score_gemma":0.00448802,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007557145,"about_ca_topic_score_gemma":0.00008051182,"domain_scores_codex":[0.9952362,0.0009578886,0.0009873109,0.0005123255,0.001922321,0.0003839788],"domain_scores_gemma":[0.9962044,0.001461476,0.000271288,0.0005971973,0.0007655248,0.0007000812],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0003182966,0.003207355,0.1406707,0.0004170574,0.0002637768,0.0002013704,0.008094703,0.000003712584,0.00003877558,0.0009918195,0.04530332,0.8004891],"study_design_scores_gemma":[0.00233638,0.00009970849,0.7809197,0.0005169234,0.0001315606,0.002599853,0.01257457,0.00451665,0.00001976841,0.0003662723,0.195674,0.000244592],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8837197,0.0007490534,0.001046072,0.1116381,0.002119373,0.0005037405,0.00000279121,0.0000844455,0.0001366695],"genre_scores_gemma":[0.9292856,0.003933421,0.0006140104,0.06066906,0.00303021,0.0002831622,0.0005169748,0.00003642893,0.001631102],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8002445,"threshold_uncertainty_score":0.9994037,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02137369388871821,"score_gpt":0.4146807371834129,"score_spread":0.3933070432946947,"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."}}