{"id":"W2013246457","doi":"10.1097/acm.0b013e3181b37b0b","title":"Use of Multimedia on the Step 1 and Step 2 Clinical Knowledge Components of USMLE: A Controlled Trial of the Impact on Item Characteristics","year":2009,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Phonocardiography and Auscultation Techniques","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Terminology; Auscultation; Multimedia; United States Medical Licensing Examination; Multiple choice; Licensure; Educational measurement; Medical education; Medicine; Computer science; Psychology; Curriculum; Linguistics; Medical school; Radiology; Pedagogy","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":[0.001705947,0.0001991073,0.001268567,0.0001669486,0.00003418134,0.00000141009,0.0001891202,0.0002576881,0.00002152115],"category_scores_gemma":[0.004541109,0.00008395383,0.0003164395,0.0003042218,0.0005759348,0.00003346177,0.00002727758,0.0007585892,8.17531e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001614937,"about_ca_system_score_gemma":0.00005687261,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002613417,"about_ca_topic_score_gemma":1.984858e-7,"domain_scores_codex":[0.9974819,0.0004517666,0.001273291,0.0001919654,0.0004453371,0.0001557684],"domain_scores_gemma":[0.9943689,0.004044699,0.0008659183,0.0004048602,0.0001967584,0.0001188295],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.5579217,0.002659549,0.09936488,0.0003869439,0.002595809,0.00001295725,0.007086652,0.000002293905,0.1459173,0.00222176,0.04900617,0.132824],"study_design_scores_gemma":[0.1814653,0.01129882,0.7915275,0.003240395,0.0008518673,0.00001091562,0.0002430807,0.002048895,0.007965753,0.0001919622,0.00100156,0.0001539845],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9954215,0.000274664,0.00005786072,0.001534158,0.0002433237,0.002079683,0.00003705065,0.00002083419,0.0003309292],"genre_scores_gemma":[0.9972852,0.001487728,0.00005490495,0.0006770712,0.0003592139,0.00001898541,0.00001667343,0.00001029252,0.0000899618],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6921626,"threshold_uncertainty_score":0.5436459,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1006634845693079,"score_gpt":0.4024753221029536,"score_spread":0.3018118375336457,"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."}}