{"id":"W4379600149","doi":"10.1097/acm.0000000000005290","title":"Deep Learning Model for Automated Trainee Assessment During High-Fidelity Simulation","year":2023,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; SickKids Foundation; Hospital for Sick Children","funders":"Medical Council of Canada","keywords":"Computer science; Inter-rater reliability; Deep learning; Modalities; Competence (human resources); Artificial intelligence; Medical physics; Simulation; Machine learning; Medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001236198,0.0002205372,0.0004504878,0.000423142,0.0002732396,0.000004695726,0.0001130752,0.0003344876,0.0001611551],"category_scores_gemma":[0.001632834,0.0001996493,0.00005906697,0.0007393307,0.00007713612,0.0001189098,0.00002618863,0.0008048718,0.00003672935],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004595285,"about_ca_system_score_gemma":0.0003138077,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00006513829,"about_ca_topic_score_gemma":0.000001642079,"domain_scores_codex":[0.9974967,0.0000891752,0.0008830712,0.0004508469,0.0006000668,0.0004800868],"domain_scores_gemma":[0.997411,0.00120269,0.0002615044,0.0003020389,0.0005123799,0.000310398],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0001008253,0.000009021699,0.04829954,0.0005670357,0.00003603223,0.000002901881,0.006111162,0.9395167,0.002461473,0.0003522521,0.001659533,0.0008835415],"study_design_scores_gemma":[0.002583164,0.0001037446,0.1879844,0.0003016482,0.00009613682,0.000004648356,0.001894966,0.8058332,0.00005406294,0.0009236949,0.00009189229,0.0001285019],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9216297,0.0001535059,0.05958498,0.01469069,0.0005574939,0.001351331,0.000006307757,0.001933177,0.00009281932],"genre_scores_gemma":[0.9932767,0.00008151467,0.001453086,0.001030345,0.001157494,0.000240581,0.0008568815,0.00006503721,0.001838329],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1396849,"threshold_uncertainty_score":0.8141463,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09712288966503065,"score_gpt":0.4675052864311365,"score_spread":0.3703823967661059,"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."}}