{"id":"W2901814463","doi":"10.22374/cjgim.v13i4.280","title":"A CanMEDS Competency-Based Assessment Tool for High-Fidelity Simulation in Internal Medicine: The Montreal Internal Medicine Evaluation Scale (MIMES)","year":2018,"lang":"en","type":"article","venue":"Canadian Journal of General Internal Medicine","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Scale (ratio); Medicine; Delphi; Delphi method; Fidelity; Rating scale; Internal validity; Usability; External validity; Medical education; Psychology; Computer science; Social psychology; Artificial intelligence; Human–computer interaction","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.005486898,0.0004776159,0.001104772,0.001449527,0.0002829418,0.00003550286,0.0006645995,0.0002276874,0.003514752],"category_scores_gemma":[0.002735569,0.0003117954,0.0001802444,0.0006602359,0.001208782,0.0002302827,0.00002506939,0.001010059,0.00000853505],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003260031,"about_ca_system_score_gemma":0.007208319,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2269642,"about_ca_topic_score_gemma":0.1256971,"domain_scores_codex":[0.9938546,0.0005919409,0.002686705,0.0005104882,0.001636602,0.0007196793],"domain_scores_gemma":[0.9901471,0.00154045,0.001325332,0.0006235586,0.005167705,0.001195838],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003573685,0.0002196997,0.7997607,0.0004311625,0.0005761541,0.0002336276,0.01311348,0.02716829,0.002877624,0.001416979,0.09812775,0.05250083],"study_design_scores_gemma":[0.01449043,0.005520919,0.6741625,0.005616616,0.0006658633,0.0003219877,0.002217924,0.2893038,0.0003229782,0.003754752,0.003291963,0.0003301997],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8504634,0.001335352,0.03132739,0.1025368,0.009269094,0.002324185,0.00005503171,0.00002453792,0.002664193],"genre_scores_gemma":[0.9761127,0.00002000187,0.002102932,0.008577587,0.01223104,0.00009460721,0.0001596083,0.00006260853,0.0006388702],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2621355,"threshold_uncertainty_score":0.9999334,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04666496802544063,"score_gpt":0.4084482448404173,"score_spread":0.3617832768149766,"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."}}