{"id":"W2084057771","doi":"10.1001/jama.296.9.1094","title":"Accuracy of Physician Self-assessment Compared With Observed Measures of Competence","year":2006,"lang":"en","type":"review","venue":"JAMA","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2341,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; PsycINFO; Competence (human resources); CINAHL; Specialty; MEDLINE; Self-assessment; Family medicine; Population; Continuing medical education; Medical education; Nursing; Continuing education; Psychological intervention; Psychology; Environmental health; Social psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.06695145,0.0004181819,0.001784408,0.00545255,0.0002786758,0.002782601,0.00115024,0.001022401,0.001848523],"category_scores_gemma":[0.3965445,0.0003824149,0.002125689,0.005768043,0.00113002,0.003050689,0.001402073,0.000583665,0.000295895],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001097786,"about_ca_system_score_gemma":0.00123829,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001826652,"about_ca_topic_score_gemma":0.002824029,"domain_scores_codex":[0.9176611,0.0497301,0.01716023,0.005359805,0.009636251,0.00045246],"domain_scores_gemma":[0.5431274,0.3626607,0.06034766,0.01165556,0.02163542,0.0005732157],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004485258,0.0001871824,0.7784094,0.03546258,0.0114477,0.0001686953,0.005285954,0.0009354528,0.0007191793,0.0008116433,0.002113453,0.1599736],"study_design_scores_gemma":[0.001126251,0.002110095,0.9372545,0.02958127,0.009151728,0.000880634,0.002863976,0.003495923,0.002316309,0.001906389,0.009144766,0.0001681898],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"review","genre_scores_codex":[0.8191633,0.1282587,0.01574999,0.001562276,0.0007676857,0.002451574,0.01148994,0.0001335991,0.02042304],"genre_scores_gemma":[0.9850559,0.007238897,0.004381622,0.0002712806,0.0001202745,0.0007096125,0.001790687,0.00001657355,0.0004151038],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.06695145,"threshold_uncertainty_score":0.3540773,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09410947641229347,"score_gpt":0.3889888834190581,"score_spread":0.2948794070067646,"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."}}