{"id":"W2123871708","doi":"10.1093/intqhc/mzg037","title":"Using standardized patients to measure professional performance of physicians","year":2003,"lang":"en","type":"article","venue":"International Journal for Quality in Health Care","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":54,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Medicine; Coding (social sciences); Standardization; Family medicine; Randomized controlled trial; MEDLINE; Cohen's kappa; Standardized test; Psychology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"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":[0.02972602,0.0005752835,0.0006479328,0.0009170964,0.000450712,0.0008229061,0.0005424391,0.001033658,0.001978486],"category_scores_gemma":[0.131012,0.0003543662,0.0007370376,0.0009113807,0.001321582,0.001042095,0.001035522,0.0009540356,0.0003319077],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007144218,"about_ca_system_score_gemma":0.001040518,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006850401,"about_ca_topic_score_gemma":0.0006132922,"domain_scores_codex":[0.9171305,0.06566793,0.005749894,0.001684576,0.009156927,0.0006100244],"domain_scores_gemma":[0.8351527,0.09652672,0.05260849,0.006025153,0.008243171,0.001443639],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0240882,0.004504126,0.7390257,0.002633859,0.001215482,0.0001492397,0.005726473,0.003892839,0.003179128,0.0007716721,0.002975371,0.2118378],"study_design_scores_gemma":[0.006794503,0.0622816,0.9046678,0.001066276,0.0005740667,0.0005622713,0.002959778,0.007658902,0.006352949,0.001153115,0.005703192,0.0002255456],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9837377,0.0008207532,0.007946295,0.0005074127,0.0001503168,0.002827858,0.00066238,0.0000482005,0.003298958],"genre_scores_gemma":[0.9904466,0.0003200012,0.005633623,0.0001862955,0.00009903255,0.002783006,0.0003020038,0.000005266734,0.0002242079],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02972602,"threshold_uncertainty_score":0.1572081,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2007847132343422,"score_gpt":0.5563869154153513,"score_spread":0.3556022021810091,"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."}}