{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002819308,0.0001815946,0.0004906639,0.0003963605,0.0007254701,0.000009457915,0.0002951371,0.000158799,0.0001011287],"category_scores_gemma":[0.00190977,0.000175722,0.0001383139,0.0002641014,0.00003227642,0.0002509524,0.00004959657,0.0007880282,0.000009210771],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003166088,"about_ca_system_score_gemma":0.002456932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004070071,"about_ca_topic_score_gemma":0.0003288712,"domain_scores_codex":[0.9938617,0.001783843,0.002154045,0.0002668511,0.0013456,0.0005879733],"domain_scores_gemma":[0.9933839,0.0006353483,0.001410624,0.0001945811,0.004160257,0.0002152663],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001008995,0.00007696966,0.9752682,0.001015244,0.00003371734,5.660232e-7,0.01323412,0.0007463858,0.00004634663,0.001444405,0.001442628,0.005682422],"study_design_scores_gemma":[0.008610518,0.0008063189,0.922006,0.006683125,0.00001103853,0.000005288233,0.01964177,0.0001903574,0.0002669862,0.000582057,0.04075703,0.0004395114],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9809119,0.0001585088,0.001562741,0.001930532,0.01244967,0.001916804,0.0008055179,0.00002064067,0.0002437366],"genre_scores_gemma":[0.9904914,0.00003545734,0.005624917,0.003227731,0.0002924382,0.0001386683,0.00006564297,0.0000315013,0.0000922477],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0532622,"threshold_uncertainty_score":0.827921,"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."}}