{"id":"W1974057428","doi":"10.1177/0163278707307924","title":"Assessment of Family Physicians' Performance Using Patient Charts","year":2007,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Healthcare Quality and Management","field":"Health Professions","cited_by":51,"is_retracted":false,"has_abstract":true,"ca_institutions":"Association des Médecins d'Urgence du Québec","funders":"","keywords":"Chart; Medicine; Inter-rater reliability; Recall; Concordance; Medical record; Reliability (semiconductor); Cohen's kappa; Family medicine; Psychology; Statistics; Rating scale","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":[0.01780286,0.0004655262,0.0006300026,0.004630769,0.0006062025,0.001337649,0.0007109518,0.0004091782,0.001196135],"category_scores_gemma":[0.07206656,0.0002451571,0.0004680733,0.001971614,0.0005599452,0.001171112,0.0008284923,0.0004339313,0.0003976082],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009758687,"about_ca_system_score_gemma":0.001317231,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00651631,"about_ca_topic_score_gemma":0.0064237,"domain_scores_codex":[0.9779211,0.01180361,0.00237121,0.0008729511,0.00662321,0.0004079536],"domain_scores_gemma":[0.9328633,0.02514526,0.01454051,0.002664081,0.02298682,0.001800061],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0007057644,0.0003547972,0.8581768,0.0003588099,0.0002677048,0.0003326646,0.006742857,0.001536173,0.002186737,0.0002556478,0.003129346,0.1259527],"study_design_scores_gemma":[0.000064966,0.001542307,0.9764382,0.0001816317,0.00009595118,0.0009622377,0.003075844,0.007401996,0.005124522,0.0002711875,0.004747474,0.00009364288],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9824405,0.001017592,0.006874078,0.0003080531,0.00004905448,0.0005403616,0.001184013,0.0002530495,0.00733322],"genre_scores_gemma":[0.9896718,0.0004828058,0.008256093,0.00003564517,0.00004877719,0.000193612,0.0005572921,0.00001273296,0.0007412917],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01780286,"threshold_uncertainty_score":0.09415168,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4189002070409529,"score_gpt":0.6045903599396676,"score_spread":0.1856901528987147,"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."}}