{"id":"W2080933952","doi":"10.1177/0272989x06290498","title":"The Impact of Unmeasured Clinical Variables on the Accuracy of Hospital Report Cards: A Monte Carlo Study","year":2006,"lang":"en","type":"article","venue":"Medical Decision Making","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto","funders":"","keywords":"Monte Carlo method; Medicine; Emergency medicine; Medical emergency; Statistics","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02375742,0.0001716962,0.0005854964,0.00008231069,0.0009269996,0.00001474951,0.0006116914,0.0003409339,0.0003577169],"category_scores_gemma":[0.08614408,0.00007505716,0.0002652,0.0003724964,0.0002294383,0.00007381818,0.0002145443,0.00144393,0.00003151966],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001372821,"about_ca_system_score_gemma":0.001400113,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001365284,"about_ca_topic_score_gemma":0.0001577885,"domain_scores_codex":[0.9917381,0.001584855,0.00332734,0.0002556764,0.002605487,0.0004885577],"domain_scores_gemma":[0.9662606,0.03033507,0.001700931,0.0008863062,0.0005891803,0.0002279032],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001204755,0.001391171,0.6118138,0.0001598431,0.0001848439,0.0001215556,0.003953647,0.0004025549,0.000002617383,0.002940655,0.1851768,0.1926478],"study_design_scores_gemma":[0.002465486,0.002196597,0.959339,0.003106524,0.00006293263,0.00000902695,0.00452653,0.01545395,0.000001681833,0.004745593,0.007899615,0.0001930784],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9870098,0.0001363696,0.003208643,0.001963163,0.001808662,0.001373572,0.00001132029,0.00004067452,0.004447814],"genre_scores_gemma":[0.9986641,0.00006213852,0.0001683543,0.0003245922,0.0005969846,0.00006968716,0.000002821981,0.00001501631,0.00009627626],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3475251,"threshold_uncertainty_score":0.9215537,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2307971450528092,"score_gpt":0.556669291106792,"score_spread":0.3258721460539828,"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."}}