{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08121941,0.0008220013,0.001195197,0.001314278,0.00090833,0.002837384,0.002069063,0.002100747,0.001448672],"category_scores_gemma":[0.3915642,0.001132364,0.001829422,0.001528295,0.003248254,0.002659955,0.001702129,0.002400545,0.0002635253],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004826555,"about_ca_system_score_gemma":0.002578485,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02092778,"about_ca_topic_score_gemma":0.009325234,"domain_scores_codex":[0.9427946,0.04686335,0.002030247,0.002926772,0.004070041,0.001315056],"domain_scores_gemma":[0.2386923,0.7112992,0.02282696,0.01891766,0.007041935,0.001221998],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004543732,0.0007988493,0.4939283,0.0002025783,0.001562209,0.000398441,0.000799931,0.464975,0.000585121,0.008542726,0.0007384513,0.02292462],"study_design_scores_gemma":[0.0004399083,0.0009460282,0.0637141,0.000121909,0.0005007866,0.000452122,0.0001858549,0.9235775,0.002337772,0.007086002,0.0005136765,0.0001244211],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9722786,0.0004852821,0.0246882,0.0005894819,0.00002208651,0.0002724834,0.0003353802,0.00007047626,0.00125786],"genre_scores_gemma":[0.9947419,0.00008156679,0.004731533,0.00008247171,0.000009169546,0.00006132173,0.000154956,0.00001324911,0.0001237479],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9187806,"threshold_uncertainty_score":0.4295344,"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."}}