{"id":"W1970322521","doi":"10.1016/j.ahj.2004.04.047","title":"Impact of the choice of benchmark on the conclusions of hospital report cards","year":2004,"lang":"en","type":"article","venue":"American Heart Journal","topic":"Acute Myocardial Infarction Research","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":false,"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre; Institute for Clinical Evaluative Sciences; Women's College Hospital; University of Toronto","funders":"","keywords":"Medicine; Benchmark (surveying); MEDLINE; Intensive care medicine","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.08072396,0.000648511,0.001485033,0.001569024,0.001263296,0.007737431,0.002277237,0.003014958,0.008575782],"category_scores_gemma":[0.4511937,0.0004583301,0.0006736221,0.001465824,0.001052796,0.002777572,0.002209283,0.003264243,0.001350946],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003238385,"about_ca_system_score_gemma":0.003156491,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006608221,"about_ca_topic_score_gemma":0.004291962,"domain_scores_codex":[0.9253229,0.05474616,0.003923082,0.003367812,0.009665078,0.002974849],"domain_scores_gemma":[0.3847218,0.5594528,0.01556644,0.01427977,0.01793442,0.008044763],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.1006132,0.006838131,0.3726516,0.0007515689,0.001660443,0.0009671028,0.001926976,0.08392426,0.004981817,0.02079105,0.03170883,0.3731849],"study_design_scores_gemma":[0.007744135,0.01721472,0.5741654,0.0009242244,0.001778123,0.0006846602,0.008391513,0.295591,0.02162456,0.05429272,0.01688279,0.0007061384],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9486709,0.001677856,0.01056965,0.01264983,0.0009936766,0.0002572751,0.001437146,0.0005910186,0.02315267],"genre_scores_gemma":[0.9954833,0.00007388466,0.00247511,0.0005541723,0.0001019249,0.00003776518,0.0003991043,0.0000453708,0.0008294114],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9192761,"threshold_uncertainty_score":0.4269142,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02616120427723125,"score_gpt":0.3843116044729369,"score_spread":0.3581504001957057,"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."}}