{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00062528,0.00009362714,0.0004256706,0.00008979149,0.000128935,0.000005345299,0.0001703049,0.00002927424,0.0001930111],"category_scores_gemma":[0.001299776,0.00004487746,0.0005777073,0.0005260211,0.0009018587,0.00003395661,0.00008947255,0.0005412798,0.000002583592],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001621877,"about_ca_system_score_gemma":0.0009651521,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003200102,"about_ca_topic_score_gemma":0.000002987664,"domain_scores_codex":[0.998298,0.00013975,0.000456977,0.00009916612,0.0008051583,0.0002008744],"domain_scores_gemma":[0.9981098,0.0002890943,0.0004526326,0.0004848933,0.0005219384,0.0001416128],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002062018,0.0004965435,0.9085923,0.00001291035,0.0009656653,0.00007112612,0.0008895198,0.0004498525,0.04412065,0.0001514212,0.04279981,0.001243979],"study_design_scores_gemma":[0.0004899542,0.003535596,0.9832261,0.0001297217,0.0000801533,0.001087107,0.000770347,0.000006690922,0.007046503,0.00006217548,0.003510117,0.00005559451],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9757399,0.00004992981,0.00001824176,0.02094483,0.0001627604,0.0002479398,0.00001930035,0.000003564877,0.002813521],"genre_scores_gemma":[0.9987797,0.00006321632,0.0001642785,0.0007097573,0.0002005812,0.000002787807,0.00000109453,0.00001167954,0.00006687875],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07463372,"threshold_uncertainty_score":0.4837618,"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."}}