{"id":"W1973080342","doi":"10.1016/j.athoracsur.2010.10.035","title":"Evaluating the Reliability and Reproducibility of the Ottawa Thoracic Morbidity and Mortality Classification System","year":2011,"lang":"en","type":"article","venue":"The Annals of Thoracic Surgery","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":62,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Medicine; Reproducibility; Reliability (semiconductor); Reliability engineering; 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":[0.07175469,0.0009265773,0.001537138,0.003738706,0.001201553,0.002721767,0.002102038,0.002149001,0.0007288529],"category_scores_gemma":[0.1789234,0.0008997318,0.002792994,0.002232529,0.002664433,0.00194594,0.002240458,0.00120756,0.0005080011],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001634403,"about_ca_system_score_gemma":0.002155125,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008051021,"about_ca_topic_score_gemma":0.01266733,"domain_scores_codex":[0.9293398,0.03803416,0.01183641,0.006731071,0.01256285,0.001495764],"domain_scores_gemma":[0.7697545,0.1489875,0.02352221,0.02085841,0.03483979,0.002037488],"domain_codex":null,"domain_gemma":"reproducibility","domain_candidate":"reproducibility","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001103171,0.00008270609,0.983859,0.0001274561,0.002326211,0.00006731901,0.00111218,0.001054185,0.0003313163,0.0003535881,0.0007268034,0.008855985],"study_design_scores_gemma":[0.0002114503,0.0008624643,0.977214,0.000141238,0.001680641,0.0003891129,0.001527905,0.01298227,0.001112479,0.00128582,0.00243031,0.0001623003],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.984463,0.002218721,0.006999303,0.0003067981,0.0006040212,0.0002521164,0.001058803,0.00009935409,0.003997848],"genre_scores_gemma":[0.996129,0.0002072093,0.002444593,0.00008281219,0.0001162352,0.0001183289,0.0005868325,0.00003829432,0.0002766647],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9282453,"threshold_uncertainty_score":0.3794796,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5794463714679673,"score_gpt":0.4969691093114348,"score_spread":0.0824772621565325,"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."}}