{"id":"W4399771702","doi":"10.2196/45973","title":"Performance Drift in Machine Learning Models for Cardiac Surgery Risk Prediction: Retrospective Analysis","year":2024,"lang":"en","type":"article","venue":"JMIRx Med","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"British Heart Foundation; University of Bristol; National Institute for Health and Care Research","keywords":"EuroSCORE; Concept drift; Medicine; Metric (unit); Cardiac surgery; Data set; Retrospective cohort study; Performance metric; Generalization; Set (abstract data type); Artificial intelligence; Machine learning; Computer science; Cardiology; Internal medicine; Engineering; Operations management; Data stream mining; Mathematics","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03417194,0.0007247266,0.0007196324,0.002203641,0.0004106308,0.001344166,0.001133134,0.0006600284,0.0004761154],"category_scores_gemma":[0.08523844,0.0003709055,0.001358643,0.002228722,0.0007472034,0.001384377,0.001354474,0.001294352,0.0002951931],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0010894,"about_ca_system_score_gemma":0.001104933,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0056123,"about_ca_topic_score_gemma":0.004586238,"domain_scores_codex":[0.9889551,0.004400187,0.001835418,0.001660296,0.002742488,0.0004064415],"domain_scores_gemma":[0.9036466,0.04836862,0.02148251,0.01242603,0.0128498,0.001226443],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002779027,0.000052591,0.9684579,0.00006273326,0.0004227912,0.0001066999,0.0002269193,0.01784353,0.0002313958,0.0002334783,0.0009150051,0.01116889],"study_design_scores_gemma":[0.00003451594,0.001140756,0.7676593,0.0001845724,0.0004270192,0.0009432995,0.000535811,0.2220615,0.002566887,0.001576391,0.002768395,0.0001014813],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9799471,0.00119796,0.01552837,0.0003422431,0.0000506642,0.0000915731,0.002245907,0.0001139583,0.0004821028],"genre_scores_gemma":[0.9936498,0.0001601596,0.003333217,0.00004170914,0.00002489054,0.00003833258,0.002635217,0.00002254432,0.00009415476],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9658281,"threshold_uncertainty_score":0.1807206,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1038413284884274,"score_gpt":0.3741849249451892,"score_spread":0.2703435964567619,"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."}}