{"id":"W4387950762","doi":"10.2196/50895","title":"Temporal Generalizability of Machine Learning Models for Predicting Postoperative Delirium Using Electronic Health Record Data: Model Development and Validation Study","year":2023,"lang":"en","type":"article","venue":"JMIR Perioperative Medicine","topic":"Intensive Care Unit Cognitive Disorders","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Japan Society for the Promotion of Science; Ministry of Health, Labour and Welfare","keywords":"Generalizability theory; Brier score; Receiver operating characteristic; Delirium; Logistic regression; Medicine; Artificial intelligence; Statistics; Machine learning; Decision tree; Computer science; Mathematics; Intensive care medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.05178168,0.001312002,0.0008290156,0.001237792,0.0004886411,0.001121359,0.001393452,0.0008999354,0.0008535665],"category_scores_gemma":[0.06800167,0.0004932242,0.002156694,0.0008053163,0.0005902768,0.001849399,0.001480665,0.00191471,0.0002872797],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001170962,"about_ca_system_score_gemma":0.001830816,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009684199,"about_ca_topic_score_gemma":0.005679979,"domain_scores_codex":[0.9914113,0.005988439,0.0005855172,0.001140999,0.00060294,0.0002706866],"domain_scores_gemma":[0.931341,0.05381102,0.003781643,0.004913737,0.00565412,0.0004984464],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.002381737,0.00143016,0.4940192,0.0002027676,0.002120347,0.0002345684,0.0003795907,0.4159401,0.001260612,0.0009479171,0.001548824,0.07953417],"study_design_scores_gemma":[0.0000514555,0.0006660986,0.02511049,0.00003486767,0.0002048214,0.00007298488,0.00006102851,0.9724969,0.0005282786,0.0005238844,0.0002268983,0.0000223963],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9483117,0.0007648506,0.04868698,0.0003587885,0.00007391602,0.0002707017,0.0005974071,0.000232545,0.0007030846],"genre_scores_gemma":[0.9891877,0.0002276927,0.009145375,0.00006776132,0.00003043553,0.0001475383,0.0009793867,0.00002475819,0.0001893273],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05178168,"threshold_uncertainty_score":0.273851,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1325988442796283,"score_gpt":0.3903482400034451,"score_spread":0.2577493957238168,"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."}}