{"id":"W4411016362","doi":"10.1097/cce.0000000000001275","title":"Mortality Prediction Performance Under Geographical, Temporal, and COVID-19 Pandemic Dataset Shift: External Validation of the Global Open-Source Severity of Illness Score Model","year":2025,"lang":"en","type":"article","venue":"Critical Care Explorations","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Generalizability theory; Medicine; Standardized mortality ratio; Cohort; Pandemic; Health care; APACHE II; Emergency medicine; Cohort study; Intensive care; Receiver operating characteristic; Predictive modelling; Coronavirus disease 2019 (COVID-19); Intensive care unit; Intensive care medicine; Disease; Internal medicine; Machine learning; Statistics; Computer science; Infectious disease (medical specialty)","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":[],"consensus_categories":[],"category_scores_codex":[0.01399437,0.001610045,0.0009073742,0.001199005,0.0005390985,0.001006478,0.001455134,0.001139156,0.001049092],"category_scores_gemma":[0.01572254,0.0002142482,0.001413153,0.0008795681,0.0008506091,0.0008530225,0.001961903,0.00138901,0.000740633],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001100616,"about_ca_system_score_gemma":0.001691138,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01588348,"about_ca_topic_score_gemma":0.01043499,"domain_scores_codex":[0.9969015,0.001387345,0.0002189844,0.0009214301,0.000355111,0.0002156322],"domain_scores_gemma":[0.9924776,0.002907871,0.0006918645,0.001577531,0.00178216,0.0005629869],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.005987153,0.00253262,0.7023721,0.0006053861,0.002415401,0.0005997187,0.0007628681,0.1567891,0.00524638,0.001070005,0.04393156,0.07768764],"study_design_scores_gemma":[0.0009412363,0.002177423,0.3750271,0.0002564711,0.0006550251,0.0005316342,0.0007610724,0.6048843,0.005284197,0.001444353,0.007850989,0.0001861459],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9814758,0.0004277386,0.003927163,0.0003756572,0.000198715,0.0001628428,0.01198787,0.0004615277,0.0009827088],"genre_scores_gemma":[0.9362086,0.0001204985,0.005210924,0.0001892817,0.00007334158,0.0001658196,0.05747653,0.00007234175,0.0004825988],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01588348,"threshold_uncertainty_score":0.07401019,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2056895533870719,"score_gpt":0.4306377547337448,"score_spread":0.224948201346673,"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."}}