{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0001717829,0.0001130923,0.0002272933,0.00004378104,0.0002523368,0.00004061062,0.0001385696,0.0000879379,0.00002141352],"category_scores_gemma":[0.0003227434,0.00008413196,0.00006051448,0.0003169962,0.0004051194,0.0002804589,0.0001752089,0.0001088945,4.969284e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001375232,"about_ca_system_score_gemma":0.0003502252,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003449893,"about_ca_topic_score_gemma":0.0001992911,"domain_scores_codex":[0.9989446,0.00008537491,0.000338887,0.0002576268,0.0002585282,0.0001150243],"domain_scores_gemma":[0.9990896,0.0001224878,0.00005889362,0.0004196714,0.0001827744,0.0001265808],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000538113,0.0003422051,0.9879625,0.0003703704,0.00004619484,6.374379e-7,0.0003583684,0.0018646,0.00002777619,0.008255481,0.0004943829,0.0002236337],"study_design_scores_gemma":[0.001614564,0.0001958634,0.972328,0.0003733097,0.0008126972,0.00001125166,0.002399394,0.01007852,0.001298811,0.01045459,0.0002973604,0.0001356247],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9690742,0.0003327812,0.02431446,0.002321222,0.00008456708,0.0005404007,0.003220102,0.00002406056,0.00008815138],"genre_scores_gemma":[0.997475,0.0001051124,0.0002473182,0.0007568737,0.00001717016,0.0001282536,0.001261606,0.000005228978,0.000003393246],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02840078,"threshold_uncertainty_score":0.3430802,"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."}}