{"id":"W4406090644","doi":"10.1186/s12911-024-02830-7","title":"External validation of AI-based scoring systems in the ICU: a systematic review and meta-analysis","year":2025,"lang":"en","type":"review","venue":"BMC Medical Informatics and Decision Making","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"Charité – Universitätsmedizin Berlin; European Commission; Alexander von Humboldt-Stiftung","keywords":"Overfitting; Medicine; Reliability (semiconductor); MEDLINE; Intensive care unit; Health informatics; External validity; Receiver operating characteristic; Intensive care; Computer science; Medical physics; Data mining; Machine learning; Intensive care medicine; Statistics; Public health; Pathology","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":[0.003932359,0.0003262778,0.006156084,0.0006587568,0.00005129108,0.0001089486,0.0001830787,0.0002032895,0.00006204355],"category_scores_gemma":[0.001797481,0.0001460916,0.001109528,0.0009393549,0.00004288915,0.0000670926,0.0000962817,0.0002926401,0.000003492109],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004977187,"about_ca_system_score_gemma":0.0003504447,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001177216,"about_ca_topic_score_gemma":0.000007355712,"domain_scores_codex":[0.9951625,0.0003589087,0.002869114,0.0001878012,0.001263493,0.0001581704],"domain_scores_gemma":[0.9910392,0.00708137,0.001063143,0.0005929326,0.0001153355,0.0001080911],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.000003103679,0.00006212579,0.00003371145,0.9642144,0.007548095,0.00001531615,0.00004016668,0.000016908,2.67345e-10,0.0001635683,0.00004801636,0.02785454],"study_design_scores_gemma":[0.0002608925,0.0000381754,0.000004638069,0.4966725,0.496807,0.00008533766,0.00007313422,0.005381713,4.255338e-8,0.00002783791,0.0005458028,0.0001029622],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00001444015,0.9448841,0.05248493,0.00002164792,0.00003405306,0.002467202,0.00001619554,0.000006812602,0.00007067357],"genre_scores_gemma":[0.0003919408,0.992837,0.005117558,0.001049286,0.00001200767,0.000548864,0.00002838648,0.000008594228,0.000006370351],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4892589,"threshold_uncertainty_score":0.5957443,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2735414778019841,"score_gpt":0.4761514850411881,"score_spread":0.202610007239204,"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."}}