{"id":"W4412062841","doi":"10.14701/ahbps.25-046","title":"A comparative study of machine learning models predicting post-hepatectomy liver failure: Enhancing risk estimation in over 25,000 National Surgical Quality Improvement Program patients","year":2025,"lang":"en","type":"article","venue":"Annals of Hepato-Biliary-Pancreatic Surgery","topic":"Hepatocellular Carcinoma Treatment and Prognosis","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto Metropolitan University; Western University","funders":"Western University","keywords":"Hepatectomy; Estimation; Quality (philosophy); Medicine; Computer science; Reliability engineering; Artificial intelligence; Machine learning; Surgery; Engineering; Systems engineering","routes":{"ca_aff":true,"ca_fund":true,"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":[0.01057583,0.0008177228,0.0007873271,0.001359328,0.0002965522,0.001006703,0.0006682953,0.0006935387,0.0009211804],"category_scores_gemma":[0.02498067,0.0002401009,0.001178307,0.0008578983,0.0003400226,0.001271798,0.0008752117,0.0008788137,0.0001829848],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001006599,"about_ca_system_score_gemma":0.0007115771,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005550249,"about_ca_topic_score_gemma":0.002958836,"domain_scores_codex":[0.9976358,0.001260275,0.0002185993,0.0004142178,0.0003024078,0.0001685927],"domain_scores_gemma":[0.9814466,0.01370149,0.001814132,0.0007553505,0.001520155,0.0007622271],"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":[0.002382722,0.0007568063,0.9590108,0.00006907765,0.0005989351,0.00007172678,0.0001418912,0.01440005,0.0001149852,0.0001254709,0.0007626416,0.02156479],"study_design_scores_gemma":[0.0002809593,0.005398801,0.5398576,0.00007798838,0.0005173621,0.0002879256,0.0007575571,0.4512522,0.0004601813,0.000545503,0.0005217965,0.00004216806],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9978561,0.0002329021,0.001076112,0.0002196794,0.00001459966,0.00003062939,0.0003207181,0.00003498537,0.0002141784],"genre_scores_gemma":[0.9982121,0.00009221374,0.0008697712,0.00003510348,0.00001836472,0.00001958487,0.0006730802,0.000004558523,0.0000752488],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01057583,"threshold_uncertainty_score":0.05593103,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1061802325981699,"score_gpt":0.3368145230559091,"score_spread":0.2306342904577393,"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."}}