An electronic health record-derived glycemic variability-enhanced triglyceride-glucose index for early mortality risk stratification in critically ill patients with acute myocardial infarction: a derivation and external validation study

Electronic health records (EHRs) contain repeated glucose measurements routinely collected during critical care, yet these data are rarely incorporated into early risk-stratification tools for acute myocardial infarction (AMI). The conventional triglyceride-glucose (TyG) index relies on a single glucose value and may not capture glycemic instability during acute illness. We developed an EHR-derived triglyceride-glucose lability index (TyGLI) and evaluated its prognostic value for 28-day mortality in critically ill patients with AMI. This was a retrospective derivation and external validation study conducted using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and the First Affiliated Hospital of Wenzhou Medical University (WMU). The primary outcome was all-cause mortality within 28 days after ICU/CCU admission. TyGLI was calculated from triglyceride levels and serial glucose measurements obtained within 72 h of ICU/CCU admission. Predictor selection was performed using LASSO regression, and multivariable logistic regression assessed independent associations. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), precision–recall AUC, calibration, Brier score, decision curve analysis, net reclassification improvement, and integrated discrimination improvement. Coefficients derived in MIMIC-IV were directly applied to the WMU cohort. Among 814 patients, 81 (10.0%) died within 28 days, including 50 of 343 patients (14.6%) in the MIMIC-IV cohort and 31 of 471 patients (6.6%) in the WMU cohort. TyGLI was significantly higher in non-survivors than survivors [7.04 (5.55–8.77) vs. 5.09 (3.94–6.60); P < 0.001]. Each unit increase in TyGLI was independently associated with mortality (OR 1.42, 95% CI 1.17–1.72; P < 0.001). Adding TyGLI improved discrimination in both cohorts (ΔAUC + 0.073, P = 0.039 in MIMIC-IV; ΔAUC + 0.115, P = 0.005 in WMU), whereas conventional TyG did not. TyGLI also enhanced risk reclassification and clinical net benefit. TyGLI was independently associated with 28-day mortality and improved risk stratification beyond conventional TyG and standard clinical predictors. Prospective studies are warranted to confirm its clinical utility.