Retrospective Cohort Analysis of TyG, TyG-SI, and TyG-Lac Indices as Predictors of 360-Day Mortality in Critically Ill Ischemic Stroke Patients.

Background: This study aimed to compare the prognostic value of three surrogate insulin resistance (IR) markers for predicting 360-day mortality in critically ill patients with ischemic stroke (IS): the triglyceride-glucose (TyG) index, TyG-shock index (TyG-SI), and TyG-lactate (TyG-Lac). Methods: The study population comprised critically ill IS patients identified from the Medical Information Mart for Intensive Care (MIMIC) IV database. The main outcome was 360-day mortality. We employed multiple analytical approaches to examine relationships between the three biomarkers and mortality outcomes, including multivariable Cox proportional hazards models (Cox models), Kaplan-Meier survival analysis, and restricted cubic spline (RCS). Furthermore, receiver operating characteristic (ROC) curve analyses were conducted to assess the predictive capacity of these three indices. We performed ROC analyses to evaluate whether the IR index improved the discriminatory ability of a base model that included baseline variables significantly different between survivors and non-survivors. Results: Altogether, 812 patients with IS were included in the analysis. In Cox proportional hazards models, the TyG index was independently associated with higher 360-day mortality (HR, 1.68; 95% CI, 1.52-1.76). Similarly, both TyG-SI and TyG-Lac indices showed significant associations with 360-day mortality, with the HR (95% CI) of 1.24 (1.05-1.38) and 1.11 (1.08-1.23), respectively. Kaplan-Meier survival curves showed a progressive elevation in cumulative 360-day mortality across ascending quartiles of each index (TyG, TyG-SI, and TyG-Lac). ROC curve analysis revealed relatively better discriminatory ability of the TyG-SI compared to TyG and TyG-Lac for all-cause 360-day mortality prediction (area under the curve: 0.605 [0.578-0.623] vs. 0.566 [0.532-0.592] vs. 0.587 [0.532-0.614]). Furthermore, incorporation of either the TyG-SI or TyG index modestly improved the 360-day mortality prognostic accuracy of the base model (area under the curve [AUC], 0.701 for the base model vs. 0.723 for the base model + TyG-SI vs. 0.716 for base model + TyG index). Conclusions: When analyzed as continuous variables, each of the three indices demonstrated significant associations with 360-day mortality risk of critically ill IS populations. Moreover, both TyG-SI and TyG can improve the 360-day mortality predictive accuracy of the base model. Among the three indices, TyG-SI showed comparatively better discriminatory performance; however, the magnitude of AUC improvement was modest, and its clinical utility should be interpreted cautiously pending external validation.