Association of stress hyperglycemia ratio trajectories and cumulative stress hyperglycemia burden with in-hospital mortality in ICU patients with heart failure: a retrospective cohort study

Heart failure (HF) is a major global health concern, and traditional prognostic markers incompletely reflect acute stress status. Although the stress hyperglycemia ratio (SHR) better assesses acute glycemic burden and correlates with adverse outcomes in critically ill HF patients, previous studies focused on single-time-point measurements, ignoring its dynamic trajectories and cumulative stress hyperglycemia ratio (CumSHR) burden. This study thus investigated their prognostic significance in intensive care unit (ICU) patients with HF using the Medical Information Mart for Intensive Care (MIMIC) database. This retrospective cohort study used data from the MIMIC-IV database, with external validation performed in the MIMIC-III database. Adult ICU patients diagnosed with HF were included. Latent class growth modeling (LCGM) was applied to identify SHR trajectories during the first 6 days after ICU admission. CumSHR was calculated using the trapezoidal rule to reflect cumulative glycemic burden. Multivariable regression, Kaplan–Meier analysis, and restricted cubic spline (RCS) models were used to evaluate the association between SHR patterns and mortality outcomes. Machine learning models were further developed using features selected by Least Absolute Shrinkage and Selection Operator (LASSO) and Boruta to predict in-hospital mortality. A total of 1,675 HF patients were included. Three distinct SHR trajectories were identified: a low-decreasing group, a moderate-decreasing group, and a persistently high-increasing group. Compared with the low-decreasing group, patients with persistently elevated SHR had significantly higher risks of in-hospital, 28-day, 90-day, and 360-day ICU mortality. CumSHR showed a significant linear association with mortality risk and demonstrated superior discriminative performance compared with single-time-point glucose indicators. Machine learning models showed superior predictive performance for in-hospital mortality; Shapley Additive Explanations (SHAP) analysis based on Extreme Gradient Boosting (XGBoost) identified CumSHR as the most important predictor. Dynamic SHR trajectories and cumulative SHR were associated with short- and long-term mortality among ICU patients with heart failure. A persistently increasing SHR trajectory was associated with the highest mortality risk. These findings suggest that CumSHR may offer complementary information for risk stratification in critically ill patients with heart failure; however, further prospective studies are needed to confirm its clinical utility.