Association between SHR and mortality in critically ill patients with CVD: a retrospective analysis and machine learning approach

BACKGROUND: The stress hyperglycemia ratio (SHR), a measure of glucose metabolism, has emerged as a novel indicator of illness severity in critically ill patients. This study aims to evaluate the association between SHR and adverse outcomes in critically ill patients with cardiovascular disease (CVD). METHODS: Clinical data of 1,913 critically ill patients with CVD were extracted from the MIMIC-IV database. The primary outcomes were 360-day, 28-day, and 7-day mortality. Restricted cubic spline (RCS) regression and Cox proportional hazards models were utilized to assess the relationship between SHR and mortality risk in critically ill patients with CVD. Kaplan-Meier survival analysis was conducted to estimate survival rates across SHR quartiles. Additionally, five predictive models were developed using machine learning (ML) algorithms, and the predictive value of SHR was assessed using the SHapley Additive exPlanation (SHAP) algorithm. RESULTS: RCS regression analyses demonstrated a positive linear association between SHR and mortality risk, indicating that higher SHR were linked to an increased risk of adverse outcomes. Kaplan-Meier curves and Cox regression further confirmed that a higher SHR was significantly associated with an elevated risk of mortality in patients with CVD compared to a lower SHR. Predictive machine learning (ML) models were constructed. EXtreme Gradient Boosting (XGBoost) algorithm demonstrated the best performance, with SHR playing a critical role in prediction as identified by SHAP analysis. CONCLUSIONS: SHR is significantly correlated with mortality in critically ill patients with CVD. Based on ML-based predictive models and ROC curves, SHR appears to be a promising indicator in this population.