Age-related variation in hemoglobin glycation index and stroke mortality: mediation and machine learning in a cohort study.

To investigate the associations between both age and the hemoglobin glycation index (HGI) and the 30-day and 1-year mortality in ischemic stroke (IS) patients and to analyze the mediating effect of the HGI on the relationship between age and mortality. A total of 3269 hospitalized patients with IS included in the Medical Information Mart for Intensive Care (MIMIC)-IV database were included in this study. The effects of age and HGI on short- (30 days) and long-term (1 year) mortality were analyzed with logistic, Cox, and least absolute shrinkage and selection operator (LASSO) regression analysis. The nonlinear relationship among the variables was further investigated via restriction cubic spline (RCS) analysis, and the mediating effects of HGI on the age-mortality relationship were confirmed via mediation analysis. Kaplan-Meier (K-M) survival curves and restricted mean survival time (RMST) analyses were used to evaluate the differences in survival among patients with different HGI levels. Finally, multiple machine learning (ML) models were constructed and subsequently evaluated in terms of predictive performance. Logistic and Cox regression analyses revealed that a lower HGI and a greater age were significantly associated with higher risks of 30-day and 1-year mortality (both P < 0.001). RCS analysis revealed a J-shaped relationship between HGI and mortality risk. Mediation analysis revealed that HGI had a negative mediating effect on the relationship between age and mortality. K-M curve and RMST analyses further revealed that patients with higher HGIs had greater probabilities of survival. ML models also confirmed the importance of HGI in predicting the risk of mortality. Age and HGI are correlated with both the 30-day and 1-year risks of mortality in IS patients. The HGI may play a partial mediating role between age and the risk of mortality.