Abstract
The hemoglobin glycation index (HGI) is a promising marker for assessing glycemic control and outcomes in critically ill patients, but its prognostic value in Trauma/Surgical Intensive Care Units (TSICU/SICU) remains unclear. This study investigates the predictive value of HGI in relation to mortality. This retrospective analysis used data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, a publicly available critical care database, focusing on TSICU/SICU patients. HGI was calculated as the difference between observed and predicted HbA1c levels. Associations between HGI quartiles and 28-day and 360-day mortality were assessed using Kaplan-Meier survival analysis, multivariate Cox regression, and restricted cubic spline (RCS) models. A stacked ensemble machine learning model validated HGI's predictive power, and mediation analysis evaluated sodium as a mediator. Kaplan-Meier analysis showed significant survival differences across HGI quartiles (log-rank p < 0.001), with the lowest quartile demonstrating the worst outcomes. Cox regression revealed that higher HGI was independently associated with lower 28-day and 360-day mortality (HR 0.76, 95% CI 0.72-0.81, p < 0.001). ROC analysis confirmed HGI outperformed HbA1c and glucose in predictive performance. The stacked ensemble model achieved an AUC of 0.85, highlighting HGI's predictive strength. Mediation analysis found sodium mediated only 3.1% of HGI's total effect on 28-day mortality, suggesting direct mechanisms. HGI is a robust predictor of short- and long-term mortality in TSICU/SICU patients, surpassing traditional glycemic markers. Incorporating HGI into risk stratification models could improve patient management, and further studies are needed to validate its clinical utility and explore underlying mechanisms.