Machine-Learning-Based Evaluation of the Prognostic Significance of the Non-High-Density Lipoprotein to High-Density Lipoprotein Cholesterol Ratio in Critical Ischemic Stroke.

BACKGROUND This study examined associations of the non-high-density lipoprotein to high-density lipoprotein cholesterol ratio (NHHR) with short-term (28-day) and long-term (365-day) mortality in critically ill patients with ischemic stroke. MATERIAL AND METHODS This retrospective cohort study utilized data from the MIMIC-IV database and focused on critically ill patients with ischemic stroke. Cox proportional hazards, restricted cubic spline, and Kaplan-Meier analyses were performed to examine the relationship between NHHR and mortality. Machine learning models were developed to improve predictive performance; model discrimination and clinical utility were evaluated using time-dependent receiver operating characteristic curves and decision curve analysis. RESULTS Overall, 2492 critically ill patients with ischemic stroke were included. For 28-day mortality, NHHR levels below 2.122 were associated with a 24.9% increase in risk per unit decrease, whereas values above 2.122 conferred a 13.0% increase in risk per unit increment. For 365-day mortality, NHHR levels below 2.111 were associated with a 20.9% increase in mortality risk per unit decrease; values above this threshold were associated with a 14.1% increase in risk per unit increase. Among 6 machine learning algorithms, the random survival forest model demonstrated the best performance, demonstrating superior discrimination, calibration, and clinical utility for predicting both short- and long-term mortality. CONCLUSIONS In critically ill patients with ischemic stroke, NHHR demonstrated a nonlinear and independent association with mortality; the lowest risk was observed at intermediate values. Random survival forest modeling supports NHHR as a robust and clinically meaningful prognostic biomarker in this population.