Risk prediction model for acute stroke in elderly population in cold regions based on machine learning

Cerebrovascular disease (CVD) profoundly affects the health and quality of life of elderly individuals, with stroke being its most prevalent manifestation, characterized by high incidence, disability, and mortality rates. In Northeast China, harsh winter conditions further exacerbate the risk of stroke. This study leverages machine learning (ML) to develop a predictive model for assessing acute stroke risk among the elderly in cold regions, aiming to enable early identification of high-risk individuals.Significant risk factors for acute stroke onset were identified through univariate and multivariate logistic regression (LR) analyses. Input variables were selected based on expert clinical recommendations and a comprehensive literature review. Three machine learning algorithms—LR, extreme gradient boosting (XGBoost), and random forest (RF)—were compared for their clinical prediction performance, with the best-performing algorithm used to construct the risk prediction model. A line chart was employed to visualize the probability of acute stroke in this population.The findings highlight diabetes history, coronary heart disease(CHD), systolic blood pressure(SBP), neutrophil count(NEUT#), total bile acid (TBA), fasting blood glucose (FBG), and homocysteine(Hcy) as independent risk factors for acute stroke in elderly individuals in cold regions. In contrast, prealbumin(PA), albumin (ALB), and high-density lipoprotein cholesterol (HDL-C) emerged as potential protective factors. The LR-based model developed in this study demonstrated robust performance in predicting acute stroke risk, providing accurate and individualized risk assessments for the target population.