Objective To predict the secondary ischemic stroke in maintenance hemodialysis (MHD) patients using different machine learning algorithms, and analyze its related factors.

Methods A retrospective case-control study was conducted. 151 patients with MHD were selected as the research subjects. According to whether the patients had ischemic stroke, they were divided into the ischemic stroke group and non-ischemic stroke group. Univariate and logistic regression analyses were used to analyze the risk factors. Based on four machine learning models, namely Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN) and Logistic Regression (LR), a prediction model for secondary ischemic stroke in MHD patients was constructed. The robustness and practicability of the model were evaluated by using accuracy rate, sensitivity, specificity, F1 and area under the ROC curve (AUC).

Results The ischemic stroke occurred in 23 cases (15.2%). The results of multivariate logistic regression analysis showed that the age (OR = 1.076, 95%CI: 1.004–1.153) and years of hemodialysis (OR = 1.213, 95%CI: 1.008–1.458), history of atrial fibrillation (OR = 4.016, 95%CI: 1.664–37.994), smoking history (OR = 12.628, 95%CI: 2.015–79.142), uric acid (OR = 1.104, 95%CI: 1.037–1.175) and serum albumin (OR = 0.781, 95%CI: 0.643–0.947) were the independent influencing factors of secondary ischemic stroke in MHD patients. The ROC curve analysis showed that among the four models, the RF model had the best predictive effects on secondary ischemic stroke in MHD patients (AUC = 0.932). Among the MHD patient prediction models established by the four algorithms, the RF model had the best comprehensive prediction efficiency, with the highest accuracy rate, sensitivity, specificity and F1 score. Among the top 6 characteristics of the RF model, the proportion of uric acid was the largest (17.40%), followed by serum albumin (12.70%), age (10.70%), years of hemodialysis (9.90%), smoking history (9.30%) and history of atrial fibrillation (6.60%).

Conclusions The uric acid, serum albumin, age, years of hemodialysis, smoking history and history of atrial fibrillation are the independent influencing factors of secondary ischemic stroke in MHD patients. Machine learning models can serve as reliable tools for predicting secondary ischemic stroke in MHD patients. The RF model has the best predictive performance, which helps clinical healthcare professionals identify high-risk patients and implement early interventions to reduce incidence.