AUTHOR=Xu Wenwen , Li Jianning , Huang Chunfang , Lu Yi , Yin Liping , Zhang Guoxin 

TITLE=A comparative study of risk factors in predictive models for cognitive dysfunction in patients with leukoaraiosis based on machine learning algorithms

JOURNAL=Frontiers in Neuroscience

VOLUME=Volume 19 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/neuroscience/articles/10.3389/fnins.2025.1654584

DOI=10.3389/fnins.2025.1654584

ISSN=1662-453X

ABSTRACT=ObjectiveTo explore the risk factors of cognitive dysfunction in patients with leukoaraiosis (LA) and to construct a predictive model using machine learning.MethodsA total of 273 patients with LA were included. Univariate analysis and multivariate logistic regression were performed to identify independent risk factors for cognitive dysfunction. The patients were divided into a training set (191 cases) and a validation set (82 cases) in a 7:3 ratio. Seven machine learning algorithms (Decision Tree, GBDT, Logistic Regression, Random Forest, SVM, KNN, XGBoost) were employed to construct predictive models. Evaluation metrics included accuracy, recall, F1 score, MCC, AUROC, and SHAP was used for model interpretation.ResultsUnivariate analysis revealed that age, LDL-C, uric acid, CRP, Fazekas score, and IASA score were associated with cognitive dysfunction (p < 0.05). Multivariate logistic regression analysis showed that age, LDL-C, uric acid, Fazekas score, and IASA score were independent risk factors (p < 0.05). Among the machine learning algorithms, the Random Forest model performed the best, with an AUROC of 0.8373 for the validation set. SHAP analysis indicated that age, LDL-C, IASA score, and Fazekas score were the most important predictors.ConclusionThe Random Forest model can be used to predict the risk of cognitive dysfunction in patients with LA, providing a reference for early warning.