Descriptive statistics for variables.

BackgroundThe high prevalence of cognitive impairment (CI) in Chronic kidney disease (CKD) patients impacts their quality of life and prognosis, yet risk prediction models for CI in this population remain underexplored.ObjectiveThis study aimed to develop a risk prediction model for CI in CKD patients using machine learning algorithms, with the objective of enhancing risk prediction accuracy and facilitating early intervention.MethodsA total of 415 CKD patients from the 2015 China Health and Retirement Longitudinal Survey (CHARLS) dataset were included in this study. Participants were categorized into two groups: the CI group (n = 53) and the non-CI group (n = 362). Binary logistic regression, encompassing both univariate and multivariate analyses, was conducted to identify influencing factors. Subsequently, a CI risk prediction model was constructed using four machine learning algorithms: Support Vector Machine (SVM), Random Forest (RF), Neural Network (NN), and Logistic Regression (LR). The optimal model was further assessed for predictor importance utilizing the SHAP method and deployed on a web platform using the Streamlit library.ResultsLogistic regression analysis identified age, hemoglobin concentration, education level, and social participation as significant factors influencing CI. Models based on NNET, RF, LR, and SVM algorithms were developed, achieving AUC of 0.918, 0.889, 0.872, and 0.760, respectively, on the test set. Calibration curves demonstrated that all models were well-calibrated. Among these, the NNET model exhibited the highest predictive performance. According to the SHAP analysis of the optimal model, the most influential predictors are age, education level, and hemoglobin concentration.ConclusionMachine learning models are valuable tools for predicting the risk of CI in CKD patients and can assist healthcare professionals in developing appropriate intervention strategies.背景 慢性肾脏病（CKD）患者中认知功能障碍（CI）的高患病率会严重影响其生活质量与预后，但目前针对该人群CI的风险预测模型仍有待深入探索。目的 本研究旨在利用机器学习算法构建针对CKD患者CI的风险预测模型，以提升风险预测精度并助力早期干预。方法 本研究纳入了2015年中国健康与养老追踪调查（CHARLS）数据集中共415名CKD患者。将受试者分为两组：CI组（n=53）与非CI组（n=362）。采用二元逻辑回归（包括单因素与多因素分析）筛选影响因素。随后，基于四种机器学习算法构建CI风险预测模型，分别为支持向量机（SVM）、随机森林（RF）、神经网络（NN）与逻辑回归（LR）。进一步采用SHAP方法对最优模型进行预测因子重要性评估，并利用Streamlit库将模型部署至Web平台。结果 逻辑回归分析显示，年龄、血红蛋白浓度、受教育水平与社会参与是影响CI的显著因素。基于NNET、RF、LR及SVM算法构建的模型在测试集上的曲线下面积（AUC）分别为0.918、0.889、0.872与0.760。校准曲线结果表明，所有模型均具备良好的校准度。其中，NNET模型的预测性能最优。通过对最优模型的SHAP分析可知，影响力最高的预测因子为年龄、受教育水平与血红蛋白浓度。结论 机器学习模型是预测CKD患者CI风险的有效工具，可协助医护人员制定合理的干预策略。