首页|期刊导航|沈阳医学院学报|基于机器学习算法的老年腹膜透析患者心血管事件发生风险预测模型构建

基于机器学习算法的老年腹膜透析患者心血管事件发生风险预测模型构建OA

Construction of a machine learning-based risk prediction model for cardiovascular events in elderly patients undergoing peritoneal dialysis

中文摘要英文摘要

目的:分析老年腹膜透析患者心血管事件(cardiovascular event,CVE)的独立危险因素并构建临床风险预测模型,为高危患者的早期识别及个体化防治提供量化依据.方法:选择2024年1月至2025年6月我院收治的行腹膜透析治疗的老年患者260例为研究对象,根据随访期间是否发生CVE分为CVE组(n=67)和非CVE组(n=193).收集患者人口学特征、临床并发症及实验室检查指标,通过单因素Logistic回归筛选潜在危险因素,经Boruta算法优化核心预测变量后,采用多因素Logistic回归分析构建预测模型.通过受试者工作特征(receiver operating characteristic,ROC)曲线、校准曲线及决策曲线分析评估模型效能,采用Bootstrap法(1 000次重复抽样)进行内部验证,结合列线图实现模型可视化,利用主成分分析(principal component analysis,PCA)及箱线图验证风险分组效果.结果:单因素Logistic回归分析结果显示,年龄、C反应蛋白、白蛋白、血红蛋白、残肾肾小球滤过率、总尿素清除指数与老年腹膜透析患者CVE发生密切相关(P<0.05);Boruta算法最终筛选出白蛋白、C反应蛋白、残肾肾小球滤过率、总尿素清除指数4项核心预测变量.多因素Logistic回归分析结果显示,C反应蛋白为CVE发生的危险因素(OR=1.203,95%CI:1.117~1.296,P<0.01),白蛋白(OR=0.776,95%CI:0.708~0.850,P<0.01)、残肾肾小球滤过率(OR=0.710,95%CI:0.573~0.881,P=0.001)、总尿素清除指数(OR=0.219,95%CI:0.099~0.486,P<0.01)为保护性因素.模型ROC曲线下面积为0.829(95%CI:0.776~0.881),当模型预测概率的最佳截断值为0.250时,敏感度为77.6%,特异度为79.2%;校准曲线显示预测概率与实际发生概率高度一致(平均绝对误差0.019);决策曲线分析表明,在风险阈值0.10~0.85范围内,模型净获益高于"全部干预"或"全部不干预"策略.根据列线图评分中位数分组,高风险组CVE发生率显著高于低风险组(P<0.01),PCA分析显示2组在特征空间中呈明显分离.结论:基于白蛋白、C反应蛋白、残肾肾小球滤过率、总尿素清除指数构建的预测模型可有效识别老年腹膜透析患者CVE发生风险,经内部验证显示预测效能良好.

Objective:To analyze the independent risk factors for cardiovascular events(CVE)in elderly peritoneal dialysis patients and to construct a clinical risk prediction model,thereby providing a quantitative basis for the early identification and individualized prevention and treatment of high-risk patients.Methods:A total of 260 elderly patients who received peritoneal dialysis in our hospital from Jan 2024 to Jun 2025 were selected as the study subjects.They were divided into a CVE group(n=67)and a non-CVE group(n=193)according to whether CVE occurred during the follow-up period.Demographic characteristics,clinical comorbidities,and laboratory parameters were collected.Potential risk factors were identified using univariate logistic regression analysis.After optimizing the core predictive variables using the Boruta algorithm,a prediction model was constructed by multivariate logistic regression analysis.The performance of the model was evaluated by receiver operating characteristic(ROC)curve,calibration curve,and decision curve analysis.Internal validation was performed using the Bootstrap method(1 000 repeated samplings).The model was visualized using a nomogram,and the risk grouping effect was verified by principal component analysis(PCA)and box plots.Results:Univariate logistic regression analysis revealed that age,C-reactive protein(CRP),albumin,hemoglobin,residual renal glomerular filtration rate(rGFR),and total Kt/V were significantly associated with the occurrence of CVE in elderly peritoneal dialysis patients(P<0.05).The Boruta algorithm ultimately identified four core predictive variables:albumin,CRP,rGFR,and total Kt/V.Multivariate logistic regression analysis demonstrated that CRP was a risk factor for CVE(OR=1.203,95%CI:1.117-1.296,P<0.01),whereas albumin(OR=0.776,95%CI:0.708-0.850,P<0.01),rGFR(OR=0.710,95%CI:0.573-0.881,P=0.001),and total Kt/V(OR=0.219,95%CI:0.099-0.486,P<0.01)were protective factors.The area under the ROC curve(AUC)of the model was 0.829(95%CI:0.776-0.881).When the optimal cut-off value of the model prediction probability was 0.250,the sensitivity was 77.6%and the specificity was 79.2%.The calibration curve showed that the predicted probability was highly consistent with the actual occurrence probability(mean absolute error 0.019).Decision curve analysis indicated that within the risk threshold range of 0.10-0.85,the net benefit of the model exceeded that of the"full intervention"or"full non-intervention"strategies.According to the median score of the nomogram,the incidence of CVE in the high-risk group was significantly higher than that in the low-risk group(P<0.01),and PCA analysis showed that the two groups were clearly separated in the feature space.Conclusion:The predictive model constructed based on albumin,CRP,rGFR,and total Kt/V can effectively identify the risk of CVE in elderly peritoneal dialysis patients,and internal validation shows good predictive performance.

韩亚男;齐甜甜;丁叶丹

江西省乐平市人民医院,江西 景德镇 333300江西省乐平市人民医院,江西 景德镇 333300江西省乐平市人民医院,江西 景德镇 333300

医药卫生

老年患者腹膜透析心血管事件机器学习风险预测模型Boruta算法

elderly patientsperitoneal dialysiscardiovascular eventsmachine learningrisk prediction modelBoruta algorithm

《沈阳医学院学报》 2026 (5)

490-497,8

景德镇市科技计划项目(No.20252SFZC033)

10.16753/j.cnki.1008-2344.2026.05.006

评论