Construction of machine learning-based risk prediction model for coronary heart disease in elderly obstructive sleep apnea patients

GAO Yinghui; CAI Weimeng; RUI Dong; ZHAO Libo; ZHAO Zhe; MA Yao; LIU Lin

doi:10.3969/j.issn.1009-0126.2025.12.011

Abstract：Objective To investigate the application value of machine learning algorithms in predicting CHD risk among elderly patients with obstructive sleep apnea(OSA).Methods A total of 1 129 OSA patients aged ≥60 years who completed polysomnography at 6 tertiary hospitals(Chinese PLA General Hospital,Peking University International Hospital,People's Hospital of Peking University,Beijing Chaoyang Hospital,No.960 Hospital of PLA,and Affiliated Hospital of Gansu University of Chinese Medicine)between January 2015 and October 2017.According to the occurrence of CHD,the participants were assigned into a CHD group(202 cases)and a non-CHD group(927 cases).Baseline data were collected,and they were divided into training(n=678),validation(n=226),and testing sets(n=225)in a 6∶2∶2 ratio.Four machine learning algorithms were used to construct predictive models,that is,logistic regression,random forest,extreme gradient boosting,and gradient boosting machine.ROC curves were plotted to evaluate the predictive performance of the four models by calculating the area under the receiver operating characteristic curve(AUC).Results The CHD group exhibited significantly older age,larger waist circumference,higher SBP and FBG levels,and higher ratios of male,hyperlipidemia,diabetes,alcohol consumption and carotid atherosclerosis than the non-CHD group(P＜0.05,P＜0.01),and statistical differences were seen in body mass index and apnea-hypopnea index between the two groups(P＜0.05).In the testing set,the random forest model demonstrated best performance with an AUC value of 0.815(95％CI:0.756-0.874),which was higher than that of logistic regression model(AUC=0.746,95％CI:0.682-0.810).The accuracy was 73.9％,the sensitivity was 84.5％,and the specificity was 69.9％in the validation set for the random forest model.Feature importance analysis revealed that diabetes,age,SBP,oxygen desaturation index and blood glucose were the top five important predictive factors.Conclusion The machine learning-based CHD prediction model demonstrates good predictive performance in elderly OSA patients.This model can be used as an effective tool for cardiovascular risk stratification and clinical decision-making for elderly OSA patients.