预测中国绝经后女性新发心脏代谢性共病风险的可解释机器学习模型研究OA

Background Cardiovascular diseases are prevalent in China,with cardiometabolic multimorbidity(CMM)being a common comorbidity pattern.Postmenopausal women represent a high-risk group for cardiovascular diseases,yet there is a lack of predictive models for CMM risk specifically in this population.Objective To develop an interpretable machine learning(ML)model to predict the risk of CMM among Chinese postmenopausal women,based on data from the China Health and Retirement Longitudinal Study(CHARLS).Methods The study included postmenopausal women aged≥45 years from the CHARLS cohort in 2011 who were free of CMM at baseline.Data on demographic characteristics,family background,health status,and laboratory indicators were collected at baseline and during follow-up in 2013,2015,2018,and 2020 to observe CMM incidence.Feature selection was performed using the least absolute shrinkage and selection operator(LASSO)algorithm.Seven ML algorithms were constructed for risk prediction.The optimal model was further optimized on the test set using a combined strategy of"class_weight='balanced'dynamic weighting+optimal threshold selection"and visually interpreted using Shapley Additive Explanations(SHAP).Model performance was evaluated using the area under the receiver operating characteristic curve(AUC),sensitivity,specificity,precision,and F1-score.Results A total of 5 575 participants completed the 4 rounds of follow-up and were included,comprising 4 363 in the non-CMM group and 1 212 in the CMM group.Over a median follow-up of 9 years,the cumulative incidence of CMM was 21.74%.LASSO regression identified 22 key features as significant predictors of CMM:self-rated health,mental disorders,arthritis,dyslipidemia,kidney disease,retirement status,systolic blood pressure(SBP),diastolic blood pressure(DBP),mean pulse rate,waist circumference,BMI,headache,lower back pain,serum creatinine(Scr),triglycerides(TG),C-reactive protein(CRP),glycated hemoglobin(HbA1c),uric acid(UA),age,Center for Epidemiologic Studies Depression Scale(CES-D)score,smoking status,and geographic region.Among the models,the Logistic regression(LR)model demonstrated the best predictive performance(test set AUC=0.758,accuracy=79.2%).The SHAP mean bar plot revealed core predictors:SBP,HbA1c,geographic region,waist circumference,CES-D score,BMI,DBP,and age.The SHAP summary plot indicated that higher values of SBP,HbA1c,waist circumference,and others were associated with increased predicted CMM risk.Conclusion This study develops a clinically interpretable prediction model for CMM in Chinese postmenopausal women,with the LR algorithm showing favorable performance.Key risk factors include SBP,HbA1c,and waist circumference.The model provides an evidence-based tool for screening high-risk individuals and guiding personalized interventions.