Comparative performance of body roundness index and traditional obesity indices in predicting cardiovascular risk: Machine learning insights from three prospective aging cohorts

Yinghuan Zhang^1,2Yuxuan Wang^1,2Shan Qiao³Xue Yang⁴Meihui Zhang²Chen Xu²Ying Wang²Fan Hu^2*Yong Cai^2,5*

• ¹Shanghai Jiao Tong University School of Public Health, Shanghai, China
• ²Tongren Hospital of Shanghai Jiao Tong University School of Medicine, Shanghai, China
• ³University of South Carolina Arnold School of Public Health, Columbia, United States
• ⁴The Chinese University of Hong Kong The Jockey Club School of Public Health and Primary Care, Hong Kong, Hong Kong, SAR China
• ⁵Shanghai Jiao Tong University China Hospital Development Institute, Shanghai, China

Objective: The burden of cardiovascular diseases (CVD) is significant, necessitating early prevention, with obesity standing out as a pivotal modifiable risk factor. We aimed to use three prospective aging cohorts to develop an obesity-focused prediction model for incident CVD risk with enhanced validation and explanation. Methods: We analyzed longitudinal data from the China Health and Retirement Longitudinal Study (CHARLS) wave 1-4, Health and Retirement Study (HRS) wave 11-14, and English Longitudinal Study of Ageing (ELSA) wave 6-9. All participants were aged 45 years or older, had no CVD at baseline, and completed follow-up assessments across three subsequent waves. The main outcome was the occurrence of CVD (self-reported physician diagnoses of either heart disease or stroke). The predictors were screened by the Least Absolute Shrinkage and Selection Operator and Random Survival Forest. A multivariate Cox regression analysis was applied to develop the prediction model. Model performance was validated using: (1) concordance index for discrimination, (2) calibration curves for risk accuracy, and (3) time-dependent Receiver Operating Characteristic curves for classification. The time-dependent feature importance plot, partial dependence survival profiles and SHapley Additive exPlanations plot were used to interpret the model. Results: The study included 5768 participants from CHARLS, 3151 from HRS and 3016 from ELSA. The CVD incidence rates of CHARLS, HRS and ELSA were 21.2%, 13.2% and 13.5% respectively. Three of the seventeen screened covariates, which were age, hypertension, systolic blood pressure (SBP), as well as body mass index (BMI) and body roundness index (BRI), were included in the prediction model. The model exhibited a valid predictive value and moderate performance, with obesity showing a pronounced effect. BRI demonstrated stronger associations with CVD than BMI in both training and validation cohorts. Conclusion: Age, hypertension, SBP, BMI, and BRI were significant predictors of incident CVD in middle-aged and older adults, highlighting the impact of obesity on CVD risk, and consequently offered a valuable model for public health strategies to prevent CVD.