Comparative evaluation of obesity-related indices for predicting incident hypertension: evidence from Chinese and UK longitudinal cohorts with machine learning interpretation

Hypertension remains a major global health burden, with excess adiposity serving as a key modifiable contributor to its development. However, conventional anthropometric measures, particularly body mass index (BMI), inadequately reflect metabolically harmful fat accumulation. Consequently, the predictive value of emerging obesity-related indices for incident hypertension remains incompletely defined. We systematically evaluated six obesity-related indices—BMI, Chinese visceral adiposity index (CVAI), triglyceride–glucose index (TyG), TyG–BMI (TGB), TyG–waist-to-height ratio (TGW), and lipid accumulation product (LA)—in relation to new-onset hypertension using data from two prospective cohorts, CHARLS and ELSA. Cox proportional hazards models, restricted cubic spline (RCS) analyses, and interpretable machine-learning methods were applied to assess associations, nonlinear patterns, and relative predictor importance. In both cohorts, all six indices were significantly associated with incident hypertension in univariate analyses, with graded risk increases across quartiles. After mutual adjustment for all indices and covariates, CVAI remained the only predictor consistently associated with hypertension risk in both CHARLS and ELSA. RCS analyses identified nonlinear associations for CVAI, TGW, and LA in CHARLS, whereas relationships in ELSA were largely monotonic. Machine-learning models showed good discrimination (AUC 0.756 in CHARLS; 0.878 in ELSA), and SHAP analysis consistently ranked CVAI, LA, and TGW as the most influential predictors. Overall, CVAI and related composite adiposity indices, particularly LA and TGW, outperform BMI and isolated metabolic markers in predicting incident hypertension. Population-specific nonlinear patterns highlight the heterogeneity of obesity phenotypes and the limitations of BMI-based risk assessment. These indices may provide additional value for risk stratification beyond BMI alone.