Anthropometry–Blood Pressure Mapping in China: Double/Debiased Machine Learning Across 38 Ethnic Groups

Background Hypertension is among the most prevalent cardiometabolic diseases globally, and abnormalities in body weight and composition represent the most modifiable and population-effective intervention targets. Ethnic groups in China exhibit substantial differences in genetic background, dietary patterns, and lifestyle, which may contribute to ethnic heterogeneity in the associations between obesity phenotypes and blood pressure. However, cross-ethnic studies examining body composition–hypertension associations remain limited. This study aimed to quantify the associations between seven body composition indices and prevalent hypertension, and to characterize heterogeneity across 38 ethnic groups in China. Methods Based on cross-sectional survey data from 30,035 adults across 38 ethnic groups in China collected between 2018 and 2025, we employed a double/debiased machine learning framework combined with generalized random forests (DML–GRF) to systematically evaluate the associations between seven body composition indices—body mass index, body roundness index, body fat percentage, body water percentage, visceral fat level, lean body mass, and muscle-to-fat ratio—and hypertension, as well as their ethnic heterogeneity. Results Body mass index, body roundness index, body fat percentage, visceral fat level, and lean body mass were positively associated with hypertension, whereas body water percentage and muscle-to-fat ratio showed inverse associations. Significant ethnic heterogeneity was observed for all seven indices (P ≤ 0.001): the Salar and Hui showed the strongest associations across multiple indices, whereas no index reached significance in the Derung, Monba, and Gelao, where limited stratum size precludes an interpretation of no association. Based on association profiles, the 38 ethnic groups were classified into four patterns: high-sensitivity (13 groups), fat-sensitive (7 groups), muscle-protective (6 groups), and relatively tolerant (12 groups), with statistically significant differences across all indices among the four patterns. Conclusions The strength of associations between body composition indices and hypertension varies substantially across Chinese ethnic groups. These findings provide an evidence base for developing ethnicity-specific hypertension prevention strategies and support a shift from universal body mass index-based screening toward precision public health approaches.