Long-term body roundness index profiles and incident diabetes in middle-aged and older Chinese adults: a longitudinal analysis of CHARLS 2011–2020

Abstract

Higher body roundness index (BRI) is associated with diabetes, but trajectory classes derived from a small number of repeated measurements may primarily reflect average BRI level. We examined whether sustained BRI profiles were associated with incident diabetes and, secondarily, whether trajectory classification added information beyond baseline and mean repeated BRI. This longitudinal analysis used China Health and Retirement Longitudinal Study data from 2011 to 2020. BRI was calculated from measured waist circumference and height in 2011, 2013, and 2015. Adults aged 45 years or older who remained free of diabetes throughout 2011–2015 and had sufficient information for strict 2018–2020 outcome classification were included. One- to five-group linear and quadratic group-based trajectory models were compared; a four-group linear model was selected as the primary parsimonious representation. Community-cluster robust logistic regression was the primary association analysis. Classification-uncertainty analyses, inverse-probability-of-observation weighting (IPOW), multiple imputation, sex-standardized and sex-specific trajectories, supplementary Cox models, and five-fold cross-validated exposure comparisons were performed. Among 4,573 participants, 388 (8.48%) developed diabetes. Four predominantly stable profiles were identified: low-stable ( n = 999; 21.8%), moderate-low stable ( n = 1,646; 36.0%), moderate-stable ( n = 1,318; 28.8%), and high-stable ( n = 610; 13.3%). Diabetes incidence increased from 5.0% to 12.6% across these profiles. Compared with the low-stable profile, primary adjusted odds ratios were 1.76 (95% confidence interval [CI] 1.23–2.52), 2.29 (1.59–3.30), and 3.12 (2.08–4.67) across progressively higher profiles. Results were consistent across alternative trajectory specifications, posterior-probability and pseudo-class analyses, IPOW, multiple imputation, supplementary Cox models, and sex-specific analyses. Five-fold cross-validated areas under the receiver operating characteristic curve were 0.587 (95% CI 0.558–0.616) for trajectory classification and 0.602 (0.573–0.632) for mean repeated BRI; the paired difference was − 0.015 (95% CI − 0.026 to − 0.005). Sustained higher BRI profiles were associated with progressively higher odds of incident diabetes. The profiles were separated mainly by long-term BRI level rather than divergent change, and trajectory classification did not improve internal discrimination beyond mean repeated BRI. Repeated monitoring of abdominal adiposity may support risk stratification, but these classes should not be treated as validated clinical prediction categories.