The global obesity epidemic necessitates improved adiposity assessment alternatives beyond conventional anthropometric indices such as body mass index (BMI), particularly given their limitations in characterizing fat distribution. This study aimed to (1) develop adiposity assessment equations based on digital anthropometry and (2) systematically evaluate our equations against conventional anthropometric indices. We analyzed 1294 participants from the International Human Phenome Project, leveraging 151 digital anthropometric phenotypes and eight DXA-derived adiposity relative indices capturing adiposity from multiple distinct aspects. After training and comparing seven machine learning algorithms, the final equations were subjected to variance inflation factor (VIF) analysis to optimize predictor selection. A systematic evaluation was conducted, comparing the newly developed equations with both BMI and conventional alternative indices via Pearson correlation analysis. The forward stepwise linear regression (FSLR) model achieved superior or comparable accuracy (R² = 0. 767 - 0. 910) to the other six machine learning algorithms. Eight ethnicity-specific equations were derived via FSLR and VIF selection, incorporating 10. 25 predictors on average. Systematic evaluation demonstrated that our equations outperformed BMI and conventional alternative indices by 2. 3–63. 0% across key adiposity metrics. Moreover, the results highlight the complementary value of combining BMI with alternative measures for comprehensive adiposity assessment. We developed adiposity assessment equations in the Chinese population using digital anthropometry, which significantly improved upon BMI-based evaluations. Furthermore, our findings provide concrete evidence for combining BMI with supplementary indices, as recently recommended by The Lancet Diabetes & Endocrinology Commission.
Zhao et al. (Wed,) studied this question.