Global external validation of fat mass prediction models for adults

Several models using weight, height and demographic variables to estimate adult fat mass (FM) have shown promising performance but require external validation within new independent settings. Here we evaluated the performance of these models globally, using data from the International Atomic Energy Agency Doubly Labelled Water Database, which includes reference-standard FM information on individuals aged 16+ years spanning 30 countries across all World Health Organization regions. Seven published models predicting adult FM, fat-free mass or FM% using combinations of height, weight (or body mass index), age, sex and/or ethnicity were identified and evaluated. FM from each model was compared with reference-standard FM using R2, calibration and root mean square error within World Health Organization regions and then summarized across regions using meta-analysis. The validation dataset contained 8,228 individuals aged 16–101 years (38% male). Median reference-standard FM was 24.6 kg. Across regions, while models exhibited moderate-to-good R2 values (range 0.69–0.84), most demonstrated miscalibration, with pooled calibration slopes between 0.78 and 1.03, and calibration-in-the-large between 0.59 kg and 3.37 kg. Root mean square error ranged between 4.3 kg and 6.6 kg. Although most models performed well overall, systematic subgroup miscalibration and heterogeneity in predictive performance was observed, suggesting they are not suitable for routine implementation. Future models should ensure consistent high model performance within key subgroups to allow implementation. Using data from 8,228 individuals from 30 countries, this benchmark analysis investigates the performance of fat mass prediction models that use only basic anthropometric features when exposed to previously unseen data.