2972-LB: Proteomics-Based Molecular Adiposity and Risk of Type 2 Diabetes in Diverse Population

Introduction and Objective: Adiposity reflects heterogeneous biological processes that are not fully captured by body mass index (BMI). Proteomic profiling may better distinguish metabolic health. We aimed to derive a proteomics-based BMI (ProtBMI) using machine learning and assess its relevance for type 2 diabetes (T2D) risk in UK and Chinese adults. Methods: ProtBMI was developed in UK Biobank (n=45,226) using 2,923 Olink proteins and a Light Gradient Boosting Machine (LightGBM) model. Boruta feature selection and recursive feature elimination identified a 30-protein signature predicting BMI (R²=0.78), which was externally validated in China Kadoorie Biobank (n=3,997; R² = 0.57). Association of ProtBMI and BMI with metabolic traits were assessed using linear regression, and with incident T2D using Cox regression adjusted for confounders. Risk prediction models were evaluated using the C-statistic, and net reclassification index (NRI). Results: Among UKB participants, ProtBMI (mean 27.4 [SD 4.2] kg/m²) showed stronger associations than BMI (mean 27.4 [4.8] kg/m²) with glycaemic, lipid, and inflammatory biomarkers. ProtBMI was more strongly associated with incident T2D (1,054 events) than BMI (HR per SD: 2.07 [95% CI 1.95-2.18] vs. 1.83 [1.73-1.93]). T2D risk increased across ProtBMI categories within BMI strata, with up to a ~9-fold difference among individuals with BMI 25-30 kg/m² and ~2-fold differences among those with BMI ≥30 kg/m². Within the normal BMI range (22.5-24.9 kg/m²), ProtBMI ≥25 was associated with 2.5-fold higher T2D risk compared with ProtBMI <25. Adding ProtBMI to a conventional risk prediction model improved the C-index from 0.821 (0.808-0.834) to 0.833 (0.820-0.845) and NRI by 34.8%. Findings were directionally consistent in CKB (mean BMI 23.9 [SD 3.6] kg/m²). Conclusion: ProtBMI captures metabolic variation beyond BMI, identifies individuals at higher T2D risk within BMI strata, including within the “normal” range, and modestly improves T2D risk prediction across diverse populations. Disclosure X. Yan: None. A. Iona: None. P. Yao: None. Z. Chen: None.