A Clinical Prediction Model Integrated With Machine Learning Algorithms Uncovers the Associations Between Dietary Omega-3, Its Components, and Aging Biomarkers

This study aimed to explore the associations between dietary omega-3 intake (including alpha-linolenic acid [ALA], eicosapentaenoic acid [EPA], and docosahexaenoic acid [DHA]) and aging biomarkers (specifically DNA methylation age [HorvathAge], telomere length [Horvathtelo], and phenotypic age) in older adults by constructing clinical prediction models through cross-sectional, machine learning (ML), and Mendelian randomization (MR) analyses. Linear regression models, supplemented by restricted cubic splines (RCS) for nonlinearity detection, were established using dietary omega-3 intake (excluding supplements) from participants in the 1999-2002 NHANES. Causal associations were further validated via MR analysis using the inverse-variance weighted. SHapley Additive exPlanations (SHAP) analysis of ML-enhanced model interpretability. High intake (> 1.631 g/day) was inversely associated with HorvathAge and phenotypic age but positively correlated with Horvathtelo. High ALA intake (≥ 1.520 g/day) exhibited similar effects, which were validated by MR causality analyses. Furthermore, high omega-3 and ALA levels were inversely associated with phenotypic age. Notably, elevated EPA and DHA levels were exclusively positively associated with Horvathtelo. The RCS models revealed nonlinear associations between omega-3 (and its components) and Horvathtelo, as well as between EPA/DHA and HorvathAge. The SHAP analysis of linear support vector machine (LSVM) ranked feature importance as omega-3 > ALA > DHA > EPA. Increasing omega-3 and ALA intake was significantly associated with reduced HorvathAge, while components (ALA/EPA/DHA) improved Horvathtelo through nonlinear dose-response effects. MR validated anti-aging associations for omega-3 and ALA. The LSVM model prioritized omega-3 as the most influential feature, followed by ALA, DHA, and EPA.