Evaluating the Oxidative Balance Score for Peripheral Artery Disease Risk: Integrating Epidemiologic Modeling and <scp>SHAP</scp> ‐Interpretable Machine Learning in <scp>NHANES</scp>

Oxidative stress plays a central role in the development of peripheral artery disease (PAD), yet composite indices quantifying its impact remain underutilized. The oxidative balance score (OBS), integrating dietary and lifestyle exposures, may offer a comprehensive approach to vascular risk stratification. We analyzed data from 7249 U.S. adults in NHANES 1999-2004 to examine the association between OBS and PAD, defined by an ankle-brachial index ≤ 0.9. Multivariable logistic regression and restricted cubic spline models were used to assess both linear and nonlinear associations. Twelve machine learning models were constructed to predict PAD status; the top-performing model (GLMNet) was interpreted using SHapley Additive exPlanations (SHAP) to identify key predictors. Higher OBS levels were significantly associated with reduced PAD prevalence (fully adjusted OR per unit increase: 0.963; 95% CI: 0.944-0.983; <i>p</i> < 0.001), with a U-shaped dose-response curve. The inverse association remained consistent across most subgroups. In sensitivity analyses focusing on diabetic individuals, the total OBS was not significantly associated with PAD, but the lifestyle component of OBS remained protective. SHAP analysis identified total folate intake, physical activity, and serum cotinine as the most influential predictors, while classic antioxidants such as vitamin C and vitamin E had limited predictive value. Integrating conventional regression and interpretable machine learning, this study demonstrates a nonlinear, inverse association between oxidative balance and PAD. OBS may serve as a practical composite biomarker for vascular risk profiling, with particular relevance for lifestyle-based interventions in high-risk populations.