Investigating the association between the food inflammation scores of individuals and stroke in adults: an extreme gradient boosting machine learning model interpreted with shapley additive explanations

Abstract Background Chronic systemic inflammation is a pivotal modifiable risk factor for stroke. The food-based Food Inflammation Index (FII) offers a novel approach to assess dietary inflammatory potential, yet the association between its derivative, the Food Inflammation Scores of Individuals (FISI), and stroke prevalence remains to be elucidated. Methods This study analyzed a cohort of 19,681 adults from the NHANES (2007–2018) database. The FISI-stroke association was assessed using multivariable logistic regression and machine learning models (XGBoost), interpreted via SHAP analysis. Results Higher FISI scores were positively associated with increased stroke prevalence in a dose-dependent manner. Specifically, a one-unit rise in FISI34, FISI26-USDA, and FISI26-CHINA corresponded to 7%, 18%, and 22% higher stroke odds, respectively. XGBoost modeling identified FISI34 as a key predictor, corroborating regression findings. Conclusions This study establishes a robust link between higher FISI, derived from the FII, and stroke risk. The FII framework surpasses nutrient-based indices by providing personalized, actionable, food-specific guidance for stroke prevention through anti-inflammatory diets. Graphical Abstract