Generating a Model of Computational Correlation Between Lithiasis and Diet: Insights into Anthropometric Predictors and the Exploratory Role of AI

Abstract

The relationship between lithiasis (kidney stone disease) and dietary habits has been widely debated but remains incompletely clarified. This study used a computational framework that combined statistical testing with AI-based exploratory modelling to assess the relative contributions of dietary and anthropometric parameters on lithiasis recurrence and stone type. Data from 144 patients with documented kidney stones were analysed. AI algorithms were not intended as predictive tools due to the small and imbalanced dataset; they served to identify the most influential variables. Classical statistical methods revealed no significant associations between relapse and isolated dietary factors such as fast-food or carbonated drinks, though higher water intake showed a protective trend (p ≈ 0.09). Anthropometric features, particularly ideal weight, were significantly associated with both relapse (p ≈ 0.006) and stone type (corrected p ≈ 0.046). Machine learning models (multinomial logistic regression, Random Forest, and XGBoost) demonstrated modest predictive performance (accuracy ≈ 55–63%), with variable importance analyses highlighting weight, ideal weight, and water intake as stronger predictors than single dietary variables. The findings suggest that anthropometric parameters, rather than isolated dietary habits, show stronger associations with stone type and recurrence, and that AI models hold promise but require larger datasets and biochemical variables to achieve clinical utility.