Social determinants and machine learning prediction of complete tooth loss in older adults across United States counties

Introduction: Complete tooth loss reflects a lifetime of accumulated oral disease and concentrates among socially disadvantaged older adults, yet county-level evidence on the determinants that best predict its geographic distribution is limited. Methods: Ecological cross-sectional study of 2,946 United States counties combining model-based prevalence of complete tooth loss among adults aged 65 years or older from CDC PLACES with American Community Survey 2017-2021 social determinant measures. A linear ridge regression and a gradient-boosted model were compared with random five-fold and spatial block cross-validation; permutation importance and partial dependence described the contributions of each predictor. Results: Mean complete tooth loss prevalence was 15.6% (SD 5.7). Each standard deviation increase in county poverty was associated with 1.77 percentage points higher prevalence (95% CI 1.52 to 2.02) and each standard deviation increase in low educational attainment with 0.67 percentage points (95% CI 0.46 to 0.88) after adjustment. Under random cross-validation the two models performed similarly (R-squared 0.72 versus 0.72), whereas under spatial cross-validation the gradient-boosted model retained an R-squared of 0.53 compared with 0.34 for the linear model. Poverty was the leading social determinant by permutation importance (0.162). Conclusions: County socioeconomic deprivation independently predicts complete tooth loss in older adults, and machine learning predictions transfer better across space than linear models. Spatial cross-validation should be routine when small-area models are intended to guide resource allocation.