Multilevel prediction of Alzheimer's disease dementia in the United States: An artificial intelligence analysis

Variables predicting Alzheimer's disease (AD) are not limited to individual-level risk factors. The purpose of this investigation is to assess multilevel predictors of AD prevalence.

US county-level datasets incorporating 45 predictor variables were analyzed cross-sectionally using artificial intelligence analytical methods. A Light Gradient-Boosting Machine model was trained to predict county-level AD after which model performance and feature importance were evaluated.

The final model retained 20 features and explained 75% (R ² = 0.75) of the variance in AD prevalence. Racial and ethnic minority status showed the highest importance value (0.848), far exceeding all other features (e.g., poor sleep ranked second with importance value of 0.153).

This study confirmed upstream factors as being significant predictors of AD prevalence and racial and ethnic minority status as being the most important. From a policy perspective, efforts to reduce population levels of AD prevalence should consider addressing racial and ethnic disparities.