(P25-040-24) Cognitive Outcome Forecasting Using Supervised Machine Learning: A Multi-Factor Analysis With Anthropometrics, Demographics, Diet Indices, and Cardiovascular Markers

Objectives: This study used supervised machine learning (ML) models to predict cognitive performance based on a comprehensive set of health and behavioral factors, including physical activity and dietary patterns, which have been linked to reduced cognitive decline in the literature. The goal was to identify key contributors to cognitive function, enabling proactive cognitive health management through personalized interventions.

Methods: Data from 380 adults (Age range: 19-82y, F: 228) was used to develop a supervised learning regression model predicting cognitive performance (reaction time) during a modified Eriksen Flanker task. Selected features included age, sex, ethnicity, income, education level, weight, height, waist circumference, body mass index (BMI), adherence to dietary indices (HEI [Healthy Eating Index], DASH [Dietary Approaches to Stop Hypertension], Mediterranean and MIND [Mediterranean-DASH Intervention for Neurodegenerative Delay]), weekly leisure-time activity, systolic and diastolic blood pressure. The dataset was split into a training and testing set. Predictive models (Decision Trees, Random Forest, Ada Boost, XG Boost) were deployed with hyperparameter tuning and cross-validation for model generalizability. Feature importance scores (0-1) were computed using Mean decrease in Impurity to identify the most salient predictors. Performance was evaluated using mean absolute (MAE) and mean squared error (MSE).

Results: Random Forest Regressor exhibited the best performance, with the lowest MAE on both the training (.599) and testing sets (.785), and lowest MSE on the training (.752) and testing sets (.985). Age was the most significant feature with the highest importance score (0.17), followed by diastolic BP (.125) and BMI (.1), WLTA (.075), HEI (0.07), DASH (0.04), MIND (.038). Ethnicity, Income, and Education levels had minimal predictive effect on the model’s performance (0-.01).

Conclusions: Understanding the collective impact of lifestyle factors on cognitive health is crucial for an aging population and the increasing prevalence of cognitive disorders. Modeling this relationship using robust ML models can inform health interventions tailored to individual risk profiles and formulate preventive strategies.

Funding Sources: Personalized Nutrition Initiative and National Center for Supercomputing Application, UIUC