TOPLINE

A machine learning model combining social, political, and behavioral factors explained 78% of the variation in obesity rates across US counties. Physical inactivity was the single strongest predictor. A simulation testing a 10% reduction in inactivity showed obesity prevalence could drop in many counties — though a few regions moved in the opposite direction.

METHODOLOGY

• While obesity prediction models increasingly use AI, most focus on clinical and individual-level data and rarely account for broader social, cultural, political, and environmental influences.
• Researchers analyzed county-level data from multiple public sources, covering 34 variables to understand what drives obesity rates across the US.
• They tested six machine learning models and selected the best-performing one (Light Gradient Boosting Machine, or LightGBM) based on lowest prediction error and consistent accuracy between training and test data. They then trained it on data from 2284 US counties; 70% of the data was used to build the model and the remaining 30% was used to test its accuracy.
• Predictor variables spanned multiple levels including culture, politics, policies, social and physical environments, behaviors, health characteristics, power dynamics, and social capital; the final model used the 29 most predictive of these 34 variables.
• The outcome variable was the 2022 age-adjusted prevalence of adults aged 18 years or older with BMI ≥ 30.

TAKEAWAY

• The optimal obesity prediction model with 29 features explained 78% of the variance in county-level obesity prevalence, demonstrating excellent predictive capability.
• Physical inactivity emerged as the strongest contributor to obesity prevalence, with increments in physical inactivity prevalence associated with a rise in predicted obesity prevalence from approximately 34.5% to 41.5%.
• Smoking, excessive drinking, culture grouping, insufficient sleep, and political ideology ranked second to sixth in importance.
• A simulation modeling a 10% reduction in physical inactivity lowered predicted obesity prevalence by up to approximately 2% in some counties, though several regional hot spots showed an increase in predicted obesity prevalence instead.

IN PRACTICE

"These study results suggest that reducing population-level physical inactivity may be important for lowering county-level obesity prevalence. Physical inactivity may reflect sustained low-energy flux states linked to metabolic dysfunction and obesity," the authors wrote.

SOURCE

The study was led by Nicolaas P. Pronk, PhD, HealthPartners Institute in Minneapolis. It was published online in Obesity.

LIMITATIONS

Key limitations include the reliance on county-level and self-reported data, a cross-sectional design, residual geographic clustering, possible feature-selection bias, and the exclusion of genetic factors. Another limitation is that feature selection was performed once on the training data.

DISCLOSURES

No study funding was explicitly mentioned. The authors declared having no conflicts of interest.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.