Physical inactivity emerged as the top predictor of diabetes prevalence across U.S. counties in a new machine‑learning study, according to researchers from the HealthPartners Institute.
Study links lack of exercise and minority status to higher diabetes rates
The cross‑sectional analysis, led by Nicolaas P. Pronk, Ph.D., examined more than 3,000 counties using 27 publicly sourced variables, including data from the CDC, County Health Rankings & Roadmaps, and the U.S. Bureau of Economic Analysis. A Light Gradient Boosting Machine model was trained, and after pruning, 17 variables remained, collectively explaining about 95 % of the variation in county‑level diabetes prevalence.
Ranking factors with Shapley values, the team found two variables that stood out. First, the share of residents reporting no leisure‑time physical activity showed the highest importance score. Second, the proportion of people identified as racial or ethnic minorities—measured through a social vulnerability index—ranked just behind, surpassing all other predictors.
Related: The history of hair transplant surgery
Other notable contributors included frequent physical distress, obesity, excessive drinking, and insufficient sleep. Upstream social and historical elements also appeared: counties with a larger historical enslaved population, lower voter turnout, and greater food insecurity tended to have higher predicted diabetes rates, though these factors ranked lower than the top four.
Simulation suggests modest gains from reducing inactivity
To explore policy implications, the researchers simulated a 10 % reduction in physical inactivity while holding other characteristics steady. The model projected an average decline of up to 0.8 percentage points in diabetes prevalence across counties.
“This exploratory model‑based scenario may be helpful to support public health practice or policy considerations for the role of county‑based physical activity levels,” the authors wrote.
Related: Get answers with a DNA test
Given that inactivity was both the strongest and most modifiable factor, the study recommends prioritizing physical activity assessment in primary and diabetes care. Clinicians might issue simple “prescriptions” encouraging patients to walk more, sit less, and, where possible, connect them with local activity resources.
Limitations and future directions
The authors caution against over‑interpreting the rankings. Many inputs rely on self‑reported data, and the ecological, county‑level design cannot establish causality. Correlations among predictors and residual spatial clustering may also influence results. Thus, the identified factors reflect relative predictive contributions at the population level rather than independent causal effects.
Nevertheless, the investigators argue that pairing an ecological framework with AI methods capable of handling complex interactions provides a new lens for pinpointing where diabetes risk concentrates.
Related: Medicaid Engagement: Turning Communication into Action for Better Health
Further investigation is needed to determine how this framework might guide interventions, identify optimal locations for public health efforts, and assess the effectiveness of specific policies.
The study adds to evidence that tackling sedentary behavior, especially in communities facing social vulnerabilities, could be a critical component of diabetes prevention strategies.
