Title : County-level social isolation in the United States: Estimating missing prevalence and identifying structural correlates of geographic variation
Abstract:
Social isolation is a public health crisis in the United States and was identified as a national epidemic by the U.S. Surgeon General in 2023. Social Isolation is influenced by social determinants of health (SDOH). However, county-level surveillance data remain incomplete, with 727 of 3,144 U.S. counties missing estimates in the Centers for Disease Control and Prevention’s Population Level Analysis and Community Estimates (PLACES) 2024 dataset because 11 states did not administer the relevant SDOH module. We aimed to generate nationwide county-level age-adjusted isolation estimates and examine structural correlates of geographic variation using county-level SDOH. Available isolation estimates for 2,417 counties from the PLACES 2024 dataset were merged with 331 SDOH predictors from the Agency for Healthcare Research and Quality (AHRQ) 2022 database. We applied a linear regression model with iterative predictor pruning to estimate isolation for 727 missing counties and to quantify the relative contribution of structural factors. SDOH predictors were classified as policy-actionable or non-actionable based on modifiability and statistical interpretability, and contribution scores identified the dominant structural correlate in each county. Model performance was compared with a random forest benchmark and evaluated using stratified validation and leave-one-state-out geographic validation. The final model performance was stable (mean absolute error = 1.49%), with comparable or lower error in leave-one-state-out validation (mean MAE = 1.27), supporting geographic generalizability. Predicted isolation values for missing counties (mean = 34.09%) closely aligned with observed estimates (mean = 34.23%). Construct validation demonstrated consistent associations between predicted isolation and related population health measures. Among policy-relevant structural correlates, broadband access most frequently exhibited the strongest statistical contribution across counties followed by public-only insurance coverage and educational attainment. These findings extend existing surveillance by enabling complete national coverage and highlight structural factors statistically associated with geographic variation in social isolation. Results of this study, although observational and not causal, highlight potential areas for future intervention and support hypothesis generation for targeted public health strategies addressing social isolation, digital inclusion, and broader social determinants of health.
