There is an ever-growing recognition that zip code is important in predicting a person’s health. This illustrates the need for data more localized than the national, state, or even county level to improve the performance of local health systems and advance population health and health equity. The lack of health data at the community level has “hampered the effective pursuit of public health goals,” because such data would be valuable for conducting community health needs assessments, documenting and identifying medically underserved populations, planning local health programs, and developing evidence-based and place-based policy.

The authors proposed a method to meet challenges in generating health estimates for granular geographic areas in which the survey sample size is extremely small. 

The authors' generalized linear mixed model predicts health outcomes using both individual-level and neighborhood-level predictors. The model’s feature of nonparametric smoothing function on neighborhood-level variables better captures the association between neighborhood environment and the outcome. Using 2011 to 2012 data from the California Health Interview Survey, authors demonstrate an empirical application of this method to estimate the fraction of residents without health insurance for ZIP Code Tabulation Areas (ZCTAs). The method generated stable estimates of uninsurance for 1519 of 1765 ZCTAs (86%) in California.

The proposed method can increase the value of health surveys by providing modeled estimates for health data at a granular geographic level. It can account for variations in health outcomes at the neighborhood level as a result of both socioeconomic characteristics and geographic locations.