METHODS: EHR data for patients >18 years old were extracted from two large health system EHRs and geocoded to Denver census tracts (CT). Depression was identified by presence of at least one notation from several depression-specific ICD-9 codes (i.e., 296.x, 298.0, 301.4, 309.x, 311), after a 2011-2012 visit. Overall coverage and crude prevalence rates for depression were calculated for each CT using EHR data for numerators and American Community Survey (ACS) data for denominators. Adjustment for differences in age, gender, and race were applied. ACS data were used to create Singh CT deprivation indices.
RESULTS: Among 169,906 patients (36.9% coverage of Denver’s > 18 year old population) geocoded to 144 CTs, coverage by CT ranged from 10.9% to 58.9%. Crude overall depression prevalence rate was 12.7% with women having higher rates (15.7%) compared to men (8.8%). By CT, depression rates ranged from 8.6% to 17.6%. Rates differed by race with whites having the highest (16.3%) compared to Asian/Pacific Islanders with the lowest (6.4%). Rates varied by age group with 18 to 24 year olds having the lowest (6.8%) compared to >75 year olds with the highest (20.7%). CT depression prevalence had a significant negative association with Singh CT deprivation index. The depression rate adjusted for gender was 12.3%, but rates adjusted for age and race were each 13.8%. Depression prevalence rate for homeless individuals was 75%.
CONCLUSIONS: Linking geo-located EHR and community-level data allowed sub-county assessment of depression prevalence rates previously not possible with federal survey data. Depression prevalence varied by CT, gender and age and was inversely associated with socioeconomic deprivation. Homeless individuals had the highest rates, but by definition were excluded from geographic analyses. Public health agencies should consider this method to identify potential “hot-spots” and inform multi-sector interventions, community engagement, and further public health monitoring.