Comparison of PM2.5 exposure estimates in the REGARDS cohort: understanding differences by community type and exposure assignment choices

Abstract

Background: There is biologic rationale for associations between ambient PM2.5 exposure and type 2 diabetes. However, studies of these associations demonstrate mixed results, potentially due to differences in: PM2.5 estimation by community types and regions; PM2.5 estimation methods that optimize temporal vs. spatial variability; and exposure lags and durations assigned to individuals for evaluation with diabetes outcomes. Methods: We evaluated several PM2.5 data sources and exposure assignment choices for 10,332 participants free of diabetes at baseline (2003-2007) and with follow-up data on diabetes in the REasons for Geographic and Racial Differences in Stroke (REGARDS) cohort. We calculated exposure durations of 2-weeks, 30-days, and 1-year; lagged 1-day, 6-months, and 1-year prior to baseline, and evaluated these by: US region (Northeast, South, Midwest, and West); community type (high density urban, low density urban, suburban and small town, and rural); and year for two sources of PM2.5: CDC EPA Downscaler model and CDC Wonder data modeled from NASA satellite observations and EPA monitor data. Results: Participants in the analysis had a mean (SD) age of 63.0 (8.5) years, were 55.8% female; 32.4% black. The mean (SD) PM2.5 estimates from CDC Wonder were 13.5 (4.2) µg/m3, 13.5 (3.6) µg/m3, and 13.3 (2.0) µg/m3, for 2-week, 30-day, and 1-year exposure periods, respectively. One-way analysis of variance (ANOVA) of PM2.5 exposure estimates of all durations showed significant differences (p ≤ 0.01) by community type, region, and year. Differences by region and community type became more pronounced with longer exposure durations. Analysis of additional exposure lags and durations is in progress. Conclusions: These results suggest that exposure assignment choices can either exacerbate or mitigate underlying spatial differences in this cohort, which can lead to differential associations between PM2.5 and diabetes. Future work should focus on better exposure classification in order to more clearly estimate this association.