Vehicular traffic effects on survival within the Washington University-EPRI veterans cohort: New estimates and sensitivity studies

We analyzed survival patterns among ∼ 70,000 U.S. male military veterans relative to vehicular traffic density in their counties of residence, by mortality period and type of exposure model. Previous analyses show traffic density to be a better predictor than concentrations of criteria air pollutants. We considered all subjects and also the subset defined by availability of air quality monitoring data from the U.S. EPA PM2.5 Speciation Trends Network (STN). Traffic density is a robust predictor of mortality in this cohort; statistically significant estimates of deaths associated with traffic range from 1.3% to 4.4%, depending on the method of analysis. This range of uncertainty is larger than the traditional 95% confidence intervals for each estimate (1-2%). Our best estimate of the relative risk for the entire follow-up period is 1.03. These deaths occurred mainly before 1997 in counties with STN air quality data, which tend to be more urban. We identified a threshold in mortality responses to traffic density, corresponding to county-average traffic flow rates of about 4000 vehicles/day. Relative risks were significantly higher in the more urban (STN) counties in the early subperiods, but this gradient appears to have diminished over time. We found larger risks by pooling results from separate portions of the overall follow-up period, relative to considering the entire period at once, which suggests temporal changes in confounding risk factors such as smoking cessation, for example. These results imply that the true uncertainties in cohort studies may exceed those indicated by the confidence intervals from a single modeling approach.