Revisiting the Veterans Cohort Mortality Study: New results and synthesis

The Veterans Cohort Mortality Study began in 1999 in collaboration with Washington University in St. Louis, comprising ~70,000 male military veterans. We published six research papers on this cohort, considering the dynamics of all-cause mortality as the subjects aged and environmental parameters changed. This paper summarizes those results and presents new results by age group. Pollutants included monitored and modeled criteria pollutants, vehicular traffic density (annual km driven per unit of county land area), and modeled nationwide levels of hazardous species. In addition to spatial relationships, we examined the effects of exposure timing through separate analyses of sequential follow-up and exposure periods from 1976 to 2001. Risks associated with peak ozone decreased with lag between exposure and response, suggesting acute effects. Risks associated with traffic were invariant over time and consistent across five exposure databases. Associations with ozone were also coherent across databases; we found no consistent associations with particulate matter. Epidemiology considers both spatial and temporal relationships; most long-term studies focus on spatial gradients at a given time, thus masking effects of cohort aging and other trends during follow-up. Our new analyses distinguished between these temporal effects by analyzing age deciles for which separate mortality risks had been estimated for nationwide levels of nitrogen oxides (NO_x), benzene, and traffic density during four sequential follow-up subperiods, thus providing 40 sets of mortality risk coefficients. We used ordinary least squares regression to define relationships with subject age and follow-up year for the data set of 40 coefficients. We found strong nonlinear relationships between subject age and mortality coefficients for smoking, climate, poverty status, and air pollution; only smoking and climate coefficients changed over time as well. We concluded that these pollutant-mortality relationships reflected differences among the veterans' residential locations rather than changes in their pollution exposures during follow-up. We saw no evidence that cleaner air reduced mortality. Implications: Recent air pollution mortality studies emphasize PM_2.5 (particulate matter with an aerodynamic diameter <2.5 μm); we show associations with many other pollutants and a measure of traffic intensity. Control policies should thus be based on multipollutant analyses. We found no reduced risks with improved air quality after distinguishing cohort aging from purely temporal effects; longitudinal studies of accountability must thus account for changes in demography and exposures. Our studies of exposure timing indicate mainly coincident responses and no evidence for cumulative effects typical of smoking; we had no information on personal exposures. We found the strongest risks were associated with high-traffic locations rather than outdoor air quality per se.