The impact of frequency and duration of air quality monitoring: Atlanta, GA, data modeling of air pollution and mortality

The purpose of this analysis is threefold. We first examine the extent to which a longer series of data improves our understanding of air pollution on human mortality in the Atlanta, GA, area by updating the findings presented in Klemm and Mason (J. Air Waste Manage. Assoc. 2000, 50, 1433-1439) and Klemm et al. (Inhal. Toxicol. 2004, 16 (Suppl 1), 131-141) with 7.5 additional years of data. We explore estimated effects on two age groups (<65 and 65+) and four categories of cause of death. Second, we investigate how enlarging the geographic area of inquiry influences the estimated effects. Third, because some air quality (AQ) measures are monitored less frequently than daily, we investigate the extent to which AQ measurement frequency can influence estimates of relationships with human mortality. Our analytical approach employs a Poisson regression model using generalized linear modeling in S-Plus to estimate the relationship between daily AQ measures and daily mortality counts. We show that the estimated effects and their associated t values vary by year for nine AQ measures (particulate matter with aerodynamic diameter < or =2.5 microm PM2.5], elemental carbon EC], organic carbon OC], NO3, SO4, O3, NO2, CO, and SO2). Several of the estimated AQ effects show downward trends during the 9-year period of study. The estimated effects tend to be strongest for the AQ measurement during the day of death and tend to decrease with additional lags. Enlarging the geographic area from two to four counties in the metropolitan area decreased the estimated effects, perhaps partly due to the fact that the measurement site is located in one of the two original counties. Estimated effects utilizing data as if the AQ were only measured every 3rd or every 6th day each week or twice per week vary from lower to higher than that estimated with daily measurements, although the t values are lower, as expected.