Fine particulate air pollution and all-cause mortality within the Harvard Six-Cities Study: variations in risk by period of exposure

PURPOSE: We used Poisson regression methods to examine the relation between temporal changes in the levels of fine particulate air pollution (PM_2.5) and the risk of mortality among participants of the Harvard Six Cities longitudinal study.

METHODS: Our analyses were based on 1430 deaths that occurred between 1974 and 1991 in a cohort that accumulated 105,714 person-years of follow-up. For each city, indices of PM_2.5 were derived using daily samples. Individual level data were collected on several risk factors including: smoking, education, body mass index (BMI), and occupational exposure to dusts. Time-dependent indices of PM_2.5 were created across 13 calendar periods (< 1979, 1979, 1980, … , 1989, ≥ 1990) to explore whether recent or chronic exposures were more important predictors of mortality.

RESULTS: The relative risk (RR) of mortality calculated using Poisson regression based on average city-specific exposures that remained constant during follow-up was 1.31 [95% confidence interval (CI) = 1.12–1.52] per 18.6 μg/m³ of PM_2.5. This result was similar to the risk calculated using the Cox model (RR = 1.26, 95% CI = 1.08–1.46). The RR of mortality was attenuated when the Poisson regression model included a time-dependent estimate of exposure (RR = 1.19, 95% CI = 1.04–1.36). There was little variation in RR across time-dependent indices of PM_2.5.

CONCLUSIONS: The attenuated risk of mortality that was observed with a time-dependent index of PM_2.5 is due to the combined influence of city-specific variations in mortality rates and decreasing levels of air pollution that occurred during follow-up. The RR of mortality associated with PM_2.5 did not depend on when exposure occurred in relation to death, possibly because of little variation between the time-dependent city-specific exposure indices.