Secondary Logo

Journal Logo


Airborne particles have been associated with early deaths and morbid events in hundreds of locations throughout the world. More mixed results have been reported with gaseous air pollutants. All of these associations have used ambient measures as metrics of exposure, as true exposures are not available for the large populations studied. This has raised questions concerning the impact of measurement error on those associations. Recently, studies have shown that changes over time in ambient particle concentrations, particularly for fine combustion particles, are good surrogates for changes over time in exposure to particles of ambient origin. The associations for gaseous air pollutants are less studied, but appear weaker. Because of the strong correlations among ambient pollutants, it is possible for ambient concentrations of one pollutant to act as a surrogate for exposure to another pollutant. This raises questions about the interpretability of multi-pollutant models. We investigated the implications of this measurement error in a simulation study. First, we conducted a longitudinal multipollutant personal exposure assessment of 56 and 43 subjects in Baltimore and Boston respectively. Using the observed covariance structure among personal exposures and corresponding ambient concentrations for each subject, we estimated a Wishart distribution in each city. The Wishart distribution is the distribution of covariances, and captures the variability in those correlations among subjects. In each city we sampled covariances from the distribution, and for each sampled pattern of association between personal and ambient, simulated ambient and personal measurements, and true risk given an assumed association (b = 0.05) between exposure and risk. We summed these risks over all of the sampled covariances, obtaining a daily count of events, for example deaths. We regressed that daily count against the ambient concentrations, simulating the analysis normally conducted in epidemiological studies, and compared the resulting coefficients to the true effect. Each simulation was repeated 100 times. In both Baltimore and Boston, we found that PM2.5 was the only pollutant where a true association with exposure resulted in an association with ambient concentration (b = 0.019, 95% CI 0.016, 0.023 for Baltimore, and 0.028 (0.024, 0.033) for Boston). In both cities, a true PM association resulted in a significant association with ambient NO2 (b = .018 (.012, .021) for Baltimore and .028 (.025, .032) for Boston), true associations with NO2 resulted in observed associations with ozone, and true associations with either O3 or SO2 resulted in no significant associations. Hence the observed epidemiological associations with air pollution are not consistent with true associations with either O3 or SO2, and are more consistent with true associations with PM2.5 than with NO2. Supported by EPA Grant R-82-7353 and EPA Contract IC-R238-NASA

(1) Harvard School Of Public Health

© 2003 Lippincott Williams & Wilkins, Inc.