Ronald E. Wyzga

Air Pollution and Mortality: Issues and Uncertainties

Results from 31 epidemiology studies linking air pollution with premature mortality are compared and synthesized. Consistent positive associations between mortality and various measures of air pollution have been shown within each of two fundamentally different types of regression studies and in many variations within these basic types; this is extremely unlikely to have occurred by chance. In this paper, the measure of risk used is the elasticity, which is a dimensionless regression coefficient defined as the percentage change in the dependent variable associated with a 1% change in an independent variable, evaluated at the means. This metric has the advantage of independence from measurement units and averaging times, and is thus suitable for comparisons within and between studies involving different pollutants. Two basic types of studies are considered: time-series studies involving daily perturbations, and cross-sectional studies involving longer-term spatial gradients. The latter include prospective studies of differences in individual survival rates in different locations and studies of the differences in annual mortality rates for various communities. For a given data set, time-series regression results will vary according to the seasonal adjustment method used, the covariates included, and the lag structure assumed. The results from both types of cross-sectional regressions are highly dependent on the methods used to control for socioeconomic and personal lifestyle factors and on data quality. A major issue for all of these studies is that of partitioning the response among collinear pollution and weather variables. Previous studies showed that the variable with the least exposure measurement error may be favored in multiple regressions; assigning precise numerical results to a single pollutant is not possible under these circumstances. We found that the mean overall elasticity as obtained from time-series studies for mortality with respect to various air pollutants entered jointly was about 0.048, with a range from 0.01 to 0.12. This implies that about 5% of daily mortality is associated with air pollution, on average. The corresponding values from population-based cross-sectional studies were similar in magnitude, but the results from the three recent prospective studies varied from zero to about five times as much. Long-term responses in excess of short-term responses might be interpreted as showing the existence of chronic effects, but the uncertainties inherent in both types of studies make such an interpretation problematic.

PM2.5 constituents and related air quality variables as predictors of survival in a cohort of U.S. military veterans.

Air quality data on trace metals, other constituents of PM2.5, and criteria air pollutants were used to examine relationships with long-term mortality in a cohort of male U.S. military veterans, along with data on vehicular traffic density (annual vehicle-miles traveled per unit of land area). The analysis used county-level environmental data for the period 1997–2002 and cohort mortality for 1997–2001. The proportional hazards model included individual data on age, race, smoking, body mass index, height, blood pressure, and selected interactions; contextual variables also controlled for climate, education, and income. In single-pollutant models, traffic density appears to be the most important predictor of survival, but potential contributions are also seen for NO2, NO3−, elemental carbon, nickel, and vanadium. The effects of the other main constituents of PM2.5, of crustal particles, and of peak levels of CO, O3, or SO2 appear to be less important. Traffic density is also consistently the most important environmental predictor in multiple-pollutant models, with combined relative risks up to about 1.2. However, from these findings it is not possible to discern which aspects of traffic (pollution, noise, stress) may be the most relevant to public health or whether an area-based predictor such as traffic density may have an inherent advantage over localized measures of ambient air quality. It is also possible that traffic density could be a marker for unmeasured pollutants or for geographic gradients per se. Pending resolution of these issues, including replication in other cohorts, it will be difficult to formulate additional cost-effective pollution control strategies that are likely to benefit public health.

Is the air pollution health research community prepared to support a multipollutant air quality management framework

Ambient air pollution is always encountered as a complex mixture, but past regulatory and research strategies largely focused on single pollutants, pollutant classes, and sources one-at-a-time. There is a trend toward managing air quality in a progressively “multipollutant” manner, with the idealized goal of controlling as many air contaminants as possible in an integrated manner to achieve the greatest total reduction of adverse health and environmental impacts. This commentary considers the current ability of the environmental air pollution exposure and health research communities to provide evidence to inform the development of multipollutant air quality management strategies and assess their effectiveness. The commentary is not a literature review, but a summary of key issues and information gaps, strategies for filling the gaps, and realistic expectations for progress that could be made during the next decade. The greatest need is for researchers and sponsors to address air quality health impacts from a truly multipollutant perspective, and the most limiting current information gap is knowledge of personal exposures of different subpopulations, considering activities and microenvironments. Emphasis is needed on clarifying the roles of a broader range of pollutants and their combinations in a more forward-looking manner; that is not driven by current regulatory structures. Although advances in research tools and outcome data will enhance progress, the greater need is to direct existing capabilities toward strategies aimed at placing into proper context the contributions of multiple pollutants and their combinations to the health burdens, and the relative contributions of pollutants and other factors influencing the same outcomes. The authors conclude that the research community has very limited ability to advise multipollutant air quality management and assess its effectiveness at this time, but that considerable progress can be made in a decade, even at current funding levels, if resources and incentives are shifted appropriately.

Air Pollution and Mortality: The Implications of Uncertainties in Regression Modeling and Exposure Measurement

In a previous paper, we showed that the mean effects on daily mortality associated with air pollution are essentially the same for gases and particulate matter (PM) and are invariant with respect to particle size and composition, based on 27 statistical studies that had been published at that time. Since then, a new analysis reported stronger mortality associations for the fine fractions of PM obtained from dichotomous samplers, relative to the coarse fractions. In this paper, we show that differential measurement errors known to be present in dichotomous sampler data preclude reliable determination of such statistical relationships by particle size. Further, it is necessary to consider gaseous pollutants simultaneously with particles to provide robust estimates of the responsibilities for the implied daily mortality gradients. Finally, certain regression model specifications may be sensitive to differences in frequency distribution characteristics according to particle size.

THE WASHINGTON UNIVERSITY- EPRI VETERANS' COHORT MORTALITY STUDY: Preliminary Results

This article presents the design of and some results from a new prospective mortality study of a national cohort of about 50,000 U.S. veterans who were diagnosed as hypertensive in the mid 1970s, based on approximately 21 yr of follow-up. This national cohort is male with an average age at recruitment of 51 ± 12 yr; 35% were black and 81% had been smokers at one time. Because the subjects have been receiving care at various U.S. Veterans Administration (VA) hospitals, access to and quality of medical care are relatively homogeneous. The health endpoints available for analysis include all-cause mortality and specific diagnoses for morbidity during VA hospitalizations; only the mortality results are discussed here. Nonpollution predictor variables in the baseline model include race, smoking (ever or at recruitment), age, systolic and diastolic blood pressure (BP), and body mass index (BMI). Interactions of BP and BMI with age were also considered. Although this study essentially controls for socioeconomic status by design because of the homogeneity of the cohort, selected ecological variables were also considered at the ZIP code and county levels, some of which were found to be significant predictors. Pollutants were averaged by year and county for TSP, PM10, CO, O3, and NO2; SO2 and Pb were considered less thoroughly. Both mean and peak levels were considered for gases. SO42 data from the AIRS database and PM2.5, coarse particles, PM15, and SO42 from the U.S. EPA Inhalable Particulate (IP) Network were also considered. Four relevant exposure periods were defined: 1974 and earlier (back to 1953 for TSP), 1975–1981, 1982–1988, and 1989–1996. Deaths during each of the three most recent exposure periods were considered separately, yielding up to 12 combinations of exposure and mortality periods for each pollutant. Associations between concurrent air quality and mortality periods were considered to relate to acute responses; delayed associations with prior exposures were considered to be emblematic of initiation of chronic disease. Preexposure mortality associations were considered to be indirect (noncausal). The implied mortality risks of long-term exposure to air pollution were found to be sensitive to the details of the regression model, the time period of exposure, the locations included, and the inclusion of ecological as well as personal variables. Both positive and negative statistically significant mortality responses were found. Fine particles as measured in the 1979–1984 U.S. EPA Inhalable Particulate Network indicated no significant (positive) excess mortality risk for this cohort in any of the models considered. Among the positive responses, indications of concurrent mortality risks were seen for NO2 and peak O3, with a similar indication of delayed risks only for NO2. The mean levels of these excess risks were in the range of 5–9%;. Peak O3 was dominant in two-pollutant models and there was some indication of a threshold in response. However, it is likely that standard errors of the regression coefficients may have been underestimated because of spatial autocorrelation among the model residuals. The significant variability of responses by period of death cohort suggests that aggregation over the entire period of follow-up obscures important aspects of the implied pollution–mortality relationships, such as early depletion of the available pool of those subjects who may be most susceptible to air pollution effects.

Infant mortality and air pollution: a comprehensive analysis of U.S. data for 1990.

ABSTRACT This paper uses U.S. linked birth and death records to explore associations between infant mortality and environmental factors, based on spatial relationships. The analysis considers a range of infant mortality end points, regression models, and environmental and socioeconomic variables. The basic analysis involves logistic regression modeling of individuals; the cohort comprises all infants born in the United States in 1990 for whom the required data are available from the matched birth and death records. These individual data include sex, race, month of birth, and birth weight of the infant, and personal data on the mother, including age, adequacy of prenatal care, and smoking and education in most instances. Ecological variables from Census and other sources are matched on the county of usual residence and include ambient air quality, elevation above sea level, climate, number of physicians per capita, median income, racial and ethnic distribution, unemployment, and population density. The air quality variables considered were 1990 annual averages of PM10, CO, SO2, SO4 2-, and “non-sulfate PM10” (NSPM10—obtained by subtracting the estimated SO4 2-mass from PM10). Because all variables were not available for all counties (especially maternal smoking), it was necessary to consider various subsets of the total cohort. We examined all infant deaths and deaths by age (neonatal and postneonatal), by birth weight (normal and low [<2500 g]), and by specific causes within these categories. Special attention was given to sudden infant death syndrome (SIDS). For comparable modeling assumptions, the results for PM10 agreed with previously published estimates; however, the associations with PM10 were not specific to probable exposures or causes of death and were not robust to changes in the model and/or the locations considered. Significant negative mortality associations were found for SO4 2-. There was no indication of a role for outdoor PM2.5, but possible contributions from indoor air pollution sources cannot be ruled out, given higher SIDS rates in winter, in the north and west, and outside of large cities.

On exposure and response relationships for health effects associated with exposure to vehicular traffic.

This work examines various metrics and models that have been used to estimate long-term health effects of exposure to vehicular traffic. Such health impacts may include effects of air pollution due to emissions of combustion products and from vehicle or roadway wear, of noise, stress, or from socioeconomic effects associated with preferred residential locations. Both categorical and continuous exposure metrics are considered, typically for distances between residences and roadways, or for traffic density or intensity. It appears that continuous measures of exposure tend to yield lower risk estimates that are also more precise than categorical measures based on arbitrary criteria. The selection of appropriate exposure increments to characterize relative risks is also important in comparing pollutants and other agents. Confounding and surrogate variables are also important issues, since studies of traffic proximity or density cannot identify the specific agents related to traffic exposures that might be responsible for the various health endpoints that have been implicated. Studies based on ambient air quality measurements are necessarily restricted to species for which data are available, some of which may be serving as markers for the actual agents of harm. Studies based on modeled air quality are limited by the accuracy of mobile source emission inventories, which may not include poorly maintained (high emitting) vehicles. Additional exposure modeling errors may result from precision limitations of geocoding methods. Studies of the health effects of traffic are progressing from establishing the existence of relationships to describing them in more detail, but effective remedies or control strategies have generally not yet been proposed in the context of these epidemiological studies. Resolution of these dose–response uncertainties is important for the development of effective public health strategies for the future.

Daily Mortality in the Philadelphia Metropolitan Area and Size-Classified Particulate Matter

ABSTRACT Time-series of daily mortality data from May 1992 to September 1995 for various portions of the seven-county Philadelphia, PA, metropolitan area were analyzed in relation to weather and a variety of ambient air quality parameters. The air quality data included measurements of size-classified PM, SO4 2-, and H+ that had been collected by the Harvard School of Public Health, as well as routine air pollution monitoring data. Because the various pollutants of interest were measured at different locations within the metropolitan area, it was necessary to test for spatial sensitivity by comparing results for different combinations of locations. Estimates are presented for single pollutants and for multiple-pollutant models, including gaseous pollutants and mutually exclusive components of PM (PM2.5 and coarse particles, SO4 2- and non-SO4 2- portions of total suspended particulate [TSP] and PM10), measured on the day of death and the previous day. We concluded that associations between air quality and mortality were not limited to data collected in the same part of the metropolitan area; that is, mortality for one part may be associated with air quality data from another, not necessarily neighboring, part. Significant associations were found for a wide variety of gaseous and particulate pollutants, especially for peak O3. Using joint regressions on peak O3 with various other pollutants, we found that the combined responses were insensitive to the specific other pollutant selected. We saw no systematic differences according to particle size or chemistry. In general, the associations between daily mortality and air pollution depended on the pollutant or the PM metric, the type of collection filter used, and the location of sampling. Although peak O3 seemed to exhibit the most consistent mortality responses, this finding should be confirmed by analyzing separate seasons and other time periods.

Daily Mortality and Air Pollution in Atlanta: Two Years of Data from ARIES

Associations between daily mortality and air pollution were investigated in Fulton and DeKalb Counties, Georgia, for the 2-yr period beginning in August 1998, as part of the Aerosol Research and Inhalation Epidemiological Study (ARIES). Mortality data were obtained directly from county offices of vital records. Air quality data were obtained from a dedicated research site in central Atlanta; 15 separate air quality indicators (AQIs) were selected from the 70 particulate and gaseous air quality parameters archived in the ARIES ambient air quality database. Daily meteorological parameters, comprising 24-h average temperatures and dewpoints, were obtained from Atlantas Hartsfield International Airport. Effects were estimated using Poisson regression with daily deaths as the response variable and time, meteorology, AQI, and days of the week as predictor variables. AQI variables entered the model in a linear fashion, while all other continuous predictor variables were smoothed via natural cubic splines using the generalized linear model (GLM) framework in S-PLUS. Knots were spaced either quarterly, monthly, or biweekly for temporal smoothing. A default model using monthly knots and AQIs averaged for lags 0 and 1 was postulated, with other models considered in sensitivity analyses. Lags up to 5 days were considered, and multipollutant models were evaluated, taking care to avoid overlapping (and thus collinear) AQIs. For this reason, PM2.5 was partitioned into its three major constituents: SO42−, carbon (EC + 1.4 OC), and the remainder; sulfate was assumed to be (NH4)2SO4 for this purpose. Initial AQI screening was based on all-cause (ICD-9 codes < 800) mortality for those aged 65 and over. For the (apparently) most important pollutants—PM2.5 and its 3 major constituents, coarse PM mass [CM], 1-h maximum CO, 8-h maximum O3—we investigated 15 mortality categories in detail. (The 15 categories result from three age groups [all ages, < 65, 65+] and five cause-of-death groups [all disease causes, cardiovascular, respiratory, cancer, and other “remainder” disease causes]). The GLM model outputs that were considered included mean AQI effects and their standard errors, and two indicators of relative model performance (deviance and deviance adjusted for the number of observations and model parameters). The latter indicator was considered to account for variations in the number of observations created by varying amounts of missing AQI data, which were not imputed. The single-AQI screening regressions on all-cause 65+ mortality show that CO, NO2, PM2.5, CM, SO2, and O3, followed by EC and OC, consistently have the best model fits, after adjusting for the number of observations. Their relative rankings, however, vary according to the smoothing knots used, and there is no correspondence between mean AQI effect and overall model fit. (Other regression runs often show that the best model fits are obtained with no AQI in the model.) There is no correspondence between mean AQI effect and statistical significance or between mean effect and serial correlation. There is a highly significant (.001 level) relationship between overall model fit and serial correlation; the best fitting models have the most frequent knot spacing and the most negative serial correlation. The regression analyses by cause of death find elderly circulatory deaths to be consistently associated with CO for all models.

The effect of air pollution upon mortality: a consideration of distributed lag models.

Abstract The relationship between daily mortality and urban air pollution is considered in a regression context. Several distributed lag models are examined in order to discover a possible delayed mortality response. The results of those models with geometrically declining lag coefficients are consistent with those of the other lag models. Adjustment is made for serial correlation, and the consequences of measurement error are explored. Application of the various models to four to six different time periods allows an examination of the consistency of these models. Results for all time periods indicate associations between daily mortality and air pollution values. The distributed lag models suggest that any lagged effect is negligible in comparison with the immediate response.