The Health Effects Institute

Statement
Particulate Air Pollution and Daily Mortality

Analysis of the Effects of Weather and Multiple Air Pollutants

The Phase I.B Report of the Particle Epidemiology Evaluation Project

This statement, prepared by the Health Effects Institute and approved by its Board of Directors, is a summary of Phase I.B of the Particle Epidemiology Evaluation Project sponsored by HEI from 1994 to 1996. Dr. Jonathan M. Samet and colleagues of the Johns Hopkins University School of Hygiene and Public Health in Baltimore, MD and the University of Delaware in Newark, conducted the study, Air Pollution, Weather, and Mortality in Philadelphia, 1973-1988. The Special Report, available from the Health Effects Institute [Publications Order Form] contains the Investigators' Report, a Commentary on the study prepared by the Oversight Committee of the Particle Epidemiology Evaluation Project, and Comments from the Health Review Committee.

BACKGROUND

The Health Effects Institute began the Particle Epidemiology Evaluation Project in 1994 to evaluate the emerging epidemiologic evidence of a relation between particulate air pollution and daily mortality. A growing number of studies reported that daily mortality rates rose in association with levels of particulate air pollution below the current National Ambient Air Quality Standard (NAAQS) for particulate matter. Some investigators interpreted their findings as indicating that particulate air pollution has a causal effect on increases in daily mortality. Critics disputed this interpretation, and questioned whether other scientists could replicate the results. Some critics asserted that the reported associations might be artifacts of the particular statistical methods used, or might be due to incomplete statistical adjustment for the effects of weather factors or other air pollutants.

This controversy had important regulatory implications because the U.S. Environmental Protection Agency (EPA) was in the midst of reviewing the scientific evidence in support of the NAAQS for particulate matter, and the studies of particulate air pollution and daily mortality had become a central focus in that review. In this context, the EPA and interested parties from private industry and nongovernmental organizations encouraged HEI to assess the validity of key epidemiologic studies and to clarify the implications of their findings.

Recognizing the tight time constraints imposed on the EPA for its review, HEI's Oversight Committee for the project separated it into two phases; Phase I was subsequently carried out in two parts by Drs. Jonathan M. Samet and Scott L. Zeger and their colleagues at the Johns Hopkins University School of Hygiene and Public Health. Briefly, in Phase I.A, the investigators (1) reconstructed from original sources the data set for Philadelphia used in earlier studies and confirmed previous numerical results from analyzing these data; (2) developed an analytic approach (including new statistical methods) based on the Philadelphia data set; and (3) applied this approach to data sets for six locations: Philadelphia; Utah Valley; St. Louis, MO; Eastern Tennessee; Birmingham, AL; and Santa Clara County, CA.

In Phase I.B, the subject of this Special Report, the investigators (1) compared approaches for controlling the effects of weather variables when analyzing the connection between air pollution and daily mortality, primarily focusing on Synoptic Weather Categories, an approach newly proposed by Dr. Laurence S. Kalkstein of the University of Delaware; and (2) evaluated the association between particulate air pollution and daily mortality in the Philadelphia metropolitan area using statistical models that included data for five pollutants regulated under the Clean Air Act Amendments of 1990 (referred to as criteria pollutants).

In Phase II, which began in December of 1996, the investigators will extend the Phase I analyses to address scientific questions with important public health and regulatory implications, such as the impact of air pollution on years of life lost, and will develop statistical methods to address the impact of errors in measuring exposure in daily time-series studies.

APPROACH

To meet the Phase I.B objectives, the investigators focused on two questions. First, could different analytic approaches used to control the effects of weather variables change the association observed between air pollution and daily mortality rates? Second, how might the association between particulate air pollution and daily mortality change if all criteria pollutants found in a given locale were analyzed simultaneously in the same statistical model?

To answer the first question, the investigators chose three different approaches to characterize weather factors in statistical models, and applied them to the Philadelphia data set from 1973 through 1980 to control the influence of weather variables in their analyses. The Philadelphia data included ambient air pollution exposure levels and daily mortality rates for relevant (cardiac and pulmonary) causes of death. The first approach reduced a relatively large number of meteorologic variables to a smaller set of summary (or synoptic) categories that describe daily weather patterns in a particular locale (referred to as the Synoptic Categories approach). The second approach used statistical regression methods to describe the temporal relation between mortality and the previous day's absolute and dew-point temperatures; a mathematical equation for that temporal relation was then incorporated into a model of daily mortality (referred to as the LOESS model; this approach also had been used in the Phase I.A analyses). The third approach incorporated into the statistical model of daily mortality the current and previous day's absolute and dew-point temperatures, plus indicator terms for hot and cold days (referred to as the S-D model because it had been used by Drs. Joel Schwartz and Douglas W. Dockery of the Harvard School of Public Health in their earlier analysis of daily mortality in Philadelphia).

To answer the second question, the investigators estimated what each individual air pollutant contributed to daily mortality in Philadelphia, a location for which data were available over a 15-year period (from 1974 through 1988) for all criteria air pollutants, and where levels of these pollutants were correlated with one another to varing degrees throughout the period of the study. The investigators developed statistical models to control weather and long-term trends in mortality, and then incorporated air pollution data into the models. These models revealed important features of the data set, such as differences in long-term mortality trends for people older than 65 years and those younger, and nonlinear patterns in the magnitude and timing of weather's effects on mortality.

Next, they estimated air pollution's effects on daily mortality, first for each of five pollutants separately (total suspended particles, sulfur dioxide, nitrogen dioxide, ozone, and carbon monoxide), then in pairs, and finally all pollutants together. Where possible, model building was guided by current knowledge of toxicology, pathophysiology, and clinical medicine, and by a statistical criterion of how well a given model described the observed daily mortality. However, the model that included all five pollutants and was "best" according to statistical criteria produced estimates for the effects of some pollutants that did not conform to prior expectations based on toxicologic knowledge. For example, nitrogen dioxide was estimated to have a statistically negative (that is, beneficial) effect on the mortality rate (meaning the rate fell, which could be misinterpreted to mean that nitrogen dioxide could somehow have a beneficial effect on human health). A result such as this most likely reflects the model's attempts to deal with inaccuracies in the measurement of an air pollutant that is closely correlated with nitrogen dioxide, and should not imply a biologic interpretation.

RESULTS AND INTERPRETATIONS

Phase I.B successfully explored the sensitivity of the air pollution–mortality relationship to different analytic approaches for statistically controlling the effects of weather on mortality, and the consequences of incorporating multiple pollutants when modeling the air pollution–mortality relationship in a single locale.

In response to the first question, the investigators found that neither the Synoptic Categories approach nor the S-D and LOESS approaches used in Phase I.A analyses substantially altered the air pollution–mortality association observed in Philadelphia. In addition, they could find no meaningful pattern of variation in the relative risks calculated under different weather conditions, regardless of the approach used to characterize weather.

In response to the second question, the multipollutant analyses confirmed two important findings. First, certain indexes of air pollution are associated with increases in daily mortality rates: ambient concentrations of total suspended particles and sulfur dioxide (which confirmed the findings of Phase I.A), and ozone (which had not been included in the Phase I.A analyses). Second, in Philadelphia during this time period, and given the correlation among a number of pollutants, no single criteria air pollutant could account completely for the observed increases in daily mortality.

However, the results of the current analyses also differ from the earlier findings. First, although the same range of concentrations of total suspended particles was still associated with a small relative increase in the daily mortality rate, the magnitude of this increase was not as large as that seen in earlier studies. This change is most likely attributable to more flexibly modeling long-term mortality trends, and particularly to allowing the long-term trends to vary for different age groups. Second, the season-specific associations between pollutants and daily mortality, observed in the Phase I.A analyses and by other analysts, were not found in Phase I.B; this may be the result of improved models for weather and long-term mortality trends. Nonetheless, seasonal variation in the effects of air pollutants remains plausible and requires further study.

The analyses of the Philadelphia data conducted during Phases I.A and I.B have limitations that are common to most studies of air pollution and daily mortality. First, we would like to be able to interpret these results in terms of how personal exposure to air pollution might affect an individual's risk of mortality. However, these analyses rely on measurements of ambient pollutant levels from centrally located monitors, which may not be accurate assessments of an individual's actual exposure. The inaccuracies in exposure measurements may introduce errors in mortality risk estimates that are difficult to quantify. Second, the amount of life-shortening that underlies the increased mortality rates associated with air pollution has not been estimated, and the statistical approaches to do so are only now being developed. Third, in Phases I.A and I.B, the detailed analyses were restricted to data from a single locale and time period in which major changes in both air pollution and mortality occurred, and during which changes in ambient levels of various air pollutants were correlated over time. It is difficult to draw conclusions about the effects of individual or combined air pollutants from such data alone. Such conclusions might be based more reliably on the results of a nationwide study using data bases on mortality and air pollution from a large number of sites with different air pollution profiles. Phase II of this project, the National Morbidity, Mortality, and Air Pollution study, will address the limitations described above.

The current findings corroborate previous results, reported by these investigators and others, that the association between air pollution and daily mortality in Philadelphia from 1973 through 1988 is not explained by other known factors associated with mortality, nor by variations in analytic approaches to adjust for weather factors. Although individual air pollutants (total suspended particles, sulfur dioxide, and ozone) are associated with increased daily mortality in these data, the broader association of pollution with daily mortality in this city cannot be reliably attributed to any single criteria air pollutant. Given the limitations discussed above, it is not possible to establish the extent to which particulate air pollution by itself is responsible for the widely observed association between mortality and air pollution in Philadelphia, but we can conclude that it appears to play a role.

CODE: PEEPI.B

Home | About HEI | What's New? | Newsletter | HEI International | Publications
Research | Funding | Meetings | Contact HEI | Links | Search | Site Map
Copyright © 2004 Health Effects Institute. Please send comments to webmaster@healtheffects.org