The Health Effects Institute
Statement
Particulate Air Pollution and Daily Mortality
Replication and Validation of Selected Studies
The Phase I.A 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.A of the Particle Epidemiology Evaluation Project sponsored by HEI from 1994 to 1995. Dr. Jonathan M. Samet and colleagues of the Johns Hopkins University School of Public Health in Baltimore, MD, conducted the study, The Association of Mortality and Particulate Air Pollution. The Special Report, available from the Health Effects Institute [Publications Order Form] contains the New Analysts' report, a Commentary on the study prepared by the Institute's Oversight Committee of the Particle Epidemiology Evaluation Project, and Comments from Original Investigators.
BACKGROUND
Scientists, regulators, and the general public are increasingly interested in the adverse health effects of particulate air pollution. This interest is motivated largely by recent epidemiologic studies that have examined both acute and longer-term effects of exposure to particulate air pollution in different cities in the United States and elsewhere in the world. Many of these studies indicate a positive association between particulate air pollution and the rate of daily mortality. The studies suggest that daily rates of respiratory illness and of mortality from respiratory and cardiovascular diseases increase in response to levels of particulate air pollution below the current National Ambient Air Quality Standard (NAAQS) for particulate matter. Critics of these studies have questioned the validity of the data sets used and the statistical techniques applied to them; have noted that studies of other cities have not found similar effects; have independently analyzed some of the same cities and come to different conclusions about the relative effects of particulate air pollution, other forms of pollution, and weather; and have questioned the biologic plausibility of the reported association between particulate air pollution and higher illness and mortality rates.
Both the studies and the critiques have engendered substantial controversy and prompted legal and regulatory action. If the original studies prove to be right, they may suggest that the current NAAQS needs to be more stringent, which would have substantial implications for those whose activities are the sources of particulate air pollution (e.g., factories, steel mills, and motor vehicles). As a result of a law suit brought by the American Lung Association, the U.S. Environmental Protection Agency (EPA) is under court order to conduct an accelerated review of the science underlying the current NAAQS for particulate matter, and the ability of the standard to protect public health with an adequate margin of safety. The scientific review must be completed in August of 1995, and a final decision on the adequacy of the standard must be reached by the end of January 1997.
In this scientific and regulatory context, the Health Effects Institute was encouraged by the EPA and a number of interested industry and environmental groups to undertake an impartial, rigorous, and timely evaluation of key portions of the epidemiologic data in order to assess the validity of the earlier results and to refine our understanding of their meaning.
APPROACH
An expert Oversight Committee, appointed by the HEI Board of Directors in July 1994, conducted an international competition and selected as New Analysts the team of Drs. Jonathan M. Samet and Scott L. Zeger of Johns Hopkins University School of Public Health to conduct this reanalysis and evaluation. Several of the original investigators (Drs. Douglas W. Dockery and Joel Schwartz of Harvard School of Public Health, C. Arden Pope III of Brigham Young University, and David Fairley of the San Francisco Bay Area Air Quality Management District) agreed to cooperate in the project and to provide their data to Samet and Zeger.
After broad public consultation in the autumn of 1994, the Oversight Committee approved an overall scope of the project:
- to evaluate through reanalysis of selected data sets the emerging evidence for an association between particulate air pollution and increased mortality from all causes and from cardiovascular and respiratory diseases;
- to evaluate the biologic plausibility and public health significance of a causal association between particulate air pollution and mortality;
- to examine how sensitive the observed associations are to key analytic assumptions and features of the data sets; and
- to the extent that a consistent, strong association is found, to evaluate the expected changes in mortality under various plausible scenarios of exposure reduction.
Recognizing the tight time constraints imposed on the EPA for its review, the New Analysts and Oversight Committee separated the project into two phases. The first phase was designed to accomplish several specific objectives in time to inform the EPA review before its August completion date:
- to reconstruct from original sources the data set for Philadelphia and attempt to confirm previous numerical results derived from analyses of these data (Philadelphia is one of the largest cities originally studied, and has been the subject of a number of analyses by others);
- to develop an analytic strategy (or model) based on the experience gained with the Philadelphia data set;
- to apply this analytic strategy to individual data sets for six locations: Philadelphia; Utah Valley; St. Louis, MO; Eastern Tennessee; Birmingham, AL; and Santa Clara County, CA;
- to complete initial sensitivity analyses of one data set to evaluate the consequences of using alternative modeling strategies and different approaches to accommodate confounding or modifying variables, such as weather; and
- to develop an analytic plan for Phase II of the project to investigate these questions in greater depth.
The New Analysts have now achieved these Phase I.A objectives, and the results of their work are presented in this report. In addition to the stated objectives, they were able to reanalyze the data of an additional investigator, Dr. Suresh Moolgavkar of the University of Washington, who has analyzed a Philadelphia data set but reached different conclusions than the Original Investigators. The New Analysts' work was reviewed by an independent Review Panel selected by the HEI Health Review Committee, and revised in response to comments from the Panel and the Oversight Committee. This report presents all of the analyses performed, a Commentary by the HEI Oversight Committee, and a response from some of the Original Investigators.
RESULTS AND IMPLICATIONS
The New Analysts were able to construct, from the same publicly available data bases used by the Original Investigators, a Philadelphia data set that closely matched that of Schwartz and Dockery. The New Analysts found a close correspondence between their own plots of raw weather and mortality data and the same plots based on the data provided by the Original Investigators. Although the New Analysts' mortality data agreed more closely with those of the Original Investigators than with the data provided by Moolgavkar, and although the New Analysts' air pollution data differed slightly from the data of both Moolgavkar and of Schwartz and Dockery, the discrepancies do not appear to be important.
When the New Analysts analyzed their data set for Philadelphia using various approaches and models, the results were in reasonable agreement with the results reported by both teams of earlier investigators. When the analyses were restricted to the same set of years for which the Moolgavkar data overlapped the data of Schwartz and Dockery, and when the effects of sulfur dioxide (SO2) were taken into account, the overall results were quite compatible, with only minor differences in the magnitude and statistical significance of the association between total suspended particulates (TSP) and increased mortality.
The most important finding of the Phase I.A reanalyses is that the new and independent analytic model developed by the New Analysts produced numerical results from the data sets for all six locations that closely agree with and, in general, confirm those of the Original Investigators.
In their sensitivity analyses of the Philadelphia data set, the New Analysts found that the original numerical results cannot be readily explained by arbitrary choices in creating statistical models. Consistent results were produced even when specifications in the analytic model, such as the form of weather variables and pollution measurements, were handled with a variety of techniques, when different assumptions regarding the lag time between the day of measured pollution and the day of mortality were evaluated, and when the models included an assumption that mortality on one day depends on mortality on previous days.
Both the New Analysts and the Original Investigators found an association in Philadelphia between TSP and increased daily mortality that persisted when SO2 was accounted for. At the same time, the New Analysts confirmed the observations of Moolgavkar in finding that mortality increased in relation to TSP and SO2 differently during the four seasons of the year, with apparently stronger relative increases associated with TSP in summer and fall, and stronger relative increases associated with SO2 in winter and spring. Other sensitivity analyses, which allowed for complex relationships among pollutants, seasons, and mortality, suggested that the Original Investigators' results simplify a more complex set of relations. Therefore, one must be cautious in assuming that the effects of air pollution on mortality in the Philadelphia data set can be attributed to particles alone.
One limitation in drawing generalized inferences from the Phase I.A analyses is the limited selection of cities for study. Although the data sets spanned different regions of the country and provided contrasts in characteristics of air pollution, seasons, and weather, four of the six data sets were relatively small, which limits the ability to evaluate further the effects of these factors and others, such as age and mortality due to specific causes.
A number of questions regarding the association between increased mortality and particulate air pollution remain to be addressed; chief among them is the question of whether the excess deaths observed on high-pollution days are causing a higher long-term excess death rate, or minor alterations in the timing of deaths that were imminent anyway due to life-threatening illness. Other remaining questions include (1) how should variables be selected and analytic models be constructed to control for the confounding effects of weather on mortality; (2) how do the changes in mortality rate differ with the seasons; (3) how do other pollutants impact the association in question; and (4) how do errors in measuring exposure and other variables influence the results? These and other questions will be addressed in Phase II of this project, and in a new program of research now underway under the auspices of the Health Effects Institute.
CONCLUSION
Given that the Phase I.A analyses agreed with those of the Original Investigators, it is reasonable to conclude that, in these six data sets, daily mortality from all causes combined, and from cardiovascular and respiratory causes in particular, increased as levels of particulate air pollution indexes increased. However, the detailed analyses of the Philadelphia data set support a call for caution in assuming that this association represents an independent effect of particles alone. Also, as the New Analysts point out in their report, Phase I.A analyses were not intended to determine if the observed association of particulate air pollution with mortality is causal. Phase II of the project will greatly extend the current analyses and, with the results of Phase I.A reported here, will provide a more thorough assessment of the evidence on air pollution and mortality.