My name is Thomas B. Starr. I am a Principal with ENVIRON International Corporation, a consulting firm headquartered in Arlington, Virginia, that specializes in health and environmental science issues related to chemical exposures, pharmaceuticals, medical devices, and food products, pesticides, and contaminants. My own consulting activities focus on the development and use of effective methods for incorporating scientific knowledge of toxic mechanisms into the quantitative risk assessment process. A brief biographical sketch is attached (Appendix A).
The comments I offer today are drawn principally from two separate consulting projects in which we have performed a critical examination of the scientific evidence for potentially causal associations between particulate matter (PM) exposure and adverse human health effects. In the first project, undertaken on behalf of the American Petroleum Institute, three world-renowned epidemiologists, Drs. Raymond Greenberg, Provost and Vice-President for Academic Affairs, Medical University of South Carolina, Jack Mandel, Chairman, Department of Environmental and Occupational Health, School of Public Health, University of Minnesota, and Harris Pastides, Chairman, Department of Epidemiology and Biostatistics, from the School of Public Health at the University of Massachusetts, were brought together as an Expert Panel to independently and objectively assess the quality of the epidemiologic evidence for associations between PM exposure and increased human morbidity and mortality.
In the second project, undertaken on behalf of Kennecott Corporation, an ENVIRON colleague, Dr. Larisa Rudenko, and I also evaluated the case for such causal associations, and, in addition, assessed the credibility of health benefits that EPA has projected would accrue from implementation of the proposed new PM standards.
The final reports from these projects were submitted to EPA and are included along with my oral testimony for your information. My remarks today briefly summarize their findings; I refer you to the full reports for additional details.
First, the issue of causality, or whether the effects observed are truly cause by the exposure to PM, specifically PM-2.5, or some other component of air pollution or lifestyle. In assessing whether the results from epidemiologic studies support the existence of a causal relationship between exposure and disease, criteria developed initially by Bradford Hill (1965) are often applied. These include the strength, consistency, coherence, specificity, and temporality of the reported association. Although not explicitly stated, a presumption exists that the validity of the association has been established prior to consideration of these criteria. What this means is that the estimates of the association's strength have been shown to be free of significant biases and not significantly confounded. The Expert Panel of epidemiologists and our independent review both concluded that the studies of PM and disease do not satisfy these conditions; they have inadequately addressed potential biases and they have failed to resolve satisfactorily the issue of confounding.
Even if the issue of validity were to be set aside, the Hill criteria would not be met. The reported associations are extremely weak and vacillate between positive and negative based on the specific regression model that has been used to characterize the dose-response relationship; as copollutants are introduced into the analyses, apparently positive associations attenuate in strength, often to non-significance. Indeed, based on the criterion of strength of association, it is difficult to imagine a weaker case for causality than that posed by the data on PM and mortality or morbidity.
Furthermore, the results of the studies are not actually as consistent as they might at first appear. For example, different exposure measures (e.g., mean daily level, maximum daily level, or some lagged estimate) have been associated with different endpoints (e.g., respiratory diseases, cardiovascular diseases, or total deaths). Also, temporal relationships between exposure variables and disease occurrence are not the same across studies, with lag times varying from concurrent day to several days earlier.
In addition, a critically important component of coherence, namely, dose-response, is, at best, weakly established in only a few studies. In virtually all of the epidemiologic studies of PM, exposure levels have not been not based on personal dosimetry, but rather on stationary samplers located in specific geographic areas. Individual subjects were thus assigned *community-wide* measures of exposure, rather than individual measures. The lack of personal exposure measures limits the ability to conclude that any individual death is linked to air pollution per se. In fact, there is a large body of data indicating that community sampler measurements rarely provide good estimates of individual exposures.
Even if a causal association were in fact to exist between PM exposure and disease occurrence at the individual level, such "ecological" exposure estimates would likely misrepresent the association's true strength. Equally important, the shape of the underlying dose-response relationship would also likely be significantly distorted by ecologic analyses, with sharp threshold-like curves being smoothed into more nearly linear curves by exposure misclassification.
Another major challenge to the case for causality relates to the nature of PM exposure, which invariably occurs in combination with exposure to other air pollutants such as ozone, carbon monoxide, SO2, H2SO4, metals, and volatile organics. Because this mixture*s composition varies according to source, season, time of day, weather conditions, and geographic region, and because PM is itself a complex and highly variable mixture, it has been virtually impossible to disentangle the potential adverse health effects of PM, or a specific PM fraction, such as PM-2.5, from those potentially attributable to other confounding copollutants.
The question of whether the coarse and/or fine particulate components of air pollution are causally related to adverse human health effects is one of great importance. If there is a causal relationship, identification and establishment of a safe and acceptable level of ambient particulate matter will be a decision with enormous consequences. However, the severe methodological limitations of existing studies prevent a conclusive judgment about the causality of associations between PM exposure and adverse health effects at the present time. EPA's proposal for new PM standards is premature.
There is an obvious need for new epidemiologic studies that collect data at the individual subject level. Carefully designed case˙2Dcontrol studies can also be effective. It is especially important that future study designs be related to clearly articulated theories about the specific mechanistic pathways through which particulate air pollution may affect human health. To serve as a basis for regulatory decision˙2Dmaking, future epidemiologic studies will be most useful if they inform us about the specific manner in which individual air pollution constituents might affect human health. The current epidemiologic literature falls well short of this goal.
The stated purpose for USEPA's proposed new PM standards is to:
"...provide increased protection against a wide range of PM-related health effects, including premature mortality and increased hospital admissions and emergency room visits (primarily in the elderly and individuals with cardiopulmonary disease); increased respiratory symptoms and disease (in children and individuals with cardiopulmonary disease such as asthma); decreased lung function (particularly in children and individuals with asthma); and alterations in lung tissue and structure and in respiratory tract defense mechanisms." (Fed Reg 61:65638)
How confident can we be that the proposed new PM standards will in fact lead to increased human health protection? The quantitative risk assessment conducted for EPA by Abt Associates, Inc. attempts to quantify the uncertainty inherent in the estimated health benefits from the new standards. This assessment is very thorough in its identification of many weaknesses in the underlying PM and health effects data, remarkably frank about its necessary reliance on numerous unproven assumptions, and surprisingly even-handed in its demonstrations, via multiple sensitivity and uncertainty analyses, that the health benefits projected from the proposed standards might well be greatly exaggerated.
Significant limitations of EPA's benefit projections either noted in the Abt Associates, Inc. risk assessment are in our critique of it and include the following:
Because correlation is not causation, the projections have had to assume causation; thus, future reductions in specific PM levels need not necessarily result in any material health benefits. This has not been acknowledged explicitly.
EPA's failure to account for the potential health effects due to simultaneous exposure to PM, other pollutants, and related weather variables almost certainly leads to substantial overstatements of both the strength and statistical significance of the apparent associations specifically with PM exposure. This issue of confounding has been explored only to a very limited extent, yet EPA has concluded that its benefit estimates are robust to the inclusion or exclusion of individual co-pollutants. This conclusion is at variance with the findings of several reanalyses that considered multiple confounding variables simultaneously. The discrepancy is almost certainly due to the fact that EPA's sensitivity analyses considered only "one-at-a-time" additions of individual co-pollutants instead of real-world multiple exposures. Thus, the true benefits that result from compliance with the proposed new PM standards may well be completely negligible.
The benefit projections assume log-linear relationships between PM exposure above natural background levels and various adverse health outcomes. Because most days of the year have low to mid-range levels of PM, the estimated health benefit over an entire calendar year of daily PM exposures is dominated by the contribution from the many days with low to moderate levels of PM. This is precisely the exposure range for which the empirically determined log-linear dose-response relationships are most uncertain. The assumption of a log-linear no-threshold dose-response relationship is not presently scientifically justified; threshold-like alternatives can not be ruled out.
EPA's sensitivity analysis using different cut points (i.e, thresholds) demonstrates the enormous impact that thresholds can have on the projected benefits from proposed new standards. High thresholds imply negligible health benefits. Nevertheless, health benefits estimated with threshold-like dose-response relationships play only a secondary role in EPA's benefits assessment. They should instead be considered at least on an equal footing with the benefits estimated with log-linear models.
EPA's regression models presume implicitly that the independent variables are known without error. Yet actual PM exposure levels are very poorly characterized and highly uncertain. EPA has acknowledged that little regional monitoring data, and virtually no personal exposure data, are available for PM-2.5 at the present time. Furthermore, recent studies have shown that only weak correlations exist between individual personal exposures and PM measures recorded by regional monitoring stations. Uncertainty about the true values of these variables, or errors in their measurement, leads to a serious "errors in variables" problem that can only be resolved with further prospective study involving adequate simultaneous measurements of both individual PM exposures and region-wide measures of air quality.
Faced with such great uncertainty in the estimated magnitude of potential health impacts of the proposed new standards, it seems far more reasonable for EPA to initiate additional data collection and analysis activities on the health effects potentially associated with various PM fractions rather than rush to promulgate and implement new standards that could well make things worse rather than better.
That completes my oral testimony. Thank you for your attention. I would be happy to answer any questions.
THOMAS B. STARR trained in theoretical physics at Hamilton College and the University of Wisconsin-Madison, receiving his Ph.D. in 1971. Following National Science Foundation postdoctoral and faculty appointments in the Institute for Environmental Studies at Wisconsin, he joined the staff of the Chemical Industry Institute of Toxicology in 1981, first as a senior scientist in the Department of Epidemiology, and then in 1987 as Director of CIIT's Program on Risk Assessment. In 1989, he joined ENVIRON International Corporation as a principal in the Health Sciences Division. His research interests have focused on means for explicitly incorporating knowledge of toxic mechanisms into the quantitative risk assessment process, and improving epidemiologic methods for assessing effects of chemical exposure on worker health. He has published over 80 scientific papers and abstracts, and given hundreds of scientific presentations.
Dr. Starr holds an adjunct faculty appointment in the Department of Environmental Sciences and Engineering in the School of Public Health at the University of North Carolina-Chapel Hill. He has been appointed to numerous advisory posts, including the Halogenated Organics Subcommittee of the U.S. Environmental Protection Agency's Science Advisory Board, the North Carolina Academy of Sciences Air Toxics Panel, and the North Carolina Environmental Management Commission Ad Hoc Committee for Air Toxics. Currently, he serves on the Methylene Chloride Risk Characterization Science Committee, and the Secretary's Scientific Advisory Board on Toxic Air Pollutants for the North Carolina Department of Environmental Health and Natural Resources. He has testified before OSHA, EPA, and other regulatory agencies regarding human health risks posed by various chemical exposures, including those to 1,3-butadiene, cadmium, dioxin-like compounds, formaldehyde, lead, methylene chloride, and environmental tobacco smoke. He is also active in professional societies, including the American Statistical Association, the Society for Epidemiological Research, the Society for Risk Analysis, and the Society of Toxicology. In 1988-89 he served as the first President of the newly formed SOT Specialty Section on Risk Assessment, and in 1989-90 as President of the Research Triangle Chapter of the Society for Risk Analysis