STATEMENT OF JONATHAN LEVY,

HARVARD SCHOOL OF PUBLIC HEALTH

The materials included in this written testimony provide support for my oral presentation regarding the implications of the PM_2.5 health literature for power plant risk calculations.

In my oral testimony, I focused on the evidence for mortality risks from particulate matter, given the important role that mortality has played in past benefits assessments of air pollution controls (such as the EPA’s benefit-cost analysis of the Clean Air Act). I also asserted that there are three crucial questions that must be answered to quantify the public health benefits of power plant pollution controls:

1. Is there a threshold below which no health effects of PM_2.5 are found, and if so, where is that threshold?

2. Do all types of particulate matter have similar health impacts, or are some particles more toxic than others?

3. Would alternative control strategies have significant impacts on the magnitude or distribution of particulate matter health impacts?

Within this document, I address these three questions in greater detail, summarizing the key studies that inform my answers to these questions. Along with this summary document, I have included copies of selected documents that provide even more information about the core issues.

Is there a threshold?

An initial point that is important to emphasize is that this is not the same question as whether PM_2.5 concentrations are above or below National Ambient Air Quality Standards. Quoting directly from the US EPA in their Final Rule for the PM_2.5 NAAQS, “The Act does not require the Administrator to establish a primary NAAQS at a zero-risk level, but rather at a level that reduces risk sufficiently so as to protect public health with an adequate margin of safety.” (p. 3). The question is therefore whether the health literature provides evidence of a threshold above current ambient concentrations.

First considering time-series studies, which evaluate the effects of changes in daily concentrations of PM on daily mortality risks, two major studies illustrate the nature of the literature (Daniels et al., 2000; Schwartz et al., 2002). The first of these studies used information from the National Morbidity, Mortality, and Air Pollution Study (NMMAPS) to evaluate whether a threshold existed for short-term exposure to PM₁₀, either for total mortality or cardiovascular/respiratory mortality. The authors showed that for daily changes in PM₁₀, linear models without thresholds were most appropriate for total or cardiovascular/respiratory mortality. When considered probabilistically, the threshold for total daily mortality appeared to be definitely below 30 mg/m³ and was most likely below 15 mg/m³. The second study used information from the Six Cities Study, considering daily mortality risks from changes in PM_2.5 concentrations. As with the NMMAPS study, the authors concluded that a linear no-threshold model was most appropriate.

Thresholds have also been examined in the cohort mortality literature, with the most recent evidence provided in the follow-up to the American Cancer Society cohort study (Pope et al., 2002). Within the range of concentrations in the study, there was no evidence of a threshold, and the relationship appeared approximately linear. The lowest concentrations reported in the study (averaged across the study period) were less than 10 mg/m³.

Thus, the epidemiological literature shows no evidence of a threshold for mortality risks at current ambient concentrations. Although this may be counter-intuitive, given the normal assumptions regarding thresholds for non-carcinogens, this relationship is biologically plausible. As explained in Schwartz et al. (2002), individuals likely have thresholds, but if those thresholds differ widely across individuals based on numerous factors, then the distribution of thresholds across the population should be normally distributed. This would imply that the population concentration-response curve would approximately a cumulative normal curve, which is linear at low concentrations. In other words, if current particle levels were below the mortality threshold for most (but not all) people, then linearity with no population threshold would be expected.

Do all types of particles have similar health effects?

Prior to evaluating the literature, it is important to frame this question appropriately. Because most of the epidemiological evidence available to date has been based on monitors that measure total particulate mass in various size ranges, it has been established that particulate matter concentrations are associated with mortality and morbidity. However, little information has been available about the relative toxicity of different types of particles, so the default assumption has been that all pollutants have equal toxicity.

While that is unlikely to be the case, to deviate from this assumption, one must be able to quantify relative toxicities and defend these quantifications. Explicitly, for the case of power plant emissions, we would need to be able to estimate how toxic a sulfate or nitrate particle is relative to average particles. Clearly, this is not a question that can be answered with certainty, nor is it one that will be definitively solved in the near term.

Focusing on epidemiological evidence, there are two types of studies available: studies that directly measured at least one of the constituents of interest (often sulfates) and studies that used statistical methods to try to determine source-specific differential toxicity. Each approach has advantages and limitations, and each can add to the body of evidence.

In cohort mortality investigations, the primary evidence arises through the analysis of sulfates along with particulate mass in various size fractions. In the Harvard Six Cities Study (Dockey et al., 1993) and American Cancer Society study (Pope et al., 2002), long-term exposure to sulfates displayed a consistent positive association with premature mortality. In the latter publication, as well as in the Health Effects Institute reanalysis (Krewski et al., 2000), the authors concluded that some combination of PM_2.5, sulfates, and possibly SO₂ were associated with mortality. In a third cohort study (McDonnell et al., 2000), sulfates were not statistically significant, although the central estimate for mortality for male nonsmokers from sulfates was between the values from the Six Cities and American Cancer Society studies.

In terms of the relative effect of sulfate versus general PM_2.5, our power plant risk assessment in Massachusetts (Levy and Spengler, 2002) found that impacts were greater if either the reported sulfate-mortality or SO₂-mortality relationship were applied rather than the PM_2.5-mortality relationship. Thus, the cohort mortality literature generally shows sulfate effects that are significant, with a concentration-response function slightly greater than general PM_2.5 effects and no direct information available on other particulate species.

In the time-series literature, much of the speciation data come from studies looking at sulfates. These studies have generally found positive associations, as indicated in the following figure (taken from the second external review draft of the Particulate Matter Criteria Document).

As indicated in the above figure, there has been preliminary evidence available from the supersite in Atlanta, which measures numerous chemical species (Klemm and Mason, 2000). This study found no statistically significant relationship for any particulate measures using one year of time-series data. Per unit concentration, the central estimates were higher for elemental carbon and sulfates than for PM_2.5 as a whole, with lower central estimates for organic carbon and nitrates (although no values were statistically significant). In interpreting these results, it is important to realize that lack of statistical significance could be related to either a lack of an effect or a lack of statistical power to find an effect, given a relatively small sample size. If we look at the body of sulfate time-series studies in the above figure, we see that the Klemm and Mason findings in fact have a central estimate in line with much of the previous literature, but with substantially wider confidence intervals. Once this study is completed, it should be combined with other available studies to determine a best estimate for the time-series relationship between sulfates and mortality, taking into account relevant site and population characteristics (e.g., air conditioning prevalence) to generalize to the US at large.

Looking at studies of source-specific effects, a study by Laden and colleagues (2000) applied statistical methods to elemental data from the Six Cities study to determine source-specific particulate matter factors. Across all six cities, they found that the motor vehicle and coal factors had statistically significant effects on premature mortality, with the motor vehicle factor approximately a factor of three greater than the coal factor (per unit concentration). A crustal factor was not significant. Although the confidence intervals were wide, there was some evidence that cardiovascular deaths were more closely related to motor vehicle particles and respiratory deaths were more closely related to coal-derived particles.

Additional factor-analytic studies include Ozkaynak and Thurston (1987) and Mar (2000). In the former study, based on cross-sectional mortality data across the US, particles from industrial sources and coal combustion had greater coefficients than those from motor vehicles or crustal sources. In the latter study in Phoenix, combustion-related pollutants (from motor vehicles and vegetative sources) and secondary sulfates were associated with cardiovascular mortality. A soil-related factor had a negative association with mortality. Thus, the findings from factor analytic studies appear to show lower toxicity of crustal particles, with significant effects from motor vehicles, power plants, and other combustion sources. However, the studies do not provide consistent quantitative evidence for greater toxicity of one combustion source category over another.

In conclusion, while it is difficult to assign specific differential toxicities to different particle types, it does appear likely that combustion particles are more toxic than crustal particles. In studies looking at both sulfates and PM, the effect per unit concentration of sulfates is generally slightly higher, but the relatively small difference and the lack of substantial toxicological evidence makes a conclusion of equal toxicity reasonable as a central estimate for risk calculations.

What are the magnitude and distribution of PM health effects from power plants?

First considering the distributional question, it is clear that the impacts from a single power plant will vary spatially (since the concentrations associated with that plant will not be uniform across the country). The crucial question is whether populations near the power plants are disproportionately at risk or whether the impacts occur at longer distances, as this will influence the formulation of optimal control strategies.

In our initial power plant analysis in Massachusetts (Levy and Spengler, 2002), we concluded that the answer to this question depended largely on how the question was framed. We distinguished between individual risk (the mortality risk to a given individual at a given location) and aggregate risk (the total public health impact associated with the facility). When we look at individual risk, the maximum occurs relatively close to the power plants – approximately 25-40 km away for the two plants studied in Massachusetts. However, because of the long-range transport of particulate matter and the number of people who are impacted at long range, most of the aggregate risk occurs at long range – more than half beyond 100 km, as illustrated in the figure below from Levy and Spengler (2002). Thus, we can conclude that individuals who live closer to a power plant are more impacted by that plant than individuals living further away, but that local populations contribute a relatively small fraction of aggregate risk.

Although this captures broad distributional trends related to distance from the source, another aspect of the distributional question is whether selected demographic groups are disproportionately affected by power plant air pollution. If this is the case, then a greater amount of the population risk occurs in a smaller set of individuals, which increases the importance of considering distributional issues.

In a recent power plant risk assessment focused on the Washington, DC area (Levy et al., in press), we identified high-risk populations for selected health outcomes and evaluated the implications for the magnitude and distribution of health benefits. For the case of premature mortality, we considered the influence of educational attainment on mortality risk, as documented in Pope et al. (2002). We concluded that if the observational evidence from the American Cancer Society cohort study were correct, then more than half of the health benefits accrued among the 25% of the population with less than high school education. Furthermore, we showed that small-scale spatial variations were significantly influenced by the incorporation of population patterns, as illustrated by Figure 4 from Levy et al. (in press), a portion of which is reproduced on the following page.

Finally, I turn to the question of estimating the magnitude of health impacts from power plant emissions. Making this estimate requires a multi-step process. First, the emissions of SO₂ and NOx are quantified (given the structure of multi-pollutant regulations and the focus on particulate matter impacts). Second, atmospheric dispersion models are used to evaluate the influence of these emissions on concentrations of PM_2.5 across a large region. These concentration changes are then combined with epidemiological evidence to quantify the public health implications.

As an example of this sort of analysis, Abt Associates (2000) used an economic model to estimate the distribution of SO₂ and NOx emissions from the power sector given proposed emission controls, applied two atmospheric dispersion models to evaluate the national PM_2.5 implications of these proposals, and linked the concentration changes with health evidence, including the mortality risk derived from the American Cancer Society cohort study. They concluded that current power plant emissions were associated with approximately 30,000 premature deaths per year, with a 75% reduction scenario yielding benefits of approximately 19,000 fewer premature deaths per year.

A critical question is whether these estimates represent reasonable central estimates or are biased in either direction. In a recent investigation (Levy, 2002), I reviewed the methodology used by Abt Associates in a similar analysis, focusing on the question of bias. I considered separately the atmospheric model and the health evidence. I concluded that the atmospheric model yielded health impact estimates that were essentially identical to those using a different model, and that the concentration-response function chosen for premature mortality was a reasonable central estimate. Thus, it appeared equally likely that the Abt Associates methodology yielded an underestimate as an overestimate, making their findings a reasonable foundation for policy decisions.

A similar methodology was used by the EPA to estimate the benefits of alternative power plant control policies. For example, the EPA estimated that the Clear Skies act would reduce premature deaths by about 12,000 per year, by combining the results of atmospheric models and epidemiological studies (see www.epa.gov/clearskies). Similarly, an earlier straw proposal from the EPA (which had more stringent caps on both SO₂ and NOx) was associated with a reduction of 19,000 premature deaths per year. Again, this was based on a similar methodology as used by Abt Associates, implying that the estimate is a reasonable central estimate.

From the above discussion, it is qualitatively clear that increased reductions of SO₂ and NOx are likely to lead to increased public health benefits. While the above public health estimates are clearly uncertain, they appear just as likely to be underestimates as overestimates. Thus, it is reasonable to assume that the Clear Skies Act would provide substantial public health benefits, but that the EPA straw proposal (which is similar to the Clean Power Act) would increase those benefits by perhaps 7,000 fewer premature deaths per year. This implies that choices between status quo emissions, the Clear Skies Act, the Clean Power Act, and other alternative formulations should depend on a comparison of the incremental costs and benefits of increased stringency.

Attached documents

I have attached a subset of the studies cited above, which either expand on the arguments in this testimony or are not yet publicly available. Attached documents include:

- Levy J. Evaluation of Methodology in “Particulate-Related Health Impacts of Eight Electric Utility Systems”. Prepared for Rockefeller Family Fund, June 2002.

- Levy JI, Greco SL, Spengler JD. The importance of population susceptibility for air pollution risk assessment: A case study of power plants near Washington, DC. Environ Health Perspect, in press, December 2002 expected.

- Levy JI, Spengler JD. Modeling the benefits of power plant emission controls in Massachusetts. J Air Waste Manage Assoc 52: 5-18 (2002).

- Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, Thurston GD. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 287:1132-1141 (2002).

- Schwartz J, Laden F, Zanobetti A. The concentration-response relationship between PM_2.5 and daily deaths. Environ Health Perspect 110: 1025-1029 (2002).

References

Abt Associates, ICF Consulting, E.H. Pechan Associates. The Particulate-Related Health Benefits of Reducing Power Plant Emissions. Available: http://www.cleartheair.org/fact/mortality/mortalityabt.pdf, 2000.

Daniels MJ, Dominici F, Samet JM, Zeger SL. Estimating particulate matter-mortality dose-response curves and threshold levels: an analysis of daily time-series for the 20 largest US cities. Am J Epidemiol 52:397-406 (2000).

Dockery DW, Pope CA III, Xu X, Spengler JD, Ware JH, Fay ME, Ferris BG. Jr., Speizer FE. An association between air pollution and mortality in six U.S. cities. N Engl J Med 329: 1753-1759 (1993).

Klemm RJ, Mason RM Jr. Aerosol research and inhalation epidemiological study (ARIES): air quality and daily mortality statistical modeling—interim results. J Air Waste Manage Assoc 50: 1433-1439 (2000).

Krewski D, Burnett RT, Goldberg MS, Hoover K, Siemiatycki J, Jerrett M, Abrahamowicz M, White WH. Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality. Cambridge, MA: Health Effects Institute (2000).

Laden F, Neas LM, Dockery DW, Schwartz J. Association of fine particulate matter from different sources with daily mortality in six U.S. cities. Environ Health Perspect 108: 941-947 (2000).

Levy J. Evaluation of Methodology in “Particulate-Related Health Impacts of Eight Electric Utility Systems”. Prepared for Rockefeller Family Fund, June 2002.

Levy JI, Greco SL, Spengler JD. The importance of population susceptibility for air pollution risk assessment: A case study of power plants near Washington, DC. Environ Health Perspect, in press, December 2002 expected.

Levy JI, Spengler JD. Modeling the benefits of power plant emission controls in Massachusetts. J Air Waste Manage Assoc 52: 5-18 (2002).

Mar TF, Norris GA, Koenig JQ, Larson TV. Associations between air pollution and mortality in Phoenix, 1995-1997. Environ Health Perspect 108: 347-353 (2000).

McDonnell WF, Nishino-Ishikawa N, Petersen FF, Chen LH, Abbey DE. Relationships of mortality with the fine and coarse fractions of long-term ambient PM₁₀ concentrations in nonsmokers. J Exposure Anal Environ Epidemiol 10: 427-436 (2000).

Özkaynak H, Thurston GD. Associations between 1980 U.S. mortality rates and alternative measures of airborne particle concentration. Risk Anal. 7: 449-461 (1987).

Pope CA III, Burnett RT, Thun MJ, Calle EE, Krewski D, Ito K, Thurston GD. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. JAMA 287:1132-1141 (2002).

Schwartz J, Laden F, Zanobetti A. The concentration-response relationship between PM_2.5 and daily deaths. Environ Health Perspect 110: 1025-1029 (2002).