Document YGvxjyvGdr2NLX48BGnvQyRVE

Texas Pl#lrP OllCJf FOUNDATION W1H0V29 PM 3: 03 Cvr7'.;?-?,F7 cXtJJi;V itC.ir.AnlAI November 9, 2017 Via Federal Express Mr. Scott Pruitt EPA Administrator United States Environmental Protection Agency EPA Headquarters Mail Code 1101A William Jefferson Clinton Building (North) 1200 Pennsylvania Avenue, N.W. Washington, D.C. 20460 RE: PETITION IN THE MATTER OF: NATIONAL AMBIENT AIR QUALITY STANDARDS FOR PARTICULATE MATTER Dear Administrator Pruitt: Enclosed please find our Administrative Petition respectfully requesting the Administrator to reconsider and make less stringent its current national ambient air quality standards ("NAAQS") for fine particulate matter ("PM2.5"), 78 Fed. Reg. 3086 (Jan. 15, 2013), because those standards are based upon faulty assumptions. Thank you in advance of your careful consideration of the enclosed Administrative Petition. Respectfully submitted. Theodore Hadzi-Antich Senior Counsel Center for the American Future Texas Public Policy Foundation Enclosure IM KODl C TION Pursuant to the Right to Petition Government Clause of the First Amendment of the United Sates Constitution,1 the Administrative Procedure Act,2 the Clean Air Act.2 and the United States Environmental Protection Agency's ("EPA's") implementing regulations. Petitioners file this Administrative Petition with EPA's Administrator and, lor the reasons set forth herein, respectfully request the Administrator to reconsider and make less stringent its current national ambient air quality standards ("NAAQS" or "standards") for Fine particulate matter ("PM2.5"), 78 Fed. Reg. 3086 (Jan. 15, 2013), because those standards are based upon faulty assumptions. Such reconsideration should he part of the current Five-year review cycle. INTEREST OK PETITIONERS Petitioner Delta Construction Company, Inc. ("Delta") is a California corporation engaged in the business of road construction, performing services such as road paving, reconstruction, shoulder widening, and fabric installation. After 73 years in business, Delta has been forced to close its doors and sell its assets mainly because of regulations governing particulate matter. Petitioner Dalton Trucking, Inc. ("Dalton") Dalton Trucking. Inc., is a California corporation engaged in the business of operating and leasing loaders, dozers, blades, and water "Congress shall make no law . . . abridging ... the right of the people ... to petition Government for a redress of grievances." U.S. Const, amend. I. The right to petition for redress of grievances is among the most precious of liberties safeguarded by the Bill of Rights. United Mine Workers ofAmerica, Dist. 12 v. Illinois State Bar Association, 389 U.S. 217, 222(1967). It shares the "preferred place" accorded in our system of government to the First Amendment freedoms and has a sanctity and sanction not permitting dubious intrusions. Thomas v. Collins, 323 U.S. 516, 530 (1945). "Any attempt to restrict those First Amendment liberties must be justified by clear public interest, threatened not doubtful or remotely, but by clear and present danger." Id. The Supreme Court has recognized that the right to petition is logically implicit in, and fundamental to, the very' idea of a republican form of government. United States v. Cruikshank. 92 U.S. (2 Otto) 542. 552 (1875). 5 U.S.C. Section 553(e). 42 U.S.C. Section 7401, el seq. (sometimes referred to here as the "CAA"). 2 trucks and performs specialized services in open top hulk transportation, lowbed, general freight on flatbcds and vans, as well as rail, intermodal. and 3PL services. Dalton is subject to the PM2.5 standards. Loggers Association of Northern California, Inc. ("LANC") is a nonprofit California trade association representing the interests of its members involved in the logging industry in Northern California. LANC members are subject to the PM2.5 standards Robinson Enterprises, Inc. ("Robinson") is a California corporation engaged in various businesses, including forest products and fuels. Robinson is a third-generation family-owned California corporation engaged in harvesting and transportation of forest products, petroleum products, and transportation of various commodities. It has suffered unnecessary financial hardship as a result of various burdensome regulatory requirements, including the PM2.5 standards. Nuckels Oil Co., Inc. dba Merit Oil Company ("Merit Oil Company") is a California corporation and is a petroleum jobber, wholesaler, and distributor. Merit Oil Company stores, transports, and wholesales a variety of petroleum products, including gasoline, diesel fuels, solvents, and kerosene, and operates a number of delivery trucks and is a family business that has operated in California for three generations. Merit oil Company is subject to the PM2.5 standards. Western States Trucking Association, Inc. ("WSTA") is a nonprofit California trade association representing the interests of over 1.000 members involved in a variety of business throughout California whose members own and operate on-road and nonroad vehicles, engines, and equipment, which are subject to the PM2.5 standards. EXECUTIVE SUMMARY On January 15. 2013. EPA published in the Federal Register a final rule reflecting the 3 results of its review of its PM NAAQS. 78 Fed. Reg. 30X6 (Jan. 15, 2013). The Final PM Rule, with an effective date of March 18. 2013, revised the level of the primary annual NAAQS for PM that is less than or equal to 2.5 microns in diameter ("PM2.5") to 12.0 micrograms per cubic meter ("pg/m3 ") and contained provisions for implementing this standard. In December 2014, EPA announced the initiation of the current periodic review of the air quality criteria for PM and of the PM2.5 and PM 10 NAAQS and issued a call for information in the Federal Register. 79 Fed. Reg. 71764 (December 3. 2014). "All of the PM NAAQS set to date are based on mass concentration and the assumption that all of the PMs in each size fraction are of equal toxicity on a mass basis. This assumption needs careful review in the current PM review cycle." Roger O. McClellan. Providing Contextfor Ambient Particulate Matter and Estimates of Attributable Mortality, Risk Analysis, 2016; 36(9): 1755-1765 at 1757. Recent scientific analyses that cast doubt on the evidence of a causal link between PM2.5 and mortality provide ample reason to reconsider the necessity of the current PM2.5 standards. Given this, the EPA Administrator should not only decline to tighten the primary annual or 24-hour NAAQS for PM2.5, but should consider making the standards less stringent. As set forth in more detail below, the PM NAAQS should be carefully reconsidered, and the Administrator should open the regulatory process to all interested stakeholders during the current five-year review, including the Petitioners. STATEMENT OF LAW AM) FACTS I. OVERVIKYV OF STATUTORY REQUIREMENTS The CAA requires the establishment and periodic revision of the PM NAAQS. Section 108 of the CAA (42 U.S.C. 7408) directs the EPA Administrator to identify and list "air pollutants" that, in his judgment, "cause or contribute to air pollution which may reasonably be 4 anticipated to endanger public health and welfare" and that the "presence [of which] ... in the ambient air results from numerous or diverse mobile or stationary sources." He is also required to issue air quality criteria for any air pollutants that are so listed. 42 U.S.C. 7408(a) & (b). These criteria are intended to "accurately reflect the latest scientific knowledge useful in indicating the kind and extent of identifiable effects on public health or welfare which may be expected from the presence of [a] pollutant in ambient air . . ." 42 U.S.C. 7408(b) (emphasis added). Section 109 (42 U.S.C. 7409) requires the Administrator to propose and issue "primary" (healthbased) and "secondary" (welfare-based) NAAQS for pollutants for which air quality criteria are issued under section 108. 42 U.S.C. 7409(a). Section 109(b)(1) defines NAAQS primary standards as those that "the attainment and maintenance of which in the judgment of the Administrator, based on such criteria and allowing an adequate margin of safety, are requisite to protect the public health." 42 U.S.C. 7409(b)(1). Section 109(b)(2) provides that secondary standards "shall specify a level of air quality the attainment and maintenance of which in the judgment of the Administrator, based on such criteria, is requisite to protect the public welfare from any known or anticipated adverse effects associated with the presence of [the] pollutant in the ambient air." 42 U.S.C. 7409(b)(2). Such welfare effects as defined in CAA section 302(h) include "effects on soils, water, crops, vegetation, man made materials, animals, wildlife, weather, visibility and climate, damage to and deterioration of property, and hazards to transportation, as well as effects on economic values and on personal comfort and well-being." 42 U.S.C. 7602(h). Section 109(d)( 1) of the CAA requires that, at five-year intervals, "the Administrator shall complete a thorough review of the criteria published under section 108 and the national ambient 5 air quality standards . . . and shall make such revisions in such criteria and standards and promulgate such new standards as may be appropriate 42 U.S.C. 7409(d)(1). Sections 109(d)(2)(A) and 109(d)(2)(B) of the Act require that an independent scientific review committee "shall complete a review of the criteria . . . and the national primary and secondary ambient air quality standards . .. and shall recommend to the Administrator any new .. . standards and revisions of existing criteria and standards as may be appropriate . . .42 U.S.C. 7409(d)(2). The Clean Air Scientific Advisory Committee ("CASAC'') conducts this review. CASAC has four responsibilities: (1) to advise the EPA Administrator of areas in which additional knowledge is required to assess the adequacy and basis of existing, new. or revised NAAQS; (2) to describe the research efforts necessary to provide the required additional information: (3) to advise the F.PA Administrator on the relative contribution to air pollution concentrations of natural and anthropogenic activity; and (4) to advise the F.PA Administrator of any adverse public health, welfare, social, economic, or energy effects which may result from various strategies for attainment and maintenance of the NAAQS. Section 109(d)(2)(C). The purpose of the primary standards is to provide an adequate margin of safety in order to take account of the inherent uncertainties due to inconclusive scientific information, and to provide a measure of protection against dangers not yet identified through research. Through the primary standards. EPA seeks to both prevent pollution levels that have been demonstrated to have adverse effects and to prevent lower pollutant levels that may pose unacceptable risks, even if those risks are. by their nature, not capable of being precisely identified as to their nature or degree. The decision on what approach to take is left to the EPA Administrator's policy judgment. The CAA does not require the Administrator to establish a primary NAAQS which eliminates all risk. 6 but rather to a level that reduces risk to the extent necessary to protect public health with an adequate margin of safety. See Lead Industries v. EPA, 647 F.2d at 1156 n.51 (D.C. Cir. 1980); Mississippi v. EPA, 723 F. 3d 246. 255, 262-63 (D.C. Cir. 2013). In establishing secondary standards, the Administrator must set standards that are neither more nor less stringent than necessary to protect public welfare from any known or anticipated adverse effects associated with the presence of PM. This policy judgment should rely on scientific evidence and analyses about the effects of PM on public welfare, as well as unquantifiable judgments about how to manage uncertainty. The CAA does not require secondary standards be set to eliminate all adverse effects on welfare. The EPA's task in setting both primary and secondary standards is to establish standards that are neither more nor less stringent than necessary, and it may not consider the costs of implementing the standards, attainability, or technological feasibility. See generally Whitman v. American /rucking Associations, 531 U.S. 457, 465-472, 475-76 (2001); American Petroleum Institute v. Castle. 665 F. 2d 1176. 1185 (D.C. Cir. 1981). 11. GENERAL SCOPE OF THE CURRENT NAAQS REVIEW In December 2014. EPA announced the initiation of the current periodic review of the air quality criteria for PM and of the PM2.5 and P\110 NAAQS. 79 Fed. Reg. 1716*1 (December 3. 2014). The multi-step review process lead to the release of the Final Integrated Review Plan for the National Ambient Air Quality Standards for Particulate Matter ("IRP") in December 2016. With regard to scope, the current review of the PM NAAQS is focused on the primary and secondary NAAQS for PM2.5 (fine particles) and PM 10 (coarse particles). The current primary and secondary PM2.5 standards are meant to protect against the health and welfare effects, respectively, that have been associated with short-term (i.e., hours up to one month) or long-term 7 (i.e., one month to years) exposures to fine particles. The primary and secondary PM10 standards are meant to protect against the effects associated with exposures to coarse particles. Important aspects of the current review include EPA's assessment of the health and welfare effects that have been associated with short- or long-term exposures to P\1 based on size fractionated P\1 mass, with a particular focus on the PM2.5 and PM 10-2.5 size fractions. In addition, as in the most recent review, EPA will assess the available scientific evidence for health or welfare effects associated with additional size fractions (e.g., ultrafme particles) and with particular PM components or groups of components, sources, or environments (e.g., urban and non-urban environments ). Based on the available scientific information, EPA is considering the extent to wrhich the current PM2.5 and PM 10 standards are requisite to protect public health and welfare, within the meaning of section 109(b) of the CAA. To the extent the available information calls into question the protection afforded by one or more of the existing PM standards, EPA has indicated tha it plans to consider potential alternatives that could be supported by the available scientific evidence and, as available, exposure-/risk-bascd information, in terms of the basic elements of the NAAQS (indicator, averaging time, form, level). ARGUMENT I. THE UNCERTAINTY OF THE SCIENCE REGARDING AMBIENT PARTICULATE MATTER CAUSING ADVERSE HEALTH EFFECTS IS GREATER THAN EPA HAS ADMITTED In the United States and some other industrial democracies, where people and their governments tend to be risk averse, legislatures, courts, and administrative entities usually create a presumption favoring more safety rather than less. The definitions of risk in law are often vague ("reasonable certainty of no harm" or "adequate margin of safety") and are likely to encourage an unrealistic belief that risks can be minimized or even eliminated altogether." - Donald Kennedy, Editor-in-Chief, .Science 309: 2137 (30 September 2005) 8 Roger O. McClellan addresses (he scientific evidence relating to NAAQS for PM2.5 in his recent works Role of Science anti Judgment in Setting National Ambient Air Quality' Standards: How Lmv Is Low Enough?. 5 Air QUALITY, ATMOSPHLRL & HEALTH 243 (2012) (questioning the unbiased nature of HPA NAAQS determinations) (hereinafter. l'Role") (attached as Exhibit A), and Providing Context for Ambient Particulate Matter and Estimates of Attributable Mortality, Risk Analysis, 2016; 36(9): 1755-1765 (specifically addressing the PM2.5 NAAQS) (hereinafter, *'Providing Context") (attached as Exhibit B). In Role, McClellan focuses on EPA's method of setting primary (health-based) NAAQS. Role at 243. The Clean Air Act in 1963 and its amendment in 1970 required "the listing of air pollutants that `may reasonably be anticipated to endanger public health and welfare."' Id. at 244. Subsequent amendments required reevaluation of the NAAQS in 1980 and ever) five years thereafter. Id. EPA also appointed an independent scientific committee called CASAC to conduct peer review for the NAAQS in 1977. Id. When creating a primary NAAQS. 42 U.S.C. 8 7409 allows the F.PA Administrator discretion to "address uncertainties associated with inconclusive scientific and technical information at the time the Standard is set" to establish an "adequate margin of safety." Id. at 245. Congress has also noted that sensitive populations, particularly those with respiratory problems who are regularly exposed to ambient air. should he accounted for. Id. Given these criteria, McClellan notes a problem with interpreting the Clean Air Act: though NAAQS are intended to mitigate risk, the Act is unclear about how' much mitigation satisfies the law. This may lead some groups to operate under the false assumption that risks from pollution in ambient air can be eliminated. Id. 9 McClellan discusses the politicized nature of such revision. For example, at its creation, the NAAQS for lead were "constrained and informed by the scientific information, but ultimately based on the policy judgment of a politically responsible decision-maker, the F.PA Administrator." Id. at 246. Earlier NAAQS were completed through informal rulemaking, which did not provide a sufficient basis for judicial review according to the United States Court of Appeals for the District of Columbia Circuit. Id. After that court struck down one of EPA's NAAQS, EPA developed a more rigorous method of documenting their decision-making process for NAAQS and making public their reasoning. Id. This reform, which was enacted subsequently by Congress in somewhat modified form in the Clean Air Act Amendments of 1977, Pub.L. No. 95-95, 305. 91 Stat. 685, sacrificed speed in rulemaking but improved transparency. McClellan notes with approval. Id. at 247. In 1997, EPA chose to set a separate PM2.5 standard for the first time. Prior to that time, PM 2.5 had been included under the standards for ambient particulate matter under 10 microns (PM 10). Id. Discussions surrounding the first PM2.5 NAAQS were "very contentious" as the scientists on the committee had "a range of views" so complex that it took a tabic to diagram them. Id. This disagreement wus magnified by the D.C. Circuit's decision in Am. Trucking Assn. v. U.S. EPA. 175 F.3d 1027 (D.C. Cir. 1999). That decision vacated the 1997 PM 10 standards largely because they included the PM2.5 standards. Further, they determined that while EPA's factors used to determine degrees of public health concern related to pollutants were "reasonable," EPA lacked any clear criterion for determining NAAQS. However, the EPA Administrator was not allowed to consider the cost of implementing NAAQS when setting them. Role at 247. 10 The Supreme Court affirmed the basic holding of Am. trucking two years later in Whitman v. Am. Trucking, 531 U.S. 457 (2(M)1). Writing for the majority. Justice Breyer clarified further that "109 does not require EPA to eliminate every healLh risk, however slight, at any economic cost, however great, to the point of `hurtling' industry over `the brink, of ruin."" Id. at 494. This sought to solve the problem posed by the Clean Air Act's risk-avoidance language: the EPA Administrator has flexibility to avoid setting standards that chill industry activity and determine "the acceptability of small risks to health." Id. Thus the EPA Administrator does not have to set NAAQS that aim at completely eliminating pollutants, as if such a thing were possible. Breyer's opinion allows the Administrator to make his determinations about what level of protection and risk is "adequate" based on his policy judgments when crafting primary and secondary NA AQS. McClellan states that a "paradigm shift" took place as the amount of scientific evidence regarding pollution's health effects grew. Role at 248. Originally, lacking human studies on the effects of pollution on health, scientists agreed that the lowest level at which pollution could be determined "statistically significant" in laboratory animal studies served as the highest level for the "adequate margin of safety." Id. at 248-49. (As an aside, in recent years, the wisdom of taking lab animal studies as determinative on this matter has been called into question, and EPA has introduced a factor in its NAAQS calculations that supposedly accounts for this discrepancy. Id. at 249.) This decision assumed that certain non-cancer health issues had a linear exposureresponse relationship to certain pollutants, an assumption which McClellan discusses further in his analysis. Id. McClellan also notes the folly of EIPA's initial inclination to "identify levels where an increase in effects is observed and then set the Standard at a lower level." Id. Eventually, EPA began linking their standards to pollutant concentrations averaged over multiple years. Id. This shift in the statistical forms underlying NAAQS produces challenges when certain studies fail to 11 provide metrics for their data that would aid F.PA in averaging. This difficulty "results in extremely stringent Standards that at best are only very loosely related to the underlying data." Id. McClellan points out that EPA's assumptions about appropriate background levels for certain pollutants, combined with ongoing acceptance of a possibly flawed statistical model for NAAQS. has hamstrung the agency's ability make NAAQS that reflect reality. Id. at 250. EPA has assumed that its practice of categorizing concentrations of pollutants above the NAAQS in a linear manner, rather than determining "whether there is a threshold level below which ihe coefficient for excess risk does or does not hold." Id. EPA's insistence on this point has extended to estimating adverse health attributable to each pollutant "down to background concentrations." Id. While admitting that he w'as originally in favor of this approach, McClellan did not expect that advocates of such quantification would take their measurements as "highly accurate projections . . . sometimes without any indication of uncertainty." Id. Due to these statistical challenges, McClellan concludes that "decisions on the selection of specific levels and averaging times for the NAAQS are policy judgments properly reserved to the Administrator informed by the available scientific knowledge." Id. at 249. In other words, the implications of Breyer s opinion in Whitman extend to the statistical modeling underlying the NAAQS determination. EPA's unreasonable decision to adopt linear modeling, in contravention of Whitman's directive that the Clean Air Act recognizes the need for policy judgment within its "adequate margin for safety" parameter is the paradigm shift McClellan previously mentioned. McClellan then discusses the PM2.5 indicator. He participated in initial CASAC discussions on the first PM2.5 NAAQS in 1997. He noted that the committee members in large part wished to create a NAAQS that "w ould mandate the monitoring of PM2.5." but also expressed reservations about setting the NAAQS too stringently given the "absence of convincing data on 12 PM2.5." Id. at 25J. He states that the Administrator's initial annual NAAQS on PM2.5 was too stringent and "very precautionary," while the 24-hour NAAQS was less so. Id. CASAC's revision of this standard in 2005 recommended a tightening of both standards, with significant pressure to provide unanimous approval. McClellan believed this tightening "was not a scientific decision. but rather a matter of policy judgment that should be left to the discretion of the Administrator." Id. He and another colleague did not join CAS AC's recommendation. The Administrator tightened the 24-hour NAAQS while leaving the annual one where it was. Id. McClellan makes it clear that it is "not appropriate for CASAC to recommend a bright line upper bound on the NAAQS." because that recommendation involves policy judgment beyond scientific analysis. Id. at 252. While the Administrator is authorized to make decisions about what constitutes appropriate risk and incorporate it into his standard-setting, the CASAC's narrow' job is to provide the Administrator with scientific information that will factor into his final decision. Id. McClellan next addresses the call for "sound science" to inform the Administrator's standard-setting decisions. He agrees wholeheartedly, and supports in principle the efforts of advocacy groups and NGOs to synthesize and submit helpful data for EPA's NAAQS process. Id. at 254. However. McClellan heavily criticizes the inclination of some groups to hold certain data as "true" or "false" based on who funded the study that produced the data, and expresses concern about the implicit expectations that "sound science" can provide perfect NAAQS: Sound science does not in and of itself make for sound decisions.... [Sjcience alone cannot identify an acceptable level of health risk, since such levels inherently represent a policy judgment call. Sound science can only inform what are ultimately policy judgments or political decisions. This is especially the case for the setting of NAAQS. in the absence of a clearly defined threshold, which involve decisions as to acceptable health risks which are linked to the level (and form) of the Standard. Id. 13 McClellan concludes that while Whitman allows the Administrator to set NAAQS in a way that accounts for policy judgment, CASAC itself may not exercise the same judgment in making its recommendations. Instead. McClellan wants CASAC members to draw on their diverse expertise to interpret and distill the vast quantity of scientific data on pollutants. Id. at 255. Most notably, McClellan believes that the Administrator would greatly benefit from CASAC's input on "the multiple factors that influence morbidity and mortality from respiratory and cardiovascular disease, the major health outcomes for key criteria pollutants." Id. at 256. He reaffirms that if Administrators seek to use the CASAC's unwarranted offering of acceptable ranges as scientific cover for their own political judgments, such action would "transform the Clean Air Scientific Advisory Committee into a de facto Clean Air Standards Setting Committee," a result not intended by Congress in enacting the Clean Air Act. Id. Moving on to McClellan's 2016 paper, he specifically addresses PM2.5 NAAQS in light of new research, analyzing the extent to which PM2.5 may or may not contribute to increased mortality based on the new findings. Providing Context at 1755. McClellan takes time to summarize the methodology of each study. Two of the four considered studies incorporate alternative methods of measuring acceptable levels of PM2.5. rather than or in addition to the commonly accepted linear concentration-response modeling that McClellan criticized in his 2012 paper. Id. at 1756. In the following section. McClellan points out that in 2012, the Administrator revised the lightened the primary annual NAAQS for PM2.5 to 12pg/cubic m. The 24-hour standard held steady. Id. at 1757. McClellan notes that both of these standards "are based on mass concentration and the assumption that all of the PMs in each size fraction are of equal toxicity on 14 a mass basis." Id. Based on new evidence, McClellan suggests that "this assumption needs careful review in the current PM review cycle." Id. McClellan begins his examination of the relation between PM2.5 and mortality by referencing a major long-term study on the subject called the Harvard Six Cities Study. It measures "changes in ambient PM2.5 concentrations in ... six cities from the mid 1970s through 2009." Id. The study demonstrates a sizable and steady decline in ambient PM2.5. Id. at 1757 58. McClellan next notes that the crude and age-adjusted death rates have seen marked improvement in the same time frame. Id. at 1758. 1 le includes another table indicating the causes of death for the United States in 2010. Id. at 1759. This table lists heart diseases as the most common cause of death, followed closely by cancer. Chronic lower respiratory' diseases are a distant third. Id. Overall, "it is widely acknowledged today . . . that the regulatory programs grounded in the CAA have had widespread positive impact" in terms of improved air quality. Id. This brings up the obvious question of whether current air quality requires stricter primary' NAAQS for PM2.5. Such a question hinges on whether PM2.5 is still a significant cause of adverse health effects, which McClellan next examines. McClellan explains that F.PA has a five-level hierarchy (ranging from "causal relationship" to "not likely to be causal relationship") to classify the weight of evidence regarding the relation between a given pollutant and a health hazard. Id. at 1760. Notably, this level based system does not speak to whether current PM2.5 levels in the United States increase the incidence of adverse health effects "over and above baseline rates." Id. Even more seriously, this system does not establish whether any given ambient PM2.5 concentration has "a causal attributable effect on health outcomes," including an increase in mortality rates simpliciter. Id. 15 Many scientists incorrectly believe the conclusions of EPA's level-based system bears some sort of implication for ambient PM2.5 concentration measurements. Id. McClellan faults the authors of the four new studies his paper examines for making a related assumption. One examined study implies that the correlation between PM2.5 levels and excess risk of adverse health effects is reliable no matter the examined concentration and risk level - a proposal with which McClellan expresses reservations. Id. at 1760. He also questions why the studies failed to question the EPA Administrator's reasoning in lowering primary annual PM2.5 NAAQS so drastically in 2012. In that instance, the Administrator considered a limited range of data in available studies as reliable evidence of a causal relationship between long-term PM2.5 exposure and increased general death rates. Id. at 1761. This conclusion conflicts with the conclusion of all four researchers, who considered all data in their studies to be reliable. Id. Since data at all concentrations did not show an equal causal relationship between long-term PM2.5 exposure and increased all-cause mortality, this is a serious omission. The Administrator also entirely failed to take into account the Six Cities Study, because it had not released numbers for PM2.5 as recently as other studies. Id. at 1760. McClellan calls the contrast between the Administrator's judgments and the seeming conclusions of the most reliable recent studies on PM2.5 "a critical issue at the interface between scientific information and policy choices." Id. at 1761. McClellan criticizes the four studies at issue further, noting that even though the data does not necessarily support the conclusion that low concentrations of PM2.5 cause an increase in death rates, none of the studies discuss this fact. Id. "[Tlhe official assumption in the last EPA review that all PM2.5 is of equal toxicity on a mass basis," McClellan notes, is especially important in a modem context, when most PM results not from direct emissions but "secondary reactions and associated changes in the chemical and size composition of PM." Id. Very little 16 data that differentiates between directly emitted and secondarily derived PM exists. Such data is necessary to determine whether a mortality increase still correlates with both kinds of PM. and in what concentrations. Id. While one study has a more extensive discussion of causality than others, McClellan calls its assumptions "simplistic and . . . naive" for oversimplifying the way that outside stressors cause an increase in mortality. Id. He especially finds the study's skepticism about a PM2.5 range of exposure where no mortality risk exists "unjustified," especially since the authors' own methods of measurement require them to "control for all other risk factors potentially associated with the disease endpoint of concern." Id. at 1762. These risks are manifold and complex. In fact, McClellan reveals, there is "a growing body of evidence of a lack of influence of ambient PM2.5 concentrations on mortality." Id. In some states, like California, the risk of increased mortality associate with PM 2.5 has decreased to the point of non-demonstrability. Id. Moreover. "[i]t is well recognized by scientists and clinicians ... that none of the individual cases carry "markers" or any characteristics that allow PM2.5 attributable cases to be distinguished from cases that are attributable to a myriad of other causes." Id. Because deaths are only attributed to PM2.5 "on a statistical and population basis," we have no hard evidence of any mortality increase directly attributable to PM2.5. Id. The authors of the studies reviewed by McClIelan do not discuss whether more well-documented risks could contribute to or account for increases in mortality currently attributed to PM2.5. Id. Given the complexity of determining what risk factors contribute to any given death (and the variance of contribution depending on time, place, and exposure level), this omission is glaring. McClellan suggests that "an expanded presentation of results" incorporating the Six Cities Study and exposure-response measurements would be more informative to future decision 17 making about PM2.5 NAAQS. Id. at 1 763. He also suggests including baseline population anti mortality data to provide context for such determinations. Id. at 1764. Regarding the most current models and studies on PM2.5, McClellan concludes that their estimates are "more likely to overestimate than underestimate the true PM2.5 attributable mortality." Id. He also wonders whether the data on mortality attributable to certain PM2.5 concentrations have been skewed by the exposure of certain individuals born in or before the 1970s to PM2.5. Id. While he agrees that it is possible that improvements in air quality contributed to reduced mortality, "the impact of PM2.5 reductions is likely very small and difficult to tease out from the myriad of other factors that were likely involved'7 in this reduction, like widespread improvement in overall socioeconomic status. Id. McClellan is not the only scientist to question the evidence of a significant link between tine particulate matter and mortality rates. James E. Enstrom's paper. Fine Particulate Air Pollution and Total Mortality Among Elderly Californians, 1973-2002. INHALATION TOXICOLOGY 2005; 17:803-816. (attached as Exhibit C), found no relationship between levels of fine particulate matter (PM2.5) arid mortality. Enstrom's analysis used proportional hazards regression and, adjusting for age. sex, cigarette smoking, and other potential confounding variables, found that "[tjhcse epidemiologic results do not support a current relationship between fine particulate pollution and total mortality in elderly Californians, but they do not rule out a small effect, particularly before 1983." Id. at 803. Enstrom's research was based on 118.094 Californians enrolled in the American Cancer Society's first Cancer Prevention Study. "For the initial period, 1973-1982, asmall positive risk was found: RR [relative risk of death] was 1.04 < 1.01--1.07) for a 10-ug/m3 increase in PM2.5. For the subsequent period, 1983-2002, this risk was no longer present: RR was 1.00 (0.98 1.02). For the entire follow-up period, RR was 1.01 (0.99-1.03)." Id. 18 at 803. Similarly, Enstrom's recent paper, Fine Paniculate Matter and Total Mortality in Cancer Prevention Study Cohort Reanalysis, DOSE-RESPONSE: An INTERNATIONAL Journal JanuarvMarch 2017:1-12. (attached as Exhibit D), independently analyzed the findings in the 1982 American Cancer Society Cancer Prevention Study (CPS II), which had earlier found a positive relationship between PM2.5 and total mortality (and has been the basis for EPA's PM2.5 NAAQS levels). Enstrom used Cox proportional hazards regression on the original questionnaire data, examining results obtained from 292,277 participants in 85 counties with 1979-1983 EPA Inhalable Particulate Network PM2.5 measurements, as well as for 212.370 participants in the 50 counties used in the original 1995 analysis. The 1982 to 1988 relative risk (RR) of death from all causes and 95% confidence interval adjusted for age, sex. race, education, and smoking status was 1.023 (0.997-1.049) for a 10 mg/m3 increase in PM2.5 in 85 counties and 1.025 (0.990-1.061) in the 50 original counties. The fully adjusted RR was null in the western and eastern portions of the United States, including in areas with somewhat higher PM2.5 levels, particularly 5 Ohio Valley states and California. Enstrom concluded there was no significant relationship between PM2.5 and total mortality in the CPS 11 cohort was found w'hcn the best available PM2.5 data were used. Contrary to the original 1995 analysis's finding of a positive relationship by selective use of CPS II and PM2.5 data Enstrom found that the underlying data raises serious doubts about the CPS II epidemiologic evidence supporting the PM2.5 NAAQS. T here have also been relevant contributions to a recent issue of Risk Analysis. Anne Smith's paper illustrates the use of alternative approaches to calculating the expected benefits of reducing the NAAQS for PM2.5 from 15 to 12 /rg/m3. Anne E. Smith, Inconsistencies in Risk Analyses for Ambient Air Pollutant Regulations, Risk ANALYSIS, 2016; 36(9): 1737-1744 19 (attached as Exhibit E). Smith describes the inconsistency between the health risk analysis that EPA uses to support its NAAQS standards and in the Regulatory Impact Analyses (RIAs) related to each NAAQS rulemaking. Risk estimates are prepared during the process of setting the NAAQS level using statistical relationships between measured pollutant concentrations and effects on human health. The final risk estimates are not directly used to set the NAAQS level, but are incorporated into a rationale for the standard intended to show compliance with the statutory requirement that the primary NAAQS protect the public health with a "margin of safety." In a separate process, EPA relies on the same risk calculations to prepare estimates of the health benefits of the rule that are reported in its R1A for the standard. Although NAAQS rules and their RIAs are released simultaneously, the rationales used to set the NAAQS have become inconsistent with their RIAs' estimates of benefits, with very large fractions of RIAs' riskreduction estimates being attributed to populations living in areas that will already be attaining the respective NAAQS. Smith's paper explains the source of this inconsistency and provides a quantitative example based on the 2012 revision of the PM2.5 primary NAAQS. Smith shows that the total risk reduction estimate (avoided premature deaths in 2020) for two approaches. The first was the traditional approach used by EPA in developing RIAs. which assumes deaths arc avoided regardless of the ambient concentrations of PM2.5. The analysis in the RIA showed 456 avoided deaths with one concentration-response function using the American Cancer Society cohort and 1.034 avoided deaths using the concentration-response function from the Six Cities Study. Smith also gave lower estimates based on the rationale that EPA used in the latest revision of the NAAQS for PM2.5, with the number of residual avoidable deaths reduced to 21 --48, dependent on the concentration-response function used. "The result is that the RIA benefits are substantially 20 overstated compared to those that would more appropriately reflect the subjective weights expressed by EPA in its rationale for setting the standard at 12 //g/m3." hi. at 1741. Smith finds that a large majority of EPA's estimated health benefit from the 2012 PM2.5 NAAQS are attributable to reductions of PM2.5 in areas that were already in attainment of the PM2.5 NAAQS. RIA calculations of risk reduction in areas already attaining the new NAAQS are given the same weight (i.e., subjective confidence level) as projected benefits from areas that would be exceeding the NAAQS. These RIA calculations are based on assumptions that are inconsistent with the rationale for that NAAQS. This causes RIAs' benefits estimates to be much more substantial than estimates of the expected benefits that could be reasonably inferred from EPA's NAAQS-setting rationale. The overstatement becomes nearly 100% for co-benefits from criteria pollutants in RIAs for non-NAAQS regulations. Id. at 1742-43. Tony Cox was invited to comment on Smith's paper (as well as other papers). Cox points out the flaws in existing models purporting to predict how7 future changes in exposure to PM2.5 affect mortality. Louis Anthony Cox. Jr.. Rethinking the Meaning of Concentration-Response Functions and the Estimated Burden of Adverse Health Effects Attributed to Exposure Concentrations, Risk ANALYSIS, 2016: 36(9): 1770-1779 (attached us Exhibit F). Basically, the modeling choices affect the concentration-response relations, but equally good varying choices lead to conflicting conclusions regarding any adverse effect from a given level of PM2.5 on mortality. This means that currently available data has questionable efficacy in predicting how future changes in PM2.5 concentrations will affect human health. Id. at 1770-75. The reduced-form regression models used to attempt to establish associations between particular PM2.5 levels and mortality are flawed, but Cox believes that other methods of modeling risk, from simulation to causal Bayesian networks, could be more efficacious in determining 21 changes in responses from changes in exposure level. Id. at 1775-77. Given the flaws in the current data used by F.PA. and the possibility of more accurate models as outlined in Cox's paper, it would be irresponsible for EPA to lighten the PM2.5 NAAQS. The analyses of McClellan, Enstrom, Smith, and Cox provide more than enough reason to reconsider the necessity of the current extremely stringent PM2.5 standards. Given that the causal link between PM2.5 and mortality is tenuous at best and indemonstrable at worst, the EPA Administrator certainly should not tighten the primary annual or 24-hour NAAQS for PM2.5; rather, the Administrator should consider making the standards less stringent. II. EPA Has Inherent Authority to Reconsider the PM NAAQS "Agencies are free to change their existing policies as long as they provide a reasoned explanation for the change. When an agency changes its existing position, it need not always provide a more detailed justification than what would suffice for a new policy created on a blank slate. But the agency must at least display awareness that it is changing position and show that there are good reasons for the new policy." Encino Motorcars. LLC v. Navarro. 136 S. Cl. 2117. 2125-26 (2016) (internal citations and quotation marks omitted). Furthermore, "|a]n initial agency interpretation is not instantly carved in stone (although] reasoned decision-making ordinarily demands that an agency acknowledge and explain the reasons for a changed interpretation. But so long as an agency adequately explains the reasons for a reversal of policy, its new interpretation of a statute cannot be rejected simply because it is new." Verizon v. FCC, 740 F.3d 623. 636 (D.C. Cir. 2014). Accordingly. EPA is free to reconsider its prior decisions on PM NAAQS. As the Supreme Court has observed, "(ajgency inconsistency is not a basis for declining to analyze the agency's interpretation under the Chevron framework. . . . [I]n Chevron itself, this Court deferred to an agency interpretation that was a recent reversal of agency policy."). Nat'l Cable & Telecomm. Ass'n v. BrandX Internet Sen\v,, 545 U.S. 967,481 -82 (2005)(citing Chevron v. NRDC, 467 U.S. 837. 857-58 (1984). Accordingly, UPA may determine in connection with the current five-year review as a matter of policy that the PM NAAQS should he made less stringent in light of new scientificstudies relating to harm to human health from PM and the new Administrator's policy judgment in evaluating the uncertainties of the evidence. See Smiley v. Citibank (South Dakota), A'. A., 517 U.S. 735. 742 (1996) ("[regulatory] change is not invalidating. . . ."); Van Hollen. Jr. v. Fed. Election Conim'n, 811 F.3d 486. 496 (D.C. Cir. 2016) ("An agency `must consider varying interpretations and the wisdom of its policy on a continuing basis.'") (quoting Brand X, 545 U.S. at 981). Therefore, F.PA is free to revisit the PM NAAQS based upon the instant Administrative Petition. CONCLUSION For these reasons. Petitioners respectfully request that, during the current five-year review, the Administrator reconsider the NAAQS PM2.5 standards in light of the issues brought to his attention in this Administrative Petition. The Petitioners also request that they be provided with the opportunity to actively participate in the five-year review as stakeholders with a keen interest in the outcome. 23 DATED: November 9, 2017 Respectfully submitted. Theodore Hadzi-Antich tha@texaspolicv.com Ryan D. Walters rwaltcrs@texaspolicv.coin TEXAS PUBLIC POLICY FOUNDATION 901 Congress Avenue Austin. Texas 78701 Telephone: (512)472-2700 Facsimile: (512)472-2728 ATTORNEYS FOR PETITIONERS cc: Neomi Rao (via Federal Express) Administrator Office of Information and Regulatory Affairs Office of Management and Budget 725 17th Street, N.W. Washington, DC 20503 Ted Boling (via Federal Express) Acting Director President's Council on Environmental Quality 722 Jackson Place, N.W. Washington, IX' 20506 Sarah Dunham (via Federal Express) Acting Assistant Administrator Office of Air and Radiation Mail Code 6101A USEPA Headquarters William Jefferson Clinton Building 1200 Pennsylvania Ave., N.W . Washington DC 20460 24 EXHIBIT A Air Qual Atmos Health (2012) 5:243-258 DO I 10.1007/s 11869-0II -0147-2 Role ol science and judgment in setting national ambient air quality standards: how low is low enough? Roger O. McClellan Received: IS March 2011 /Accepted: 22 May 2011 /Published online: I June 2011 ' The Authors) 2011. This article is published with open access at Spnngerhnk com Abstract The Clean Air Act (CAA) requires listing as criteria air pollutants those pollutants that arise from multiple sources and are found across the United States. The original list included carbon monoxide, nitrogen oxides, sulfur oxides, particulate matter, photochemical oxidants (later regulated as ozone), and hydrocarbons. Later, the listing of hydrocarbons was revoked and lead was listed. The CAA requires the EPA Administrator to set National Ambient Air Quality Standards (NAAQS) for these pollutants using the latest scientific knowledge" at levels that, in the judgment of the Administrator, are "requisite to protect public health" while "allowing an adequate margin of safety" without considering the cost of implementing the NAAQS The NAAQS are set using scientific knowledge to inform the Administrator's policy judgments on each NAAQS. Recently, there has been increasing tension and debate over the role of scientific knowledge versus policy ludgment in the setting of NAAQS This paper reviews key elements of this debate drawing on the opinion of Supreme Court Justice Stephen Breyer, in Whitman v. Americ an Trucking Associations, to resolve the conundrum posed by the CAA language. I conclude that scientists should carefully distinguish between their interpretations of scientific knowledge on This paper was presented in the concluding plenary session on 'Regulatory and Policy Implications' at the American Association for Aerosol Research International Specialty Conference: Air Pollution and Health' Bridging the Gap from Sources to Health Outcomes." March 22-26, 2010, San Diego. CA. R O. McClellan (K) Toxicology and Human Health Risk Analysts, 13701 Quaking Aspen PI NE, Albuquerque. NM 87111, USA c-mail. roger.o.nicclellaxn@att.net specific pollutants and their personal preferences as to a given policy outcome (i.e.. specific level and form of the NAAQS), recognizing that these are policy judgments as to acceptable levels of risk if the science does not identify a threshold level below which there are no identifiable health risks. These policy judgments are exclusively delegated by the CAA to the EPA Administrator who needs to articulate the basis for their policy judgments on the level and form of the NAAQS and associated level of acceptable risk Keywords Clean Air Act Criteria pollutants Ozone Paniculate matter Policy Risk Regulations Introduction In thi s paper, I briefly review key aspects of the Clean Air Act (I L>70) with regard to the setting of National Ambient Air Quality Standards (NAAQS) for criteria pollutants noting various landmark decisions. I address the primary or health-based Standards and do not consider the secondary or welfare-based Standards, although the core concepts arc also relevant to the setting of the secondary Standards I highlight actions of the last two EPA Administrators (Stephen Johnson and Lisa Jackson) and the Clean Air Scientific Advisory Committee (CASAC) related to the setting of NAAQS for particulate matter and ozone that serve to illustrate the growing tension and debate over the role of scientific knowledge and policy judgments in the setting of NAAQS 1 conclude with recommendations tor the role of CASAC in synthesizing and interpreting the science on criteria pollutants and offering scientific advice that informs the EPA Administrator's policy judgments on acceptable health risks that, in turn, are linked to the level and statistical fonu of the NAAQS primary Standard. C Springer 244 Air Qua, Atm.* Health (2012) 5:243 258 The Clean Air Act The C lean Air Act (CAA), initially passed in 1963, is the principal national statute in the United States concerned with air quality. The original CAA (19631 directed the then Department of Health, Education and Welfare (HEW) to prepare, '"compile and publish criteria on the effects of air pollutants," hence the identification of "criteria pollutants''' and `'criteria documents" summarizing the scientific knowl edge on certain air pollutants arising from multiple sources and found across the United States as a basis for Standard setting. The National Air Pollution Control Administration (NAPCA) within HEW was assigned responsibility for administering the CAA. When the U S. Environmental Protection Agency (EPA) was created in 1970. responsibil ity for administering the CAA was transferred from NAPCA to the new agency. Bachmann (2007) provides an in-depth review of the evolution of Air Quality Management in the United States from 1900 through 2006, with emphasis on the NAAQS, for those readers interested in an in-depth coverage of the topic. John Bachmann prepared his historical review soon after he retired from liPA's Office of Air Quality Planning and Standards where he had a central role for more than three decades in the setting of NAAQS for all the criteria pollutants. Readers interested m legal details of the CAA will find the summary of Martineau and Novello (2004) useful. In 1970, amendments to the CAA (1970) were passed that required ihe listing of air pollutants that "may reasonably be anticipated to endanger public health and welfare" and to issue air quality' criteria for them. These air quality criteria are to "accurately' teflect the latest scientific knowledge useful in indicating the kind and extent of all identifiable effects on public health or welfare which may be expected from the presence of [a] pollutant in the ambient air. in varying quantities." The pollutants originally designated as "entena pollutants" because of their ubiquitous distribution and potential to endanger health were photochemical oxidants (later regulated as ozone), particulate matter (later regulated as total sus pended particulates, then as PM10, and PM?.?), carbon monoxide, sulfur oxides (regulated as sulfur dioxide), nitrogen oxides (regulated as N02h and non-methane hydrocarbons (later dropped as a criteria pollutant) The fiPA (1971) established NAAQS for these pollutants, soon after the Agency' was created, using existing scientific documentation, i e . entena As I will discuss below, the EPA later added lead as a criteria pollutant with legal prodding front the National Resources Defense Council- In 1977, several key amendments were made to the CAA (1977) Concern about slow action of the EPA in preparing entena documents and reassessing NAAQSs prompted a legislated requirement that the NAAQSs be reevaluated not later than January 1. 1980. and at 5-year interv als thereafter. Reevaluation was not intended to automatically result in changes in the NAAQSs for a pollutant; rather, reevaluation was intended to ensure that the scientific database was reviewed and that the NAAQSs were consistent with current knowledge. To my knowledge, ibis requirement for mandatory review every 5 years is unique to ihe setting of the NAAQS in the United States. Indeed, I know of no other stabile calling tor an updating of the science and reconsideration of the Standard every 3 years. Peer review of the earliest criteria documents prepared by the EPA was carried out by various committees of the agency's Science Advisory Board as I will discuss later. A 1977 amendment to the CAA institutionalized the peer-review process for the NAAQS (CAA 1977). The amendment requires the EPA Administrator to appoint an independent scientific committee, composed of seven members, including at least one member ct the National Academy of Sciences, one physician, and one person representing state air pollution control agencies to advise the Administrator on the science informing the policy judgments made in setting the NAAQS. The EPA has implemented this provision of the CAA by appointing a Committee, which designated itself as the CASAC The CASAC is directly responsible to the EPA Administrator, although it functions administratively as one of the standing committees of the EPA Science Advisory Board. Traditionally, the requirement for one CASAC member to be a member of the National Academy of Sciences has been broadly interpreted to also include membership in eithct (he National Academy of Engineering or the Institute ol Medicine. To complement the expertise of regular members of the CASAC, consultants with specialized expertise usually have been added to the review panels for specific pollutants. The CAA was amended again in 1990 (CAA 199(>) Although major changes were made in the CAA with these amendments, especially with regard to the regulation of hazardous air pollutants, there were no changes in the fundamental approach to dealing with the setting of NAAQS for criteria pollutants. However, there were changes in the CAA that have had major impact on the regulation of emissions of PM and precursors especially from large power plants. National Ambient Air Quality Standards Section 109 of the CAA (1970) directs the Administrator to propose and promulgate "primary" and "secondary" NAAQSs for criteria pollutants identified under Section 108. The primary Standards are to be set to protect public health; secondary Standards are to be set to protect the Cl Springer Air Qua! Atmos Health (2012) 5:24.v-25f< 245 public welfare such as effects on soils, water, crops, visibility, and deterioration of property In this paper. 1 focus on the use of scientific knowledge arid judgment in the setting of die primary Standards. However, the issues discussed arc also broadly applicable to the setting of secondary Standards. Section 109(b)(1) defines apntnary NAAQS as one that ``the attainment and maintenance of which in the judgment of the Administrator, based on the criteria and allowing an adequate margin of safety, is requisite to protect the public health " The margin of safety, as interpreted by the EPA, is intended to address uncertainties associated with inconclu sive scientific and technical information at the time the Standard is set and to account for hazards that research has not yet identified. The primary Standards are intended to protect against "adverse effects, not necessarily against all identifiable effects of changes produced by a pollutants." Although Congress did not rigorously define an adverse effect, it did provide general guidance in the legislative history of the debate on the CAA (Library of Congress 1974). Congress was concerned with effects ranging from cancer, metabolic and respiratory disease, and impairment of mental processes to headaches, dizziness, and nausea. Congress also noted concern foT sensitive population groups in setting the NAAQSs. In particular. Congress noted that the Standards should protect "particularly sensitive citizens such as bronchial asthmatics and those with emphysema who in the normal course of daily activity are exposed to the ambient environment." This has been interpreted to exclude individuals who aie not performing normal activities, such as individuals who are hospital ized. Further guidance was given noting that the Standard is statutorially sufficient whenever there is "an absence of adverse effect on the health of a statistically related sample of persons in sensitive groups from exposure to the ambient air." The challenge of interpreting the language of the CAA was noted in an editorial by Donald Kennedy on "Risk versus Risk" published when he served as Editor in-Chief of Science (Kennedy 2005). He wrote--"In the United States and some other industrial democracies, where people and their governments tend lo be risk averse, legislatures, courts, and administrative entities usually create a presumption favoring more safely rather than less. The definitions of risk in law are often vague ("reasonable certainty of no harm" or "adequate margin of safety") and are likely to encourage an unrealistic belief that risks can be minimized or even eliminated altogether." I think Kennedy has captured the conundrum posed by the language of the CAA, a conundrum that has been addressed by Supreme Court Justice Stephen Breyer as I w ill relate later. Standard-setting process The process for developing and issuing NAAQS is quite complex Key elements of the process, as used until quite recently include preparation and review of (a) criteria document, (b) staff paper, (c) more recently a risk assessment, and (d) a regulator* decision package leading to the Administrator's policy judgment decisions as to the proposed and final NAAQS which are published in the Federal Register. Traditionally. CASAC focused its attention on reviewing the Criteria Documents and Staff Papers and, more recently, a formal Risk Assessment. As an aside, the process was changed at the end of 2006 (Peacock 2006) with an Integrated Science Assessment and Policy Assessment Document replacing the Criteria Document and Staff Paper. Time will tell if these changes really improve the overall process. In addition to the documents noted above, the Agency now prepares a Regulatory Impact Analysis which is required under Executive Order 12866 issued by President Clinton (1993) that applies to economically significant rules that have "an annual effect on the economy of S100 million or more or adversely effect in a material way the economy, a sector of the economy, productivity, competition, jobs, the environment, public health or safety or site, local, or tnbal governments or communities." The Regulatory Impact Analysis is not considered during the NA AQS rulemaking process given the prohibition of consideration of cost in the setting of the NAAQS, as will be discussed later. The first Criteria Document prepared and released by the EPA addressed lead as a criteria air pollutant. This document was prepared and the review initiated before a Clean Air Scientific Advisory Committee was mandated by the CAA Amendments of 1977. Lead was not one of the original criteria pollutants In 1975, the Natural Resources Defense Council (NRDC), with legal leadership from Attorney David Schoenbrod, sued EPA to have lead listed as a criteria pollutant The EPA argued that it was already dealing effectively with reducing lead in air through its program to remove lead from gasoline. The Second Circuit Court disagreed (NRDC v. Train 19"6) and on March 1, 1976, ordered EPA to identify lead as a criteria pollutant and begin the process of developing a NAAQS At the lime, EPA's Science Advisory Board (SAB 1 was in the process of assuming review responsibility for scientific activities across the Agency consolidating review functions brought to EPA from its predecessor organizations such as the National Air Pollution Control Administration (NAPCA). The EPA had just disbanded the National Air Quality Criteria Advisory Committee which had operated under NAPCA as well as other media specific advisory committees in favor of a series of discipline-oriented Committees; e g., health, engineering and ecology. Springer 24h Air Qua] Atmos Health (2(>l 2) 5.21 > 258 In 1976,1 was asked, as a member of the SAB Executive Committee, to chair an ad hoc Committee to review the criteria document on lead. Preparation of this document had already been initiated by EPA in anticipation of the Second Circuit Court decision. It was prepared bv a Criteria and Special Studies Office within the Office of Research and Development located at EPA's Health Effects Laboratory in Research Triangle Park. NC. The first draft, released November 18, 1976. was viewed as unacceptable by the Ad Hoe Committee. The Committee was concerned with the poor scientific quality of the document In addition, as noted by Bachmann (2007), the Committee was concerned that the document recommended a specific numerical Standard, a value of 5 }ig/m\ which was inconsistent with the intent of the CAA to separate the scientific assessment of the relevant criteria and the setting of the specific NAAQS. The views of the Ad Hoc Committee members varied. Indeed, some members wanted the Committee to assume responsibility for re-writing the Criteria Document and recommending a specific Standard. As Chair, I emphasized our role was advisory to the Administrator, not to serv e as substitutes for EPA staff to prepare the Criteria Document. The EPA proceeded to prepare a second draft which was released on May 27. 1977. The Committee viewed it as unproved, but felt it was still not adequate for setting a lead Standard. The Agency proceeded to develop a third draft released on August 22. 1977. The Committee offered modest comments on the third draft which were considered by the Agency as it prepared the final criteria document released on December 14, 1977 (EPA 1977y) winch served as a basis for the proposed lead NAAQS (EPA 1977b). As Chair. I conveyed to the Agency the view that the final version--"accurately reflected the available scientific litera ture and provided an adequate scientific basis for promulga tion and issuance of a Standard foi airborne lead. " The first lead NAAQS was issued in 1978 (EPA 1978) The experience with the lead criteria document served as a stimulus for EPA to create a sepaiate Environmental Criteria Assessment l tfilce within the Agency's Office of Research and Development. For three decades, this office was headed by Lester Grant. Grant originally came to the EPA from the University of North Caiolina-Chapel Hill as an Inter-Govemment Personnel Act assignee to assist w ith revision of the criteria document on lead. As noted by Bachntann (2007), the Office of Air Quality Planning and Standards (OAQPS) prepared an analysis to support the Lead Standard which was reviewed by EPA scientists, policymakers and the public. However, it was not reviewed by the SAB Ad IIoc Committee. That analysis served as a basis for the proposed NAAQS for lead (EPA 197'b) and the final lead NAAQS (EPA 1978). Bachmann (2007) has noted "As for all NAAQS decisions, the final choice on the Standard was constrained and informed by the scientific information, but ultimately based on the policy judgment of a politically responsible decision-makei the EPA Administrator After consideration of and reaction to public comments, and review and discussion on the final package by OMB, the Administrator promulgated a Pb Standard of I 5 pg nr' quarterly average in TSP." 1 stronglv agree with Bachmann's first sentence assessment of the role of scientific information informing the policy judgments of the EPA Administrator. This will be a recurring theme in the remainder of this paper In many ways, the experience EPA gained in setting the lead NAAQS influenced the NAAQS process for subsequent NAAQS decisions. The OAQPS analysis evolved into preparation of formal Staff Papers that w'ould be subjected to review by the CASAC. The first activity of the newly created CASAC, initially chaired by Sheldon Friedlandcr. was the review of a combined criteria document for particulate matter and Sulfur Oxides. Subsequently, separate addenda were prepared for Sulfur Oxides and particulate matter and separate Standards issued for the two pollutants. Sulfur Dioxide was identified as the indicator for Sulfur Oxides and Total Suspended Particulate (TSP) as the indicator for particulate matter. Without going into the administrative or legal details, it is important to note that EPA, in carrying out mandated NAAQS actions in the early days, used an "informal rulemaking process" to propose and promulgate Standards (Bachmann 2007). The informal process focused on the end product, the NAAQS. The process was not always well documented as to how decisions were reached on the four elements of each NAAQS; the indicator, averaging time, specific numerical concentration and the statistical form. The DC Circuit Court of Appeals subsequently found that the record of this informal process did not give the Court a sufficient basis to complete its judicial review of the rules that were promulgated. This led to the final rule for the secondary Sulfur Dioxide Standard being revoked in 197? as recounted by Berry (1984) in his review of NAAQS decision-making. This judicial decision led EPA to develop more rigorous procedures, including documentation, for the setting of each NAAQS (Pedersen 1975). As noted by Bachmann (2007), these procedures addressed the follow ing points: "(1) EPA was to make available to the public ihe information and technical methodologies it relied upon by the time of proposal; (2) the preambles to proposal and final rules were to provide a detailed explanation of EPA's decision; (3) EPA was required to respond to all "significant" comments on the proposal by the time it issues its final rule; and (4) all of the above documents, analyses, preambles, and responses constituted the record that the court would examine in reviewing the final Standard decision Objections not raised <0 Springer \u Qujl Atmos Health (2012) 5:243-258 :47 in the record could rot he raised in court. The halcyon days of a speedy NAAQS process were over." 1 agree that the speed of the process was reduced, however. I would add that the transparency of the process was also substantially improved. Congress apparently agreed and these provisions were substantially codified by the CAA Amendments of 1977 EPA's implementation of the CAA, especially ns setting of NAAQS even with improved documentation, has been a matter of continuing controversy and litigation (some persons might argue that controversy and litigation were enhanced by improved documentation in the record). Bachmann (2007) summarizes many of the key legal cases in his review. In this paper. I will only highlight certain of the key legal cases. The 1997 revisions of the Ozone NAAQS (EPA 1997a) and Particulate Matter NAAQS (EPA 1997b) proved to be very contentious, including the discussions within CASAC. The CASAC PM Panel members had a range of views on the PMt 5 Standard that was being set for the first time supplementing the PMm Standard. This range of views w as clearly articulated in the CASAC Chair's letter (Wolff 1996) to the Administrator by including a Table showing the views of each individual. The contentious nature of the debate over these revised NAAQS prompted Administrator Browner to involve President Clinton Bachmann (2007) recounts that Admin istrator Browner had a 1-h meeting on these Standards with the President--"she reported that the President quickly accepted her decision and spent much of the time discussing how to reduce unnecessary burdens in the implementation process Flits resulted in some of us writing ihe first drafi of a letter that was later sent by President Clinton (Clinton 1997) to EPA directing implementation be carried out so as to "maximize common sense, flexibility, and cost effectiveness."" Not suiprisingly, President Clinton (New' York 1997) had a role in announcing the tighter Standards which included for the first time a separate PM? ^ Standard to supplement the PM,o Standard and a shift from a 1-h averaging time to an 8-h averaging time Standard for Ozone The issuance of a revised PM NAAQS triggered the case of American Tntcking Associations v. EPA (ATA 1999) The Court found `the growing, empirical evidence demonstrat ing a relationship between fine particle pollution and adverse health effects amply justifies establishment of new fine particulate Standards." The Court went on to find "ample support" for EPA's decision to regulate coarse particulate pollution, but vacated the 1997 PM in Standards, concluding in part that PM m is a "poorly matched indicator for coarse particulate pollution" because it includes fine particulates which were separately regulated as PMi < Subsequently, EPA removed the vacated 1997 PMio Standard allowing the 1987 PMm Standard to remain in place along with the new PM;.$. In addition, the three judge panel held, two to one. that EPA's approach to setting the level of the PM and < )zone Standards in 1997 effected "an unconstitutional delegation of legislative authority." The Judicial Panel found that "the factors EPA uses in determining the degree of public health concern associated with different levels of ozone and particulate matter are reasonable." However, it remanded the rule to EPA. The Judicial Panel stated that when the Agency considers these factors tor potential non-threshold pollutants "what EPA lacks is any determinate criterion for drawing lines" to determine the level at which the Standards should be set. The Judicial Panel also found that the Administrator, under the CAA. is not permitted to consider the cost of implementing these Standards in setting them. Not surprisingly, the nature of the Circuit Court opinion resulted in cross appeals being filed on the several issues. The Supreme Court in February 2001 issued a unanimous opinion upholding EPA's position on both the Constitu tional and cost issues (Whitman v: American Trucking Associations 2001). On the Constitutional issue, the Supreme Court held that the statutory requirement that the NAAQS be "requisite" to protect public health with an adequate margin of safety sufficiently guided EPA's discretion, affirming EPA's approach of setting Standards that are neither more nor less stringent than necessary. Supreme Court Justice Breyer. who participated in the Whitman v. American Trucking Associations Case, is well known and highly regarded for his opinions and writings on risk assessment and regulation (Breyer 1982. 1^93). Thus, it is not surprising that he took the opportunity in Whitman r. American Trucking Associations (2(301) to offer comments on the Standard-setting process and. specifically, the identification of the level of the NAAQS and the associated level of health risk While concurring that EPA cannot consider the costs of unplementing the NAAQS, he went on to note- this interpretation of 109 does not require the EPA to eliminate every health risk, however slight, at any economic cost, however great, to the point of "hurtling" industry over "the brink of ruin," or even forcing "deindustrialization." (Id. At 494; Breyer, J.. concurring in part and concurring in judgment; citations omitted). Rather, as Justice Breyer explained: "The statute, by its express terms, does not compel the elimination of all risk: and it grants the Admin istrator sufficient flexibility to avoid setting ambient air quality Standards ruinous to industry. Section 109(b)(1) directs the Administrator to set Standards that are "requisite to protect the public health" with "an adequate margin of safely." But these words do not describe a world that is free of all risk an impossible and undesirable objective (citation omitted). Nor are the words "requisite" and "public 2 Springer health" to be understood independent of context. W'c consider football equipment "safe" even if its use entails a level of risk that would make drinking water "unsafe" for consumption. And what counts as "requisite" to piotectmg the public health will similarly vary with background circumstances, such as the public's ordinary tolerance of the particular health risk in the particular context at issue. The Administrator can consider such background circum stances when "deciding what risks are acceptable in the world in which we live." (citation omitted). The statute also permits the Administrator to take account of comparative health risks. That is to say, she may consider whether a proposed rule promotes safety overall. A rule likely to cause more harm to health than it prevents is not a rule that is "requisite to protect the public health." For example, as the Court of Appeals held and the parlies do not contest, the Administrator has the authority to determine to what extent possible health risks stemming front reductions in tropospheric ozone (which, it is claimed, helps prevent cataracts and skin cancer) should be taken into account in setting the ambient air quality Standard for ozone (citation omitted). The statute ultimately specifies lhat the Standard set must be "requisite to protect the public health" "in the judgment of the Administrator," 109(b)(I), 84 Stat. 1680 (emphasis added), a phrase that grants the Administrator considerable discretionary' Standard setting authority. The statute's words, then, authorize the Administrator to consider the severity of a pollutant's potential adverse health effects, the number of those likely to be affected, the distribution of the adverse effects, and the uncertainties surrounding each estimate (citation omitted). They permit the Administrator to take account of comparative health consequences They allow her to take account of context when determin ing the acceptability of small risks to health. And they give her considerable discretion when she docs so The discretion would seem sufficient to avoid the extreme results that some of the industry' parties fear. After all, the EPA, in setting Standards that "protect the public health" with "an adequate margin of safety," retains discretionary authority to avoid regu lating risks that it reasonably concludes are trivial in context. Nor need regulation lead to deindustrializa tion. Pre-industrial society was not a very healthy society; hence a Standard demanding the return of the Stone Age would not prove "requisite to protect the public health." Although 1 rely more heavily than does the Court upon legislative history and alternative sources of Springer Au Qual Atmos Health (2012) 5:24.' 258 statutory flexibility. 1 reach the same ultimate conclu sion, Section 109 does not delegate to the EPA authority to b3se the national ambient air quality Standards, in whole or in part, upon the economic costs of compliance." The case of Whitman i . American Trucking Associations (2001) is widely cited for the conclusion that EPA cannot consider the economic costs of compliance in the setting of NAAQS. Unfortunately, m my opinion, insufficient atten tion is giv en to the thoughtful guidance of Justice Breyer on exercising policy judgment in deciding on an acceptable level of health risk, a judgment that in turn determines the level and statistical form of each NAAQS. It is interesting that Justice Brever's opinion appeared in Administrator Johnson's notice of the Ozone NAAQS (EPA 2008). but did not appear in Administrator Jackson's "reconsideration" proposal lor ozone (EPA 2010a) which will be discussed later. Paradigm shift At this juncture, it is appropriate to note that it is my view' that a paradigm shift has taken place in the use of scientific knowledge and policy judgments in the selection of the level and form of each NAAQS over the past four decades In my opinion, the paradigm shift has been driven in pan by the nature of (lie growing body of scientific evidence of pollution effects. In the 1970s. most scientists and regulators viewed the criteria pollutants as having a threshold in the concentration-response relationship for 11011-cancer endpoints, the major concern for the criteria pollutants. This was different than the prevailing view for cancer causing agents which were assumed to have linear, noil-threshold, concentration-response relationships In the early 1970s. the available data on each criteria pollutant were quite modest, with attention in the review process focusing on only a few epidemiological studies Tor those few studies, attention often focused on whether a relative risk on the order of 2.0 was observed and whether it was statistically significant or rot For a given criteria pollutant there were few . if any. controlled human exposure studies. The data from laboratory animal studies had frequently been acquired in short-term studies with expo sure concentrations much higher than ambient concentra tions. This raised questions about extrapolation from laboratory animals to humans and high to low exposure concentrations. The general approach taken to evaluating the published studies was to identify the lowest levels where effects were statistically significant and assume this w as the inflection point in the concentration-response relationship. It could then be readily argued that setting the Standard at a lower concentration than that at which Air Qua I Aunts Heallh (2012) 5:243-258 24`> effects were observed satisfied the requirement for "an adequate margin of safety " In contrast, the most recent reviews of the criteria pollutants have involved thousands of papers with obser vations ranging from the human population level to studies of intact laboratory animals to studies of effects of air pollutants on cells and molecules. Despite the huge number of published studies, the focus has ultimately centered in the Staff Paper on the results of a few studies where attention turns to the relevance of the results for informing policy judgments on the level and statistical form of the Standard. For the epidemiological studies, the debate often focuses on whether relative risks of less than 1.1 for excess morbidity and mortality are significant. Of course, the specific relative risk number is dependent on the denomi nator being used. For controlled exposure clinical studies, attention has focused on the lowest levels with statistically significant changes and whether the changes arc adverse. A new's report (Taube 19951 in Science, that 1 view as a classic report, highlighted the issues involved in the search for subtle links between diet, lifestyle, or environmental factors and disease, especially using retrospective observa tional smdies. 1 especially liked the quote at the end attributed to UCLA Professor Greenland in offering advice to his "most sensible, level-headed, estimatable colleagues.-' Remember, he says--"there is nothing sinful about going out and getting evidence, like asking people how much do you drink and checking breast cancer records There's nothing sinful about seeing if that evidence correlates. There's nothing sinful about check ing for confounding variables. The sm comes in believing a casual hypothesis is true because your study came up with a positive result, or believing the opposite because your study was negative." It is interesting to note that CASAC discussions of criteria pollutant effects have frequently focused initially on the level of the Standard, dev oid of any consideration of the statistical form of the level. This approach was in keeping with traditional practice in the setting of Standards such as Threshold Limit Values for occupational exposures to chemicals (McClellan 1999, 2010c), That approach has traditionally involved a review of the available human data on a toxic chemical to determine a no-observed effect level, or the low est observed effect level, and then use of a safety factor to arrive at an acceptable exposure level set at a lower level. In the absence of adequate human data, laboratory animal data are used and an additional safety factor applied to account for the potential that the animal observations might not adequately predict human effects. This approach was routinely used for a w ide range of health responses that were assumed to have an exposure-response relationship that exhibited cither a true or practical threshold, an excess of effects above some level and an absence of effects below that level A review of the earliest Criteria Documents and. indeed, also the Staff Papers, documents that a similar line of reasoning was used in the setting of the NAAQS identify levels where an increase in effects is observed and then set the Standard at a lower level. I he implementation of Standards set with this approach soon revealed that if the Standard was to be rigorously enforced, i.e., no exceedances of the specific level of the Standard, the practical effect would be to cause average levels of the pollutant to be reduced to levels far below the Standard so as to avoid the occasional high concentration exceeding the Standard Fortunately, common sense pre vailed and the EPA, over time, moved to the practice of routinely linking attainment of the specific level of the Standard to a statistical form such as the 98th percentile 24-h concentration averaged over 3 years, or the fourth highest 8-h average concentration during a 3-ycar period. In my experience, most of the attention of the CASAC in the NAAQS-setting process has focused on the level of the Standard with limited discussion of the statistical form of tile Standard In doing so, there has been a failure to recognize that the stringency of the Standard and the degree of health protection provided depends on both the lev el and statistical form of the Standard for a particular indicator and averaging time In fact, there have been occasions when CASAC has deliberated at length on the level of a prospective Standard and, then in a casual manner, turned its attention to what would be the appropriate statistical form for that level That this is the case is not surprising since few scientific papers discuss the implications of the reported results in terms of the frequency with which a given health effect may be observed The challenges of selecting appropriate averaging times and statistical forms for the NAAQS are substantial. The original epidemiological and toxicological studies that provide the scientific information that should inform the setting of the NAAQS do not always repoit results with an averaging time that is the same as used for the Standard. Hence, the need to make extrapolations from results reported based on one metric, such as average daily exposure, to second metric, such as an 8-h or shorter averaging tune. The setting of Standards at extreme v alues, the 98th percentile for NOi (EPA 2010b) and the 99th percentile form as done with the 1-h averaging time Standard for S02 (EPA 2010c), results in extremely stringent Standards that at best are only very loosely related to the underlying data. In my view, decisions on the selection of specific levels and averaging times for the NAAQS are policy judgments properly reserved to the Administrator informed by the available scientific know-ledge. In the 1990s, concurrent with the increasingly w idespread use of formal nsk analysis Ti Springer 250 \n Quji Atmos Health i2012) ? 245 258 procedures across society (McClellan 1999. 2010c). EPA moved to quantify the health benefits associated with setting the NAAQS at various levels, with an associated statistical form. I must admit to being an early advocate of formal quantification of health benefits of various levels and forms of the prospective Standards I viewed the approach then and l still do today, as a way to synthesize the science so it could provide useful guidance to the Administrator for making policy decisions. 1 did not envision that some advocates of quantitative risk analysis would actually view the results of the analyses as being highly accurate projections of potential health benefits expressed to two ot more significant figures, sometimes without any indication of uncertainty. The quantification of health effects potentially associated with various levels and forms of the Standards requires several kinds of input. First and foremost, it requires some knowledge of the nature of the concentration-response relationships for various temporal metrics for the pollutant in question. Typically, the response term is expressed as excess risk per unit of increased concentration over some range of ambient concentrations. The question then becomes one of whether the relationship is linear and whether there is a threshold level below which the coefficient for excess risk does or does not hold. The issue of whether there are or are not thresholds for non-cancer health endpoints is very contentious and a subject ofon-going debate (White et al 2009; Rhomberg et al 2011) A related issue becomes the selection of suitable reference baseline statistics foT the particular health effects An additional question becomes the appropriate population to be evaluated a single city, multiple cities or the population of the United States. It is obvious that there are substantial uncertainties associated with each component of the analyses. With the use of linear, no-threshold, concentration response models, the EPA has on some occasions calculated estimated excess morbidity and mortality effects attribut able to the specific pollutant down to background concen trations The Health Risk Assessment (EPA 2007b) and the Regulatory Impact Analysis (EPA 2007d) for the 2008 Ozone NAAQS serve as examples. Further, dependent on the assumptions made with regard to how ambient concen trations of the pollutant would change in response to various levels and forms of the Standard, estimated health effects avoided (i.c. health benefit) may be calculated. A key consideration as to whether these benefits can be realized relates to whether the roll-back in air concen trations that is assumed in the analysis as a result of implementation of the new Standard can actually be realized In part, the validity of the analyses relate to how realistic the assumptions have been with regard to back ground levels. A discussion of this issue for ozone can be found in McClellan et al. (2009). Indeed, as the levels of the Standards are ratcheted down toward background levels, there ts increasing uncertainly as to whether there are any health effects attributable to single pollutants and even greater uncertainty as to the magnitude of the health benefits associated with any new lower Standard. The use of single pollutant models for estimating benefits also raises the issue of double-counting of benefits as the benefits of the individual pollutants are aggregated. Hence, the paradigm shift. It is apparent that in setting the earliest NAAQS some individuals, including CASAC members, envisioned that the Standards were being set at levels protective of public health with an adequate margin of safety based on threshold concentration-response models. In short, if there were health effects at the level and form of the selected NAAQS, they were viewed as de minimis. In contrast, more recent NAAQS have been set at levels which the CASAC and EPA characterize as having residual health effects even if the Standard were to be attained. The central question remains--how low is low enough? 1 view the answer as a policy judgment informed by science that can only be made by the EPA Administrator. Recent action on revision of the particulate matter and ozone NAAQS It is instructive to now turn our attention to the most recent actions of EPA with regard to the revision of the PM NAAQS in 2006 (EPA 2006bl, the revision of the Ozone NAAQS in 2008 (EPA 2008) and the "reconsideration" proposal (EPA 2010a) for a further revision of the Ozone NAAQS in 2011 In the initial discussion, I will focus on the EPA's 2006 revision of the P\-l2. Standard The science that informed the setting of that Standard was summarized in a Criteria Document (EPA 2004). This, in turn, provided the basis for the Staff Paper (FPA 2005) The central issue was the level and associated form of the two different averaging time Standards, a 24-h averaging time and an annual Standard The first Standards using PM2 s as an indicatoi were set in 1997 (FPA 1997a). The 24-h averaging time Standard was set at 65 pg/m\ The 24-h PM ., Standard of 65 pg m1 was attained when the 3-year average of the 98th percentile of the concentrations at each population-oriented monitor was not exceeded. The Annual Standard was set at an annual arithmetic mean of 15 pg/m3. The annual Standard was attained when the 3-year average of the weighted PV1, 5 concentration from single or multiple community-oriented monitors did not exceed 15 pg/nr. Recall that the 1997 PM2 s Standard was originally intended to supplement and, in part, replace the PM io (annual arithmetic mean of 50 pg/trr and 24 h average of 150 pgrii:) Standard set in 1987 That PMjo Standard had replaced the earlier Total Suspended Particulate Standard promulgated in 1971. ) Springer Air Qual Atmos Health (2012) 5 24' 258 251 I participated as a member ot the CASAC Panel that provided advice on the setting of the PM? 5 Standard in 1997 There was much discussion about the uncertainty associated with the shift from a PM m to a PM? 5 Standard, especially the uncertainty in a shift from dependence on only the PM!0 indicator to PM?* indicator. There was strong scientific support for introducing the PM? s indicator, although at the time, there was limited epidemiological data from studies in which PM had actually been measured. There was no clear scientific evidence on the presence or absence of a threshold in the concentration-response relationship for either acute or chronic responses. The big issues related to the levels and associated form--"how low was low enough?" The prevailing tone in hallway con versations focused on two points. First, it was argued that it was important to introduce a PM; 5 indicator which, in turn, would mandate the monitoring of PM; 5. The availability of the PM? 5 monitoring data would then allow the conduct of epidemiological studies to directly evaluate a potential concentration-response association for this indicator. Sec ond. it was argued that in the absence of convincing data on PM; j the final action contemplated by the Agency should not represent a drastic increase in the stringency of the PM Standard. In my opinion the new PM2 5 annual Standard set at 15 pg/nr did increase the stringency of the PM Standard and represented a policy judgment call on the part of the Administrator that was very precautionary. In contrast, in niy opinion, the setting of PM?S 24-h averaging time Standard at 65 pg/m' was much less precautionary The level and form of the new Standards was as follows: (1) The annual PM> , Standard is met when the 3 year average of the annual arithmetic mean PM? s concern nations, from single or multiple community-oriented monitors, is less than or equal to 15 pg/nv', with fractional parts of 0.05 or greater rounded up (2) The 24-h PM2.5 Standard is met w'hcn the 3-ycar average of the 98th perccutile of 24-h PM? 5 concen trations at each population-oriented monitor within an area is less than or equal to 65 pg ni5. with fractional parts of 0.5 or greater rounded up (3) The form of the previous 24-h PM10 Standard is revised to be based on the 3-ycar average of the 99th percentile of 24-h PM|0 concentrations at each monitor within an area. Review of the PM Standard that would lead to revision of the 1997 PM Standard moved forward in the early 2000s. In 2004, as the new Criteria Document for PM was reviewed, it was decided that the CASAC would abandon CASACs practice of issuing "closure letters." "Closure Letters" had traditionally been sent by the CASAC Chair to the EPA Administrator at key junctures, such as completion of revision of a Criteria Document or Staff Paper, signifying the work product was scientifically acceptable for regulatory decision-making. Some individuals had viewed the "closure letters" as a way by which CASAC impeded progress in the setting of NAAQS in a timely manner. I viewed the "closure letters" as an effective approach to ensuring that FPA was prepanng documents that included the latest scientific information and analyses, even if it required the Agency to develop Revisions or Addendums. After reviewing and commenting on the Cnteiia Document (EPA 2004) and Staff Papct (EPA 2005). CASAC recom mended that the 24-h PM? < Standard be set in the range of 25-35 pg/nri and the annual PM? 5 Standard be set in the range of 13-14 pgm3 (Henderson 2005. 2006a; Table 1). There was strong pressure within the CASAC PM Panel to provide consensus advice to the Administrator In the end, two consultant members of the PM Panel who had both served as Chair of CASAC (myself and another) did not deem it appropriate to join w ith other members of the Panel in endorsing the specific levels others wished to recommend to the Administrator. I held strongly to the view that the difference between leaving the Standard at 15 pg m' and reducing it to 14 pg/m3 u'as not a scientific decision, but rather a matter of policy judgment that should be left to the discretion of the Administrator. In my opinion. Administrator Johnson, as the politically responsible decision-maker (using the words of John Bachmann 2007 in describing the 1974 Lead NAAQS decision] was not bound by the recommendations of CASAC as they were an advisory committee. In my opinion, the Administrator alone had the authority to make policy judgment calls in retaining or revising the annual PM-* <, Standard, then at 15 pg/m' and the 24-h PM? 5 Standard, then at 65 pg/m' (EPA 1997a). The Administrator issued a final mlc with the annua! PM?< Standard retained at 15 pg/m3 and the 24 h Standard reduced to 35 pg/nr (LPA 2006b) Table 1 National ambient air quality standards for PM -, 3nd ozone, the old standard, CASAC recommendations and administrator's final rule Indicator (unit) Old standard CASAC New standard PM.i-24 h Uiguri) 65J 30-35b 351 Annua) (pg/nr' 1 15* 13-14b 15* Ozone--8 li (ppb) S411 60-70* 75r a EPA 1997a, b l'Henderson 2006a, b; Henderson et al. 2606c "EPA 2006b J EPA 1997a, b. set at 0.08 ppm which by rounding cons ention equals 84 ppb 'Henderson 2007. 2008 rEPA 2008 C2 Springer :52 Air Qua! Atmos Health (20121 5 24' 258 After the final PM rule was issued in 2006 (EPA 2006b), the seven formal members of CASAC (Henderson et al 2006c) sent a letter to the Administrator expressing concern that the EPA Administrator had not decreased the PM2.5 Annual Standard from 15 ugrrr to 13-14 (.tg'm in combination with the setting of the 24-h Standard at 35 pg/nv, the upper end of the ranges they had recom mended. In iny view, the CASAC recommendation that the Administrator had to reduce the annual Standard by at least 1 (.ig/nr' indicated that the CASAC failed to appreciate that the setting of any NAAQS involves policy judgments, reserved by the CAA to the EPA Administrator, informed by the science. Presumably, the CASAC would have found it acceptable if the Administrator had reduced the Annual PM? 5 Standard from 15 to 14 ug/nr', or even to 13 pg/m'. Perhaps it would be useful for me to elaborate on why I think it is not appropriate for CASAC to recommend a bright line upper bound on the NAAQS. even assuming no change in the statistical form of the Standard. The Committee, when commenting on the science under girding the Standard, had noted that it had not identified a threshold in the ambient exposure concentration-response relationship for PM? 5. Consistent with this assessment of the science, the EPA in its Risk Assessment had used a linear exposure concentration-response model to estimate risk that would be avoided and risks that would remain if the Standards were set at various specific levels and with an assumed statistical form. There were estimated risks associated with retaining the Standard at 15 pg/m' and reducing it to 14 or 13 tig nr By endorsing a level of 14 pg/nv' for the annual Standard, the CASAC was indicating its support for setting the Standard at a particular level of estimated risk. In my opinion, a decision on acceptable risk (i.e., the residual risk level when the Standard is attained) is a policy decision left to the discretion of the EPA Administrator under the authority of the CAA The Committee's blended scientific and policy judgment advice would have been clearer if they had stated their specific advice by indicating both the specific numerical level and the associated morbidity and mortality. Of course, the estimates of morbidity and mortality should have had an indication of the associated uncertainties Let us now turn to revision of the Ozone NAAQS. Final action on revision of the Ozone Standard set in 1997 (EPA 1997b) followed almost 2 years after the decision on the PM? 5 Standard. The ozone review included a Criteria Document (EPA 2006c) which summarized publications through 2005. This document served as the basis for a subsequent staff paper (EPA 2007a) and risk assessment (EPA 2007b). Again. CASAC (Henderson 2006b, 2007) offered very prescriptive advice ou the level of the Standard indicating that the level of the revised 8-h averaging time Standard should be lowered to no greater than 0.070 ppm down from the 1997 Standard of 0 0.X ppm which by rounding convention was effectively 0 084 ppm. The 1997 Standard is met when the 4th highest S-h average value over a 3-year period does not exceed 0.084 ppm (Table 1). The CASAC letter on the Ozone Staff Paper (Henderson 2007) commented on policy relevant background (PRB) noting "`the Final Ozone Staff Paper does not provide a sufficient base of evidence from the peer-reviewed litera ture to suggest that the current approach to determining a PRB is the best method to make this estimation." The letter concludes w ith the statement "Thus. PRB is irrelevant to the discussion of where along the concentration response function a NAAQS with an averaging time that provides enhanced public health protection should be." The CASAC apparently failed to appreciate that identification of scien tifically valid levels for PRB for different sections of the country can have a profound influence on realizable public health benefits (see discussion in McClellan et al. 2009) and the calculated benefit and residual risks for various levels and forms of the Standards. As the Agency's activities on revision of the Ozone NAAQS were proceeding. I participated in June 2007 with a small group of scientists at a meeting held in Rochester, NY to discuss critical considerations in evaluating scientific evidence of health effects of ambient ozone. The discus sions at the Rochester Conference focused on the scientific interpretation of the data available on the health effects of exposure fo ambient concentrations of ozone, controlled ozone exposure studies with human volunteers, long-term epidemiological studies, time-senes epidemiological studies, human panel studies, and toxicological inves tigations The deliberations also dealt with ihe issue of background levels of ozone of non-anthropogemc origin and issues involved with conducting formal risk assess ment of the health impacts of current and prospective levels of ambient ozone. The participants, while offering comments on the science informing the revision of the Ozone NAAQS. did not feel it appropriate to offer policy judgments on the level and form of the Ozone NAAQS then under consideration. A report based on the Rochester Conference has been published (McClellan ct al. 2009). The deliberations at the Rochester Conference were sumniai ized and included with my comments (McClellan 2007) submitted to the EPA Ozone Docket on the proposed Ozone Standard (EPA 2007c). Administratoi Johnson, in March 2008 (EPA 2008), issued a final revised Standard for Ozone with the primary 8-h average Standard set at 75 ppb retaining the statistical form the same as the 1997 primary Standard--the Standard is attained when the fourth highest 8-h average value over a 3-year period does not exceed 75 ppb. The CASAC was displeased with the policy judgment of Administrator Johnson to set the Standard at 75 ppb rather than heeding Springer Air <>al Atmos Health (2012) 5:243-258 253 their recommendation to set the Standard in the range of 0 06(L 0 070 ppm (Henderson 2008). As an aside. Admin istrator Johnson also decided to set the secondary Standard for Ozone equal to the primary Standard. In doing so, he did not heed C'ASAC's advice to set a secondary Standard with a different cumulative form. The CASAC had recommended a sigmoidally weighted W126 index, accu mulated over 12 "daylight" hours and over at least the three maximum ozone months of the summer growing season (Henderson 2008). Some CASAC members have argued that by giving the EPA Administrator a range (0060-0.070 ppm), the CASAC had not taken away the Administrator's discretion in making policy judgments on the level and form of the NAAQS. To the contrary, I argue that the upper value in the range is in effect a bright line that CASAC has indicated the Administrator should not go above based on the science. In short, under the new paradigm. CASAC has defined for the Administrator the upper level of excess risk that CASAC deems acceptable, even though they have not clearly identified the specific health risk level associated with the 0.070 ppm level. 1 firmly believe that Administrator Johnson's decisions on both the primary and secondary ozone Standards were consistent with the legislative authority accorded the Administrator undei the CAA Much was made of the fact that in the setting of the Ozone Standards, discussions took place between White House staff and, perhaps then President Bush, as the Standard was finalized. This is hardly surprising Recall Bachmann (2007) recounted the discussions between President Clinton and Administrator Browner in 1997 and the draft memo to EPA Administrator Browner prepared by EPA staff for ultimate issuance oxer the signature of President Clinton (Clinton 1997). As soon as President Obama was sworn in on January 20, 2009. the then-Whitc House Chief of Staff, Rahm Emanuel, issued a memorandum (Emanuel 2009) stating --"It is important that President Obama's appointees and designees have the opportunity to review and approve any now or pending regulations.'' The Emanuel memorandum then proceeded to outline explicit conditions for what qualified as new or pending regulations-- for example, "all proposed or final regulations that have not been published in the Federal Register" and "consider extending for 60 days the effective date of regulations that have been published in the Federal Register but not yet taken effect." The revised NAAQS for ambient ozone, published in the Federal Register, March 12, 2008 (EPA 2008), could hardly be viewed as new or pending in January 2009. Indeed, in the fall of 2008, the EPA had already initiated action on the next review of the Ozone NAAQS (Martin 2008). In initiating the new review, it was noted that CASAC advice on the previous review of the Standard represented "a mixture of scientific and policy considerations." Nonethe less. EPA Administrator Lisa Jackson in late 2009, decided to proceed with "reconsideration" of the final Ozone NAAQS rule issued in March 2008 (EPA 2008). The decision to proceed with a reconsideration" proposal was formally announced in the Federal Register in January 2010 (EPA 2010a). The "reconsideration" proposal noted--"With respect to CASACs recommended ranee of standard levels, EPA observed that the basis for CASAC's recommendation appears to be a mixture of scientific and policy consideration " Administrator Jackson has stated that the "reconsidera tion" rule will be based on the same record used to propose the 2008 Standard, essentially the scientific infonnation available through late 2005 and included in the 2006 Cntena Document (EPA 2006a). Recall the earlier dis cussion of EPA moving to a formal rulemaking process at the insistence of the Court. The approach of using the "old scientific record" was apparently taken with a view that it offered a "fast track" to a revision of the Ozone Standard without creating a new record. The "reconsideration" proposal (EPA 2010a) states that consideration will be given to setting the primary Standard set in the range of 60 to 70 ppb. The announced date for release of the final "reconsideration" Standard has continually shitted from August 2010 to October 2010 to December 2010 to July 2011. In accord w ith the review plan laid out in October 2008. the EPA staff proceeded with preparation of the Integrated Science Assessment reviewing the new scientific infonnation to be considered in the next 5-sear res less triggered by promulgation of the March 2008 Ozone NAAQS. Ironically, the Integrated Science Assessment, the document replacing the old criteria document for ozone, was released on March 2, 2011 (EPA 2011a), all while EPA's reconsideration of the old record remains pending 1 offered comments (McClellan 2010a) on the appropri ateness of the Administrator proceeding with a "reconsid eration" Standard for ozone and offered comments (McClellan 2010b) to the EPA Ozone Docket on the specifics of the proposal, hi my view, the pioposal for the Administrator to reconsider a rulemaking, die setting of a NAAQS, formally completed 9 months earlier by the previous Administrator tn another Administration is with out precedent. It has the potential to serxe as a bad precedent with es erv change in Presidential Administration triggering a review of actions completed by the previous Administration with a viexv to potentially reconsidering the rules. In short, the new Administrator is saying "if I had been in office before I was appointed, I would have made a different policy judgment call." Administrator Jackson's use of the CASAC position in 2008 to justify the "reconsidera tion" action, in my opinion, moves CASAC out of its scientific advisory role into a strategic, policy-driving ^ Springer 254 <\ir Qual Atmos Health (2012) 5:24.' 25X Standard-setting role. This is troubling since Administrator Johnson, in issuing the 7008 Standard, had noted (perhaps with ireptdation) that the CASAC recommendation "appears to be a mixture of scientific and policy considerations,'' a view informed by EPA staff analysis (Martin 2008). 1 agree with the assessment that CASAC, in recommending specific levels, is on a path of mixing scientific interpretations with policy judgments. Administrator Jackson, in early 2011 (EPA 2011b), called on the CASAC to offer further clarification of the views it expressed earlier. The specific advice being solicited by the Administrator from CASAC is detailed in a memorandum from Lydia Wegman. Office of Air Quality Planning and Standards to CASAC (Wegman 2011). Many of the questions appear to be directed at attempting to distinguish between CASAC's interpretation of the old science and the policy judgments that resulted in CASAC's 60-70 ppb recommended range for the Standard. It proved challenging for CASAC to address these questions based only on the "old record" of pre-2006 science while ignoring the new scientific information on ozone (Samet 20) 1), The substantial new- scientific information on ozone that has been published in the 5 years since the Criteria Document (EPA 2006c) was prepared is documented in the recently released Integrated Science Assessment (EPA 2011b). The current drama over the "reconsideration" ozone rule has the potential to damage the credibility of CASAC by drawing it more tightly into the "regulatory web of policy judgments" that arc the exclusive dominion of the Administrator under the authority of the CAA. My advice (McClellan 2011) to the Administrator and CASAC was to withdraw the "reconsideration" proposal and ask CASAC to expeditiously proceed with review of the new' science now available in the Integrated Science Assessment (FPA 201 la). Call for sound science Over the last several decades, there have been increasingly loud calls from multiple quarters for using "sound science" to make regulatory decisions such as the setting of NAAQS. The call has come from both Non-Government Organizations (NGOs) representing multiple sectors, from Industry and from the scientific community. In my opinion, all of these groups and the individuals within them have difficulty separating the science from their policy-driven preferred outcomes. As a scientist and as a citizen, I strongly support the use of all the available scientific information lo inform public policy decisions. In general, I ihink the efforis of individuals and organizations to critically review and synthesize relevant scientific informa tion for the various Agency rulemaking activities has had a positive impact. This includes the situations in which original scientific data files were made available (actions that I applaud) and re-analyses conducted Indeed. I think more such analyses should he conducted, especially when the original data were acquired with public funding. By the same token, I would urge industry groups to make available to other investigators data acquired under industry sponsorship. What 1 decry, however, is the desire by some to label certain reviews or analyses as either "acceptable" or "dead on arrival" based on the source of funding without regard to scientific quality of the review or analyses. Over my career. I have encountered exceptionally high-quality reviews and analyses performed by scientists in academic, industrial, and environmental organizations with sponsorship from government. NGOs. and industry. I have also noted some reviews and analyses from these same quarters that I thought were of inferior scientific quality. In my opinion, scientific quality and rigor is riot defined by the source of funding for the work. I have great concern that the advocates of "sound science," be it NGO, academics or industry, may have unrealistic expectations as to what "sound science" can deliver. Sound science does not in and of itself make for sound decisions. As I have noted in this paper, science alone cannot identify an acceptable level of health risk, since such levels inherently represent a policy judgment call. Sound science can only inform what are ultimately policy judgments or political decisions. This is especially the case for the setting of NAAQS. ill the absence of a clearly defined threshold, which involve decisions as to acceptable health risks which aie linked lo the level (and form) of the Standard. Setting NAAQS at acceptable levels of risk Let us now return to the critical issue of "how low is low enough?" for setting a specific NAAQS. It is apparent that the body of science on any given criteria pollutant today is such that it is difficult to argue that the current Standards, if attained, would result in a world that is free of any risk of adverse effects from air pollution on the populations oi the United States. As Justice Brever wrote, wc live in a world that is not free of all risk. I draw guidance from Justice Brever's statement on his interpretation of the words of the CAA--"They permit the Administrator to take account of comparative health consequences. They allow her lo take account of context when determining the acceptability of small risks to health. And they give her considerable discretion when she does so." The "her" in Justice Brcyer's opinion is a reference to past EPA Administrator Cbristtne Whitman. However, in ray opinion, the discretion that Justice Breyer assigns to the EPA Administrator does not extend to Springer \ir Qual Atmos Health ('012) 5:243-258 the CASAC. either as individuals or acting collectively. Each of the individuals serving on CASAC may be an extraordinarily competent scientist or engineet or have other specialized know ledge of air quality and its health and environmental effects. Because of this special expertise, these individuals have a special role in interpreting the scientific knowledge that the Administrator will use m making policy judgments on the level and form of the Standard recognizing that the level and fonn, in turn, determine the level of acceptable risk that it is estimated Society will bear for that specific pollutant. As broadly knowledgeable health and environmental scientists, CASAC members are in a unique position to offer advice to the Administrator that will provide the "comparative health consequences" context that Justice Breyer has called for in his opinion. For example, it would be refreshing if CASAC members were to more broadly draw on their experience as health specialists. In doing so, when debate begins on the public health significance of an excess risk of 0.1 for some health endpoint per 10 ppb increase in ozone at 60, 70. or 80 ppb averaged over 8 h, they could offer comments on the multiple factors that influence the health risks for that endpoint. This discussion, in my opinion, should even be extended to recognize that complex factors such as the socio-economic status of individuals have a profound influence on health (Table 2, Steenland et al. 20041. I will readily admit that differences in air quality associated with socio-economic status may have a role in the differences reported by Steenland et al. (2004) and other investigators However, that admission does not serve as a basis for not providing scientific context to decisions on "how low is low enough" in setting NAAQS. 1 suspect that this was the kind of input Administrator Bill Ruckelshaus was seeking when he noted in 1983 that a decision on the PM Standard "could not be made solely on Table 2 The impact of socio-economic status uii mortality (Sleeuland el al. 2004) Mortality Men Women All causes Heart disease Stroke Diabetes COPD Lung cancer Breast cancer Colorectal cancer External causes 2.02 (1.95-209)* 1.88 (1.83-1.93) 2.25 (2.14-2.37) 2 19 (2.07-2.32) 3.59 (3 35-3 83) 2.15 (2.07-2.23) 1.21 (U6 1.27) 2.67 (2.58-2.78) 1.29 (1.25-1 32) 1.84 (1.76-1.93) 1.53 (1.44 1 62) 1 85 (1.72-2.00) 2 09 (1.91-2 30) 1.3) (1.25-1 39) 0 76 (0.73-0.79) 0.91 (0.86-0.96) 1.41 (1.35--1.48) Mortality rate ratio ' 95% confidence interval ; of socioeconomical slants science, and asked if under the statute "is there room to consider other non .scientific factors in making the major social policy judgment of picking a precise number from a range of scientifically justified values" (Bachmann 2007) Justice Breyer Iras answered former Administrator Ruckelshaus' question in the affirmative. Indeed, Justice Breyer has recommended the use of comparative health consequences as a context for Standard setting. In doing so, he has indicated that the boundaries of the relevant science for setting a NAAQS are not restricted exclusively to the health effects of the specific pollutant under consideration. This common sense approach has not been evident in many of the recent CASAC deliberations or the policy judgments of the Administration. Conclusions The United States now has nearly a half century of experience of improving air quality under the federal statute, the Clean Air Act. first enacted in 1963. The amendments of 1970. 1977 and 1990 substantially strengthened the CAA. Remarkable progress has been made in improving air quality as assessed using multiple criterion. The establishment of National Ambient Air Quality Standards for criteria pollutants by the EPA and the implementation programs of the individual Stales have contributed significantly to that success. Every decade front 1970 to the present has seen major actions with regard to the NAAQS and. in general, more stringent Standards. In many instances. Standards have been attained or nearly attained, and then a new more stringent Standard has been introduced. As some have said, we were almost there and then they moved the goal posts, i.e. lowered the Standards Now. more than at any time in the past, the policy judgment question must be asked "How low is low enough?" for each of the NAAQS. In my opinion, the guidance of Justice Breyer provides the Administrator broad latitude to make policy judgments consistent with our common goal of enhancing the health of all Americans Whatever path is chosen to go forward, there will remain a need for policy judgments informed by the best available scientific information. In creating new scientific information, I urge scientists to think broadly and adopt a strong comparative health benefit orientation. For example, when conducting epidemiological investiga tions, include multiple air pollutants and other factors, including socio-economic status that may influence the health endpoints being evaluated. Then report on all of the tested associations, not just the results for a single air pollutant The resulting broader base of knowledge will allow Society to make decisions as to what actions will yield the most improvement in health at the lowest net cost to Society. *j Springer 256 Air Qual Aimes Health (2012) 5:243-25* When future Integrated Science Assessment Docu ments are prepared, I urge that they include information that will help put the reported health effects of the specific pollutant in context One approach to this might he the development of a generic document that reviews current knowledge on the multiple factors that influence morbidity and mortality' from respiratory and cardiovas cular disease, the major health outcomes for key criteria pollutants. This information could then be used in multiple Policy Assessment Documents. Both the Inte grated Science Assessment and Policy Assessment Docu ments should more clearly identify' and characterize the health effects role of the specific pollutant under consideration as well as the role of co-pollutants and other factors influencing the health outcomes evaluated Policy Assessment Documents need to include ``determi nate criterion for drawing lines" as called for by the DC Circuit Court in its American Trucking Associations v. EPA (1999) opinion. These are needed to provide a clearer basis for the Administrator's policy judgments on the level and form of the Standard. These criteria, along with a strong comparative healih context, should provide an improved basis for the Administrator's policy decisions. I also strongly urge the CASAC to focus on the scientific rigor of the scientific content and analyses in the Integrated Science Assessment and Policy Assessment Document, and av oid the temptation of offering policy judgments as to a specific upper-bound level and form of the Standard or what they view as acceptable ranges If CASAC cannot avoid this temptation to stay out of the "policy judgment thicket," then it needs to be clear as to the specific scientific knowledge that informs their personal policy preferences. CASAC is required to comment to the Administrator under CAA 109(d)(2)(B) "on any new national ambient air quality Standards and revisions of existing criteria or Standards as may be appropriate." However, in offering comments. CASAC needs to very carefully articulate where CASAC scientific interpretations leave off and CASAC policy judgments begin. Moreover, it is important for EPA Administrators to recognize they need not be bound by CASAC's specific policy preferences or range of policy preference outcomes. While the CASAC members arc citizens and arc certainly entitled, just like any citizen, to have personal preferences as to policy outcomes, CASAC members, acting in that role, should not view' themselves as broadly representative of Society at large. It is critically important that EPA Administrators recognize, as Administrator William Ruckelshaus so clearly did in 1983, that Standards cannot be set solely on science and that the ultimate decision on a level and form of a Standard necessarily reflects policy judgments. Adminis trator should not seek to find "scientific cover" for these policy judgments in the deliberations offered by CASAC. If tins is done, it has the potential to transform the Clean Air Scientific Advisory Committee into a de facto Clean Air Standards Setting Committee, thereby usurping the policy role of the Administrator. I do not think that is consistent with the language of the CAA Hie Administrator as a public official appointed by the President and confirmed by the Senate, is expected to have a broad perspective reflective of all of Society, not just a specific scientific constituency, when making policy judgments in setting National Ambient An Quality Standards. Declaration of Interest I have participated, beginning in the mid1970s. as a member of numerous CASAC Panels providing advice io the EPA Administrator on the setting of the NAAQS for all Ihc criteria pollutants- 1 served as Chair of CASAC in 1988-1992 when ihe debate began on shifting the averaging time for the ozone Standard from 1 to 8 h 1 served on the CASAC PM Panels that provided advice on the PM25 Standards promulgated in 1997 and 2006.1 serv ed on the CASAC Ozone Panel that provided advice on the Standard promul gated in 1997 I did not verve on the CASAC Ozone Panel that provided advice io the FPA Administrator on the Standard promul gated in 2008 However, I did follow ihat activity closelv and offered comments to CASAC and EPA on the science informing the Administrator's judgments on the Ozone N A AQS. Tire views 1 share in this paper are my own professional v lews based on three decades of experience participating in the NAAQS setting process I regularly serve as an advisor to both public and private organizations on atr quality issues. This includes the American Petroleum Institute (API) and various companies in the energy and transportation sectors The views 1 have expressed arc not necessarily those of the API or any organ izaiion 1 advise Open Access This article is distributed under the terms nf :he Crestive Commons Attribution Noncommcicial License winch permits any noncommercial use. dislnbution. and reproduction in any medium, provided the original aulhorjj) and source are crediled. References American Trucking Associations, Inc vs U.S. Environmental Protection Agency (1999) DC Circuit. May 19. |99Q Bachmann J (2007) Will the circle he unbroken: a history of the lr S National Ambient An Quality Standards. J Air Waste Manage Assoc 57-652-697 Berry MA (1984) A method for examining policy implementation: a study of decision-making for the national ambient air quality standards. 1964-1984: U.S. Environmental Protection Agency. Research Triangle Park. NC Brcyer S (1982) Regulation and reform. Harvard University. Cambridge. MA Brcycr S (1993) Breaking the vicious circle: toward effective risk regulation (Based on 1992 Oliver Wendell Holmes leciurcs al Harvard Law School). Harvard University, Cambridge. MA CAA (196?) The Clean Air Act of 1963 CAA (19701 The Clean Air Act amendments of 1970-P.L. 91-04 (December 31. 1970), Plus technical amendments made by PL. 92-157 (November 18, 1971) CAA (1977) The Clean Air Act amendments of 1977 4) Springer Air Qu a I .Atmos Health (2012) 5:243 258 CAA (I WO i ihc Clean Air Art amendments t.i 1490. Public Law No 101-549: 104 STAT 1499. 1990 Clinton WJ. President 11993) Executive Order 12866--regulatory planning and review. Fed Regis 5S.51735 51744 Clinton WJ. President (1997) Memorandum for the administrator of the environmental protection agency implementation of revised air quality standards for ozone and particulate matter Emanuel R (2009) Assistant to the President and Chief of Staff. Memorandum to the Herds of Executive Departments ami Agencies Fed Reg 74( 131:4435-4436 Henderson Rf7 (2005) Letter to administrator on paniculate matter standard from CASAC Chair, Rugene Henderson to EPA AdniimstTator Stephen Johnson. June 6. 2005 Henderson RF (2006a) Clean air scientific advisory committee recommendations concerning the proposed national ambien! air quality standards for particulate mutter: Letter from CASAC Chair, Rogerie Henderson to EPA .Administrator Stephen Johnson. March 21. 2006 Henderson RF (2006b) Clean air scientific advisory committee's review of the agency's draft ozone staff paper letter from CASAC chair. Rugene Henderson to administrator Stephen Johnson dated October 24. 2006, EPA-CASAC-07-001 Henderson RF Cowling LB, Crapo JD. Miller FJ. Poirot RL. Spcizcr FE, Zieiinska B (2006c) Clean air scientific advisory committee recommendations concerning the proposed national ambient an quality' standards for particulate matter, letter to EPA administra tor Stephen L, Johnson, EPA-CASAC-LTR-06-002. March 21, 2006: EPA clean air scientific advisory committee. Washington, DC Henderson RF (2007) Clean air scientific advisory committee's review of the agency's final ozone staff paper: letter from CASAC Chair. Rogene Henderson to administrator Stephen Johnson dated March 26, 2007, EPA-CASAC-07-002 Henderson RE (2008! Clean air scientific advisory committee recommendations concerning the final rule foi the national ambient air quality standards for ozone: letter from CASH Chair. Rogene Henderson, to EPA Administrator Stephen Johnson, April 7, 2008, F,PACASAC-084)09 Kennedy D(20f)5) Risk versusnsk Science 309:2137, 30 Sqitemher 2005 l ibrary of Congress (1974) A legislative history of the clean air amendments of 1970, Vol I: Serial number 93-- 18; Environmental Policy Decision. Congressional Research Service. Library of Congress. Washington. DC Martin K (2008) Building on the last ozone KAAQS Review key policy relevant issues. Presentation to EPA Workshop on Science/ Policy Issues for Next Penodic Review of Ozone NAAQS, October 29. 2008 Maruneau R, Novcllo D (2004) The Clean An Act handbook 2nd edition McClellan RO (1999) Human health risk assessment a historical overview and alternative paths forward. Inh.il Tox 11 477 518 McClellan RO (2007) Comments on national ambient air quality standards for ozone: proposed role Submitted on October 9, 2007 to Docket ID No. EPA-HQ-OAR-2005-0172 McClellan RO. Frarnpton MW, Koutrakis P. McDonnell WF. Moolgavkar S, North D\V_ Smith AE Smith RL Utell MJ (2009) Critical considerations in evaluating scientific evidence of health effects of ambient ozone, a conference report I oho I Toxicol 21(2):!-36 McClellan RO (201 Oa) Comments on proposed "Reconsideration" national ambient air quality standard for ozone for presentation at the public meeting. U.S, Environmental Protection Agency, Arlington, VA. February 2. 2010 McClellan RO i2010b) Comments on the proposed "Reconsideration ' national ambient air quality standard submitted on March 9, 2097 to i he Ozone Docket McClellan RO (2010c) Hazard and risk assessment and management In: Jon Ayres G, Roy Harrison M Maynard R. Gordon Nichols L (F.ds.), Textbook of Environmental Medicine Chaplet 4 1st Ed Hodder Education London, UK pp 59-88 McClellan RO (2011) Written Maiununit for consideration by the clean air scientific advisory committee (CASAC) Ozone review panel at the panel's teleconference meeting on February IS. 201) New York Times (1997) Clinton sharply tightens air pollution regulations despite concern over costs. NY Times. June 26. 1997 NRDC vs Tram (1976) 411F, Supp. 864 IS D.N.Y.l affd 545F.2d 320 (2nd Ctr. 1976) Peacock M (2006) Memorandum from Marcus Peacock, deputy administrator to Dt. George Gray and Bill Wehrun. Process for reviewing the national ambient air quality standards, December 7, 2006; available at hujv'Avww.epa goviiniirtaqs/memn process for_revicwing_naaqs.pdf (accepted 2007) Pedersen WF (1975) Formal records and informal rulemaking Yale Law J 38:38-88 Rhomberg LR. Goodman JE, Haber LT. Dougson M. Andersen MT. Khiurng JE. Meek B, Price PS. McClellan CSM (2011) Linear low-dose extrapolation for non-cancer health effects is the exception, nol (he rule Crit Rev Toxicol 4|(1),1-19: available at http: informahealthcare com < :xc 31 i Samet JM (2011) Clean air scientific advisory committee (CASAC) Response to charge questions on the reconsideration of the 2008 Ozone national ambient air quality standard EPA-CASAC-114)04 Letter from CASAC Chair. Jonathan M. Samet to Administrator Lisa P Jackson. March 30. 2011 Steenland K. Hu S. Walker J (2004) All-cause and cause-specific mortality by socioeconomic status among employed persons ip 27 U.S. Stales. 1984- 1997 Am J Public Health 94:1037-1042 Tsube G (1995) Epidemiology faces its limits Science 269 104-169 U S Environmental Protection Agency (1971 j National primary and secondary ambient air quality standards. Fed Reg 36:8186 U S. Environmental Protection Agency (1977a) Air quality criteria for lead (released December 14. 1977). F PA/600 8-77 017 U.S. Environmental Prelection Agency (1977) Lead ambient air quality standard proposal. Fed Regis 42:63076 U.S Environmental Protection Agency (197Si National ambient an quality standard for lead Fed Regis 43:46246 U.S. Environmental Protection Agency (1997a) National ambient air quality standard for particulate matter Final Rule Fed Reg 62.38652-38760, July 18. 1997 U.S. Environmental Protection Agency (1997b) National ambient air quality standard for ozone. Final Rule Fed Reg 62( I3$):3885538896. July 18. 1997 U S I nvinmmen:a Protection Agency (2004) Air quality for particulate matter (October 2004). 60fl/P-99.'002aF-bF. Washington, DC U S. Environmental protection Agency (2905) Review of the national ambient air quality standards for particulate matter: policy assessment of scientific and technical information (December 20(15). EPA-452 R-05-0O5a. Washington, DC U.S. Environmental Protection Agency (2CKlba) National ambient air quality standards for particulate matter proposed rule Fed Regis 71:2620-2708 U.S. Environmental Protection Agency (2006b) National ambient air quality standards for particulate matter: final rule. Fed Regis 71:61144-61233 U.S. Environmental Protection Agency (2006c) Air quality criteria for ozone and related photochemical oxidants (Final). Washington, IX' EPA/600. TUO5- 004aF-c F U.S. Environmental Protection Agency (2007a) Review of the national ambient air quality standards for ozone: policy assessment of scientific and technical information U.S. Environmental Protection Agency (2007b) Ozone he.: I ill risk assessment for selected urban areas. EPA-452/R-07-009 ) Springer :58 Air Qual Atmos Health (.2012) 5:213-258 U S. Environrucotal Protection Agency (2007) National ambient air quality standards for ozone: proposed rule 40 ( HR part 50 Fed Regis 72:3781R 37010 US. Environmental Protection Agency (2007d) Regulatory impact analyses 7007 Proposed revisions to the national ambient air quality standards for ground-level 07011c Released August 2. 2007. www.cpa.gov ttn4vcas/riaJitml#ria2007 U.S. Environmental Protection Agency (2008) National ambient air quality standards for ozone: final rule 40 CFR paits 50 and 58 Fed Regis 73:16476- 16514 U.S. Environmental Protection Agency (2010a) National ambient air quality standards for ozone Proposed Rule Fed Reg 75 2938. Januaty 19.2010 U.S. Environmental Protection Agency (2010b| Primary national ambient air quality standards for nitrogen dioxides; final rule. Fed Reg 75:6474. February 9, 2010 U.S. Environmental Protection Agency (2010c) Primary national ambient air quality standard for sulfur dioxide: final rule Fed Regis 75:35520-35603. June 22. 2010 U.S Environmental Protection Agency (2011a) Integrated science assessment for ozone and related photochemical oxidants (First Fxtemal Review Draft) U S Environmental Protection Agency. Washington, DC, EPA/60WR-10/076A U.S. Environmental Protection Agency (201 lb) Notification of Public Teleconference ol the Clean Air Scientific Advisory Committee (CASAC): Ozone Review Panel for Reconsideration of the 2008 National Ambient Air Quality Standard Fed Reg 7> 1089' February 28, 2011 Wcgman LN (2011) Solicitation of CASAC Advice on EPA's Reconsideration of the 3008 Primary Ozone National Ambient Air Quality Standard Letter to Holly Stallworth. Designated Federal Official. CASAC, January 75. 3011 While RJ1. Cote 1. Zeise L. Fox M. DotnmJCi F. Butke TA. White PD. Hattis DB. Samct JM (2009) State-of-the-Sciencc Work shop Report: Issues and Approaches in Low Dose-Response Extrapolation for Environmental Health Risk Assessment Available at: hup wwvv chpotilinc nrg members 2(108 11502 I I502.pdf. Environmental Health Perspectives I17(2V doi.i 0.1289 clip. 11502 Wliitman vs American Trucking Associations (2001) 531 U.S. 457. 121 S Ct 903149L. Ed. 2d! Wolff OT (1996) Letter to Carol M. Browner. U.S. Environmental Prelection Agency Administrator closure on draft Office of Air Quality Planning and Standards (OAPQS Staff Paper) (Review of the National Ambient Air Quality Standards for Particulate Matter: Policy Assessment of Science and Technical Information) Springer EXHIBIT B Risk Analysis. Val Vo. <?. 20lti doi !onn/n.i:h74 Invited Commentary Providing Context for Ambient Particulate Matter and Estimates of Attributable Mortality/ Roger O. McClellan* Four papers on fine particulate matter (PM: <) by Anenberg et at., Fann et at, Shin ft at, and Smith contribute to a growing body of literature on estimated epidemiological associations between ambient PM; ; concentrations and increases in health responses relative to baseline notes. This article provides context for the foui articles, including a historical review of provi sions of the U.S. Clean Air Act as amended in 1970, requiring the setting of National Ambient Air Quality Standards (NAAQS) for criteria pollutants such as particulate matter (PM). The substantial improvements in both air quality for PM and population health as measured by decreased mortality rates are illustrated. The most recent revision of the NAAQS for PM; in 2013 by the Environmental Protection Agency distinguished between (1) uncertainties in characterizing PM;; as having a causal association with various health endpoints, and as all cause mortality, and (2) uncertainties in concentration--excess health response relationships at low' ambient PM concentrations below the majority of annual concentrations studied in the United States in the past. In future reviews, and potential revisions, of the NAAQS for PM;it will be even more important to distinguish between uncertainties in (1) char acterizing the causal associations between ambient PMj < concentrations and specific health outcomes, such as all-source mortality, irrespective of the concentrations. (2) characterizing the potency of major constituents of PM 5. and (?) uncertainties in the association between ambient PM; concentrations and specific health outcomes at various ambient PM;., concen trations The latter uncertainties are of special concern as ambient PM; 1 concentrations and health morbidity and mortality rates approach background or baseline rates. KEY WORDS: Clean Ait Act: criteria pollutants: National Ambient Air Quality Standards; particulate matter; PM2 5 The purpose of this commentary is to provide context and perspective for considering the contents and conclusions of four articles in this issue of Risk Analysis concerned with ambient fine particulate matter, 2.5 micron (PM; s) and estimates of PM; 5 at tributable mortality. 1. KEY ELEMENTS OF FOUR ARTICLES Before offering my comments, I will briefly sum marize w'hat I view as key aspects of the four articles. "Toxicology nnd Human Health Risk Analysis. Albuquerque. NM. L SA: rogero.mcclellani? att.net. Anenberg el <7/.(" provide a useful review of 12 air pollution health impact assessment tools that have been extensively used internationally. The tools use common data sources for the key inputs: (1) ambient PM; < concentration-response association functions, (2) measured or estimated ambient con centrations of PM'5, (3) populations evaluated, and (4) baseline mortality rates. The models are all grounded in linear concentration-response functions. Fann et al.t2) focus on the strengths and weak nesses of four research synthesis approaches to characterizing the long-term ambient PM; s concen tration-response functions. They note "whether 1755 0272-4.332/16/010fl-175<>$22 00/1 < 2016 Society for Risk Analysis 12I9-AR86-COMM-21 -7, McClellan 2016 1756 McClellan implicitly or explicitly, all require considerable judgment by scientists," an admonishment that should be heeded by both scientists and policy makers. Their locus is on linear models of ambient PM2.5 concentration-response relationships. They provide some useful examples of estimated PM < attributable premature deaths based on different ambient concentration response functions, estimates for which 1 will provide context. Shin etal}-] review meta-analysis methods for es timating the shape and uncertainty in the association between long-term exposure to ambient particulate matter (PM) and all-cause mortality. Their article considers both linear concentration-response mod els and alternative models extending to higher con centrations as required for some global applications. This is an especially important consideration when addressing the global range of PM; concentrations from current low ambient concentrations observed in countries such as the United States and Canada that have aggressively regulated air pollutants for half a century to countries like China and India with re cent rapid industrialization, more limited regulations, and very high ambient concentrations of PM-,.5 and other air pollutants. Equally as important and not addressed by Shin et al are the remarkable differ ences in various characteristics among the countries and within countries, including population character istics such as baseline mortality rates, which are key inputs to the models. The fourth article by Smith1'1 illustrates the use of alternative approaches to calculating expected benefits of reducing the U.S. annual National Am bient Air Quality Standard for PM-^ from 15 to 12 /ig/m\ This article contains useful examples of the marked differences in estimates of avoided prema ture deaths dependent on the assumptions used in the calculations, including whether deaths are pro jected to occur below the current U.S. annual stan dard of 12 fipm . I will provide context to those cal culated estimates of avoided premature deaths. 2. THE CLEAN AIR ACT AS CONTEXT The subject matter of the articles is grounded in the Clean Air Act (CAA) originally passed in 1963/'" extensively amended in 1970(61 and again in 1990.11 The CAA is the primary legislative basis for addressing air quality in the United States. Key sections of the CAA that require the U.S. Environmental Protection Agency (EPA) to set National Ambient Air Quality Standards (NAAQS) for certain air pollutants found across the United States and attributable to multiple sources, based on scientific criteria; hence, in common usage they are called criteria pollutants. The CAA identifies two types of NAAQS. Primary standards are intended to protect public health, including protection of `sen sitive" populations such as asthmatics, children, and the elderly. The primary standards are the focus of this commentary. The CAA also calls for secondary standards to protect public welfare, which includes visibility and damage to animals, crops, vegetation, and buildings. Bachman1 M reviewed the long history of the NAAQS. a paper that should be read by all who are interested in this topic. It is useful to recall that passage of the CAA was motivated by widespread recognition in the 1950s and 1960s that the United States had serious air qual ity problems arising from a marked increase in indus trial activity during and after World War II. In ad dition, it was recognized that air quality was being increasingly impacted by expanded use of motor ve hicles. It was generally accepted that poor air quality was impacting the health of the populace. Initial at tempts to control air pollution were grounded in local and state legislation. It soon became apparent that these actions were inadequate; hence, the CAA, as passed in 1963. was national in scope. Indeed, it spec ified the creation of a National Air Pollution Con trol Agency. This agency would ultimately become the "air office" component of the U.S. EPA when it was created on December 2,1970. The CAA amendments of 1970 substantially el evated the federal role in improving air quality, in cluding the setting of NAAQS. Ihe amended CAA (1970) delegates to the EPA Administrator respon sibility for policy decisions on setting the four ele ments of each NAAQS (the indicator such as PM; s, the averaging time such as annual 01 24 hour, the con centration, and the statistical form used to determine when the standard is attained). It is important to recognize that the CAA gives the EPA Administra tor broad policy-making discretion for setting each NAAQS. The primary or health-based NAAQS are standards set so as to provide requisite protection, neither more nor less stringent than is necessary to protect public health, with an adequate margin of safety. The CAA does not specify a quantitative goal for setting each NAAQS based on some specific level of health protection, i.e., an acceptable level of risk. Thus, the level of risk protection embedded in each NAAQS is a policy judgment delegated to the EPA Administrator. Further, the U.S. Supreme Court in Invited Commentary 1757 Whitman vs. American Trucking Association' ruled that in setting the NAAQS, the Administrator can not consider the costs of achieving the standards. The six original criteria pollutants were PM. pho tochemical oxidants, carbon monoxide, sulfur diox ides, nitrogen oxrdes, and hydrocarbons. It was later determined that the hydrocarbons were more appro priately addressed as individual pollutants under the hazardous air pollutants section of the CAA. Le gal action in the 1970s initiated by the National Re sources Defense Council forced HPA to list lead as a criteria air pollutant. NAAQS have been set for each of the criteria pollutants and the science under girding each NAAQS periodically reviewed. Most re views have concluded with revision of the NAAQS. In addition, a national network of monitors has been established, primarily for regulatory compliance pur poses. These monitors also provide the data that have been key to the conduct of most long-term epidemi ological studies. PM is a generic term for a broad class of chem ically and physically diverse substances that exist as discrete particles (liquid droplets or solids) over a range of sizes in the ambient air. It is important to recall that the original NAAQS for PM set in 1971 used "total suspended particles" (TSP) as an in dicator. TSP samples are collected with a high vol ume sampler and include particles up to 25-45 mi crons in size. Standards were set for both 24-hour and annual averaging time. The latter was set at 75 /ig/m\ annual geometric mean form. In the dis cussion that follows, the focus will be on the annual standard. After an extensive review process initiated m the late 1970s. the PM NAAQS was revised in 1987 with the TSP indicator replaced with a particu late matter, 10 microns (PM-o) indicator. It is impor tant to recognize that the PM1(> fraction is included within the size range of TSP samples. The new an nual PMjn NAAQS was set at 50 /ig/m and the form changed to an annual arithmetic mean, aver aged over three years. A contentious review concluded in 1997 resulted in a revision of the PM NAAQS with the addi tion of a PM25 indicator despite there being very limited PM'5 ambient concentration-response data available for setting the NAAQS with 2.5 micron PMj 5, indicators. Keep in mind that the PM? 5 frac tion is included within the size range of the PM10 fraction. The PM2S annual NAAQS wras set at 15 /rg/nr\ annual arithmetic mean, averaged over three years. To give impetus to the adoption of a PM-i 5 in dicator, one EPA official commented: "If you want monitoring data on PM25 for epidemiological stud ies, you need to support setting a NAAQS for PM_vs, we only monitor what is regulated." Tn 200b, after an other contentious review, the PM NAAQS was re vised with a reduction in the 24-hour standard from 65 to 35 pg/m and no change in the annual standard. In 2012, after another review, the PM standard was again revised with a reduction in the primary annual NAAQS to 12 pg/m'. annual arithmetic mean aver aged over three years. The next cycle of review of the PM NAAQS is already underway. If the agency were to conform wfith a five-year review cycle, the next re view should be concluded by 2018. The agency has al ready acknowledged that it will not meet that sched ule and instead has announced a schedule for release of the final PM rate in 2021. In my opinion, the changes in the annual PM NAAQS over the decades have been driven largely by (1) improved scientific knowledge on the role of particle size governing the deposition and retention of airborne particles, hence the serial shift from a TSP to PM10 to PM15 indicator, and (2) improved knowledge from epidemiological studies of human populations such as those under discussion in the four articles. The policy decision of the EPA Admin istrator on the level and form of the NAAQS for PM has largely been informed by the information from epidemiological studies. All of the PM NAAQS set to date are based on mass concentration and the assumption that all of the PMs in each size fraction are of equal toxicity on a mass basis. This assumption needs careful review in the current PM review' cycle. 3. HISTORIC CHANGES IN PM25 AND MORTALITY To provide context for considering the contents of the four articles, it is useful to consider the sub stantial historic changes in ambient PM2, and mor tality rate in the United States. One of the major long-term studies of the association between ambient PM and mortality is the Harvard Six Cities Study, a study conceived by Professor Benjamin Ferns in the 1970s w'hen revision of the NAAQS set in 1971 was under review. Updated findings from this study have been periodically published. The recent paper by Lepeule et alP0) provides a useful summary of the changes in ambient PM2s concentrations in the six cities from the mid 1970s through 2009. The range of ambient concentrations shown (Fig. 1) is a rea sonable representation of the downward trend in 1758 "R D 1= * % S G. a* McClellan * SuiUbfHJsflf --* Uin-gsiioin-rtarrrmssr * St touts' | - -Jr - VtfatOrlQWp - -* - Tapgfci Pwa^-wvm-ratcttMllc >o ?S74 '.975 1976 1977 197 1979 1930 1581 155? 1953 '984 195 1986 *987 1988 1983 19% 1991 >997 19% 1994 'MS 19% 199? 1995 1999 70QC MC ?Q0? 7903 HC4 7DCS 2306 ?Q&7 23tti ?OOQ fig, I. Annua! mean PMj s levels daring 1974-2009 in the Harvard Six Cities Study. (Adapted from I epeule tt al^"' The data points pre-3 997 tor PM; < have been extrapolated from TSP and PMjn measurements. > urban areas seen across the United States over this time period. In reviewing the figure, keep in mind that the PM indicator from 1971 to 1978 was TSP and from 1978 to 1997 was PM ,, with the PM?5 indicator added in 1997. The PMa concentrations shown in the figure for the earliest years are extrap olations from other indicators. The reductions in am bient PM- are impressive, especially for the three cities that originally had concentrations of 25 jug/'m and higher. It is reasonable to assume that these cities experienced even higher concentrations of PM; 5 and coarse particles (PMp minus PM; 5 ) at earlier times. During the last three-quarters of a century, theie have also been impressive improvements in mortality rates across the United States, with con tinuous reductions in crude death rates and even more impressive reductions in age-adjusted death rates.111'1 Data for the period 1960 2010 are shown in Tig. 2.<12' It is important to note that these are na tional statistics with important substantial differences in both crude and age-adjusted death rates (deaths per 100.000 population) among different states and racial groups. For example, the age-adjusted death rate (all causes) in 2010 ranged from 590 in Hawaii to 962 in Mississippi. Further context is provided by the data in Table I as to cause of death for mortality in the United Stales in 2010.n:* Consideration of these multiple causes of death provides insight into potential opportunities NOTES' Cruris death rales- are on art annual basis pat 100.0QU ttqpufat&n. aas3d[iJSJteri rates are par lOO.OW U j standard populanort; sse teciiaicai Wuta$. Rglas Tor i?W1 -2009' shte revised iisifi$ ujxto-ieci omercensafi population estmaaes and! rngjf (fsITor kwerases prewitiiisty pubUsbed: see Technical Metes. SOURCE. CDC.iNCHS, Piaflbnal Vila! Staffs&eiSystaw rAewtahry Kg. 2. Crude and age-adjusted death rates: United States. 1960?.ntO (Adapted from Murphy errrftl;i) for improving the health of the li.S. population, our ultimate goal. 4. COMMFNTER'S BACKGROUND FOR CONTEXT It is important to recognize that provision of any context, to a large extent, is dependent on the commenters' backgrounds and how they view the Invited Commentary 1759 I able I. Causes of Death for the United St ates for 2010 by Major ( auses11-' Rank Cause of Death (Based on ICD-iP. 2004) Number All causes 2.468.435 1 Diseases of heart 597,689 2 Malignant neoplasms 574.743 3 Chronic lowci respiratory diseases 138.080 4 Cerebrovascular diseases 129.476 5 Accidents (unintentional injuries) 120.859 6 Alzheimer s disease 83.494 7 Diabetes mellitus 69.071 8 Nephritis, nephrotic syndrome, 50.476 and nephrosis 9 Influenza and pneumonia 50.097 10 Intentional self-harm (suicide) 38.364 11 Septicemia 34,812 12 Chronic liver disease and cirrhosis 31.903 13 Essential hypertension and 26.634 hypertensive renal disease 14 Parkinson's disease 22.032 15 Pneumonitis due to solids and 17.011 liquids All other causes 487.694 application of the work being reviewed. The context and perspective 1 offer is grounded in my experience as a scientist, research manager, and advisor on the use of science to inform public policy decisions. 1 have been studying the health effects of airborne ma terials for over half a century, initially focusing on ra dioactive materials, as might be released in a nucleai reactor accident, and later on airborne emissions from various energy technologies, especially diesel compression ignition engines. Soon after passage of the CAA, I began advising both public agencies and private organizations on air quality issues at the interface between science and public policy. Much of that activity has involved the setting of NAAQS for criteria air pollutants, including PM and implementation of strategies to attain the NAAQS. This service included chairing the LPA's review committee for the first criteria document on airborne lead and later the HPA ( lean Air Scientific Advisory Committee (CASAC) and service on the CASAC Panels that reviewed the science undergirding the 1987, 1997. and 2006 revisions of the PM NAAQS. I offered independent comments on the 2013 revision. Based on my personal experience in the NAAQS setting process, I am firmly convinced that science should inform the policy decisions that are re quired in the setting of the NAAQS.'1 -'1 However, a corollary is that both scientists and policymakers should recognize that the science alone is not suffi cient for making policy decisions. This is particularly the case, in the absence of a quantitative goal or target for acceptable risk. The alternative appioach embedded in the CAA is a policy judgment by the EPA Administrator as to how low is low enough. Tensions develop when scientists want to enter the policy arena and specify numerical standards that implicitly involve policy judgments. Tensions also arise when policymakers cast their policy judgments as being dictated by the science and abdicate their policy judgment role. I addressed those issues in the paper `Role of Science and Judgment in Setting National Ambient Air Quality Standards: How Low is Low Enough?"' The passage of the CAA had substantial impact on the research enterprise in the United States, with substantial federal funding provided for investiga tion of pollutants from their movement from their sources at smoke stacks and tail pipes through the atmosphere to people and the development of an improved understanding of the health effects of air borne pollutants. A national network of monitors has been deployed, primarily for regulatory compliance purposes and secondarily for research purposes. Sub stantial investments of public and private funds have been made to develop and improve a wide range of technologies to reduce emissions of both regulated and nonregulated air pollutants from various sources. It is widely acknowledged today by multiple par ties. the public, government agencies, mdustiy. and politicians that the regulatory programs grounded in the CAA have had widespread positive impact. Ail quality in the United States today is markedly im proved from that observed in the 1970s and earlier, l'his leads to a critical question today as to what ex tent current air quality has anv adverse impact on human health and, if so, are even more stringent NAAQS required? The first three articles under con sideration address the science that informs policy de cisions on the question posed. The fourth article by Smith'4' is at the interface of the science and pol icy. Some readers may be alarmed by my raising the issue of whether current air quality in the United States has adverse health impacts and requires more stringent standards In my opinion, addressing that complex issue is at the interface of science and pol icy and is one reason why the four articles and related commentaries should be of interest to a wide audience of scientists, policymakers, and the public. 1T fiO McClellan 5. EVALUATING CAUSALITY A critical issue related to assembling, integrat ing. synthesizing, and communicating the science on the health effects of PNLs revolves around whether there is a "causal" link between exposure to ambi ent PM25 and a range of health endpoints includ ing all-cause mortality and specific causes of death such as ischemic heart disease, stroke, chronic oh structive pulmonary disease, and lung cancer. To aid in addressing this issue in 3n organized way, the EPA has developed a five-level hierarchy that classifies the overall weight of evidence drawn from integration of evidence across epidemiological, con trolled human exposure studies, and toxicological studies and the related uncertainties that ultimately influence our understanding of the evidence. The five categories are: (1) causal relationship. (2) likely to be causal relationship. (3) suggestive of a caasal re lationship. (4) inadequate to infer a causal relation ship. and (5) not likely to be causal relationship, 51 This approach is analogous to the hazard identi fication methodology widely used for decades in addressing cancer hazards of various agents. The Federal Register announcement of the National Am bient Air Quality Standards for Particulate Matter: Final Rule'15' has extensive discussion of the use of this qualitative categorical hazard hierarchy in in forming Ihc policy judgments supporting the decision (1) to lower the annual NAAQS for PM from 15 to ! 2 /ig/m! and (2) to retain the 24-hour averaging time NAAQS set at 35 pg/m' with a 98th percentile statis tical form for attainment purposes. It is noteworthy lhat this "causal" calcgorization process, by ns very nature, emphasizes positive findings, which, in turn, emphasize the findings from studies at the highest ambient PM concentrations, ll is important to recognize that the categorization process does not rigorously address tlie equally im poilant question of whether PNE 5 at levels currently found in the United States have increased associated morbidity and mortaliu rates for specific health out comes over and above baseline rates. Thai is a critical issue in the review of the science for a policy decision on any potential revision of the NAAQS for PM> v The issue of what ambient concentrations of PM;s have a causal attributable effect on health outcomes such as an increase in ail-cause mortality over and above background or baseline rates is not addressed by the five-level causal hazard hierarchy. This is a separate and extremely important issue It is my opinion that many scientists, perhaps including some of the authors of the four articles, are confused and view the causa! hazard hierarchy as extending to ambient PM concentration-response functions. Shin et /.iyi touch on this issue when they note the lowest concentration studied in Ihc American Cancer Society (ACS) cohort was 5.8 pg/m\ the 5th percentile was 8.8 /4g/m\ and the 95th percentile is below 20 f/uon'- They note ``reliable estimates of risk from the available studies can only be made using the data in the 5th to 95th percentile of exposure, i.e,, estimates of ihe shape in the lower 5th and upper 95th percentile are both imprecise and likely to be inaccurate." I question the implication that the sta tistical association between ambient concentrations of PM; 5 and excess risk is equally reliable over the full range from the 5th to the 95th percentile of PM; concentrations. It was disappointing that Shin et al. did not more rigorously address the basis for their fo cus on the 5th percentile in view of EPA's approach to the last N \AQS revision. Specifically, it would have been of interest to readers it Shin et al.1 ' had offered a rigorous cri tique of the related methodology used by the EPA Administrator to make the policy decision lower ing the annual PMj< NAAQS from 15 to 12 ;tg/m effective from March 18, 2013. ' In reaching that policy decision, the final rule stated "In consider ing the evidence, the Policy Assessment recognized lhat NAAQS arc standards set so as to provide req uisite protection, neither more nor less stringent than necessary to protect public health, with an adequate margin of safety. This judgment ultimately made by the Administrator involves weighing the strength of the evidence and the inherent uncertainties and lim itations of that evidence." As summarized in the Fi nal Rule fur the PM NAAQS.1 ' the Administra tor gave special attention to four multicity studies for which distributional statistics of I'M; ; ambient concentrations were available This did not include the Harvard Six (flies Study, for which the Lepeulc et al.1'1" paper is the last update apparently, because the investigators would not release their data on anv btent PMj concentrations for the populations stud ied in six cities. The Rule noted: "By considering this approach one could focus on the range of PM. concentrations below the long-term mean ambient concentrations over which we continue to have con fidence in the associations observed in epidemiolog ical studies (e.g,, above the 25ih percentile) where commensurate public health protection could be ob tained for PM2 5-related effects and. conversely, iden tify the range in the distribution below which our Invited Commentary 1761 confidence in the associations is appreciably less, to identify alternative annual standard levels." It is clear that this approach accepts the categorization of some long-term exposure studies as evidence of a causal or likely causal relationship foi all-cause mortality; however, only above the 25 percentile of ambient PMis concentrations in the four studies. Most impor tantly, the EPA Administrator viewed the evidence below the 25th peicentile as uncertain and not sup portive of a causal or likely causal relationship. This contrasts w'ith the conclusions of Shin et al.tM It is very' likely that this issue will be raised again in the next review of the PM2 5 NAAQS. This is a critical issue at the interface between scientific information and policy choices. It is important to recognize that each review' docs not have to necessarily conclude with a revision of the NAAQS. 6. ASSOCIATIONS VERSUS CAUSALITY AT LOW P\T..5 CONCENTRATIONS All four of the articles most often referred to the "association" between ambient PM?.j and health re sponses. Unfortunately, the tone of three of the arti cles was that this association represented a causal re lationship. As revealed in the earlier discussion of the EPA approach to setting the PM? 5 (annual) NAAQS at 12 /tg/m\ it is important to not assume that causal ity extends to the lowest ambient PM^ concen trations studied based on a Linear model and the lowest ambient PM2 5 concentrations studied. At a minimum, this issue deserves rigorous discussion and debate. Unfortunately, none of the articles contain a ro bust discussion of the many biomedical uncertainties inherent in ambient PM35 concentration-response associations over a range of ambient PM? * concen trations. These uncertainties are multifold, including the official assumption in the last EPA review that all PM? s is of equal toxicity on a mass basis. The assumption of equal toxicity is especially uncertain when one recognizes that PM reduction strategies have been highly effective in the United States over the past half-century in reducing mass emissions and reduced ambient concentrations of PMw and PM? 5. These reductions have resulted in a shift from PM resulting from direct emissions to PM formed from secondary reactions and associated changes in the chemical and size composition of PM. It is impor tant to recognize that these changes arc embedded in the ambient PM concentration data used in the ma jor long-term epidemiological studies with the ambi ent PM for the earliest time periods in the studies being different from the ambient PM for the most recent updates of the studies. Unfortunately, speciated PM? > data have rarely been obtained over long periods of lime at multiple monitoring sites. Data on speciated PM?, aie necessary to test hypothe ses on whether different PM? ; components have dif ferent potencies for causing an increase in differ ent health effects. A closely related issue is whether ambient PM? s concentration-response functions de rived from the study of populations in one part of the United States are applicable to populations in other parts of the United States. The importance of this is sue was underscored by the results reported by Zeger etalJ16' They found an association between increases in PM? 5 and increases in mortality in the eastern and central regions of the United States and no evidence of an association in the western United States for the period 2000-2005. It is also important to recognize that the U.S. populations studied in recent decades w'ere not likely exposed to PM of the composition and high concentrations encountered today in some countries such as China and India. The Shin et al.1-1 article has the most extensive discussion of the issue of causality. However, in my opinion, much of this discussion is quite simplistic and. indeed, naive with regard to the actual com plexity of disease processes. This is illustrated with the statement: "`There is now experimental and clini cal evidence that exposure to fine particulate mattei causes biological responses such as oxidative stress leading to chronic inflammation, which in lurn, can lead to increased mortality from chronic cardiovascu lar and respiratory disease and lung cancer, thereby shortening the lifespan." In my opinion, this is an excessivelv broad conclusion. I would agree that oxidative stress is one of the current fads in the biomedical sciences: however, such fads come and go. Unfortunately, disease processes are much more complex than this statement indicates, and a sin gle step in complex multistep disease processes has rarely proved to be overwhelmingly dominant across a population afflicted with a particular disease. Shin era/.''* use the term "causalmodels" at several places in their article, including reference to the paper of Pope et al.11 These modeling exercises are useful; however, the models fall short of describing the myr iad of complex steps by which responses over many decades to a single risk factor, such as PM?>. unde fined as to chemical composition, cause a very small increase in the relative risk of death from a particular disease in a large population. 1762 McClellan The Shin et a!2'1 article contains what I view as an unjustified statement that: "There is no biological reason to believe that there exists a range in expo sure for which no mortality risks exists (i.e., thresh old).'' It is noteworthy that the authors provided several figures in which data were plotted as hazard ratios or relative risk. The above quote apparently fails to recognize that the hazard ratio or relative risk of 1.0 is not an absence of mortality, it is the baseline mortality rate against which an increase in mortality attributable to the putative risk factor being exam ined is evaluated, in this case--PM? ?--after attempt ing to control for all other risk factors potentially associated with the disease endpoint of concern. The diseases that are of concern for chronic exposure to PM?.5 are very common causes of death (recall Table I) and arise from multiple risk factors. For deaths occurring late in life, many of these risk factors have interacted with normal biological pro cesses. including damage, repair, and homeostatic processes, for decades throughout the individual's life. At the risk of sounding trite, life from conception to death is full of competing risks. The challenge for biomedical scientists, including statisticians, is to de termine under what PM. exposure conditions over a lifetime of exposures there is a significant role for PM? in altermg those complex processes and im pacting morbidity and mortality rates. The challenge is even more difficult because manv of the risk factors identified to date for the diseases of concern do have impact over the individual's total lifetime. As noted earlier (Fig. 1), there has been continuous improve ment in mortality rates in the United Stales over the past half-century. Attempting to tease out the rela tive importance of a multitude of risk factors for this improvement in health is complex and beyond the scope of this commentary. In this commentary, 1 have not discussed a grow ing body of evidence of a lack of influence of am bient PM?5 concentrations on mortality. An ex ample is the paper by Greven et that uses ambient PM? monitoring data for 2000-2006 and data on time of death and age for 18.2 million individ uals in a Medicare cohort. They developed both na tional and local coefficients to exarmne trends. Based on the local coefficient alone, they were not able to demonstrate any change in life expectancy for a re duction in ambient PM-? These results suggest the need foi caution in using national values for esti mating PM25 attributable effects in specific regions of the United States, including California. In this regard, a number of studies have been developed on California populations, some of which suggest an absence of excess risk for recent ambient PM;? concentration.'1619) It is well recognized by scientists and clinicians knowledgeable of the biology and pathobiology of the health endpoints of concern that none of the individual cases carry "markers" or any characteris tics that allow PM;? attributable cases to be distin guished from cases that are attributable to a myriad of other causes. The attribution of deaths associated with PM? ? exposure is done on a statistical and pop ulation basis. The statistical models used typically are proportional hazard models that estimate for the population a given portion of the cases over and above the baseline mortality rates attributed to other causes. To provide a context for considering the es timated PM?s attributable deaths, it is always help ful to present the baseline mortality rate, which, as discussed earlier, varies with time, place, and popu lation as influenced by multiple factors. 1 will return to that point later. In addition to showing the excess risk attributed to PM? 5, it would be informative if the analysts also showed the excess risks estimated for other well-recognized risk factors, such as smoking and socioeconomic status, that must be controlled for in the analyses to develop reliable estimates of excess PM?? attributable risks. This information would be valuable to the analysts and to other parties to help understand if the calculated results for PM?? make sense. An array of attributable risk results for dif ferent risk factors also provides valuable context for policymakers and the public concerned with how best to positively impact human health. In my opinion, it is important to periodically recall the goal to improve public health; the regulation and control of specific risk factors such as PM; 5 is just one means to that end. 7. EXPANDED PRESENTATION OFRESUI TS TO PROVIDE CONTEXT In this section, I will illustrate how an ex panded presentation of results can provide useful context and perspective. Fann et a/.<21 use their Fig. 3 to graphically illustrate the estimated premature deaths avoided" based on different ambient PM- ? concentration-response functions. The locus is on comparison of the results using functions from the Harvard Six Cities Study'1 and the ACS study.'20' The graph also showed estimates developed from Invited Commentary 1763 lalilt- II. Comparison of 2014 Estimated Premature Deaths Avoided Using Alternative Ambient PM3 5 Concentration-Response lunctions (Adapted from Fann ci nL,-) and Fann |pei serial communication]) Source of Function Baseline Mortality (Deaths) Estimated Premature Deaths Avoided (Deaths 95% Confidence Interval) I larvard Six Cities t.epeule etnL<,tn AC'S Krewski el at. (200V) Pooled experts Mela-analvsis (beyond 2006) Mcta-anal\sis (through 2006) Integrated exposure response 2.565.1691 2.565.160'' 2-565169- 2,565.169' 2.565.169364.408* 10,373 (6.010.14,698) 4,582 (3.334, 5.821) 8.327 (1.492.18289) 5,852 (2.527. 9.150) 5,530 (3,287. 7.756) .3.931 (1,935.4241) 1 All cause. ''Ischemic heart disease. functions elicited from 12 experts, a meta-analyses of literature through 2006 and beyond 2006. and a pooling of the 12 experts based on all-cause mor tality. Also shown is an estimate from an integrated exposure-response analysis for ischemic heart dis ease. In I'able II. the original estimates of Fann et are shown complemented by baseline mor tality data added to provide context. Fn my opin ion. showing the baseline mortality values helps the reader to understand this mathematical exercise. The table would be even more informative if it included the total population for 2014. Smith141 provides an excellent example of how the assumptions used in estimating benefits can have major impact on the results In her paper. Table 1 showed the total risk reduction estimate (avoided premature deaths in 2020) for two approaches. One approach was the traditional approach used by EPA in developing regulatory impact analyses (RIAs). That approach assumes deaths are avoided irrespective of the ambient concentrations of PM-,. Table III yields 456 avoided deaths with the national concentration-response function that was developed by Krevvskt et al.Cd>) using the ACS cohort and 1.034 avoided deaths using the concentration response function that was developed by Lepeule et al.`'w from the Six Cities Study. Smith*41 also gave lower estimates based on the approach that EPA used in the latest revision of the NAAQS for PMj? described earlier in this commentary. As shown in Table III. the number of residual avoidable deaths is reduced to 21-47. dependent on the concentration response function used and involves an impacted population of less than 1 million. Alleged benefits in the RIA, of 456-1.034 (or 460-1,000 using the RIA's rounding convention) avoidable deaths, disappear if one uses the qualitative policy judgment used by the EPA Administrator in revising the NAAQS for PM- s. Indeed, a strong argument can be made that there are no avoided PM;s attributable deaths in California based on the report of Zeger et n/.(lr" Recall that they reported no finding of evidence of an association between ambient PM; - and mortality in the western United States. They noted "this lack of association is largely because the Los Angeles Basin counties (California) have higher PM levels than other West Coast urban centers hut not higher adjusted mortality rates." As an aside. California was the only state for which benefits of TiMr III. Estimates of Avoided Premature Deaths in California in 2020 Estimated for PM? 5 N'A AOS with a Reduction in the Annual Standard from 15 to 12 /ig'nr' Ptci|eeled Using BetiMAF121* and Smith (personal communication) Population Baseline Mortality (#) Avoided Deaths (#} Krewski3 Lepeule*' Krewski Lepeule Krewski Lepeule Not attaining/above margin (>13Mg/mJ) Not attaining/in margin {>12-13 u g'm3) Already attaining (<12 /ig'mJ) Total (30-99) 763,104 3,841,464 7.560,163 12.164.732 (25-99) 875.086 4,419.703 8437.984 13.832.773 (30-99) 7.574 41,853 86,913 136440 (25-99) 7.681 42 342 87.735 137,758 (30-99) 21 117 318 456 (25-99) 47 266 721 1.054 `Krewski et alP1^ evaluate the population from age 30 to 99 years, ''t.epeule el n/""' evaluate the population from age 25 to 99 years. 1764 McClellan avoided mortality were projected to occur with a low ering of the PM?; NAAOS. Other areas had already attained the PM? 5 NAAQS. Again, the inclusion of the baseline population and mortality data helps provide context and perspective to the calculated benefits. Note that the population and baseline mortality values are based on actual data rather than hypothesized relationships and, thus, are much more certain than the calculated benefits. This broader array of data not only gives perspective to the calcu lated benefits, i.e,, avoidable deaths, for a PM2.5 stan dard, but invites questions as to where society at large can gain the greatest benefits in improved health. To give a broader perspective to the estimated avoidable deaths, it is useful to recall Table I. which provides detailed mortality data by causes for 2010. As discussed earlier, consideration of calculated esti mates of PM? 5 attributable deaths along w ith an ar ray of mortality data by multiple causes opens the door to a broader discussion of options for improv ing the health and quality of life for society at large moving beyond a singular focus on PM? 5. The above discussion has focused on providing information on three key inputs: (1) the population under consideration, (2) baseline mortality rate, and (3) the ambient PM s concentration response functions (and the associated uncertainties at var ious PM?s levels). Tt is also useful to have a more complete exposition of the ambient PM? , data being used as input as illustrated by Smith. The above discussion also carries with it impor tant implications for setting priorities for research that wall help improve human health. Let me first address the adequacy of current models of ambient PM? 5 concentration-response functions. In my view, the models currently available provide reasonable upper-bound estimates of PM > attributable mor tality, i.e., more likely to overestimate than under estimate the true PM?.? attributable mortality The estimated ambient PM; 5 concentration-response functions and PM;; attributable mortality calculated for those studies are likely related to the exposure of the populations over a lifetime beginning early in life, i.e., in the 1970s and earlier for the vast majority of the deaths. Ambient concentrations of PM? 5 have steadily declined across the United States from that time to the present; recall Fig. 1. In addition, the U.S. age-adjusted death rate has steadily decreased, as shown in Fig. 2, related to many factors. Let me quickly note that some individuals may suggest that improved air quality had a role in the observed reduced death rates. That may be true; Table IV. The Impact of Socioeconomic status on Mortality Rate R,alio ( Adapted from Steenland cl it/.*22') Mortality Men Women All causes Heart disease Stroke Diabetes COPT) Lune cancer Breast cancer Colorectal cancer External causes 2.92 (1 95-209)' 1.88 (1.83-193) 2.25 (2.14-2.37 2.19 (207-2 32) 3.59 <335-3 83) 2.15 (2.07-2.23) - 1.21 (1.16-1 27) 2.67 (2.58-2.78) 1.29 (1)25-132) 1.84(1.76-1.93) 1.55 (1.44-162) 1.85(1,72-2 00) 3.0*4 < 1 q|_?l(J) 1.31 (1.25-1.39) (1.76 (0.73-0.79) 0 91 (0.S6-0.96) 1,41 (1 35-1 48) Sole Mortality rate ratio = mortality for lowest quartile of socioe conomic status Mortality for highest quartile of socioeconomic status. "95% Confidence interval. however, I suggest the impact of PM?? reductions is likely very small and difficult to tease out from the myriad of other factors that were likely involved in reducing mortality rates. To provide further perspective, it 1$ useful to note the substantial impact of socioeconomic status on mortality122> (Table IV). The mortality rate ratio for the lowest quartile over the highest quartile of socioeconomic status is high compared to small changes attributed to PM s. It is obvious that many individual risk factors are in cluded within socioeconomic states. All ot these fac tors create "noise" that makes it challenging to iden tify any small signal attributed to PM?; This speaks for caution in interpreting and using the small signals attributed to PM? x in these statistical exercises The overall point 1 wish to make is that disease processes are very complex and are influenced by multiple risk factors. For any attempt to tease out the effects of a single risk factor, like PM;,5, to be success ful it needs to take account of the other risk factors. I urge the investigators who have focused their energy on PM 5 issues to broaden the scope of their research to give greater attention to identifying and char acterizing multiple risk factors. In my opinion, this broader perspective offers the best opportunity for having a positive impact on the health of populations. 8. CONCLUSIONS The information presented in the four articles and discussed here also has important implications for setting future PM NAAQS and for research to better understand mechanisms of disease causation, approaches to mitigation of disease, and treatment of Invited Commentary 1765 disease. A review of the data presented here, with a focus on the United States, indicates that any health effects attributable to I'M:, are quite small when considered in the context of the total disease burden, A corollary' is the need for caution in advocating for more PM; 5 focused research. In my opinion, a better return on societal investment is likely to come from a broader consideration of the complex pathways of disease causation common to multiple risk factors and, perhaps, amplified by certain risk factors, REFERENCES 1 Anenberg SC, Belova A. Riandt J, Kami N, Greco $. Guliikunja S. Heroics Mfc Hurley F Krzyzanowski M, Medina S, Millet H pandey K, Ross J, Va Dm pencil R Survey of air pollution health usk assessment tools Risk Analysis, 201ft-, 36{9):171S--1736. 2 Fann N, Gilmore HA, Walker K Characterizing the long term PM concentration-response function: Comparing the strengths and weaknesses of research synthesis approaches Risk Analysis, 2015: 36<9):1693-1707 3 Shin HH. Cohen At. Pope CA Ezzati M, Lim SS, Hubbell BJ. Bui nett RT Meta-analysis methods to estimate the shape anti imcertainty in the association between long-term exposure to ambient fine particulate matter and cause-specific mortality over the global concentration range. Risk Analysis, 2016; 36(8) (this issue ). 4 Smith ALL Inconsistencies in risk analyses for ambienl air pol lutant regulations Risk Analysis 201 b: 36(91:1737-1744. 5 C'AA I he Clean Air Act of i*i63. 77 Stal. 3<>2: 1763 ft CAA. The Clean Atr Acl Amendirlc.nlsof 1070, 84 Stal. 1676, P.I 71-604 (December 31.19 70J. Plus technical amendments made by P 1 72-157 (November IK. 1771). 1970. 7 CAA. Clean Air Act Amendments of 1990. 104 Slat 2468, PL 10L-549,1990. 8 Bachman J Will the. circle be unbroken: A history of the U S National Ambient Air Quality Standards. Journal of the Air & Waste Management Association. 2007;57:652-697. 9. Whitman vs American Trucking Associations. 531 U S. 457, 121 Supreme Court 903149 L Ed. 2<t 1.2001 It). Lepcule J. Laden F, Dockery D, Schwartz J Chtonic exposure to fine particles and mortality: An extended follow-up of the Harvard Six Cities Study from 1974 to 2009. Environmental Health Perspectives. 201?: 32(1):,31-95. 11. Hoycrt DL. 75 Years of Mortality in the United States. 1935 2010. National Center tor Health Statistics Data Brie I No. 88, Hyatlsville. MD.March 2ul2. 12 Murphy SL Xu J. KocharlekKO Deaths: Final data for 2tUH National Vital Statistics Report. 2013; 6i| "I): 1--117 13, McClellan RO. Role of science and judgment in setting National Ambient Air Quality Standards How low is low enough? Air Quality. Atmosphere and Health 20] 2: 5:243 258. ' 14. F.PA (U.S. Environmental Protection Agency) Integ'aled Science Assessment for Particulate Matter Final Report Re search Triangle Park. NC: National Center for Environmental Assessment - RTP Division, Office of Research and Develop ment, EPA/600/R-08/139F. December 2009. Available at: http //vvww epa.gov 'tintnaaqs/staretards pm A_pm_2flft7_isa_h tml. Accessed September 6, 2016, 15 FPA (U,S. Environmental Protection Agency), National Am bient Air Quality Standards for particulate matter f inal rule Federal Register 2012; 78:3086-3287. 1ft. Zegcr SL. Dominki F. McDermott A. Samel JM. Mortality in ihe Medicare population and chronic exposure to line panicu late ait pollution in urban centers (2000-2005). Environmental Health Perspectives,2008: 116:15)4-1619 17. Pope CA III. Burnett RT. Krewski D Jerrett M Shi \ Calk EE Thun MJ. Cardiovascular mortality and exposure to fine paniculate matter from air pollution and cigarette smoke Shape of the exposure-response relationship. Circu lation. 2009:120:941-948. 18. Grevcn S. Dominki F. Zegcr S, An approach to ihe estimation of chronic air pollution efleets using spatio-temporal inlorma lion. Journal of ihe American Statistical Association. 2U11 J06:396-406. 494 19 Fnslrom .IF bine particulate air pollution and total mortality among eldertv Californians 1783-7092. Inhalation Toxicology. 2005; 17 8113-81ft 20. Krewski D, Jerrett M. Burnell RT. Mn R, Hughes E. Shi Y, Turner Ml Pope CA 3rd. Thurston CL Calk I F Hum MJ, Beckerman B. Def.uca P, Finkelstein N, Jto K. Moore DK. New-bold KB. Ramsay T. Ross L, Shin H. Iempalski B. Ex tended follow-up and spatial analysts of the American Can cer Society Study linking particulate air pollution and mortal ity. Research Report Health Effects Institute. 2009; 140:5-114, Accessed September 6,2016 21. BenMAP Version 4 0.67. Available at: http://www.epa.gQu air, benmap/download.hlml 2? Steenlaml K 1 lu 5 Walker J. All-cause and cause-speeiiic mortalily by socioeconomic status among employed persons in 27 US states, 1784-13)77 American Journal of Public Health, 200-4: 94(6): 1037-1042, EXHIBIT C Inhalation Toxicology. 17:803 816. 2005 C opyright <> Taylor ami Francs* Inc. ISSN 0895-8378 pnni/ 1091-7691 online DOI 10 1080/089583 70500740413 Taylor & Francis Uyto t> frr*c- Croup Fine Particulate Air Pollution and Total Mortality Among Elderly Californians, 1973-2002 James E. Enstrom Jmuson Comprehensive Cancer Center. University of California, Ixts Angeles. California, USA, and Scientific Integrity Institute, Los Angeles California, USA Fine particulate air pollution has been associated with increases in long-term mortality in selected cohort studies, and this association has been influential in the establishment of air quality regulations for fine particles ( PM1.5). However, this epidemiologic evidence has been questioned because of methodological issues, conflicting findings, and lack of an accepted causal mechanism. To further evaluate this association, the long-term relation between tine particulate air pollution and total mortality w as examined in a cohort of 49,975 elderly Californians, w ith a mean age of 65 yr as of 1973. These subjects, who resided in 25 California counties, were enrolled in 1959. recontacted in 1972, and followed from 1973 through 2002; 39,846 deaths were identified. Proportional hazards regression models were used to determine their relative risk of death i RR i and 95% confidence interval (Cl) during 1973-2002 by county of residence. The models adjusted for age. sex. cigarette smoking, race, education, marital status, body mass index, occupational exposure, exercise, and a dietary factor. For the 35,789 subjects residing in 11 of these counties, county-wide exposure to fine particles was estimated from outdoor ambient concentrations measured during 1979-1983 and RRs were calculated as a function of these PM!S levels (mean of 23.4 pg/nr'). For the initial period. 1973-1982, a small positive risk was found: KR was 1.04 (1.01-1.07) for a lO-pg/m1 increase in PM?For the subsequent period, 1983--2002, this risk was no longer present: RR was 1.00 (0.98-1.02). For the entire follow-up period. RR was 1.01 (0.99-1.03). The RRs varied somewhat among major subgroups defined hy sex, age. education level, smoking status, and health status. None of the subgroups that had significantly elevated RRs during 1973-1982 hud significantly elevated RRs during 1983-2002. The RRs showed no substantial variation by counts of residence during any of the three follow-up periods. Subjects in the two counties with the highest PM; s levels (mean of 36.1 f/g/nr > had no greater risk of death than those in the two counties with the lowest PM: 5 levels (mean of 13.1 pg/m' i. These epidemiologic results do not support a current relationship between fine particulate pollution and total mortality in elderly Californians, Hut they do nof rule out a small effect, particularly before 1983. Received 28 February 2005, accepted 21 June 2005 The extended mortality follow-up and analyses piesented in this ar ticle have been funded by the F.lectnc Power Research Institute (EPR1). The entire funding history of C \ CPS I pnor to this analysis has been described elsewhere < Enstrom & Heath. 1999; Enstrom & Kabat. 2003). The author is responsible for all aspects of the article and declares no competing interests relevant to its contents. In the spirit of the Data Quality Act (OMB, 2003; Steinbrook, 2004), the author is willing to facilitate independent analysis of all the data used in the article. The author thanks Dr. Frederick W. Lipfert for proposing air pollution anal yses with CA CPS I data and for contributing extensively to numerous versions of the text and tables. Dr Ronald E. Wyzga for critiques of the article and for suggestions about making it as comprehensive and objective as possible, and Dr Lingqi Tang for statistical assistance. Address correspondence to James E Enstrom. University of California, Jonsson Comprehensive Cancer Center. Los Angeles, CA 90095. USA. E-mail; jenstrom@ucla.edu Many observational epidemiological studies have reported associations between air pollution from combustion sources and human health (Lipfert, 1994) During past severe air pollution events, such as the 1952 London fog incident (Logan & Clasg, 1953), extremely high concentrations of particulate air pollution were accompanied by major increases in coincident mortality In more recent years, health effects have also been associated with much lower concentrations of particulate air pollution (Pope & Dockery', 1999) While much of the recent research has focused on short-term exposures, several studies suggest that long-term exposures may be more important. In particular, results from two major cohorts (Dockery etal., 1993; Pope etal, 1995.2002) have shown significant mortality associations with outdoor concen trations of fine particles (PM: 5. median aerodynamic diameter 803 804 J. E ENSTROM less than 2.5 fi m) Other cohort studies have also examined mor tality associations with PM? 5 and other pollutants (McDonnell et aL 2000; Lipfcrt et al.. 2000, 2003). with somewhat different findings. The major cohort studies have been used to support new na tional ambient air quality standards for fine panicles issued by the U S. Environmental Protection Agency (U.S. EPA, 1997). These standards are specific with respect to particle size, but not with respect to chemical composition PM? 5 is a variable mixture, rather than a defined chemical compound as in the case of gaseous air pollutants. Fine particles are generated mainly by combustion processes and their atmospheric sequelae, and all such particles measured by the approved methods are con sidered equally harmful. However, the chemical composition of airborne particulate matter varies appreciably across the nation and within metropolitan areas. Although national ambient air quality standards are intended to apply throughout the ration, it is not clear that the selected epidemiological studies on which those standards are based are equally applicable nationwide The associations of particulate air pollution with long term mortality remain controversial (Phalen. 2002; Moolgavkar. 2005; Kaiser, 2005). This is in large part because the epidemio logic studies that have examined these health effects are subject to a number of methodological limitations (Greenbaum et al., 2001; Moolgavkar, 1996; Gamble. 1998; Krewski et al., 2000; Lipfert, 2003). Actual exposures to air pollution arc difficult to determine accurately in large cohorts. Indeed, the exposure of each individual has not been directly measured in these stud ies, but has been assumed to equal the ambient outdoor PM? 5 concentration for the individual's county or metropolitan area of residence. Also, one national cohort study has found largely neg ative associations between PM? 5 and mortality (Lipfert et al., 2000, 2003). California is a large, diverse state that has long been con cerned about the health effects of air pollution and that has recently issued new stricter ambient PMjs standards (CARB, 2003), based in large part on the national standards. However, no previous cohort study has focused on mortality with respect to measured PM? 5 levels in California. This article used a large co hort of elderly Californians to examine in detail Ihc relationship between PM?s levels measured during 1979-1983 and mortality from all causes during 1973-2002. METHODS California Cancer Prevention Study The California Cancer Prevention Study (CA CPS 1) is the extended follow-up of the 118,094 California subjects from the original Cancer Prevention Study (CPS I) of 1,078.894 adults from 25 states. CPS 1 was initiated by the American Cancer So ciety (ACS) beginning in 1959, and CA CPS I has been indepen dently conducted at the University of California. Los Angeles (UCLA), since 1991, as described in detail elsewhere (Enstrom (fk Heath, 1994; Enstrom & Rabat. 2003). The conduct of CA CPS I has been approved by the UCLA. Office for Protection of Research Subjects during this entire period. The subjects in this prospective cohort study were enrolled from October 1959 through February' 1960 using a detailed four-page questionnaire. Surviving cohort members completed short questionnaires in late 1961. 1963, 1965, and 1972, and a two-page questionnaire in mid 1999. Deaths through 1972 were identified primarily by surviving study subjects and were confirmed with death certifi cates. The later deaths were identified primarily from comput erized and manual matches with the California death file and the nationwide Social Security Death Index, using name and other identifying variables (Enstrom & Heath, 1999; Enstrom & Rabat. 2003). About 86% of the later deaths were identified on the California death file and the remainder were identified on the Social Security Death Index file. The only pnoT analysis of the CPS I cohort with respect to air pollution found no relationship between suspended particulate matter and lung cancer mortal ity during the 1960s and had no PM 15 data (Hammond, 1972; Hammond & Garfinkel. 1980). This article analyzes those CA CPS 1 subjects who reported their cigarette smoking status in both the 1959 and 1972 ques tionnaires and who were alive as of January 1, 1973 Respon dents to both questionnaires were traced more easily than those who responded only to the 1959 questionnaire. The 1972 ques tionnaire updated their cigarette smoking status, the most impor tant confounding variable. The early years of follow up (January I. I960-December 31, 1972) have not been included in this ar ticle because there are no statewide PM? s data before 1979 Results for this early period and for the entire follow-up period since 1960 will be presented in a subsequent article dealing with other air pollutants This analysis is limited to the 25 counties with the largest number of CA CPS I subjects, which ranged from 325 to 17.340 per county. About 95% ol the CA Cl'S I subjects resided in these 25 counties. TTiere were 49,975 trace able subjects alive as of January 1. 1973, of whom 39.846 died as of December 31. 2002. There were 35,789 traceable subjects alive as of January 1.1983 in ihe 11 counties with PM? 5 data, of whotn28.441 died as of Deceriibet 31.2002. An additional 2.735 subjects in these counties lost since January l. 1973 (7.6-1%), have been omitted from further analysis. The 1999 addresses for most of the traceable subjects alive as of January 1, 1999. were determined from a match with California driver's license (Dl.) identifying information, and about 33% of the subjects responded to a two-page smoking and lifestyle questionnaire that was mailed in mid 1999 to their DL address (Enstrom & Heath, 1999; Enstrom & Rabat, 2003). Based on the questionnaire information in late 1972. the 1999 DL address information, and the death information, the county of residence and county of death were determined for most subjects as of late 1972 and early 1999. The residential mo bility of subjects was assessed by calculating the percentage of subjects who lived or died 111 the same county from 1972 to 1999. FINE PART1CUI ATE POl J .1 "HON AND MORTAIITY 805 1979-1983 PM2 5 Data and 1973-2002 Mortality Data The independent variable in this analysts is PM7.5, as mea sured during 1979-1983 in 11 California counties by the EPA as part of the Inhalable Particulate Network (1PN) {Hinton eta! , 1984, 1986), also known as the Inhalable Particle Monitor ing Network (IPMN) (Sune, 1999: Pope et ah, 2002) These data have been used in several previous epidemiological studies (Ozkaynak et ah. 1987; Lipfert et ah, 1988, 2000, 2003; Pope et ah, 1995,2002). In this article, the PM; s data for each county were averaged over lime and across the available monitoring stations, and are assumed to indicate the average long-term ex posure level for all subjects in the county. No routinely measured PM2.5 data in California exist before 1979 or during 1984-1998; routine statewide measurements in California were resumed in 1999. The average county-level PM2.s value was assigned to the traceable subjects alive as of January 1. 1973, based on their county of residence as of late 1972. This analysis was based on the deaths from January 1, 1973, to December 31, 2002, a 30-year follow-up period that includes the 5-yr period of the 1979-1983 PM2.5 data. Additional analyses have been done for deaths from January I, 1973, to December 31, 1982, and from January 1, 1983, to December 31. 2002. This latter period is roughly the same as the period (September 1. 1982 December 31. 1998) used in the recent national cohort study (Pope et ah. 2002). Analysis by Proportional Hazards Regression The age- and sex-adjusted relative risk of death (RR) and 95% confidence interval (Cl) were calculated using Cox pro portional hazards regression, specifically the SAS PHREG pro cedure (SAS. 2004), including age at baseline in 1-yr intervals and sex, as a function of PM; 5 level in units of 10 jzg/nr. This type of analysis is similar to the one used recently (Pope et ah, 2002). Fully adjusted relative risks were calculated using a Cox model that includes age, sex, and eight potential confounding variables at baseline: cigarette smoking status (never, fonneras of 1959 and 1972, 1-9. 10-19, 20, 21-39, 40+ cigarettes per day as of 1972), race (white, nonw+ite), education level (<12. 12, >12 yr), marital status (married, widowed, single, sepa rated, divorced), body mass index (<20, 20-22.99, 23-25.09, TABLE 1 Demographic and lifestyle characteristics in 1959 for California CPS I male subjects as of 1/1/1973 who resided in the 11 counties having 1979-1983 PM2-i measurements and who provided 10/1/1972 cigarette smoking status Characteristic 1959 value (11 PM2 5 counties) 1959 value (2 highest PM25 counties) 1959 value (2 lowest PM21 counties) 1959 value for 1999 respondents 1999 value for 1999 respondents Mean level of 1979-1983 PM; 5 (/ig/m*) Number of subjects alive as of 1/1/1973 Lost to follow-up since 1/1/1973 (%> Number of subjects alive 1/1/1973 and not lost since 1/1/1973 Age as of 1/1/1973 (mean, years) Age as of 1/1/1983 (mean, years) Race (% white) Marital status (%> married) Education (% >12 yr) Height (mean, inches) Weight (mean, pounds) History of serious diseases (% yes) Cancer Heart disease Stroke Sick at the present time (% yes) Occupation (% professional) Residence location (% urban) Exercise (% moderate or heavy) Cigarette smoking (% current in 1959) Cigarette smoking {% current in 1972) Fruit/fruit juices (7+ times/week) 23.4 16.296 4.432 15,574 65,7 73.8 98 4 97 3 71.8 69.4 173 0 9.7 4.6 4.6 0.6 6.8 10.5 98.1 72.5 41.5 23.3 63 2 36.1 1043 2.987 1012 67.1 74.9 99.0 97.4 70.7 69.5 172.7 12.7 6.9 4.6 1.2 6.8 11.9 99.2 73.9 40.5 24.2 60.6 13.1 1040 4.840 990 64.5 72.4 97.5 98.0 79.8 69.9 174.5 7.5 3.8 3.5 0.2 5.4 9.8 98.6 76.1 45.3 25.9 63.7 1314 (alive 1999) 58.4 68.4 98.5 96.3 90.3 69.9 172.9 4.6 3.1 1.2 0.3 6.2 17.5 98.1 67.1 41.9 14.9 66.2 1314 (alive 1999) 58.4 68.4 980 75.6 92 6 69.3 168 9 42.5 25.9 61.7 1.8 1.8 59.0 Note. Values in 1959 and 1999 for male subjects in 11 counties who responded to 1999 questionnaire. 806 J. E. ENSTROM TABLE 2 Demographic and lifestyle characteristics in 1959 for California CPS I female subjects as of 1/1/1975 who resided in the 11 counties having 1979-1983 PM, 5 measurements and provided 1972 cigarette smoking status Characteristic 1959 value (11 PM, 5 counties) 1959 value (2 highest PM, s counties) 1959 value (2 lowest PM, 5 counties) 1959 value for 1999 respondents 1999 value for 1999 respondents Mean level of 1979-1983 PM, , < // g/m3) Number of subjects alive or lost as of 1/1/1973 Lost to follow-up since 1/1/1973 (9r) Number of subjects alive 1/1/1973 and not lost since 1/1/1973 23.4 22,228 9.058 20,215 36.1 1491 9.276 1353 13.1 1313 10.252 1178 2877 2877 (alive 1999) (alive 1999) Age as of 1/1/1973 (mean, years) Age as of 1/1/1983 (mean, years) Race (% white) Marital status (% married) Education (% >12 yr) Height (mean, inches) Weight (mean, pounds) History of serious diseases (% yes) Cancer Heart disease Stroke Sick at the present time (% yes) Occupation (% professional) Residence location (% urban) Exercise (% moderate or heavy) Cigarette smoking (% current in 1959) Cigarette smoking (% current in 1972) Fruit/fruit juices (7+ times/week) 64.9 72.9 98.3 83.1 76.8 63.8 137.1 9.9 5.9 3.5 0.5 8.4 15.9 97.7 80.2 32.5 22.8 74.1 66.3 74.3 99.3 81.8 76.6 63.8 138.4 10.6 7.0 3.0 0.6 8.1 21.4 99.1 82 8 28.7 196 75.4 64.0 72.0 97.9 86.3 82.6 64.0 135.3 9.8 6.0 3.1 0.7 5.6 16.0 98.0 83.3 404 29.1 74.3 Note Values in 19V) and 1999 for female subjects in 11 counties who responded to 1999 questionnaire 57.1 67.1 98.1 90.0 89.9 64.1 133.0 5.5 3.8 1.5 0.2 7.2 18.6 96.9 77.9 31.1 19.8 74.8 57.1 67.1 97.7 32.3 93.3 63.5 137 4 36.8 22.0 63.4 3.6 3.6 60.1 26-29.99, >30 kg/nr), male occupational exposure (no, yes), exercise (none/slight, moderate, heavy), and fruit/fruit juice in take (0,1,2,3,4,5,6,7 days/wk). One additional variable, health status at entry (good, fair, pooi. ill. sick/cancer/CHD/strokc). was evaluated in a sensitivity analysis. The confounding vari ables are defined at original entry into study in late 1959, except for curarette smoking status, which was updated in late 1972. All of the confounding variables were measured again in a large sample of survivors during 1999. Subgroup analy ses were done by sex, year of birth (1873-- 1907. 1908-1929, representing ages 43-64 and 65-99 as of January 1, 1973), education level (<12, 12, 12+ yr), cigarette smoking status (never, former, current as of October 1, 1972), and health status (healthy, unhealthy as of October l, 1959), as well as by decade of follow-up (January 1,1973-December 31, 1982, January 1, 1983-December 31. 1992, January 1, 1993December 31, 2002). In addition, the relative mortality rates by county of residence were calculated using PHREG as an alter native method to assess the influence of different county-wide pollution levels. The Los Angeles county subjects arc used as the referent group in estimating the fully-adjusted RRs during 1973 2002, 1973 1982. and 1983 2002 for each of the other 24 counties RESUITS Demographic Characteristics and 1979-1983 PM2 ^ Data The late 1959 demographic and lifestyle characteristics of the CA CPS 1 subjects in the 11 counties w ith 1979-1983 PM,5 data (mean of 23.4/rg/m1) are shown in Table 1 for I5.574males and in Table 2 for 20,218 females. These tables also show the corresponding characteristics for the subjects in the two counties (Kern and Riverside) with the highest PVI2 5 levels (mean of 36.1 /rg/m') and in the two counties (Contra Costa and Santa Barbara) with the lowest PM, s levels (mean of 13.1 /tg/rn '). The characteristics of subjects are quite similar, irrespective of their mean pollution levels. The mean age of the subjects alive as of January 1, 1973, was 65.7 yr for males and 64.9 yr for TABLE 3 Fully adjusted relative risk of death from all causes (RR and 95% Cl) during 1/1/1973- 12/31/2002 and 1/1/1983-12/3 1/2002 for both sexes, by county of residence relative to Los Angeles county lor the 25 counties with the most California CPS I subjects, based on 10/1/1972 county of residence County of residence as of 10/1/1972 1973-2002 Deaths/ 1973 subjects Percent alive or dead same co in 1999 Fully adjusted 1973 -2002 RR(95% Cl) Fully adjusted 1973-1982 RR<95% Cl) Fully adjusted 1983-2002 RR(95% Cl) Alameda Butte Contra Costa Fresno Humboldt Kern Marin Napa Orange Riverside Sacramento San Bernardino San Diego San Francisco San Joaquin San Mateo Santa Barbara Santa Clara Santa Cruz Solano Sonoma Stanislaus Tulare Ventura Los Angeles Total Chi-square test of homogeneity (24 degrees of freedom) 3380/4294 462/534 1260/1652 840/1085 424/507 630/790 641/805 500/627 2453/3050 1311/1575 1370/1721 1340/1622 2958/3615 1597/2043 248/325 1403/1789 411/516 1851/2345 295/372 402/505 482/581 551/691 921/1117 369/474 13,747/17,340 39,846/49,975 60.7 0.962 (0.926-0.999) 0.948 (0.886-1.015) 0.967 (0.924-1.012) 73.1 0.999(0.910-1.096) 0.899(0.763-1.060) 1.051 (0.939-1.176) 60.1 0.999(0.943-1.058) 0.989(0.890-1.100) 1.004 (0.937-1.076) 80.0 0.935(0.872-1.002) 0.896(0.786-1.021) 0 951 (0.876-1.033) 79.3 0.992(0.900-1.092) 0.985 (0.830-1.168) 0.992 (0.882-1.115) 79.5 0.944 (0.872-1.023) 0.950 (0.824-1.096) 0.941 (0 854-1.036) 57.2 0.939 (0.867-1.016) 1.006(0.875-1.158) 0.908 (0.825-1.000) 73.4 0.949(0.868-1.038) 0.817(0.687-0.972) 1.006 (0.906-1.117) 65.1 0.990(0.948-1.034) 0.962(0.891-1.038) 1.003(0.952-1.056) 59.6 0.959(0.906-1.015) 1.022 (0 928-1.064) 0.926 (0.863-0.993) 77.2 0.998 (0.944 1.055) 0 960(0.867-1.064) 1 013(0.948-1.083) 63.5 0.992 (0.938-1.049) 0.932(0.841-1.033) 1.018 (0 951-1.088) 84.4 0.992 (0.954-1.033) 0.911 (0.847-0.979) 1 028 (0.981-1 078) 48.3 0.963 (0.914-1.014) 0.985(0.899 1.080) 0.952(0 894-1.014) 71.9 0.925 (0.816-1.049) 0.847 (0.670-1.071) 0.965(0.832 1 120) 58.0 0.949(0.899-1.003) 0.897 (0.808-0.997) 0.971 (0.910-1.035) 67.4 0.968 (0.878-1.068) 0.832(0.690-1.003) 1.030(0.918 1.156) 63.5 0.955 (0.910- 1.003) 0.961 (0.880-1.049) 0.954(0.900-1.012) 64.7 0.890 (0.793-0.999) 0.980(0.805-1.193) 0.849 (0.736-0.979) 59.8 0.901 (0.815-0.995) 0.823 (0.685-0.989) 0.934(0.830-1 051) 75.7 0.968 (0.884-1.060) 0.919(0.781-1.082) 0.987 (0.885-1.102) 83.7 0.984 (0.904-1.072) 0.981 (0.841-1.144) 0.981 (0.885-1.087) 78.7 1.047 (0.979-1.119) 1.031 (0.918-1.158) 1 054(0.972-1 144) 69.1 0.967 (0.872-1.072) 0.774(0.629-0.951) 1.053(0.935-1 187) 64.4 1 000 1.000 1.000 66.4 X M II II O' X2 = 27.48 p = .283 X2 = 32.21 p = .122 ooc TABLE4 Fully adjusted relative risk of death from all causes (RR and 95% Cl) by county of residence relative to Los Angeles county, during 1973-2002 and 1983-2002 for both sexes, for the 35,789 California CPS 1 subjects in rank order of 1979-1983 PM2.5 level for the 11 counties and groups of these counties County of residence as of 10/1/1972 1/1/1973 subjects Fully adjusted 1973 2002 RR(95% Cl) Fully adjusted 1973-1982 RR(95% Cl) Fully adjusted 1983-2002 RR(95% Cl) PM2.5 Wm1) 1979-1983 Rank Santa Barbara 516 0.968(0.878-1.068) 0.832 (0.690-1.003) 1.030(0.918-1.156) 10.6 \ Contra Costa 1652 0 999(0.943-1.058) 0.989(0.890-1 100) 1.004 (0 937-1.076) 13 9 2 Alameda 4294 0 962 (0.926-0.999) 0.948 (0.886-1 015) 0.967 <0.924-1.012) 14 4 3 Butte San Francisco 534 0 999 (0.910-1 096) 0.899(0 763-1 060) 1.051 (0.939-1.176) 15.5 4 2043 0 963 (0 914-1.014) 0.985 <0.899-1 080) 0 952(0.894-1.014) 16 4 S Santa Clara 2345 0.955(0.910-1.003) 0.961 <0.880-1 049) 0.954 (0.900-1.012) 17.8 6 Fresno 1085 0.935(0.872-1.002) 0.896(0.786-1.021) 0.951 (0.876-1.033) 184 7 San Diego 3615 0.992 (0.954-1.033) 0.911 (0.847-0.979) 1.028(0.981-1.078) 18 9 8 Los Angeles 17.340 1.000 1.000 1.000 28.2 9 Kern 790 0.944 (0.872-1.023) 0.950(0.824-1.096) 0.941 (0.854-1.036) 30.9 10 Riverside 1575 0.959(0.906-1.015) 1.022(0.928-1.064) 0.926 (0.863-0.993) 42.0 n Total 35.789 23.4 Chi-square test of homogeneity X2 = 12.65 p = .244 X~ = 14.27 p = .161 X2 = 16.35 p = .090 (10 degrees of freedom) Two lowest PM2.5 counties 0.991 (0.942-1.043) 0.948(0.863-1 041) 1.010 (0.951-1.073) 13.1 (Contra Costa and Santa Barbara) Next lowest PM2 5 counties 0.965 (0.935-0.996) 0.955 (0.903-1.010) 0.969 (0.933-1.006) 15 1 (Alameda, Butte, San Francisco) Next highest PM2.5 counties 0.971 (0.942-1.002) 0 926(0.875-0.980) 0.991 (0.955-1.029) 18.5 (Fresno, San Diego, Santa Clara) Reference county (Los Angeles) 1.000 1.000 1.000 28.2 Two highest PM2 5 counties (Kern 0.954(0.910-1.001) 1.000(0.921 1087) 0.931 (0.878-0.986) 36 1 and Riverside) Chi-square test of homogeneity X2 = 8.48 p = .075 X2 = 8.72 p = 069 X2 = 8.02 p = .091 (4 degrees of freedom) FINE PARTICULATE POLLUTION AND MORTALITY 809 females ami their minimum age was 43 yr The mean age of the subjects alive as of January 1, 1983, was 73.8 yr for males and 72,9 yr for females, and their minimum age was 33 yr. The 1979 1983 PMyj data from the Il'N arc shown m Ap pendix Table 1 for 11 California counties, with details for the 15 monitoring sites at which measurements were made. The 1999 follow-up questionnaire provided important infor mation about the confounding variables for survivors 40 yr alter they originally enrolled in the study. Although these survivors were the younger members of the cohort, with a mean age of about 57 yr as of January 1,1973, they provide a good indication of the risk factor changes that have occurred. A comparison of their 1959 and 1999 responses in Tables 1 and 2 shows that the v ariables of race, education level, exercise, body mass index, and fruit/fruit juice intake changed very little and were similar in the high and low PM25 counties. The percentage of married subjects declined substantially in ail counties because of the targe frac tion of spouses who died. The percentage of current cigarette smokers declined dramatically and uniformly in all counties, reflecting the large degree of smoking cessation that has al ready been documented in this cohort (Enstrom & Heath, 1999). Health status, used as an additional variable in a sensitivity anal ysis, also declined substantially among subjects in all counties, because the aging survivors had a much higher prevalence of cancer and other diseases in 1999 than they did in 1959. Relative Risks by County of Residence Table 3 shows the 1973-2002, 1973-1982, and 1983-2002 mortality risks relative to Los Angeles county, adjusted for age, sex, and eight confounding variables, for the 25 counties with the most CA CPS i subjects, including the 11 counties with 1979-1983 PM; 5 data. Overall, the RRs were quite consistent w ith each other and most had a 95% Cl that included 1.0. None of the RRs were greatly different from 1 0, Wald chi-square Tests conducted on these RRs did not reject the hypothesis of homogeneity during any of the three follow-up periods. Also, fable 3 shows that, based on their counties of residence and death from 1972 to 1999, about 56% of the subjects remained in the same county during this period, indicating relative stability of residence. In particular, the stability of residence was similar in the two highest PMxs counties (66%), the two lowest PM2.5 counties (62%). and in Los Angeles county (64%). Table 4 shows the mortality risks relative to Los Angeles county for 11 counties ranked in order by their 1979-1983 f'Mi; value. During 1973-2002, the two counties (Kern and River side) with highest PM2.5 levels (mean of 36.1 p,g/rrr) had an RR of 0.954 (0.910-1,001), whereas the two counties (Contra Costa and Santa Barbara) with the lowest PMj 5 levels (mean of 13.1 pg/nv) had a slightly higher RR of 0.991 (0.942-1.043). During 1983--2002 there was a larger difference, with corre sponding RRs of 0.931 (0.878-0.986) and 1.010(0.951-1.073). During 1973-1982 there was a reverse pattern, with correspond ing RRs of 1.000 (0.921-1.087) and 0.948 (0.863-1.0411. Two groups of the six counties with medium PM> 5 levels had inter mediate RRs that were consistent with the RRs for the high and low PVT t counties. Although thereis some variation, Wald chisquare tests conducted on these R Rs did not reject the hypothesis of homogeneity during any of the three follow-up periods for the individual counties or for the groups of counties These findings are consistent with those in Table 3. Relative Risks by 1979-1983 PM2 5 l evel Table 5 shows the relationship of 1973-2002 mortality to 1979-1983 PM; 5 level, by decade of follow-up, based on assigning each subject the PMs s level of the county in which they resided as of late 1972 Both the age- and sex-adjusted and fully adjusted 1973-2002 RRs are shown. Also show n are the 1973-2002 RRs for selected subgroups defined by sex, age (year of birth), education level, cigarette smoking status, and health status. These RRs were calculated based on a unit in crease in PM2.5 of 10 pg/m3. The age- and sex-adjusted RRs and the fully adjusted RRs were slightly elevated (-- 1.04 > dur ing the first decade, 1973 1982, but were essentially 1 0 during the next two decades Among the subgroups, the fully adjusted 1973-2002 RRs were slightly elevated (1.03) for females and younger subjects (those bom during 1908-1929) and consistent with 1.0 for the others. Table 6 shows the relationship of 1973-1982 mortality to 1979-1983 PM>.; levels for all subjects and for the same selected subgroups. ITte fully adjusted RRs were significantly elevated above 1.0 for all subjects (1.04), for females (1.05), for younger subjects (1.06), for the least educated () .07). for former smokers (1.06), and for the healthy (1.05). Table 7 shows the relation ship of 1983-2002 mortality to 1979-1983 PM, 5 levels for all subjects and for the subgroups. Results for the entire 20 years are show n because they are virtually the same as the separate results for 1983-1992 and 1993-2002. The fully adjusted RRs were not elevated above 1.0 for any subgroup during 1983-2002. Taken as a whole, these results suggest there was a weak rela tionship between fine particulate pollution and mortality during 1973--1982, but none during 1983 2002 However, because of statistical fluctuation, effects of up to a 2% increase in mortality per 10 .g/nv' of PM25 cannot be ruled out during 1983-2002. Sensitivity Analysis of Relative Risks Table 8 presents a sensitivity analysis to determine the ex tent to which the RRs in Tables 5-7 were influenced by the confounding variables that were used. It shows the RRs during 1973-2002. 1973-1982, and 1983-2002 based on sequential proportional hazards regression models, beginning with 1979-- 1983 PM25 as the only independent variable and then adding age, sex, and nine confounding variables, one at a time The results were not particularly sensitive to the addition of any of the variables except age and cigarette smoking status. It is unknown how much the RRs would be changed if PM; s data were available for years before and after 1979-19.83, None of the RRs during 1973-2002 and 1983 2002 were significantly elevated above 1.0 after adjustment for age, bill (he RRs have 810 J E. ENSTROM TABLE 5 Relative risk of death from all causes (HR and 95% C'l) during 1/1/197.V-12/31/2002 associated with change of 10 j.tg/m' in 1979-1983 PMfcj, with subgroups defined by decade of follow-up. sex, year of birth, education level, cigarette smoking status as of 10/1/1972, and health status as of 10/1/1959 Model based on 1979-1483 PM> 5 Subgroups Deaths/subjects Age-sex-adjusted RR (95% Cl) Fully adjusted RR (95% Cl) All subjects during each decade of full follow-up period (1/1/1973- 12/31/2002) 1/1/1973--12/31/1982 8795/35,783 1.032(1.003-1.062) 1/1/1983-12/31/1992 10,821/26.988 0.989(0.964-1.015) 1/1/1993-12/31/2002 8825/16.167 0.997(0.969-1.026) Subjects during full follow-up period (1/1/1973- 12/31/2002) All subjects 28.441/35,783 All males 13.532/15.573 All females 14,909/20,210 Bom 1908-1929 (1973 age 43-64) 13,354/20,086 Born 1873-1907 (1973 age 65-99) 15,082/15,697 < 12 yr education 8025/9079 12 vr education 6346/8557 > 12 yr education 14,070/18.147 Never smoker 11,528/15,181 Former smoker 10,074/12,400 Current smoker 6839/8202 Healthy 22,234/28,461 Unhealthy 5456/6439 1.005 (0.989-1.021) 0.996 (0.973-1.020) 1.013 (0.991-1.035) 1.022 (0.998-1.046) 0.991 (0.970-1.013) 1.016(0.987-1.047) 1.002 (0.969-1.037) 0.998 (0.976-1.021) 1.020(0.995-1.045) 1.005(0.978 1.032) 1.003(0.971-1.036) 1.006(0.988-1.024) 0 981 (0.945-1.018) 1.039(1.010-1.069) 0.996(0.970-1.022) 0.999(0.970-1.028) 1.010(0.994-1.026) 0.993 (0.970-1.016) 1.027 (1.005-1.050) 1.027 (1.003-1.052) 0.996 (0.975-1.018) 1.018 (0.989-1.049) 1.005 (0.972-1.040) 1.007 (0.984-1.030) 1.019 (0.994-1.044) 1.005 (0.978-1.032) 0.999(0.967-1.032) 1.010(0.992-1.028) 0.99) (0 957 1.030) TABLE 6 Relative risk of death from all causes (RR and 95% Cl) during 1/1/1973-12/31/1982 associated with change of 10 /zg/m! in 1979-1983 PMj 5, with subgroups defined by sex. year of birth, education level, cigarette smoking status as of 10/1/1972, and health status as of 10/1/1959 Model based on 1979-1983 PVL j Subgroups Deaths/subjects Age sex adjusted RR (95% Cl) Fully adjusted RR (95% Cl) Subjects during initial decade of follow-up (1/1/1973- 12/31/1982) All subjects 8795/35.783 All males 4701/15,573 All females 4094/20,210 Born 1908-1929 (1973 age 43-64) 2637/20,086 Bom 1873-1907 (1973 age 65-99) 6158/15,697 < 12 yr education 3123/9079 12 yr education 1686/8557 > 12 yr education 3986/18,147 Never smoker 3425/15.181 Former smoker 3264/12,400 Current smoker 2106/8202 Healthy 6432/28.461 Unhealthy 2104/6439 1.032(1.003 1.062) 1.027(0.988 1.067) 1.039(0.996 1.083) 1.062(1.006-1.120) 1.021 (0.988-1.056) 1 064(1 015-1.115) 1.045(0.980-1.115) 1.001 (0.960-1.045) 1.038(0.993-1.086) 1.059(1.011-1.110) 1.014(0.957-1.075) 1.043 (1.009-1.078) 0.981 (0.925-1.040) 1.039(1.010-1.069) 1.029(0.990 1.069) 1.052 (1,009 1.096) 1.064(1.008-1.122) 1.031 (0.997-1.066) 1.072(1 023-1.124) 1.046 (0.981-1.116) 1.011 (0.969-1.055) 1.038 (0.992-1.085) 1.058 (1 010-1.109) 1.009 (0 952-1.069) 1.050 (1.016-1.085) 0.991 (0.935-1 051) LLNE PARTICULATE POLLUTION AND MORTALITY 811 TABLE 7 Relative risk of death front all causes (HR and 95% Cl) during 1/1/1983-12/31/2002 associated with change of 10 /ig/nv* in 1979-1983 PM.; s. with subgroups defined by sex, year of birth, education level, cigarette smoking status as of 10/1/1972, and health status as of 10/1 /l 959 Model based on 1979 1983 PM2 5 -Subgroups Deaths/subjects Age-sex-adjusted RR (95% Cl) Fully adjusted RR (95% Cl) Subjects during last two decades of follow-up (1/1/1983-12/31/2002) All subjects 19,646/26.988 All males 8831/10.872 All females 10,815/16,116 Bom 1908-1929 (1983 age 53-74) 10,717/17.449 Bom 1873-1907 (1983 age 75-99) 8929/9539 < 12 yr education 4902/5956 12 yr education 4660/6871 > 12 yr education 10,084/14.161 Never smoker 8103/11.756 Former smoker 6810/9136 Current smoker 4733/6096 Healthy 15802/22.029 Unhealthy 3352/4335 0.992 (0.973-1.011) 0.979 (0.951-1.008) 1.003 (0.978-1.029) 1.012(0.986-1.040) 0.972 <0 945-0 999) 0.987 (0.950-1.025) 0.985 (0.947-1.026) 0.997 (0.970-1.024) 1.011 (0.982-1.042) 0.979 (0.947-1.012) 0.999(0.961-1.039) 0.990 <0.969-1.012) 0.980(0.935-1.028) 0.997 (0.978-1.016) 0.974 (0.947-1.003) 1.018(0.992-1.044) 1.018(0.992-1.046) 0.975(0 948-1 002) 0.986 (0.949-1.024) 0.990(0.951-1.030) 1.005(0.978-1.032) 1.011 (0.981-1.041) 0.980(0.949-1.013) 0.996(0.958-1.036) 0.994 (0.973-1.015) 0.995 (0 949-1.043) upper confidence intervals as high as 1.028. All of the RRs dur ing 1973- 1982 were significantly elevated about 1.0. with upper confidence intervals as high as 1.071. Based on their large / ' values, the variables of age. sex, and cigarette smoking status were by far the most important variables in the model After in clusion of age, sex. and cigarette smoking status, the addition of the next 7 independent variables changed the RRs hy only about 0.1%. When initial health status was entered as an additional TABLE 8 Relative risk of death from all causes (RR and 95% Cl) during 1/1/1973-12/31/2002, 1/1/1973-12/31/1982, and 1/1/1983-12/31/2002 associated with change of 10 /tg/itr' in 1979--198 5 PMi 5, hy individual confounding variables defined as of 1959, except for 1972 cigarette smoking. Age. sex, and nine confounding variables are added to the ptopoitional hazards regression model one variable at a time Cumulative PHREG model based on adding one variable at a time 1973-2002 Chi-square 1973-2002 RR (95% Cl) 1973-1982 RR (95% Cl) 1983-2002 RR (95% Cl) 1979-1983 PM2.5 -LAge -fSex -t-Cigaiette smoking 4-Race 4- Education 4-Martial status 4-Body mass index 4-Occupational exposure 4-Exercise 4-Fruit/fruit juice intake 4-Health status 0.68 14.445.92 464.16 1.610.59 0.93 53.67 26.16 91.87 3.28 0.02 42.54 156.15 1.029(1.012-1.045) 1.003(0.987 1.019) 1.005 (0 989-1.021)'' 1.011 (0.995-1.028) 1.011 (0.995 1.027) 1.011 (0.995-1.027) 1.011 (0.995-1.027) 1.010(0.994-1.026) 1.010(0.994-1.026) 1.010(0.994-1.026) 1.010 (0.994-1.026)* 1.007(0.991 1.023) 1.058 < 1.028-1 089) 1.029(1.000-1.059) 1.032 <1.003-1.062)'' 1.041 (1.012-1.071) 1.041 (1.012-1.071) 1.041 (1.012-1.071) 1.041 (1.012-1.071) 1.040(1.011-1.070) 1.040(1.011-1.070) 1.039(1.010-1.069) 1.039(1.010-1.069)* 1.036 (1.006-1.066) 1.016(0.996-1.035) 0.991 (0.972-1.010) 0.992 (0.973-1.011)' 0.998 (0.979-1.017) 0.997 (0.978-1.017) 0.997 (0.978-1.016) 0.997 (0.978-1.016) 0.996 (0.977-1.016) 0.996 (0.977-1 016) 0.997 (0.978-1.016) 0.997 (0.978-1.016)' 0.994 (0.975-1.013) Age-sex-adjusted RR Eight-variable fully adjusted RR 812 J. E ENSTROM TABl.F. 9 Fully-adjusted relative risk of death from all causes (RR and 95% Cl) by cigarette smoking status as of 10/1/1972, during 1973-2002, 1973-1982. and 1983-2002 for both sexes, for the California CPS subjects in II counties with 1979-1983 PM? 5 measurements, which were used as one of the confounding variables Cigarette smoking status as of 10/1/1972 Fully adjusted Fully adjusted Fully adjusted 1973-2002 RR (95% Cl I 1973-1982 RR (95% Cl) 1983-2002 RR (95% Cl) Deaths/subjects Never (as of 1959 and 1972) Former (as of 1959 and 1972) Former (as of 1972 only) Current: 1-9 cpd (as of 1972) Current: 10-19 cpd (as of 1972) Current: 20 cpd (as of 1972) Current: 21-39 cpd (as of 1972) Current: 40-1- cpd (as of 1972) Chi-square test of homogeneity (7 degrees of freedom) Note, cpd, cigarettes per day. 28,447/35,789 1.000 1.054(1.014-1.096) 1.253(1.212-1.295) 1.239(1.150-1.336) 1.597(1.510-1.688) 1.871 (1.791-1.953) 2.068 (1.936-2.210) 2.543 (2.375-2.723) X2 = 1701.75 p < .0001 8801/35.789 1.000 1.061 (0.987-1.140) 1.312(1.236-1.392) 1.227 (1.065-1.414) 1.667 (1.508-1.842) 1.829(1.689-1.980) 1.889(1.666-2.140) 2.460 (2.189-2.765) X2 = 478.14 p < .000! 19.646/26.988 1 .<.8)0 1 049 <1.001-1.100) 1.224(1.176-1.273) 1.243(1.138-1.357) 1.566(1.465-1.675) 1.887(1.792-1.987) 2.145 (1.984-2.320) 2.587(2.378-2.814) X2 = 1232.17 p < .0001 independent variable, the RRs decreased by 0.3%'. However, ini tial health status was not one of the eight confounding variables used in calculating the fully adjusted RRs in the other tables, be cause it may have been influenced by exposure to air pollution before entry into the study. Relative Risks by 1972 Cigarette Smoking Level Tabic 9 shows the fully adiusted RRs for eight levels of 1972 cigarette smoking, the strongest confounding variable in this study Note there was a strong and clear dose-response rela tionship during 1973-2002. The dose-response relationship re mained as strong during 1983-2002 as it was during 1973 -1982, in spite of the large degree of smoking cessation that occurred from 1972 to 1999. as documented in Tables 1 and 2. This com parison supports the findings of an earlier paper, which exam ined smoking cessation and mortality trends in this cohort during 1960-1997 (Enstrom & Heath, 1999) The Wald chi-squarc test of homogeneity for each of the three follow-up periods, where X - > 478 for 7 degrees of freedom, clearly rejects the hypothesis that the RRs are equal (homogeneous). The large RRs related to increasing cigarette smoking level are shown here in order to put the RRs related to increasing PM2.5 level in Table 4 in perspective. DISCUSSION Strengths and Uncertainties of This Study This study has several important strengths: a large, diverse cohort of males and females distributed throughout California, a large number of deaths, extensive baseline and follow-up data on demographic and lifestyle characteristics, long-term follow up of a high percentage of subjects, relative stability of sub jects based on their residential address history', and availability of PM2 5 measurements for over 70% of the subjects. In addi tion, there is a wide range of PMt 5 levels (10.6 to 42.0 /ag/nv) available for subjects in 11 counties. Although the CA CPS I cohort is not a random sample of the California population, previous examination has shown that mortality ratios based on cigarette smoking status are similar in this cohort (Enstrom & Heath, 1999) and in a cohort representative of the US population (Enstrom, 1999). The results of this study, as in all epidemiology studies, arc dependent upon the underlying data and the analytical methods that were used. Major uncertainties include the extent to which the available air quality data represent actual exposures, rhe va lidity of the proportional hazards regression calculations of ihe RRs. and the potentially important confounders that may have been omitted from the analysis. The PiVU 5 air quality data are limited to 11 California coun ties and 8 of these counties had only one outdoor monitoring station each. The period of monitoring was from July 1979 to December 1983 and most counties had data forjust a two to three year period. These sample data are assumed 10 represent long term personal exposures of each subject based on their county of residence in late 1972. The validity of this assumption has not heen confirmed, but these same limited data have been used in other major cohort studies (Pope et al,, 1995, 2002; Lipfert et al,, 2000. 2003). The assumption that individual exposures are the same as county-wide averages, as measured by a few' centrally-located monitors, can result in the "ecological fallacy," where results based on group averages differ from those based on individual exposures (Piantadosi et al,, 1988). However, it is impractical to monitor individual exposures for a large cohort, especially I'ABLE 10 Relative risk (RR) and 95% confidence interval (Cl) tor long-term all-cause mortality per IO-/tg/m* increase in PM25 for U.S. cohort studies based on PM; 5 data, circa 1980 PM; * Study characteristics Study (author, year) Data period Mean (range) (jxgAn ') Cohort geographic definition Follow-up period Mean entry age for period Number entered in cohort Deaths in follow-up period RR (95% Cl) Males Dockery et al., 1993 Pope et al.. 1995 McDonnell et al.. 2000 Lipfcrt et al., 2000 Pope et al., 2002 Enstrom, 2005 Females Dockery et al., 1993 Pope ct al., 1995 McDonnell et al.. 2000 Pope et al., 2002 Enstrom, 2005 Both Sexes Dockery et al., 1993 Popect al., 1995 Pope et al.. 2002 Enstrom, 2005 1979-1985 1979-1981 1973-1977 1979-1981 1982-1984 1982-1984 1979-1983 1979-1983 1979-1983 1979-1985 1979-1981 1973-1977 1979-1983 1979-1983 1979-1983 1979-1985 1979-1981 1979 1983 1979-1983 1979-1983 19 (11-30) 18(9-34) 32 (17-45) 24 (6-12) 22 (8-11) 22 (8-11) 21(10-30) 24(11-42) 24 (11-42) 19(11-30) 18(9-34) 32(17-45) 21 (10-30) 24(11-42) 24(11-12) 19(11-30) 18(9-34) 21(10-30) 24(11-42) 24(11-42) "Obtained from supplementary data (Krewski et al.. 2000). "Recalculated from published data (U.S EPA, 2004). ' Obtained from the author. 6 U.S. cities 50 U.S. SMSAs 9 CA airsheds 42 U S counties 61 U.S. SMSAs 11 CA counties 6 U S cities 50 U.S. SMSAs 9 CA airsheds 61 U.S. SMSAs 11 CA counties 6 U.S cities 50 U.S. SMSAs 61 U.S. SMSAs 11 CA counties 1975-1989 1982-1989 1976-1992 1975-1981 1982-1988 1989-1996 1982-1998 1973-1982 1983-2002 1975-1989 1982-1989 1976-1992 1982-1998 1973-1982 1983-2002 1975-1989 1982-1989 1982-1998 1973-1982 1983-2002 ~50 57 58 51 57 63 57 66 74 -50 57 58 57 65 73 --50 57 57 65 73 3671" 130,310" <1347 26.067 ~21.467 -15,367 -159,000" 15,573 10,872 830" -12.400" <375 --4600" --61OO --5765" -36,000" 4701 8831 4440" 164,913" <2422 --200,000" 20,210 16,116 599" -8365" <568 -24,000" 4094 10,815 8111 295,223 --359.000 35.783 26,988 1430 20,765 --60.000 8795 19,646 1.15 (1.02-1.30 )h 1.07 (1.03-1.11 /' 1.09 (0.98-1.21 )b 0.95 (0 89-1.01 )r 0.94 (0 90-0.98)" 0.89 (0 85-0.95 f 1.05(1.01-1.10) 1.03 (0.99-1.07) 0.97 (0.95-1.00) 1.12 (0.96-1.30/' 1.06 (1.01-1.12)* --1.00 (assumed) 1.02 (0.98-1.06) 1.05 (1.01-1.10) 1.02 (0.99-1.04) 1.13 (1.04-1.23 )h 1.07(1.04-1.10/' 1.04 (1.01-1.08) 1.04(1.01-1.07) 1.00(0 98-1 02) 814 J. E, ENSTROM over the long-term, This analysis used the smallest practical geographic unit (counties)* given typical mobility of the subjects, in hopes of minimizing exposure errors. The relative risk results are all based on Cox proportional hazards regression (PI1REG), which has been used in numerous cohort survival analyses. This statistical methodology depends on the assumption of proportionality (SAS, 2004). The validity of this assumption with respect to the PM; 5 variable has been confirmed for most RRs in Tables 3-9 by the Kolmogorov-type supremuni test of functional form (SAS. 2004). Also, previ ous findings on cigarette smoking and mortality in this cohort showed that relative risks based on proportional hazards regres sion were similar to relative risks based on life table survival analysis (Enstrom Si Heath, 1999). With respect to the impact of additional potential confound ing variables, ecological variables at the county level, such as climate* were explored and found to be uncritical. The analyses based on ihe eight-variable model found that after age and sex adjustment only one confounding variable, cigarette smoking status, had much impact on the RRs. Indeed, each fully-adjusted R R was within 1.5% of the corresponding age-sex-adjusted RR. Comparisons with Other Cohort Studies In Table 10 the major findings in this study are compared with those of the other U.S. cohort studies based on PM; 5 data, circa 1980. Hie basic characteristics and relative risks for all-cause mortality are given for all six studies, where each relative risk (RR and 95% Cl) is based on a 10 /ig/ni ' increase in PM2.5 The RRs were standardized to the extent possible by using the same conversion of publ ished results that was used by the FPA in their latest criteria document on particulate matter (US EPA. 2004). For instance, the RR of 1.07 (1.04-1.10) during 1982-1989 for both sexes in the ACS CPS 11 study (Pope et ah, 1995) is based on a mortality difference of 17% between the highest and lowest PM* 5 areas and a PM; 5 difference of 25 /rg/m' By comparison, Ihe RR of 1 (X) (0-98 1.02) during 1983-2002 among both sexes in the current CA CPS 1 study is based on a PM; 5 variation up to 31 /ig/ni among 11 counties. The RRs in Table 10 range from 0.89 to 1.15 and each cor responding 95% Cl either includes 1.0 or is within 0.05 of in cluding 1.0. Thus, the relationship between PM; 5 and mortality is very weak and near the limit of detectability by epidemio logic methods. In order to define this relationship as accurately as possible, it is important to understand the differences that exist between the RRs. These differences could be due to the epidemiologic methodology used or they could be due to the characteristics of the study cohorts, such as their geographic lo cation, follow-up period, demographics, and size. For instance, the RR of 1.07 in the ACS C'PS II study is the average relation ship between PM; 5 level and mortality in 50 areas of (he U.S. However, detailed reanalysis of this study reveals substantial ge ographic variation in the relationship (Krewski et al,, 2000). In particular, a map of PM; 5 levels and relative risk of mortality throughout the U.S. (Figure 21) shows that most areas of Califor nia had medium mortality risk and no areas had high mortality risk T his pattern suggests that the relationship between PM;.? and mortality among the California subjects in the CPS II cohort was weaker than the RR of 1.07 and consistent with the RR of 1 .(X) found in the CA CPS 1 cohort. It is clear from Table 10 that no single result can adequately describe the relationship between PM; * and mortality for the entire country. The complete body of epidemiologic evidence should be used to assess this relationship as accurately as pos sible within the limitations of epidemiologic methodology. A full comparative examination of the available cohort studies is warranted. Ideally, a standardized method of analy sis should be applied to the underlying data in each cohort and the results should be presented in a standardized way. Such an analysts would make a substantial contribution to the research priorities for particulate matter (National Research Council, 2004). REFERENCES California Air Resources Board. 2003, Ambient air qualin standards for suspended particulate matter t PMt and sul fates. California Environmental Protection Agency. Sacramento. CA. July 5. (ftp://ftp.arb.ca.gov/carbis/regact/aaqspm/ixor.pdf and www.arb.ca.gov/regaet/aaqspm/aaqspm-htnri. Dockery, D W., Pope, C. A , 111, Xu, X Spongier. J. D , Ware, j if. Fay. M. E.. Ferris. B. G.. and Speizer, F E 1993 An association between air pollution and mortality in six U.S. cities. N. Engl. J. Med 329:1753-1759. Enstrom, J. H. 1999. Smoking cessation and mortality trends among two United States populations. J Clin F-pidemwt 52:813 Enstrom, J. E , and Heath, C W,, Jr. 1999 Smoking cessation and mor tality (remix among 118,(XX) Californians. 1960-97. Epidemiology 10:500-512. Enstrom, J E.. and Kabai, G C. 2003 Environmental tobacco smoke and tobacco related mortality in a prospective study oi Californians. 1960-98. Hi Med J. 326: 1057-1061 Gamble, J F 1998. PMj s and mortality in long-term prospective co hort studies. Cause-effect or statistical associations, Environ, Health Prospect. 106X35-549 Grecnbaum, D. S., Bachmann, J. D., Krewski. D., Saniet, J. M,, White. R , and Wyzga. R. E. 2001. Particulate air pollution standards and morbidity and mortality: Case study. Am. J. Epidemiol. 154:S78 S90. Hammond, t. C. 1972. Smoking habits and air pollution in rela tion to lung cancer In Environmental Factors r;i Respirators Dis ease (Edited by Dll Lee). Fogerty International Centet Proceedings No. 11 New1 York: Academic Press, 1972, Chapter 12. pages ] 77-- 198. Hammond. E. C,, and Garfinkel. L. 1980. General air pollution arid cancer in the United States. Prev. Med 9:206-211 Hinton, 13 O., Sunc, J. M., Suggs. J. C., and Barnard, W. F. 1984. Inhalable Particulate Network Report: Operation and Data Summary {Mass Concentrations Only). Volume I. April 1979December 1982. EPA-600/4-84-088a. Research Tiiangle Park, NC U.S. Environmental Protection Agency, November 1984, particularly pages 108-113. Hinton. D. O., Sune. J. M,, Suggs. J. C . and Barnard. W. F 1986 Inhalable Paniculate Network Report: Data Summitry (Mass FINE PARTICULATE POL LUTION AND MORTALITY 815 Concert tuitions only) Volume TIL January 1983-December 1984, KPA-600/4-86/019 Research Triangle Park. NC U S. Environmen la I Prolection Agency. April 1986. particularly pages 53-55. Kaiser. J. 2005. Mounting evidence indicts fine-particle pollution. Sci ence 307:1858 1961 (http://www.sciencemag.org/cgi/reprint/307/ 5717/1858a.pdf). Krewski, D.. Burnett. R T.. Goldberg. M. S.. Hoover. K.. Siemiatycki. J.. Abrahamowicz. M., White, W. H and others 2000 Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality; Special report Cambridge. MA: Health Effects Institute. Part I Replication and Val idation (http://wwvt.healtheffecLs.org/Pubs/Rean-partl .pdf) and Part II. Sensitivity Analyses (http://wwwhcaltheffccts.org/Pubs/Reanpart2.pdf), particularly Figure 21 on p. 197 Lipfert, F. W 2003 Commentary on the HEI Reanalysis of the Har vard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality J Toxicol. Environ. Health A 66:1705-1714. Lipfert, F. W. 1994. Air pollution and community health. A critical review and data sourcebook New York: van Nostrand Reinhold. Lipfert, F. W,, Malone. R (., Daum.M. L , Mendell.N. R., and Yang. C. C. 1988. A statistical study ofthe macroepidemiology ofair pollution and total mortality Brookhaven National Laboratory, Upton. NY. Report BNL 52122, April Liplert, F. W.. Perry. H. M., Jr . Miller. J P. Baty. J. LL. Wyzga, R. E . and Carmody. S. E. 2000. The Washington University EPRI vet erans' cohort mortality study: Preliminary results. Inha! Toxicol. 12[S4J:41-73. Lipfert. F W . Perry, H. M.. Jr.. Miller. J. P.. Baty. J. D.. Wyzga. R. E .and Carmody. S. E. 2003 Air pollution, blood pressure, and thetr long term associations with mortality. Inhal. Toxicol 15(5) 493-512. Logan, W. P D. and Glasg M. D. 1953. Mortality in London fog inci dent. 1952 Lancet 1:336-338. McDonnell, W F.. Nishino-lshikawa. N . Petersen. F. F.. Chen, L. H , and Abbey, D E. 2000. Relationship of mortality with the fine and coarse fractions of long term ambient PM]a concentrations in non smokers. /. Expos Anal. Environ Epidemiol. 10:427-436 Moolgavkar. S. H. 1996 A critical review of the evidence on particulate air pollution and mortality. Epidemiology 7:420-428. National Research Council. Committee on Research Priorities for Airborne Particulate Matter 2004. Research priorities for air borne particulate matter: TV. Continuing research progress. Washington, DC: National Academics Press. (http7/www nap edu/ books/0309091993/html). Office of Management and Budget. 2003. Proposed bulletin on peer review and information tfutility. September 15. 68FR 54023-9. (http://www whttehouse gov /omb/mforcg/agency_info_quality Jinks html). Ozkaynak, H . and Thurston, G. D. 1987. Associations between 1980 U S. mortality rates and alternative measures of airborne particle concentration. RiskAnaly. 7:449-461 Phalen, R. F 2002. The paniculate air pollution controversy: A case sludv and lessons learned Boston Kluger Academic Piantadosi, S , Byar, D. P., and Green. S. B. 1988. The ecological fallacy. Am. J. Epidemiol. 127 893-904. Pope. C. A. III. and Dockery, D. W 1999. Epidemiology of particle effects. In Air pollution and health, eds. S. T Holgate. H. Korcn, R Maynard, and J. Saniet, pp. 673-705. London: Academic Press. Pope. C A. 111. Thun, M J., Namboodiri. M. M., Dockery, D. W.. Evans. J. S.. Speizer, F. E., and Heath. C W , Jr. 1995, Particulate air pollution as a predictor of mortality in a prospective study of U.S. adults. Am J. Respir. Crit. Care Med. 151:669-674. Pope. C. A III. Burnett. R T.. Thun. M J.. Callc, E. E . Krewski. D., Ito, K , and Thurston. G. D. 2002. Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution J Am. Med. Assoc. 287:1132-1141. SAS Institute, Inc. 2004 The PHREG Procedure SAS/STAT 9.1 user's guide pp. 3213-3332 (http://supportsas.com/documentation/ onlinedoc/91 pdf/sasdoc_91/stat_ug_7313 pdf). Cary. NC SAS Insti tute. Steinbrook. R 2004 Peer review and federal regulations N. Engl .1 Med 350:103-104 Sune, J. 1999. Personal communication with FW Ltpifert. Inhalable Particle Monitoring Network data (iptnnshrt.exe) Research Triangle Park. NC: U S. Environmental Protection Agency. U S. Environmental Protection Agency 1997. Air quality criteria for particulate matter. Fed Reg. 62:38676. U.S. Environmental Protection Agency 2004 Air Quality Crite ria for Particulate Matter. Volume 1 (EPA/600/P-99/002aF) and Volume II (EPA/600/P-99/002bF). Washington. DC. October 2004. particularly Table 8-12 on page 8-117 (http.//cfpub.cpa.gov/ncca/ cfm/partmatt.cfm). APPENDIX TABLE 1 1979-1983 PM2.5 ("F1NE15") Jala in California from the F.PA Inlialable Particulate Network by county anJ monitoring site (Hinton cl al.. 1984. pages 108-113; Hinton et al.. 1986, pages 53-55; Sunc, 1999). County City (Street) SAROAD site Monitoring period Samples Site PM2.5 Oig(m') Mean SD County PM2.5 (/rg/ml) Mean Santa Barbara Contra Costa Alameda Butte San Francisco Santa Clara Fresno San Diego Los Angeles Kern Riverside Lompoc Richmond Livermore (Railroad Ave) Livermore (Old First St) Chico San Francisco East San Jose Fresno (East Olive) Five Points El Cajon Azusa (Loren Ave) West Los Angeles Pasadena Bakersfield (Chester Ave) Rubidoux (Mission Blvd) 05408(X)02A07 0563110003A07 054020002A07 054020003A07 051260002A07 056860003A07 056980004A07 052800005A07 052820002A07 05222(XX)3A07 050500002A07 054180103A07 055760004A07 050520004A07 056535001A07 03-29-81 to 12-31-82 01-06-83 to 03-01-83 09-24-79 to 10-14-82 09-24-79 to 12-17-80 03-29-81 to 08-21-82 03-24-82 to 08-15-82 01-24-83 to 12-26-83 11-22-79 to 09-14-82 09-30-79 to 10-14-82 09-14-82 to 09-20-82 09-30-79 to 10-20-82 09-25-81 to 12-31-82 01-06-83 to 08-04-83 10-18-79 to 11-07-82 07-08-79 to 10-14-82 10-18-79 to 12-12-81 11-17-80 to 12-01-82 01-18-83 to 01-30-83 08-19-79 to 12-31-82 01-06-83 to 06-05-83 91 10.631 4.649 8 10.586 4.062 167 13.920 11.333 58 17.398 15.547 82 12.260 8.654 25 9.992 4.073 50 18.183 10.793 115 16.352 12.676 174 17.788 16.487 2 10.315 0.728 153 18.478 17.535 66 19.873 13.929 29 16.747 11.992 92 28.765 19.200 147 26.752 18 667 28 34.174 16.857 53 30.803 29.190 3 31.920 30.178 127 43.337 28.669 15 30.791 33.864 10627 13.920 14.389 15.453 16.352 17.788 18.373 18.919 28.224 30.863 42.012 EXHIBIT D Original Article Fine Particulate Matter and Total Mortality in Cancer Prevention Study Cohort Reanalysis James E. Enstrom1 Dose-Response: An international Journal January-March 2017:1-12 C- The Author(s) 2017 Reprints and permission: sagepub.com/journalsPermi5sions.nav DOi: 10.1 177/1559325817693345 journals.sagepub.com/home/dos dSAGE Abstract Background: In 1997 the US Environmental Protection Agency (EPA) established the National Ambient Air Quality Standard (NAAQS) for fine particulate matter (PM2 5), largely because of its positive relationship to total mortality in the 1982 American Cancer Society Cancer Prevention Study (CPS II) cohort. Subsequently, EPA has used this relationship as the primary justification for many costly regulations, most recently the Clean Power Plan. An independent analysis of the CPS II data was conducted in order to test the validity of this relationship. Methods: The original CPS II questionnaire data, including 1982 to 1988 mortality follow-up, were analyzed using Cox pro portional hazards regression. Results were obtained for 292 277 participants in 85 counties with 1979-1983 EPA Inhalable Particulate Network PM25 measurements, as well as for 212 370 participants in the 50 counties used in the original 1995 analysis. Results: The 1982 to 1988 relative risk (RR) of death from all causes and 95% confidence interval adjusted for age, sex, race, education, and smoking status was 1.023 (0.997-1.049) for a 10 pg/m3 increase in PM2,5 in 85 counties and 1.025 (0.990-1.061) in the 50 original counties. The fully adjusted RR was null in the western and eastern portions of the United States, including in areas with somewhat higher PM25 levels, particularly 5 Ohio Valley states and California. Conclusion: No significant relationship between PM2 5 and total mortality in the CPS II cohort was found when the best available PM2S data were used. The original 1995 analysis found a positive relationship by selective use of CPS II and PM2 5 data. This independent analysis of underlying data raises serious doubts about the CPS II epidemiologic evidence supporting the PM2S NAAQS. These findings provide strong justification for further independent analysis of the CPS II data. Keywords epidemiology, PM2 s, deaths, CPS II, reanalysis Introduction In 1997 the US Environmental Protection Agency (EPA) estab lished the National Ambient Air Quality Standard (NAAQS) for fine particulate matter (PM2 5), largely because of its pos itive relationship to total mortality in the 1982 American Can cer Society (ACS) Cancer Prevention Study (CPS II) cohort, as published in 1995 by Pope et al.1 The EPA uses this positive relationship to claim that PM2 5 causes premature deaths. How ever, the validity of this finding was immediately challenged with detailed and well-reasoned criticism.2"4 The relationship still remains contested and much of the original criticism has never been properly addressed, particularly the need for truly independent analysis of the CPS II data. The EPA claim that PM2j causes premature deaths is implausible because no etiologic mechanism has ever been established and because it involves the lifetime inhalation of only about 5 g of particles that are less than 2.5 pm in dia meter.'2 The PM2.5 mortality relationship has been further chal lenged because the small increased risk could be due to wellknown epidemiological biases, such as, the ecological fallacy, inaccurate exposure measurements, and confounding variables like copollutants. In addition, there is extensive evidence of spatial and temporal variation in PM2.5 mortality risk (MR) that does not support 1 national standard for PM2 5. 1 University of California, Los Angeles and Scientific Integrity Institute, Los Angeles, CA, USA Corresponding Author: James E. Enstrom, University of California, Los Angeles and Scientific Integrity Institute, 907 Westwood Boulevard #200, Los Angeles, CA 90024, USA. Email: jenstrom@uda.edu Creative Commons CC BY-NC: This article is distributed under the terms of die Creative Commons Attribution-NonCommercial 3.0 License cc {http://www.creativecommons.org/licenses/by-nc/3-0/) which permits non-commercial use, reproduction and distribution of the work without further permission provided the original work is attributed as specified on the SAGE and Open Access pages {https://us.sagepub.com/en-us/nam/open-access-at-sage). Enstrom 3 Gamble IF. PM; 5 and mortality in long-term prospective cohort studies: cause-effect or statistical associations? Environ Health Perspeci. 1998: 106(9):535-549 doi: 10.1289/chp 98106535 4. Phalen RF. The particulate air pollution controversy. Nonlinearity Biot Toxicol Med. 2004;2<4):259-292. doi 10 1080/1540142049090024< Accessed February 20, 2017. 5 Enstrom JE. Young SS, Dunn JD, et al. Particulate Matter Does Not Cause Premature Deaths. August 17, 2015. https://www nas. org/imaces/documents'PM; <; pdf Within Wood P. Concerns about National Academy of Sciences and Scientific Dissent National Association of Scholars. December 15, 2015 https:// www.nas.org/articles/nas_letter Accessed February 20, 2017. 6. Enstrom JE. EPA's Clean Power Plan and PMj j-related Co-Benefits Tenth Internationa! Conference on Climate Change. Panel 8 Heartland Institute. Washington, DC. 2015 http:// climateconferences.heartland.org james-enstrom-icccl0-panel-8/, and http://www.scientificintegrityinstitute.org/JEElCCC061115. pdf Accessed February 20. 2017 7 Krewski D, Burnett RT, Goldberg MS, et al Reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particulate Air Pollution and Mortality: Special Report. Cambridge. MA: Health Effects Institute; 2000. Part I Replica tion and Validation and Part II Sensitivity Analyses, particularly Figure 5 on page 161, Figure 13 on page 89. and Figure 21 on page 197 and Appendix D and Appendix F https:/www.healtheffects org/publication/rcanalysis-harvard-six-cities-study-and-amcricancanccr-socicty-study-particulate-air Accessed February 20. 2017. 8. Enstrom JE Fine particulate air pollution and total mortality among elderly Californians. 1973-2002 Inhal Toxicol. 2005: ]7(14>:803-8i6 PMID:16282I58. http: '/scientificintegritymstitute .org/lT 121505 pdf 9. Enstrom JE Response to "A Critique of *Fme Particulate Air Pol lution and Total Mortality Among Elderly Californians, 1973 2002'" hy Bert Rnmekreef. PhD. and Gerald Hock. PhD Inhal Toxicol 2006; 18(21:509-514 http /scientificintegritymstitute. org/lT060106 pdf 10 Enstrom JE Particulate Matter is Not Killing Californians. Pro ceedings of the American Statistical Association 2012 Joint Sta tistical Meeting, Section on Risk Analysis, San Diego, CA 2012: pages 2324-2336 htips www.am.slal org/nieetings/jsm/2012' proceedings cfm, and http://www.scientificintegritymstitute.org/ ASAS092812 pdf 11. July 22, 2013 US House Science Committee Final Request to EPA for ACS CPS II Data https://science.house.gov/news/ press-releases/committee-threatens-subpoena-epa-secret-science, and https://sciencc.house.gov sites/republicans science.house gov/files/documents/07-22-2013l7f20Smith%20and'3120Stewart %20to%20.McCarthy.pdf. Accessed February 20. 2017. 12. August I, 2013 US House Science Committee Subpoena to EPA Requesting ACS CPS 11 Data, https://science.house.gov/news/ press-releases/smith-subpoenas-epa-s-secret-science, and https:// science.housc.gov/sites/republicans.science.house.gov/files, documents/Subpoena%201mk.pdf Accessed February 20, 2017 13. Brawley OW August 19, 2013 ACS Brawley Letter to EPA Refusing to Cooperate with August 1, 2013 US House Sci ence Committee Subpoena of ACS CPS II Data. http://www. scientificmtegrityinstitute.org'Braw lcy081913.pdf. Accessed February 20, 2017 14. Gapstur SP September 20, 2013 ACS Letter to Enstrom Deny ing CPS II Collaboration as Proposed in September 16, 2013 Enstrom Fmai I http://www.scientificintegntyinstitute.org/ GapsturEns092013.pdf. Accessed February 20, 2017 15. Greenbaum D. October 4, 2013 HE! Response to September 26. 2013 Enstrom Email Declining to Conduct Special Analyses of ACS CPS II re 2000 HEI Reanalysis Report http seientificinlegrityinstitute.org/Greenbauml00413.pdf. Accessed February 20,2017. 16. Hinton DO, Sune JM, Suggs JC, Barnard WF lnhalable Particu late Network Report: Operation and Data Summary (Mass Concentrations Only). Volume 1. April 1979-December 1982. EPA-600.4-84-088a. Research Triangle Park, NC: U S. Environ mental Protection Agency, November 1984, particularly pages 102-160 of 210 total pages. http;//nepis.epa.gov'Exe/ZyPDF.cgi/ 200150U3.PDF/Dockey=2001 SOU3.PDF. Accessed February 20. 2017 17 Hinton DO, Sune JM, Suggs JC. Barnard WF lnhalable Particu late Network Report: Data Summary (Mass Concentrations Only). Volume III. January 1983-December 1984 EPA-600/486/019 Research Triangle Park. NC: U.S. Environmental Protec tion Agency, April 1986: particularly pages 51-80 of 227 total pages http://nepis.epa.gov/Exe/ZyPDF.cgi/9101R4E8.PDF? Dockey=9!01 R4L8.PDF 18 SAS, PI IRF.G and REGRESSION Procedures.SAS/STAT 9.4 User's Guide Cary. NC: SAS Institute Inc http' support.sas.com/ documentation'94 index.himl. Accessed February 20, 2017. 19 Centers for Disease Control. National Center for Health Statistics. 1980 CDC WONDER On-line Database, compiled from Com pressed Mortality File CMF 1968-1988. http://wonder.cdc.gov/ cmf-icd9 html Assessed April 15, 2016 20 Krew ski D. Jenett M. Burnett RT. et al. Extended Follow-Up and Spatial Analysis of the American Cancer Society Study I inking Particulate Air Pollution and Mortality HEI Research Report 140, Health Effects Institute. Boston, MA- 2009, particularly fable 34 https www healtheffects org pubiication/extended-follow andspatial-analysis-american-cancer-society-study-linking-particulatc Accessed February' 20, 2017. 21 Krewski D August 31,2010 Letter to HEI re Special .Analysis of California Subjects Within ACS CPS II Cohort Based on 2009 HEI Research Report 140 Methodology, http://www.arb.ca.gov/ research'health/pm-mort/HEI_Correspondence.pdf Accessed February 20, 2017 22. Lipfert FW, Malone RG, Daum ML, Mendell NR, Yang CC A Statistical Study of the Macroepidemiology of Air Pollution and Total Mortality. Brookhaven National Laboratory Upton, NY Report No. BNL 52122, April 1988, 136 pages, http:, www osti gov/scitech'servlets/purl/7028097. Accessed February 20. 2017 23. Lipfert F. Commentary on the HEI reanalysis of the Harvard Six Cities Study and the American Cancer Society Study of Particu late Air Pollution and Mortality J Toxicol Environ Health A. 2003.66(16-19): 1705-1714, discussion 1715-1722. doi 10.1080/ 15287390306443. 24. Enstrom JE. July 11, 2008 CARD PM2.5 Premature Deaths Teleconference Involving Enstrom, Pope, Jerrett. and 12 Dose-Response An International Journal Burnett Transcript and Audio File http://www scientificintegntyinstitute.org/CARB071108.pdf Accessed February 20, 2017. 25. Enstroni JF.. Analysis of HFI 2000 Figures 5 and 21 to Identify PM2 5 Mortality Risk in 49 US Cities Used in Pope 1995 and 11EI 2000 September JO, 2010. http/wuw.scientific integrityinstitute org/Htl FTgure5093010.pdf 26 Enstrom JE. Submission to UCl.A Research Integrity Officer Karagozian Challenging Jcrrcft ct al. PM;5 Mortality Findings and Karagozian Response. December 19. 2016. http Vscientiticintegrityinstitute.ore/RIOJerrettAII 121916.pdf 27 Enstrom Compilation of Rejections of This Paper by Seven Major Journals. http7/www.sciemiftcintegrityinstitute.org/ CPSIIRej 122716.pdf. Accessed February 20, 20P. EXHIBIT E Risk Analysis, Vol 16, SIa 0. 2tlf6 DOI 10.11 llTLsa.l-^17 Inconsistencies in Risk Analyses for Ambient Air Pollutant Regulations Anne E. Smith' This article describes inconsistencies between health risk analyses that the U S. Environmen tal Protection Agency (EPA) uses to support its decisions on primary National Ambient Air Quality Standards (NAAQS), and in the associated Regulatory Impact Analyses (RIAs) that accompany each NAAQS rulemaking. Quantitative risk estimates arc prepared during the NA AQS-setting deliberations using inputs derived from statistical associations between mea sured pollutant concentrations and health effects. The resulting risk estimates are not directly used to set a NAAQS, but incorporated into a broader evidence-based rationale for the stan dard that is intended to demonstrate conformity with the statutory requirement that primary NAAQS protect the public health with a margin of safety. In a separate process, EPA staff rely on the same risk calculations to prepare estimates of the benefits of the iule that are reported in its RIA for the standard. Although NAAQS rules and their RIAs are released simultaneously, the rationales used to set the NAAQS have become inconsistent with their RI As" estimates of benefits, with very large fractions of RIAs' risk-reduction estimates being attributed to populations living in areas that will already be attaining the respective NAAQS This article explains the source of this inconsistency and provides a quantitative example based on the 2012 revision of the fine particulate matter (PM ;) primarv NAAQS. This arti cle also demonstrates how this inconsistency is amplified when criteria pollutant co-benefits arc calculated in RIAs for non NAAQS rules, using quantitative examples from the 2011 Mercury and Air Toxics Standards and the currently proposed ('lean Power Plan. KEY WORDS: Renetiis: co-benefits; NAAQS: o/one: PM regulatory impact analysis 1. BACKGROUND When the primary particulate matter (PM;s) National Ambient Air Quality Standards (NAAQS) were first established in 1907 (one for annual aver age and one for daily average ambient PM;? con centrations), the principal basis for those standards w'as epidemiological evidence of positive statistical associations between ambient PM;? levels and ad verse health effects, including premature death risk. Address correspondence to Anne E Smith. Senior Vice Presi dent. NERA Economic Consulting. 125 5 23rd Street N\V. Suite 600 Washington. DC 200.17. USA; anne.smith@'nera.Com. ITiese reported associations, combined with a pre sumption that they represented a causal relationship, were also used to calculate quantitative public health risk estimates to supplemenl reasoning on setting the NAAQS. Quantitative risk analyses based on epi demiological evidence have continued to be a cen tral feature of the review process for revisions of the PM;,> NAAQS since then, and have also been a salient consideration in revisions of the NAAQS for ozone. This article focuses on a quantitative in consistency that has emerged between the rationale that U S. Environmental Protection Agency (EPA) Administrators use for setting a NAAQS when rely ing primarily on cpidemiologically-based health risk evidence, and the estimates of public health benefits 1737 0272-4332/16/OlOO-17~7$22.(IO/l C 2015 Society foi Riik Analysis 1738 Smith from those rules that EPA staff produces in its Reg ulatory Impact Analyses (RIAs).1 2. THE K VTIONALE H>K SETTING A PRIMARY NAAQS The Clean Air Act requires EPA2 to set the primary NAAQS for each criteria pollutant at levels that "are requisite to protect the public health" while "allowing an adequate margin of safety."1'' This determination must be made without regard to the potential cost of meeting the standard,12-' and legal rationales for choosing a NAAQS tra ditionally involved a balanced consideration of three attributes: (1) size of affected population, (2) severity of effect, and (3) certainty of effect.1131 However, the evolution since 1997 towards greater reliance on epidemiological evidence in setting a NAAQS forced a shift in how the rationale could be constructed, particularly for PNC 5. This was because the available epidemiological studies on several clearly adverse types of health effects (such as premature death) have not been able to identify a "threshold" or any other less sharp delineation of a point where the risk per unit increment of concentration appears to attenuate.' This situation eliminates the first two of the three above-mentioned considerations that EPA had typically relied on in ! A separate point of discussion rccarding the quantitative risk es limates is whethet the full body of scientific evidence is sufficient 111 aive confidence that these epidemiological associations reflect a causal relationship between the pollutant and health endpoint studied. This article does not attempt to add to that discussion 'Formally, under the Clean Air AcL llte responsibility fot deciding where to seta NAAOS is s-ested specifically 111 the Administrator. Throughout this article, when I use. the term TPA," I am refer ring to llte EPA Administrator When not referring to the Ad ministrator specifically. I use the terms EPA staff" or Agency." 'EPA staff and others often refer to this as a '"threshold" for ef fects, but the phenomenon being sought to help identify a pro tective level for a particular adverse effect need not be a point of sharp delineation where all population-wide effects end. Even evidence of diminishmtmi in the slope of the association would be helpful but has not been consistently found Lack of detection of such a dirmnishment in an association, even if the detected association is causal at relatively high concentrations, does not mean one does not exist at some relatively low concentration (sec Ref. -t p. 3R2). This is because the epidemiological techniques available have very limited ability to reliably discern the shape of a potential concentration-response relationship, and thus to in form the question of where or whether the association may- end. It is theoretically established that unavoidable inaccuracies in mea surement of an explanatory variable (e g., pollutant exposure) make it difficult to statistically delect a threshold or other non linearity at low concentrations even when it actually exists.1-1 NAAQS-setting rationales. I hat is. (1) the entire U.S. population is now implicated as at risk at every potential NAAQS level, and (2) the seventy of effect can no longer be seen to be changing as lower potential NAAQS levels are considered. As a result, consideration (3)-- uncertainty about the reliability of the epidemiologtcallv estimated association has become the only consideration remaining available to EPA for setting a primary NAAQS above zero that can be argued to be adequately protective of the public health as required by the statute. This shift in the nature of the scientific evidence for setting a NAAQS was so profound that the U.S. Court of Appeals ruled that the setting of a NAAQS under these circumstances amounted to an uncon stitutional delegation of legislative power to the Administrator unless she would first articulate an "intelligible principle" for how to draw that However, the Supreme Court overruled this finding,' with the result being that since then EPA's rationales for at least two of the NAAQS (i.e.. PNC v and ozone) have largely emphasized identifying a level at which continuation of the nonthreshold statistical health associations becomes too uncertain to indicate an actionable level of further public health nsk. The preamble for the 2012 PM>5 NAAQS decision provides an example. It starts bv noting that setting a standard based on epidemiological studies that cannot identify a population threshold requires a decision-making approach that "includes considera tion of how to weigh the uncertainties in the reported associations across the distributions of PM;, con centrations in the studies and the uncertainties in quantitative estimates of risk, in the context of the entire body of evidence before the Agency." M Later, the document states. "[i]n reaching decisions on alternative standard levels to propose, the Ad ministrator judged that it was most appiopnate to examine where the evidence of associations observed in the epidemiological studies was strongest and. con versely, where she had appreciably less confidence in the associations observed in the epidemiological studies,"(t| and after a detailed discussion of the epidemiological information states, "(t]he Adminis trator views this information as helpful in guiding her determination as to where her confidence in the mag nitude and significance of the associations is reduced to such a degree (emphasis added] that a standard set at a lower level would not be warranted to provide requisite protection that is neither more nor Inconsistencies in NAAQS Risk Analyses 1739 less than needed to provide an adequate margin of safely."'101 Similarly, in 2008 EPA used lack of confidence in continuation of the epidemiological associations to lower levels as its rationale for not setting the ozone NAAOS lower than 0.075 ppm despite clinical evidence in the record of health responses at yet lower concentrations. The ozone NAAQS preamble states: "A standard set at a level lower than 0.075 would only result in significant further public health protection if. in fact, there is a continuum of health risks in areas with 8-hour average O, concentrations that are well below the concentrations observed in the key controlled human exposure studies and if the reported associations observed in epidemiological studies are, in fact, causally related to O- at those lower levels. Based on the available evidence, the Administrator is not prepared to make these as sumptions. Taking into account the uncertainties that remain in interpreting the evidence from available controlled human exposure and epidemiological studies at very low levels, the Administrator notes that the likelihood of obtaining benefits to public health with a standard set below 0.075 ppm O decreases [emphasis added], while the likelihood of requiring reductions in ambient concentrations that go beyond those that arc needed to protect public health increases."'11 The U.S. Court of Appeals for the District of Columbia Circuit accepted this rationale and upheld the standard in 2013.* Although the NAAOS rationales are not written to conform to the terminology of probability or ex pected values, readers with decision analytic or other risk analysis training would be inclined to interpret the above quotes as expressing subjective judgments about the probability that the health relationships apparent in statistical associations cease to exist at some point on the continuum of lower and lower am bient pollutant concentrations. A decision-analytic interpretation of the above statements might be as follows. In order for a selected NAAQS level to be deemed as requisite to protect the public health, EPA's subjective probability that the relationship exists at and below the selected NAAQS level must, logically, be very nearly zero. (Indeed, the subjective probability of continued effects must fall to nearly zero at an ambient concentration somewhere above the selected NAAQS level. This is because the NAAQS needs to include at least some margin of safety, and thus must be set at least somewhat lower than the level where expected risk is deemed to become too small to be considered a public health concern.) 3. THE RESULTING INCONSISTENCY IN BENEFITS ESTIMATES FOR A NAAQS Thus, in setting NAAQS using epidemiological evidence, EPA has deemed quantitative estimates of health risks for concentrations below the NAAQS far less reliable and more inaccurate than the numerical precision with which they are reported. In essence, the NAAQS rationales give little or no weight to the subset of the quantitative risk estimates the Agency has placed in the record that have been calculated for pollutant concentrations below the selected NAAQS level. This lack of confidence in risk estimates from that below'-the-NAAQS range does not, however, make its way into the RIAs that accompany the re lease of the final rules. RIAs are documents that report on the benefits and costs of each major new regulation, such as a revised NAAQS. Federal regulatory agencies are required to prepare RIAs by Executive Order of the President. 14144 Although this requirement is unrelated to the legal requirements of the statute that motivates the regulation (such as the Clean Air Act in the case of air pollutant regulations). EPA's RIAs for air regulations adopt the same epidemiologically-based method of quantifying health risks used when deliberating where to set the NAAQS.4 The consistency ends there, however. At the same time that EPA is setting NAAQS at levels where it has minimal confidence that the public health is affected at lower concentrations, the Agency's RIAs arc giving the same weight to risks calculated for population exposures below the NAAQS level as they do to risks calculated tor population exposures above the NAAQS level. That is, RIAs assume elevated hazards exist with 100% certainty for all ambient pollutant exposure levels down to a zero concentration, inconsistent with EPA's judgments (formed when assessing those pol lutants' hazards), which imply nearly 0% certainty. EPA does not explain or try to justify why data that are too uncertain to use in the NAAQS preamble context are certain enough to use in the R1A con text. Although different certainty standards may be 4White the ''benefits" in an RIA are stated as a monetary value to be compared lo the regulation's costs, they are directly derived from quantitative estimates of physical health effects 1741) Smith justified in the context of decisions with different consequences, the contexts of a NAAOS preamble and that NAAOS's R1A are not very different at all. This inconsistency was not always as pronounced as it is now. Until 2009. risk reduction calculations used in air RIAs were at least truncated for pollutant concentrations below the lowest concentration level measured in the epidemiological study being used to make the risk estimates. RIAs would still include risk reduction estimates below the prevailing NAAOS level, as NAAOS levels have always been set at lev els above the lowest levels measured in the studies. However, from 2009 onwards. RIAs eliminated even that truncation, which resulted in a sudden and large increase in R1A benefits estimates for PM; - and ozone pollutant changes.'15 The fact that RTAs cal culate health risk reductions below the NAAOS, and now' down to zero, is widely known but the following examples quantify the extent to which this practice results in upward-biased nsk and benefits estimates. This author recommends that EPA staff more clearly communicate subjective epistemic uncertainty in its R1A benefits estimates. More specifically, the author recommends that the Agency's central estimates of benefits in its RIA be made consistent with the science policy judgments EPA makes in setting the criteria pollutant standards. This recommendation is in line with the need for more effective sensitivity analysis capabilities for health risk analyses, as described by Smith and Gans.,|r' 4. OVERSTATEMENT OF EXPECTED BENEFITS OF THE 2012 PM25 PRIMARY NAAQS REVISION The implications of this inconsistency are illus trated using as an example the RIA for the 2012 PM NAAQS rulemaking.1 In this rulemaking, the an nual primary standard for PM; - was tightened from an annual average of 15 to 12 /zg/m'. In the asso ciated RIA. a range of 460 to 1,000 fewer prema ture deaths per year was estimated from tightening the standard to 12 pg/m\ This range w;as derived by applying two different concentration-response func tions to the Agency's standard risk calculation for mula. The concentration response coefficient for the lower end of the range was derived using a coeffi cient from Krewski el and the upper end of the range was derived using a coefficient from Lepeule et ait,Vl A yet wider range of uncertainty in potential mortality risk reductions exists, as explained in Ref. 16, but the following discussion addresses only how the Agency's own range changes when the assump tions of the RIA's nsk analysis arc made consistent with EPA's reasoning when choosing how stringently to set the standard. Calculations were performed using EPA's Ben MAP model, which is a PC-based program that en ables users to compute health risks associated with criteria pollutants using the standard formulas that EPA uses in its own RIAs, and using EPA's or their own input files and other assumptions.'*01 The air quality input files that had been used for this RIA's calculations were obtained from EPA staff. After confirming that BenMAP does indeed repli cate the mortality reduction estimates reported in the RIA using those data, the same files were then used to assess the portion of the RIA's premature mortality estimates that are associated with the lin ear, no-threshold assumption that assumes that the risk relationship continues to exist below the selected NAAOS. This analysis found that 70% of the bene fits for the standard of 12 pglm wrere due to reduc tions in PM2.i from baseline levels that were already attaining (i.e., lower than) that standard. Given that the choice of a NAAQS level of 12 /zg/m' meant that EPA assigned too little con fidence in the continuation of health effects below 12 fig/W to warrant setting the NAAQS at a low;er level, standard decision analysis would assign negligi ble probability to calculations of benefits from reduc tions that would be occurring from levels below that NAAQS. That is, the expected values for 70% of the Agency's risk calculations should be approximately zero. When a threshold is assumed at 12 pglm Ben MAP calculates that the expected risk reduction of that NAAQS would be 138 to 313 fewer premature deaths per year, considerably lower than the 460 to 1,000 deaths reported in the RIA. (Dollar values of the benefits also fall proportionally.) As noted above, the rationale for the NAAQS arguably implies that some of the benefits de rived from locations with concentrations just above 12 /zg/m' also should be given less than 100% weight because of EPA's assurance that exposures to annual average concentrations of 12 pg/m'' are protective with an adequate margin of safety. EPA rarely if ever defines the magnitude of its margin of safety quan titatively. However, ranges for its magnitude could be tested with sensitivity analyses. If, for example, the margin of safety is taken to be about 1 /zg/m', and a threshold is assumed in the risk relationship 13 figjm'y BenMAP calculates the expected benefits associated with the selected NAAQS of 12 /zg/m are Inconsistencies in NAAQS Risk Analyses 1741 Table I. Fst rotates of Avoided Premature Deaths in 2020 tor the 12 /g/m' PM < N A AOS' RIA Assumptions Compared lo Alternative Views Suggested by EPA's Rationale foi that NAAQS Confidence Category (baseline PM; concentration i NAAQS Based Risk Reduction Estimate R lA-Based Risk Reduction E si mute (";. oi unal) Already attaining (<12 /tg/m!) Not attaining'in margin (e g. -12 to 13 /tiym ) Not attaining(above margin (e g.. >13 //pm I Confidence weighted Total usk reduction estimate Approximately 0 0-117 21 21-117 31b (7(1%) 117(26%) 21 (5%i 456 only 21 to 48 deaths, less than 5% of the RIA's esti mate of benefits from that standard. Whether the particular assumptions in this analysis about where the concentration-response re lationship begins to exist are reasonable or should be refined, its point is that the RIA's benefits estimates are very sensitive in the downward direction to expressions of declining confidence in continuation of the association at or just above the selected NAAQS level. The result is that the RIA benefits are substantially overstated compared to those that would more appropriately reflect the subjective weights expressed by EPA in its rationale for setting the standard at 12 uglm\ Table 1 contrasts the results of the RIA with judgments about confidence in those risk calculations that one might infer from the NAAQS rationale, and illustrates one way that RIAs could be enhanced to better communicate to the public the implications of the judgments made in setting the NAAQS for the rule's benefits estimates. For simplicity. Table 1 summarizes only the lower bound benefits estimate of 460 deaths (which BenMAP calculates more precisely as 456 deaths).' In this table, the risk estimates are divided into three "confidence categories." The lowest confidence cate gory is for risk reductions attributed to populations already residing in areas of attainment (i.e., with annual average concentrations less than 12 /ig/m3). (liven the NAAQS rationale, the public health risk is de minimis, and in weighted terms, would be nearly zero, while in the RIA. which gives 100% weight to all such risk calculations, benefits equal to about 318 deaths per year are assigned. The middle con fidence category' is for risk reductions attributed to populations in areas that are just above the NAAQS before the standard is implemented, but close 'The upper-bountl risk estimates would faU into the three rows in the table in the same proportions as seen for the lower-bound estimates in the table. Fig. I. Areas projected in the PMys NAAQS RIA to experience health benclits under the selected NAAQS of 12 fig'm' (456-1.03? avoided premature deaths, rounded to nearest death). enough to attainment that they might be viewed as being within the (undefined) "margin of safety." (For purposes of constructing the illustrative tabular sum mary, the margin of safety is assumed to be about 1 /zg/m* meaning that less than the NAAQS based weights would be declining or perhaps nearly zero even w ithin this category of baseline exposures.) To reflect risk estimates that fall in this category, the N AAQS-based risk reduction estimate is listed as be ing somewhere between 0 and 117, while the RIA would assign it 117 with 100% confidence. Finally, there are 21 avoided premature deaths estimated for populations living in areas well above the NAAQS. For this third category, the RIA's ben efits estimates can be considered consistent with the NAAQS-based rationale. Note that for the PMiv NAAQS RIA, this category accounts for only about 5% of the total RIA benefits estimate. It is recom mended that RIAs provide their benefits estimates for criteria pollutants in a format such as Table 1, and more explicitly provide weighted benefits estimates for confidence categories that are defined with respect to the NAAQS level. Geographical representation of where these health benefits are expected to occur is also interest ing to explore. The PM - NAAQS RIA calculated reductions in premature mortality only for areas that 1742 Smilli tig. 2. Sensitivity analysis of areas projected to expe rience health bent tits under the 12 NAAQS: (a) assuming benefits tor all baseline t*V12 5 levels; (b| assuming nsks exist only if baseline P\t2 5 is above 12 iig ms; (c) assuming nsks exist only tl baseline PV12.5 exceeds the selected standard by more than 1 erg/m3. 456-1033 avoided premature deatns 138-313 avoided premature deaths 21-48 avoided premature deaths were within 50 km of a monitor that the RIA's air quality analysis projected would not attain the newstandard under baseline conditions. Fig. 1 shows the locations in which the RIA's estimate of 460-1,000 avoided premature deaths occur. It is notable that all of those benefits occur in California. Fig. 2 zooms in on California to show: (a) the areas in Fig. 1 where benefits are attributed to reductions in PM: s at any level (i.e., showing the same areas as in Fig. 1): (b) the more limited areas projected to experience a health benefit when only reductions in PM: > that start above the 12 yug/nT NAAQS are considered; and (c) the even more limited areas if a 1 ^g/m-1 margin of safety is assumed to be associated with the selected standard of 12 nglm\ t hat is. Fig. 2(c) only gives weight to risks below 13 ng/m'' Both Figs. 2(b) and (c) reveal a far smaller area of at-risk populations than assumed in the RIA (i.e., than in Fig. 2(a)). This example from the PM NAAQS RIA brings to light another important uncertainty in its mortality benefits. All of the benefits estimates for the NAAQS of 12 /vg/m ' are based on PM changes in California. The risk calculations for changes in PM; 5 in California are performed using relative risk estimates denved from the entire United States, yet the epidemiological evidence that an association between PM: ; and all-cause mortality risk exists in California is tenuous/ Hence all of the above risk estimates might actually be zero, even if one does "The PMy RIA*171 piles seven California-specific PM;< cohort studies with alt-cause risk estimates and notes that tour have in significant associations while three have larger coefficients (Ref. 17 at p. 5. A-13). However, one of the three positive findings cited (i.e-, Oslro ei til.. 2010) was erroneous, according to an erratum published the following year (Ostro el h).. 2011). and the cor rected estimate of association was found to be insignificant. The remaining two positive findings cited were from the same cohort, one estimate being just an update of the other, t hus, the evidence for an all-cause mortality association in California alone consists of five null findings and one cohort with a positive finding. not wish to discount risks in areas already below the NAAQS. In other words, the much tighter 2012 PMy s NAAQS was set on the basis of projected mortality reductions that occur only in a part of the United States where the evidence of heightened mortality risk from PMy s appears to be weaker than in other parts of the United States. 5. ()V K R STAT KM ENT OK CRITK R1A POLLUTANT CO-BENEFITS IN NON-NA AQS RULEMAKINGS As explained in Ref. 15, epidemiologically- based estimates ol co-benefits from coincidental reductions of ambient criteria pollutants (especially PM ;) have also driven statements about regulatory benefits for a majority of non NAAQS air rulemakings in recent years. The upward bias in RIA benefits estimates becomes even more pronounced when co-benefits are calculated from coincidental criteria pollutant re ductions under regulations that do not relate to the NAAQS or regulations to help attain a NAAQS. Prominent examples are the RIAs for the Mercury and Air Toxics Standards (MATS) for electricity generating units promulgated in December 201T 11 and the Clean Power Plan (CPP) proposed in June 2014/-2' The MATS RIA projected PM s co-benefits in the hundreds of billions of dollars per year, based almost entirely on estimates of reduced premature mortality from reductions in PMy: 4.200 to 11,000 deaths per year. The reductions in PM_ in the MATS RIA are projected to occur when generating units are forced to install controls to reduce acid gas emissions, which will also reduce SO; emissions, a precursor to ambient PM: s formation. A figure in the MATS RIA reveals that over 99% of those projected benefits are projected to occur in areas where the PM; 5 levels will already be below' the PM: . NAAQS Inconsistencies in NAAQS Risk Analyses 1743 of 12 /ug/m:' (Figure 5-15 on p. 5-102 of Ref. 21). If the MATS rule's co-benefits arc calculated prob abilistically, accounting for the very low subjective probability that EPA assigned to the existence of the PM health effects relationships at levels below the NAAQS, the resulting estimate of expected ben efits from the MATS rule becomes nearly zero. The fraction of the PM? , co-benefits calculated below the NAAQS is much higher in the MATS RIA than the already high level of 70% that we have found for the benefits calculated for the PM?, NAAQS rule itself. This is due to the fact that bene fits in the RIA for the NAAQS rule were calculated only in areas within 50 km of a monitor that was pro jected to be out of attainment. By letting projected nonattainment constrain the geographical area over which benefits will be calculated, one ensures that a larger fraction of the resulting benefits will indeed be from areas above the NAAQS. However, when co benefits of some other rule arc assessed using PM?.risk relationships, no such constraint is applied. In the MATS rule, co-benefits were calculated across the entire nation, and furthermore, the units where acid gas controls were incremental to baseline con trols were more likely to be in areas already attain ing the NAAQS. As a result, nearly all of the PM?, co-benefits are projected in NAAQS-attaining areas. For these reasons, the bias in PM-, co-benefits es timates in RIAs for non PM?s rulemakings will tend to be much greater than the bias in the direct benefits estimates in RIAs for PM? s regulations. The same magnitude of overstatement of co benefits is apparent in the RIA for the proposed CPP RIA. which includes co-benefits for both PM2.5 and ozone. In the CPP RIA (focusing, for simplicity, on its Option 1 with state-level implementation) the PM- v co-benefits of the rule are estimated to be up to 4,100 deaths in 2020 and up to 6,200 deaths in 2030, and the ozone co-benefits are estimated to be up to 170 and 440 in those respective years (Tables 4 16 through 4-18 on pp. 4-34 to 4-36 of Ref. 22). Unlike the MATS RIA. the CPP RIA does not provide any information on the fraction of these co-benefits that are calculated for areas already at taining those two NAAQS, but they can be inferred by replicating the co-benefits calculations from other data in the RIA. Recalling that the PM'; NAAQS RIA indicates that only California will be exceeding This involves using (Jala on emissions reductions of the PM? 5 and ozone precursor emissions in the RIAs Table 4-10. and multi plying them by the incidei>ce-per-ton estimates in Tables 4A-5 through 4A-7 the PM? s NAAQS in 2020. only California-based PM? s co benefits estimates could be associated with exposures in the above the NAAQS category: less than 1% of the CPP RIA's PM?.s co-benefits are attributable to changes in emissions in California in 2020. Furthermore, the PM. ^ NAAQS is supposed to be fully attained by 2020, so even that sliver of the PM-, co-benefits attributable to California are sup posedly in an attainment area. Although California is not projected to attain the ozone NAAQS before 2030, less than 0.5% of the ozone-related co-benefits are associated with changes in ozone precursors in California. Thus, in the CPP RIA as well in the MATS Ri A, more than 99% of the co-benefits would be discounted if health risks below the NAAQS are assigned a much lower probability (or confidence weight) than risks above the NAAQS. 6. CONCLUSION In conclusion, we find that a large majority of the Agency's estimated health benefit from the 2012 PM? NAAQS are attributable to reductions of PM? ? in areas that are already in attainment of the PM NAAQS. RIA calculations of risk reduc tion in areas already attaining the new NAAOS are given the same weight (i.e., subjective confidence level) as projected benefits from areas that would be exceeding the NAAQS. These RIA calculations are based on assumptions that are inconsistent with the rationale for that NAAQS. The above sensi tivity analyses show that this causes RIAs' benefits estimates to be much larger than estimates of the expected benefits that can be reasonably inferred from EPA's NAAQS-setting lationale. I he over statement becomes nearly 100% for co-benefits from criteria pollutants in RIAs for non-NAAQS regula tions, such as the MATS rule and the proposed CPP rule. RIAs should be written to reflect consistency with EPA's NAAQS policy judgments. Precise con fidence weights will likely never be articulated, but this article has shown that the quantitative impor tance of such policy judgments for benefits estimates can be communicated to RIA readers in simple formats. It is the opinion of this author that such quantitative disclosure is important to maintaining credibility and trust in the Agency's RIAs. ACKNOWLEDGMENTS This work was conducted with funding from the Electric Power Research Institute. The author thanks Ms. Reshma Patel for her analytical support 1744 Smith in preparing the analyses used in this article. The au thor also thanks three anonymous reviewers for their comments and suggestions. Any errors remain the author's sole responsibility. REFERENCES 1 Section ICW (h)(l). Clean An Act 42 USC 1)7409 2. Whitman \. American Trucking Associations. 531 U.S 47?- 476(2001) ?. Lead Industries Association Inc v Environmental Protection Agency. 647 F.2d 1130(1980) 4. Smith AE. Response to commentary bv Fann et at. on ``En hancing ihe characterization of epistemic uncertainties in PM; s risk analyses." Risk Analysis. 2015; 35(3):3R1-3S4 5. Brauer M. Brumm J Veda I S Petkau At. Exposure misclassificatiun and ilireshold concentrations in time series analyses of air pollution health effects Risk Analysis 2007- 22(5):11831193 6. American Trucking Associations v EPA 175 F 3d 1027 1034-1037 (D C Cir. 1999). rehearing granted in part and de nied in part. 195 F 3d 4 (D C. Cir 1999). affirmed in part and reversed in pan. Whitman v. American Trucking Asso ciations. 531 U.S 457 (2001). 7. Whitman v. American Trucking Associations. 531 U.S. 473470(2001). S. 7S Fed Reg. 3086 at 3098. 9. 78 Fed Reg. 3086 at 3139. 10. 78 Fed Reg. 3086 at 3161. IT 76Fed Rec. 16436at I64S? 12 Mississippi's EPA. 744 F.3d 1334 (D.C. Cir. 2013). 13. Executive Order 12866. 19a? Regulatory Planning and Re view.'' s8 Fed. Her. 51735. October 4 Available at www whitehouse.gov,'omb'inforeg/eol2866 pdf 14. Executive Order 13563 2011 "Improvine Regulation and Regulatory Review." 76 Fed. Reg. 3821. Januaty 18. As ailable at: http www.regulations cos 'exchange'siics default dies doc .files President%27s"..20Executis e%200i dei't^>2013563 0 pdf 15. Smith, AE An evaluation of ihe PM; x health benefits esti mates in regulatory impact analyses foi recent air regulations. Reporl prepared for the Utility Air Regulatory Group, December 2011. at p. 24. Available at: http www neta.com/ cc'nlcnt/dam tie nrPublications'archive? PUB_Smith_Ozone NAAQS 07ll.pd! 16. Smith AE, Cians W. Enhancing the characterization of epis temic uncertainties in PMi s risk analyses. Risk Anals sis. 2015: 35(3):361-78. 17 EPA. Regulaloiv impact analysis for the final revisions to the National Ambient Aii Quality Standards for particulate mat ter. EPA-452/R-12-1X13. December. 2012 IS Kress ski D. JcjtcU M. Burned RT. Ma R, Hughes E. Shi Y. Turner MC. Pope CA 3rd. Thurston G. Oalle EE. Thun MJ. Beekerman B. DcLuca P Finkclstcin N lio K. Moore DK. Newhold KB. Ramsay T. Ross Z. Shin II. Tempalski B Extended follow-up and spatial analysis of ihe American Can cer Society Study linking particulate air pollution anil mortal ity. Research Repoit Healih Effects Institute. 2009:1405-114. 19 Lepeule J. Laden F Dockery D. Schwartz J Chronic exposure to fine panicles and mortality. An extended follow up of the Harvard Six Cities Study from 1974 to 2009 Environmental Health Peispectives. 2012:32(1 ):K 1-95. 20. BenMAP version 4.0.67 Available at httpt'-'www epa.gov-'air/ benmapAiownload.html. 21 EPA Regulatory impact analysis for the final mercury and air toxics standards. EPA-452/ R-l 1-011. December 2011 22. EPA. Regulatory impact analysis for the proposed carbon pollution guidelines for existing power plants and emission standards for modified and reconstructed powei plants, EPA 542/RI4 DO? June 2014 EXHIBIT F Ri-l Analysis. VoL 36. No. 9. _W6 DOT: 10.1111 risa. 1^670 Invited Commentary Rethinking the Meaning of Concentration-Response Functions and the Estimated Burden of Adverse Health Effects Attributed to Exposure Concentrations Louis Anthony (Tony) Cox Jr.' Four articles by Ancnbcrg ct ah, Fann ft at.. Shin etal., and Smith contribute valuable perspec tives and syntheses to a large and growing literature that estimates the burden of mortality risks attributed to fine particulate matter (PM2.5) based on estimated epidemiological as sociations, summarized as concentration-response (C-R) relations. This comment questions the use of C-R relations to predict or estimate how changing exposure concentrations would change responses in a population. C-R associations typically reflect modeling choices, and equally good choices can commonly lead to conflicting conclusions about the signs, signifi cance, and magnitudes of C-R relations and regression coefficients. This indicates that C-R relations do not necessarily reflect underlying stable causal laws useful for making risk pre dictions. but only choices about how to describe past data, with no uniquely correct choice being determined by the data. Similarly, currently available C-R data typically do not suffice to make valid predictions about how future changes in concentrations will affect responses. These difficulties can be substantially overcome by model ensemble and causal graph model ing and time series methods, blit these require different data and knowledge- for example, knowledge of how multiple variables depend on each other, rather than only of how one dependent variable is associated with multiple explanatory variables--than that captured by traditional C-R models or expressible by any single C-R coefficient or curve. KEY WORDS; Air pollution, ambiguous association;causalitv. causation: concentration response tunelion: line particulate matter 1. INTRODUCTION To what extent can historical exposure-response (C-R) associations be used to predict correctly how future changes in exposure would affect responses? A voluminous literature, including considerable au thoritative expert opinion, assumes that observed C-R relations and associations can be used to give useful guidance to policy makers about how changes in concentrations should be expected to change health responses in populations. If the C-R model is some regression curve or function response -- concentration), then this view holds that chang ing the concentration from old level x to new level 'Cox Associates and University of Colorado: tcoxdenver aol.com. y should change the population response from /(.v) to f{y)- The purpose of this comment is to dispute that view. An excellent indeed, even a perfect descriptive model of the relation between past levels of exposure concentrations and responses does not necessarily or usually allow us to predict how chang ing concentration would change responses. C-R functions estimated from past data are widely used to estimate the human health burdens of different exposures and to project how changes in exposures would change health impacts. For example, in this issue of Risk Analysis, Shin et aV mention that "[o]ur approach to characterizing the shape of the exposure-response function is based on only summary information available in the open literature: relative risk estimates and the exposure 1770 0272-4332/16/0100-1770522JMVI 2016 Society for Risk Analysis Invited (oiiimentan 1771 distribution for each study." They suggest that their approach based oil statistical associations (relative risk estimates) is "more useful for burden estima tion" and can be used to "present an example of mortality risk due to long term exposure to ambi ent fine particulate matter," reflecting a belief that relative risks provide a useful basis for estimating dis ease burdens and mortality risks due to (presumably meaning caused by, and reducible by reducing) ex posure concentrations. Similarly, Fann et al.i2) state: "At the core of these assessments are judgments about the likelihood that PM2.5 is a causal factor in mortality and about the choices made to characterize the C-R function that quantified the relationship between changes in concentrations of ambient PM2.5 and the risk of premature mortality. ... Asa final evaluation, we examine the implications of any differences among the quantitative methods for esti mating the number of avoided PM2.5-related prema ture deaths, including uncertainty, for an illustrative policy scenario." Again, this appears to reflect a belief that C-R functions can tell us how "changes in concentrations of ambient PM2.5" would affect "risk of premature mortality" and "the number of avoided PM2.5-related premature deaths." Smith notes that "|q|uantitative risk estimates are prepared during the NAAQS-setting deliberations using inputs de rived from statistical associations between measured pollutant concentrations and health effects. The resulting risk estimates are not directly used to set a NAAQS. but incorporated into a broader evidence-based rationale for the standard that is intended to demonstrate conformity with the statu tory requirement that primary NAAQS protect the public health with a margin of safety.Anenberg era//4' state that their article "reviews 12 current and publicly available multinational tools that combine air quality information, epidemiologically-derived concentration-response associations, and demo graphic data sets to estimate air-pollution-related health risks." These descriptions correctly reflect the reality that current quantitative risk estimates used to inform regulatory policy deliberations about health harms caused by exposures to pollutants are based on epidemiologically-derived statistical associations (e.g., C-R regression coefficients and relative risks or odds ratios), perhaps augmented with expert judgments about whether exposure is "a causal factor," rather than on quantification and validation of direct and indirect causal im pacts of changes in concentrations on changes in responses. The purpose of this comment is to challenge the belief that statistical associations and C R relations between historically observed exposure concentra tions and responses provide the information needed to draw valid causal inferences about how changing exposure concentrations would change responses. This typically cannot be determined from data on past C R associations, even with the help of expert judgments. Instead, such inferences require understanding how changes in causes will change effects. This requires data, knowledge, and analyses different from those in C-R functions. To fix ideas and to concretely illustrate methodological points, we use both simple examples and a publicly available data set from the Los Angeles air basin, kindly pro vided by Dr. Stanley Young. The data, and statistical tools for analyzing it, are included in the free Causal Analytics Toolkit (CAT) software described at https://reguIatorystudies.columbian.gwu.edu/causalanalytics-toolkit-cat. 1.1. Which C-R Relation? Signs, Magnitudes, and Significance of Associations Depend on Modeling Choices When an associational analysis of a data set, such as a regression model or a relative risk, odds ratio, or slope factor calculation, reveals a significant posi tive C-R relation, it is tempting to think that one has thereby learned something about the real world: that higher concentrations are associated with higher re sponse rates. But this natural interpretation is often mistaken. Usually, a significant positive association shows only that the investigator has selected a model that gives a significant positive C-R association (e.g.. slope or regression coefficient) when applied to the data set. Selecting different, equally good (or better) models (by any criterion) for the same data might produce no significant positive association, or a sig nificant negative association. This poses a method ological challenge, recognized by Dominici et al.} wrho noted that associational methods are unreliable in general, as their results can be reversed by making different modeling choices. Table I illustrates an extreme case of this point using a simple hypothetical example for three communities, A. B, and C, having different concen trations of PM2.5 and different elderly mortality rates in 1980. The response rate of elderly mortalities per 1,000 people over the age of 75 per year is observed to be proportional to (and, on these scales, double) the concentration of PM2.5. Does this justify 1772 Cox Tahiti. Whal Conclusions Do These Data Warrant About How Changing PMZ 5 Would CTnrye Elderly Mortality Rates' Community A B C l'M: 5 in 1980 0/g/m') 4 8 12 Elderly Mortality Runin 1980 (per 1,000 People Over Ape 25 per Year) 8 16 24 20 mortalities per 1.000 elderly people per year. On the other hand, a different investigator, persuaded that income matters a lot hut agnostic about the effects of ambient levels of pollution matter, might fit the alternative Model 2 to the same data. Average Elderly Mortality Rate 35--0.5 x P M2.5 Concentration -0.25 x Income (Model 2) Titbit II. Should the Estimated PM2.5-Mortality C-R Relation Be the Same as for Table T? Community PM2.5 in 1980 </ig'm!l Income Elderly Mortality Rate in 1980 A 4 100 S B B 60 16 C 12 20 24 an inference that reducing exposure concentration would reduce elderly mortality rates? Does it justify the stronger conclusion that every 10 qg/m ` change in PM2.5 should be expected to produce a corre sponding change of 20 mortalities per 1.000 elderly people per year? (Assume very large sample sizes, as the point of this example is not to quibble about the hypothetical numbers, but to scrutinize the logic of whal may be validly inferred from such data.) Before answering, consider Table fl, which shows Ihe same data as Table 1, augmented bv an additional column for average per household income per year in each community (in thousands of dollars). The income numbers are lower where the PM2.5 numbers are higher, so that higher PM2.5 is associ ated with lower income. Given these data, an investigator who believes lhat income is irrelevant and that PM2.5 is a poten tially important cause of elderly mortality might tit the following structural equation (causal) model to the data: Average Elderly Mortality Rate = 2 x PM2.5 Concentration (Model 1) Interpreting this causally, a change in the right hand side variable, PM2.5 concentration, is predicted to cause an adjustment in the left-hand side (depen dent) variable, elderly mortality rate, until equality is restored. Every 10 KgOn ' increase in PM2.5 is then predicted to produce a corresponding increase of Interpreted causally. Model 2 predicts that each 10 pg/m increase in PM2.5 will cause a reduction of 5 mortalities per 1.000 elderly people per year. (Empirically, in a study of 27 U.S. communities, about a third of observed estimated C R relations for PM2.5 and mortality were negative, three of them significantly so.16' Toxicologically, relatively low lev els of exposure to particulate matter can result in upregulation ol the production of endogenous protec tive antioxidants in the lung, although higher expo sure concentrations overwhelm this protective effect with increased reactive oxygen and nitrogen species (ROS and RNS).(71 Thus negative coefficients may not be altogether far-fetched, notwithstanding the claim of Shin et al. that "any reported negative statis tical estimates of the relationship between PM2.5 and mortality must be due solely to statistical error ") Which C-R regression coefficient, 2 in Model 1 or --0.5 in Model 2. should be considered to charac terize "the" C R relation between PM2.5 and elderlymortality? In principle, the answer is not determined by the data: Models 1 and 2 both tit the data equally well. In practice, it is determined by which model the investigator chooses. (In practice, also, it is unlikely that different models will all explain the data petfectly, as in this example. But the same key point holds: different models that fit the data approximately equally well, and better than other models, often yield very different conclusions.) Thus, a study that cites Model l's regression coefficient of 2 as evidence that reducing PM2.5 would reduce elderly mortality should not be accepted as credible, as it shows only that the investigator selected Model 1 instead of Model 2. It does not necessarily reveal how future real-world changes in PM2.5 would change real-world mortality rates. Similarly, con cluding on the basis of Model 2 that reducing PM2.5 would increase elderly mortality would simply reflect a choice of a model that implies this conclusion instead of a different model that implies its opposite. Such ambiguous associations--that is, associations that depend on modeling choices, and that can easily Imitcd (omiiienUin 1773 be reversed by varying the modeling assumptions-- make the conclusions from association-based meth ods (including traditional C R modeling) unreliable. Real-world data frequently exhibit the key feature of this example: signs and magnitudes of the estimated C R relation vary widely with modeling choices.1 For example. Table 111 shows a Poisson regression model for number of daily mortalities among people 75 or older (a count variable) fit to the Los Angeles data set (available in the previously mentioned CAT package), using the generalized linear modeling ((glm)) package in R. To facilitate replication and reanalysis for interested readers who do not use R, the Appendix provides an Excel alternative to the R analyses. Table III shows the result of regressing daily mortality counts for people aged 75 or older, denoted by AllCause75, against same-day values of minimum and maximum daily temperature {trnin and tmax), maximum relative humidity (MAXRH), and the month and year for the observations. The regression coefficient for PM2.5 is significantly positive, p = 0.00038. If an alternative Poisson regression model is lit to the same data, with month treated as a discrete factor instead of as a con tinuous predictor (with possible values of JanuaryDecember coded as 0-1 dummy variables by replac ing month with as.factor(month) in the R model), then PM2.5 is no longer a significant predictor of el derly mortality at the conventional 0.05 significance level (p -- 0.06 in a Poisson model,p - 0.09 in a quasiPoisson model). If values of PM?..5 lagged by one. two, or three days are included as predictors, then the only significant C R coefficient between PM2.5 and elderly mortality counts is negative [p 0.006 if month is treated as continuous; p = 0.03 if month is treated as a discrete factor), at a lag of three days. Thus, the finding of a significant positive C-R re lation between PM2.5 and elderly mortality counts in Table III is in a very real sense created by model ing choices to "control" for the effects of month us ing a relatively inflexible (e.g., linear) model and to include only same-day values of variables as predic tors. Different modeling choices remove or reverse the finding, consistent with the warning of Dominici et al.''1 In practice, investigators often use more sophis ticated models (e.g., splines with an investigatorspecified stiffness and number of knots to account for the smoothed effects of seasons and trends; or vary ing combinations of lags for different predictors). But the fundamental methodological problem remains: any significant C-R associations found may simply reflect the particular modeling choices made This is not a problem that can easily be overcome by appeals Lo expert judgments, for example, by inviting selected experts to opine about the causal interpretation of reported findings, insofar as the experts themselves do not know what other models would have shown. Sensitivity analyses for a selected model may reveal the sensitivity of its conclusions to variations in its as sumptions, but without revealing whether taking very different models as a starting point would have led to very different conclusions. Model ensemble meth ods, especially with nonparametric models (such as the popular random forest algorithm, discussed later) provide a possible constructive solution to these chal lenges by examining the distribution of C-R esti mates from hundreds or thousands of models, but have been criticized on the grounds that their results do not necessarily reflect the beliefs (or "substantive knowledge") of subject matter experts.'1*1 2. ASSOCIATIONS BETWEEN LEVELS DO NOT PREDICT ASSOCIATIONS BETWEEN CHANGES Suppose that all PM2.5 levels in Table II were to be cut in half between 1980 and 1990, as shown in Fable IV. Given the data available in 1980. can the effects on elderly mortality of this reduction in PM2.5 be predicted? Specifically, what will the mor tality rates for communities A C in 1990 be, and how sure can we be? This prediction challenge highlights the distinc tion between finding an associational C-R relation that describes the association between past levels of exposure concentrations and past levels of responses (as Models 1 and 2 do for the data in the top half of Table IV) and developing a valid causal C-R re lation that predicts how changing the levels of expo sure concentrations would change the responses, al lowing the ' values in the bottom half of Table IV to be predicted correctly. Different models and meth ods are typically needed for these two different pur poses. But the C-R literature to which Shin et al.,' Fann et al.,{ U Smith,*' and Anenberg et a/.141 refer thoroughly conflates them. Although it may be reasonable for policymakers to ask risk analysts how halving exposure would af fect mortality rates, the unfortunate truth is that the data in the top half of Table IV7 do not permit valid answers to the question: they place no constraints on the possible values of the three unknown (in 1980) quantities in the lower right, that is, the 1990 1774 Cox Table 111. A Poisson Regression Model for Daily Elderly Mortality Counts (AllCttuye?*) Dependent Variable: AllCau**75 Estimated Coefficients (Intercept) PM2.5 tmin irnax MAXRH year month Estimate 3.684524 0.(100745 -0.00.i820 -0.001776 -01000961 o.ooos?? -0.009686 Std. Error 4.126877 0.000210 0.000517 0.000369 0.000194 0.002056 0.000668 7. Value 0.89 3.55 -7,38 -4.81 --1,96 0.41 -14.50 Fr(,|-|i 1137195 0.000'S 1 .fie-13 1.5c-06 7.0e-07 " 0 68536 <2e-lfi "" Significance codes: 0"' A 0.001 " : 0 (H 0.05 V; 0.1 -T Source. The numbers in Table III can be obtained by applying the following R commands applied to the La data: fit - glmt AlICause75 - PM2.5 + tmin 4 tmac MAXRH -- month t year, data -- data.lramelPM2j . tmin. tmax, MAXRH, month, year), family -- poissnnOI: Mimmary(fit). Table I\. Effects on Mortality of Future Changes in Exposure Are Underdetemiined by Past Exposure Response Data Com m unity A B C A b c Year 1980 1980 1980 1990 1990 1990 PM2.5 t/jg/m ' 4 8 12 2 4 6 Income too 60 20 too 60 20 Elderb Mortality Rate in 1980 S 16 24 9 9 9 mortalily rates in the three communities. For exam pie, if the values of these three quantities (from top to bottom) turn out m 1990 to be 8. 16. and 24. the same as in 1980, then we will have learned that changes m PM2.5 appear to have had no impact, and that perhaps only income matters for predicting mortality rates. If instead they turn out to be 4. 8. and 12, then we might conclude that income has no discernible effect, and that halving exposure concentrations halves elderly mortality rates, in accordance with Model 1. If the numbers instead are eventually revealed to be 9, 18, 27, then we could conclude that Model 2 appears to have been right, proving that a positive association between C-R levels docs not logically imply a positive association between future changes in their levels, (Decreases in PM2.5 of -2, -4. and -6 pg/nr in communities A. B. and C from 1980 to 1990 would correspond to increases in daily mortality of 1, 2. and 3 deaths per day, respectively, even though daily mortality is proportional to PM2.5 in 1980.) Other outcomes in which both income and pollution affect mortality in various ways can readily be envisioned. The key point is that the data avail able in 1980 provide no sound logical or statistical basis for predicting what the three values denoted bv "?" in the lower right of Table IV will turn out to be in 1990. Specifically, it would be incorrect to assume that a C-R relation such as Model 1 or Model 2 es timated from data in the top half of Table IV allows prediction of how changes in PM2.5 would change el derly mortality rates in the bottom half of Table IV. (In the terminology of econometrics, this assump tion confuses a "reduced-form" regression equation, which is what air pollution researchers typically work with, with a "structural" or causal equation, which is what policymakers need.) It is instead necessary to learn about how the world works--specifically, how changing one or more variables changes others that depend on them rather than assuming that past associations can be used to predict future changes. Asking experts whether they think that historical associations are causal, or whether concentration is "a causal factor" (to an unspecified degree) for response, in the terminology of Farm et all1 does Invited C ommentary 1775 Hfc. 1 A Bayesian network (BN) mode! for the LA data, show ing statistical dependencies among variables, indicates thai AllCatt.uS75 does not depend directly on PM2-5. Sourer, Fig- 1 was generated by running R package |bnleam\ on the LA da(a using the default hill-climbing (hr) method, as follows: libraryfhnleam) library! Rgraph viz): bn <-- hc(tlata.frame (year, mnnlh AllOius-75 PM2.5, imin rmax, MAXRH)). graphvi?.,p]i)t(bn. shape - `'rectangle"). not solve the problem that associational models do not in general reveal how future changes in concentrations will affect future changes in response. 3. TOWARD MORE RELIABLE AND PREDICTIVELV USEFUL CAUSAL C R RELATIONS The foregoing comments do not imply that mod els cannot or should not be used to predict (or at least to constrain predictions about) how changes in exposure concentrations would change responses. They only imply that quantifying C R associations between past concentration and response levels, for example, using reduced-form regression models, and then extrapolating these C-R relations to predict future changes is not the way to do it. But many other risk analysis methods and models, from simu lation to causal Bayesian networks, are available to help identify, test, and quantify the changes in re sponses caused directly and indirectly by changes in exposures.w For example, Fig. 1 shows the structure of a Bayesian network (BN) model learned directly from the LA air basin data for Table III via a datamining algorithm that searches through a set of many possible models to find one that best explains the data. All variables in the LA data set axe included, although day of the month turns out not to signiftcantly affect the frequency distribution of any other variable in the data set: thus, it appears as an isolated node in Fig. 1. In Fig. 1. the nodes (rectangles) represent the variables, that is, columns in the underlying data set. The arrows signify "is not statistically independent of" (or. more colloquially, "is informative about"). An arrow from X to Y implies that the conditional frequency distribution of Y differs significantly for at least some different values of A (and for at least some configuration of values of other variables pointing into V, if any); otherwise, they remain unlinked Ar rows between variables can typically point in either direction, by Bayes's Rule. (Technically, the arrowdirections in Fig. 1 indicate one of many ways to compute the joint distribution of the variables from the marginal distributions of the input variables-- those with only outward-pointing arrows [here, year and month]--and the conditional probability tables for all other variables, specifying the conditional probability for the value of each variable for each combination of values of the variables that point into it, Continuous variables are automatically binned into deciles to permit these conditional probability calculations. For more details on Bayesian networks and structure learning algorithms, see https://cran.rprojeet.org/web/packages/bnlearn/bnleam.pdf and the references therein.) In Fig. I, AHCaii$e75 does not dctectably depend on PM2.5 once the daily minimum and maximum temperatures tmin and unax and month are known; more accurately, PM2.5 provides no further information that improves ability to pre dict AllCause75. In this sense, the observed data provide no evidence that PM2.5 directly affects AllCause75. Interventions that change the daily temperatures experienced by elderly people might affect both PM2.5 and AUCausc75, but interventions that change only PM2.5 and not daily temper alurc should not be expected to change elderly mortality rates, even if regression models show a statistically significant C R association between them, as in Table III, Relations between conditional independence, interventions, and prediction of causal effects are discussed further in the causal modeling and causal graph literature and supporting software packages (see e.g., https://cran.r-project. org/web/packages/causaleffect/causaleffect.pdf and references therein, and Cox|()|), 1776 Cox Partial Dependence on "tmax-' Fig. 2, A partial dependence plot for AUCause75 versus tmax, showing that daily elderly mortality counts are predicted to be smallest for days with maximum daily temperature of about 86 'C. Source: Fig. 2 is generated by applying the partialPlat() func tion in the \nmdomForest) R package to the LA data, as fol lows' library!randoniForesi): data - dala.framefvear, month, day. AIICause75. PM2 5. tmin. tmax. MAXRIf); partialPlot (randomForeshdaiit. All('ause75). pred.data = data, war "tmax") Although the algorithms that search through al tentative Bayesian network models to identify those that best explain the data can be complex, and the interpretation of network diagrams such as Big. 1 can he subtle, it ts easy to understand the main insights from such automated analyses: that temperature and humidity are associated with (more precisely, informative about) PM2.5 and that it mortality is driven primarily by these meteorological variables and not PM2.5 levels, then PM2.5 will be associated with mortality rates, but weather variations, rather than fluctuations in PM2.5, contribute causally to fluctuations in mortality. Beginning with such data-driven findings about the qualitative structure of dependencies among vari ables, causal modelers and risk analysts can then quantify the relations between outcomes of inter est and factors that might affect them (e.g., those to which they are linked by arrows in a BN) without having to commit to any specific model or small set of models. For example, Fig. 2 shows a partial depen dence plot for how elderly mortality (AltCause75) varies with maximum daily temperature {tmax) when only tmax is assigned alternative (counterfactual) values and all other variables have their actual val ues in the data set. (Further background on random forest al gorithms, which are among the most successful machine-learning techniques, and partial depen dence plots is provided at hUps://cran.r-project.org/ web/packages/random Forest/random Forest, pdf. The partial dependence plot is similar to an added vari ables plot for a regression model, but uses a random forest model ensemble of many classification and re gression trees instead of a single regression model to predict the value of a dependent variable for differ ent values of an independent variable. For the LA data set. the random forest mode! ensemble explains 41% of ihe variance in AllCause75, as contrasted with about 32% explained by the best single linear regression model.) The nonparametric model ensemble that gener ated Fig. 2 easily detects and quantifies the nonlinear dependence of elderly mortality counts on tmax. It could as easily do the same for PM2.5 if elderly mor tality counts depended on PM2.5 levels. It is based on an ensemble of several hundred automatically gener ated nonparametric models (about 90 suffice to sat urate the predictive accuracy of the ensemble), and thus avoids the need to make model selection and specification choices that might otherwise undermine the reliability of the results. Ensembles of BN models can be developed and used similarly. The challenge that future changes in responses caused by changes in concentrations are under determined bv data (Table IV) can be substantially met with the help of causal graph algorithms devel oped to check sufficient conditions for transporting observed conditional probability relations across contexts (e.g.. the [causaleffeci] package in R) and time series methods developed to check the stationarity of relations among different time series (e.g.. the \changepoint\ and [ecp] packages in R). Fi nallv. BNs and related graphical models offer a more nuanced description of causation than "judgments about ihe likelihood that PM2.3 is a causal faetoi in mortality"^ by making explicit both direct and indirect paths between exposure concentration and response variables and by allowing die fraction of a OR association that is due to a direct link be tween them (if any) lo be estimated distinctly from the fractions due to other pathways, These methods allow genuine prediction of how changes in expo sures will change effects once genuine, stable causal relations or laws have been identified and validated. They can augment older regression-based models lo Invited Commentary 1777 substantially address the objections raised in the ex amples of Tables I IV. ACKNOWLEDGMENTS I thank Area Editor Warner North for a close reading and very valuable suggestions that helped simplify and clarify my exposition of several tech nical points. Warner also suggested introducing an appendix and references for more detailed technical exposition. The final exposition has been improved by his insightful advice. APPENDIX: CAUSAL AN ALYTICS TOOLKIT (CAT) FOR EXCEL USERS A free Causal Analytics Toolkit (CAT) pack age may be downloaded from https://regulatory studies.columbian.gwu.edu/causal-analytics-toolkitcat to give readers access to the Los Angeles data set provided by Dr. Stanley Young and to enable independent replication, new analyses, or exten sions of the analyses summarized in Table III and Figs. 1 and 2. CAT is an Excel add-in that provides relatively simple commands and a point-andclick interface for doing advanced analytics from Excel using R packages, even if the user does not know R. A user guide that describes how to download and install CAT is available here: http://cox-associates.com/CAT/UserGuide.pdf. Once CAT has been downloaded and installed, it can be enabled as an Excel add-in (select File > Options > Add-Ins > Go and check Causal Analytics Toolkit, then click OK to add it to the Excel toolbar). To install the LA data used in Table III and Figs. 1 and 2, open a new Excel workbook and click on the "Excel to R" button at the far left of the CAT ribbon; then select "Samplel" to download the data. The newly created data set should look as shown in the screen shot (with 1,401 records) on an automatically created sheet named "Data." Select/highlight the data (columns A H) and click on the Excel-to-R button at the far left of the CAT ribbon to import the data to R. (Click "Yes" to accept the default of creating an R data frame called "Data" corresponding to the data on the Data work sheet.) The data are now ready for analyses using CAT. To replicate the Poisson regression model in Ta ble III, select the dependent variable, AllCause75 (i.e.. click on column D to highlight it). Then use Ctrl -click to select the independent variables. (Multi ple adjacent columns E-H can be selected by swiping across them while holding down the Ctrl key. Then, go to column B and use Ctrl-click to select month for inclusion in the model.) Click on P Poisson in the Regression Models area of the CAT ribbon to gen erate a new sheet with the results of the Poisson re gression model; this is how Table III was produced. For users who would rather type than point and click, entering R: CAT_poisson(AllCause75, PM2.5, tmin. tmax, MAXRF1) in anv cell will produce the same Poisson regression model output immediately below that cell. (Once CAT is installed, entering "R:" into any Excel cell makes it behave like an R command console, ready to receive and process R commands.) Returning to the Data sheet with the columns al ready highlighted, and using Ctrl-Click to add year and day (or simply reselecting all columns A- H), and MAI___ _ i, PE Ix--it- Ole .jgiK'OtM n a- C-v*n DU Smpiri a fe^jl .. 7 2007 3 2007 1 4 2007 s 2007 i 6 2007 i 7 2007 1 & 2007 1 9 2007 i 10 2007 i 11 2007 i ' c i 2 3 4 5 6 7 a 9 10 _*** Ckartaft. .4 . m* >t* I?'orftUtwm ''N. lognlv A4i*ruu#fW H ***><* -uaire*) 0 i 151 158 139 164 135 152 160 14B 188 169 fMI S 384 174 199 646 6.1 18 8 19.1 138 14.6 396 G M lm*x MAXRH 36 72 60 0 36 75 40 9 44 75 61 3 37 68 87 9 40 61 47 5 39 69 39 41 78 40.9 41 S3 33 7 41 84 37 5 41 78 63 2 1778 Cox then clicking on B Bayesian network in the Causal Models area of the CAT ribbon will iun the [bnlearn] package and generate Fig. 1 (using ellipses rather than rectangles as the default shape for nodes). Fig. 2 can be generated either by selecting the columns in the order AllCause75, tmax, and then the remaining variables in any order and clicking on Sensitivity Plots in the CAT ribbon; or by entering the following R commands successively into any three Excel cells. R; library(randomForest) R: data <-- data.frame(year, month, day, All- Cause75. PM2.5, tmin. tmax. MAXRH) G: partialPlot(randomForest(data, AllCause 75), pred.data = data, x.var = "tmax") CAT uses the "G:" prefix to direct graphics output to the spreadsheet: if "R:" is used in stead. then it will appear in a separate win dow. The preceding instructions should suffice to al low replication of Table III and Figs. 1 and 2. For new analyses, the following CAT functions are use ful. (Many of these can also be accessed through thepoint-and-click ribbon features, as just described, and also through a Function Builder interface that allows users to select the name of the CAT function and then its arguments from drop-down lists.) CATjdescrihe{X) generates summary' statistics and plots (e.g. frequency distribution his tograms) for selected variable X. (The "Data Explorer ' feature of CA I allows such results to be viewed simply by passing the cursor over a column with data in it.) CAT-correlations^..and CAT'associations (...) will display graphs and tables showing various types of correlations and associations among the variables (e.g., AllCause75, tmin, tmax), generically indicated by "..." specified by the user. CAT-regression(Y, X, ...) will automatically select an appropriate regression model (linear if all variables are continuous, and also by de fault; logistic if T is 0-1, and Poisson if Y is a count variable), fit it to the data, and generate various tables and plots with Y as the dependent variable and A", ... (the names of one or more other variables) as independent variables. CATjree(Y, A, ...) will generate a classifica tion and regression tree for dependent variable Y (typically a health effect) and independent variables X, ... (typically an exposure variable and other covariates). CATJbnLearn(...) will display the structure of a Bayesian network, given a list of commaseparated variable names (e.g., AllCause75, tmin. tmax), generically indicated by "__ " to be included in the model. CATshow3d(Y. X. Z) generates a 3D scatter plot of response Y against variables X and Z and fits a smooth surface for the expected value of Y given X and Z. (Multiple surfaces for different values of a discrete fourth variable. W, can be generated with CAT_show3d(K X. Z. W}.) CA T.grangerTests(X, Y) will assess whether X is a Granger-cause of Y over a horizon (default is 7 time steps) that the user can specify, if Xand Y are both time series variables. Any of these functions can be invoked by typing the prefix R: followed by the CAT function in any cell of an Excel spreadsheet, once the CAT add-in is installed. CAT is intended to make advanced analytics readily available to Excel users, and it can be used to install new' R packages as desired and access them through the CAT interfaces (specifically Function Builder, which works with all R packages and func tions). New functions may be added to the CA F rib bon over time. REFERENCES 1. Shin Itlt. Cohen AJ. Pope CA .lid. Ezrali M Lim SS. Ilubhell BJ, Burnell RT Meta-analysis methods lo estimate the shape ami uncertainty in the association between long-term exposure to ambient fine particulate mallei and cause specific moitatity over the global concentration range Risk Analysis. 2016(9) (this issue) 2 Fann N. Gilmore t A. Walker K Characterizing ihe Iona trim PM concentration-response function: Comparing the strengths and weaknesses of research synthesis approaches. Risk Analy sis. 2015; 36{9):1693-17U7. 3. Smith AE. Inconsistencies in risk analyses for ambiem air pol lutant regulations. Risk Analysis. 2016: 36(9):1737-1744 4. Anenberg SC. Belova A. Brandi J. el at. Survey of ail pollution health risk assessment tools Risk Analysis. 2016 PV,(t)):l7|S_ 1736. 3 Dominki F. Greenstone VI. Sunstein CR Science and regula tion. Particulate matter matters Science. 2014: 344(6ISl't:257259 6. Franklin M. Zeka A. Schwartz J Association between PM2 5 and all-cause and specific-cause mortality in 27 I S communities. Journal of Exposure Science and Environmental Epidemiology. 2(XI7: 17(3)079-287 Invited C'niiiiiienlarv 1779 7, CDC (Centers for Disease Control and Prevention (US)). National Center for Chronic Disease Prevention and Health Promotion (US): Office on Smoking and Health (US). How tobacco Smoke Causes Disease; The Biologs and Behav ioral Basis for Smoking-Attributable Disease A Report ot the Surgeon General Atlanta. GA: Centers for Dis ease Control and Prevention (US). DO it). 7. Pulmonary Dis eases. Available at: http://www.nctn.nlnuiih.gflv/boaksfNBK 53021.. Accessed August 1.2016 8. Thomas DC. Jerrell M. kuen/h N Louis I X. Dominici P. Zcgcr S.SchwartzJ. Burnett RT. Krewski D. Bates D Bayesian model averaging in time series studies of air pollution and mortalitv. Journal of Toxicology and Environmental Health. 2IXr7: 70(3 41:3 II 315. 9. Cox LA Jr. Quantifying and reducing uncertainty about causal ity in improving public health and safety. In Graham R. Higdon D, Ovvhadi I I (eds) Handbook of Uncertainty Quantification. New York: Springer, 2(ll<> V ORIGIN ID AUSA (512)4/2-2/00 MICHELLE GROSCHE TEXAS PUBLIC POLICY FOUNDAMON 901 CONGRESS AVENUE AUSTIN, TX 78701 UNITED STATES US SHIP DATE. 28NOV17 ACTWGT 1.20 LB CAD 107341640/INET 3920 BILL SENDER to SCOTT PRUITT US EPA HEADQUARTERS MAIL CODE 1101A 1200 PENNSYLVANIA AVENUE NW WASHINGTON DC 20460 (512) 572-2700 WV PO REF NATIONAL AMBIENT DEPT FedEx Ship Manager - Print Your Label(s) of' 7708 5022 5582 XI RDVA WED - 29 NOV 8:00A FIRST OVERNIGHT ASR 20460 DC-US IAD A fte r printing th is label 11/28/2017 1