UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON D.C. 20460
THE ADMINISTRATOR
EPA-COUNCIL-AD V-04-001
Dr. Trudy Cameron and Dr. Bart Ostro
Science Advisory Board
Mail Code 1400F
1200 Pennsylvania Ave, NW
Washington, DC 20460
Dear Drs. Cameron and Ostro:
The Agency greatly appreciates the thoughtful and thorough advisory reports submitted
by the Council/HES in March and by the Council in June, pertaining to the analytic blueprint for
the second section 812 prospective study. The purpose of this letter is to request clarification of
the Council/HES and Council advice pertaining to a particular issue: the structure of cessation
lags associated with reductions in PM2.5 exposure.
The Agency recognizes that the cessation lag issue is particularly difficult due to the lack
of definitive scientific evidence supporting specification of any particular lag structure. In the
absence of a definitive, consensus lag structure, the Agency has continued to employ a 5-year
distributed lag for our base estimates, but we have also evaluated an alternative structure through
sensitivity analysis. The Agency intends to address the issue of cessation lag in the upcoming
full expert elicitation on PM mortality; however, this process will likely take at least one year.
In the interim, given that the Council has opined on the appropriate structure of the cessation lag,
the Agency needs to decide whether to continue using the current 5-year lag structure in our base
estimates or to employ the alternative structure, which was modeled after the HES-proposed 3-
segment lag approach. Therefore, the Agency requests that you consider providing clarification
of your advice by indicating whether this particular alternative lag structure is more scientifically
defensible than the 5-year lag structure that we have relied upon in the past and whether you
have any recommended modifications to that alternative.
In the remainder of this letter, we summarize the background information and relevant
review history of Council, SAB Arsenic Rule Benefits Review Panel (ARBRP), and NAS advice
on cessation lags and describe the alternative cessation lag structure. Attachments to this letter
provide the full text of 812 Council and Council HES, NAS, and ARBRP advice on cessation
lags excerpted from the relevant reports.
The Agency acknowledges that actual analytical application of one or more cessation lag
-------
structures will not be required for the section 812 study until late 2005. However, the Agency
has several PM2.5-related rulemaking analyses, which will be initiated or completed during the
coming months, and near-term clarification of the Council and Council/HES advice may
significantly improve these rulemaking analyses.
Background
In the previous section 812 study, and in recent rulemaking analyses, EPA has employed
a 5-year distributed lag structure to estimate the temporal path of incidence of premature
mortality associated with reductions in PM2.5. The lag structure assumes 25 percent of the
incidence reduction is manifest in each of the first two years, and the remaining 50 percent of the
incidence reduction is spread evenly over the succeeding three years.
This 5-year distributed lag structure was adopted in 1999 after the Agency posed a charge
question to the HEES regarding the use of a 15-year lag assumption based on a 1996 study by
the World Health Organization (i.e., the Brunekreef study), which implied that EPA should adopt
a 15-year lag assumption. Current Agency practice at the time was to assume no cessation lag
given the lack of empirical data supporting any specific quantitative lag structure. During the
subsequent HEES review, EPA presented three options for consideration: (1) zero lag, (2) 15-
year lag, and (3) the 5-year distributed lag, which had been incorporated in the Tier II
rulemaking analysis as a sensitivity test.
Initially, the HEES advised that the state of the relevant science was such that (a)
adoption of any particular lag structure would be arbitrary, (b) inclusion of pollutant-related time
lags in mortality at this time may therefore be premature, and (c) evaluation of the effects of
long-term downward trends in pollutant concentrations presents an important research
opportunity, [see Attachment A: HEES February 10, 1999 advisory report].
During their June 28-29, 1999 review meeting, the HEES panelists subsequently stated
that they preferred the 5-year distributed lag to the zero and 15-year options, both of which they
considered implausible. They further stated that the 5-year distributed lag was generally
consistent with available data on smoking cessation.
The HEES then provided an in-depth response in their October 1999 report which,
among other advice, stated that the HEES "...concurs with the [5-year distributed lag] approach
proposed by the EPA ... as the best estimate for use in the 1999 Section 812 report. HEES also
recommends that a sensitivity analysis of the time lag issue should also be presented in the
report. The sensitivity analysis should include a higher end and a lower end mortality estimate
(e.g., 0, 8, 15-year lags)." [see Attachment B: HEES October 29, 1999 advisory letter].
In August 2001, the SAB Arsenic Rule Benefits Review Panel (ARBRP) published their
review of the EPA Office of Water (OW) arsenic rule benefits analysis. We include excerpts
from their report as Attachment C, since their review included significant advice pertaining to
lag structures. In addition, the panel clarified the distinction between latency periods and
-2-
-------
"cessation lags", a term they coined, and pointed out that estimates for cessation lag and latency
may be significantly different.
In 2002, the NRC evaluated EPA's use of the five-year lag model in the context of air
pollution benefits analysis and stated that the NRC panel "found little justification for the 5-year
time course of exposure and outcome assumed ... and recommends that EPA more fully account
for the uncertainty regarding lags in health effects by incorporating a range of assumptions and
probabilities on the temporal relationship." [see Attachment D: NRC 2002 report]
In response to the NRC report, the Agency identified three alternative options in the
analytic blueprint for the second section 812 prospective study: (1) the currently employed 5-
year distributed lag, (2) an alternative based on a range of lag structures from 0 to 20-30 years,
and (3) construction of a 3-parameter Weibull distribution configured to match (undefined)
expected low, most likely, and expected high values. EPA then submitted a charge question to
the Council/HES requesting comment on these three approaches.
In the March 2004 advisory report, the Council/HES provided an in-depth assessment of
the cessation lag issue and the three options. The report echoes the earlier HEES and NAS
reports by noting, "Empirical evidence is lacking to inform the choice of the lag distribution
directly and agrees with the NAS report that there is little empirical justification for the 5-year
cessation lag structure used in the previous analyses." The Council / HES report then "urges the
Agency to begin to move from the relatively arbitrary assumptions of the 5-year lag structure to
an approach based on some plausible models of the disease processes involved", and states that
"[l]acking direct information from the cohort studies themselves, new insights regarding the
shape of the cessation lag can only come from improved understanding of the mechanism of the
exposure-response relationship." The Council / HES report discusses several potentially
relevant disease modeling approaches and suggests consideration of a 3-segmented lag structure
reflecting acute effect (0-6 month), medium-term effect (2-5 year), and long-term effect (15-25
year) patterns of exposure-response, using either a Weibull distribution or a simpler
distributional form with a smoother to mitigate discontinuities. Finally, the Council/HES report
recommends that cessation lags "be considered for inclusion in future expert elicitation efforts
and/or sensitivity analyses." [see Attachment E: March 2004 Council / HES advisory report]
In response to this cumulative advice, the Agency intends to address the cessation lag
question in the full expert elicitation on estimation of PM mortality. In the interim, our joint
collaboration with OMB on the nonroad diesel rule, led to the identification of an alternative lag
structure, which assumes 20 percent of the incidence reduction occurs in the first year of a
reduction in PM exposure, another 50 percent of the incidence reduction is evenly spread among
years 2 through 5 (i.e., 12.5 percent each year), and the remaining 30 percent of the incidence
reduction is evenly spread out among years 6 through 20 (i.e., 2 percent each year). This
alternative lag structure was evaluated as part of the sensitivity analysis for this rule.
-------
The central premise of the alternative lag structure is that estimates of the size of the
cessation lag should be based on our current understanding of the mechanism associated with
PM2.5 related-mortality and the empirical results of a variety of epidemiological and clinical
studies. As noted by the Council/HEES (EPA-SAB-COlMCIL-ADV-00-001, 1999), "some of
the mortality effects of cumulative exposures will occur over short periods of time in individuals
with compromised health status, but other effects are likely to occur among individuals who, at
baseline, have reasonably good health that will deteriorate because of continued exposure. No
animal models have yet been developed to quantify these cumulative effects, nor are there
epidemiologic studies bearing on this question." In its recently published fourth volume, the
NRC's Committee on Research Priorities for Airborne Particulate Matter (NRC 2004) concludes
that in addition to exacerbation of chronic respiratory disease, "results from epidemiological,
clinical, and animal studies are converging to indicate that PM exposures, both to PM2.5 and
ultrafine particles, have adverse cardiovascular effects." The Committee highlights clinical and
animal studies that suggest changes in heart rate variability, cardiac arrhythmias, ischemic
events, and congestive health failure should be considered among particle-related health
outcomes. Pope et. al. (2004) presents epidemiological evidence regarding cardiovascular
mortality and long-term exposure to air pollution to particles. Therefore, the distribution of
deaths over the latency period is intended to reflect the contribution of short term exposures in
the first year, cardiopulmonary deaths in the 2 to 5 year period, and longer term lung disease and
lung cancer in the 6 to 20 year period. The relative magnitudes of these segments are proposed
as interim estimates, to be used in benefits analyses until expert elicitation among those who
specialize in the natural course of disease can be completed, and/or whether modifications to that
alternative are recommended.
Request for Clarification of Council / HES Advice on Cessation Lag Structure
The Agency respectfully requests that the Council consider providing clarification of its
existing advice pertaining to cessation lags; specifically, whether the alternative lag structure
described above is more scientifically defensible than the five year lag structure the Agency has
used previously for its base estimates, and whether there are any modifications to this alternative
the Council would recommend.
Conclusion
EPA continues to place a very high value on the sound and thoughtful advice of the
Council and its technical subcommittees, and we appreciate your willingness to consider
providing additional elaboration and clarification of your advice so that we may continue to
improve our analyses.
-4-
-------
Sincerely,
/Signed/ /Signed/
Jeffrey Holmstead Jessica Furey
Assistant Administrator for Associate Administrator for Policy,
Air and Radiation Economics, and Innovation
cc: Vanessa Vu, Director, SAB Staff
Holly Stallworth, SAB Staff, Council DFO
5 attachments
-5-
-------
Attachment A
EPA-SAB-COUNCIL-ADV-99-005
Council / HEES February 1999 advisory report
February 10, 1999
[Cover letter from Cropper and Lioy, page 2:]
With regard to mortality time lags, the HEES agrees with the Agency that current studies on
animal mortality do not have an implied time lag, and the inclusion of pollutant-related time lags
in mortality at this time is premature.
[page 12:]
3.5.4 Modeling Time Lags for Cumulative Effects of Long-Term Exposure
FLEES agrees that consideration of time lags on annual mortality outcomes might be premature.
The current studies on animal mortality do not have an implied time lag, and selection of a value
for such a time lag would be arbitrary. The long-term downward trend in pollutant
concentrations, especially for PM, presents an important research opportunity for revisiting the
issue of time lags using already assembled data bases, and would be a good candidate for
sensitivity analysis. An effort of this nature, however, is most likely beyond the scope of the
current prospective study.
-6-
-------
Attachment B
EPA-SAB-COUNCIL-ADV-00-001
Council / HEES October 1999 advisory letter
October 29, 1999
[pages 8 to 10:]
15-Year Lag for Particulate Matter Effects
Charge question: "It has been suggested to the Agency that the WHO (1996) study provides
scientific evidence of the existence of a 15 year lag between changes in PM exposure and
changes in associated adverse health effects. Heretofore, however, the Agency has interpreted
the WHO authors' summing of incidences at the end of the 15 exposure period of the Dockery
study as a matter of mathematical convenience, not evidence of the WHO authors' belief in the
existence or magnitude of a lag between changes in exposure and changes in risk of adverse
health effect. What is the SAB HEES view regarding the proper interpretation and use of the
WHO (1996) study? Specifically, does the HEES believe it is reasonable to assume that, based
on the WHO (1996) study or other evidence, there is no reduction in risk of adverse health
consequences until 15 years following a reduction in PM exposure?
Response: Contrary to the June 17, 1999 letter from Arbuckle and Blank to Donald Barnes,2
there are no statements in the 1996 World Health Organization (WHO) report to suggest that
there is any scientific evidence for the existence of a 15-year lag between changes in PM
exposure and mortality.3 On page 35 of the WHO report (last paragraph, third line from
bottom), the authors state that "for simplification [emphasis added], it was assumed that the
effect of particulate matter only started to become manifest after 15 years in subjects who were
27.5 [years of age] initially ..." No citations from the published literature are given to support
the 15-year lag assumption, nor is the issue further discussed within the WHO report. Thus it is
clear that the authors of the WHO report used a 15-year lag assumption strictly "for
simplification," which can be interpreted as a convenient statistical device for estimating the
mortality effects from chronic exposure of the population to parti culate air pollution.
There is considerable evidence, cited in both the WHO report and EPA's 1995 Air Quality
Criteria Document for Particulate Matter,4 that daily variations in PM have an immediate effect
on mortality risk within a one to five day interval between elevated PM concentrations and
excess mortality. This effect was particularly apparent for cardiovascular and respiratory causes
of death among the elderly. These observations are commonly interpreted as implying that the
2
Donald R. Arbuckle and Rebecca M. Blank to Donald G. Barnes, June 18, 1999, Science Advisory
Board, HEES Meeting, 6/28&29/1999.
World Health Organization (WHO), "Final Consultation on Updating and Revision of the Air Quality
Guidelines for Europe." Bilothoven, The Netherlands, 28-31 October, 1996 ICP EHH 018 VD 96 2.11.
US Environmental Protection Agency, Air Quality Criteria for Particulate Matter, EPA/600/P-95/001aF-
CF.
-7-
-------
Attachment B
acute mortality effect of PM occurs among a particularly susceptible segment of the population
whose health status is already compromised by pre-existing disease. Thus with a reduction in
PM levels, it is reasonable to expect that there will be some immediate benefits from mortality
reductions among susceptible individuals.
However, the magnitude of estimated mortality effects from the cohort studies of Dockery et al.5
and Pope et al.6 are different than the estimates from the time-series studies. The WHO report
estimates a 10% mortality increase per 10|ig/m3 annual difference in PM from the cohort
studies, whereas the time-series studies show an overall 1-2% mortality increase per 10|ig/m3
daily variation in PM. The different estimates from the cohort studies, even when they are
adjusted for the differences in time duration, may be attributable to three consequences of PM
exposures: (1) cumulative PM exposures of the entire population may result in a PM-induced
increase in the number of individuals who become susceptible to the acute mortality effects
observed in the time series studies; (2) cumulative PM exposure may cause chronic diseases
which increase the mortality rate of the population, but the deaths of a portion of these
chronically ill persons may occur independently of the daily variations in PM exposure, and
these latter deaths are not captured by the time series studies; and (3) a 10|ig/m3 change in
annual average concentration may be associated with a much larger change in peak 24-hour
exposure levels.
Given that the mortality effect of cumulative air pollution exposure exceeds that of daily
variations in exposure, the question becomes, over what time period does the excess effect
manifest itself in the population? As noted above, some of the mortality effects of cumulative
exposures will occur over short periods of time in individuals with compromised health status,
but other effects are likely to occur among individuals who, at baseline, have reasonably good
health that will deteriorate because of continued exposure. No animal models have yet been
developed to quantify these cumulative effects, nor are there epidemiologic studies bearing on
this question. As the HEES previously stated, "consideration of time lags on annual mortality
outcomes might be premature".7 Neither the 1996 WHO report nor do any recently published
studies provide reasons to revise this statement.
Although there is substantial evidence that a portion of the mortality effect of PM is manifest
within a short period of time, i.e., less than one year, it can be argued that, if no a lag assumption
is made, the entire mortality excess observed in the cohort studies will be analyzed as immediate
effects, and this will result in an overestimate of the health benefits of improved air quality.
5 Dockery, D.W., C.A. Pope, X.P. Xu, J.D. Spengler, J.H. Ware, M.E. Fay, E.G. Ferris andF.E. Speizer,
""An association between air pollution and mortality in six U.S. cities," NEngl JMed, 329(24): 1753-1759.
6 Pope, C.A. Ill; Thun, MJ.Namboodiri, M; Dockery, D.W.; Evans, J.S.; Speizer, F.E., and Heath, C.W,
Jr. Paniculate Air Pollution is a Predictor of Mortality in a Prospective Study of U. S. Adults. Am. J. Respir. Care
Med., Vol. 151, March 1995, pp. 669-674.
"Clean Air Act Amendments (1990) Section 812 Prospective Study Health & Ecological Effects Initial
Studies," EPA-SAB-COUNCIL-ADV-99-005.
-------
Attachment B
Thus some time lag is appropriate for distributing the cumulative mortality effect of PM in the
population. The HEES concurs with the approach proposed by EPA at the June 29th meeting on
this issue, and recommends that the Tier 2 SA Lag estimates as presented at the meeting (Table
entitled "Sensitivity to Lag Assumption" Attached in Appendix A) be considered as the best
estimate for use in the 1999 Section 812 report. HEES also recommends that a sensitivity
analysis of the time lag issue should also be presented in the report. The sensitivity analysis
should include a higher end and a lower end mortality estimate (e.g., 0, 8, 15-year lags), in which
the higher end estimate would include a no-lag assumption, as given in the second column of the
above table, and the lower end estimate would replicate the analysis used in the 1996 WHO
report. The latter analysis has been published in the peer-reviewed literature.8 The Brunekreef
analysis clearly results in an underestimate of the immediate mortality effect of PM, since, as
discussed above, there is ample evidence for a short term mortality effect of PM, but the 15-year
lag analysis presented by Brunekreef provides a statistically simplified approach to estimating
the potential delayed effect of PM exposures for a young and relatively healthy segment of the
population.
o
Brunekreef B., "Air pollution and life expectancy: is there a relation?" Occupational Environmental
Medicine, 1997; 54:781-4.
-9-
-------
Attachment C
EPA-SAB-EC-01-008
SAB Arsenic Rule Benefits Panel (ARBRP)
Arsenic Rule Benefits Analysis: An SAB Review
August 2001
[pages 3 to 7:]
2. RESPONSE TO THE CHARGE QUESTIONS
2.1 The Impact of the Timing of Exposure on Avoided Cancers
Charge Question 1: How should latency be addressed in the benefits estimates when
existing literature does not provide specific quantitative estimates of latency periods
associated with exposure to arsenic in drinking water?
2.1.1. Introduction
A central component in analyzing the benefits of reduced exposure to a carcinogen is the
prediction of the annual reduction in cancer cases following reduction in exposure. If a
population previously exposed to 50 g/L of arsenic in drinking water is exposed, beginning in
2006, to only 10 g/L, cancer risks in the population will eventually decline to a steady-state
level associated with a lifetime of exposure to 10 g/L. How fast this reduction in risk occurs
depends on the cessation-lag following reduction in exposure. We believe that this is more
appropriately termed a "cessation-lag," rather than "latency." This distinction is clarified below.
In order to explain what should be done when the length of this cessation-lag is unknown,
we must describe how the timing of the relationship between exposure and response (death due
to cancer) should be treated in a benefits analysis. We emphasize that we believe that this is how
such an analysis is conducted; it does not refer to the approach taken in the arsenic benefits
analysis. As in the case of arsenic, we analyze a policy that would reduce exposure from a
current level of dO (e.g., 50 g/L) to dN (e.g., 10 g/L). We assume that this policy would
continue into the indefinite future.
For a benefits analysis we would like to:
a) Calculate the expected number of cancer fatalities avoided each year, as a result
of the policy, beginning with the year in which the policy is implemented and continuing
into the future.
If benefits are to be monetized in accordance with conventional economic practice:
b) The expected number of cancer fatalities avoided each year should be multiplied by
the value of a statistical life in that year. This will give the dollar value of benefits each
-10-
-------
Attachment C
year, beginning with the year in which the policy in implemented.
The dollar value of benefits in each year should be discounted to the year in which the policy is
implemented and summed. The present discounted value of benefits, so calculated, should be
compared with the present discounted value of costs, calculated over the same period.
The timing of the relationship between exposure and cancer mortality is implicit in the
calculations in (a). As described more fully below, if the lag between reduction in exposure and
reduction in risk of death is long, there will be fewer cancer fatalities avoided in years
immediately following the policy than if the lag were shorter. Uncertainties in the timing of the
exposure-response relationship will be reflected in uncertainties in the number of cancer
fatalities reduced each year after the policy is implemented. These uncertainties should be treated
as described in the answer to Charge Question 5.
2.1.2 Calculation of Reduced Cancer Fatalities Associated with Reduced Exposure to a
Carcinogen
The approach taken here is to relate the age-adjusted risk of death due to cancer to the
history of exposure to the carcinogen. This relationship, together with information on the age
distribution of the population affected by the policy, can be used to calculate the expected
number of cancer fatalities avoided by the policy.
The epidemiology underlying the arsenic benefits analysis (Morales et al. 2000) assumes that
the conditional probability of dying from cancer at age t, h(t) is related to cumulative exposure to
a carcinogen as of age t, xt, by a proportional hazard model:
(1) h(t,x) = h0(t)g(xt)
where ho (t) = baseline risk of dying from cancer at age t (assuming survival to age t) and g(xt)
represents the impact of exposure incurred up to age t on risk of death.2
2.1.2.1 The Timing of the Exposure-Response Relationship
The key question is how cumulative exposure (xt) depends on the dose of arsenic
received at ages 0 through t. Let d; = dose received at age i. A general form that this relationship
could take is:3
2 A proportional hazard model (Pope et al. 1995) is also used to measure the association between particulate matter
and all-cause mortality in The Benefits and Costs of the Clean Air Act 1970-1990 (USEPA 1997) and The Benefits
and Costs of the Clean Air Act 1990-2010 (USEPA 1999). The issue of the length of the cessation-lag after a
reduction in exposure also arises in these studies.
3 The function ft () could also be conditioned on other factors such as smoking.
-11-
-------
Attachment C
(2) xt = ft(do,di,-,dt)
The exact form of this function reflects the answers to the following four questions (Tollerud et
al. 1999):
(a) How long does it take after an exposure until an increase in risk is observed?
(b) How long does the effect of an exposure last after exposure has terminated?
(c) How does the effect of exposure vary by the age at which it was received?
(d) Does the exposure act at an early or late stage in the carcinogenic process?
The relevant questions for the implementation of changes in the drinking water standard
for arsenic are questions (b)-(d). In contrast, most of the epidemiologic literature addressing the
issue of latency has focused on question (a), which is the usual definition of latency. The
committee wishes to underscore that data addressing question (a) do not necessarily provide
information answering questions (b)-(d). Unfortunately, much less work has been done to
evaluate questions (b)-(d) in the epidemiologic literature in general, and in the research on
arsenic carcinogenicity in particular.
The NAS report Veterans and Agent Orange: Update 1998 (Tollerud et al. 1999)
addresses the second question, regarding how long effects last after cessation of exposure. With
respect to arsenic in drinking water, the charge of our committee is an expansion of this
question: when does the excess risk (compared to a lifetime of exposure to dN (e.g., 10 g/L))
begin to attenuate and how long does it take until all of the excess is eliminated? Since the term
latency has a traditional usage that is not the charge given to this committee, we have coined the
phrase "cessation-lag" to clarify and emphasize the difference.
An important point is that the time to benefits from reducing arsenic in drinking water
may not equal the estimated time since first exposure to an adverse effect. A good example is
cigarette smoking: the latency between initiation of exposure and an increase in lung cancer risk
is approximately 20 years. However, after cessation of exposure, risk for lung cancer begins to
decline rather quickly. A benefits analysis of smoking cessation programs based on the observed
latency would greatly underestimate the actual benefits. We return to the issue of how to
estimate the length of the cessation-lag below.
2.1.2.2 Calculating the Time Path of Cancer Cases Avoided
If the relationships in (1) and (2) are known, it is, in principle, a simple matter to compute
the expected number of cancer fatalities avoided at age t (and, by analogy, for all other ages) in
each year following the policy. In the first year of the policy it is only the most recent dose of the
carcinogen (dt for persons who are age t in the year the policy is implemented) that is affected by
the policy. The expected reduction in risk of death due to cancer at age t in the first year of the
policy is:
(3) h0 (t)[g(ft (do0,di°,..,dt0)) - g(ft (do°,di°,..,dt'))]
-12-
-------
Attachment C
where the superscripts 0 and N refer to doses with and without the policy, respectively. In the
second year of the policy, for persons of age t, both dt-1 and dt are affected by the policy, and
the formula in (3) would be adjusted accordingly. Eventually, a steady-state will be reached in
which persons of age t face the same mortality risk from cancer as people who have been
exposed to the lower level of the carcinogen (dN) throughout their lifetime.
In each year, the number of fatalities avoided by the policy among persons of age t would
be the expression similar to (3) multiplied by the number of persons of age t. Similar
computations would be performed for persons of all ages. In this manner, it should be possible to
compute the expected number of fatalities avoided, by age (or age-group), in each year following
the implementation of the policy. Because the age distribution of avoided cancer fatalities is
calculated, it should be reported in a benefits analysis even if information on the age distribution
of avoided fatalities is not used in valuing avoided mortality.
2.1.3 Quantifying the Relationship Between Exposure and Mortality Risk
Most epidemiologic studies ignore the time pattern of exposure in estimating the
proportional hazard model in equation (1). For example, Morales et al. (2000) effectively assume
that
t
(4) xt= di.
i=0
Given sufficient data, the time pattern of exposure and effect can be estimated in the
context of equations (1) and (2).4 In order to properly study effects of protracted exposures,
detailed exposure histories for each study subject, including the dates and ages when the
individual was exposed and the level of exposure at all points in time, are needed. Appropriate
statistical methods have been developed for an investigation of the effect of exposure accrued as
a function of time since that exposure (Thomas 1983; Breslow and Day 1987; Thomas 1988). In
general, the ability to investigate the issues of timing of exposure in a given data set will depend
on the quality of the exposure measure, the quality of the timing of exposure information, the
number of people developing the disease of interest, and variation of exposure over time within
the study group. These aspects of study quality are, of course, important in evaluating any
epidemiologic investigation. But there are special problems that arise in the evaluation of time
related factors (Enterline and Henderson 1973; Thomas 1987).
If possible, it would be desirable to use information about the mechanism by which
cancer occurs in estimating the length of the cessation-lag.5 For example, if arsenic primarily
4 Latencies and cessation-lags would be expected to vary by cancer site, would probably be shorter for
cardiovascular disease than for cancer, and may be shortest for reproductive effects. We emphasize that the same
model should be used to estimate the time pattern of exposure and response as is used to estimate the potency of the
carcinogen.
5 We emphasize that the same model should be used to estimate the time pattern of exposure and response as is
used to estimate the potency of the carcinogen.
-13-
-------
Attachment C
exerts a late-stage effect in the cancer formation process, the cessation-lag will be shorter than if
arsenic primarily exerts an early-stage effect. Appendix 2.1 to this report discusses how the time
pattern of exposure and response could be estimated in the context of the multi-stage model of
cancer formation.
In addition, two published studies have attempted to address either latency or cessation
lag, or the stage at which arsenic acts in the carcinogenic pathway. Brown and Chu (1983, 1987)
attempted an analysis based on one of the arsenic-exposed occupational cohorts and
demonstrated that two models provided good fit to the data: one with only a late-stage effect and
the other with both an early- and late-stage effect. There was a slightly better fit for the model
with only a late-stage effect but the difference in fit was not sufficient to exclude an early-stage
effect. A more recent analysis (Hazelton et al. 2000) examined an occupational cohort with
exposures to arsenic, radon and tobacco using biologically based models. They evaluated the
time between generation of the first malignant cell and death from lung cancer. This would
appear to assume an early-stage effect only; nevertheless, it is notable that the best fit was given
for a gamma distribution of lags that had a mean of 4.1 years and a variance of 2.9 years. Under
this distribution, which is consistent with a minimal first stage effect of arsenic, the bulk of the
benefit following cessation would be expected to occur within the first five years after exposure
is reduced.
It thus appears that some information about the length of the cessation-lag is available in
the case of arsenic. Additional information on the length of the cessation-lag could be evaluated
from data on arsenic-exposed populations in Taiwan and Chile, and we urge that such research
be undertaken. In Taiwan, the water supply was changed in the early 1970's, thereby eliminating
the arsenic exposure. In Antofagasta, Chile, water treatment beginning in 1970 reduced the
arsenic concentration from 800 to 110 g/L within a short time, and over a few more years to
40-50 g/L.
If, however, such information were not available (as the charge question assumes), what
could be done? One extreme assumption that would yield an upper bound to the benefits of the
program is to assume that the program immediately attains the steady-state result, i.e., that the
reduction in the age-t mortality rate is given by:
(5) h0 (t)[g(ft (do°,di°,.,dt0)) - g(ft (do'.di1,...,^))]
This is the assumption made in the Agency's primary analysis.
If it should prove infeasible to estimate the cessation-lag and account for it as described
above, it would still be desirable to examine the influence of this lag by performing sensitivity
analyses similar to those carried out for the PM-mortality relationship in the Agency's analysis
of The Benefits and Costs of the Clean Air Act: 1990-2010 (USEPA 1999). In the context of the
multi-stage model described in Appendix 2.1, we would suggest that the testing of extreme cases
of potential mechanisms (i.e., arsenic's effects being exerted entirely at an early stage v. all at a
late stage) be done as part of the uncertainty analysis.
-14-
-------
Attachment D
NRC Report: Estimating The Public Health Benefits Of Proposed Air Pollution Regulations
Committee on Estimating the Health-Risk-Reduction Benefits of Proposed Air Pollution
Regulations
National Research Council of the National Academies
2002
[page 64:]
Finally, the health benefits of reducing emissions in a single year might not occur solely in that
year but might occur in subsequent years because of physiological and other lags. The analyses
should carefully state and document the lag relationships between pollution reductions and
health improvements that have been used (see Chapter 4).
[pages 114 to 115:]
Effect Lags and EPA's Assumptions
Understanding long-term disease processes is important for benefits analysis. For example,
certain health benefits resulting from a change in air quality may occur only after several years.
Although it appears that mortality following short-term exposure to PM occurs within a
relatively short time, little is known about the temporal relationship between longer-term
exposure and mortality as demonstrated in the prospective cohort studies. For example, the ACS
study (Pope et al. 1995) provided little information as to whether the observed geographic
differences in mortality risks are due to a 1-year average or some multiyear history of PM
exposures preceding mortality. Thus, it is not known which period of exposure is the most
important and how quickly benefits from air pollution reductions will appear in the case of long-
term disease processes. In the Swedish lung cancer study (Nyberg et al. 2000), effects were
strongest for the exposure 20-30 years ago. For other outcomes, other time periods may be
relevant.
The time course relating exposure to outcome is an important assumption in benefits analysis,
especially when long-term mortality effects dominate the analysis, as occurs in PM analyses. It
is important because health benefits that occur far into the future may count less based on the
way the benefits are monetized. In EPA's benefits analyses for the Tier 2 rule and the HD engine
and diesel-fuel rule, EPA assumed a weighted 5-year time course of benefits in which 25% of the
PM-related mortality benefits were assumed to occur in the first and second year, and 16.7%
were assumed to occur in each of the remaining 3 years. Although recommended by EPA's
Science Advisory Board, the committee found little justification for a 5-year time course and
recommends that future benefits analyses more fully account for the uncertainty regarding lags in
health effects by incorporating a range of assumptions and probabilities on the temporal
relationship.
-15-
-------
Attachment D
[page 118:]
The committee found little justification for the 5-year time course of exposure and
outcome assumed in the more recent EPA analyses and recommends that EPA more fully
account for the uncertainty regarding lags in health effects by incorporating a range of
assumptions and probabilities on the temporal relationship.
[pages 136-137:]
EPA expressed much less certainty about alternative lag structures than it did about thresholds in
the Tier 2 analysis. The lag structure used in the primary analysis was recommended by the
Science Advisory Board (EPA 1999a, pp. 4-6, 4-7), but the agency considered a range of
alternative lag structures plausible. Here a probabilistic weighting of alternative lag structures
based on expert judgment might have led to a more appreciable widening of the health benefit
probability distribution.
Although EPA considered alternative lag structures to vary in plausibility, these variations were
not, but could have been, approximately captured by subjective probability distributions. The
incorporation of these distributions into the final probability distribution for the primary analysis
would have resulted in a more realistic presentation of acknowledged sources of uncertainty.
-16-
-------
Attachment E
EPA-SAB-COUNCIL-ADV-04-002
Council / HES March 2004 advisory report
March 2004
[Executive Summary, page 2:]
The HES also provides advice on how to address the question of cessation lag, which is
the time lag between reductions in concentrations of air pollutants and manifestation of health
benefits in the population. The HES notes that for long-term PM effects, empirical evidence is
lacking to estimate the lags. Given this problem, the HES recommends that the Agency consider
developing models for each cause of death category expected to make up PM mortality, since the
lag structure most likely differs for different PM-associated disease processes. Although specific
causes of death would not be specifically calculated in the base case, the literature provides
enough information to guide estimates of the likely proportion of PM mortality by disease type
(Pope et al., 2002, 2004).
[pages 22 to 24:]
3.6. Agency Charge Question 16: Cessation Lag.
Charge Question 16. In recent EPA rulemakings, EPA's "base estimate" of benefit from
PM control has been based on cohort epidemiological studies that characterize the chronic
effects of pollution exposure on premature death as well as capturing a fraction of acute
premature mortality effects. If these chronic effects occur only after repeated, long-term
exposures, there could be a substantial latency period and associated cessation lag. As such, a
proper benefits analysis must consider any time delay between reductions in exposure and
reductions in mortality rates. For the acute effects, such as those considered in EPA's alternative
benefit analyses, the delays between elevated exposure and death are short (less than two
months), and thus time-preference adjustments are not necessary.
(1) In the previous 812 analysis and in recent rulemakings, EPA assumed a
weighted 5-year time course of benefits in which 25% of the PM-related mortality
benefits were assumed to occur in the first and second year, and 16.7% were
assumed to occur in each of the remaining 3 years. Although this procedure was
endorsed by SAB, the recent NAS report (2002) found "little justification" for a
5-year time course and recommended that a range of assumptions be made with
associated probabilities for their plausibility. Do you agree with the NAS report
that EPA should no longer use the deterministic, 5-year time course?
(2) One alternative EPA is considering is to use a range of lag structures from
0 to 20-30 years, with the latter mentioned by NAS in reference to the Nyberg et
al. PM lung cancer study, with 10 or 15 years selected as the mid-point value until
more definitive information becomes available. If this simple approach is used,
should it be applied to the entire mortality association characterized in the cohort
-17-
-------
Attachment E
studies, or only to the difference between the larger mortality effect characterized
in the cohort studies and the somewhat smaller effect found in the time series
studies of acute exposure? Should judgmental probabilities be applied to different
lags, as suggested by NAS?
(3) Another option under consideration is to construct a 3-parameter Weibull
probability distribution for the population mean duration of the PM mortality
cessation lag. The Weibull distribution is commonly used to represent
probabilities based on expert judgment, with the 3-parameter version allowing the
shaping of the probability density function to match expected low, most likely,
and expected high values. EPA is still considering appropriate values for the low,
most likely, and expected high values -and therefore for the Weibull shape and
location parameters- and EPA is interested in any advice the Council wishes to
provide pertaining to the merits of this approach and/or reasonable values for the
probability distribution.
HES Response: Given the purpose of the 812 Studies (to estimate a future situation), the
cessation lag is a very important issue. As noted by EPA, for short-term effects (including time-
series based observations of mortality) this is not a problem, and there is even published
evidence that these short-term effects closely follow changes in the pollution, thus, benefits are
'immediate' (on the annual aggregate level). For long-term effects, the HES notes that empirical
evidence is lacking to inform the choice of the lag distribution directly and agrees with the NAS
report that there is little empirical justification for the 5-year cessation lag structure used in the
previous analyses. This is because the cohort mortality studies reported to-date have lacked data
on the long-term time-course of exposures for each cohort member; such data, if available, might
enable testing hypotheses regarding alternative exposure lag structures, if sufficient statistical
power was available. However, the HES notes the importance of developing some estimates of
the cessation lag rather than assuming there is no lag and urges the Agency begin to move from
the relatively arbitrary assumptions of the 5-year lag structure to an approach based on some
plausible models of the disease processes involved. Lacking direct information from the cohort
studies themselves, new insights regarding the shape of the cessation lag can only come from
improved understanding of the mechanism of the exposure-response relationship. Information
that may prove valuable in this regard could include results from clinical, experimental animal,
and in-vitro studies, and analogies with the health effects of other long-term inhalation
exposures, such as cigarette smoking. The clinical intervention literature (e.g., cardiovascular
trials) or smoking cessation data may be useful.
The HES recommends that the Agency consider developing models for each cause of
death category expected to make up PM mortality, since the lag structure most likely differs for
different PM-associated disease processes. Although specific causes of death would not be
specifically calculated in the base case, the literature provides enough information to guide
estimates of the likely proportion of PM mortality by disease type (Pope et al., 2002, 2004).
As a general rule, one may assume that the longer the air-pollution-sustained disease process
is, the longer the delay. This may be true whether pollution is an initiator or a promoter. For
example, if inhalation of carcinogens from ambient air contributes to the incidence of lung
-18-
-------
Attachment E
cancer, the pathophysiologic process between exposure and death may take many years (for the
average case) and the benefit of a reduction in carcinogenic constituents in PM between the
year 2000 and the year 2010 may lead to a reduction in lung cancer rates only after many years.
For effects of long-term PM exposures on pulmonary disease (e.g., COPD), a useful model
may be the change in the natural history of lung function with exposure to air pollution.
Several studies show effects of long-term PM exposures on decreased lung function (e.g.,
Gauderman et al., 2002)). By analogy with cigarette smoking, this may put people on steeper
trajectories of lung function decline, which is a known risk factor for premature mortality.
This might imply distributed lags extending over a substantial fraction of a lifetime. On the
other extreme, some cardiovascular deaths captured in the cohort studies may be due to air
pollution during the last months to years prior to death whereas the underlying susceptibility to
a cardiovascular death may be due to non-air pollution causes (e.g., diabetes). Lifetime lost,
captured in the cohort, may still be rather long (see comments in response to Charge Question
17). Clean air policies would bring a rather immediate benefit for such kind of cases. For
example, Lightwood and Glantz (1997) conducted a meta-analysis of studies to determine how
excess risks of myocardial infarction and stroke in smokers decline after quitting. They
reported that risks would be reduced after roughly 1.5 years. Finally, to the extent that cohort
results capture a portion of the acute time-series mortality effects of PM, there may be an even
shorter lag.
EPA staff has presented several alternative lag structures, including the use of a flexible
Weibull distribution spanning up to 25 years. It would be useful to utilize a distribution that
could incorporate time lag to benefits based on different patterns of exposure-response
consistent with models developed of the various response mechanisms. For example, acute
effects may be reduced within the first 6 months of an exposure change, medium-term effects
may be reduced within 2 to 5 years, and long-term effects may be reduced after 15 to 25 years.
Thus, the HES supports either the use of a Weibull distribution or a simpler distributional form
made up of several segments to cover the response mechanisms outlined above, given our lack
of knowledge on the specific form of the distributions. An important question to be resolved is
what the relative magnitudes of these segments should be, and how many of the acute effects
are assumed to be included in the cohort effect estimate. The Subcommittee suggests that a
smoother might be applied to the lag function to smooth the discontinuities. Given the current
lack of direct data upon which to specify the lag function, the FIES recommends that this
question be considered for inclusion in future expert elicitation efforts and/or sensitivity
analyses. As noted, time lag to benefits may depend on the cause of death and the underlying
morbidity processes that ultimately lead to premature death.
[pages 28 to 29 :]
If the Agency adopts the approach discussed in response to Charge Question 16 of
modeling the exposure-response processes to estimate the range of cessation lags, a similar
approach could be used to estimate life-years saved. Just as likely ranges of cessation lags may
be estimated by looking at what is known about different causes of death and how PM may be
contributing to the disease processes and attempting to build some models/ranges of that
-19-
-------
Attachment E
process, ranges of life-years lost could be similarly estimated. Whether the Agency uses a
static or dynamic life table, the assumption made in the life tables approach is that the average
remaining disease-specific life expectancy for the people whose deaths are predicated on air
pollution exposure is the same as the average remaining life expectancy for all individuals (i.e.,
where deaths are both related and non-related to air pollution) of the same age and gender. This
may result in an overestimate of life-years saved due to PM reductions if the disease profile of
the subgroup impacted by air pollution is different from the profile of the full group (i.e., if the
air pollution-impacted people with previous cardiovascular disease are more frail than people
who die from cardiovascular disease, in general). It would be reasonable to assume, consistent
with the cessation lag estimates, that some share of the deaths are among people with lower
than average life expectancy. The Agency could use available information on causes of death
and likely disease processes to propose a set of reasonable assumptions for both cessation lags
and life-years saved that are consistent with one another. For example, some share of the
COPD deaths associated with PM exposure consists of individuals who developed COPD
because of long-term PM exposure. In this instance the cessation lag may be many years and
the life-years lost are consistent with standard life tables. In another category, there may be
heart attack deaths associated with PM exposure that include individuals who had already
existing coronary heart disease. In this case the cessation lag may be quite short and the life-
years saved, although substantial, may be less than the standard life table's calculation because
of the pre-existing disease. In yet another category, there may be PM-related deaths due to
pneumonia in individuals with rates of pre-existing disease comparable to the general
population. If in the absence of PM exposure a full recovery would have been made, then the
cessation lag is quite short and the life-years saved is consistent with the standard life tables.
The HES acknowledges, however, that uncertainties remain, given that no study has
formally analyzed the years of life lost and the dependence of years of life lost on causes of
death, pre-existing diseases, and the underlying distributions of other susceptibilities. Even
though a considerable amount of judgment would be involved, an approach that uses available
information to estimate the shares of PM-associated deaths in each of several categories may
provide a more defensible set of assumptions for estimating both cessation lags and life-years
saved than more arbitrary assumptions.
###
-20-
------- |