Environmental Protection Agency: Use of Precautionary Assumptions in
Health Risk Assessments and Benefits Estimates (Letter Report,
10/16/2000, GAO/GAO-01-55).

Some of the Environmental Protection Agency's (EPA) regulations set
standards that limit environmental contaminants to levels that are
determined, in large part, on the basis of the health risks they pose.
However, when EPA assesses the health risks of contaminants, the agency
is faced with uncertainties and gaps in scientific knowledge and data.
This report summarizes GAO's findings on whether EPA's benefits
estimates for major environmental regulations that establish
health-based standards reflect precautionary assumptions about health
risks. Three key factors influence EPA's use of precautionary
assumptions in assessing health risks. First, EPA is influenced by its
mission to protect human health and safeguard the natural environment.
Second, EPA is influenced by the nature and extent of relevant data.
Finally, EPA is influenced by the nature of the health risk being
evaluated.

--------------------------- Indexing Terms -----------------------------

 REPORTNUM:  GAO-01-55
     TITLE:  Environmental Protection Agency: Use of Precautionary
	     Assumptions in Health Risk Assessments and Benefits
	     Estimates
      DATE:  10/16/2000
   SUBJECT:  Safety standards
	     Environmental policies
	     Pollution control
	     Environmental law
	     Cost effectiveness analysis
	     Health hazards

******************************************************************
** This file contains an ASCII representation of the text of a  **
** GAO Testimony.                                               **
**                                                              **
** No attempt has been made to display graphic images, although **
** figure captions are reproduced.  Tables are included, but    **
** may not resemble those in the printed version.               **
**                                                              **
** Please see the PDF (Portable Document Format) file, when     **
** available, for a complete electronic file of the printed     **
** document's contents.                                         **
**                                                              **
******************************************************************
GAO-01-55

Report to Congressional Requesters

October 2000 ENVIRONMENTAL PROTECTION AGENCY

Use of Precautionary Assumptions in Health Risk Assessments and Benefits
Estimates

GAO- 01- 55

Letter 3 Appendixes Appendix I: Objectives, Scope, and Methodology 34

Appendix II: Dose- Response Relationships Based on Varying Assumptions 37

Appendix III: Summary of EPA's 1997 Particulate Matter Standards 38 Appendix
IV: Issues Associated With EPA's Setting of the 1997

Particulate Matter Standards 39 Tables Table 1: 1997 Particulate Matter
Standards 38 Figures Figure 1: Typical Sequence of Risk Assessment and Risk
Management Processes 8

Figure 2: EPA's Standard (Default) Dose- Response Relationship for
Carcinogens 16 Figure 3: Dose- Response Relationships Based on Varying

Assumptions 37

Abbreviations

EPA Environmental Protection Agency

Lett er

October 16, 2000 The Honorable Bud Shuster Chairman Committee on
Transportation

and Infrastructure House of Representatives

The Honorable Frank R. Lautenberg United States Senate

Some of the Environmental Protection Agency's (EPA) regulations set
standards that limit environmental contaminants 1 to levels that are
determined, in large part, on the basis of the health risks they pose. For
example, EPA sets health- based air quality standards under the Clean Air
Act. For such actions, EPA also estimates the benefits of the health- based
standards. 2 These benefits primarily represent the estimated dollar value
of reductions in assessed risks to human health- illnesses and deaths
avoided as a result of decreased pollution. 3 However, when EPA assesses the
health risks of contaminants, the agency is faced with uncertainties and
gaps in scientific knowledge and data. As a result, EPA's risk assessments
include assumptions about the relationship between specific contaminants and

health effects, some of which are precautionary- that is, they are intended
to ensure that the agency does not underestimate health risks. But using
such precautionary assumptions to estimate benefits could produce overly
optimistic estimates of the benefits of regulatory actions.

Because of concerns about the potential impact of precautionary assumptions
on benefits estimates, you asked us to examine whether EPA's benefits
estimates for major environmental regulations that establish health- based
standards reflect precautionary assumptions about health risks. As agreed
with your offices, this report identifies (1) key factors that 1 In this
report, we use the term contaminants to refer to substances that harm human
health.

EPA uses this and other terms, including toxins, toxic substances, and
pollutants, to describe these substances.

2 Analyses of benefits and costs are required for all regulations that are
“economically significant” and includes those expected to have
an annual impact on the economy of $100 million or more. These are also
referred to as major rules or major regulations. 3 Using the concept of the
“value of a statistical life,” economists have developed several
methods to estimate a value of reductions in mortality risk.

explain why EPA uses precautionary assumptions in assessing health risks and
(2) whether EPA used and identified precautionary assumptions in estimating
the health risks and benefits of recent major regulations setting

health- based standards. In recent years, EPA has proposed or finalized a
small number of major regulations establishing health- based standards, and
these have been issued under two statutes- the Clean Air Act and the Safe
Drinking Water Act. 4 To illustrate EPA's use of precautionary assumptions
in setting healthbased standards and estimating their benefits, we reviewed
two of these regulations: (1) air quality standards for particulate matter
(commonly called soot) and (2) drinking water standards for arsenic. As a
result, our

findings on these two rules may not be generalized to all of EPA's major
regulations that set health- based standards. Consistent with our
objectives, we did not review the economic aspects of EPA's benefits
estimates, such as the methods used to estimate dollar values for lives
saved and illnesses avoided, or the extent of EPA's compliance with guidance
from the Office of Management and Budget and EPA on the preparation of
benefits estimates. (See app. I for a detailed discussion of our scope and
methodology.) We are also reviewing the health risk assessment procedures
and assumptions of four agencies, including EPA, and plan to provide the
results of this review

in a forthcoming report to the Chairman, House Committee on Commerce.
Results in Brief Three key factors influence EPA's use of precautionary
assumptions in

assessing health risks. First, EPA is influenced by its mission to protect
human health and safeguard the natural environment. For example, in some
instances, environmental statutes require EPA to protect the public health
with an “adequate margin of safety.” Second, EPA is influenced
by the nature and extent of relevant data- in particular, whether studies of
a contaminant's effects on people are available or whether the agency must
extrapolate from studies using other animal species. Finally, EPA is
influenced by the nature of the health risk being evaluated, such as whether

the contaminant is suspected of causing cancer. The two regulations that we
examined differed in the extent to which they used precautionary assumptions
in estimating health risks and benefits.

4 According to EPA, the majority of its major regulations in recent years
have established technology- based or performance- based standards that
reduce pollution using available pollution control technology.

EPA generally did not rely on precautionary assumptions in estimating either
the health risks or the benefits of the particulate matter regulation
because the agency had better data than is often the case. In contrast,
because of scientific uncertainties and data gaps, EPA relied on several
precautionary assumptions in estimating the health risks and the benefits of
the arsenic regulation. As a result, EPA's estimates of the risk of bladder
cancer associated with arsenic in drinking water and the related benefits of

the proposed rule may be overstated for this health risk. However, EPA used
new analytical techniques in estimating the health risks and benefits that
removed certain precautionary assumptions used in the past, thereby reducing
the extent to which the benefits of reductions in bladder cancer

may be overstated. In its proposed arsenic regulation, EPA identified the
key health uncertainties and precautionary assumptions it used in assessing
the risks of arsenic in drinking water but was less complete in identifying
them in its formal cost- benefit analysis. Furthermore, although its
guidance on cost- benefit analyses calls for assessing uncertainty using
sensitivity analysis, EPA did not perform sensitivity analysis that could

have shown how the estimated benefits would change depending upon the health
assumptions used. We are recommending that, in developing its final rule on
arsenic, EPA fully disclose and analyze the impact of the key precautionary
health assumptions used in its benefits estimate.

Background EPA assesses human health risks in the context of great
uncertainty- both in terms of scientific theory and data- about the adverse
effects on human

health posed by a wide variety of environmental contaminants, the
relationship between toxicity and dose, and the extent of people's exposure
to contaminants. As a result, EPA must use assumptions, based on general
scientific knowledge and policy judgments, when it lacks more specific
scientific knowledge and data needed to assess the health risks posed by
contaminants. The National Research Council of the National Academy of

Sciences 5 has noted that risk assessment inevitably includes policy
judgments as well as science. 6

EPA has issued, over the past 25 years, a series of guidance documents that
describe the principles, policies, and practices the agency employs in
evaluating the health risks (toxicity) of environmental contaminants. EPA's
guidelines cover many topics, including cancer, reproductive and
developmental toxicity, neurotoxicity, exposure assessment, and
mutagenicity, that is, the capacity to cause sudden change in the genetic
material of a cell. EPA personnel who conduct risk assessments are to use
EPA's guidelines to ensure consistency in the interpretation of scientific
information across all of the agency's programs and regions.

EPA's risk assessment guidelines set forth “default”
assumptions- generic approaches based on general scientific knowledge and
policy judgment that are applied to various elements of the risk assessment
process when specific scientific information is not available. In this
report, we refer to the default assumptions that are intended to avoid
underestimating risk as precautionary assumptions. There is an ongoing
scientific and policy debate concerning whether and under what circumstances
some of the assumptions used in estimating human health risks are
precautionary. According to some analysts, using a series of protective
default assumptions within the same risk assessment might produce results
that seriously overstate actual risks. Other analysts believe that the
degree of precaution may not be as great as some argue- or that the risk
assessment 5 The National Academy of Sciences, a private, nonprofit
organization composed of scholars, is engaged in scientific and engineering
research to further knowledge and advise the

federal government. The National Research Council, the principal operating
agency for the National Academy, provides services to the government, the
public, and the scientific and engineering communities.

6 See Risk Assessment in the Federal Government: Managing the Process (1983)
and Science and Judgment in Risk Assessment (1994).

methodologies may in some cases actually underestimate risks- and that it is
appropriate to be precautionary about health risks. EPA's practice of
including precautionary assumptions in its risk assessment policies and
practices has been recognized and affirmed by the National Research Council.
According to the Council, EPA's risk assessment practices rely heavily on
default options or generic approaches. The Council stated that these default
options, or assumptions, are for the most part chosen to “lead to risk
estimates that, although plausible, are believed to be more likely to
overestimate than to underestimate the risk to human health and the
environment.” 7 The Council said, however, that the choice of default
assumptions should have a decreasing impact on regulatory decision- making
over time because, as scientific knowledge increases, uncertainty diminishes
and risk assessments should be less dependent on such assumptions.

Risk assessments of chemicals examine the types of adverse health effects
that might occur in humans and wildlife following chemical exposure (hazard
identification), how the effects vary with the degree of exposure (dose-
response assessment), and the degree to which exposure actually occurs
(exposure assessment). Combining this information enables the

overall risk to be described for decisionmakers (risk characterization).
Once the risk is characterized, risk management involves deciding what
actions, if any, are needed to prevent or reduce the risk, such as limiting

pollutant emissions. As shown in figure 1, the risk management decision
considers other information in addition to the risk characterization. For
example, the risk management decision may be affected by control options-
that is, the

technologies that are available to implement a standard- legal
considerations, and economic factors. Economic information that may be
considered includes cost- benefit analyses, which are required for
economically significant regulations. 8 EPA's regulations that establish
national health- based standards, such as drinking water standards under the
Safe Drinking Water Act, typically have a significant effect on the

7 See Science and Judgment in Risk Assessment (1994). 8 Executive Order
12866 requires detailed cost- benefit analyses for all economically
significant regulations that include those expected to have an annual impact
on the economy of $100 million or more. EPA is also directed by the Unfunded
Mandates Reform Act of 1995 to conduct regulatory cost- benefit analyses
under certain circumstances.

economy. Therefore, EPA conducts both a risk assessment and a national
benefits estimate in these cases, while in many other cases the agency
conducts a risk assessment but conducts only a limited benefits estimate or

none.

Figure 1: Typical Sequence of Risk Assessment and Risk Management Processes

Risk Risk

assessment management Dose- response

Control Legal

assessment

options considerations

Hazard Risk

Risk identification

characterization management decisions

Exposure assessment

Other economic and social factors

Source: Research and Development: Fiscal Years 1997- 1998 Research
Accomplishments, EPA (Dec. 1999).

EPA's various program offices, including those responsible for pesticides,
toxic substances, and air and water pollution, conduct many health risk
assessments that vary in purpose and the availability of data. They range
from single- purpose screening assessments that receive limited review to

fully developed, peer- reviewed assessments that serve as the basis for
major regulations. The program offices may use the hazard identification and
dose- response assessments conducted by EPA's Office of Research and
Development. 9 However, the program offices usually conduct the exposure
assessment and risk management phases and are also usually responsible for
preparing the cost- benefit analyses for major rules. 9 Some of the program
offices, such as the Office of Pesticide Programs and the Office of

Water, do all or some of their own hazard identification and dose- response
assessments.

EPA's methodologies for conducting risk assessments and benefits estimates
have changed in recent years in response to a number of factors, including
recommendations from the National Research Council, changes in the agency's
environmental statutes, the availability of more sophisticated mathematical
and computer models, and new scientific data on and increased understanding
of how cancer develops. For example, in its 1994 report on EPA's risk
assessment activities, the National Research

Council recommended that EPA's risk assessment guidelines identify the
specific assumptions that are default options and clearly state the
scientific and policy basis for each default assumption used. EPA has
proposed revisions to its risk assessment guidelines for carcinogens
(cancer- causing

substances) that call for identifying the default assumptions used and for
highlighting significant issues; they also provide some clarification on
departing from default assumptions. The revised guidelines have been
peerreviewed by EPA's Science Advisory Board but are not yet final. EPA has
started to incorporate some aspects of these new guidelines into some risk
assessments.

Other changes include guidance on EPA's exposure policies and/ or practices
aimed at reducing the use of some precautionary assumptions. For instance,
EPA's estimates of individuals' exposures to contaminants can be
precautionary if the estimates assume that individuals are exposed at the
highest levels. Past exposure assessment and health risk assessment

practices at EPA have sometimes relied on exposure estimates derived from a
hypothetical “maximally exposed individual” who might spend, for
example, a 70- year lifetime drinking only groundwater with the highest
concentrations of contaminants detected. According to the 1997 report of the
Presidential/ Congressional Commission on Risk Assessment and Risk

Management, this approach was often based on such unrealistic assumptions
that using it impaired the scientific credibility of risk assessments. EPA,
like other federal agencies, has moved away from

exposure assessments relying on such maximally exposed individuals. For
example, EPA's exposure assessment guidelines have adopted the use of
distributions of individual exposures. EPA's current guidance indicates that
risk assessments should include both central estimates of exposure (based on
either the mean or the median exposure) and estimates of the exposures that
are expected to occur in small, but definable, “high- end”
segments of the population.

Three Key Factors Three key factors principally influence EPA's use of
precautionary Affect EPA's Use of

assumptions in assessing health risks: (1) the agency's mission to protect
human health and safeguard the environment; (2) the nature and extent of
Precautionary

relevant data- in particular, whether studies of a contaminant's effects on
Assumptions

humans are available or whether the agency must extrapolate from studies
using other animal species; and (3) the nature of the health risk being
evaluated, such as whether the contaminant is thought to cause cancer. For
example, in assessing the risks of contaminants that may cause cancer, the
agency has typically made the precautionary assumption that there is no safe
level of exposure- that is, that any exposure poses some risk of

developing cancer. EPA's Mission to Protect EPA's mission, articulated in
its strategic plan and reflected in statutes, Human Health and the

agency policies, and practices, is to protect human health and safeguard
Environment the natural environment. This mission is a key factor
encouraging the

agency to use precautionary health risk assumptions in the absence of
convincing scientific knowledge. For example, to avoid underestimating risks
to human health, EPA has incorporated a number of precautionary assumptions
in its risk assessment guidelines to address scientific uncertainties.

In some instances, environmental statutes require EPA to protect the public
health with a margin of safety either in assessing risks or in setting
health- based standards. Under the 1996 Food Quality Protection Act, for
example, EPA is required to give special consideration to children's
susceptibility to pesticide residues when the agency sets allowable levels
for such residues in food. Among other things, the statute requires EPA to

make precautionary assumptions in its risk assessments about safe levels of
pesticide residues for children when data are incomplete or unreliable. 10
Under the Clean Air Act, EPA is to establish national standards for ambient
(outdoor) air quality to protect the public health from the effects of
certain widespread air pollutants, such as carbon monoxide and particulate
matter. The act requires that these standards be set at levels that allow
for an “adequate margin of safety.”

10 Children and Pesticides: New Approach to Considering Risk Is Partly in
Place (GAO/ HEHS- 00- 175, Sept. 11, 2000).

The Type of Data, The assumptions that EPA uses in assessing the health
risks of

Particularly Whether contaminants depend largely on whether the agency has
adequate studies Human or Animal Studies of their effects on people, known
as epidemiological studies, or only Are Available

studies of other animals conducted under controlled laboratory conditions,
called toxicological studies. Epidemiologists compare two or more groups of
people to determine which characteristics, such as exposure to contaminants,
distinguish people who get disease from people who do not.

Data from epidemiological studies are preferred for characterizing human
health risks because they can provide the most direct evidence that a
substance poses health risks to people. If these studies are extensive and
of good quality, EPA generally gives greater weight to epidemiological data
than to animal data. Nonetheless, epidemiological studies require the use of
assumptions to address uncertainties. Among the key assumptions in risk

assessments based on epidemiological studies that may be precautionary are
the following: Causality. A key challenge inherent in the use of
epidemiological studies is establishing a causal relationship between the
contaminant being

assessed and the identified health effect. That is, epidemiological studies
may show that a particular substance is associated with a higher incidence
of disease in an exposed population, but generally the studies do not
provide clear evidence that the substance causes the disease. Furthermore,
other scientific information that would help establish a causal
relationship- such as how a contaminant causes the health effect, referred
to as the “mode of action”- often does not exist. In

addition, simultaneous exposures to other contaminants can reduce the
certainty that exposure to a specific contaminant is producing the health
effect that has been identified. Such exposures, referred to as confounding
factors, can only be recognized, controlled, and measured to a certain
extent. Because it is difficult to establish causation on the

basis of epidemiological evidence, EPA must determine whether it can infer a
causal relationship between exposure to a contaminant and observed health
effects through its review of the available epidemiological and other data.
Extrapolation from high doses. The populations analyzed in epidemiological
studies may be exposed to doses of contaminants that

are higher than the doses normally occurring in the environment. For
example, epidemiologists often study more highly exposed populations, such
as factory workers. As a result, EPA must make assumptions when it
extrapolates the effects of high doses to the lower dose levels to which the
general population may be exposed. These assumptions may

or may not be precautionary.

Extrapolation from a study population to other populations. Using
epidemiological data from one population to estimate the health risks for
another population is sometimes precautionary. For example, epidemiological
data on the health risks of arsenic have been developed

principally from a Taiwanese population whose diets may make them more
susceptible than the U. S. population to cancer from exposure to arsenic.
EPA used the Taiwanese data to estimate health risks in the United States
but acknowledges that extrapolating from the Taiwanese data to the United
States may tend to overstate the risk to the U. S. population. On the other
hand, extrapolating from other epidemiological studies, such as those of U.
S. workers, may tend to understate some

health risks for the U. S. population because the workers in the studies
would not include individuals with higher health risks, such as children,
the frail elderly, and those with weakened immune systems that make them
more susceptible to disease.

For most contaminants, epidemiological studies are less commonly available
than animal studies. As a result, EPA relies primarily on studies of
laboratory animals to support its health risk assessments. 11 Laboratory
studies of animals can be controlled, and thus establishing causation is
generally not an issue. Another advantage of animal studies is that they can

provide information on the toxicity of contaminants before they are used,
whereas epidemiological data can be collected only after human exposure.
When using these toxicological studies to assess human health risks,
however, risk assessors must rely on a number of assumptions that may be
precautionary. EPA's assumptions relating to the use of toxicological data
in risk assessment include the following:

Species- to- species inference. The use of toxicological studies in
assessing health risks relies on the assumption that laboratory animals,
such as rats, mice, and monkeys, are surrogates for humans. According to the
National Research Council, extrapolation between different species is
supported by biological principles and empirical observations

for many forms of biological responses, but the scientific basis of such
extrapolation is not established with sufficient rigor to allow broad and
definitive generalizations. The Council has also stated that toxicity is
very often a function of chemical metabolism and that differences among
animal species in metabolic handling of a chemical are not 11 According to
EPA officials, a growing literature suggests that cancer health risks
derived

from both human data and animal data are similar.

uncommon. The Council has further noted that, in most cases, information on
a chemical's metabolic profile in humans is lacking and identifying the
animal species and toxic response most likely to predict the human response
accurately is generally not possible. While risk assessors assume that
health effects seen in laboratory animals are likely to be seen in humans as
well, some tests on laboratory animals have not identified human health
problems. Perhaps the best known

case is that of the drug thalidomide. No adverse health effects were found
in animal testing; however, in humans it caused severe birth defects in the
children of women who took the substance. Similarly, while epidemiological
studies have shown that arsenic is a human

carcinogen, test animals have not developed cancer from exposure to arsenic.
Extrapolation from high doses to low doses. Animal studies must use

much higher doses than the doses that people are typically exposed to
because millions of animals would have to be exposed to low doses in order
to detect adverse health effects. Although some critics question

the validity of extrapolating the effects of high doses of contaminants
given to research animals to low doses that people encounter in the
environment, the National Research Council recommended that EPA continue to
assume adverse effects from lower doses in the absence of other information.
Use of highly sensitive animal species. Toxicology studies often use animals
that are highly sensitive to the contaminant being studied in

order to ensure a detectable response. Similarly, when there are multiple
studies assessing the toxicity of a substance but information is lacking on
which species responds most like humans, EPA uses the most sensitive species
in assessing human risk. 12

In addition to assumptions inherent in the use of toxicological data that
may be precautionary, there are other assumptions, such as the following,
that are generally seen as not being precautionary: Studies account for
exposure to only one chemical. Animal studies usually address an individual
chemical, even though people are often exposed to multiple contaminants in
the environment. There is evidence that for some contaminants, combinations
of exposures may increase

health risks to higher levels than would be estimated by simply adding the
individual risks together. Therefore, a toxicity assessment that relies 12
Reference Dose (RfD): Description and Use in Health Risk Assessments, EPA
(Mar. 1993).

on animal studies that address a single contaminant may tend to
underestimate toxicity to people who are exposed to multiple contaminants.

Studies account for exposure through only one route. Similarly, people may
be exposed to a contaminant by more than one route, which could mean that
risks assessed on the basis of an animal study using only one such route
could underestimate the risks to people. Studies generally use mature
animals. According to EPA officials, most toxicology studies use mature
animals, thereby ignoring effects of exposure on the developing animal-
which may be more frequent, more severe, or very different in nature from
the effects on mature animals.

The Type of Health Risk Over time, EPA has developed different policies and
methodologies for

Being Assessed assessing different types of health risks. Most notably,
EPA's approach for

cancer risk assessment has differed from its approaches for noncancer risk
assessment. Both types of risk assessments rely on assumptions, including
precautionary assumptions; however, the specific assumptions used differ
with the type of risk being assessed. Central to the development of EPA's
approach to risk assessment for carcinogens has been the theory that even

a small number of changes in a single cell can lead to the uncontrolled
growth of cells known as cancer. This theory implies that there is no safe
level of exposure, or threshold, below which the contaminant does not

pose a risk. In contrast, for health problems other than cancer, EPA has
generally posited that there is some safe level of exposure to a contaminant
before health effects occur.

EPA's Assessment of Cancer In assessing cancer risks, EPA develops a
quantitative estimate of the Risks

expected increase in the incidence of cancer resulting from varying
exposures to a contaminant. This estimate is called the dose- response
relationship. In dose- response assessment, EPA has not traditionally
speculated as to how the potential carcinogen induces cancer, and such

data have generally not been available, according to the director of the
quantitative risk methods group within EPA's Office of Research and
Development. In developing the dose- response relationship, EPA generally
uses two key precautionary assumptions:

A linear, no- threshold relationship between the dose and the health effects
at low doses. This assumption posits no safe level of exposure to a
carcinogen- that is, any exposure presents some risk of developing cancer.
However, for some contaminants, a nonlinear dose- response relationship is
believed to exist, while other contaminants are believed

to have a threshold (see app. II). In addition, this assumption relies on
the observed relationship between high doses and the incidence of cancer-
based on either epidemiological or toxicological data- to predict or
extrapolate the cancer risk at the much lower levels for which no data on
health responses are available. When EPA does not have sufficient data or a
model for extrapolating the cancer risk at lower doses, EPA's guidelines for
cancer risk assessment call for estimating a

linear relationship between dose and health effects. That is, each
additional increment of exposure is assumed to produce the same proportional
change in the health effect.

“ Upper bound” estimates of risk. EPA uses statistical
procedures to develop an estimate of the dose- response relationship that is
very unlikely to be exceeded by the true risk. This estimate, typically set
at the 95- percent upper confidence limit, represents an upper bound on risk
(see fig. 2). 13 According to a report prepared for the National

Commission on Risk Assessment and Risk Management, 14 if the unknown dose-
response relationship is linear at low doses, this procedure overestimates
the true risk by a relatively small factor, usually two- to threefold. If,
however, the true relationship at low doses is nonlinear, this approach will
overestimate risks by larger factors that increase as the dose levels
decrease.

13 The 95- percent upper confidence limit is a statistically derived upper-
limit estimate of risk that is designed to overstate rather than understate
human risk. 14 A Survey of Methods for Chemical Health Risk Assessment Among
Federal Regulatory Agencies, prepared for the National Commission on Risk
Assessment and Risk Management by Lorenz R. Rhomberg, Ph. D., Harvard Center
for Risk Analysis, Harvard School of Public Health (1996).

Figure 2: EPA's Standard (Default) Dose- Response Relationship for
Carcinogens

Increasing response (health risk)

EPA chooses the 95- percent upper confidence limit

Central estimate

Increasing dose (exposure)

Note: The figure above is for a hypothetical contaminant. The relative
difference between the 95- percent upper confidence limit and the central
estimate varies across contaminants.

The National Research Council stated in 1994 that EPA should continue to use
upper- bound estimates of lifetime cancer risks. It noted, however, that
whenever possible, this estimate should be supplemented with other
descriptions of cancer risk that more fully reflect the uncertainty

associated with these estimates. EPA's cancer risk assessment guidelines,
issued in 1986, emphasize that its default assumptions lead to a plausible
upper limit on the risk that is consistent with some proposed mechanisms for
how cancer develops. The

guidance also states, however, that the estimate is not necessarily a
realistic predictor of risk, since the true value of the risk is unknown and
may be as low as zero.

In addition to the precautionary assumptions that relate to the doseresponse
relationship, EPA's risk assessments for cancer reflect other precautionary
assumptions such as the following:

A substance that is carcinogenic in animals is likely to be a human
carcinogen. When benign and malignant tumors are observed in animals, the

incidence of both is combined to represent the substance's carcinogenic
potential in humans. In the absence of information indicating which species
responds most like humans, the animal species exhibiting the greatest
carcinogenic sensitivity is given the greatest emphasis in developing
estimates of human cancer risk.

As for other types of risk assessment, EPA's policies and practices for
assessing cancer risks have been changing as new analytical techniques are
developed and scientific knowledge increases. For example, the agency's
guidelines for assessing carcinogenic risk are being revised. Among the
issues addressed in the draft guidelines is the no- threshold assumption for
carcinogens. As more research into the mechanisms of how cancer develops has
become available, there have been challenges to the theory that there is no
safe level of exposure to a carcinogen. For example, in 1994

the National Research Council reported that risk models that use a threshold
are plausible for many carcinogens.

In March 2000, a federal appeals court ruled that in setting a health- based
maximum contaminant level goal for chloroform (a chemical byproduct of
chlorination, the most widely used technique for ensuring the safety of

drinking water) under the Safe Drinking Water Act, EPA “openly
overrode the ‘best available' scientific evidence,” which
suggested that there is a threshold for the carcinogenic effects of
chloroform. 15 In 1998, EPA

concluded that the assumption of a nonlinear relationship, which is
permitted under its existing carcinogenic risk guidelines issued in 1986,
would be more appropriate than a linear assumption. However, in the final
rule, EPA used a no- threshold, linear assumption. According to EPA
officials, staff responsible for conducting risk assessments have been
reluctant to depart from the standard cancer defaults in EPA's existing

policy because of uncertainties about when such a departure would be
appropriate. 16 The criteria for departing from the defaults have not been
clearly articulated in the past, and there is ongoing debate over whether
departing would be protective of sensitive populations, including children.

The revised guidelines, which are not finalized as of October 2000, will
offer more direction to risk assessors in terms of when and how to depart
from the traditional no- threshold assumption, according to EPA officials.
15 Chlorine Chemistry Council v. Environmental Protection Agency, 206 F. 3d
1286, 1290 (D. C. Cir. 2000). 16 In the case of chloroform, the risk
assessors departed from the standard cancer default, but EPA made the policy
decision to use the standard default, in part, because not using the

default would represent a “significant and precedential”
application of new science that had important implications for other
contaminants regulated under the Safe Drinking Water Act.

EPA's Assessment of Noncancer In addition to assessing carcinogenic risks,
EPA assesses other threats to Health Risks

human health, such as respiratory problems associated with air pollution.
Because of the wide variety of noncancer health effects and the diversity of
ways in which contaminants are suspected of working, EPA does not have a
single approach for assessing noncancer risks comparable to its agencywide
guidelines for assessing cancer risks. 17 However, one

approach- using what is called a reference dose, or RfD- is used most
frequently to determine a threshold or safe level of exposure to a
contaminant. The reference dose is an estimate of a daily exposure level
that is not likely to cause “appreciable risk of deleterious effects
during a lifetime.” 18 In estimating a safe level of exposure to the
noncarcinogenic effects of a contaminant, EPA first determines the dose
level at which no adverse

effects have been observed. This level is then reduced because of
uncertainties in the data- generally toxicological data from animal studies.
That is, the dose at which no adverse effects have been observed is reduced
(divided) by one or more uncertainty factors (sometimes called safety
factors). The uncertainty factors account for the possibility that people
might need a lower level of exposure to better ensure safety. 19 Each
uncertainty factor typically reduces the level at which no adverse effects
have been observed to one- tenth the original dose. The following two

uncertainty factors are used most frequently: A factor of 10 is generally
used to account for variation in sensitivity

among people, such as the elderly and other populations that are more
susceptible to diseases. A factor of up to 10 is generally used to account
for the uncertainty associated with using the results of laboratory animal
studies to estimate the health effects expected in people. This factor,
usually set at 10, stems from the concern that people could be more
sensitive to the

toxic effects of a contaminant than are laboratory animals. For example, 17
The agency has, however, issued risk assessment guidelines for several types
of noncancer health threats, including developmental toxicity and
reproductive toxicity. 18 EPA also develops reference concentrations (RfC),
estimates of a daily exposure level (in terms of air concentrations rather
than dose) that is likely to be without an appreciable risk of adverse
noncancer effects during a lifetime. 19 EPA uses the term “uncertainty
factor” rather than “safety factor” because of concerns
that the latter term implies an absolutely safe level, an assurance the
agency does not believe it can provide.

small animals have faster metabolic processes, allowing them to eliminate
contaminants from their bodies more quickly than people can. Other factors
may be added to account for such uncertainties as a safe level for lifetime
human exposure when only short- term animal studies are available. According
to a report prepared for the National Commission on Risk Assessment and Risk
Management, two or three factors are typically used in assessing noncancer
risks. This could result in as much as a 100- or 1,000- fold decrease in the
estimated safe level of exposure.

As for cancer risk assessment, EPA's approaches for noncancer risk
assessment are changing over time with increases in scientific understanding
of how contaminants cause adverse health effects and the relationship
between the dose of a contaminant and the increased health risk expected. In
some instances, EPA has enough information on the general human population
to estimate a quantitative relationship between exposures to varying
concentrations of the contaminant and the expected

increased health risks. In these cases, EPA does not have to use the
reference dose approach that relies on various factors to reflect
uncertainties and data gaps. Such is the case with six widespread air
pollutants for which EPA has established national health- based standards
under the Clean Air Act, including particulate matter.

EPA's Use and The two major health- based regulations we examined differed
in the extent

Disclosure of to which they used precautionary assumptions in estimating
health risks

and benefits. EPA used such assumptions only to a limited extent in the
Precautionary Risk case of the 1997 air quality standards for particulate
matter but relied on Assumptions in

them more in its recently proposed drinking water standard for arsenic.
Estimating Benefits for

However, in assessing the health risks of arsenic and estimating the related
benefits of the proposed standard, EPA used new analytical techniques that
Two Major Regulations removed other precautionary assumptions used in the
past. In its proposed arsenic regulation, EPA identified the key health
uncertainties and precautionary assumptions it used in assessing the risks
of arsenic in drinking water but was less complete in identifying them in
its formal costbenefit

analysis. Furthermore, although its guidance on cost- benefit analyses calls
for assessing uncertainty using sensitivity analysis, EPA did not perform
sensitivity analysis that could have provided information on the potential
impact of the precautionary health assumptions that underlie its benefits
estimate.

EPA's 1997 National Under the Clean Air Act, EPA sets health- based National
Ambient Air

Ambient Air Quality Quality Standards for certain widespread air pollutants
including

Standards for Particulate particulate matter- the generic name for a mixture
of air pollutants

commonly found across the United States. 20 EPA is required to review the
Matter

existing standards every 5 years to ensure that the standards are based on
the most recent scientific information. EPA last revised the particulate
matter standards in July 1997, at which time it added standards for fine
particulate matter (particles less than 2.5 micrometers in diameter). 21
(See app. III for information on the several standards EPA set for
particulate matter.) EPA estimated that the new standards for fine particles
could save between 3, 300 and 15,600 lives annually. EPA's monetary estimate
of the health benefits of the new standards, which included not only lives
saved but also other benefits, such as cases of chronic bronchitis avoided,
ranged from $14. 5 billion to $96.1 billion per year. 22

When EPA evaluated the health risks from particulate matter in developing
its 1997 standards, it generally did not rely on precautionary assumptions.
This was largely because EPA had better data at relevant exposure levels
than is often the case when the agency assesses risks. The strengths of the
data included the following:

EPA had a large body of epidemiological research upon which to base its risk
assessment. This contrasts with most of EPA's risk assessments, which must
rely on animal studies. Thus, EPA did not have to make the assumptions
needed when animal studies are used to predict human health risks. Moreover,
the epidemiological research showed largely consistent associations between
particulate matter in the outdoor air

and a variety of health problems, including premature death from respiratory
and cardiovascular causes, particularly among the elderly; exacerbation of
cardiopulmonary and respiratory illnesses; an increased 20 Particulate
matter can include many chemically and physically diverse substances and can
vary significantly by location. A number of activities typically add to the
concentration of particulate matter in the air, including combustion from
power plants, industrial facilities, cars, trucks, and wood stoves;
construction and demolition activities; and road dust. Among the major
chemical components of particulate matter are sulfates, nitrates, acids,
metal compounds, and water. 21 A micrometer, also known as a micron, is one-
millionth of a meter.

22 The particulate matter rule accounts for as much as 54 percent of the
total benefits of the 48 major rules issued governmentwide from Apr. 1,
1995, to Mar. 31, 1999, for which federal agencies estimated benefits.

incidence of bronchitis in children; and an increased number of childhood
asthma attacks. EPA had sufficient data to estimate a range of risks showing
how risk

would vary with different concentrations of particulate matter in the air.
Typically called a dose- response relationship, this is referred to as a
concentration- response relationship in the case of air pollutants, where
the observed relationship is between the concentration of the pollutant in
outdoor air and the given health effects. In contrast, for most other
noncarcinogenic pollutants, EPA would typically make only one estimate- that
of a dose likely to be safe (a reference dose discussed

above). In estimating a dose likely to be safe, EPA would typically reduce
the allowable exposure by applying uncertainty factors to account for such
things as people's differing susceptibilities to the pollutant's health
effects. Furthermore, EPA had information about directly relevant

concentration levels of particulate matter in the air, that is, at levels to
which the population was actually exposed. Thus, the agency did not have to
extrapolate exposure information from studies using higher doses. Because
the studies EPA relied on were based on U. S. data for a number of urban
areas, EPA did not have to extrapolate the effects observed in one
population to estimate the health effects in another population. Moreover,
in assessing the health effects of particulate

matter on the basis of findings from numerous epidemiological studies, EPA
used risk levels that represented central tendency estimates rather than
upper- bound estimates of the health effects. Despite the strengths of the
available epidemiological data, there were significant scientific
uncertainties in EPA's evaluation of particulate matter's health effects,
which led EPA to use at least one precautionary assumption. EPA's
sensitivity analyses showed that the most significant uncertainty was
whether a threshold concentration existed, that is,

whether the health effects would be associated with particulate matter at
any level of exposure. In the quantitative health risk assessment used in
the standard- setting process, EPA made the precautionary assumption that
there was no threshold for the health problems associated with particulate
matter. EPA officials said they made this precautionary assumption

because they did not have information that indicated the existence of a
threshold for the various health effects included in the risk assessment.

However, EPA's methodology did not estimate the health risks down to a
concentration of zero. EPA estimated the health effects associated with
particulate matter starting at the larger of

the level of naturally occurring particulate matter (also called the
background level) 23 or the lowest average annual level of particulate
matter observed in the

epidemiological studies EPA relied upon. 24 EPA's estimate of adverse health
effects was lower than it would have been had the agency included health
risks at the lowest levels. However, EPA officials believe it was
appropriate to measure health effects only above the naturally occurring
levels of particulate matter because it is unlikely that concentrations of
particulate matter could be reduced below such levels. Also, EPA did not
estimate health effects below the levels reported in the epidemiological
studies because of the uncertainties about health effects at such levels.

23 EPA made the following “base case” assumptions about
background concentrations of particulate matter (expressed in terms of
micrograms per cubic meter of air) in the two locations included in its risk
analysis: fine particulate matter (also called PM2. 5)- 3. 5 in Philadelphia
County, Pennsylvania, and 2. 5 in Southeast Los Angeles County, California;
course and fine particulate matter (also called PM10)- 8 in Philadelphia
County and 6 in Southeast Los Angeles County. EPA also conducted sensitivity
analyses looking at the

effects of alternative assumptions about background levels. 24 For example,
the lowest median annual concentration of particulate matter in a key study
assessing the mortality risks associated with this pollutant was 9
micrograms per cubic meter of air.

When EPA estimated the health benefits of the particulate matter standards
as part of its cost- benefit analysis, it used essentially the same
doseresponse relationships used in the risk assessment. EPA's cost- benefit
analysis also identified the scientific uncertainties associated with this
dose- response relationship. However, in estimating the benefits of this
rule, EPA did not include the risk assessment's precautionary assumption
that

there was no threshold (above certain levels) for the health effects.
Rather, EPA estimated benefits using two different assumptions about
thresholds above background levels- at concentrations of 12 and 15
micrograms of fine particulate matter per cubic meter of air. According to
EPA officials,

the additional benefits from risk reductions at concentrations of
particulate matter below 12 micrograms per cubic meter would not be as
significant as at the higher levels because few geographic locations would
achieve concentrations at the lowest pollution levels. 25

Another assumption that EPA made in evaluating particulate matter's health
risks and in estimating benefits, an assumption that is typically used when
epidemiological data are involved, was causality- that is, EPA assumed that
there is a causal relationship between particulate matter and the health
effects with which it has been associated. Some scientists and other
commenters on EPA's proposed particulate matter regulation believed that the
causality assumption was precautionary. EPA, however, did not characterize
this assumption as precautionary because it believed

the consistency of the results from a large number of locations and the
coherent nature of the results suggest a likely causal role of particulate
matter in contributing to these health effects.

Nonetheless, many questions remain about how particulate matter may be
causing premature death and other adverse health effects, including
questions of whether there may be confounding agents, such as other air
pollutants (e. g., sulfur dioxide and ozone), that may be causing at least
some of the problems. The National Research Council has cited the
“relatively consistent but poorly understood associations between
ambient

particulate matter concentrations and various adverse health effects.”
26 Because of the many unknowns, EPA's Clean Air Scientific Advisory 25 As a
result of guidance from its economic advisory committee, EPA's more recent
benefits estimates done for other regulations, such as its estimate of the
prospective benefits of the Clean Air Act, have assumed, among other
scenarios, that there is no threshold above the background level of
particulate matter. 26 See Research Priorities for Airborne Particulate
Matter, Vol. 1 (1998), p. 19.

Committee strongly recommended that EPA implement a targeted research
program to address these unanswered questions and uncertainties. In its
fiscal year 1998 appropriations, the Congress provided funding of $49. 6
million specifically for research on particulate matter. The Congress also
directed the EPA Administrator to arrange for an independent study to, among
other things, monitor and report on progress toward improved understanding
of the relationship between particulate matter and its health effects. The
National Research Council is carrying out this mandated work.

Some research findings have recently been released that buttress the
findings in the studies EPA relied on in making its decisions to regulate
fine particles in 1997. For example, in July 2000, the Health Effects
Institute reported that its reanalysis and additional analysis of the
underlying data essentially validated two key studies, as well as these
studies' findings of an

association between particulate matter and mortality, that EPA had relied on
in issuing its 1997 rule. 27

Although EPA develops what it considers best estimates rather than
precautionary estimates of the risks from particulate matter, the Clean Air
Act requires that the health- based standards be set with an “adequate
margin of safety.” According to officials from EPA's Office of Air and
Radiation, building in the required margin of safety for the particulate
matter standards was a risk management decision, not part of the

assessment of risk. That is, the margin of safety was incorporated into the
allowable concentration levels, or standards, chosen by the Administrator.
These standards have been challenged in court. A summary of the issues
associated with the decision over the levels at which the standards were

set, as well as the court case, is included in appendix IV. While the
judicial review of the 1997 particulate matter standards proceeds, the next
5- year mandated review of the standards, expected to be completed in 2002,
is also under way. 27 The Health Effects Institute is a nonpartisan,
independent research organization whose

major funding is from EPA and the auto industry. See its Reanalysis of the
Harvard Six Cities Study and the American Cancer Society Study of
Particulate Air Pollution and Mortality. A Special Report of the Institute's
Particle Epidemiology Reanalysis Project (July 2000).

EPA's June 2000 Proposed Under the Safe Drinking Water Act, EPA sets health-
based, legally

National Primary Drinking enforceable drinking water standards limiting the
level of contaminants in Water Regulation for the nation's drinking water
systems that can adversely affect public health.

Arsenic First, EPA establishes a health- based goal at a level at which no
known or

anticipated adverse health effects occur and that allows an “adequate
margin of safety.” If a contaminant, such as arsenic, is likely to
cause cancer, EPA generally sets the goal at zero. After setting the goal,
EPA

typically establishes an enforceable standard, called a maximum contaminant
level, that is as close to the health- based goal as is feasible,
considering the available technology and costs. 28 EPA is also to complete
an economic analysis to determine whether the benefits of the standard
justify the costs. If the benefits do not appear to be justified, EPA may
adjust the standard to a level that “maximizes health risk reduction
benefits at a cost that is justified by the benefits.” 29 This latter
provision is a new authority the Congress gave to EPA under amendments to
the Safe Drinking Water Act in 1996.

The 1996 amendments to the Safe Drinking Water Act also require EPA to
propose a new arsenic standard by January 1, 2000, and to issue the final
rule by January 1, 2001. Recognized as a toxic element for centuries,

arsenic has also been associated with skin, bladder, and other cancers. 30
EPA issued its proposed arsenic standard in June 2000. Specifically, EPA
proposed setting the maximum contaminant level goal at zero and lowering the
enforceable contaminant standard from 50 micrograms per liter to 5
micrograms per liter. 31 EPA is using its new authority to set the standard
at a higher level than the technologically feasible level of 3 micrograms
per liter both because (1) it did not believe that the costs were justified
by the benefits at that level and (2) there were a number of uncertainties,
including scientific uncertainty about the health effects of arsenic at low
levels of exposure in drinking water. EPA also requested public comment on
alternative standards of 3, 10, and 20 micrograms per liter. 28 Drinking
water standards also often apply to contaminated groundwater at hazardous
waste sites regulated under EPA's Resource Conservation and Recovery Act and
Superfund programs, and arsenic is a key contaminant at many of these sites.

29 P. L. 104- 182, 110 Stat. 1613, 1624 (1996). 30 These cancers have been
associated with arsenic present in drinking water at concentrations higher
than those observed in U. S. drinking water supplies.

31 A microgram is one- millionth of a gram. The Public Health Service first
established the standard of 50 micrograms per liter in 1942.

Arsenic occurs naturally in the environment, for example, in rocks, soil,
and groundwater, and has been used in a variety of commercial activities,
most notably today as a component of wood preservatives. Because arsenic
does not degrade in the environment, contamination from

historical releases is cumulative. EPA estimates that about 10 percent of
groundwater and surface water systems have average arsenic levels above 5
micrograms per liter, and about 4.5 percent have average arsenic levels
above 10 micrograms. Arsenic concentrations in groundwater generally are

highest in the Western United States, where EPA estimates that 12 percent of
groundwater systems exceed 10 micrograms per liter. In setting this
standard, EPA sought to protect public health at a level such that at least
90 percent of the exposed population would face a lifetime risk of less than
1 in 10,000 for developing bladder cancer. EPA estimates that the new
standard will avoid 16 to 36 bladder cancer cases each year, 4 to 9 of which
would be fatal. As discussed below, EPA reported that several other health
benefits are expected to result from the proposed standard, including

reductions in lung cancer, that it cannot reliably quantify at this time.
For the proposed arsenic standard, EPA estimated the health risks of bladder
cancer on the basis of epidemiological studies, 32 relying primarily on a
review of arsenic health effects research conducted by the National Research
Council at EPA's request. The Safe Drinking Water Act requires EPA to
consider peer- reviewed scientific information on health effects in setting
drinking water standards, and EPA considered the Council's study

as presenting the best available peer- reviewed science. EPA used four
alternative dose- response relationships reported by the National Research
Council. These relationships were based on epidemiological data on bladder
cancer mortality in a high- arsenic region in Taiwan. The median arsenic
concentrations in the 42 Taiwanese villages studied ranged from 10
micrograms per liter to 934 micrograms per liter; 29 of the villages had
median arsenic concentrations at or above 100 micrograms per liter. The
National Research Council noted that studies in Chile and Argentina observed
risks of lung and bladder cancer of the same magnitude as those reported in
Taiwan at comparable levels of exposure. EPA's estimate of the reductions in
health risks and the benefits of its proposed arsenic standard relied on the
following key precautionary assumptions:

32 As noted earlier, arsenic is one of the few contaminants that causes
cancer in humans but not in laboratory animals.

EPA assumed that the dose- response relationship for arsenic is linear
without a threshold, even at low concentrations, rather than sublinear and/
or with a threshold. (For either of these alternative assumptions, the
response at low doses would have been less than that predicted by a linear
dose- response relationship.) EPA made this assumption because it concluded
that the scientific evidence suggesting a sublinear doseresponse
relationship is not strong enough to depart from its default assumption of a
linear, no- threshold dose- response relationship. While the National
Research Council noted that the most plausible scientific evidence on how
arsenic causes cancer supports a sublinear doseresponse

relationship, the Council concluded that the available evidence was
inconclusive and did not meet EPA's stated criteria in its 1996 proposed
cancer risk assessment guidelines for departure from the default assumption
of linearity. In its proposed rule, EPA solicited comments on this issue.
EPA assumed that the National Research Council's risk estimates based

on epidemiological data for a rural population in Taiwan could apply to the
U. S. population, even though the Taiwanese diet, compared to the U. S.
diet, includes (1) higher levels of arsenic and (2) lower levels of
selenium, which has been shown to moderate the adverse health effects of
arsenic in animal studies. The Council noted that available data

suggest that arsenic intake from food is higher in Taiwan than in the United
States, and EPA stated in its proposed rule that “arsenic intake (by
persons in the Taiwanese study region) from sources other than drinking
water would overestimate the unit risk calculated from the Taiwan
study.” The Council also noted that these differences “could
affect the relevance of the results” of its Taiwanese- based risk
estimates for a risk assessment for the U. S. population. In light of this
concern, the Council recommended that EPA investigate the relationship
between nutritional factors in study populations and susceptibility to
arsenicinduced cancer. Without this information, the Council indicated that
it might be appropriate to be precautionary in risk assessments of arsenic.

Because of data limitations, EPA relied on the National Research Council's
assumption that the individuals in the Taiwanese study who got bladder
cancer were exposed to the median (50th percentile) level of arsenic found
in the water of their villages' wells. In its notice of the

proposed standard, EPA acknowledged that this assumption is precautionary,
citing its Expert Panel on Arsenic Carcinogenicity, which said that biases
from using average doses for groups lead to overestimation of risk.

Each of these assumptions tends to overestimate the risk of arsenic exposure
and the benefits associated with reducing those risks. EPA's proposed rule
clearly identifies the precautionary assumptions discussed above and also
indicates that the agency will reexamine the use of these

assumptions before issuing the final rule. 33 The proposed rule also
indicates that EPA's use of these assumptions may have resulted in a
significant overestimation of the risk of bladder cancer. Consequently,
EPA's estimate of the related benefits may also be significantly overstated.
EPA officials told us that they did not attempt to determine the effects of
varying these precautionary assumptions by conducting sensitivity analyses
because they did not have sufficient information to do so. We note, however,
that even in the absence of sufficient data, sensitivity analyses can use
“what if” assumptions to assess the potential impact of
precautionary assumptions. Furthermore, EPA's guidelines for preparing

economic analyses state that sensitivity analyses should be performed on key
assumptions, if feasible, in assessing and presenting uncertainty. 34
Finally, EPA's cost- benefit analysis report identified the precautionary
assumption of a linear dose- response relationship but did not identify the
other precautionary assumptions discussed above. 35 Thus, some key
scientific uncertainties associated with the arsenic benefits estimates are
not reflected in this report.

33 EPA's proposed rule represents the agency's documentation of its risk
assessment on arsenic at this time. EPA also cites the report of the
National Research Council from which it derived its risk estimates, Arsenic
in Drinking Water, National Academy Press (1999). 34 This guidance is
consistent with GAO's 1997 recommendation to EPA to ensure that the agency's
cost- benefit analyses identify the sensitivity of benefit and cost
estimates when there are major sources of uncertainty. See Air Pollution:
Information Contained in EPA's Regulatory Impact Analyses Can Be Made
Clearer (GAO/ RCED- 97- 38, Apr. 14, 1997). We have also recommended that
the Office of Management and Budget amend its guidance to strengthen the
clarity and credibility of cost- benefit analyses. See Regulatory Reform:
Agencies Could Improve Development, Documentation, and Clarity of Regulatory
Economic Analyses (GAO/ RCED- 98- 142, May 26, 1998).

35 Proposed Arsenic in Drinking Water Rule: Regulatory Impact Analysis, EPA
815- R- 00- 013 (June 2000).

The National Research Council concluded that risk assessments of bladder,
lung, and other cancers associated with arsenic can be performed and also
recommended further study to characterize a number of adverse noncancer
effects, including heart disease and diabetes. Because EPA based its risk
estimate only on bladder cancer, the agency believes that it underestimated
the combined risk of all arsenic- induced health effects. In its proposed
rule, EPA said that its decision to use only bladder cancer was a judgment
call that was guided by the best available science. In addition, EPA did
develop a rough estimate of the benefits of reductions in lung cancer, using
what it termed a “what if” scenario, and it identified them as
potential benefits in its cost- benefit comparisons. 36 EPA expects to have
a

peer- reviewed quantitative analysis of lung cancer risk available for its
risk assessment and benefits estimate for the final rule. EPA's cost-
benefit analysis report also indicates that the monetary benefits associated
with

reductions in skin cancer would be minimal. Finally, EPA does not believe it
has sufficient information to quantify the numerous noncancer health effects
associated with arsenic.

Although EPA's risk assessment and benefits estimate included several
precautionary assumptions, EPA also used new analytical techniques that
removed other precautionary assumptions used by the agency in the past: 37
According to officials in EPA's Office of Water, past risk assessments and

benefits estimates for drinking water standards typically relied on the
precautionary assumption that all individuals consume 2 liters of drinking
water per day over his or her lifetime, which is more than the

average individual is believed to consume. This assumption tended to
overestimate risk because health risks from contaminants in drinking water
increase with water consumption. For the proposed arsenic rule,

however, EPA assumed that different individuals consume different amounts of
water. EPA's estimates of these varying amounts were based on recent survey
data indicating that, on average, an individual consumes only about 1 liter
of drinking water a day. By using more

36 For its rough estimate of lung cancer benefits, EPA relied on the
National Research Council's statement that “some studies have shown
that excess lung cancer deaths attributed to arsenic are 2- to 5- fold
greater than the excess bladder cancer deaths.” EPA acknowledged that
this estimate is probably too high because a reanalysis of the studies
indicates that the excess lung cancer deaths attributed to arsenic are
approximately equal to the excess bladder cancer deaths. 37 EPA calculated
risks using Monte Carlo analysis, which employs computer simulations to
calculate a range of risk values rather than a single “point
estimate” of risk.

realistic water consumption assumptions, EPA produced both risk and benefits
estimates that were significantly less than they would have been if it had
continued to use the precautionary assumption about water consumption.

EPA's risk assessment and benefits estimate for arsenic also did not include
another significant precautionary assumption that EPA typically uses for
carcinogens. Although EPA did make the precautionary

assumption that the dose- response relationship is linear with no threshold,
it did not use the upper- bound estimate of this linear relationship.
Instead, it used ranges of linear dose- response

relationships derived from a range of risk estimates reported by the
National Research Council. Because EPA used ranges of dose- response
relationships, its risk estimates and benefits estimates were significantly
less than they would have been if it had used its typical precautionary
assumption of an upper- bound dose- response relationship.

Conclusions The particulate matter and arsenic rules demonstrate some of the
key uncertainties that EPA must address in evaluating the health effects of
contaminants and determining the appropriate actions to take. In terms of
the types of precautionary assumptions that EPA may use to address
scientific data gaps and uncertainties, these cases are perhaps atypical in
that sufficient epidemiological data are available in both cases to evaluate
human health risks, whereas EPA generally must extrapolate from animal data.
However, they are quite typical in that their fundamental uncertainty is
about adverse health effects at the lowest levels of exposure- specifically,
whether there is a threshold below which adverse health effects do not
occur. Neither epidemiological nor animal data can provide complete
information on health effects at the low levels to which people

are typically exposed, and therefore assumptions about these health effects
must be made. Greater uncertainty about the effects at low levels may exist
for arsenic than for particulate matter because the epidemiological
information on particulate matter is based on actual exposure levels in the
United States, whereas the data on arsenic are primarily from Taiwanese

populations exposed to much higher levels of arsenic than U. S. populations.
However, the case of particulate matter illustrates how difficult it is to
scientifically resolve uncertainties about causality associated with the use
of epidemiological data- how challenging it is to

conclusively demonstrate that the health effects associated with particulate
matter are, in fact, primarily caused by exposure to the particulate matter
rather than other factors, including other pollutants.

While EPA identified in its proposed regulation the key health uncertainties
and precautionary assumptions it used in assessing the risks of arsenic in
drinking water, it did not identify all of them in its formal cost- benefit
analysis. Because this regulation is still being developed, EPA has the
opportunity both to fully disclose the precautionary assumptions used and to
incorporate some additional analyses into its final regulation. For example,
sensitivity analyses showing the impact of precautionary assumptions on the
risk assessment and benefits estimate could help the Administrator in
setting the standard and could also provide other

interested parties with a more complete understanding of the potential range
of benefits of the standard.

Recommendation for To better inform decisionmakers about the effects of
precautionary health Executive Action

assumptions on EPA's estimates of the benefits associated with the arsenic
rule, we recommend that the EPA Administrator ensure that EPA's costbenefit
report for the final rule fully disclose the precautionary assumptions used
and provide sensitivity analysis on the key precautionary assumptions
included in the agency's benefits estimate.

Agency Comments We provided a draft of this report to EPA for review and
comment. In commenting on the draft, officials from the National Center for

Environmental Assessment, Office of Research and Development, stated that
the report presents a fair and informative discussion of the characteristics
of data typically available for risk assessment. In addition, the officials
supported the recommendation that the cost- benefit report for the arsenic
rule disclose precautionary assumptions and provide sensitivity analysis,
subject to the availability of data pertinent to such an analysis.

Officials from the Offices of Air and Radiation; Water; and Policy,
Economics, and Innovation provided technical comments and clarifications
that we incorporated as appropriate.

The scope and methodology for our work are discussed in appendix I. We
conducted our work from October 1999 through October 2000 in accordance with
generally accepted government auditing standards. We will send copies of
this report to the Honorable Carol M. Browner, Administrator, EPA, and to
other interested parties. We will also make copies available to others on
request. Please call me or Christine Fishkin at

(202) 512- 6111 if you or your staff have any questions. Other key
contributors to this report are David Goldstein, Bruce Skud, and Susan
Swearingen.

David G. Wood Director, Natural Resources and

Environment

Appendi Appendi xes x I

Objectives, Scope, and Methodology Objectives The Honorable Bud Shuster,
Chairman of the House Committee on Transportation and Infrastructure, and
Senator Frank R. Lautenberg asked us to examine whether the Environmental
Protection Agency's (EPA) benefits estimates for major environmental
regulations that establish health- based standards reflect precautionary
assumptions about health risks. As agreed with our requesters, we focused
our work on (1) key factors that affect EPA's use of precautionary
assumptions in assessing health risks and (2) whether EPA used and
identified precautionary assumptions in estimating the health risks and
benefits of recent major regulations setting health- based standards.

Scope The issue of whether EPA's estimates of the benefits of health- based
environmental standards reflect precautionary health risk assumptions is
relevant to major regulations that set health- based standards and estimate
health benefits. The benefits of major regulations- that include those with
an impact on the economy of $100 million or more- are provided in economic
analyses (also referred to as regulatory impact analyses) required by
Executive Order 12866. In fiscal years 1997 through 1999, four of EPA's
major regulations set new health- based standards- two each under the Clean
Air Act and the Safe Drinking Water Act. In fiscal year 2000 through June
2000, EPA proposed two health- based standards under the

Safe Drinking Water Act. We selected one rule each from EPA's air and water
programs in order to report on the agency's current methodologies and
practices for estimating health risks and benefits: EPA's regulations (1)
establishing air quality standards for particulate matter in July 1997 and
(2)

proposing a drinking water standard for arsenic in June 2000. These
regulations generally reflect EPA's current practices and procedures for
risk assessments and benefits estimates for major categories of contaminants
regulated by EPA under the Clean Air Act and the Safe Drinking Water Act.
For example, the particulate matter rule covers one of the six widespread
air pollutants that EPA is required to set standards for and to review every
5 years. EPA uses similar approaches in estimating

health risks and health benefits for the other criteria pollutants,
including the ozone rule that was also promulgated in 1997. As of June 2000,
EPA had not proposed or finalized other health- based standards for criteria
pollutants since 1997. Similarly, we reviewed EPA's proposed rule to revise
standards for arsenic in drinking water under the Safe Drinking Water Act,
issued for comment in June 2000, because it incorporates new approaches to
setting health- based standards and estimating benefits reflecting the

requirements of the 1996 amendments to the Safe Drinking Water Act. This
rule incorporates the general approach the agency plans to use in setting
health- based standards under the act.

Although these regulations provide examples of the differing extent to which
EPA may use precautionary assumptions in risk assessments and benefits
estimates, the results may not be generalizable to all of EPA's major
regulations setting health- based standards. In addition, our review of

the benefits estimates is limited to examining the extent to which EPA used
precautionary assumptions in assessing health risks and in estimating the
benefits of the proposed or final health- based standards. As a result, we
did not review the economic aspects of EPA's benefits estimates, such as the
methods used to estimate dollar values for lives saved and illnesses
avoided. Finally, we did not assess the extent to which EPA's benefits
estimates comply with the agency's and the Office of Management and Budget's
guidance for preparing economic analyses of significant regulatory actions,
which include guidance on the treatment of risk and uncertainty and
reporting the “best” or most likely estimate.

Methodology To determine the key factors that affect EPA's use of
precautionary assumptions in assessing health risks, we reviewed EPA
statutes that address health- based standards, EPA policy guidance on risk
assessment and risk characterization, and relevant reports by the National
Academy of Science's National Research Council and the Presidential/
Congressional Commission on Risk Assessment and Risk Management. We also
interviewed officials in EPA's Offices of Policy, Economics, and Innovation;
Water; Air and Radiation; and Research and Development. To obtain
perspectives on the use of precautionary assumptions in health risk
assessment, we interviewed experts from the Health Effects Institute, the
Harvard School of Public Health, Resources for the Future, and an official
of the American Industrial Health Council. We also reviewed a variety of

scientific articles on health- based risk assessment. Finally, we attended
risk assessment and environmental regulation seminars at the Harvard School
of Public Health's Center for Risk Analysis, from which we obtained
information incorporated into this report.

To determine how precautionary assumptions affected the health- based
standards and the estimated benefits of recent major regulations and whether
EPA clearly identified these assumptions, we reviewed (1) the proposed and
final rules for particulate matter, the quantitative health risk assessment,
the regulatory impact analysis, and other key documents,

including EPA's criteria document on particulate matter and recent judicial
findings on the particulate matter rule and (2) the proposed rule for
arsenic, regulatory impact analysis, and other supporting documents,
including the 1999 report, Arsenic in Drinking Water, by the National
Research Council that EPA relied on for its risk assessment. We attended

the June 2000 meeting of EPA's Science Advisory Board on the proposed
arsenic regulation. We met with officials in the air and water program
offices to obtain, among other information, their views on the use of
precautionary assumptions in these rules.

Dose- Response Relationships Based on

Appendi x II

Varying Assumptions In assessing cancer risks, EPA generally assumes a
linear, no- threshold relationship between the dose of a contaminant and its
health effect at low doses. This assumption posits that (1) dose and health
effect vary linearly, that is, if a dose is doubled, the health effect
doubles; and (2) there is no safe level, or threshold, of exposure to a
carcinogen- that is, any exposure presents some risk of developing cancer.
However, for many contaminants, a threshold and/ or a sublinear dose-
response relationship is believed to be plausible. Both sublinear and
threshold dose- response relationships typically lead to lower risk
estimates at the lowest dose levels than does a

linear, no- threshold dose- response relationship. 1 These varying
doseresponse relationships are shown in figure 3.

Figure 3: Dose- Response Relationships Based on Varying Assumptions

Increasing response (health risk) Increasing dose (exposure)

Linear, no- threshold Threshold Sublinear

1 According to EPA officials, some contaminants, such as vinyl chloride,
pose a higher risk at low doses than that predicted by a linear dose-
response relationship. The dose- response relationship for such contaminants
is referred to as supralinear.

Summary of EPA's 1997 Particulate Matter

Appendi x II I Standards In July 1997, EPA established its most recent
National Ambient Air Quality Standards for particulate matter (commonly
called soot), which included new standards for the fine fraction of
particulate matter, also known as fine particulate matter. 1 The standards
are for annual and daily concentrations of particulate matter measured at
various locations across the United States. The annual standards are
intended to provide protection against typical day- to- day exposures as
well as longer- term exposures. The daily

standards are intended to provide protection against days with high peak
concentrations of particulate matter, localized “hot spots,” and
risks from seasonal emissions that would not be well controlled by national
annual standards.

Table 1: 1997 Particulate Matter Standards

Standards in micrograms per cubic meter

Type of particulate matter Annual standard 24- hour (daily) standard

Particulate matter (PM10) a 50 b 150 c Fine particulate matter

15 d 65 e (PM2. 5) a Includes both coarse and fine particles.

b Based on the 3- year average of annual arithmetic mean concentrations of
PM10 at specified monitors. c Based on the 3- year average of the 99th
percentile of 24- hour concentrations of PM10 at specified monitors. d Based
on the 3- year average of annual arithmetic mean concentrations of PM2.5 at
specified

monitors. e Based on the 3- year average of the 98th percentile of 24- hour
concentrations of PM2.5 at specified

monitors.

1 These are the primary standards, intended to protect the public health.
The Clean Air Act also calls for secondary standards, set at a level to
protect the public welfare, e. g., to address effects on visibility,
vegetation, and wildlife.

Issues Associated With EPA's Setting of the

Appendi x V I 1997 Particulate Matter Standards Although EPA develops what
it considers best estimates rather than precautionary estimates of the risks
from particulate matter, the Clean Air Act requires that the health- based
standards be set with an “adequate margin of safety.” According
to EPA, the margin of safety was incorporated into the allowable
concentration levels, or standards, chosen by the

Administrator. Under EPA's interpretation of the Clean Air Act, the agency
is not required to set air quality standards at a zero- risk level to
achieve an adequate margin of safety, but simply at a level that avoids
unacceptable risks. For example, to build in this margin of safety for the
standard for the allowable annual concentration of particulate matter (also
called the

“annual standard”), the Administrator chose 15 micrograms of
particulate matter per cubic meter of air. This concentration was somewhat
lower- that is, it was more protective of health- than the concentration at
which the epidemiological evidence of adverse health effects, particularly
mortality, was most consistent and

coherent. For example, several key epidemiological studies reported
statistically significant associations between mortality and illness and
annual particulate matter concentrations of about 16 to about 21 micrograms
per cubic meter of air. According to EPA, even differences of 1 microgram
per cubic meter of air in the annual concentration in this range,

such as the difference between 16 and 15, can significantly affect health
risks. The Administrator believed that an annual standard of 15 micrograms
per cubic meter of air would provide an adequate margin of safety against
the health problems reported in the scientific literature. Although some

studies did show health effects at lower levels, EPA believed that the
scientific uncertainties about the health effects at such levels were too
great to support standards at those concentrations.

As soon as the particulate matter standards were promulgated in July 1997,
they were challenged in court by industry representatives, small businesses,
and some states, which were concerned about the potential economic impact of
the stricter standards, among other things. Many of

these parties had fought the issuance of the standards. Critics had argued,
for example, that the costs of implementing the standards could run as high
as $46 billion per year and would cause serious financial harm to key
segments of the U. S. economy without providing significant health benefits.
The parties challenged the rule on a variety of grounds, including questions
about the scientific uncertainties associated with the health

effects of particulate matter. In May 1999, a U. S. Court of Appeals
remanded the particulate matter standards, as well as the ozone standards
issued at the same time, to EPA for further consideration. 1 EPA appealed
this ruling to the Supreme Court, and in May 2000, the Court agreed to
review the case. Oral arguments in the case are scheduled for November 2000.

A key concern expressed by the appeals court in its ruling remanding the
standards dealt with EPA's rationale for setting the allowable annual
concentration of particulate matter at 15 micrograms per cubic meter of air.
The court found that the factors EPA used in determining the degree of

public health concerns associated with different concentrations of
particulate matter were reasonable, but that neither the Clean Air Act nor
EPA had articulated an “intelligible principle,” or definitive
criterion, to direct how the agency would apply these factors. Specifically,
for particulate matter, the court indicated that EPA had not adequately
articulated how a standard of 15 micrograms per cubic meter could meet the
Clean Air Act's requirement to set National Ambient Air Quality

Standards with a margin of safety when it is also EPA's judgment that
particulate matter poses health risks at any level above zero. Because no
intelligible principle was articulated, the court held that EPA's
construction of the Clean Air Act in setting the particulate matter and
ozone standards resulted in an unconstitutional delegation of legislative
authority. Shortly after the Supreme Court agreed to review the case, it
also agreed to consider an industry petition arguing that EPA could provide
the intelligible principle needed for setting the air quality standards, at
least in part, by

taking the costs of implementing the regulations into consideration when
setting the standards. While the judicial review of the 1997 particulate 1
American Trucking Associations v. EPA, 175 F. 3d 1027 (D. C. Cir. 1999).

matter standards proceeds, the next 5- year mandated review of the
standards, expected to be completed in 2002, is also under way. (160509)
Lett er

GAO United States General Accounting Office

Page 1 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Contents

Page 2 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 3 GAO- 01- 55 EPA's Use of Precautionary Assumptions United States
General Accounting Office

Washington, D. C. 20548 Page 3 GAO- 01- 55 EPA's Use of Precautionary
Assumptions

Page 4 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 5 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 6 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 7 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 8 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 9 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 10 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 11 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 12 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 13 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 14 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 15 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 16 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 17 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 18 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 19 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 20 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 21 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 22 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 23 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 24 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 25 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 26 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 27 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 28 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 29 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 30 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 31 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 32 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 33 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 34 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix I

Appendix I Objectives, Scope, and Methodology

Page 35 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix I Objectives, Scope, and Methodology

Page 36 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Page 37 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix II

Page 38 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix III

Page 39 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix IV

Appendix IV Issues Associated With EPA's Setting of the 1997 Particulate
Matter Standards

Page 40 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Appendix IV Issues Associated With EPA's Setting of the 1997 Particulate
Matter Standards

Page 41 GAO- 01- 55 EPA's Use of Precautionary Assumptions

Ordering Information The first copy of each GAO report is free. Additional
copies of reports are $2 each. A check or money order should be made out to

the Superintendent of Documents. VISA and MasterCard credit cards are
accepted, also. Orders for 100 or more copies to be mailed to a single
address are discounted 25 percent.

Orders by mail: U. S. General Accounting Office P. O. Box 37050 Washington,
DC 20013

Orders by visiting: Room 1100 700 4th St. NW (corner of 4th and G Sts. NW)
U. S. General Accounting Office Washington, DC

Orders by phone: (202) 512- 6000 fax: (202) 512- 6061 TDD (202) 512- 2537

Each day, GAO issues a list of newly available reports and testimony. To
receive facsimile copies of the daily list or any list from the past 30
days, please call (202) 512- 6000 using a touchtone phone. A recorded menu
will provide information on how to obtain these lists.

Orders by Internet: For information on how to access GAO reports on the
Internet, send an e- mail message with “info” in the body to:
info@ www. gao. gov or visit GAO's World Wide Web home page at: http:// www.
gao. gov

To Report Fraud, Waste, or Abuse in Federal Programs

Contact one: Web site: http:// www. gao. gov/ fraudnet/ fraudnet. htm e-
mail: fraudnet@ gao. gov 1- 800- 424- 5454 (automated answering system)

United States General Accounting Office Washington, D. C. 20548- 0001

Official Business Penalty for Private Use $300

Address Correction Requested Bulk Rate

Postage & Fees Paid GAO Permit No. GI00
*** End of document. ***