CHAPTER 14

RISK ASSESSMENT

C.P. Gerba

Structure of risk analysis.

212
 C. P. Gerba 213

14.1 THE CONCEPT OF RISK ASSESSMENT inant, say, lead or carbon tetrachloride, and documenting
its toxic effects on humans.
Risk, which is common to all life, is an inherent property of • Exposure assessment—Determining the concentration
everyday human existence. It is therefore a key factor in all of a contaminating agent in the environment and estimat-
decision making. Risk assessment or analysis, however, ing its rate of intake in target organisms. An example
means different things to different people: Wall Street ana- would be finding the concentration of aflatoxin in peanut
lysts assess financial risks and insurance companies calculate butter and determining the dose an “average” person
actuarial risks, while regulatory agencies estimate the risks of would receive.
fatalities from nuclear plant accidents, the incidence of can- • Dose–response assessment—Quantifying the adverse
cer from industrial emissions, and habitat loss associated with effects arising from exposure to a hazardous agent based
increases in human populations. What all these seemingly on the degree of exposure. This assessment is usually ex-
disparate activities have in common is the concept of a mea- pressed mathematically as a plot showing the response in
surable phenomenon called risk that can be expressed in living organisms to increasing doses of the agent.
terms of probability. Thus, we can define risk assessment as • Risk characterization—Estimating the potential impact
the process of estimating both the probability that an event of a hazard based on the severity of its effects and the
will occur, and the probable magnitude of its adverse amount of exposure.
effects—economic, health/safety-related, or ecological—
Once the risks are characterized, various regulatory op-
over a specified time period. For example, we might deter-
tions are evaluated in a process called risk management,
mine the probability that a chemical reactor will fail and the
which includes consideration of social, political, and eco-
probable effect of its sudden release of contents on the im-
nomic issues, as well as the engineering problems inherent in
mediate area in terms of injuries and property loss over a pe-
a proposed solution. One important component of risk man-
riod of days. In addition, we might estimate the probable in-
agement is risk communication, which is the interactive
cidence of cancer in the community where the chemical was
process of information and opinion exchange among individ-
spilled. Or, in yet another type of risk assessment, we might
uals, groups, and institutions. Risk communication includes
calculate the health risks associated with the presence of
the transfer of risk information from expert to nonexpert au-
pathogens in drinking water or pesticide in food.
diences. In order to be effective, risk communication must
There are, of course, several varieties of risk assessment.
provide a forum for balanced discussions of the nature of the
Risk assessment as a formal discipline emerged in the 1940s
risk, lending a perspective that allows the benefits of reduc-
and 1950s, paralleling the rise of the nuclear industry. Safety-
ing the risk to be weighed against the costs.
hazard analyses have been used since at least the 1950s in the
In the United States, the passage of federal and state
nuclear, petroleum-refining, and chemical-processing indus-
laws to protect public health and the environment has ex-
tries, as well as in aerospace. Health-risk assessments, how-
panded the application of risk assessment. Major federal
ever, had their beginnings in 1976 with the EPA’s publication
agencies that routinely use risk analysis include the Food and
of the Carcinogenic Risk Assessment Guidelines.
Drug Administration (FDA), the Environmental Protection
In this chapter, we are concerned with two types of risk
Agency (EPA), and the Occupational Safety and Health Ad-
assessment:
ministration (OSHA). Together with state agencies, these
• Health-based risks. For these risks, the focus is on gen- regulatory agencies use risk assessment in a variety of situa-
eral human health, mainly outside the workplace. Health- tions (Information Box 14.1).
based risks typically involve high-probability, low- Risk assessment provides an effective framework for de-
consequence, chronic exposures whose long latency termining the relative urgency of problems and the allocation
periods and delayed effects make cause-and-effect of resources to reduce risks. Using the results of risk analyses,
relationships difficult to establish. This category also we can target prevention, remediation, and control efforts
includes microbial risks, which usually have acute short- toward areas, sources, or situations in which the greatest risk
term effects. However, the consequences of microbial reductions can be achieved with the resources available. How-
infection can persist throughout an individual’s lifetime. ever, risk assessment is not an absolute procedure carried out
in a vacuum; rather, it is an evaluative, multifaceted, compara-
• Ecological risks. For these risks, the focus is on the
tive process. Thus, to evaluate risk, we must inevitably com-
myriad interactions among populations, communities,
and ecosystems (including food chains) at both the micro pare one risk to a host of others. In fact, the comparison of
and the macro level. Ecological risks typically involve potential risks associated with several problems or issues has
both short-term catastrophes, such as oil spills, and long- developed into a subset of risk assessment called comparative
term exposures to hazardous substances. risk assessment. Some commonplace risks are shown in Table
14.1. Here we see, for example, that risks from chemical expo-
Whatever its focus, the risk assessment process consists of sure are fairly small relative to those associated with driving a
four basic steps: car or smoking cigarettes.
• Hazard identification—Defining the hazard and nature Comparing different risks allows us to comprehend
of the harm; for example, identifying a chemical contam- the uncommon magnitudes involved and to understand the
214 Chapter 14 • Risk Assessment

TABLE 14.1 Examples of some commonplace risks in the

United States.*
INFORMATION BOX 14.1 RISK LIFETIME RISK OF MORTALITY
Cancer from cigarette smoking 1:4
Applications of Risk Assessment
(one pack per day)
• Setting standards for concentrations of toxic chemicals Death in a motor vehicle 2:100
or pathogenic microorganisms in water or food. accident
• Conducting baseline analyses of contaminated sites or Homicide 1:100
facilities to determine the need for remedial action and Home accident deaths 1:100
the extent of cleanup required. Cancer from exposure to 3:1,000
• Performing cost/benefit analyses of contaminated-site radon in homes
cleanup or treatment options (including treatment pro- Exposure to the pesticide 6:10,000
cesses to reduce exposure to pathogens). aflatoxin in peanut butter
• Developing cleanup goals for contaminants for which Diarrhea from rotavirus 1:10,000
no federal or state authorities have promulgated Exposure to typical EPA 1:10,000–1:10,000,000
numerical standards; evaluating acceptable variance maximum chemical
from promulgated standards and guidelines (e.g., contaminant levels
approving alternative concentration limits). *Based on data in Wilson and Crouch (1987) and Gerba and Rose (1992).
• Constructing “what-if” scenarios to compare the From Pollution Science © 1996, Academic Press, San Diego, CA.
potential impact of remedial or treatment alternatives
and to set priorities for corrective action.
• Evaluating existing and new technologies for effective are widely perceived as qualitatively different. Rather, we
prevention, control, or mitigation of hazards and risks.
• Articulating community public health concerns and must take into account certain qualitative factors that affect
developing consistent public health expectations risk perception and evaluation when selecting risks to be
among different localities. compared. Some of these qualifying factors are listed in
Table 14.2. We must also understand the underlying
premise that voluntary risk is always more acceptable than
involuntary risk. For example, the same people who cheer-
level, or magnitude, of risk associated with a particular haz- fully drive their cars every day—thus incurring a 2:100
ard. But comparison with other risks cannot itself establish lifetime risk of death by automobile—are quite capable of
the acceptability of a risk. Thus, the fact that the chance of refusing to accept the 6:10,000 involuntary risk of eating
death from a previously unknown risk is about the same as peanut butter contaminated with aflatoxin.
that from a known risk does not necessarily imply that the In considering risk, then, we must also understand an-
two risks are equally acceptable. Generally, comparing other principle—the de minimis principle, which means that
risks along a single dimension is not helpful when the risks there are some levels of risk so trivial that they are not worth

TABLE 14.2 Factors affecting risk perception and risk analysis.

FACTOR CONDITIONS ASSOCIATED WITH CONDITIONS ASSOCIATED WITH
INCREASED PUBLIC CONCERN DECREASED PUBLIC CONCERN
Catastrophic potential Fatalities and injuries grouped in time and space Fatalities and injuries scattered and random
Familiarity Unfamiliar Familiar
Understanding Mechanisms or process not understood Mechanisms or process understood
Controllability (personal) Uncontrollable Controllable
Voluntariness of exposure Involuntary Voluntary
Effects on children Children specifically at risk Children not specifically at risk
Effects manifestation Delayed effects Immediate effects
Effects on future generations Risk to future generations No risk to future generations
Victim identity Identifiable victims Statistical victims
Dread Effects dreaded Effects not dreaded
Trust in institutions Lack of trust in responsible institutions Trust in responsible institutions
Media attention Much media attention Little media attention
Accident history Major and sometimes minor accidents No major or minor accidents
Equity Inequitable distribution of risks and benefits Equitable distribution of risks and benefits
Benefits Unclear benefits Clear benefits
Reversibility Effects irreversible Effects reversible
Origin Caused by human actions or failures Caused by acts of nature
Source: Covello et al. (1988). From Pollution Science © 1996, Academic Press, San Diego, CA.
 C. P. Gerba 215

bothering about. However attractive, this concept is difficult • Inadequate evidence

to define, especially if we are trying to find a de minimis
level acceptable to an entire society. Understandably, • No data available
regulatory authorities are reluctant to be explicit about an • No evidence of carcinogenicity
“acceptable” risk. (How much aflatoxin would you consider The available information on animal and human studies
acceptable in your peanut butter and jelly sandwich? How is then combined into a weight-of-evidence classification
many dead insect parts?) But it is generally agreed that a life- scheme to assess the likelihood of carcinogenicity. This
time risk on the order of one in a million (or in the range of scheme—which is like that developed by the EPA—gives
106 to 104) is trivial enough to be acceptable for the gen- more weight to human than to animal evidence (when it is
eral public. Although the origins and precise meaning of a available) and includes several groupings (Table 14.3).
one-in-a-million acceptable risk remain obscure, its impact Clinical studies of disease can be used to identify very
on product choices, operations, and costs is very real— large risks (between 1/10 and 1/100), most epidemiological
running, for example, into hundreds of billions of dollars in studies can detect risks down to 1/1,000, and very large epi-
hazardous waste site cleanup decisions alone. The levels of demiological studies can examine risks in the 1/10,000
acceptable risk can vary within this range. Levels of risk at range. However, risks lower than 1/10,000 cannot be studied
the higher end of the range (104 rather than 106) may be with much certainty using epidemiological approaches.
acceptable if just a few people are exposed rather than the Since regulatory policy objectives generally strive to limit
entire populace. For example, workers dealing with food ad- risks below 1/100,000 for life-threatening diseases like can-
ditives can often tolerate higher levels of risk than can the cer, these lower risks are often estimated by extrapolating
public at large. These higher levels are justified because from the effects of high doses given to animals.
workers tend to be a relatively homogeneous, healthy group
and because employment is voluntary; however, the sum
14.2.2 Exposure Assessment
level of risks would not be acceptable for those same food
additives in general. Exposure assessment is the process of measuring or estimat-
ing the intensity, frequency, and duration of human exposure
to an environmental agent. Exposure to contaminants can oc-
14.2 THE PROCESS OF RISK ASSESSMENT
cur via inhalation, ingestion of water or food, or absorption
through the skin upon dermal contact. Contaminant sources,
14.2.1 Hazard Identification release mechanisms, transport, and transformation charac-
teristics are all important aspects of exposure assessment, as
The first step in risk assessment is to determine the nature of
are the nature, location, and activity patterns of the exposed
the hazard. For pollution-related problems, the hazard in
population. This explains why it is critical to understand the
question is usually a specific chemical, a physical agent
factors and processes influencing the transport and fate of a
(such as irradiation), or a microorganism identified with a
contaminant (see Chapters 6, 7, 8, and 17).
specific illness or disease. Thus the hazard identification
An exposure pathway is the course that a hazardous
component of a pollution risk assessment consists of a re-
agent takes from a source to a receptor (e.g., human or ani-
view of all relevant biological and chemical information
mal) via environmental carriers or media—generally, air
bearing on whether or not an agent poses a specific threat.
(volatile compounds, particulates) or water (soluble
For example, in the Guidelines for Carcinogen Risk Assess-
compounds) (Figure 14.1). An exception is electromagnetic
ment (U.S. EPA, 1986), the following information is evalu-
radiation, which needs no medium (see Chapter 21). The
ated for a potential carcinogen:
exposure route, or intake pathway, is the mechanism by
• Physical/chemical properties, routes, and patterns of which the transfer occurs—usually by inhalation, ingestion,
exposure and/or dermal contact. Direct contact can result in a local ef-
• Structure/activity relationships of the substance fect at the point of entry and/or in a systemic effect.
• Absorption, distribution, metabolism, and excretion
characteristics of the substance in the body
• The influence of other toxicological effects TABLE 14.3 EPA categories for carcinogenic groups.
• Data from short-term tests in living organisms CLASS DESCRIPTION
• Data from long-term animal studies A Human carcinogen
• Data from human studies B Probable carcinogen
B1 Linked human data
Once these data are reviewed, the animal and human B2 No evidence in humans
data are both separated into groups characterized by degree C Possible carcinogen
of evidence: D No classification
• Sufficient evidence of carcinogenicity E No evidence

• Limited evidence of carcinogenicity From U.S. EPA, 1986.

216 Chapter 14 • Risk Assessment

Figure 14.1 Exposure pathways for potential contaminants.

Modified from Straub et al., 1993.
Figure 14.2 Average tap water ingestion rates in the United
States by age. From Roseberry and Burmaster, 1992.
The quantification of exposure, intake, or potential dose
can involve equations with three sets of variables:
• Concentrations of chemicals or microbes in the media water, we first have to determine the average daily consump-
tion of that water. But this isn’t as easy as it sounds. Studies
• Exposure rates (magnitude, frequency, duration) have shown that daily fluid intake varies greatly from indi-
• Quantified biological characteristics of receptors (e.g., vidual to individual. Moreover, tap water intake depends on
body weight, absorption capacity for chemicals; level of how much fluid is consumed as tap water, and how much is
immunity to microbial pathogens) ingested in the form of soft drinks and other non-tap-water
Exposure concentrations are derived from measured sources. Tap water intake also changes significantly with age
and/or modeled data. Ideally, exposure concentrations (Figure 14.2), body weight, diet, and climate. Because these
should be measured at the points of contact between the factors are so variable, the EPA has suggested a number of
environmental media and current or potential receptors. It very conservative “default” exposure values that can be used
is usually possible to identify potential receptors and when assessing contaminants in tap water, vegetables, soil,
exposure points from field observations and other infor- and the like (Table 14.4).
mation. However, it is seldom possible to anticipate all One important route of exposure is the food supply. Toxic
potential exposure points and measure all environmental substances are often bioaccumulated, or concentrated, in plant
concentrations under all conditions. In practice, a combi- and animal tissues, thereby exposing humans who ingest those
nation of monitoring and modeling data, together with a tissues as food. Moreover, many toxic substances tend to be
great deal of professional judgment, is required to esti- biomagnified in the food chain, so that animal tissues contain
mate exposure concentrations. relatively high concentrations of toxins. Take fish, for exam-
In order to assess exposure rates via different exposure ple. It is relatively straightforward to estimate concentrations
pathways, we have to consider and weigh many factors. For of contaminants in water. Thus, we can use a bioconcentration
example, in estimating exposure to a substance via drinking factor (BCF) to estimate the tendency for a substance in water

TABLE 14.4 EPA standard default exposure factors.

LAND USE EXPOSURE PATHWAY DAILY INTAKE EXPOSURE FREQUENCY EXPOSURE DURATION
(DAYS/YEAR) (YEARS)
Residential Ingestion of potable water 2 L day1 350 30
Ingestion of soil and dust 200 mg (child) 350 6
100 mg (adult) 24
Inhalation of contaminants 20 m3 (total) 350 30
15 m3 (indoor)
Industrial and Ingestion of potable water 1 liter 250 25
commercial Ingestion of soil and dust 50 mg 250 25
Inhalation of contaminants 20 m3 (workday) 250 25
Agricultural Consumption of homegrown produce 42 g (fruit) 350 30
80 g (vegetable)
Recreational Consumption of locally caught fish 54 g 350 30
Modified from Kolluru (1993). From Pollution Science ©1996, Academic Press, San Diego, CA.
 C. P. Gerba 217

to accumulate in fish tissue. The concentration of a chemical in

fish can be estimated by multiplying its concentration in water
by the BCF. The greater the value of the BCF, the more the
chemical accumulates in the fish and the higher the risk of

d
ol
exposure to humans.

sh
re
The units of BCF—liters per kilogram (L kg1) —are

th
o
N
chosen to allow the concentration of a chemical to be

Risk
expressed as milligrams per liter (mg L1) of water and
the concentration in fish to be in milligrams per kilogram (mg
kg1) of fish body weight. In Table 14.5, we see the BCFs of
several common organic and inorganic chemicals. Note the
high values of BCF for the chlorinated hydrocarbon pesti-
Threshold
cides like dichlorodiphenyltrichloroethane (DDT) and poly-
chlorinated biphenyls (PCBs). This exemplifies the concern
we have with such compounds, as discussed in Chapter 10. Dose

14.2.3 Dose–Response Assessment Figure 14.3 Relationship between a threshold and nonthresh-
old response.
Chemicals and other contaminants are not equal in their ca-
pacity to cause adverse effects. To determine the capacity of
agents to cause harm, we need quantitative toxicity data. Some The goal of a dose–response assessment is to obtain a
toxicity data are derived from occupational, clinical, and epi- mathematical relationship between the amount (concentra-
demiological studies. Most toxicity data, however, come from tion) of a toxicant or microorganism to which a human is
animal experiments in which researchers expose laboratory exposed and the risk of an adverse outcome from that dose.
animals, mostly mice and rats, to increasingly higher con- The data resulting from experimental studies is presented
centrations or doses and observe their corresponding effects. as a dose–response curve, as shown in Figure 14.3. The
The result of these experiments is the dose–response relation- abscissa describes the dose, while the ordinate measures the
ship—a quantitative relationship that indicates the agent’s risk that some adverse health effect will occur. In the case of
degree of toxicity to exposed species. Dose is normalized as a pathogen, for instance, the ordinate may represent the risk
milligrams of substance or pathogen ingested, inhaled, or ab- of infection, and not necessarily illness.
sorbed (in the case of chemicals) through the skin per kilogram Dose–response curves derived from animal studies must
of body weight per day (mg kg1 day1). Responses or effects be interpreted with care. The data for these curves are neces-
can vary widely—from no observable effect, to temporary and sarily obtained by examining the effects of large doses on test
reversible effects (e.g., enzyme depression caused by some animals. Because of the costs involved, researchers are limited
pesticides or diarrhea caused by viruses), to permanent organ in the numbers of test animals they can use—it is both im-
injury (e.g., liver and kidney damage caused by chlorinated sol- practical and cost-prohibitive to use thousands (even millions)
vents, heavy metals, or viruses), to chronic functional impair- of animals to observe just a few individuals that show adverse
ment (e.g., bronchitis or emphysema arising from smoke dam- effects at low doses (e.g., risks of 1:1,000 or 1:10,000). Re-
age), to death. searchers must therefore extrapolate low-dose responses from
their high-dose data. And therein lies the rub: Dose–response
curves are subject to controversy because their results change
TABLE 14.5 Bioconcentration factors (BCFs) for various depending on the method chosen to extrapolate from the high
organic and inorganic compounds. doses actually administered to laboratory test subjects to the
low doses humans are likely to receive in the course of every-
CHEMICAL BCF (L kg1)
day living.
Aldrin 28
This controversy revolves around the choice of several
Benzene 44
mathematical models that have been proposed for extrapola-
Cadmium 81
Chlordane 14,000 tion to low doses. Unfortunately, no model can be proved or
Chloroform 3.75 disproved from the data, so there is no way to know which
Copper 200 model is the most accurate. The choice of models is therefore
DDT 54,000 strictly a policy decision, which is usually based on
Formaldehyde 0 understandably conservative assumptions. Thus, for noncar-
Nickel 47 cinogenic chemical responses, the assumption is that some
PCBs 100,000 threshold exists below which there is no toxic response; that is,
Trichloroethylene 10.6 no adverse effects will occur below some very low dose (say,
Vinyl chloride 1.17 one in a million) (Figure 14.3). Carcinogens, however, are con-
From U.S. EPA, 1990. sidered nonthreshold—that is, the conservative assumption is
218 Chapter 14 • Risk Assessment

TABLE 14.6 Primary models used for assessment of TABLE 14.7 Lifetime risks of cancer derived from different
nonthreshold effects. extrapolation models.

MODELa COMMENTS MODEL APPLIED LIFETIME RISK (1.0 mg kg1 day1)

OF TOXIC CHEMICALa
One-hit Assumes (1) a single stage for cancer and
(2) malignant change induced by one One-hit 6.0  105 (1 in 17,000)
molecular or radiation interaction Multistage 6.0  106 (1 in 167,000)
Very conservative Multihit 4.4  107 (1 in 2.3 million)
Linear Assumes multiple stages for cancer Probit 1.9  1010 (1 in 5.3 billion)
multistage Fits curve to the experimental data a
All risks for a full lifetime of daily exposure. The lifetime is used as the
Multihit Assumes several interactions needed before unit of risk measurement, because the experimental data reflect the risk
cell becomes transformed experienced by animals over their full lifetimes. The values shown are
Least conservative model upper confidence limits on risk.
Probit Assumes probit (lognormal) distribution for Source: U.S. EPA, 1990. From Pollution Science © 1996, Academic
tolerances of exposed population Press, San Diego, CA.
Appropriate for acute toxicity; questionable for
cancer
generally much flatter at low doses and consequently predict
a
All these models assume that exposure to the pollutant will always a lower risk than the multistage model.
produce an effect, regardless of dose.
The probit model is not derived from mechanistic as-
Modified from Cockerham and Shane, 1994. From Pollution Science © sumptions about the cancer process. It may be thought of as
1996, Academic Press, San Diego, CA.
representing distributions of tolerances to carcinogens in a
large population. The model assumes that the probability of
the response (cancer) is a linear function of the log of the
that exposure to any amount of carcinogen creates some likeli- dose (log normal). While these models may be appropriate
hood of cancer. This means that the only “safe” amount of car- for acute toxicity they are considered questionable for car-
cinogen is zero, so the dose–response plot is required to go cinogens. These models would predict the lowest level of
through the origin (0), as shown in Figure 14.3. risk of all the models.
There are many mathematical models to choose from, The effect of models on estimating risk for a given
including the one-hit model, the multistage model, the multi- chemical is shown in Table 14.7 and Figure 14.4. As we can
hit model, and the probit model. The characteristics of these see, the choice of models results in order-of-magnitude dif-
models for nonthreshold effects are listed in Table 14.6. ferences in estimating the risk at low levels of exposure.
The one-hit model is the simplest model of carcino- The linear multistage model, a modified version of the
genesis in which it is assumed: multistage model, is the EPA’s model of choice, because this
agency chooses to err on the side of safety and overempha-
1. That a single chemical “hit,” or exposure, is capable of size risk. This model assumes that there are multiple stages
inducing malignant change (i.e., a single hit causes for cancer (i.e., a series of mutations or biotransformations)
irreversible damage of DNA, leading to tumor develop- involving many carcinogens, co-carcinogens, and promoters
ment). Once the biological target is hit, the process lead- (see Chapter 13) that can best be modeled by a series of
ing to tumor formation continues independently of dose. mathematical functions. At low doses, the slope of the
2. That this change occurs in a single stage.
The multistage model assumes that tumors are the
result of a sequence of biological events, or stages. In sim-
plistic terms, the biological rationale for the multistage
model is that there are a series of biological stages that a
chemical must pass through (e.g., metabolism, covalent
bonding, DNA repair, and so on) without being deactivated,
before the expression of a tumor is possible.
The rate at which the cell passes through one or more of
these stages is a function of the dose rate. The multistage
model also has the desirable feature of producing a linear re-
lationship between risk and dose.
The multihit model assumes that a number of dose-
related hits are needed before a cell becomes malignant. The
most important difference between the multistage and multi-
hit model is that in the multihit model, all hits must result
from the dose, whereas in the multistage model, passage Figure 14.4 Extrapolation of dose–response curves. Adapted
through some of the stages can occur spontaneously. The from U.S. EPA, 1990. From Pollution Science © 1996, Academic Press,
practical implication of this is that the multihit models are San Diego, CA.
 C. P. Gerba 219

TABLE 14.8 Chemical RfDs for chronic noncarcinogenic

effects of selected chemicals.

CHEMICAL RfD (mg kg1 day1)

Acetone 0.1
Cadmium 0.0005
Chloroform 0.01
Methylene chloride 0.06
Phenol 0.04
Polychlorinated biphenyl 0.0001
Toluene 0.3
Xylene 2.0
From U.S. EPA, 1990.

Figure 14.5 Potency factor is the slope of the dose–response In general, substances with relatively high slope factors
curve at low doses. At low doses, the slope of the dose– and low reference doses tend to be associated with higher
response curve produced by the multistage model is called the toxicities. The RfD is obtained by dividing the NOAEL (see
potency factor. It is the risk produced by a lifetime average Chapter 13) by an appropriate uncertainty factor, sometimes
dose of 1 mg kg1 day1. Adapted from U.S. EPA, 1990. From called a safety factor or uncertainty factor. A 10-fold
Pollution Science © 1996, Academic Press, San Diego, CA. uncertainty factor is used to account for differences in sensi-
tivity between the most sensitive individuals in an exposed
human population. These include pregnant women, young
children, and the elderly, who are more sensitive than
dose–response curve produced by the linear multistage “average” people. Another factor of 10 is added when the
model is called the potency factor (PF) or slope factor (SF) NOAEL is based on animal data that are extrapolated to hu-
(Figure 14.5), which is the reciprocal of the concentration of mans. In addition, another factor of 10 is sometimes applied
chemical measured in milligrams per kilogram of animal when questionable or limited human and animal data are
body weight per day, that is, 1/(mg kg1 day1), or the risk available. The general formula for deriving an RfD is
produced by a lifetime average dose (AD) of 1 mg kg1
day1. Thus the dose–response equation for a carcinogen is NOAEL
RfD   (Eq. 14.2)
VF1  VF2 ...  VFn
Lifetime Risk  AD  PF (Eq. 14.1)

The probability of getting cancer (not the probability of

dying of cancer) and the associated dose, consist of an aver-
age taken over an assumed 70-year human lifetime. This
dose is called the lifetime average daily dose or chronic
daily intake.
The dose–response effects for noncarcinogens allow for
the existence of thresholds, that is, a certain quantity of a
substance or dose below which there is no observable toxic
effect (NOAEL; see Chapter 13) by virtue of the body’s
natural repair and detoxifying capacity. If a NOAEL is not
available, a LOAEL (lowest observed adverse effect level)
may be used, which is the lowest observed dose or concen-
tration of a substance at which there is a detectable adverse
health effect. When a LOAEL is used instead of a NOAEL,
an additional uncertainty factor is normally applied. Exam-
ples of toxic substances that have thresholds are heavy met-
als and polychlorinated biphenyls (PCBs). These thresholds
are represented by the reference dose, or RfD, of a sub-
stance, which is the intake or dose of the substance per unit
body weight per day (mg kg1 day1) that is likely to pose
no appreciable risk to human populations, including such
sensitive groups as children (Table 14.8). A dose–response
plot for carcinogens therefore goes through this reference Figure 14.6 Relationships between RfD, NOAEL, and LOAEL
point (Figure 14.6). for noncarcinogens.
220 Chapter 14 • Risk Assessment

where VFi are the uncertainty factors. As the data become TABLE 14.9 Toxicity data for selected potential carcinogens.
more uncertain, higher safety factors are applied. For exam-
CHEMICAL POTENCY FACTOR ORAL ROUTE
ple, if data are available from a high-quality epidemiological (mg kg day1)
study, a simple uncertainty factor of 10 may be used by sim-
Arsenic 1.75
ply dividing the original value for RfD by 10 to arrive at a Benzene 2.9  102
new value of RfD, which reflects the concern for safety. The Carbon tetrachloride 0.13
RfDs of several noncarcinogenic chemicals are shown in Chloroform 6.1  103
Table 14.8. DDT 0.34
The RfD (Figure 14.6) can be used in quantitative risk Dieldrin 30
assessments by using the following relationship: Heptachlor 3.4
Methylene chloride 7.5  103
Risk  PF (CDI  RfD) (Eq. 14.3) Polychlorinated biphenyls 7.7
where CDI is the chronic daily intake, and the potency factor (PCBs)
2,3,7,8-TCDD (dioxin) 1.56  105
(PF) is the slope of the dose–response curve. Table 14.9 con-
Tetrachloroethylene 5.1  102
tains potency factors for some potential carcinogens:
Trichloroethylene (TCE) 1.1  102
CDI (mg kg1 day1)  Vinyl chloride 2.3
Average daily dose (mg day1) From U.S. EPA, [Link]/iris.
 (Eq. 14.4)
Body weight (kg)

This type of risk calculation is rarely performed. In most

cases, the RfD is used as a simple indicator of potential risk
in practice. That is, the chronic daily intake is simply com- The mean exposure concentration of contaminants is
pared with the RfD, then, if the CDI is below the RfD, it is used with exposed population variables and the assessment
assumed that the risk is negligible for almost all members of determined variables to estimate contaminant intake. The
an exposed population. general equation for chemical intake is
C  CR  EFD 1
CDI     (Eq. 14.6)
BW AT
14.2.4 Risk Characterization
where:
The final phase of risk assessment process is risk characteri-
zation. In this phase, exposure and dose–response assess- CDI  chronic daily intake; the amount of chemi-
ments are integrated to yield probabilities of effects occurring cal at the exchange boundary (mg/kg-day)
in humans under specific exposure conditions. Quantitative C average exposure concentration over the
risks are calculated for appropriate media and pathways. For period (e.g., mg/L for water or mg/m3 for
example, the risks of lead in water are estimated over a life- air)
time assuming: (1) that the exposure of 2 liters of water per CR  contact rate, the amount of contaminated
day is ingested over a 70-year lifetime; and (2) that different medium contacted per unit time (L/day or
concentrations of lead occur in the drinking water. This m3/day)
information can then be used by risk managers to develop
EFD  exposure frequency and duration, a
standards or guidelines for specific toxic chemicals or infec-
variable that describes how long and how
tious microorganisms in different media, such as the drinking
often exposure occurs. The EFD is
water or food supply.
usually divided into two terms:
EF  exposure frequency (days/year) and
[Link] Cancer risks ED  exposure duration (years)
BW  average body mass over the exposure
If the dose–response curve is assumed to be linear at low
period (kg)
doses for a carcinogen, then:
AT  averaging time; the period over which the
Incremental lifetime risk of cancer  (CDI) (PF) (Eq. 14.5) exposure is averaged (days)
The linearized multistage model assumptions (see Table Determination of accurate intake data is sometimes diffi-
14.6) estimates the risk of getting cancer, which is not nec- cult; for example, exposure frequency and duration vary
essarily the same as the risk of dying of cancer, so it should among individuals and must often be estimated; site-specific
be even more conservative as an upper-bound estimate of information may be available; and professional judgment may
cancer deaths. Potency factors can be found in the EPA be necessary. Equations for estimating daily contamination in-
database on toxic substances called the Integrated Risk In- take rates from drinking water, the air, and contaminated food,
formation System (IRIS) (see Information Box 14.2). Table and for dermal exposure while swimming, have been reported
14.9 contains the potency factor for some of these chemicals. by the EPA. Two of the most common routes of exposure are
 C. P. Gerba 221

INFORMATION BOX 14.2

Integrated Risk Information System (IRIS)
The Integrated Risk Information System (IRIS), prepared and maintained by the U.S. Environmental Protection Agency (U.S.
EPA), is an electronic database containing information on human health effects that may result from exposure to various
chemicals in the environment ([Link]/iris). IRIS was initially developed for EPA staff in response to a growing demand
for consistent information on chemical substances for use in risk assessments, decision-making, and regulatory activities. The
information in IRIS is intended for those without extensive training in toxicology, but with some knowledge of health sciences.
The heart of the IRIS system is its collection of computer files covering individual chemicals. These chemical files contain de-
scriptive and quantitative information in the following categories:
• Oral reference doses and inhalation reference concentrations (RfDs) for chronic noncarcinogenic health effects.
• Hazard identification, oral slope factors, and oral and inhalation unit risks for carcinogenic effects.

Oral RfD Summary for Arsenic.

CRITICAL EFFECT EXPERIMENTAL DOSES* UF RFD

Hyperpigmentation, keratosis and NOAEL: 0.009 mg/L converted to 0.0008 mg/kg-day 3 3E-4 mg/kg-day
possible vascular complications
Human chronic oral exposure LOAEL: 0.17 mg/L converted to 0.014 mg/kg-day
Tseng, 1977; Tseng et al., 1968
*Conversion Factors—NOAEL was based on an arithmetic mean of 0.009 mg/L in a range of arsenic concentration of 0.001 to 0.017 mg/L. This
NOAEL also included estimation of arsenic from food. Since experimental data were missing, arsenic concentrations in sweet potatoes and rice
were estimated as 0.002 mg/day. Other assumptions included consumption of 4.5 L water/day and 55 kg body weight (Abernathy et al., 1989).
NOAEL  [(0.009 mg/L  4.5 L/day)  0.002 mg/day]/55 kg  0.0008 mg/kg-day. The LOAEL dose was estimated using the same assump-
tions as the NOAEL starting with an arithmetic mean water concentration from Tseng (1977) of 0.17 mg/L. LOAEL  [(0.17 mg/L  4.5 L/day)
 0.002 mg/day]/55 kg  0.014 mg/kg-day.
UF  Uncertainty Factor or Safety Factor.

through drinking contaminated water and breathing contami- (90th percentile) at one residence]; 9
nated air. The intake for ingestion of waterborne chemicals is years [national median time (50th per-
CW  IR  EF  ED centile) at one residence]
CDI   BW  AT (Eq. 14.7) BW: 70 kg (adult, average); Age-specific values

where AT: pathway-specific period of exposure for

noncarcinogenic effects (i.e., ED  365
CDI  chronic daily intake by ingestion (mg/kg- days/year), and 70-year lifetime for car-
day) cinogenic effects (i.e., 70 years  365
CW  chemical concentration in water (mg/L) days/year), averaging time.
IR  ingestion rate (L/day)
EF  exposure frequency (days/year)
ED  exposure duration (years)
BW  body weight (kg)
AT  averaging time (period over which the EXAMPLE 14.1
exposure is averaged—days) Estimation of an Oral Chronic Daily Intake

Some of the values used in Equation 14.5 are The mean concentration of 1,2-dichlorobenzene in a water
supply is 1.7 g/L. Determine the chronic daily intake for a
CW: site-specific measured or modeled value 70-kg adult. Assume that 2 L of water are consumed per day.
IR: 2 L/day (adult, 90th percentile); 1.4 L/day Solution
(adult, average) The chronic daily intake (CDI) may be calculated using
Equation 14.7.
EF: pathway-specific value (dependent on the
C  CR  EF  ED
frequency of exposure-related activities) 
CDI  BW  AT
ED: 70 years (lifetime; by convention);
30 years [national upper-bound time
222 Chapter 14 • Risk Assessment

where EXAMPLE 14.2

CDI  chronic daily intake (mg/kg-day) Application of Hazard Index and Incremental
C  1.7 g/L  0.0017 mg/L
CR  2 L/day Carcinogenic Risk Associated with Chemical Exposure
EF  365 days/year
A drinking water supply is found to contain 0.1 mg L1 of
ED  30 years (standard exposure duration for an
acetone and 0.1 mg L1 of chloroform. A 70-kg adult drinks
adult exposed to a noncarcinogenic)
2 L per day of this water for 5 years. What would be the haz-
BW  70 kg
ard index and the carcinogenic risk from drinking this water?
AT  365 days/year  30 years  10,950 days
First, we need to determine the average daily doses
Substituting values into the equation yields the chronic daily (ADDs) for each of the chemicals and then their individual
intake. hazard quotients.
0.0017  2  365  30
 For Acetone
CDI  70  10,950  4.86  105 mg/kg-day
(0.1 mg L1) (2L day1)
ADD  
70 kg
 2.9  103 mg kg1 day1
[Link] Noncancer risks From Table 14.5, the RfD for acetone is 0.1 mg kg1 day1
2.9  103 mg kg1 day1
Noncancer risks are expressed in terms of a hazard quotient Hazard quotient (HQ)  
0.1
(HQ) for a single substance, or hazard index (HI) for multi-
ple substances and/or exposure pathways.  0.029
Hazard quotient (HQ)  For Chloroform
(0.1 mg L1) (2L day1)
Average daily dose during exposure period (Eq. 14.8) ADD   70 kg
(mg kg1 day1)
  2.9  103 mg kg1 day1
RfD (mg kg1 day1
Unlike a carcinogen, the toxicity is important only during the From Table 14.5, the RfD value for chloroform is 0.01 mg
kg1 day1
time of exposure, which may be one day, a few days, or 2.9  103 mg kg1 day1

years. The HQ has been defined so that if it is less than 1.0, HQ  0.01
there should be no significant risk or systemic toxicity. Ra-
tios above 1.0 could represent a potential risk, but there is no  0.029
way to establish that risk with any certainty. Thus,
When exposure involves more than one chemical, the
Hazard index  0.029  0.29  0.319
sum of the individual hazard quotients for each chemical is
used as a measure of the potential for harm. This sum is Since the hazard index is less than 1.0, the water is safe. No-
called the hazard index (HI): tice that we did not need to take into consideration that the
person drank the water for 5 years.
HI  Sum of hazard quotients (Eq. 14.9) The incremental carcinogenic risk associated with chloro-
form is determined as follows.

[Link] Uncertainty analysis Risk  (CDI) (Potency factor)

(0.1 mg L1) (2 L day1 (365 days yr1) (5 yrs)
Uncertainty is inherent in every step of the risk assessment CDI  
(70 kg) (365 days yr1) (70 yrs)
process. Thus, before we can begin to characterize any risk,
we need some idea of the nature and magnitude of uncer-  4.19  105 mg kg1 day1
tainty in the risk estimate. Sources of uncertainty include: From Table 14.6, the potency factor for chloroform is
• Extrapolation from high to low doses 6.1  103
• Extrapolation from animal to human responses Risk  (CDI) (Potency factor)
• Extrapolation from one route of exposure to another
• Limitations of analytical methods Risk  (4.19  105 mg kg1 day1)
• Estimates of exposure (6.1  103 mg kg1 day1)  2.55  2.55  107
Although the uncertainties are generally much larger in esti-
mates of exposure and the relationships between dose and re- From a cancer risk standpoint, the risk over this period of
sponse (e.g., the percent mortality), it is important to include exposure is less than the 106 goal.
the uncertainties originating from all steps in a risk assess-
ment in risk characterization.
 C. P. Gerba 223

Two approaches commonly used to characterize un- ment may result in a high risk due to release of wind-blown
certainty are sensitivity analyses and Monte Carlo simula- dusts that may expose workers at the site (Watts, 1998). In
tions. In sensitivity analyses, we simply vary the uncertain contrast, a soil contaminated with 10 mg of hazardous mate-
quantities of each parameter (e.g., average values, high and rial per kg of soil may be considered a greater risk if the site
low estimates), usually one at a time, to find out how has sandy soil, shallow groundwater, and nearby drinking
changes in these quantities affect the final risk estimate. water wells, and is located near a school. Cleanup to low lev-
This procedure gives us a range of possible values for the els would be necessary in this case to protect human health
overall risk and tells us which parameters are most crucial (Figure 14.7).
in determining the size of the risk. In a Monte Carlo simu-
lation, however, we assume that all parameters are random 14.3 ECOLOGICAL RISK ASSESSMENT
or uncertain.
Thus, instead of varying one parameter at a time, we use Ecological risk assessment is a process that evaluates the prob-
a computer program to select parameter distributions ran- ability that adverse ecological effects will occur as the result
domly every time the model equations are solved, the proce- of exposure to one or more stressors. A stressor (or agent) is
dure being repeated many times. The resulting output can be a substance, circumstance, or energy field that has the inher-
used to identify values of exposure or risk corresponding to ent ability to impose adverse effects upon a biological system.
a specified probability, say, the 50th percentile or 95th per- The environment is subject to many different stressors, in-
centile. cluding chemicals, genetically engineered microorganisms,
ionizing radiation, and rapid changes in temperatures. Ecolog-
[Link] Risk projections and management
The final phase of the risk assessment process is risk
characterization. In this phase, exposure and dose–response
assessments are integrated to yield probabilities of effects
occurring in humans under specific exposure conditions.
Quantitative risks are calculated for appropriate media and
pathways. For example, the risks of lead in water are esti-
Municipal Water
mated over a lifetime, assuming (1) that the exposure is 2 Surface Soil Supply Well
liters of water ingested per day over a 70-year lifetime and Contaminated SCHOOL

(2) that different concentrations of lead occur in the drinking with 10 mg/kg PCB 50 m

water. This information can then be used by risk managers to

develop standards or guidelines for specific toxic chemicals 10m Sandy Soils
to groundwater
or infectious microorganisms in different media, such as the
drinking water or food supply.
Groundwater
Flow
[Link] Hazardous waste risk assessment
Site A-High Risk
Hazardous waste risk assessments are a key part of the Com-
prehensive Environmental Response, Compensation, and Li-
ability Act (CERCLA). Risk assessments are performed to
assess health and ecological risks at Superfund sites and to 20m
evaluate the effectiveness of remedial alternatives in attain- 10,000 mg/kg
PCB
ing a record of decision (ROD). Since specific cleanup re-
quirements have not been established for most contaminants
Clay Subsurface
under CERCLA, each site is assessed on an individual basis Materials
100 m
and cleaned up to a predetermined level of risk, such as 1
cancer case per 1,000,000 people. Risks may be different
from one site to the next, depending on characteristics of the
site and the potential for exposure. Saline Groundwater
For example, at one site, a high level of contaminants
may be present (10,000 mg per kg of soil), but there is no
nearby population, there is a large distance to groundwater, Site B-Low Risk
and the soils are of low permeability. Based on a risk assess- Figure 14.7 Two extremes of potential risk from contaminated
ment, the best remedial action for the site may be to leave the sites. Site A is a high-risk site with potential for migration from
contaminated soil in place, where natural attenuation pro- the source to nearby receptors. Site B, although characterized by
cesses will eventually result in its degradation. Removing the a higher source concentration, has minimal potential for
contaminated soil with disposal in a landfill or in situ treat- contaminant migration and risk. Modified from Watts, 1998.
224 Chapter 14 • Risk Assessment

ical risk assessment may evaluate one or more stressors and

ecological components (e.g., specific organisms, populations,
communities, or ecosystems). Ecological risks may be ex-
pressed as true probabilistic estimates of adverse effects (as is
done with carcinogens in human health risk assessment), or
they may be expressed in a more qualitative manner.
In the United States, the Comprehensive Environmental
Response, Compensation, and Liability Act (CERCLA)
(otherwise known as the Superfund), the Resource Conser-
vation and Recovery Act (RCRA), and other regulations
require an ecological assessment as part of all remedial in-
vestigation and feasibility studies (see also Section [Link]).
Pesticide registration, which is required under the Federal
Insecticide, Fungicide, and Rodenticide Act (FIFRA), must
also include an ecological assessment (see Section 14.3).
In the CERCLA/RCRA context, a typical objective is to
determine and document actual or potential effects of con-
taminants on ecological receptors and habitats as a basis for
evaluating remedial alternatives in a scientifically defensible
manner.
The four major phases or steps in ecological assessment Figure 14.9 Ecological risk assessment.
(Figure 14.8) are as follows:
• Problem formulation and hazard identification
• Exposure assessment process begins with an evaluation of the stressor characteris-
• Ecological effects/toxicity assessment tics, the ecosystem at risk, and the likely ecological effects. An
• Risk characterization endpoint is then selected. An endpoint (Figure 14.9) is a char-
An ecological risk assessment may be initiated under many acteristic of an ecological component, e.g., the mortality of
circumstances—the manufacturing of a new chemical, evalu- fish) that may be affected by a stressor. Two types of endpoints
ation of cleanup options for a contaminated site, or the planned are generally used: assessment endpoints and measurement
filling of a marsh, among others. The problem-formulation endpoints. Assessment endpoints are particular environmen-
tal values to be protected. Such endpoints, which are recog-
nized and valued by the public, drive the decisions made by
official risk managers. Measurement endpoints are qualita-
tively or quantitatively measurable factors. Suppose, for ex-
ample, a community that values the quality of sports fishing in
the area is worried about the effluent from a nearby paper mill.
In this case, a decline in the trout population might serve as the
assessment endpoint, while the increased mortality of min-
nows, as evaluated by laboratory studies, might be the mea-
surement endpoint. Thus, risk managers would use the quanti-
tative data gathered on the surrogate minnow population to
develop management strategies designed to protect the trout
population.
Exposure assessment is a determination of the environ-
mental concentration range of a particular stressor and the ac-
tual dose received by the biota (all the plants and animals) in a
given area. The most common approach to exposure analysis
is to measure actual concentrations of a stressor and combine
these measurements with assumptions about contact and up-
take by the biota. For example, the exposure of simple aquatic
organisms to chemicals can often be measured simply as the
concentration of that chemical in the water because the physi-
ologic systems of these organisms are assumed to be in equi-
librium with the surrounding water. Stressor measurements
Figure 14.8 Framework for ecological risk assessment. can also be combined with quantitative parameters describing
Adapted from U.S. EPA, 1992a. the frequency and magnitude of contact. For example, concen-
 C. P. Gerba 225

trations of chemicals or microorganisms in food items can be trapolated to the community and ecosystem level. One
combined with ingestion rates to estimate dietary exposure. of the difficulties in the quantification of the stressor–re-
Exposure assignment is, however, rarely straightforward. Bio- sponse profile is that many of the quantitative extrapolations
transformations may occur, especially for heavy metals such as are drawn from information that is qualitative in nature. For
mercury (see Section 13.5.5). Such transformations may result example, when we use phylogenic extrapolation to transfer
in the formation of even more toxic forms of the stressor. Re- toxicity data from one species to another species—or even to
searchers must therefore use mathematical models to predict a whole class of organisms—we are assuming a degree of
the fate and resultant exposure to a stressor and to determine similarity based on qualitative characteristics. Thus, when
the outcome of a variety of scenarios. we use green algal toxicity test data to represent all photo-
The purpose of evaluating ecological effects is to iden- synthetic eukaryotes (which we often do), we must remem-
tify and quantify the adverse effects elicited by a stressor ber that all photosynthetic eukaryotes are not, in fact, green
and, to the extent possible, to determine cause-and-effect re- algae. Because many of the responses are extrapolations
lationships. During this phase, toxicity data are usually com- based on models ranging from the molecular to the ecosys-
piled and compared. tem level, it is critically important that uncertainties and as-
sumptions be clearly delineated.
Risk assessment consists of comparing the exposure
EXAMPLE 14.3 and stressor–response profiles to estimate the probability of
Examples of a Management Goal, Assessment Endpoint, effects, given the distribution of the stressor within the sys-
tem. As you might expect, this process is extraordinarily dif-
and Measures
ficult to accomplish. In fact, our efforts at predicting adverse
Goal: Viable, self-sustaining coho salmon population that effects have been likened to the weather forecaster’s predic-
supports a subsistence and sport fishery. tion of rain (Landis and Ho-Yu, 1995). Thus, the predictive
Assessment Endpoint: Coho salmon breeding success, fry process in ecological risk assessment is still very much an art
survival, and adult return rates. form, largely dependent on professional judgment.
Measures of Effects Conceptual model diagrams can be used to better visu-
• Egg and fry response to low dissolved oxygen alize potential impacts (Figure 14.10). They may be based on
• Adult behavior in response to obstacles
theory and logic, empirical data, mathematical models, or
• Spawning behavior and egg survival with changes in
sedimentation
probability models. These diagrams are useful tools for
Measures of Ecosystem and Receptor Characteristics communicating important pathways in a clear and concise
• Water temperature, water velocity, and physical obstruc- way. They can be used to ask new questions about relation-
tions ships that help generate plausible risk hypothesis.
• Abundance and distribution of suitable breeding substrate
• Abundance and distribution of suitable food sources for fry
• Feeding, resting, and breeding behavior 14.4 MICROBIAL RISK ASSESSMENT
• Natural reproduction, growth, and mortality rates
Measures of Exposure Outbreaks of waterborne disease caused by microorganisms
• Number of hydroelectric dams and associated ease of fish usually occur when the water supply has been obviously and
passage
significantly contaminated. In such high-level cases, the ex-
• Toxic chemical concentrations in water, sediment, and
fish tissue
posure is manifest, and cause and effect are relatively easy to
• Nutrient and dissolve oxygen levels in ambient waters determine. However, exposure to low-level microbial con-
• Riparian cover, sediment loading, and water temperature tamination is difficult to determine epidemiologically. We
know, for example, that long-term exposure to microbes can
have a significant impact on the health of individuals within
a community, but we need a way to measure that impact.
Generally, there are acute and chronic data for the stres- For some time, methods have been available to detect
sor acting on one or several species. Field observations can the presence of low levels (1 organism per 1000 liters) of
provide additional data, and so can controlled-microcosm pathogenic organisms in water, including enteric viruses,
and large-scale tests. bacteria, and protozoan parasites. The trouble is that the risks
The process of developing a stressor–response profile is posed to the community by these low levels of pathogens in
complex because it inevitably requires models, assumptions, a water supply over time are not like those posed by low lev-
and extrapolations. For example, the relationship between els of chemical toxins or carcinogens. For example, it takes
measurement and assessment endpoint is an assumption. It is just one amoeba in the wrong place at the wrong time to in-
often expressly stated in the model used, but when it is not fect one individual, whereas that same individual would
specifically stated, it is left to professional judgment. In ad- have to consume some quantity of a toxic chemical to be
dition, the stressor–response profile is analogous to a comparably harmed. Microbial risk assessment is therefore a
dose–response curve in the sense that it involves extrapola- process that allows us to estimate responses in terms of the
tions; in this case, though, a single-species toxicity test is ex- risk of infection in a quantitative fashion. Microbial risk gen-
226 Chapter 14 • Risk Assessment

Source
(e.g., logging plan)

Primary Stressor
(e.g., building
Interaction with logging roads)
ecosystem
(e.g., slope, soil type)
(No exposure of receptor
by this pathway)
X
Secondary Stressor
(e.g., increased
situation of stream
Figure 14.11 Outcomes of enteric viral exposure. From
Exposure of Pollution Science © 1996, Academic Press, San Diego, CA.
receptor Primary Effect
(e.g., smothering of
benthic insects)
others—without ever getting sick themselves. The ratio of
Interspecies interaction clinical to subclinical infection varies from pathogen to
(e.g., food, habitat, pathogen, especially in viruses, as shown in Table 14.10. Po-
competition)
liovirus infections, for instance, seldom result in obvious
clinical symptoms; in fact, the proportion of individuals de-
Secondary (Indirect)
veloping clinical illness may be less than 1%. However,
Effect other enteroviruses, such as the coxsackie viruses, may
(e.g., decreased exhibit a greater proportion. In many cases, as in that of
abundance of
insectivorous fish) rotaviruses, the probability of developing clinical illness
appears to be completely unrelated to the dose an individual
Figure 14.10 Conceptual model for logging. Source: receives via ingestion. Rather, the likelihood of developing
[Link].

TABLE 14.10 Ratio of clinical to subclinical infections with

enteric viruses.
erally follows the steps used in other health-based risk
assessments—hazard identification, exposure assessment, VIRUS FREQUENCY OF CLINICAL ILLNESSa (%)
dose–response, and risk characterization. The differences are Poliovirus 1 0.1–1
in the specific assumptions, models, and extrapolation Coxsackie
methods used. A16 50
Hazard identification in the case of pathogens is com- B2 11–50
plicated because several outcomes—from asymptomatic in- B3 29–96
fection to death (see Figure 14.11) —are possible, and these B4 30–70
outcomes depend upon the complex interaction between B5 5–40
the pathogenic agent (the “infector”) and the host (the Echovirus
Overall 50
“infectee”). This interaction, in turn, depends on the charac-
9 15–60
teristics of the host as well as the nature of the pathogen. 18 Rare–20
Host factors, for example, include preexisting immunity, 20 33
age, nutrition, ability to mount an immune response, and 25 30
other nonspecific host factors. Agent factors include type 30 50
and strain of the organism as well as its capacity to elicit an Hepatitis A (adults) 75
immune response. Rotavirus
Among the various outcomes of infection is the possi- (Adults) 56–60
bility of subclinical infection. Subclinical (asymptomatic) (Children) 28
infections are those in which the infection (growth of the mi- Astrovirus (adults) 12–50
croorganism within the human body) results in no obvious a
The percentage of the individuals infected who develop clinical illness.
illness such as fever, headache, or diarrhea. That is, individ- From Gerba and Rose (1993). From Pollution Science © 1996, Academic
uals can host a pathogen microorganism—and transmit it to Press, San Diego, CA.
 C. P. Gerba 227

clinical illness depends upon the type and strain of the virus TABLE 14.12 Case-fatality rates for enteric viruses and
as well as host age, nonspecific host factors, and possibly bacteria.
preexisting immunity. The incidence of clinical infection can ORGANISM CASE-FATALITY RATE (%)
also vary from year to year for the same virus, depending on
Viruses
the emergence of new strains.
Poliovirus 1 0.90
Another outcome of infection is the development of Coxsackie
clinical illness. Several host factors play a major role in this A2 0.50
outcome. The age of the host is often a determining factor. In A4 0.50
the case of hepatitis A, for example, clinical illness can vary A9 0.26
from about 5% in children less than 5 years of age to 75% in A15 0.12
adults. Similarly, children are more likely to develop rotavi- Cosxsackie B 0.59–0.94
ral gastroenteritis than are adults. Immunity is also an im- Echovirus
portant factor, albeit a variable one. That is, immunity may 6 0.29
or may not provide long-term protection from reinfection, 9 0.27
Hepatitis A 0.30
depending on the enteric pathogen. It does not, for example,
Rotavirus
provide long-term protection against the development of
(Total) 0.01
clinical illness in the case of the Norwalk virus or Giardia. (Hospitalized) 0.12
However, for most enteroviruses and for the hepatitis A Norwalk 0.0001
virus, immunity from reinfection is believed to be lifelong. Astrovirus 0.01
Other undefined host factors may also control the odds of de- Bacteria
veloping illness. For example, in experiments with the Nor- Shigella 0.2
walk virus (norovirus), human volunteers who did not be- Salmonella 0.1
come infected upon an initial exposure to the virus also did Escherichia coli 0157:H7 0.2
not respond to a second exposure. In contrast, those volun- Campylobacter jejuni 0.1
teers who developed gastroenteritis upon the first exposure From Gerba and Rose (1993) and Gerba et al. (1996). From Pollution
also developed illness after the second exposure. Science © 1996, Academic Press, San Diego, CA.
The ultimate outcome of infection—mortality—can be
caused by nearly all enteric organisms. The factors that
control the prospect of mortality are largely the same fac-
tors that control the development of clinical illness. Host fatality rate for common enteric bacteria ranges from 0.1 to
age, for example, is significant. Thus, mortality for hepati- 0.2% in the general population. Enteric bacterial diseases
tis A and poliovirus is greater in adults than in children. In can be treated with antibiotics, but no treatment is available
general, however, one can say that the very young, the el- for enteric viruses.
derly, and the immunocompromised are at the greatest risk Recognizing that microbial risk involves a myriad of
of a fatal outcome of most illnesses (Gerba et al., 1996). pathogenic organisms capable of producing a variety of out-
For example, the case-fatality rate (%) for Salmonella in comes that depend on a number of factors—many of which
the general population is 0.1%, but it has been observed to are undefined—one must now face the problem of exposure
be as high as 3.8% in nursing homes (Table 14.11). In assessment, which has complications of its own. Unlike
North America and Europe, the reported case-fatality rates chemical-contaminated water, microorganism-contaminated
(i.e., the ratio of cases to fatalities reported as a percentage water does not have to be consumed to cause harm. That is,
of persons who die) for enterovirus infections range from individuals who do not actually drink, or even touch, con-
less than 0.1 to 0.94%, as shown in Table 14.12. The case- taminated water also risk infection because pathogens—
particularly viruses—may be spread by person-to-person
contact or subsequent contact with contaminated inanimate
objects (such as toys). This phenomenon is described as the
TABLE 14.11 Case fatality observed for enteric pathogens in secondary attack rate, which is reported as a percentage.
nursing homes versus general population. For example, one person infected with poliovirus can trans-
ORGANISM CASE FATALITY (%) CASE FATALITY (%)
mit it to 90% of the persons with whom he or she associates.
IN GENERAL IN NURSING HOMES This secondary spread of viruses has been well documented
POPULATION for waterborne outbreaks of several diseases, including that
Campylobacter 0.1 1.1 caused by Norwalk virus, whose secondary attack rate is
jejuni about 30%.
Escherichia coli 0.2 11.8 The question of dose is another problem in exposure
0157:H7 assessment. How does one define “dose” in this context?
Salmonella 0.1 3.8 To answer this question, researchers have conducted a
Rotavirus 0.01 1.0 number of studies to determine the infectious dose of en-
Modified from Gerba et al. (1996). teric microorganisms in human volunteers. Such human
228 Chapter 14 • Risk Assessment

experimentation is necessary because determination of the model may be used to describe the probability of infection
infectious dose in animals and extrapolation to humans is in human subjects for many enteric microorganisms (Haas,
often impossible. In some cases, for example, humans are 1983). These models have been found to best fit experi-
the primary or only known host. In other cases, such as that mental data. For the beta-Poisson model, the probability of
of Shigella or norovirus, infection can be induced in labo- infection from a single exposure, P, can be described as
ratory-held primates, but it is not known whether the infec- follows:
tious dose data can be extrapolated to humans. Much of
the existing data on infectious doses of viruses has been P  1  (1  N/) (Eq. 14.10)
obtained with attenuated vaccine viruses or with aviru-
where N is the number of organisms ingested per exposure
lent laboratory-grown strains, so that the likelihood of seri-
and and  represent parameters characterizing the host-
ous illness is minimized. An example of a dose–response
virus interaction (dose–response curve). Some values for
curve for a human feeding study with rotavirus is shown in
and  for several enteric waterborne pathogens are shown in
Figure 14.12.
Table 14.13; these values were determined from human
In the microbiological literature, the term minimum in-
studies. For some microorganisms, an exponential model
fectious dose is used frequently, implying that a threshold
may better represent the probability of infection.
dose exists for microorganisms. In reality, the term used usu-
ally refers to the ID50 dose at which 50% of the animals or P  1  exp(rN) (Eq. 14.11)
humans exposed became infected or exhibit any symptoms
of an illness. Existing infectious dose data are compatible In this equation, r is the fraction of the ingested microorgan-
with nonthreshold responses, and the term “infectivity” is isms that survive to initiate infections (host-microorganism
probably more appropriate when referring to differences in interaction probability). Table 14.13 shows examples of re-
the likelihood of an organism causing an infection. For ex- sults of both models for several organisms.
ample, the probability of a given number of ingested ro- These models define the probability of the microor-
taviruses causing diarrhea is greater than that for Salmonella. ganisms overcoming the host defenses (including stomach
Thus, the infectivity of rotavirus is greater than that of pH, finding a susceptible cell, nonspecific immunity, and
Salmonella. so on) to establish an infection in the host. When one uses
Next, one must choose a dose–response model, whose these models, one estimates the probability of becoming in-
abscissa is the dose and whose ordinate is the risk of infec- fected after ingestion of various concentrations of
tion (see Figure 14.12). The choice of model is critical so pathogens. For example, Example 14.5 shows how to cal-
that risks are not greatly overestimated or underestimated. culate the risk of acquiring a viral infection from consump-
A modified exponential (beta-Poisson distribution) or a log- tion of contaminated drinking water containing echovirus
probit (simple lognormal, or exponential, distribution) 12 using Equation 14.10.

Figure 14.12 Dose–response for human rotavirus by oral ingestion.

 C. P. Gerba 229

TABLE 14.13 Best-fit dose-response parameters for enteric

pathogen ingestion studies. EXAMPLE 14.5
MICROORGANISM BEST MODEL MODEL PARAMETERS Risk Assessment for Rotavirus in Drinking Water
Echovirus 12 Beta-Poisson  0.374 Pathogen identified Rotavirus
  186,69 ↓ ↓
Rotavirus Beta-Poisson  0.26 Dose Response Model best fit for data is the
  0.42 (based on human Beta Poisson Model
Poliovirus 1 Exponential r  0.009102 ingestion studies) P  (1  N/)
Poliovirus 1 Beta-Poisson  0.1097  0.2631
  1524   0.42
Poliovirus 3 Beta-Poisson  0.409 ↓ ↓
  0.788 Exposure (field studies 4 rotavirus/1,000 liters
Cryptosporidium Exponential r  0.004191 on concentration in
Giardia lamblia Exponential r  0.02 drinking water)
Salmonella Exponential r  0.00752 ↓ ↓
Escherichia coli Beta-Poisson  0.1705 Risk Characterization Risk of Infection
  1.61  106 Assumes: 2 liters/day of
Modified from Regli et al. (1991). From Pollution Science © 1996, drinking water ingested.
Academic Press, San Diego, CA. Thus,
N  0.008/day
Risk of Infection/day  1:200
Risk of Infection/year
PA  1  (1  P)365
EXAMPLE 14.4 PA  1:2

Application of a Virus Risk Model to Characterize Risks

from Consuming Shellfish
It is well known that infectious hepatitis and viral gas-
troenteritis are caused by consumption of raw or, in some
cases, cooked clams and oysters. The concentration of
echovirus 12 was found to be 8 plaque-forming units per Annual and lifetime risks can also be determined, again
100 g in oysters collected from coastal New England wa- assuming a Poisson distribution of the virus in the water
ters. What are the risks of becoming infected and ill from consumed (assuming daily exposure to a constant concentra-
echovirus 12 if the oysters are consumed? Assume that a tion of viral contamination), as follows:
person usually consumes 60 g of oyster meat in a single
serving: PA  1  (1  P)365 (Eq. 14.12)

8 PFU N where PA is the annual risk (365 days) of contracting one or

   N  4.8 PFU consumed
100 g 60 g more infections, and
From Table 14.13,  0.374,   186.64. The probability
PL  1  (1  P)25,550 (Eq. 14.13)
of infection from Equation 14.10 is then
where PL is the lifetime risk (assuming a lifetime of 70 years
 
4.8 0.374
P  1  1  ——  9.4  103  25,550 days) of contracting one or more infections.
186.69
Risks of clinical illness and mortality can then be deter-
If the percent of infections that result in risk of clinical ill- mined by incorporating terms for the percentage of clinical
ness is 50%, then from Equation 14.14 one can calculate the illness and mortality associated with each particular virus:
risk of clinical illness:
Risk of clinical illness  PI (Eq. 14.14)
Risk of clinical illness  (9.4  103) (0.50)  4.7  103
Risk of mortality  PIM (Eq. 14.15)
If the case-fatality rate is 0.001%, then from Equation 14.15
where I is the percentage of infections that result in clinical
Risk of mortality  (9.4  103) (0.50) (0.001) illness and M is the percentage of clinical cases that result in
 4.7  106 mortality.
If a person consumes oysters 10 times a year with 4.8 PFU Application of this model allows estimation of the risks
per serving, then one can calculate the risk of infection in of infection, development of clinical illness, and mortality
one year from Equation 14.12 for different levels of exposure. As shown in Table 14.14, for
example, the estimated risk of infection from 1 rotavirus in
100 liters of drinking water (assuming ingestion of 2 liters
per day) is 1.2  103, or almost 1 in 1,000 for a single-day
230 Chapter 14 • Risk Assessment

TABLE 14.14 Risk of infection, disease, and mortality for TABLE 14.15 Comparison of outbreak data to model
rotavirus. predictions for assessment of risks associated
with exposure to Salmonella.
VIRUS RISK
CONCENTRATION FOOD DOSE CFU AMOUNT ATTACK PREDICTED
PER 100 LITERS DAILY ANNUAL CONSUMED RATE (%) P (%)

Infection Water 17 1L 12 12
Pancretin 200 7 doses 100 77
100 9.6  102 1.0 Ice cream 102 1 portion 52 54
1 1.2  103 3.6  101 Cheese 100–500 28 g 28–36 53–98
0.1 1.2  104 4.4  102 Cheese 105 100 g 100 99.99
Disease Ham 106 50–100 g 100 99.99
100 5.3  102 5.3  101 Source: Rose et al., 1995. From Environmental Microbiology © 2000,
1 6.6  104 2.0  101 Academic Press, San Diego, CA.
0.1 6.6  105 2.5  102
Mortality source in terms of the concentration of a disease-causing or-
100 5.3  106 5.3  105 ganism in that supply. Thus, the more contaminated the raw
1 6.6  108 2.0  105 water source, the more treatment is required to reduce the
0.1 6.6  109 2.5  106 risk to an acceptable level. An example of this application is
Modified from Gerba and Rose (1992). From Pollution Science © 1996, shown in Figure 14.13. The plausibility of validation of mi-
Academic Press, San Diego, CA. crobial risk assessment models has been examined by using
data from foodborne outbreaks in which information has
been available on exposure and outcomes (Rose et al., 1995;
100 liters of drinking water (assuming ingestion of 2 liters Crockett et al., 1996). These studies suggest that microbial
per day) is 1.2  103, or almost 1 in 1,000 for a single-day risk assessment can give reasonable estimates of illness from
exposure. This risk would increase to 3.6  101, or ap- exposure to contaminated foods (Table 14.15).
proximately one in three, on an annual basis. As can be seen In summary, risk assessment is a major tool for decision
from this table, the risk of developing a clinical illness also making in the regulatory arena. This approach is used to
appears to be significant for exposure to low levels of ro- explain chemical and microbial risks, as well as ecosystem
tavirus in drinking water. impacts. The results of such assessments can be used to in-
The EPA recommends that any drinking water treat- form risk managers of the probability and extent of environ-
ment process should be designed to ensure than human pop- mental impacts resulting from exposure to different levels of
ulations are not subjected to risk of infection greater than stress (contaminants). Moreover, this process, which allows
1:10,000 for a yearly exposure. To achieve this goal, it the quantification and comparison of diverse risks, lets
would appear from the data shown in Table 14.10 that the risk managers utilize the maximum amount of complex in-
virus concentration in drinking water would have to be less formation in the decision-making process. This information
than 1 per 1,000 liters. Thus, if the average concentration of can also be used to weigh the cost and benefits of control
enteric viruses in untreated water is 1,400/1,000 liters, treat- options and to develop standards or treatment options (see
ment plants should be designed to remove at least 99.99% of Example 14.6).

5
95 percent confidence limits EXAMPLE 14.6
4 How Do We Set Standards for Pathogens in
oocysts required
Log reduction of

Acceptable Drinking Water?

3
In 1974, the U.S. Congress passed the Safe Drinking Water
2
Act, giving the U.S. Environmental Protection Agency (EPA)
the authority to establish standards for contaminants in drink-
Unacceptable
ing water. Through a risk analysis approach, standards have
1
been set for many chemical contaminants in drinking water.
Setting standards for microbial contaminants proved more dif-
0 ficult because (1) methods for the detection of many pathogens
0.01 0.1 1 10 100
are not available, (2) days to weeks are sometimes required to
Source water concentration (oocysts/100 l) obtain results, and (3) costly and time-consuming methods are
required. To overcome these difficulties, coliform bacteria had
Figure 14.13 Relationship of influent Cryptosporidium concen-
been used historically to assess the microbial quality of drink-
tration and log reduction by treatment necessary to produce
acceptable water. From Haas et al., 1996.
 C. P. Gerba 231

ing water. However, by the 1980s it had become quite clear to the annual risk of infection from waterborne disease
that coliform bacteria did not indicate the presence of outbreaks in the U. S. (4  103). Based on the estimated
pathogenic waterborne Giardia or enteric viruses. Numerous concentration of Giardia and enteric viruses in surface wa-
outbreaks had occurred in which coliform standards were met, ters in the United States from the data available at the time,
because of the greater resistance of viruses and Giardia to dis- it was required that all drinking water treatment plants be
infection. A new approach was needed to ensure the microbial capable of removing 99.9% of the Giardia and 99.99% of
safety of drinking water. the viruses. In this manner it was hoped that the risk of in-
To achieve this goal a new treatment approach was fection of 104 per year would be achieved. The STR went
developed called the Surface Treatment Rule (STR). As into effect in 1991.
part of the STR, all water utilities that use surface waters To better assess whether the degree of treatment re-
as their source of potable water would be required to pro- quired is adequate, the EPA developed the Information Col-
vide filtration to remove Giardia and enough disinfection lection Rule, which required major drinking water utilities
to kill viruses. The problem facing the EPA was how much that use surface waters to analyze these surface water for the
removal should be required. To deal with this issue, the presence of Giardia, Cryptosporidium, and enteric viruses
EPA for the first time used a microbial risk assessment ap- for a period of almost 2 years. From this information, the
proach. The STR established that the goal of treatment was EPA set treatment control requirements to ensure that the
to ensure that microbial illness from Giardia lamblia in- 104 yearly risk is met. Utilities that have heavily contami-
fection should not be any greater than 1 per 10,000 ex- nated source water are required to achieve greater levels of
posed persons annually (104 per year). This value is close treatment (see Figure 14.13).

QUESTIONS AND PROBLEMS

1. List the four steps in a formal risk assessment. 9. What is a NOAEL and how does it differ from a LOAEL?
2. Why do we use safety factors in risk assessment? 10. If 10 oocysts of Cryptosporidium are detected in 100 L of
3. What is the most conservative dose–response curve? What surface water, how much reduction (in log10) by a water
does it mean? treatment plant is required to achieve a 1:10,000 annual risk of
4. What is the difference between risk assessment and risk infection?
management? 11. Give an example of a nonthreshold response for a chemical
5. What are some of the differences between the risks posed by toxin.
chemicals and those posed by microorganisms? 12. What is the difference between a stressor and a receptor?
6. Suppose a 50-kg individual drinks 2 L day1 of chloroform Give an example of a chemical stressor and a receptor in an
and 0.1 mg L1 phenol. What is the hazard index? Is there aquatic system. What endpoint would you use?
cause for concern? 13. Draw an exposure pathway for pathogens for the disposal of
7. Estimate the cancer risk for a 70-kg individual consuming 1.5 raw sewage into the ocean. Consider likely routes of inges-
liters of water containing trichloroethylene (TCE) per day for tions and inhalation. As a risk manager, what options may
70 days. you have to reduce the risks of exposure?
8. Calculate the risk of infection from rotavirus during swimming 14. Using the U.S. Environmental Protection Agency IRIS
in polluted water. Assume 30 ml of water is ingested during database ([Link]/iris), find the critical effect, uncer-
swimming and the concentration of rotavirus was 1 per 100 tainty factor, and NOAEL, LOAEL, and RfD for mercury,
liters. What would the risk be in a year if a person went swim- chromium, and chloroform. In drinking water, which one
ming 5 times and 10 times in the same water with the same would be the most toxic?
concentration of virus?

REFERENCES AND ADDITIONAL READING

Cockerham L.G. and Shane B.S. (1994) Basic Environmental Gerba C.P. and Rose J.B. (1992) Estimating viral disease risk
Toxicology. CRC Press, Boca Raton, Florida. from drinking water. In Comparative Environmental Risks, C.R.
Covello V., von Winterfieldt D. and Slovic P. (1986) Risk Cothern (ed.) Lewis Publishers, Boca Raton, Florida.
Communication: A review of the literature. Risk Analysis Gerba C.P., Rose J.B. and Haas C.N. (1996) Sensitive populations:
3,171–182. Who is at the greatest risk? Int. J. Food Microbiol. 301, 113–123.
Crockett C.S., Haas C.N., Fazil A., Rose J.B. and Gerba C.P. Gerba C.P., Rose J.B., Haas C.N. and Crabtree K.D. (1997)
(1996) Prevalence of shigellosis: Consistency with dose-response Waterborne rotavirus: A risk assessment. Water Res. 12,
information. Int. J. Food Prot. 30, 87–99. 2929–2940.
232 Chapter 14 • Risk Assessment

Haas C.N. (1983) Estimation of risk due to low levels of microor- Roseberry A.M. and Burmaster D.E. (1992) Lognormal Distribu-
ganisms: A comparison of alternative methodologies. Am. J. tions for Water Intake by Children and Adults. Risk Anal. 12,
Epidemiology. 118, 573–582. 99–104.
Haas C.N., Crockett C.S., Rose J.B., Gerba C.P. and Fazil A.M. Straub T.M., Pepper I.L. and Gerba C.P. (1993) Hazards from
(1996) Assessing the risk posed by oocysts in drinking water. J. pathogenic microorganisms in land-disposed sewage sludge. Rev.
Am. Water Works Assoc. 88, 113–123. Environ. Contamination Toxicology 132, 55–91.
Haas C.N., Rose J.B. and Gerba C.P. (1999) Quantitative Micro- Tseng W.P. (1977) Effects and dose response relationships of skin
bial Risk Assessment. John Wiley & Sons, New York. cancer and blackfoot disease with arsenic. Environ. Hlth.
ILSI Risk Science Institute Pathogen Risk Assessment Working Perspect. 19, 109–119.
Group. (1996) A conceptual framework to assess the risks of human Tseng W.P., Chu H.M., How S.W., Fong J.M., Lin C.S. and Yeh
disease following exposure to pathogens. Risk Anal. 16, 841–848. S. (1968) Prevalence of skin cancer in an endemic area of chronic
Kammen D.M. and Hassenzahl D.M. Should We Risk It? Prince- arsenic in Taiwan. J. Natl. Cancer Inst. 40, 453–463.
ton University Press, Princeton, New Jersey. United States Environmental Protection Agency (U.S. EPA)
Kolluru R.V. (1993) Environmental Strategies Handbook. (1990) Risk Assessment, Management and Communication of
McGraw-Hill, New York. Drinking Water Contamination. EPA/625/4–89/024, Washington,
Landis W.G. and Yu M.H. (1995) Introduction to Environmental D.C.
Toxicology. Lewis Publishers, Boca Raton, Florida. United States Environmental Protection Agency (U.S. EPA)
National Research Council. (1983) Risk Assessment in the Federal (1992a) Framework for Ecological Risk Assessment. EPA
Government: Managing the Process. National Academy Press. 1630/R–92/001, Washington, D.C.
Washington, D.C. United States Environmental Protection Agency (U.S. EPA)
National Research Council. (1989) Improving Risk Communica- (1992b) Dermal Exposure Assessment: Principles and Applica-
tion. National Academy Press. Washington, D.C. tions. EPA 600/8–91/011B, Washington, D.C.
National Research Council. (1991) Frontiers in Assessing Human United States Environmental Protection Agency. (U.S. EPA)
Exposure. National Academy Press, Washington, D.C. (1986) Guidelines for Carcinogen Risk Assessment. Federal
Register, September 24, 1986.
Regli S., Rose J.B., Haas C.N. and Gerba C.P. (1991) Modeling
the risk from Giardia and viruses in drinking water. J. Am. Water Ward R.L., Berstein D.I. and Young E.C. (1986) Human
Works Assoc. 83,76–84. rotavirus studies in volunteers of infectious dose and serological
response to infection. J. Infect. Dis. 154, 871–877.
Rodricks J.V. (1992) Calculated Risks. Understanding the Toxic-
ity and Human Health Risks of Chemicals in Our Environment. Watts, R.J. (1998) Hazardous Wastes: Sources, Pathways,
Cambridge University Press, Cambridge, England. Receptors. John Wiley & Sons, New York.
Rose J.B., Haas C.N. and Gerba C.P. (1995) Linking microbiolog- Wilson R. and Crouch E.A.C. (1987) Risk assessment and
ical criteria for foods with quantitative risk assessment. J. Food comparisons: An introduction. Science 236, 267–270.
Safety 15, 111–132.