United States
Environmental Protection Office of Water EPA 815-R-98-005
Agency 4607 August 1998
*EPA U.S. EPA and CDC Workshop on
Waterborne Disease Occurrence
Studies - Summary Report
March 12-13, 1997
Atlanta, Georgia
-------
Summary Report
U.S. Environmental Protection Agency and the Centers for Disease Control and
Prevention
Workshop on Design of Water borne Disease Occurrence Studies
Workshop Held in Atlanta, Georgia, March 12-13,1997
l
Workshop Organizers: Sue Binder, Fred Hauchman , and Ron Hoffer
Report prepared by Gunther Craun
Draft Report Reviewed by Rebecca Calderon, William MacKenzie, Donald Reasoner
-------
Table of Contents
Introduction
Workshop Agenda 3
Epidemic and Endemic Waterborne Disease and a National Estimate of its Occurrence .... 4
Epidemiological Study Designs '. 6
Table 1. Types of Epidemiological Studies* ; 6
Systematic Bias in Epidemiological Studies 8
w
Random Error 9
Measures of the Magnitude of Risk ' 10
Selection of a Study Design 10
Table 2. Advantages and Disadvantages of Epidemiological Study Designs to
Estimate the National Occurrence of Waterbome Disease 13
Table 3. Possible Sources of Systematic Basis and Methods for Control in
Studies to Estimate the National Occurrence of Waterborne Disease 13
Selection of a Health Outcome for Study 14
Populations to be Studied 15
Selection of a Study Site 16
1 •
-------
Table 4 Matrix of Water Quality and Treatment/Compliance Considerations for Selecting a
study Site 17
Water Exposures and Home Treatment Devices To Consider for Intervention Studies .... 17
Table 5 Evaluation of Bottled Water and Home Water Treatment Units for Use in
Intervention Studies - 19
Appendix A. Participant List 20
Appendix B. Abstracts of Presentations 25
Appendix C. Draft Request for Proposal 36
Appendix D. Causality of an Epidemiological Association 38
2,
-------
Introduction
The Safe Drinking Water Act (SOWA) Amendments of 1996, Section 1458(d), require
the Director of the Centers for Disease Control and Prevention (CDC), and the administrator of
the U.S. Environmental Protection Agency (EPA) to jointly:
within 2 years after the date of enactment of this section, conduct pilot waterborne
disease occurrence studies for at least 5 major United States communities or public
water systems; and within 5 years after the date of enactment of this section, prepare a
report on the findings of the pilot studies, and a national estimate of waterborne disease
occurrence.
EPA has set aside research funds to begin this collaborative effort with CDC, and several
interagency discussions have taken place. The next step in the planning of this research was to
convene a work group to discuss the types of epidemiological studies that may be appropriate to
estimate the prevalence of waterborne disease and identify the various issues related to their
design and feasibility. Invited epidemiologists, scientists, physicians, and engineers from EPA,
CDC, state public health agencies, and universities met in Atlanta on March 12 and 13, 1997 (list
of participants in Appendix A). Participants were asked to address the strengths and weaknesses
of various study designs and other important issues, such as the selection of geographic areas,
population groups, health outcomes, and water exposures which should be studied.
Considering EPA's overall strategy for implementing the SDWA and the legislative
history of the SDWA amendments, the following assumptions were made regarding the proposed
pilot studies:
• Studies should be designed primarily to provide a national estimate of waterborne disease
occurrence rather than address other issues surrounding waterborne disease or
environmental epidemiological methods.
• Studies should focus on diseases caused by both well-known and "emerging" waterborne
pathogens (bacteria, viruses, and protozoa).
• Because of recent concerns about microbial contaminants and limited funding, studies
will hot address effects of acute or chronic exposure to chemical contaminants in water.
Workshop Agenda
Technical presentations at the beginning of the workshop provided participants with
current information to assist in the design of the proposed epidemiological studies:
• Water sources and treatment used in the United States.
Occurrence of waterborne pathogens in water sources and tap water.
• Etiologic agents causing waterbome disease.
• Water system deficiencies associated with reported waterbome outbreaks.
• Analytical methods for detecting pathogens in water samples and clinical specimens.
• Human dose response issues for Norwalk virus and Cryptosporidium.
• 3
-------
Participants were also informed about the objectives, study designs, and preliminary
results of recently completed and ongoing waterborne disease epidemiological studies:
Study of risk factors including water for cryptosporidiosis in an HIV-infected and a general
population cohort at Cornell Medical Center, New York City.
An incidence case-control study of waterborne and other risks of cryptosporidiosis in HIV-
infected persons in New Orleans and Los Angeles.
• Study of the sero-prevalence of Cryptosporidium-specific antibodies in populations using
surface and groundwater systems and an NHANES national population.
• Longitudinal enteric disease studies in communities using surface water sources where water
filtration facilities are being installed.
• Community intervention studies of endemic waterborne disease in Canada.
Abstracts of these presentations are included in Appendix B.
A general discussion of the fundamental questions surrounding the proposed
epidemiological studies followed the technical presentations. Participants then met separately in
three break-out groups for further discussions. Each group summarized their deliberations at the
end of the first day. During the second day' workshop participants identified and discussed the
important issues that should be considered in designing and conducting the proposed
epidemiological studies. This report summarizes these issues.
Epidemic and Endemic Waterborne Disease and a National Estimate of its Occurrence
Workshop participants were reminded that not all waterborne disease results in diarrhea!
illness and that to fully understand waterborne disease risks requires the consideration of other
symptoms. For example, nausea and vomiting are the primary symptoms of illness caused by
Norwalk virus, an important waterborne pathogen, and hepatitis A; however, symptoms do not
usually include gastroenteritis. In addition, many waterborne infections result in mild illness or
may be asymptomatic, and ingestion is not the only route of exposure. The important route of
exposure for several potential waterborne pathogens (e.g., Mycobacteria, Legionelld) can be
from the inhalation of aerosolized organisms.
Waterborne illness can be both epidemic (outbreaks) and endemic. Several workshop
participants cautioned that in the United States unrecognized endemic waterborne illness may be
the greater problem and this should be the focus of the proposed epidemiological studies. Little
information is currently available about the endemic waterborne disease risks. It is not known
whether the endemic incidence of waterborne illness is a relatively low or high, the proportion of
gastroenteritis cases is relatively constant over time, or whether illness occurs as sporadic cases
due to very small outbreaks. These small outbreaks may occur relatively frequently but are not
likely detected by current disease surveillance systems, and epidemiological studies must be
•4 '
-------
specifically designed to determine the endemic waterborne risk. Although several studies have
assessed endemic waterbome disease risks, the informatidri is not sufficient to estimate the
occurrence of waterbome disease in the United States with any degree of certainty. Waterborne
outbreaks are voluntarily reported, and information about their occurrence and causes has been
systematically collected by CDC and EPA since 1971. Analysis of waterbome outbreak data are
primarily used to help identify important etiologic agents and water system deficiencies.
National estimates of waterborne disease occurrence have been reported by several
investigators. Hauschild and Bryan (1980) estimated the annual incidence of both food- and
waterborne disease in the U.S. for 1974 and 1975 to be 1,400,000 to 3,400,000 cases. Morris and
Levin (1994) estimated that 1.8 million cases of waterborne disease and 1,800 deaths occur annually,
and Bennett et al. (1987) estimated the annual incidence of waterborne disease to be more than
900,000 cases and almost 900 deaths. These estimates were not based on epidemic logical studies
specifically designed to obtain such an estimate, and the estimates are highly uncertain. For example
the estimate prepared by Bennett et al. (1987), the most often cited reference, was based solely on
expert opinions about the underreporting of specific enteric diseases and the proportion of cases
likely to be waterbome. Workshop participants agreed that a better estimate is needed for the
occurrence of waterbome illness and this estimate should be obtained from appropriately designed
epidemiological studies.
To estimate the national occurrence «of waterborne disease requires two components — the
population attributable risk (PAR) of illness due to drinking water exposures and the incidence of
acute gastroenteritis (AGI) or specific enteric diseases from all causes1 . Well known standard
procedures are available to design and conduct a national survey of AGI incidence or other selected
microbial enteric disease. However, more careful thought must be given to choosing and designing
the kinds of epidemiological studies needed to estimate the PAR.
The PAR measures the relative proportion of disease attributed to water compared to
other exposures, and this measure will likely differ for each etiologic agent and each geographic
area (major city, region, and water system type or source). These possible differences should be
considered in the proposed study designs and must be taken into account in computing the
national estimate. The selection of health outcome measures, and other issues such as the
consideration of symptomatic and asymptomatic cases and herd immunity must be also be
considered. For example, the entire U.S. population using public water systems may not be at
risk of waterbome infection because of recent infection and immunity to certain .etiologic agents.
This may be an especially important when evaluating the risk for waterbome cryptosporidiosis.
For example, if it is found that only 50% of the population is at potential risk from illness due to
Cryptosporidium water exposures, the overall PAR for all waterbome disease will be greatly
reduced. The workshop examined the designs of epidemiological studies that should be
considered to estimate the PAR, disease or health related endpoints that should be selected for
study, water exposures that should be evaluated, and how sites should be selected. These issues
are further discussed in a later section of this report.
1 Waterbome Disease = (PAR) x (Disease from all Causes)
-------
Workshop participants felt that a single study design for all geographic areas may not be
the best approach to provide information for a national estimate of waterborne disease. Rather,
more convincing evidence of a waterborne disease risk and its magnitude would be provided by
results from several different study designs, especially if results were consistent among the
different studies. -
EpidemiologicaL Study Designs
Observational and experimental studies can be conducted to assess associations between
waterborne infectious disease risks and drinking water sources, treatment, and contaminants.
This section describes the kinds of studies that were considered. Because of slight differences in
the terminology used to describe epidemiological study designs, the scheme proposed in Table 1
is proposed to avoid confusion about the types of studies.
Table 1. Types of Epidemiological Studies*
I. Experimental
A. Clinical
B. Population-Based Intervention
II. Observational
A. Descriptive (lnclude»Community-Based Intervention)
1. Disease Surveillance and Prevalence Surveys
2. Ecological *
B. Analytical (Includes Community-Based Intervention)
1. Longitudinal
a. Cohort or Follow-up
a. Case-Control
2. Cross-sectional
* Adapted from Monson. 1980
Estimates of the magnitude of the waterborne disease risk can be obtained from
moderately large, well-designed, and well-conducted experimental and observational
epidemiological studies (see Appendix C). Guidelines are available to assess the causality of an
epidemiological association (see Appendix D), and each study must be carefully evaluated to
determine whether its design is appropriate, the study size and power are adequate to detect a
meaningful difference in risk of illness and the systematic bias.
Experimental Studies. Experimental epidemiological studies include population-based
intervention studies and clinical trials. The experimental epidemiological study design is the
only design approved by the Food and Drug Administration to determine the effectiveness of
new pharmaceuticals. If this study design is found to be scientifically and economically feasible
to evaluate water exposures, it can provide solid epidemiological evidence for an estimate of
waterborne disease in the United States.
This type study considers the effect of varying some characteristic or exposure which is
under the investigator's control, much like a lexicological study. Comparable individuals are
randomly assigned to a treatment or intervention group (e.g., bottled, tap, and home filtered
water groups) and observed for a specific health-related outcome (e.g., signs and symptoms of
-------
illness, positive stool specimens to identify cases of giardiasis, antibodies in serum or saliva).
The study is longitudinal in nature, much as a cohort study, as persons in each group are
followed prospectively. Ethical concerns must be fully addressed, and periodic analyses may be
necessary to assess risks throughout the study. Once sufficient information is available to
suggest an increased or decreased risk in any group, participants must be informed. Once
informed, participants may choose to drop out of the study.
Observational Studies. Two basic kinds of observational epidemiological studies can be
conducted, descriptive and analytical. These approaches differ primarily in the supportive
evidence they can provide about a possible causal association between drinking water and
disease. An ecological study does not link individual health outcome events to individual
exposure or confounding characteristics; an analytical study does. In an ecological study,
information about exposure and disease is available only for groups of people but not for
individuals that comprise the groups. Thus, information regarding possible confounders
(alternative explanations) of associations between exposure and disease are not available.
Investigators have examined limitations of these studies and when and under what assumptions
this type study may be appropriate (Piantadosi, 1994; Greenland and Robins, 1994a, 1994b;
Susser, 1994a, 1994b; Poole, 1994). Ecological studies can help develop hypotheses and may be
useful for studying infectious disease where the overall community effect is the important
consideration. For example, they can provide useful information about herd immunity in the
community and possible changes in community disease rates associated with changes in water
sources or treatment methods. These studies will not, however, provide reliable quantitative
information about the magnitude of risk for a national estimate.
Analytical epidemiological studies can provide information about the causality of an
association and estimate its magnitude. Information is obtained about disease and exposures for
each person included in the study. Analytical studies are either longitudinal or cross-sectional.
In a longitudinal study, the time sequence can be inferred between exposure and disease, i.e.,
exposure precedes disease. In a cross-sectional study, exposure and disease information relate to
the same time period, and it is not always easy to determine if exposure precedes disease or vice-
versa. Cross-sectional studies of infectious disease are less of a problem in this regard, but
results must be evaluated to ensure that exposure caused disease.
Longitudinal studies are of two opposite approaches: the cohort study and the case-
control study. The cohort study begins with the identification of individuals having an exposure
of interest (e.g., persons using tap water, untreated groundwater, or unfiltered surface water) and
an unexposed population (e.g., persons using bottled, disinfected groundwater, or filtered surface
water) for comparison. Disease (e.g., AGI, cryptosporidiosis) or other health-related outcomes
(e.g., serum or saliva antibody levels) are then determined for each group. A number of different
diseases or symptoms can be studied. A cohort study can be either retrospective or prospective
and sometimes a combination retrospective-prospective approach is used. Two or more groups
of people are assembled for study strictly according to their exposure status at the present time
(prospective) or in the past (retrospective). Disease incidence rates are compared between
exposed and unexposed groups.
In a case-control study, the investigator identifies persons having a disease (e.g., AGI,
cryptosporidiosis) or health outcome of interest (antibody level, stool positivity) and a control or
-------
comparison group of individuals without the disease of interest. Past water exposures and other
risk factors are evaluated in these persons. A variety of types of exposures (e.g., bottled water,
tap water, filtered water, volume of water consumed) and potential risk factors (e.g., foreign
travel, consuming untreated water while camping) can be studied. Controls (the comparison
group) can be randomly selected from the either the general population or a special population
within a specified geographic area (e.g., hospitals, HMOs, or nested in a selected cohort), but
procedures must be adequate to ensure controls are free of disease. Different types of water
exposures must also be available in the selected geographical area. There are several design
variations and options for obtaining information (e.g., how cases and controls are selected; the
geographical area studied; how information on exposures, risk factors, and confounding factors is
obtained; who is interviewed; etc.). Both cohort and case-control studies can take advantage of a
v natural experiment'; that is, a community that changes water treatment or water sources or uses
water sources of different quality for distribution to several distinct areas. In thesex natural
experiments' exposure is assessed for each person included in the study rather than on an
ecological level.
Systematic Bias in Epidemiological Studies
Bias can occur because of the way ^Tstudy is designed and conducted, leading to a false or
spurious association or a poor measure of risk (one that departs systematically from the true
value). Procedures in the study's design and conduct are used to prevent or reduce possible
systematic bias. When information for exposure arid disease is collected by methods that are not
comparable for each participant (e.g., selective recall of disease by exposed persons), an incorrect
association will be due to observation bias. When the criteria used to enroll individuals in the
study are not comparable, the observed association between exposure and disease will be due to
selection bias. A wrong diagnosis of disease or assessment of exposure can result in
misclassification bias. Random or non-differential misclassification bias almost always biases a
study toward not observing an effect or observing a smaller change in risk than may actually be
present, but non-random or differential misclassification can result in either higher or lower
estimates of risk, depending on the distribution of misclassification.
Confounding bias may convey the appearance of an association; that is, a confounding
characteristic rather than the putative cause or exposure may be responsible for all or much of the
observed association. To cause confounding, a characteristic must be associated with the
exposure being evaluated, a risk factor for the disease, and not part of the exposure-disease
pathway. Information on known or suspected confounding characteristics is collected to evaluate
and control this bias during data analysis. Matching is a technique used to prevent confounding
bias. For example, if a child in day care is thought to be a possible confounding characteristic,
an equal number or proportion of families with children in day care can be selected for study in
order to avoid confounding bias from this exposure.
In an experimental epidemiological study, randomization is possible; that is, each
individual in the study has an equal or random chance of being assigned to an exposed or
unexposed group. Because of this random assignment of exposure, all characteristics,
confounding or not, tend to be distributed equally between the selected study groups of different
-------
exposure. This is the major reason experimental studies are held in such high esteem.
Random Error
Random error is governed by chance and influenced primarily by the size of the study
(number of study participants). The likelihood that a positive association is due to random error
is assessed by calculating the level of statistical significance ("p" value) or confidence interval
(C.I.)- A. small "p" value or a C.I. that does not include unity (1.0) suggests chance may be an
unlikely explanation for an observed association but does not completely rule it out. A
statistically significant association may still be spurious, however, because of observation,
selection, misclassification, and confounding biases discussed previously.
A study should be to conducted within a reasonable time period and without
overwhelming logistical problems. The number of study participants should be reasonably large
based on an expectation'of detecting a meaningful difference in risk of illness between the
exposed and unexposed groups.
When computing sample size requirements, either two-tailed or one-tailed statistical tests
can be considered. One-tailed tests require fewer study participants for the same level of
significance and can be justified because there is no anticipated beneficial effect of drinking
contaminated water. But in an intervention study, the possibility of introducing harm by the
•intervention, although this is considered highly unlikely, should be taken into account in
designing study size. Sample size requirements primarily introduce logistical problems - how to
collect sufficient numbers of participants in a reasonable time period. If a particular study design
is-desired and it is not possible to obtain the required number of participants within a reasonable
time, either the study period can be extended or the study area expanded in order to collect the
necessary cases. Another less desirable option is to lower expectations about the magnitude of
risk that can be detected. Study design changes can also be considered. For example, use of
more sensitive measures of illness might allow detection of the same level of risk with fewer
participants. Sample size requirements can be greatly reduced by use of more sensitive
indicators, but supporting biological data may be needed to show the relationship between serum
or saliva antibodies and illness or infection for selected diseases. These considerations are all
part of assessing the feasibility of a particular study design.
Investigators must consider what magnitude of risk is desired to be detected and the
number of participants required to detect that risk can be computed for each type of study.
Determining the magnitude of risk that may be meaningful is often difficult, but the anticipated
risk differences should be at levels sufficient for establishing public health policy. For drinking
water associated risks, a relatively small difference in illness between the exposed (tap water)
and unexposed groups should be considered. In the recently completed intervention study by
Payment et al., the study size was computed based on the detection of a 5% increased risk of
AGI due to tap water consumption; a 14 to 40% PAR (depending on the age group) was
observed in the study.
Designing a study to detect very small risks is especially important in the event no
association is observed in one or more of the proposed studies. If systematic error is not
suspected in a reasonably large study, the study's power to observe an association should be
-------
considered. A priori expectations of the magnitude of risk and assumptions about its variability
are used to determine an appropriate number of study participants. Upon completion of the
study, information from that study is available to evaluate the study' s power to detect an
association of a certain magnitude. For example, studies in which an association is not observed
often include a statement similar to the following, "no association was observed but the power of
this study could only detect a RR of 1.9 or greater." This suggests the possibility that there may
be an association, however, the risk of that association is less than could be detected in this
study.
Measures of the Magnitude of Risk
The two basic measures of an association between exposure and disease in analytical
studies are: the rate ratio or relative risk (RR) and exposure-odds ratio (OR). A RR or OR of
1.0 indicates no association; any other ratio signifies either a positive or negative association.
For example, a RR or OR of 1.8 indicates an 80 percent increased risk among the exposed. The
size of the RR and OR is also used to help assess an observed association. Based on Monson's
(1980) experience, a RR or OR of 0.9 to 1.2 is considered too weak to be detected by
epidemiological methods. It is difficult to interpret a RR or OR of 1.2 to 1.5 because one or
more confounding characteristics can easily lead to a weak association between exposure and
disease, and it is difficult to identify, measure, or control weak confounding bias. A larger RR or
OR, however, is unlikely to be completely explained by some uncontrolled or unidentified
confounding characteristic. Because the RR or OR for endemic waterborne disease is expected
be less than 1.5, unidentified weak confounding bias will be an important consideration. The
size of a RR has no relevance for assessing the occurrence of observation, selection, or
misclassification bias.
Selection of a Study Design
Discussion of appropriate study designs to provide a national estimate of waterborne
disease focused on five designs:
1) Population-based intervention study where persons in a community are randomly assigned to
drink tap water or an alternative source of water that is known to be free of pathogens (bottled,
boiled, or further filtered and disinfected at the home).
2) Cohort study where persons in a specified community are enrolled into several groups based
upon their current water consumption patterns (always drink bottled water; frequently drink
bottled water; always drink home filtered water which is effective in removing pathogens;
always drink tap water without further treatment; drink only coffee, tea, soft drinks, etc.).
3) Case-control study where cases (AGI, severe AGI, persons with certain antibody levels, or a
specifically-diagnosed disease) are selected from a broad geographic area with different types
of community water systems or in an area or community where the only difference in water
exposures will be individual choices about use of the community supply or alternatives
(bottled, filtered, etc. as noted previously).
10
-------
4) Cross-sectional study where the prevalence of illness or a health-related outcome is
determined for populations residing in communities or areas with various water sources and
types of treatment. Community-based intervention study (analytical rather than ecological
design). In this study, the selection of communities depends upon planned improvements to
water quality, increased water treatment, changes in water sources, or communities where
separate water sources serve distinct areas. The study can be cross-sectional, cohort or case-
control. Each of these study designs can provide evidence for a national estimate, and each
offers certain advantages in terms of scientific and economic feasibility. However, each also
has certain disadvantages and limitations. Participants generally agreed that the population-
based intervention study would provide the strongest epidemiological evidence of waterborne
disease risk and best estimate of the PAR for drinking water exposures, but it was also
acknowledged that for both economic and scientific reasons several different designs could be
employed. In the intervention study, it may be difficult to completely control systematic bias
by not effectively blinding study participants about water exposures. Results of studies of a
different design (and different sources of possible bias) can be used for comparison with
results from intervention studies.
5) The population-based intervention study is the most expensive to conduct, and funds may not
be available to conduct this type study in all five cities. Similar evidence of an association
can be obtained from the cohort or case-control design at less cost, but the feasibility of
conducting other types of studies depends largely on the number of persons in that area who
do not use tap water without further treatment or use bottled water. In the intervention study,
a equal number of persons will be assigned to the various water exposure groups, whereas, in
cross-sectional, cohort, and case-control studies, 30% or fewer in the population studied may
by choice consume bottled or water other than tap water. The cost of a cohort study will be
much less than an intervention study because the cost of alternative drinking water sources is
not provided by investigators, and because of this lower prevalence of exposure, a larger
number of participants may be needed to detect the same magnitude of risk that can be
detected by the intervention study. The study size of a cohort study may be reasonable to
study certain health outcomes (AGI or persons with increased antibody levels), but this design
may not be feasible to study specific waterborne diseases such as cryptosporidiosis, giardiasis,
or shigellosis, as a very large number of study participants will be required. Sample size
requirements, however, may be reasonable for a case-control study of specific waterborne
diseases or severe cases AGI, and these studies can be conducted within a cohort study. A
possible disadvantage of the non-intervention study designs is the self selection of persons
who regularly consume bottled water. Are they different in ways that may cause confounding
bias? This question needs to be answered before these studies are initiated, and an appropriate
study of bottled water users should be conducted to obtain this information.
Workshop participants recommended that each of these types of study be further
evaluated for possible systematic biases that may need to be controlled, methods available for
their control, number of participants needed for a statistically stable estimate of risk, and other
advantages and disadvantages. Each of the study designs should also be evaluated for their
ability to consider each of several possible health outcomes (see subsequent section on selecting
a health outcome). For example, if specific diseases such as giardiasis, cryptosporidiosis, or
11
-------
Norwalk vims caused illness are to be studied, a case-control design (population-based, hospital-
based, or nested within a cohort) may be more feasible than a cohort design and may be the most
informative approach, whereas, if symptoms of AGI are to be studied one of the other study
designs may be more appropriate. Discussions about the case-control design suggest that this
design is probably useful primarily as a complement to the cohort study, but its use as a way to
obtain information about AGI (mild or severe) should not be completely dismissed without
further consideration. A possible problem with the case-control approach is the selection of
appropriate controls who do not have the illness or health condition being studied.
An advantage of the cross-sectional study design is that a random sample of the U.S.
population can be selected for study so that results will be representative of the general
population. The cost of this study is also less than the cohort because persons are not followed
over time, but a disadvantage is that the health and exposure of a person are determined at a
single point in time. This problem can be remedied by conducting a series of cross-sectional
studies in the area over a selected time period, but the costs would then approach a cohort study.
Another advantage is that the study allows investigators to determine base-line levels of health,
monitor water quality or other measures of effective treatment for adverse changes, and then
conduct health monitoring studies as water quality changes. Serological measures can be
incorporated in the cross-sectional design, but because paired sera are necessary to assess illness
caused by Norwalk-like viruses, this study design can not be used to assess this illness; the
cohort and case-control study could. It was'suggested that a cross-sectional study could be
conducted as part of NHANES by including" questions about drinking water use in the NHANES
questionnaire. This possibility should be explored, but the difficulty of collaborating with
NHANES was also noted.
The following tables should be completed and used as a guide for selecting one or more
study designs. Table 3 presents a systematic approach to identifying the advantages and
disadvantages of each of the study designs, and Table 2 allows for an evaluation of possible
systematic bias and methods for its control. The desired features of an epidemiological study of
waterborne disease are presented as candidates, for possible consideration.
After a study design(s) is selected, its scientific and economic feasibility must be assessed.
Certain information may be required before further development of a particular study design.
For example, information may be needed about the proportion of the population that regularly
uses bottled water and their characteristics to help design the study and assess its feasibility.
Procedures for reducing systematic bias, collecting information about the disease or health-
outcome measure, exposure, and other risk factors, and selecting and enrolling participants
should also be tested before a series of large epidemiological studies are initiated. Good
management procedures and controls should also be tested and developed. Overall management
of the studies through a multi-center approach should be investigated.
12
-------
Table 2. Advantages and Disadvantages of Epidemiological Study Designs to
Estimate the National Occurrence of Waterborne Disease
Important Features of a Study Design
Control of Observation and Selection Bias
Control Confounding Bias
Minimal Exposure Misclassification .
Incorporate Dose-Response Assessment
Minimal Disease Misclassification
Sample Size Required to Detect a Reasonable
Risk
Time Required to Collect Sufficient Number of
Study Participants
Likely High Rate of Participation in Study
Incorporate a Measure of Disease Severity
Study AGI
|| Study Specific Diseases of Interest
^Incorporate a Quantitative Measure of Exposure
Bo Waterborne Pathogens or Overall Microbial
If Contamination
|| Ability to Generalize Results
1 Measure Worst Case and Average Exposures
1 Include Subgroups of the Population
Individual-based
Intervention
* Discuss advantages and
disadvantages of study
design in this regard
w
i
Community-
based
Intervention
Cohort
Case-
Control
Cross-
Sectional
Table 3. Possible Sources of Systematic Basis and Methods for Control in
Epidemiological Studies to Estimate the National Occurrence of Waterborne Disease
Possible Sources
of Bias and
Comments
Misclassification
of Exposure
Misclassification
of Disease
"Selection bias .
Observation bias
Confounding bias
Individual-based
Intervention
* Discuss sources and
how likely control
efforts might be
successful
Community-
based
Intervention
-
Cohort
Case-
Control
Cross-
Sectional
1.3
-------
Selection of a Health Outcome for Study
In selecting the health outcome(s) that should be considered, it is important to remember
that waterbome pathogens can be transmitted by both the ingestion and inhalation routes of
exposure and that nausea and vomiting are important symptoms in addition to diarrhea. These
are important considerations for the design of epidemic logical studies. For example, if the study
compares bottled and tap water exposures and a high proportion of persons are found to have
symptoms associated with exposure to Mycobacteria or Legionella are noted in the study there
will be no way to evaluate whether these symptoms are associated with the selected water
exposures unless appropriate methods are included in the design of the study. Questions can be
included to evaluate inhalation risks, and questionnaire development for assessing risk factors
and identifying exposure will be an important part of all study designs.
There are really two important questions of concern in estimating the national occurrence
of waterborne disease ~ what is the magnitude of risk (PAR) and what are the important
etiologic agents. The implications for water treatment are quite different if most waterbome
illness is viral compared to bacterial or protozoan. Thus, it is important to consider the inclusion
of procedures to identify the etiologic agents. Stool, serum, or saliva specimens can be collected
and analyzed. One method is to bank sera from selected persons in the study so they can be
tested at a later date as new procedures become available for identifying the agents. This should
be considered at one or two sites. More information is needed to better assess the relationship
between an antibody response to Cryptosporidium exposure, infection, and illness.
Work group participants felt the following health outcomes should be considered for
inclusion in one or more epidemiological studies:
• Signs and symptoms of acute gastrointestinal illness (AGI), probably limited to highly
credible gastroenteritis illness (HCGI) as defined by Payment et al.
• Respiratory symptoms.
• Antibody response (serum or saliva specimens). Collection of a single specimen to look for
antibodies will not be appropriate for waterborne viruses except hepatitis. At the current
time, Norwalk-like viruses require acute and convalescent sera.
Related issues include internal controls, herd immunity or the immune status of the population
studied, secondary attack rates, severity of illness, and access to health care. Each must be
evaluated for possible inclusion in the proposed study. Internal controls (e.g., if questions are
asked about symptoms include symptoms that are unrelated to water exposures) may be needed
to help assess possible observation bias. The immune status of the population will affect the size
of population considered to be at risk and if immunity levels are high, a larger number of study
participants may be needed to detect the desired levels of risk. The question, of secondary attack
rates relates to whether secondary transmission of illness should be considered waterborne. If so,
assessment of secondary cases should be included in the study design. Participants generally felt
that if the initial case within a family is waterbome and the illness is then transmitted person to
person, the secondary cases should be considered waterborne.
.14
-------
Symptoms of waterborne disease are generally mild and many persons do not seek
medical care. Persons can also become infected but show no symptoms. Thus, should the
national estimate focus on waterborne infection, symptomatic illness, or severe illness?
Evaluation of the severity of waterborne illness requires additional discussion because it may be
useful to identify the economic impact of severe waterborne illness. Severe consequences of
waterbome disease in the United States may be limited to special; high risk groups (young,
elderly, immunocompromised), and these groups can be studied separately from the general
population. There is concern that if the risk of only severe cases is considered (e.g., hospitalized
cases of AGI) in an epidemiological study of the general population, little will be learned about
the much larger number of mild cases of waterbome illness in this population. Mild illness in a
small percentage,of the general population can also result in large economic losses if persons
must miss work or school due to the illness. Another concern about the study of only severe
cases is that the waterborne transmission of hospitalized cases may be much different than the
waterborne transmission of mild AGI. Severe cases may not be representative of all cases nor do
they necessarily cause the greatest economic impact (e.g., in the Milwaukee outbreak the highest
cost was associated with cases not hospitalized). The study population's access to health care
should be evaluated in studies of hospitalized cases or cases from HMO's. There may be
additional severe cases in the community that are not hospitalized because the persons .do not
have health care coverage. Participants feluthat the primary source of study participants should
be population based and at the household level; selecting individuals or families from a HMO
may introduce bias due to health care access.
Populations to be Studied
Different populations may be at high risk for different pathogens, and the primary
questions to be considered about selection of populations to study are:
• Should the proposed study be representative of the general population or representative of
the types of water systems used by the U.S. population?
• Should studies focus on possible high-risk populations groups, such as young children, the
elderly, or immunocompromised persons?
• Should studies be conducted to determine waterborne disease in populations exposed to water
systems with highly contaminated water or average water quality?
To obtain results that are generalizable to the U.S. population, a random sample of the
population is needed, and the cross-section study would be the study of choice. To obtain results
that are representative of the water system types or water quality received by the general
population requires a stratified random sample. Since these type studies consider water systems
with good as well as poor water quality, large numbers of study participants are required to
detect the likely very small risk. To optimize detection of an effect, populations with the poorest
water quality should be selected (see next section). This is the worst case scenario and will likely
define the upper bounds of the risk estimate.
The question of studying special populations that may be of high risk should be examined
further, but participants agreed that the argument is in favor of first studying the general
15
-------
population. If you find a risk in the general population, the risk in the immunocompromised
population can be assumed as least as high. If you find no risk in the general population, you
cannot conclude there is no risk in the immunocompromised population. If you study only the
immunocompromised population, no conclusions can be made about the general population
regardless of the findings. The study of children may be difficult, as approval from Review
Boards may not be given ( no invasive samples). However, children can likely be included in
non-intervention studies where questions are asked about symptoms of illness.
Selection of a Study Site
The health outcome(s) selected for study, the study design, and study area all affect the
feasibility of conducting a study. Criteria have been suggested for selecting a study design(s)
and health outcomes, and similar criteria should be developed for selecting suitable study
locations. Much of the success of the proposed study will depend on the area chosen, and its
location should be selected with care. The most optimal study sites may not necessarily be in
those areas where enhanced surveillance activities are being conducted for emerging pathogens,
enteric disease, or food- or waterbome outbreak surveillance. The primary site selection criterion
is water source/treatment/quality, and a major concern is to conduct the study in an area where
water exposures are optimum for detecting an association (high levels of contamination or non-
compliance and large numbers of people exposed). Since this represents the worst case situation
and not the average or a representative exposure, the national occurrence of waterbome disease
would be over-estimated. This must be balanced with the likely results from studies in areas
with little contamination. Observing no associations in these study areas provides little or no
information to estimate waterbome disease with any certainty. Results may be uninterpretable.
To ensure that a difference in risk is detected requires that the most polluted water source be
studied. Selecting unpolluted water sources reduces the likelihood of observing a risk. Since
sites are pathogen specific and risk-specific it is important to have a diversity of water sources
and quality. The alternative is to select a site where the water source is continuously impacted
by sewage. It may be important to include a control community, that is a community with a high
quality water source and good treatment. Additional discussion of this is needed, since it may be
difficult to select this control community.
Participants generally agreed that a study site should:
• Provide a representative sample of the general population's exposure to water sources and
treatment.
• Be contaminated to the extent that a health effect can be observed with an
epidemioiogical study.
However, it may not be possible to meet both conditions.
Most people are served by large community water systems but small communities may be
at greater risk because most water quality violations occur in small water systems.. It was
suggested that at least one study area contain a variety of community water systems sizes.
Possible sources of information about water sources, treatment, and quality include the EPA
regulatory data bases, EPA special water quality surveys, EPA inventory, USGS data bases on
source water quality, water quality information from state agencies, the AWWA and AWWA
16
-------
Research Foundation, and the ICR. The methods for pathogen detection in the ICR are likely to
be too poor to be of use in assessing exposures in epidemiological studies. However, the mini
ICR will collect more frequent samples and use improved laboratory techniques and thus, may be
useful. It may be possible to coordinate the selection of epidemiological study sites with
selection of sites for the mini ICR. A number of water quality data bases are available and these
should be further examined for useful information to select study sites. Individual water
treatment facilities and state regulatory agencies or their regional district offices are the best
sources of information about water quality. These sources should be used to obtain routine water
quality monitoring data and other information about water exposures.for use in the
epidemiological studies.
Communities are also planning to change water sources and treatment technologies and
these locations can be considered for possible study. Information was presented about water
systems changing disinfection practices, installing filtration (e.g., granular, membrane filtration).
The question was also raised about possible health effects associated with water distribution
systems, and this should be considered in the design if possible.
Highly contaminated water systems may define high exposure but this does not take into
account the susceptibility of the population at risk. It may be necessary to chose sites based on a
matrix of considerations. Table 4 considers a matrix of water contamination and water
treatment/regulatory compliance. A similar Jable should be developed for water quality and
population characteristics.
Table 4 Matrix of Water Quality and Treatment/Compliance Considerations for Selecting a study Site
No Contamination
Low Contamination
High Contamination
No Water Treatment
*
*
Few Systems
Limited Water Treatment or Not in
Compliance with Regulations
*
*
**»*
Good treatment and
in Compliance.
*
*
***** '
'Study to be costly and unlikely to observe an increased risk
****Good candidate sites for epidemiological studies
Water Exposures and Home Treatment Devices To Consider for Intervention Studies
In intervention studies there is concern about systematic bias in reporting symptoms of
illness because of knowledge about water exposures. Persons assigned to the tap water group
may know their source of water and provide a biased response about the occurrence of illness.
Likewise, persons assigned to the group using bottled or further treated tap water may know their
source of water and also provide a biased response about the occurrence of illness. It is
important to blind study participants about their source of drinking water. Internal controls, as
discussed earlier, can also be included to help evaluate possible observation bias. Persons in the
intervention study should also not change their normal behavior patterns related to the amount of
water consumed and other behavior habits. These changes can be documented during the study
for evaluation in the analysis.
Thus, any home water treatment unit considered for an intervention study should remove
bacteria, protozoa, and viruses, but also should be able to be modified so that a sham water
17
-------
treatment unit can also be installed. The sham unit must not modify the tap water quality in any
way. Household members should also be unable to detect a difference in water quality from the
sham or treatment units. The exterior of the units should be similar in appearance and not change
the taste or appearance of the water. Routine service and maintenance must also be performed on
both units and at the same, frequency. A question was raised about whether water for the entire
house or only the water at the kitchen sink should be treated. Since some people may drink at
other than the kitchen sink, the study should specify that water be treated water for the entire
house. Since whole house treatment is much more expensive, studies should be conducted to
determine the compliance of study participants who agree to drink water only from the kitchen
tap. It may be feasible to consider treating water only at the kitchen sink.
A discussion of the home water treatment units that should be considered for an
intervention study identified the most likely candidates (Table 5). Use of these units to further
treat tap water and use of bottled water sources should be evaluated. Ultraviolet light
disinfection will destroy bacteria and viruses but is not very effective for protozoa. An exception
is a newly developed pulsed unit, but this may not be readily available for purchase. An
undetectable sham unit can be installed to operate at an ineffective light-wave length.
Cartridge filters can remove particles, bacteria, and protozoa depending on their effective size but
must be followed by ultraviolet light for destruction of viruses. Disinfection will be more
effective when combined with cartridge filtration because of it ability to remove particles and
turbidity. A sham cartridge filter unit can be installed. Flow must be restricted to simulate the
pressure drop associated with the filtration process. A cartridge with a larger effective size
should not be employed as a sham filter because it will filter some particles, and the water
leaving the unit will not be the same quality as delivered to the tap. Reverse osmosis units are
effective in removing chlorine so there may be problems of bacterial regrowth. The water
storage reservoir may also become contaminated with bacteria which will represent a different
type water exposure. If these units are installed, health department may have concerns because
of HPC bacteria levels in treated water. Reverse osmosis treated water has a different taste so a
sham unit without the reverse osmosis membrane may be recognized by the homeowner. Reverse
osmosis also results in lots of wasted water.
An important point was made regarding the funding of the installation of the water
treatment units. Manufacturers may assist in providing the units at or below cost. Workshop
participants agreed that this could save money, but the study will be more credible if the industry
that may profit from the results is not involved in its funding or conduct.
-------
Table 5 Evaluation of Bottled Water and Home Water Treatment Units for Use in Intervention Studies
Home Treatment Unit
or Alternative Water
Source
Bottled Water
Ultraviolet Light
Cartridge Filter
Cartridge Filter + UV
Reverse Osmosis
Removal
Should Test for Quality if Purchased
Not Effective for Protozoa
Particle, Bacterial, Protozoan Removal
Depends on Effective Size; Cannot
remove all viruses
Good for All Pathogens if <2 Micron
Pore Size
Excellent for All Pathogens; Lots of
waste water; improves Taste
Remarks about Sham Unit
Wavelength can be Changed to be
Ineffective
Filter housing can be installed with a
flow restriction device to simulate
pressure drop through filter
See Above
May be Difficult to Disguise Sham Unit
as An operating Unit
References
Beaglehole R., Bonita R., and Kjellstrom T., "Basic Epidemiology" World Health
Organization, 1993
Bennett et al. Infectious and parasitic diseases. In Closing the Gap: The Burden of
Unnecessary Illness, Amler, R.W. and Dull, H.B., Eds., Oxford Univ. Press, 102, 1987.
Hauschild A.H.W. and Bryan F.L. "Estimate of cases of food and waterborne illness in
Canada and the United States. J. Food. Prot. 43: 435-440,1980.
Hill A.B., "The environment and disease: association or causation?" Proceedings of the
Roval Society of Medicine 58:295-300, 1965.
Morris R. and Levin R. " Estimating the incidence of waterborne infectious disease
related to drinking water in the United States." , Proceedings of the International Symposium,
International Assoc. Hydrological Sciences, Rome, September 13-17, 1994.
Rothman K.J., "Causal Inference, in Epidemiology" Modern Epidemiology pp 7-21,
1986.
19
-------
Appendix A. Participant List
EPA/CDC Waterborne Disease Workshop Participants
Name/Address
Ben Beard
DPD, NCID, CDC
Mail Stop F- 12
4770 Buford Hwy N.E.
Chamble,Ga 30341
Sue Binder, MD
DPD/NCID/CDC, MS F22
4770 Buford Highway, NE
Atlanta, GA 30341
Paul Blake, MD, MPH
Epidemiology and Prevention Branch
Division of Public Health
Georgia Dept. Oh Human Resources
2 Peachtree St., NW, Room 325
Atlanta, GA 30303
Joe Bresee, MD
DVRD/NCID/CDC, MSG17
600 Clifton Road, NE
Atlanta, GA 30333
Dr. Rebecca Calderon
U.S. EPA
ORD/NHEERL (MD-58A)
Research Triangle Park, NC 27711
Dr. Cynthia Chappell
University of Texas at Houston
School of Public Health
1200 Herman Pressler
P.O. Box 20 186
Houston, TX 77225
Paul Cieslak, MD
Oregon Health Division
800 NE Oregon St., #21, Suite 772
Portland, OR 97232
Dr. Bob Clark
U.S. EPA
ORD/NRMRL (MD-689)
26 W. Martin Luther King Dr.
Cincinnati. OH 45268
Telephone #
770
488-4939
%
770
488-7750
404
657-2588
404-
639-4651
919-
966-0617
713-
500-9372
503-
731-4024
513
569-7201
FAX#
770
488-4454
770
488-7794
404
657-2608
404-
639-3866
919-
966-7584
713-
500-9364
503-
731-4798
513
569-7685
E-mail address
cbbO@CDC.GOV
SCB1@CDC.GOV
PAB1@PH.DHR.STATE.G <
A.US
JSB6@CDC.GOV
CALDERON.REBECCA®
EPAMAIL.EPA.GOV
CCHAPPELL@UTSPH.
SPH.UTH.TMC.EDU
CIESLAK@STATE.ORUS
CLARK.ROBERT@EPAM
A1L.EPA.GOV
20
-------
Jack Colford, MD PhD
Assistant Professor
School of Public Health
UC Berkeley
140 Warren Hall
Berkeley, CA 94720
Mike Cox
U.S. EPA
OW/OGWDW (4603)
401MSt., S. W.
Washington, DC 20460
Gunther Craun
Gunther Craun and Associates
101 West Frederick St., Suite 104
Staunton, VA 24401
Vance Dietz MD
DPD,CDC
Mail Stop F22
4770 Buford Hwy N.E.
Atlanta, GA 30341
Dr. Al Dufour
U.S. EPA
ORD/NERL (592)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Kim Fox
U.S. EPA
ORD/NRMRL (B-24)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Dr. Floyd Frost
Lovelace Medical Foundation
Center for Health and Population Research
2425 Ridgecrest Dr., S.E.
Albuquerque, MM 87108
Roger Glass, MD, PhD
DVRD/NCID/CDC
Mail Stop G-04
1600 Clifton Road N.E.
Atlanta, GA 30333
Lee H. Harrison, MD
Associate Professor
Epidemiology and Medicine
University of Pittsburgh
521 Parran Hall
1 30 DeSoto Street
Pittsburgh, PA 15261
510
643-1076
202
260-1445
540
8861939
770
488-7771
513
569-7303
513
569-7820
505
262-3471
404
639-3577
412
624-3332
510
643-5163
202
260-3762
540
886-1939
770
488-7761
513
569-7464
513
569-7172
505
262-7598
404
639-3645
412
624-2256
JCOLFORD@UCLINK2.B
ERKELEY.EDU
COX.MICHAEL@EPA
MA1L.EPA.GOV
vdxO@CIDDPD2.
EM.CDC.GOV
FOX.KIM@EPAMAIL.
EPA.GOV
FLOYD@AUDREY.TLI.
ORG
RIG2@CIDDVD1.EM.
CDC.GOV
LHARRISO@EDC1.
GSPH.PITT.EDU
21
-------
Dr. Fred Hauchman
U.S. EPA
ORD/NHEERL (MD-S i A)
Research Triangle Park, NC 2771 1
Craig Hedberg, PhD
Surveillance and Disease Investigations Unit
Minnesota Department of Health
7 1 7 Delaware Street, SE
P.O. Box 9441
Minneapolis, MN 55440-9441
Ron Hoffer
U.S. EPA
OW/OGWDW (4607)
401 M St., S. W.
Washington, DC 20460
Dr. Christen Hurst
U.S. EPA
ORD/NRMRL (MD-387)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Dr. Walt Jakubowski
U.S. EPA ••*
ORD/NERL (592)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Dennis Juranek, DVM, MPH
DPD/NCID/CDC, MS F22
4770 Buford Highway, NE
Atlanta, GA 30341
Jan Keithly, Ph.D.
Wadsworth Center, NYS DOH
David Axelrod Inst. Pub. Health
P.O. Box 22002
120 New Scotland Avenue
Albany, NY 12201-2002
Laura A Klug, MPH
OD/NCID/CDC
Mail Stop C- 12
1600 Clifton Road N.E.
Atlanta, GA 30333
Pat Lammie, PhD
DPD/NCID/CDC, MS F13
4770 Buford Highway, NE
Atlanta, GA 30341
Bill Mac Kenzie, MD
DPD/NCID/CDC, MS F22
4770 Buford Highway, NE
Atlanta, GA 30341
919
541-3893
612-
623-5414
202
260-7096
513
569-7461
513
569-7385
770
488-7783
518
473-2692
404
639-4733
770
488-4054
770
488-7784
919
541-0642
612
623-5743 .
202
260-3762
513
569-7411
770
488-7761
518
473-6150
404
639-4698
770
488-4108
770
488-7761
HAUCHMAN.FRED@
EPAMAIL.EPA.GOV
CRAIGH@MDH-DPC.
HEALTH.STATE.MN.US
HOFFER.RON@EPA
MAIL.EPA.GOV
HURST.CHRISTON®
EPAMAIL.EPA.GOV
JAKUBOWSKLWALTE
R
@EPAMAIL.EPA.GOV
DDJ1@CDC.GOV
KEITHLY®
WADSWORTH.ORG
LBK1@CDC.GOV
PJL1@CIDDPD2.
EM.CDC.GOV
wrmO@CDC.GOV
22
-------
AltafLal
DPD, NCID, CDC
Mail Stop F- 1 2
4770 Buford Hwy N.E.
Chamble.Ga 30341
Dr. Christine Moe, PhD
Dept. of Epidemiology
School of Public Health
Univ. of North Carolina at Chapel Hill
CB-7400
Chapel Hill, NC 27599-7400
Steve Monroe, PhD.
DVRD/NCID/CDC, MS G04
1 600 Clifton Road, NE
Atlanta, GA 30333
Anne Moore, MD, PhDDPD/NCID/CDC, MS
F22
4770 Buford Highway, NE
Atlanta, GA 30341
Randall Nelson, DVM
Dept. of Public Health
4 1 0 Capital Avenue, -
Mail Stop 1 1 EPI
Hartford, CT 06 134
Dr. Patricia Murphy
U.S. EPA
NCEA(MD-190)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Pierre Payment, PhD
Institut Armand-Frappier
53 1 boulevard des Prairies
Laval, Quebec HTN 4Z3
Canada
Bob Pinner, MD
OD/NCID/CDC, MS C12
1600 Clifton Road, NE '
Atlanta, GA 30333
Dr. Don Reasoner
U.S. EPA
ORD/NRMRL (MD-387)
26 W. Martin Luther King Dr.
Cincinnati, OH 45268
Sudha Reddy
DBMD/FDDB
Mail Stop A-3 8
1600 Clifton Road N.E.
Atlanta, GA 30333
770
488-4047
919
966-1420
404
639-2391
770 .
488-7776
860
509-7994
513
569-7226
514
687-5010
1
404
639-2859
513
569-7234
404
639-4399
770
488-4454
919
966-2089
404
639-3645
^770
V 488-7761
»
860
509-8286
513
569-7916
514
686-5626
404
639-4698
513
569-7328
404
639-4080
AAL1@CDC.GOV
CHRISTINE MOE@
UNC.EDU
STM2@CDC.GOV
AYM2@CDC.GOV
NELS 127W@WONDER.E
M. CDC.GOV
MURPHY.PATRICIA®
EPAMAIL.EPA.GOV
PIERRE.PAYMENT®
IAF.UQUEBEC.CA
RWP1 ©CDC.GOV
REASONER.DONALD@
EPAMAIL.EPA.GOV
/
VBR8@CDC.GOV
•23
-------
Robin Ryder, MD
Yale University School of Medicine
60 College Street
P.O. Box 208034
New Haven, CT 06520-8034
Dr. Steve Schaub
U.S. EPA
OW/OST (4304)
401 MSt., S. W.
Washington, DC 20460
Drew Voetsch, MPH
DBMD/FDDB
Mail Stop A3 8
Atlanta, GA 30333
Due Vugia, MD, MPH
Division of Comm. Dis. Control
California Dept. Of Health Services
2151 Berkeley Way
Berkeley, CA 94704
Debra Walsh
U.S. EPA
HSD (MD-58C) »'
Research Triangle Park, NC 27711
Julie Ziegler
SRA Technologies
8 110 Gatehouse Rd.
Suite 600 West
Falls Church, VA 22042
203
785-2919
202
260-7591
404
639-4637
510
540-2566
919
966-0636
703
205-8664
203 -
785-7552
202
260-1036
404
639-4080
510
540-2570
919
966-0655
703
205-6260
RYDERR W@MASPO3 .
MAS.YALE.EDU
SCHAUB.STEPHEN®
EPAMA1L.EPA.GOV
AAV6@CDC.GOV
DVUGIA@HW1.
CAHWNET.GOV
WALSH.DEBRA@
EPAMAIL.EPA.GOV
JZIEGLER@SRATECH.
COM
24
-------
Appendix B. Abstracts of Presentations
DRINKING WATER TREATMENT AND DISTRIBUTION:
ITS ROLE IN PREVENTING WATERBORNE OUTBREAKS
Robert M. Clark", Donald A. Reasoner5, Kim R. Foxc and Christen J. Hurst4
For the last 100 years, drinking water utilities in the United States (U.S.) have played a
major role in protecting public health through the reduction of waterborne disease. For example,
in the 1880s for one year, the typhoid death rate was 158 deaths per 100,000 in Pittsburgh,
Pennsylvania but by 1935 the typhoid death rate had declined to 5 per 100,000. These reductions
in waterborne disease outbreaks were brought about by the use of sand filtration, disinfection and
the application of drinking water standards. Despite this excellent record, occasional drinking
water quality problems are a reminder of the need for constant vigilance. For example, in 1993,
Milwaukee, Wisconsin suffered a cryptosporidiosis outbreak in which it was estimated that more
than 400,000 people were ill. In July of 1993, Manhattan, New York was placed on a boil water
order, as was Washington, D.C. in December of 1993.
Nearly all of the utilities in the U.S. that use surface water practice some type of
treatment and many practice what might be called conventional treatment. A standard water
treatment train that makes up conventional treatment usually consists of: coagulant chemical
feed, rapid mix, flocculation, sedimentation, filtration, and disinfection.
In addition to being concerned over the role of water treatment in preventing waterborne
disease, we should also be concerned over the role of distribution systems in causing or
preventing waterborne disease. An illustration of the water quality problems associated with
failures in distribution comes from two recent studies. During the period December 15, 1989 to
January 20, 1990, residents and visitors to Cabool MO, population 2090, experienced 240 cases
of diarrhea and 7 deaths. Escherichia coli serotype 0157:H7, a bacterium associated with the
feces of healthy, dairy cattle, was isolated in many of the stool samples of ill people. Following
an investigation by the Centers for Disease Control (CDC), with assistance by U.S. EPA's Water
Supply and Water Resources Division, it was concluded that the illness was caused by
waterborne contaminants that entered the distribution system through a series of pipe breaks and
meter replacements that occurred during unusually cold weather. Another example of
infrastructure failure occurred in Gideon, Missouri in November 1993, when 400-500 people of a
population of 1000, contracted Salmonella typhimurium. The Salmonella outbreak contributed to
the death of 6 elderly individuals. It is presumed that bird droppings contaminated water in
storage tanks. As with Cabool, the city used a nondisinfected ground water. One consequence
of these investigations is the finding that immunocompromised people (elderly and young) are
often less capable of surviving waterbome illness than younger, healthier people. The issue of
water borne disease from municipal water supplies is of growing concern in the U.S.
a. Director, Water Supply and Water Resources Division, NRMRL, U.S. Environmental
Protection Agency, 26 W. Martin L. King Dr., Cincinnati, OH 45268.
b. Microbial Contaminants Control Branch, NRMRL, U.S. Environmental Protection
Agency, 26 W. Martin L. King Dr., Cincinnati, OH 45268.
c. Treatment Technology Evaluation Branch, NRMRL, U.S. Environmental Protection
Agency, 26 W. Martin L. King Dr., Cincinnati, OH 45268.
25
-------
d. Microbial Contaminants Control Branch, NRMRL, U.S. Environmental Protection
Agency, 26 W. Martin L. King Dr., Cincinnati, OH 45268.
Pathogen Occurrence
Walter Jakubowski, NERL, USEPA, Cincinnati, OH
EPA and CDC maintain a database of waterbome disease occurrence. This database was
examined for the time period 1971-94 to determine what microbiological agents have been
associated with waterbome disease outbreaks in the United States. For this time period, 737
waterborne disease outbreaks were reported. Eighty-nine of these were in recreational water and
69 were chemical etiologies. Removing these resulted in 579 drinking water outbreaks having a
microbiological or unknown etiology producing acute gastrointestinal illness (AGI).
Drinking water outbreaks were reported from non-community (50%), community (40%)
and individual (10%) supplies. About 60% of the outbreaks were associated with well or spring
waters, 25% with rivers or lakes and 15% with a mixture of surface or ground waters, or the
supply was unknown.' Most of the outbreaks in non-community supplies were associated with
well/spring waters. In the community water supplies, about 42% were from surface water
sources, 39% from ground waters and 20% from mixtures or unknown sources.
AGI with no specific agent identified^accounted for 56% of the outbreaks; protozoa were
responsible for 22%, bacteria for 14%, and"viruses for 8%. The bacterial agents identified were:
Shigella (51% of outbreaks in the bacterial'category), Salmonella (23%), Campylobacter (20%),
Yersinia (3%), Vibrio cholerae (3%), and Escherichia coli (1%). The protozoan agents identified
were: Giardia (90% of the protozoan outbreaks), Cryptosporidium (8%), Entamoeba histolytica
(1%), and Cyclospora (1%). Viruses identified were: hepatitis A (57% of viral outbreaks),
Norwalk-like (41%), and rotavirus (2%). Information on the densities of these organisms in
feces, sewage/wastewater, surface water, ground water and drinking water will be summarized.
26
-------
DETERMINATION OF THE INFECTIOUS DOSE OF NORWALK VIRUS
IN HUMAN VOLUNTEERS
Christine Moe, Department of Epidemiology, University of North Carolina
The objectives of this study are to 1) determine the dose-response range of Norwalk Virus
(NV) infectivity in adult human volunteers with various levels of preexisting antibodies; and 2)
determine the risk of Norwalk virus at low levels of virus that are representative of levels in
contaminated water and food. Norwalk and related viruses are predominant waterbome and
foodbome viral pathogens, as evidenced by their role in numerous outbreaks and the high
seroprevalence rates in the population. The first NV challenge study was in 1971. Since then,
numerous human volunteer studies have been conducted to examine histopathology, patterns of
illness and immune response associated with NV infection. However, none of these studies
systematically addressed the issue of infectious dose because existing technology was inadequate
to accurately titer the inoculum. The limited dose-response data from these studies is
inconsistent and contradictory. .
The study will test 45 human volunteers by having them ingest safety-tested, PCR-titered
NV inoculum at different dose levels. Conducted in three rounds of 15 volunteers, each round
tests three doses (five volunteers per dose) which approximate the ID10, ID 50 and ID 90. Samples
collected include sera, stool, vomitus and saliva. Outcome measures are clinical illness (nausea,
vomiting, diarrhea, abdominal paid), detection of NV in stool and/or vomitus by PCR and
seroconversion.
Based on the results of this study a s£cond phase is proposed which will be a closer
examination of the response in the critical low dose region of the dose-response curve. The
second phase is necessary to estimate the risk of NV infection from the low levels of virus that
may occur in water and food. The lower two or three dose levels where volunteer response in
observed in Phase 1 will be evaluated with a larger number of volunteers and more closely space
dilutions, such as half-log or three-fold dilutions. The results of these studies will be invaluable
for estimating the risk of NV gastroenteritis associated with exposure to contaminated water,
food and other vehicles.
Timeline: July 1994 - June 1997
27
-------
FIELD SURVEY OF CRYPTOSPORIDIOSIS:
RISK FROM CONTAMINATED DRINKING WATER
Floyd Frost, Lovelace Medical Institute, Albuquerque, NM
Failure to detect cryptosporidiosis outbreaks among persons consuming oocysts
contaminated drinking water may result from the insensitivity of disease surveillance programs,
which are likely to only detect large outbreaks. Alternatively, it is possible that the presence of
detected oocysts in the drinking water may be unrelated to the risk of disease outbreaks. A larger
issue is whether elevated endemic levels of disease or infection, which do not result from disease
outbreaks, are found among residents of cities consuming surface drinking water. The focus of
this series of studies is to develop information to evaluate the endemic risk of infection among
consumers of surface drinking water and to compare this risk with that for persons consuming
municipals drinking water derived from deep wells.
The serological method used is based on a serological response to Cryptosporidium
infection. Following infection, a large fraction of persons will develop antibodies to proteins
contained on the surface of the Cryptosporidium merozoite. Serum IgG antibodies appear
several weeks to several months following infection and remain, in many individuals, elevated
for months to several years. Subsequent Cryptosporidium infections appear to increase the levels
of these antibodies. Two different antibodies appear most commonly among previously infected
people and are used in this study as markers'for prior infection. These antibodies are
characterized by molecular weight of the antigen or protein to which they attach The two
antigens have wights of 17 kDa and 27 kDa.
Several studies to estimate the prevalence of Cryptosporidium antibodies have begun
using blood bank volunteer donors as the source of sera. Blood bank donors have several
advantages for the study. Access to these people is obtained through local blood banks. Donors
are screened to exclude persons with behaviors that may increase certain risks of
Cryptosporidium infection. Donors also tend to be in an age group that is less likely to have
small children in diapers. By using phoresis donors, who regularly donate blood, it is possible to
follow a large fraction of these individuals over an extended period of time. The sera from these
donors also have a low risk of HIV and are, therefore, relatively safe to work within the
laboratory.
In this ongoing study, sera are drawn from similar sub-populations in each of several
communities. These communities represent different drinking water sources, such as cities
which use unfiltered surface water, filtered surface water and deep well water. The sub-
population in each city will be phoresis blood donors. Although these donors are not
representative of the general community population, they should be similar between
communities in all characteristics other than their drinking water source. Compared with the
general population, they are likely to be easier to follow over time because of their strong
community ties and they will likely have fewer other exposures that increase their risk of
infection.
28 ,
-------
Each participant completes a two-page questionnaire on possible exposures (e.g., livestock.
children, foreign travel, drinking water source and duration of residence in the community) and
demographic information (age, sex). The questionnaire information is linked to the serology
results and analyzed to determine whether residents consuming surface water have higher
serological evidence of prior infection, after adjusting for other risk factors. In some
communities, a second blood draw is performed and a second, much shorter, questionnaire is
completed.
29
-------
COMMUNITY ENTERIC DISEASE STUDY
Rebecca L. Calderon, Epidemiology & Biomarkers Branch, NHEERL, USEPA
Microbial organisms that cause enteric disease and their sources are a major concern for
EPA. To conduct risk assessments or determine environmental health policy, information is
needed on the level of disease, factors that influence that level, specific microbial organisms that
cause illness, and possible sources of those organisms. Approximately 50% of food and
waterborne disease outbreaks are of unknown etiology. Current surveillance programs do not
provide adequate information on background rates of enteric illness and the relative source
contribution of environmental sources of organisms that cause disease. In addition, current
surveillance does not provide information on the effectiveness of environmental policy or
management decisions in lowering exposure or reducing disease.
The goal of this study is to obtain information on enteric disease rates in the U.S. Enteric
disease rates are needed to determine environmental health policy and management strategies for
environmental sources of microorganisms. The objectives of this study are to: 1) determine the
enteric disease rates in various communities across the country; 2) determine the relative source
contribution of environmental factors associated with enteric disease; 3) determine etiologic
agents associated with enteric disease; and 4) evaluate methods of surveillance. These studies
will examine alternative surveillance methods versus longitudinal studies as means to obtain
information for trend analysis.
Vf
Site selection. To vary environmental ranges of environmental factors, communities of different
geographic location, size, drinking water sources and drinking water treatment have been
identified. Ideal communities are those served by utilities that are about to change either source
or treatment. Studies in these communities would provide additional information on the health
benefit and cost-effectiveness of changes.
Timeline; June 1996 to December 1997 for completion of first community.
Site criteria: Community utility is either changing source water or changing treatment (e.g.,
adding filtration, changing disinfectants). The population served by this utility should be 15,000.
Communities with one source water type are preferred.
Measures of enteric disease:
Longitudinal health study. Approximately 300 families with children between the ages
of two .and 10 will be asked to record gastrointestinal symptomatology on a daily basis. At
the beginning of the study, those individuals will be asked to report baseline information
on current use of medicines, occupation, underlying health conditions and childcare
arrangements. Those individuals reporting symptomatology will be ask extensive.
environmental factors questions, i.e., eat out, recreational activities, contact with animals,
travel, use of antibiotics, contact with diapered children and contact with ill individuals.
These families will be monitored from June to December of 1996 and again June to
December of 1997.
30
-------
Hospital admissions. Admission rates for gastrointestinal disease (including emergency
room visits) will be collected starting in June 1996 and ending in December 1997.
Nursing home surveillance. Select nursing homes will be asked to report incidences of
gastrointestinal illness based on defined symptomatology. This surveillance will be
conducted from June 1996 to December 1997. An addition a pilot study of a community
retirement center will be conducted to correlate their reported symptomatology with
nursing home and longitudinal measurements.
Antidiarrheal sales. Pharmaceutical distributors will be asked to report sales information
from June 1996 to November 1997.
Clinical laboratory reporting. Clinical diagnostic laboratories will be asked to report
number of stool specimens reported and results of those analyses.
Analysis of etiologic agents. Approximately 30 individuals from the health diary study will be
asked to provide baseline sera and stools for etiologic analysis. During the course of the study,
those individuals will be asked to provide stools when ill and acute and convalescent sera. At the
end of the health diary, all individuals will be asked to provide sera and stools for analysis.
.•=*-.
•*<
Cross sectional serologic survey. Sera from approximately 260 college students will be
collected the first year to determine seroprevalence to known viral and protozoan agents. These
students will be asked to donate twice more in six month intervals.
31
-------
Serologic Assays for Cryptosporidium
Patrick Lammie, Ph.D.
The use of parasitologic techniques for surveillance of cryptosporidiosis underestimates
the true prevalence of infection; oocytst shedding is intermittent and tests to detect
Cryptosporidium oocysts-in stool are insensitive. Serologic assays may provide more sensitive
methods for detection of Cryptosporidium infection.
CDC investigators have developed an immunoblot to detect antibody responses directed
against two specific Cryptosporidium antigens. Monoclonal antibodies directed against the 27-
and 17- kDa antigens localize these proteins on the surface of sporozoites, the infective stage of
the parasite. Antibody responses directed against these antigens are associated with protection
from symptomatic infection. IgA and IgG responses to the 27- and 17-kDa antigens develop
within 2 to 3 weeks after infection and persist for weeks and months, respectively. Bases on
these preliminary studies, the immunoblot assay shows promise as a tool for epidemiologic
studies designed to define the risk of waterborne cryptosporidiosis. For example, monitoring
IgA responses to specific Cryptosporidium antigens may provide a measure of the incidence of
infection. Unfortunately, the laborious nature of the immunoblot assay and the requirement for
trained laboratory personnel to perform it may restrict the size of such studies and their ability to
provide definitive estimates of risk. For these reasons, the current consensus is that it would be
advantageous to convert the immunoblot to £n ELISA format. Considerable progress has been
make in the development of techniques to purify the native 27- and 17-kDa antigens and in the
characterization of these antigens. In additfon, one of these antigens has now been cloned and
expressed in a bacterial expression system; thus, it is now feasible to determine whether simple
assays based on this recombinant antigen can be developed. Additional work is now required to
determine 'whether the sensitivity and specificity of the immunoblot is retained when purified or
recombinant Cryptosporidium antigens are used in an ELISA format. It also will be necessary to
determine the time course of Cryptosporidium antibody responses using these assays and the
influence of repeated exposures to Cryptosporidium on antibody responses.
32
-------
Molecular Approaches for Environment Monitoring of Cryptosporidium spp.
Altaf Lai, Ph.D.
DNA and RNA-based diagnostic procedures have the potential to detect Cryptosporidium spp in
a species and isolate specific manner. The advantages of nucleic acid-based procedures include
greater sensitivity, rapid analysis of many samples, relatively low cost, simultaneous detection of
many pathogens, and the ability to differentiate between viable and non-viable parasites. The
availability of species/isolate specific and sensitive diagnostic tests would address the question of
: 1) what is the source of contamination in environmental samples? 2) to what extent are surface
and finish water contaminated with parasites incapable of infecting humans? and 3) what is the
zoonotic potential of most animal isolates?
Accordingly, CDC's Cryptosporidium molecular diagnostics program focuses on the
development of diagnostic that would allow: 1) detection of C. parvum in environmental
samples; 2) differentiation of C. parvum strain in both outbreak and non-outbreak settings; 3)
differentiation of C. parvum strains from C. muris, C. \vrairi, C. meleagridis, C. baileyi, C.
serpentis, and C. nasorum; and 4) differentiation of viable from non-viable parasites.
We have developed PCR diagnosis procedures that use small subunit rRNA and antigen
genes genus specific detection of Cryptosporidium. In addition to agarose gel-based diagnostic
read out, PCR procedure has been adapted^) bio luminescence and ELISA format. We are also
investigating the use of ribosomal rRNA gene-based primers, developed by Australian
investigators, to differentiate between human and non-human C. parvum strains. We have also
invasion-PCR-based tests that allow identification of viable parasites.
DPD investigators are also focusing on sequence characterization of multiple genes of C.
parvum (from human and non-human), C. muris, C. baileyi, C. serpentis toward the development
of a robust, sensitive species and isolate specific diagnostic test. Preliminary analysis of the
TRAP-C2 gene suggests that several point mutations separate humans C. parvum from non-
humans parasites.
The existing resources at DPD allow us to support epidemiologic investigation by
molecular diagnosis or pathogen. We are interested in working collaboratively with the water
.treatment facilities to test the presence of Cryptosporidium in source and filtered water. The test
available would allow us to identify the presence of Cryptosporidium in water samples, and the
tests in the process of development are designed to reveal the source of water contamination.
The present staff and resources would allow us to handle between 10-15 samples a week. In
some of these samples we will also be able to determine the viability to oocytsts. Analysis of
larger sample size, as may be required in systematic monitoring of water samples, would require
additional funding.
In about a four-six months time frame, we will test whether procedures developed in
Australia would allow differentiation of C. parvum strains. In about a years time, using probes
and primers developed through our multi-locus gene analysis, we may be able to distinguish C.
parvum from non-C. parvum parasites. These procedures are being developed with a turn around
time such that water utilities can use information developed by laboratory investigations as part
of a decision making algorithm.
-------
Prevalence of Risk Factor for Gryptosporidiosis:
Drinking Water Usage Patterns and Other Exposure Risks for NYC Patients
Dennis Juranek, Rosemary Soave, and Larry Davis
Objective: To assess the prevalence of known and theoretical exposure risks for
Cryptosporidium parvum in HIV-infected individuals and a general population
cohort at Cornell Medical Center.
Methods: A 17 page questionnaire was administered to 160 HIV infected patients and 153
ambulatory outpatients (representing the general population) at The New York
Hospital-Cornell Medical Center. Persons representing the general population were
recruited from general medicine, nephrology, and otorhinolaryngolgy medicine
departments/services.
34
-------
Incident Cases of Cryptosporidiosis in HIV-positive Persons: Risk Factors Associated with
Infection and Development of Laboratory Methods for Detection
Anne Moore, John Ward, Frank Sorvillo, Susan Troxler, Pat Lammie, Ben Beard, Norman
Pieniazek, Gilden Beall, Andrea Kovacs, Elisabeth Clark
Objectives: The study will characterize the epidemiology and risk factors for acute
Cryptosporidiosis in HIV-infected persons, characterize the serologic response to infection, and
determine the prevalence of latent colonization with Cryptosporidium after acute illness and its
relationship to CD4 and viral load.
Project Description: From a cohort of-2600 active HIV clinic patients currently followed in
New Orleans (60%) and Los Angeles (40%), this case-control study will enroll persons with
acute laboratory-confirmed Cryptosporidiosis along with two controls (one without diarrhea and
one with acute, non-cryptosporidial diarrhea). Demographic and risk factor data will be
obtained. Serum and stool specimens will be collected to facilitate the evaluation of PCR,
molecular typing, and serologic assays for Cryptosporidium now in development.
Study implementation: A study questionnaire has been collaboratively developed with the
participating sites and has been piloted in New Orleans and Los Angeles. Procedures for
identification, contact, and interview of study subjects have been outlined. Methods of specimen
acquisition, handling, storage, and shipping have been specified for each site. Study coordinators
have been hired hi each location. IRB clearance has been obtained at two participating hospital
clinics (LAC/USC, Beer Medical Group). "Clearance by CDC, and the remaining two sites
(Harbor/UCLA, LSUMC) is pending and e'xpected shortly. Study implementation with
collection of specimens is targeted to commence February 15,1997.
Progress in laboratory method development: PCR methods in development are specific for
Cryptosporidium; other parasites tested (e.g. Cyclospord) are nonreactive. Molecular typing has
been assessed at 3 genetic loci in specimens from 7 sources. All 3 loci show nucleotide variation
among isolates from different sources, one position may show strain-specific variation for the
Milwaukee outbreak, and intra-isolate variation is detected among multiple samples from the
same outbreak. The data are promising, but are too preliminary to assign patterns.
-------
Appendix C. Draft Study Proposals
Evaluating the Burden of Waterborne Diseases
EIP sites may apply for one or more of the following studies
A. Recipient activities for cross-sectional studies of the population to estimate the incidence of
gastrointestinal illness and water consumption patterns
1. Supplementation of random digit dialing surveys already being conducted by the EIP
site in the FOODNet activities to include information regarding gastrointestinal illness and
water consumption (e.g., quantity of unboiled tap water consumed, quantity of bottled
water consumed, public or private water supply, approximate fraction of water consumed
at various sites - home vs. work, etc.)
2. Willingness to work collaboratively with EPA, CDC, and other EIP sites in the
formulation of questions for cross-sectional surveys
3. Use of clinical signs as outcomes for gastrointestinal illness [e.g., a) clinically defined
diarrhea, b) vomiting].
B. Recipient Activities for Feasibility Evaluation of Household Intervention Studies
1. Work collaboratively with CDC and EPA to identify intervention devices that can
eliminate viable pathogens from wafer, can be altered to be ineffective in a blinded.
manner, do not change participant drinking habits, and that can be used in a limited
feasibility study. Evaluate the cost of obtaining the intervention devices and installing and
maintaining the devices.
2. Identify a cohort of randomly or systematically selected households of
immunocompetent persons with at least one child in the household to participate in a
feasibility study
3. Conduct a limited, blinded household intervention trial (feasibility study) to evaluate the
effectiveness of blinding participants to .the effectiveness of the intervention device,
logistical considerations regarding maintenance of the intervention device, frequency of
follow-up for data collection and collection of specimens (if desired). At the end of the
study provide cost estimates for conducting specific portions of the intervention trial (e.g.
data and specimen'collection, specimen testing, etc.)
4. Measurement of outcomes during the feasibility study - Examples of such outcomes
could include: a) clinically defined diarrhea, b) vomiting, c) laboratory studies of stool
from cooperative, ill participants which would be tested broadly for bacterial, parasitic,
and viral pathogens, and/or d) antibody response to specific pathogens such as
Cryptosporidium and Calici viruses in study participants willing to give specimens (serum,
saliva).
5. The recipient(s) will be required by EPA to document their Quality Assurance Project
Plan (QAPP) for assuring that the environmental data collected are of the expected quality
for their intended use and are properly assessed.
6. Work with CDC and EPA in identifying specific characteristics (e.g. water supply)
would constitute a "good site" for doing future household intervention studies.
36
-------
;r Studies contributing to determining the burden of waterbome disease may also be
ed. Examples of such studies include case-control studies using cases of severe outcomes
gastroenteritis or community intervention studies to evaluate changes in the incidence of
:ome (e.g. gastrointestinal illness, antibody response to a specific pathogen) before and
change in the community's water treatment.
37
-------
Appendix D. Causality of an Epidemiological Association
For each epidemic logical study, the validity of the association observed between
exposure and disease must be assured before its causality can be evaluated. Because of the
observational nature of epidemiology, no single study can provide a definite answer about
causality even if systematic bias is minimal. A body of evidence must be obtained from studies
conducted in different geographic areas and populations. Guidelines are available to help
epidemiologists assess the possible causality of associations observed in well-designed and
conducted studies. The interpretation of epidemiological data should be made with caution and
in the context of all available scientific information. Epidemiologists apply the following criteria
to assess evidence about causality (Beagle et a/., 1993; Rothman, 1986; Hill, 1965):
- Biological Plausibility. When the association is supported by evidence from clinical research
or toxicology about biological behavior or mechanisms, an inference of causality is strengthened.
- Temporal Association. Exposure must precede the disease, and in most epidemiological studies
this can be inferred. When exposure and disease are measured simultaneously, it is possible that
exposure has been modified by the presence of disease.
- Study Precision and Validity. Individual studies which provide evidence of an association are
well designed with an adequate number of study participants (good precision) and well
conducted with valid results (i.e., the association is not likely due to systematic bias).
- Strength of Association. The larger the RR or OR the less likely the association is to be
spurious or due to confounding bias. HoweVer, a causal association cannot be ruled out simply
because a weak association is observed!
- Consistency. Repeated observation of an association under different study conditions supports
an inference of causality, but the absence of consistency does not rule out causality.
- Specificity. A putative cause or exposure leads to a specific effect. The presence of specificity
argues for causality, but its absence does not rule it out.
- Dose-Response Relationship. A causal interpretation is more plausible when a risk gradient is
found (e.g., higher risk is associated with larger exposures).
- Reversibility. An observed association leads to some preventive action, and removal of the
possible cause leads to a reduction of disease or risk of disease.
38
------- |