United States       Science Advisory       EPA-SAB-DWC-99-016
      Environmental       Board (1400)           August 1999
      Protection Agency   Washington DC        ivww.epa.gov/sab

&EPA AN SAB REPORT ON THE
      NATIONAL CENTER FOR
      ENVIRONMENTAL
      ASSESSMENT'S COMPARATIVE
      RISK FRAMEWORK
      METHODOLOGY
      A Review by the Drinking Water
      Committee

-------
                   UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
                                 WASHINGTON, D.C. 20460
                                                          OFFICE OF THE ADMINISTRATOR
                                                          SCIENCE ADVISORY BOARD
                                  August 12, 1999
EPA-SAB-DWC-99-016
The Honorable Carol Browner
Administrator
United States Environmental Protection Agency
401 M Street, SW
Washington, DC 20460
             Subject:
An SAB Report on the National Center for Environmental
Assessment's Comparative Risk Framework Methodology
Dear Ms. Browner:
      This Report was developed by the Drinking Water Committee (DWC) of the Science
Advisory Board (SAB) in response to interactions with the Agency's National Center for
Environmental Assessment (NCEA) during the December 1998 and February 1999 DWC
meetings.  This Report provides the SAB's reactions to this important effort.

      The DWC was pleased that the Agency has taken an important step in beginning the
development of an integrated and structured approach for considering complex environmental
issues.  NCEA illustrated the methodology by evaluating alternative treatment approaches for
reducing endemic Cryptosporidium risk relative to chemical risks associated with drinking water
treatment. We believe that these approaches are needed both from the perspective of
science, as well as from the perspective of good environmental public policy making in that
they will help to make that process more transparent to the individuals who make up our
democratic society.

      Even though the Committee is encouraged by the progress on this evaluative
approach, it is important to recognize that significant work remains to be done on the
methodology.  Much of the prior work that has been done is based on narrow evaluations of
very specific medical and public health interventions.  It is not clear that the approaches for
developing common metrics in these settings are appropriate for making environmental health
decisions. A specific issue in this regard is the need to insure that the metrics are based on
the opinions of an informed public, not just professionals, scientists, or groups that have
special interest in the problem. Also, significant problems exist when we bring forth diverse
sets of seemingly relevant health risk data in a form that can be aggregated for evaluation

-------
using common metrics. As the DWC's review of the case study demonstrated, significant
efforts will need to be focused on the peer review of the individual elements of these analyses
(i.e., microbial risks, chemical risks, and the reliability of metrics used), as well as the overall
outcome, to ensure that critical information is not lost or misused in such integrative
assessment approaches.

      The DWC found that the case study provided to illustrate the application of this
methodology was very useful for identifying weaknesses in the methodology that are obscured
by the somewhat dry textual descriptions of the approach.  Despite these concerns we
encourage EPA to further explore this methodology with additional case studies in order to
help it identify additional areas for improvement. The Committee stands ready to provide
additional review and assistance as EPA further develops the basic approach that was
outlined by EPA in the documents submitted for this review.  We look forward to the response
to these comments from the Assistant Administrator for the Office of Research and
Development.
                                 Sincerely,
      /signed/                                             /signed/
Dr. Joan M. Daisey, Chair                             Dr. Richard J. Bull, Chair
Science Advisory Board                               Drinking Water Committee
                                                    Science Advisory Board

-------
                                       NOTICE
       This report has been written as part of the activities of the Science Advisory Board, a
public advisory group providing extramural scientific information and advice to the
Administrator and other officials of the Environmental Protection Agency. The Board is
structured to provide balanced, expert assessment of scientific matters related to problems
facing the Agency.  This report has not been reviewed for approval by the Agency and, hence,
the contents of this report do not necessarily represent the views and policies of the
Environmental Protection, nor of other agencies in the Executive Branch of the Federal
government, nor does mention of trade names or commercial products constitute a
recommendation for use.
Distribution and Availability: This Science Advisory Board report is provided to the EPA
Administrator, senior Agency management, appropriate program staff, interested members of
the public, and is posted on the SAB website (www.epa.gov/sab).  Information on its
availability is also provided in the SAB's monthly newsletter (Happenings at the  Science
Advisory Board). Additional copies and further information are available from the SAB staff.

-------
                                    ABSTRACT

      The Drinking Water Committee (DWC) of the Science Advisory Board (SAB) reviewed
a methodology developed by the US Environmental Protection Agency's (EPA) National
Center for Environmental Assessment, Cincinnati (NCEA) entitled Comparative Risk
Framework Methodology and Case Study. The document presents a methodology intended
for analyzing, and describing in comparable terms, disparate health risks associated with
alternative drinking  water treatment approaches. The Committee supported the continued
development of this method and the research necessary to allow its further development.

      The Committee noted that the proposed methodology presents a potentially powerful
tool that provides a  structural framework for identifying important variables that influence the
nature and extent of complex environmental  problems.  The case study that was conducted to
illustrate the method's application, while demonstrating its promise, highlighted the difficulties
that can be anticipated when such a framework is applied.  The Committee suggested that with
further development, the Comparative Risk Framework Methodology has the potential to
provide valuable insights to officials responsible for local and national decisions on the most
appropriate intervention to apply to control human health risks associated with drinking water.
The text of the report provides advice that highlights the further efforts that will be necessary
for its development  and use by the Agency.

Keywords: Comparative risk, economic metrics, common metrics, risk-risk tradeoffs,  drinking
water risk comparisons

-------
                       U.S. Environmental Protection Agency
                              Science Advisory Board
                             Drinking Water Committee
                  Comparative Risk Framework Methodology Panel

Chair
DR. Richard J. Bull, Battelle Pacific Northwest Laboratories, Richland, WA

Members
Dr. David B. Baker, Water Quality Laboratory, Heidelberg College, Tiffin, OH.
Dr. Mary Davis, Department of Pharmacology & Toxicology, Robert C. Byrd Health Sciences
      Center, West Virginia University, Morgantown, WV
Dr. Yvonne Dragan, Ohio State University, Columbus, OH
Dr. John Evans, Program in Environmental Science and Risk Management Harvard School of
      Public  Health, Cambridge,  MA.
Dr. Anna Fan-Cheuk, California Environmental Protection Agency, Office of Environmental
      Health  Hazard Assessment, Oakland,  CA
Dr. Lee D. McMullen,  Des Moines Water Works, Des Moines, IA.
Dr. Christine Moe, Assistant Professor, Department of Epidemiology University of North
      Carolina, McGavran-Greenberg, Chapel Hill, NC
Dr. Charles O'Melia, Department of Geography and Environmental Engineering, The Johns
      Hopkins University, Baltimore, MD
Dr. Gary A. Toranzos, Department of Biology, University of Puerto Rico, San Juan, PR
Dr. Rhodes Trussell, Montgomery Watson, Pasadena, CA
Dr. Marylynn V. Yates , Department of Soil &  Environmental  Sciences, Geology, University of
      California, Riverside, CA

Consultants
Dr. Judy A. Bean, Children's Hospital Medical Center, Cincinnati, OH.
Dr. Lenore S.  Clesceri, Rensselaer Polytechnic Institute, Materials Research Center, Troy, NY
Dr. Richard Gilbert, Battelle Washington, Washington, DC
Dr. Winston Harrington , Resources for the Future, Washington, DC
Dr. Edo D. Pellizzari, Research Triangle Institute, Research  Triangle Park, NC
Dr. Joel G. Pounds, Institute of Chemical  Toxicology, Wayne State University, Detroit, Ml.
Dr. Verne A. Ray,  Medical Research Laboratory, Pfizer Inc.,  Groton, CT

Science Advisory Board Staff
Mr. Thomas O. Miller, Designated Federal Officer, US EPA, Science Advisory Board, 401 M
      Street,  S.W. (1400), Washington, DC
Ms. Dorothy M. Clark, Management Assistant, US EPA, Science Advisory Board, 401 M
      Street,  S.W. (1400), Washington, DC
                                        in

-------
                              TABLE OF CONTENTS

1.  Executive Summary	1

2.  Introduction and Charge 	3
      2.1.  Introduction	3
      2.2.  Charge  	6

3.  Comments on the Framework	9
      3.1.  General Comments on the Comparative Risk Framework Methodology	9
      3.2.  Specific Comments on the Comparative Risk Framework Methodology	13
             3.2.1.  Combining Risk Assessment and Cost-Effectiveness Approaches   . . 13
             3.2.2.  Step 1: Initial Risk Assessment	15
             3.2.3.  Step 2: Translation of Risk to Health Conditions  	16
             3.2.4.  Step 3: Translation of Conditions to a Common Metric	19
             3.2.5.  Step 4: Cost Evaluation  	25

4.  The Case Study	26
      4.1.  General Issues for the Case Study	26
             4.1.1.  Is the Case Study an Appropriate Application of the Framework
                   Methodology	26
             4.1.2.  Statistical Approaches and Uncertainty  	28
             4.1.3.  Case Study  Limitations	30

5.  Results	31
      5.1.  Are the Results Presented Appropriate	31

6.  Research Needs  	32
      6.1.  Are the Research Needs Clearly Defined	32
      6.2.  Research Needs to Support Continued Development of the CRFM	33
      6.3.  Research Specific to the Problem Evaluated  	34

7.  General  Recommendations	35
      7.1  Other Specific Recommendations  	35

Appendix A. Specific Comments on the Case Study	A-1

1 .Engineering/Water Treatment	A-1
      a. Treatment Scenarios and Cost	A-1
      b. Treatment System Efficacy  	A-2

2.  Risk Characterization  	A-4
      a. Cancer Risks 	A-4
      b. Microbiological Risks 	A-5

3.  Chemical Dose-Response Assessment	A-8
      a. Appropriateness of the  Discussion of Mechanistic Toxicity for Chemicals	A-8
      b. Dose-response Analysis of Individual DBPs  	A-9

                                         iv

-------
      c. Appropriateness of Assumptions, Techniques and Models for DBFs	A-9
      d. Usefulness of Mixtures Risk Assessment Approach  	A-10
      e. Appropriateness of Assessment of Risks of TOX  	A-11
      f.  Appropriateness of DBFs Addressed 	A-11
      g. Use of Epidemiology Data 	A-12

4.  Exposure 	A-13
      a. General Issues Related to the Evaluation of Exposure  	A-13
      b. Distribution of Drinking Water Consumption Rates  	A-14
      c. Unheated Drinking Water Fraction	A-15
      d. Pathogen Survival Relative to Preparation Methods  	A-15
      e. DBP Changes in the Distribution System	A-15
      f.  Other Routes of Exposure	A-16

5.  Health Conditions	A-16
      a. Health Conditions and Cryptosporidiosis	A-16
      b. Appropriateness of Developmental, Reproductive and Cancer Risks  	A-16
      c. Uncertainties in the Process	A-17

6.  Common Health Metric 	A-18
      a. QALY Appropriateness	A-18
      b. QALY Assignment 	A-18

-------
 AN SAB REPORT ON THE NATIONAL CENTER FOR ENVIRONMENTAL ASSESSMENT'S
                 COMPARATIVE RISK FRAMEWORK METHODOLOGY
                              1. EXECUTIVE SUM MARY
      The Drinking Water Committee (DWC) of the Science Advisory Board (SAB) reviewed
a methodology developed by the US Environmental Protection Agency's (EPA) National
Center for Environmental Assessment, Cincinnati (NCEA) entitled Comparative Risk
Framework Methodology and Case Study. The document presents information on a
Comparative Risk Framework Methodology (hereafter, the CRFM or the methodology), that
would be used for analyzing, and describing in comparable terms, disparate health risks
associated with alternative drinking water treatment approaches. The Committee appreciates
the efforts of the NCEA staff in developing this integrated approach for evaluating complex
environmental issues and supports the further development of this methodology and instituting
the research necessary to support its application by the Agency.

      The proposed methodology is a potentially powerful tool that provides a structural
framework for identifying  important variables that influence the nature and extent of complex
environmental problems.  The case study that was conducted to illustrate the application of
such a methodology served well to identify the promise of the approach as well as the
difficulties that can be anticipated when such a framework is applied. With further
development, the Comparative Risk Framework Methodology has the potential to provide
valuable insights to officials responsible for local and national decisions on the most
appropriate intervention to apply to control human health risks associated with drinking water.
This SAB report provides advice on the methodological and data development efforts that will
be necessary to further develop this approach for more extensive use by the Agency.

      One important need is for the Agency to more clearly and concisely describe how the
issue of uncertainty in the model and in data used in the model is accommodated throughout
the method.  The Agency should provide a complete and detailed description of the methods
used to conduct the uncertainty and sensitivity analyses.  Specifically, it should state how
parameters are determined to be uncertain or variable, the methods used to propagate the
uncertainty of the model structure, the methods used to propagate the uncertainty or variability
of model parameters in order to approximate the cumulative probability distribution of the
model output. An outcome  of this  analysis should be guidance on how to determine which
parameters contribute the most to  the overall uncertainty in the model output.

      It is extremely important that the Agency abandon the use of upper bounds of risk in
these analyses. Central estimates should be used or systematic error will be introduced into
the overall risk analysis. This does not preclude the use of confidence  intervals in the
uncertainty analysis.

-------
       Additional important issues to be made a part of the CRFM through methods, guidance,
or data development are:

       a) a way to add methods that specifically address the risk of disease outbreaks from
       drinking water system upsets as well as chronic low-level microbial risk,

       b) additional definition on the processes used to convert continuous data into risk
       estimates and then to translate risks into health conditions recognizable to the lay
       public,

       c) research that could establish valuations of health conditions by the public (e.g.,
       Quality Adjusted  Life Year-QALY) based on infectious disease morbidity and mortality
       associated with an environmental risk.  There is a need to broaden evaluations in
       environmental health rather than relying on prior work with medical interventions and
       preventive medicine.

       d) a better method for dealing with acute nonfatal disease in the derivation of the
       common metric,

       e)  more succinct guidelines for translating adverse health effects from animal
       experiments to human conditions, especially guidance on how mechanism or mode of
       action is to be used  in these translations.

       f) better foundation for the criteria used to determine incremental exposures from
       disinfection including disinfection byproduct (DBP) identification, distribution of
       disinfection byproduct concentrations over time and space, data aggregation for scaling
       up to the  national level, and use of contaminant occurrence information in risk
       assessment,

       g) a way to consider source water characteristics as a major variable in analyses that
       go beyond consideration of single water supplies,

       h) ways to consider  microbial risks in significantly greater depth. Cryptosporidium is not
       a good prototype for other waterborne pathogens due to its resistance, its behavior in
       the distribution system, and its minimal risk for secondary spread (also, different
       classes of microbes have differing sensitivities to different interventions, and different
       individuals have different susceptibility to disease (genetic susceptibilities and
       generalized susceptibilities such as those based on age).

       Additional specific comments are included in the sections that follow in this report of the
Science Advisory Board's Drinking Water Committee.

-------
                          2.  INTRODUCTION AND CHARGE

2.1  Introduction

      The Drinking Water Committee (DWC) reviewed the US Environmental Protection
Agency (EPA) National Center for Environmental Assessment, Cincinnati (NCEA) document
entitled Comparative Risk Framework Methodology and Case study. The document presents
information on a comparative risk framework methodology (hereafter, the CRFM or the
methodology) that would be used for analyzing, and describing in comparable terms, disparate
health risks associated with alternative drinking water treatment approaches.

      The NCEA made a number of suggestions about what they consider to be the utility
and importance of the methodology. They suggested that:

      a)  "Use of the CRFM should be beneficial to the local water purveyor who must
      evaluate treatment options."

      b)  "Use of the CRFM should also be beneficial to risk assessors and managers at the
      national level who must develop data and draft regulations."

      c)  "The CRFM...can help provide a sound scientific basis for determining whether or
      not to go beyond the November 1998 Stage 1 DBP rule to additional regulations for
      DBPs or microbes."

      d)  "EPA believes that the proposed  CRFM will assist the Agency in determining the
      balance between adequate water treatment to control and minimize microbial risk and
      the creation  of unacceptably high  levels of countervailing risks from DBPs.  The CRFM
      presented here is intended to support and strengthen traditional and existing risk
      assessment and risk management activities."

      The CRFM integrates cost-effectiveness analysis, as applied to public health
interventions, with the 1983 NAS Risk Assessment Paradigm. Figure  1 provides a graphic
overview of the basic CRFM.  First,  traditional risk assessments are conducted for the
microbial and DBP contaminants in treated water.  Next, to compare these chemical and
microbial risks, the effects or consequences described in the risk characterization are
translated  or expressed in terms of measurable human health conditions, such as cases of
cancer, and infection and illness from infectious diseases.  The range of potential human
health conditions are then converted into a common health metric. In this framework, a Quality
Adjusted Life Year (QALY) is used as the common metric to capture changes in the length and
quality of life associated with the different health conditions.  In the final stage of the
framework, alternative strategies are compared by assessing their expected impact on health
and economic outcomes.

      In addition, the document includes a case study to illustrate how the methodology  might
be applied to evaluate the health risk tradeoffs associated with two alternative water treatment
approaches that focus on reducing risks from the protozoan, Cryptosporidium parvum.
According  to the authors, "The case study is not intended to provide definitive answers

-------
because changes to the specific assumptions used in the case study could alter the results."
However, it

-------
              Chemical
              Risk
              Assessment
              Intervention
              Cost
                       Alternative
                       Intervention
                       Strategies
Assessment
    of
Intervention
   Cost
              Microbial
              Risk
              Assessment
                                                            Health
                                                           Condition
                                                                             Common
                                                                              Health
                                                                              Metric
                     Cost
                  Associated
                  With Health
                   Condition
    Cost
Effectiveness
  Analysis
                                                                 Health
                                                               Condition
"...can be
                             Figure 1.  Comparative Risk Assessment Framework Overview (NCEA, 1998).

                                                         5

-------
used to recognize which parameters and assumptions most affect analysis results, while
identifying data gaps and uncertainties."

       To consider the potential utility of the CRFM, it is important to understand the statutory
process that EPA must follow to regulate drinking water contaminants. First, and foremost,
EPA must ensure that drinking water treatment is protective against waterborne microbial
diseases.  This leads to treatment requirements that control waterborne pathogens in
processed water and in the systems used to distribute this drinking water.

       EPA is also required to control risks from  byproducts of the drinking water treatment
process (i.e., disinfection byproducts-DBPs). The EPA Administrator must publish a
maximum contaminant limit goal (MCLG) and promulgate a National Primary Drinking Water
Regulation (NPDWR) for physical contaminants:

       a) that may have an adverse effect on the health of persons,

       b) that occur in public water systems with a frequency and at levels of public health
       concern, and

       c) for which regulation presents a meaningful opportunity for health risk reduction.

       MCLGs are to be set at a level at which no known or anticipated adverse health effects
occur (allowing for an adequate margin of safety).  The regulation must specify a Maximum
Contaminant Level (MCL) which is set as close to the health goal (MCLG) as is feasible with
the use of best technology, treatment techniques, and other efficacious means available while
taking cost into consideration.  If the determination of contaminant levels is not economically or
technologically feasible, EPA can  promulgate a Treatment Technique in lieu of an MCL in its
regulation. The Treatment Technique must prevent adverse health effects to the extent
feasible.

       To deal with risk/risk trade-off concerns, the Administrator is permitted to establish an
MCL for a contaminant at other than a feasible level if the technology, treatment technique, or
other means used to determine a feasible level would result in an increase in health risk by
increasing concentrations of other contaminants in drinking water or by interfering with the
efficacy of other treatment techniques  or processes used to comply with drinking water
regulations.

       Finally, if a regulation specifies an MCL, EPA must publish and seek comments on
analyses that EPA conducts on:

       a) quantifiable and nonquantifiable  health  risk reduction benefits likely to occur as a
       result of treatment to comply with the level,

       b) quantifiable and nonquantifiable  health  risk reduction benefits likely to occur as a
       result of reductions in co-occurring  contaminants associated with the new regulation,

       c) quantifiable and nonquantifiable  costs likely to occur as a result of treatment to
       comply with the level (monitoring, treatment, other),

-------
       d) incremental costs and benefits associated with each alternative MCL considered,

       e) effects of the contaminant on the general population and groups within the general
       population (infants, children, pregnant women, elderly, individuals with history of
       serious illness, other subpopulations at greater risk),

       f) any increased health risk as a result of compliance (including risks from co-occurring
       contaminants), and

       g) other relevant factors (quality/extent of information, uncertainties in the analyses in
       these special analyses, factors with respect to the degree and nature of the risk).

2.2 Charge

       The Agency charge to the Drinking Water Committee asked for comments on both the
methodology and the case study.  Generally, the Agency asked if:

       a) the purposes of these two components are clearly presented and the document's
       objectives/ purposes consistent with the rest of the document? Do the materials in
       Chapters 1 and 2 sufficiently orient readers to the rest of the document?  Is the
       information presented in the document clear?

       For the Comparative Risk Framework Methodology section, the Agency asked if:

       a) it is appropriate to combine the  MAS 1983 risk assessment paradigm with a cost
       effectiveness analysis approach?  Further, they asked if the method would be useful for
       environmental and public health decision makers faced with alternative intervention
       technologies and disparate risks, and if the methodology was presented in a logical
       format?

       b) initial application of the MAS 1983 risk assessment paradigm to evaluate risks from
       chemicals and microbes, separately is an appropriate first step, and if other steps
       should  be evaluated?

       c) the second step, translating predicted, separate, risks into human health conditions
       which are measurable on the population level is appropriate, or if other interim steps
       should  be included?

       d) the third step, translating human health conditions into a common metric, is
       appropriate or if other interim steps should be included?

       e) the final step, evaluating the financial costs of treatment technologies,  medical costs,
       and a cost effectiveness ratio is appropriate, or if there are other interim steps that
       should  be evaluated?

       The Agency prepared the case study to demonstrate the use of the CRFM to evaluate
and compare realistic, central tendency risk estimates of  both DBFs and one microbial

-------
contaminant, Cryptosporidium parvum, across various treatment technologies.  EPA asserts
that the case study uses hypothetical but plausible assumptions for source water
concentrations of Cryptosporidium, the efficacy of each treatment system in eliminating this
pathogen, concentrations of selected DBFs in finished drinking water for each treatment
system, the size of the population served by the facility, the size of the immunocompromised
group within the general population, the distribution of drinking water consumption rates, dose-
response curves for the DBFs and C. parvum exposures, and the financial costs of the
treatment options.  EPA used a response addition model in the case study because of data
limitations on the mechanism of action for the majority of DBPs. Other approaches, for
example, another mixtures approach such as dose addition or an assessment of individual
chemical constituents of the mixture, could have been developed.

       For the case study, the Agency asked if:

       a) this is an appropriate application of this methodology; the initial assumptions are
       plausible; other factors should be considered within the  scope of the case study?

       b) the statistical  methods and assumptions used  to develop parameter distributions are
       reasonable  and  correctly applied; the simulation procedures and sensitivity analyses
       are appropriate for the data; there are other statistical techniques that could be applied
       to improve the CRFM?

       c) the limitations of the case study are appropriate and clearly presented? There are
       other significant limitations not identified?

       d) the choices of water treatment scenarios for evaluation relative to the "real world"
       scenarios are appropriate? The construction and operation cost estimates adequately
       reflect real-world facilities and point-of-use device costs?

       e) the accuracy of water treatment system  efficacy estimates, including the assumed
       distribution are accurate and the treatment system assumptions concerning DBP
       concentrations in finished drinking water are appropriate and reasonable?

       f) overall, the cancer risk estimates, based on exposures to pathogens and DBP
       mixtures, are plausible and reasonable given the goals and limitations of the case
       study; the uncertainties presented in the assessments of pathogens and DBP mixtures
       are adequately described and characterized;  there are significant uncertainties that are
       not identified; the presentations and discussions of variability appropriate?

       g) the assessment of microbial risk based on  only a single type of microorganism,
       Cryptosporidium parvum, is appropriate; other organisms should be evaluated with this
       approach in future applications of the CRFM; the  identified hazards related to exposure
       to the protozoan parasite, and the probabilities of infection and disease conditional on
       exposure appropriate for the two populations considered (i.e., the general population
       and the AIDS subpopulation), are appropriate; all possible outcome categories have
       been identified; the scientific basis of the dose-response model for predicting pathogen
       risks is well-founded and adequately described and supported; other subpopulations
       (e.g., the elderly) should be evaluated in future applications of the CRFM?

                                          8

-------
h) the discussion of mechanistic toxicity data for individual chemicals and/or mixtures is
appropriate?

i) the assumptions, techniques and models used in the estimation of risks posed by
DBP mixtures, including both statistical and biological considerations, are appropriate?

j) the mixtures risk assessment approach for comparing health risks across drinking
water treatment scenarios and levels of DBFs is useful?

k) the assessment of risks posed by the unidentified halogenated fraction (TOX) is
appropriate?

I) the appropriate DBFs have been addressed and if it is appropriate to not address the
carcinogenicity of chloroform in the analysis because of the evidence (Drinking Water
NODA, EPA,  1998) indicating that this chemical's mechanism of action exhibits a
threshold?

m) the DWC would offer suggestions for the use of epidemiologic data in future
applications of the methodology?

n) the distribution of drinking water consumption rates are appropriate?

o) the identification of the fraction of unheated drinking water consumed is
appropriate?
p) the assumption that the pathogen would not survive the pathways from tap to
consumption  is valid?

q) the validity and significance of the assumption that DBP concentrations do not
change as a result of transport through the distribution system and many pathways
through which water is consumed (e.g., boiling to prepare tea)?

r) other routes of exposure should be included (e.g., dermal and inhalation routes) and
whether their contribution would be significant?

s) the assignment and definition of health conditions are appropriate for the
progression of cryptosporidiosis?

t) the assignments and definition of health conditions for developmental, reproductive
and cancer risks are appropriate and if latency periods and reversibility issues are
handled reasonably?

u) the uncertainties in the process are identified?

v) the use of Quality Adjusted Life Years (QALYs) is appropriate?

w) the assignment of QALYs given the health conditions are appropriate?

x) the uncertainties related to the use of QALYs in this application are identified?

                                    9

-------
       y) the results are presented appropriately and the interpretations of the results are
       appropriate? Additional discussion items should be presented?

       z) the research needs are clearly defined?
                        3. COMMENTS ON THE FRAMEWORK

3.1  General Comments on the Comparative Risk Framework Methodology (CRFM)

       The draft Comparative Risk Framework Methodology (CRFM) is among the first
methods to systematically compare the risk of microbial diseases with diseases that could
occur from exposure to disinfection byproducts (DBFs). Thus, the CRFM has the potential to
provide a generally useful context for evaluating decision inputs by individual communities, or
for EPA risk managers,  by comprehensively framing the impacts of alternative forms of
regulatory intervention.  When fully developed, the methodology will provide transparency and
an integrating structure to the analytical process in support of decision making. It will also
provide valuable insights for local and national decision makers who must decide upon
alternative interventions that can be applied to deliver safe drinking water. For now,  though,
significant additional research and development will be necessary for both the methodology
and for case specific elements that serve as inputs to the methodology.

       Figure 2 shows the CRFM structure as viewed by the Drinking Water Committee. Seen
this way, the framework consists of several layers and,  like computer software, individual
kernels in each layer can be developed and examined independently.  The methodology
section of the EPA document addresses the uppermost layers, Levels 6 and 7. The  case
study illustrates how the requirements outlined in Levels 1 through 5 might be fulfilled for a
particular situation.

       The CRFM is a potentially powerful tool that is well suited to helping identify important
variables that are presented by complex environmental problems. The methodology can be a
significant advance for evaluating environmental problems and the Committee is enthusiastic
about the overall analytical structure that the framework provides, in particular, the concepts
described in Levels 6 and 7 (i.e., the description of the "quality-adjusted life year"--QALY
method). For these levels, major remaining tasks are to compare alternative choices for the
common health metric to determine which is most suitable, and then to develop defensible
weighting parameters for the metric once selected.  The description of the QALY method for
comparing health risks is clear and the arguments for the use of some common measure of
health impact reflecting  both longevity and quality of life are persuasive. Even though the
Committee is not questioning whether the QALY is the most suitable metric, it believes that
alternatives should be considered, adequately discussed in the document, and a clear case
made for the metric selected.

       The case study served well to identify both the promise of the CRFM and the
complexities that can be anticipated in its application.  However, before the method is adopted
for the analysis of  national policy issues (such as the microbial risk/disinfection byproduct risk
issue) further methodological development will be necessary to ensure that the complexity
                                         10

-------
introduced by site-to-site variation in raw water quality, treatment costs, and community values
can be accommodated.  Refinement of Levels 6 and 7, alone, will not be sufficient for
successful application.  Data and methodology gaps will also need to be filled at Levels 1
through 5. Fortunately the bulk of the nation's current DBP/Disinfection research is directed at
gathering information that will support the development of methodologies at levels 1 through 4.

       The CRFM uses a common metric to compare all potential risks associated with
situations being evaluated.  However, there is a need for research to develop measures that
are valid for environmental health issues. Methods to translate the risks estimated in Level 4
into the probability distribution function (PDF) of health outcomes required in Level 5 will be
critical (especially the identification of relevant health outcomes in human populations on the
basis of data from animal studies). Another key issue needing further attention is the
treatment of uncertainty ~ particularly model uncertainty.

       Many CRFM applications will require analysts to predict human disease outcomes
based on evidence from animal toxicology studies.  Because the "policy importance" of a
disease in the QALY approach is determined, in part, by the loss of quality adjusted life
expectancy associated with the disease, the correct identification of  disease conditions will be
important. However, the relationship between the effects seen in animal studies and the
diseases apparently detected in human populations exposed to DBFs is not always clear.  For
example, the case study assumed that the diseases resulting from human exposure to
disinfection byproducts would be bladder and rectal  cancer ~ i.e., the diseases which have
been seen in human epidemiological studies.  However, the quantitative risk estimates for
these diseases were based on extrapolation of results from animal studies which, instead of
observing these disease outcomes, observed liver and kidney cancers in study animals. It is
not clear that these outcomes from animal studies will adequately extrapolate and translate to
the equivalent QALY weights in humans.

       The Committee understands and agrees with the logic underlying the decision to
compute the present value of health impacts.  Further, we believe that in the current case
study the particular choice of a discounting  rate would not substantially influence the results.
However, in other cases the choice of the discounting rate is likely to be important. Because
of this decision, and in view of both the wide-spread misunderstanding of the rationale for
discounting health impacts, the legitimate ethical arguments surrounding this  issue, and the
political sensitivity of the issue, we recommend that the document include a more prominent
and nuanced discussion of the general issue and that it provide stronger support for the
particular choice of  3% as the discount rate used in the analysis underlying the case study.

       The Agency document does not describe the methods used to conduct the uncertainty
and sensitivity analyses in sufficient detail.  A complete and detailed description should be
provided that discusses how parameters are determined to be  uncertain or variable, the
methods used to propagate the uncertainty of the model structure, the methods used to
propagate the uncertainty or variability of model parameters in  order to approximate the
cumulative probability distribution of the model output, and the  methods used to determine
which parameters contribute the most to the overall uncertainty in the model output.  The
Agency may wish to place the detailed description in an appendix. However, the major steps
of the process should be explained in the body of the report with  the aid of flow-charts or
similar graphical methods.

                                          11

-------
  Level 3
                                    Cost Effectiveness Analysis
                      Common Health Metric
                       II
      II
                 PDF* [Health
                 Conditions] -
                 from exposure
                 to chemicals
                       PDF* [Health
                       Conditions] -
                      from exposure
                       to microbes
                       II
                            II
Chemical Risk
 Assessment
                                       Microbial Risk
                                        Assessment
                       II
                            II
    PDF of
  chemicals
    after
 intervention
   PDF of
microbes after
 intervention
                              II
                       II
                              Proposed
                             Intervention
                                  II
Quality of
Water
Source
State of the
Current
Treatment
State of the
Watershed
Case Conditions
                                                                          II
                                                     Cost of
                                                 Scenario Under
                                                  Examination
                                                              II
                     Sum[PDF*(Cost
                        of Health
                       Conditions)]
                                           Cost of proposed
                                             intervention
Figure 2. DWC Overview of the Proposed EPA Comparative Risk Framework Methodology

 *PDF = Probability Distribution Function
                                        12

-------
       Of special concern to the DWC is the lack of an adequate discussion, in both the
general methodology and the case study, of how to assess and incorporate the magnitude of
model uncertainty in the total uncertainty of the model outputs.  The case study focuses on
propagating uncertainty in some model parameters, but it appears that the model structure
itself is assumed to be known with certainty. This assumption is not realistic and some method
is needed for including model uncertainty in the uncertainty analysis.  Also, additional effort is
needed to separate and properly take into account the effects of variability across populations
and uncertainty in models and parameters. A general methodology for this process should be
developed and illustrated in the case study. Other issues that should be addressed in more
detail are the rationale for the selected probability distributions for uncertain parameter values
and variable quantities, the methods used to take into account the correlations among key
parameters of the main model and any submodels, the methods that should be used to obtain
expert opinion about alternative model structures, exposure scenarios, and the probability
distributions of uncertain or variable quantities, and when methods such as Latin Hypercube
Sampling for generating random numbers from specified probability distributions of variable or
uncertainty quantities are preferred to simple random sampling.  Just as important is that
parameters involved in assigning utility to changes in  health status, such as QALY weights,
may be uncertain and should be considered as such in the uncertainty/sensitivity analysis.

       The Committee suggests that the methodology be cast in the most general case.
Effects and exposure data vary and they generally follow a distribution rather than being
dichotomized. For example, sensitive populations should be  cast as a distribution of
sensitivities as the baseline  in the methodology.  As the CRFM is applied to specific cases, it
may be necessary to dichotomize the risks. For example, the dichotomization of human
sensitivity to Cryptosporidium into "normal" versus HIV-infected individuals in the case study
was appropriate because that was the nature of the data that were available (i.e., a
comparison of a specific population  relative to the norm). But it must be remembered that HIV-
infected individuals make up only a portion of the sensitive population. This had implications  in
the case study because treatment of sensitivity only in this selected population tends to
minimize the effectiveness of better water treatment.

       The methodology does not effectively address the problems of outbreaks of microbial
disease. Outbreaks are more visible to the public at large, and their effects differ in that they
are not just estimates, but are confirmed reality.  Therefore, the methodology should consider
the risks from outbreaks of waterborne disease separately from the risks of endemic
waterborne disease.  In general, endemic waterborne disease may be more related to
distribution system problems while outbreaks of waterborne disease are usually due to
treatment failures. Data from Payment's studies (1991, 1997) suggest that viruses may be
responsible for a large portion of endemic waterborne disease.  Outbreaks of waterborne
disease in the US are due to both protozoan and viral agents. Most identified etiologic agents
of waterborne disease outbreaks are protozoa (Giardia and Cryptosporidium). However, the
etiologic agent is never identified for approximately half of the waterborne disease outbreaks.
These outbreaks are  likely due to viral agents based on their  epidemiologic characteristics.
They likely are not identified because of our limited capacity to detect viral agents in clinical
and environmental samples.

       It is the Committee's opinion that omitting outbreaks will have real implications for the
risk manager. The measures that will most effectively address endemic disease are not

                                          13

-------
necessarily the measures that will most effectively address outbreaks. Reducing endemic
disease requires lowering the mean level of pathogens in the water supply over long periods of
time.  Preventing outbreaks requires improving the reliability of the barriers between the
sources of pathogens and the consumer's tap.

       The Comparative Risk Framework Methodology developed by the Agency is a broad
outline of an approach to compare risks of a disparate nature. However, application of the
method in an illustrative case study revealed that there are additional layers of infrastructure
needed that require careful consideration before the method can be successfully implemented.
The Committee  recommends that EPA conduct additional case studies of substantially
different conditions than those reflected in the existing case study, in order to ensure that
solutions are developed for the practical implementation  problems noted by the DWC.
Properly chosen, these case studies might provide useful insights for decision-making.  In this
effort, the DWC  recommends that the elements of the analyses (e.g., DBP risk, microbial risk,
and economics be subject to rigorous peer review.  This wold ensure that all pertinent
information is captured and that the uncertainties are appropriately propagated in the overall
application of the methodology

3.2 Specific Comments on the Comparative Risk Framework Methodology

       The subsections that follow address specific EPA charge  questions on the
methodology. In Section 4 of the report, we provide general comments on the case study.
Later, in Appendix A, we provide comments on details of the case study that are not necessary
to address in the body of the document, which addresses the Comparative Risk Framework
Methodology itself.

       3.2.1  Combining Risk Assessment and Cost-Effectiveness Approaches :
             The proposed CRFM in Chapter 4 combines the  NAS 1983 risk assessment
             paradigm with a cost effectiveness analysis approach. Is this combination
             appropriate? Will the CRFM be useful for environmental and public health
             decision  makers faced with alternative intervention technologies and
             disparate risks?  Is the CRFM presented in a logical format?  If not, please
             suggest an alternative organization.

       a) General:

       The Drinking Water Committee concluded that the combination of risk assessment and
a cost effectiveness analysis is appropriate in the methodology, and that it has the potential to
be quite useful in giving new insights to decision makers who are considering appropriate
types of environmental control, as well as to those prioritizing research efforts that target
critical decision making information.

       The Committee is aware that in many real cases the framework will fail to strongly
differentiate among the alternative control strategies (i.e., on the  basis of their cost-
effectiveness).  If good science and regulatory analysis cannot clearly differentiate among
strategies showing one to be superior to another on the basis of  economic efficiency, then the
decision maker may have a legitimate basis for making the decision on other grounds.  It was


                                         14

-------
apparent to the Committee that authorizing legislation may preclude the application of the
CRFM in some situations. For example, it is inappropriate under the Safe Drinking Water Act
(SDWA or the Act) to establish maximum contaminant level goals (MCLGs) for a contaminant
in this way. The MCLG is to be established strictly on the available health effects information,
based on agreed upon policies. On the other hand, the method would be suited to identify
issues that are important to establishing a Maximum Contaminant Level (MCL),  since the Act
allows for consideration of a variety of practical factors in setting the MCL. An example would
be a lack of an available (affordable) treatment process  to meet the MCLG. The methodology
would also have an application in determining the alternatives that would be pursued in
obtaining variances or exemptions from MCLs.

      The Committee endorses comprehensive frameworks that integrate risk  analysis within
decision making, and that consider simultaneously the health consequences and the control
costs of decisions.  However, this endorsement does not imply that we believe cost-
effectiveness analysis to be superior to other approaches such as benefit-cost  analysis.  In
fact, we believe that increased communication among the EPA offices that have used or are
considering comprehensive frameworks would be beneficial. This might encourage cross-
fertilization of ideas and consistency in the approaches taken to evaluate regulatory  programs.

      b) Specific Comments:

      EPA's Office of Policy (OP, formerly the Office of Policy, Planning, and Evaluation) has
a long history of leadership in developing and applying a comprehensive economic framework
to analyze public decisions-specifically, they have worked to develop benefit-cost methods for
evaluating proposed environmental regulations. Conceptually, that method values an
improvement to an individual's health status as the  maximum sum of money the individual is
willing to pay for the improvement.

      The CRFM differs from the Office  of Policy's typical regulatory analyses  because it
relies on cost-effectiveness analysis rather than benefit-cost analysis.  Only human health
effects are considered, and no attempt is made to monetize  those effects. Instead, all health
effects are valued using a common health metric. Each of the two approaches  has strengths
and weaknesses and the use of QALYs does not preclude the use of benefit-cost analysis. A
discussion among representatives from NCEA, OP, and potential users of the method could be
helpful in establishing the place for each approach.

      The innovative aspect of this proposal is not on this risk assessment approach perse,
but in putting diverse information on risk into a framework better suited for decision-making.
Essentially, this involves developing a common metric - Quality Adjusted Life Year
(QALY-discussed later). While this does not change the fundamentals of the MAS 1983 risk
assessment paradigm, one must be sure that the process is  internally consistent with arriving
at outputs  that are compatible with the development of the common  metric.

      The CRFM team has made a commendable effort to  integrate and coordinate the
diverse concepts, processes, and data presented in this document. However, the document
prepared by EPA was very complex, and the Committee found that its organization and style
of  presentation made it difficult to easily identify information on specific topics or the
mechanics of addressing and folding the components into a  comparative risk assessment.  An

                                         15

-------
effective approach might be to organize the general methodology and the case study so that
the section and topic headings correspond to the schematic illustrations of the method. It
would also be helpful if subsequent versions of the document make more effective use of
tables and graphics to highlight critical aspects of the process.

       Finally, the Committee emphasizes its concern that the components of these analyses
be subjected to appropriate peer review. The complexity of this approach as it has been
applied to the microbial and disinfectant byproduct example will be a powerful incentive to
ignore  data not clearly understood by the analyst. Making certain that the technical
underpinnings of the analysis are correct and up-to-date will be the major difficulty in bringing a
completed analysis to the decision maker.

       3.2.2  Step 1: Initial Risk Assessment
             The proposed CRFM (Chapter 4) initially applies the NAS 1983 risk
             assessment paradigm to evaluate risks from D/DBP chemicals and
             microbes, separately. Is this an appropriate initial step in the methodology?
             Are there other steps that should be evaluated?

       Risks from disinfectants/disinfection byproducts and  microbial agents  should be
independently evaluated, as should different endpoints such as cancer and reproductive
toxicity, before proceeding with a comparative assessment.  The 1983 NAS risk assessment
paradigm is an obvious place to start for both types of agents.  However, there are distinct
issues  with each type of agent and the degree of understanding  of interdependent parameters
of risk and disease that must be kept segregated to make the analysis transparent.

       A microbial risk assessment paradigm that is consistent with the general 1983 NAS risk
paradigm, yet which expands on and shows the unique characteristics of microbial risk can be
found in ILSI (1996). Further, a paper by Sobsey, Dufour, et a/. (1993) presents a systematic
strategy for identifying, analyzing, quantifying and characterizing microbial risks. These papers
are relevant to the development of the current CRFM approach, and the Committee
recommends that the EPA staff also consider them in assessing  microbial risks.

       There have been  many advances in our understanding of how disease is produced by
microbial and chemical agents that were taken into account, at least in part, in the document.
However, some of the fundamental issues that underlie the use of research data to arrive at
estimates of risk are not readily apparent in the document.  This  is due, in part, to the
document's style.  The Committee is concerned that some users and reviewers of an analysis
based  on the framework may not recognize the nature of the assumptions that underlie most
risk assessments. Most analysts are unlikely to be sufficiently well grounded  in toxicology or
microbiology to understand the actual magnitude of some of the  uncertainties that can be
glossed over in such an analysis. While somewhat mundane and certainly covered in other
EPA documents, the Committee considers it to be important to make these technical issues
explicit in the CRFM document.

       The methodology should also consider the role of secondary person-to-person
transmission of waterborne infectious agents in the calculations of disease impact. For many
microbial pathogens, secondary transmission is responsible for significant morbidity.  Data on


                                         16

-------
probabilities of secondary transmission can be collected from reports of outbreak
investigations and from investigations conducted by the Centers for Disease Control (CDC).

       Host susceptibility to infection from various waterborne pathogens is a complex issue
and can vary widely depending on the pathogens under consideration.  For some pathogens,
such as Hepatitis A virus, infection results in life-long immunity.  For other pathogens, such as
Norwalk virus, the majority of the population has serum antibodies; however, these are not
protective and re-infection occurs frequently.   For pathogens such as Cryptosporidium there is
great uncertainty about what method to use to evaluate seropositivity, what constitutes a
positive serologic test, what percentage of the population has Cryptosporidium antibodies and
how the presence of antibodies affects the risk of re-infection, disease occurrence, and
disease severity and duration. The Committee recommends that where the uncertainty is
large, then the methodology consider a range of host susceptibility in the model. This
recommendation also applies to the reinfection period used in the model. When there are no
data from human challenge studies or outbreak investigations, it seems prudent to treat the
infection  period as an uncertain parameter with values ranging from perhaps 3 months to 1
year.

       The availability and quality of dose-response data for waterborne pathogens varies.
The Committee recommends that, when available, the median infectious dose be used in the
CRFM  models, and that the  uncertainty associated with the estimate be recognized. The
median infectious dose can vary significantly,  depending upon the strain of the microorganism
under consideration.

       It is important to recognize the limited availability of important data on pathogen risk.
This is  especially true in the case of pathogen occurrence. This situation is further
exacerbated by the poor sensitivity,  recovery,  and detection limits of the methods  that have
been used to develop the occurrence data that do exist.  In contrast, DBP occurrence data is
relatively large and it is certainly accessible. Thus the methodology may generate a biased
result.

       3.2.3  Step 2: Translation of Risk to Health Conditions
             The second step is to translate the predicted, separate, risks into human
             health conditions  which are measurable on the population level. Is this an
             appropriate next step in the methodology? Are there other interim steps
             that should be included?

       The translation of the predicted risks into human health conditions is a logical and
appropriate next step in the methodology.  However, at the February 1999 Committee meeting
it became apparent that the  methodology did not identify all the assumptions and  steps
necessary to translate the available  data from animal toxicology research into human health
conditions.  Examples  are given below to illustrate these problems. One involves  the use of
information derived from studies in which parameters were measured as continuous responses
in the derivation of risk estimates expressed as stochastic variables, another the use of
alternative assessment approaches, and the third the effective use of both  animal toxicology
and human epidemiology data in the identification of health conditions.
                                          17

-------
       The first problem arises because much of the most useful toxicological research
measures continuous responses associated with dosing by some chemical agent (i.e.,
increasing dose to an individual increases the magnitude of a response in that individual).  The
Agency frequently uses this type of information in developing MCLs, and NCEA did include
such data in this analysis.  However,  risk is  expressed stochastically (i.e., increases in dose
increase the number of individuals with the disease, in other words, individuals either respond
or they do not respond at a given dose). The conversion between the data types is generally
based on some arbitrary assignment of a point on the continuous scale that will be considered
the dividing line between positive and negative stochastic responses. This can potentially
confuse the derivation of the common metric which ultimately depends upon human
evaluations of the quality of life associated with the health conditions ascribed to that positive
response.

       Therefore, the factors considered in the conversion  of continuous to stochastic
responses can not be arbitrary. An example of how this has been approached for neurotoxicity
endpoints can be found in Kodell  et al. (1995). The probability that a given alteration in the
continuous variable will lead to the development of an adverse impact that is defined in
stochastic terms must be explicitly considered. Then comes the assessment of risk (i.e., the
frequency of the adverse impact in a  population  arrayed against dose or exposure). Finally,
the adverse impact must be translated into a disease condition that a lay person can  recognize
and assign some negative  value to which can then be converted into the metric that will be
used to normalize different health effects. All these intermediate steps must be explicitly
considered for the application of the common metric to be scientifically sound.

       Another way of looking at this issue is illustrated by the diagram in Figure 3.
Continuous variables in toxicological  experiments are measurements of effects on enzyme
rates or other protein function that are involved in normal biological functions.  These
parameters can often be measured within isolated cells or intact animals, but generally not in
humans.  To be most useful the parameters measured should be involved in the mechanism
by which  the chemical produces harm.  However, there are many factors that affect the
expression of such effects  as a frank disease. Most important is that the shape of the dose-
response curve describing  the probability of disease is determined by the distribution of
sensitivities of individuals in the population to the modifications in the biochemical system
measured.  In most cases,  the distribution of these susceptibilities in the population will be
unknown. As pointed out in the CRFM document, continuous variables can be converted to a
population-based figure by some  relatively simple mathematical conventions, so something like
an incidence term can be developed. However,  such arbitrary treatments can trivialize the
translation processes because they can gloss over fundamental flaws in our understanding of
the mechanisms leading to environmentally induced disease. As such they can introduce
substantial uncertainties into the risk  assessment process that need to be accounted for by the
methodology.

       As results of animal studies are integrated with epidemiological studies, it becomes
important to understand how the analysis will be calibrated  in terms of attributable risk. This
will be difficult, but it must always be  attempted.  Are the projections of risk reasonable
considering other recognized causes of a particular disease? Are the projections of the animal
data consistent with the epidemiological literature and are the epidemiological  data consistent
with the toxicological data? The handling of these problems in the CRFM was cursory and not

                                          18

-------
very defensible.
                                              Endocrine &
                                              Paracrine Factors
                                    Pharmacokinetics
                                                                      Continuous or
                                                                      stochasticresponses
                                                                            Validation of in vitro
                                                                            to in vivo extrapolation
                                                                            (PK and TD)
                                                         Continuous responses
Validation of
Toxicodynamic
Variables
                              Incidence of disease
                              (stochastic')
                                 Extrapolation
                                                                                                       Continuous responses
                Continuous responses
                       Figure 3.  Relationship between different indicators of adverse effects of chemical and physical agents that can be
                       measured. Toxicological research frequently involves measurement of biochemical or molecular parameters whose
                       quantitative relationship to the incidence or severity of disease is not explicitly understood. Measures of these 'key events'
                       indicate a genuine potential for producing disease. Such measures are frequently used to develop regulations for individual
                       chemicals. However, if they cannot be converted into a probability that a certain condition will occur in a population, it will
                       be difficult to normalize the effects of agents that produce diverse outcomes.
                                                                            19

-------
       A second point is that more attention should also be paid to how policy affects the way
in which agents are considered in the CFRM. Clearly, the risk manager may be constrained by
Agency policy. The Committee encourages the comparison of results from alternative
approaches to risk assessment.  This would make apparent to the Agency whether certain
policy prescriptions would have significant impacts. This might also serve to identify areas
where Agency research might be targeted.

       The third problem involves translations of a different sort. Animal data were used to
estimate  quantitative risks, but the results of epidemiological studies were used to develop the
common  metric (e.g. the Quality of Adjusted Life Years). Part of the problem arises from the
very different quantitative risks that are projected from epidemiology studies (Morris et al.,
1992; Poole, 1994, 1997) versus those risks based on available toxicological data on DBFs
(Bull and Kopfler,  1991). Toxicology provides the  type of information that is needed to deal
with individual  byproducts,  so it is possible to estimate risks from the concentration of individual
DBFs and proceed accordingly.  However, the toxicological data address only a few of the
DBFs that are  found in disinfected water. Thus, it is  entirely possible that both estimates are
accurate  within their areas  of investigation.  It is very possible, or even likely, that the two
approaches are measuring the effects of different  byproducts or different byproduct
combinations.

       These examples point out the temptation (or even need) to oversimplify that will always
be encountered in general cases of comparable complexity. It is imperative that some
instructions be provided to indicate the need to comprehensively develop such cases and then
to explicitly identify the simplifications that have been made and to  include this in the
uncertainty analysis. Therefore, it is critical  to explicitly address these issues when it is
necessary to extend a large body of information on health effects of a number of agents to
make predictions about a larger set of unknowns.

       The Committee recommends that more explicit guidance be provided in Section 4.4 of
the Agency document about decision criteria and paths  to be  taken in exercising this aspect of
the CRFM. The interim steps that are taken to make translations between different data types
(e.g., continuous responses underlying "key events") have to  be viewed carefully in terms of
their potential impact on the translation of potential human health effects into common
measures.  In turn, the contribution of the assumptions that these translations have on the
uncertainties that are embedded in the final analysis must be  acknowledged.

       3.2.4   Step 3: Translation of Conditions to a Common Metric
              The third step is the translation of the human health conditions into a
              common metric.  Is this an appropriate next step in the methodology? Are
             there other interim steps that should be included?

       The third step, the translation of the human health conditions into a common metric, is
a necessary and appropriate step in this cost-effectiveness analysis. The quality adjusted life
year (QALY) was the metric chosen by NCEA.  Theoretically,  this metric would allow
quantitative comparisons of different kinds of adverse health effects by quantifying subjective
responses to hypothetical questions (an example of the  calculation of the QALY is  given later
in this section).
                                          20

-------
       The QALY concept has been used by international agencies (such as the World Health
Organization and UNICEF) to weigh the benefits of various public health interventions in
developing countries.  For example, in efforts to improve children's health (i.e., reduce
childhood morbidity and mortality due to a variety of infectious diseases) such agencies must
determine which of several possible interventions (e.g., immunization programs to prevent
measles and typhoid, oral rehydration solutions to reduce diarrhea mortality, pesticide
impregnated bed nets to reduce  malaria, improved water and sanitation to reduce diarrheal
diseases,  and primary health care clinics) to pursue. Faced with this problem, such agencies
have relied on the QALY (or other similar metrics).  Because this approach is new for EPA, the
SAB strongly recommends that EPA consult with experts from other public health/development
agencies who are experienced in this type of analysis.

       The CRFM is not the first EPA effort to use a comprehensive economic framework to
analyze public decisions having a health component.  The EPA's Office of Policy referred to
earlier in this document has developed methods to  estimate the value of the beneficial effects
of environmental programs. A collection of EPA case studies on the use of benefit-cost
analysis is contained  in Morgenstern (1997).  Especially noteworthy are the cases dealing with
regulation of atmospheric lead and lead in drinking water.  The benefit-cost analysis (BCA)
method uses monetary units as the common metric for valuing all benefits, including the health
benefits.  Conceptually, a given improvement to an  individual's health status is valued by the
maximum sum of money the individual is willing to pay for the improvement- i.e., the individual
is indifferent between having the lower health status with the money and the higher health
status without the money. To be  clear, the QALY cost of a health condition is "the life years in
perfect health" that an individual  is willing to sacrifice in exchange for elimination of the health
condition under consideration. For example, a 25 year-old who is indifferent between living 50
years with her current ailments and living 47.5 years in a state of ideal health would have a
cost associated with her current condition of 2.5 QALYs (50.0 - 47.5 = 2.5) and each year in
her current condition would the be  valued at 47.5/50 = 0.95 QALY/year.

       Like willingness to pay (WTP), the QALY is a utility-based measure of health effects.
But instead of measuring an individual's indifference between various states of health and
money, the QALY measures the individual's indifference between years with various diseases
or disabilities and years of "perfect" health. It combines impacts on longevity with morbidity
outcomes using an intuitive framework - i.e., the number of years of life loss that one would
trade for a year with any specific ailment.

       Each of the two approaches (i.e., cost-effectiveness analysis and benefit-cost analysis)
has strengths and weaknesses, and a dialogue between users of each could be profitable to
both.  The greatest advantage of the dollar metric is its total comparability with costs. Money is
the only metric that allows measurement of all the outcomes of a policy intervention in the
same metric and thus allows the calculation of "net  benefits." Its biggest disadvantage is that it
seems to reduce everything to a matter of dollars and cents. Applications of benefit-cost
analysis to health issues frequently run into objections, from lay persons as well  as health
professionals who are uncomfortable about attaching dollar values to changes in health status.
       Some may find cost-effectiveness analysis more "palatable" than benefit-cost analysis
because cost-effectiveness analysis does not require that health impacts be evaluated in

                                          21

-------
monetary terms.  Public health decision makers faced with choices among alternative
intervention technologies may find it easier to communicate the logic behind their decisions to
stakeholders and the public if they rely on cost-effectiveness rather than benefit-cost analysis.

       In fact, QALYs might offer a number of other advantages over the direct monetization
of health benefits to practitioners of BCA. First, QALYs have undergone extensive study by
public health researchers, medical decision analysts and health economists, and there is a
large literature to draw on for determining the QALY scores of various health states.  In
principle, QALYs also have some transferability from one situation to another. QALY weights
in one study can in many cases be used with some plausibility in another, avoiding the
expense of an additional survey. Finally, the QALY concept offers a way of treating mortality
and morbidity consistently within a common framework. This is missing in the usual approach
to health benefit estimation, where individual disease conditions are treated separately and on
an ad hoc basis.  In particular, mortality is frequently valued by means of the "value of
statistical life," which is based on individual willingness to pay for very small changes in the risk
of dying. QALYs may be a useful device for permitting the comprehensive treatment of
mortality and chronic disease in BCA.

       Despite these advantages, there are circumstances where cost-effectiveness analysis
based on QALYs may be inadequate.  For example, where the benefits from regulatory
interventions include appreciable impacts on species other than humans or on the environment
itself, it becomes necessary to rely on "common metrics" which are more broadly based - such
as dollars.

       Furthermore, although the QALY is a comprehensive health measure,  it has some
weaknesses, and it may be weakest in the valuation of acute health effects. The approach
taken in the CRFM to estimate QALYs for relatively minor,  short-duration illnesses involves
many un-tested assumptions.  Further research is needed  to directly evaluate individual
tradeoffs between episodes of short-term illness (e.g., vomiting and diarrhea)  and  small
reductions in life span or small increases in the risk of immediate death. Quite possibly, it is an
easier conceptual task to trade off minor acute illnesses against money than against later
chronic disease or risk of death.

       QALY estimates should not become the sole basis  for weighting the strengths and
limitations of various water treatment interventions.  For example, it may be tempting for some
decision-makers to look at the bottom line of the cost-effectiveness analysis and choose the
intervention that gives the lowest cost per QALY.  Such a choice may provide the most cost-
effective health benefit but may miss other important considerations (such as time, or access
to/availability of health care, etc.) that need to be weighed  in the decision.  For example: In
1979, an important analysis by Walsh and Warren concluded that a primary health care
package was a much more cost-effective intervention to reduce childhood mortality in
developing countries than improved water and sanitation and this analysis formed the basis for
WHO policy for several years. However, Okun pointed out that this analysis only looked at
childhood mortality and ignored the benefits of improved water and sanitation for free time for
women (the main water collectors), improved economic development, improved community
organization, improved domestic agriculture, etc.

       When EPA uses this approach to weigh the value of several water treatment options,

                                         22

-------
the agency should explicitly list all possible strengths and limitations associated with the
proposed interventions that should be considered in addition to the QALY estimates.  In the
interventions considered in the case study, home filters may involve aspects of maintenance
time and effort and difficulty monitoring finished water quality that are not reflected in the
financial cost of the intervention. Water taste is another consideration that is hard to quantify
in this type of analysis.

       It is important to note that the use of QALYs does not preclude the use of benefit-cost
analysis. There is no reason in principle why a dollar value cannot be attached to a QALY. It
would represent the individual's valuation of a year of perfect health. It might even be useful to
compare the direct valuations of health states in monetary terms with their valuations as
determined in a two-step process, first determining their QALY score and then converting to
monetary units by means of an estimate of the monetary value of a QALY.

       The Committee identified a number of specific issues associated with the calculation of
QALYs that need further consideration. Some of these issues simply require more explanation
in the text; others may require  a more extensive survey of the health cost literature.

       a) The elicitation method used to derive QALY weights deserves more discussion.  In
       chapter 4 the Agency poses the time tradeoff as the way to calculate QALYs.  The
       CRFM report mentions  two alternative approaches,  the standard gamble and the use
       of subjective rating scales, but argued that the flaws in each of these precluded their
       use.

       The time tradeoff method is not accepted as the best method by all researchers. The
       time tradeoff method seems to embody some hidden assumptions about time
       preference, considering that the years of perfect health that one trades off necessarily
       occur at the end of one's expected life. When this question was raised at the DWC
       meeting on December  10,  1998, the Agency responded that the question of time
       preference is never brought up during surveys to estimate QALYs by the time-tradeoff
       method. Is it correct, i.e., are there  no debriefing exercises among survey respondents
       to determine whether they  were taking  discounting into account in valuing the reduced
       life span?

       b)  Which populations should  be used to derive QALY estimates of disease severity?
       Should  QALY weights be determined from a random sample of the population; a
       sample of those suffering from the disease  of interest; or a group of experts such as
       doctors and nurses who treat patients suffering with the disease of interest? All have
       been used  in studies designed to elicit QALY  values.

       For example, in a study by Hamilton, a source of QALY weights for the CRFM  case
       study, the respondents  are parents  of children identified as attending publicly funded
       schools. While one of the  three disease outcomes of interest in the CRFM case study
       was change in fertility, the  Hamilton study population, excluded a potentially relevant
       group of respondents - i.e., childless adults.  This excluded (or at least under-
       represented) group may value fertility impacts quite differently than other members of
       the community.
                                         23

-------
It seems that each approach would have certain advantages and disadvantages. The
SAB urges the Agency to more fully consider these and other alternatives and to
explain to potential users of the methodology the advantages and disadvantages of
each approach.

c) The concept of "perfect health" seems to suffer from a lack of definition. Is "perfect
health" different for a 20 year old than it is for a 60, 70, or 80 year old person? How is
that taken into account?

d) The report should discuss whether the studies that have been used in the past to
establish QALY weights are appropriate for the kind of environmental decision-making
envisioned by EPA for the CRFM. QALY weights are established by means of surveys
that ask respondents to make tradeoffs about their own health.  Would the answers be
the same if the respondents were asked to make decisions about public health
interventions,  in a format that more nearly matched the manner in which these
questions are raised in the political arena?  This would require questions that, for
example, asked respondents whether they would support a program that would prevent
X cases of disease A or an equally costly program that prevented Y cases of disease
B. Are the authors aware of any QALY studies that have tried to rank interventions in
this way, and  if so, compared the rankings thus generated with those generated by a
set of QALY weights?

e) The report and its authors should be prepared for questions  to be raised about the
appropriateness and possibly even the legality of using QALYs.  Several years ago the
State of Oregon proposed to use QALYs to determine the kinds of health interventions
that would be  reimbursed under Medicaid (i.e., possible treatments were ranked by a
value derived  by a measure of the improvements to QALYs they achieved divided by
the cost of the treatment).  The Oregon approach was twice rejected by the US
Department of Health and Human Services (DHHS) based on their determination that
the Oregon procedure for deriving QALY weights violated the Americans with
Disabilities Act (i.e., QALYs were not derived from persons having disabilities; QALYs
associated with disabled persons were underestimated in the Oregon approach). A
more complete description of the Oregon example can be found in US Court of Appeals
for the District of Columbia Circuit (1999). There, the Court did  not judge the
appropriateness of the DHHS denial of the Oregon approach, or the approach itself.
However, they did cite the QALY as applied  by Oregon as evidence of the existence of
approaches that might be used by EPA as a starting point for its own use in deriving a
method to determine a decision  point for use in National Ambient Air Quality Standards
decision making.

There are distinctions between the State of Oregon proposal, which affected the
availability of medical treatment to identifiable individuals, and the policy uses
contemplated  by NCEA's CRFM, which deals with public health  impacts to individuals
whose identity is not known (or perhaps knowable). Although the DWC doubts that this
kind of legal-ethical  objection would apply to decisions about investments in drinking
water quality,  it recommends that NCEA consider this issue and determine whether
there are likely to be unanticipated legal-ethical dilemmas arising from the use of quality
of life weighted longevity measures of disease impact.

                                   24

-------
f) The document should provide more information about both the variability and
uncertainty in the QALY weights.  It would be good to know, for example, (i) the
variability of the QALY weights for the same conditions across individual respondents in
the studies under consideration, (ii) the stability of the QALY weights for the same
conditions from one study to another, (iii) the sensitivity of the QALY estimates to the
set of conditions being estimated, (iv) the sensitivity to the method chosen for elicitation
of the QALY weights (time tradeoff, standard gamble, subjective  rating scales etc.), and
v) whether there are genetic traits or environmental conditions within the human
population that are potentially underestimated in the  population samples that were
studied.

A related issue is whether QALY weights for most anticipated EPA applications of the
CFRM can be taken from the existing technical literature or whether they will  need to be
developed from field studies in the communities, or the specific populations, affected by
the proposed  interventions.  Subsequent versions of the report should address this
issue, which may substantially affect the feasibility and cost of implementing the CFRM.
It would also be interesting to know whether EPA anticipates cases in which it would be
appropriate to develop common metrics through, for  example, focus group work with
stakeholders.

g) There is a need for a better method for dealing with acute nonfatal disease. The
assignment of QALY costs to health outcomes required a great many assumptions,
combining results from a number of different studies  and adapting results derived in
one situation to a quite different situation. For example, the study used the QALY cost
of a year in severe pain derived in the Hamilton, Ontario study to estimate the QALY
cost of a bout of severe acute illness (such as diarrhea or vomiting), by  dividing the
one-year cost by 365 to get a one-day figure, and then multiplying by 14 for a two-week
illness.

This procedure requires the assumption that pain produces "disutility" for a person at a
constant rate.  There is no empirical or theoretical  basis for assuming proportionality, or
for that matter, for assuming any alternative form such as marginal increase or
decrease.  In addition,  the disparity in scale between a day and a year is very great; it  is
doubtful that the value to an individual of anything of a day's duration can be inferred
from information about the value of experiencing that same thing for a full year. Some
things that can be borne for a day (with little impact)  can lead to despair if they must be
endured for much  longer periods.   Other things, perhaps, we can get used to or can
find ways to cushion ourselves from - in which case  the cost of a single day might be
much  greater than the  average  daily cost of a year.

In addition, the DWC believes that the method used to estimate QALYs for acute
nonfatal illness is somewhat ad hoc.  If such diseases account for a large share of the
health  impact - as they do for many of the scenarios of the current case study - the
QALY assignments for this class of health effects may require further attention. There
are at least two alternative approaches to this question. The first approach would
involve a search of the literature for studies that address directly  the tradeoff  between
acute  nonfatal disease and more serious, long-term diseases, for which more reliable

                                   25

-------
       QALY estimates already exist.  If no such studies exist, It would be useful to do a
       comparative risk survey that directly pits against each other the two risks examined in
       the case study (acute Gl disease and cancer after some period of latency). The survey
       results could be used to estimate directly the QALY cost of the various outcomes of
       these diseases, which could then be compared to the QALY costs estimated  in the
       current study.  A close concordance between the survey-based QALY costs and those
       estimated in the case study would help build confidence in the method used to
       calculate QALYs. If there were no close correspondence, then survey results would
       nonetheless help analysts develop better estimates of the QALY cost of acute nonfatal
       disease. A second option, would involve expression of acute nonfatal disease costs in
       monetary terms, possibly with the assistance of researchers in the Office of Policy.

       h) Accurate portrayal of the severity of microbial and chemically  induced disease
       states is a critical issues in  the CRFM. For microbial agents, our knowledge  of
       infection, morbidity and mortality provides a reasonably dependable point of departure
       for such determinations because  the information can be developed directly from human
       experience. If the effects of chemicals are expressed in the probability of contracting a
       specific disease it is possible to parameterize similar issues in the context of QALYs.
       However, the information developed from animal toxicological studies is neither as
       complete nor as  convertible to "states" of health that are as easily recognized by the
       average lay person.  The influence of this on meaningful QALY development is unclear.
       This problem may be best illustrated by the following example. Assume that
       chemically-induced changes in circulating estradiol concentrations in an experimental
       animal lead to a variety of "adverse" outcomes of different severity.  These effects
       would include a variety of symptoms easily identified in a humans that might be
       described as varying degrees of discomfort.  Clearly, these effects should impact the
       QALY at some level, but they may not be apparent from animal data. Conversely
       sustained exposures to effective levels can lead to a variety of serious health effects
       related to impaired reproductive capacity, developmental toxicities, or cancer in a
       variety of organs. In these cases the QALY may be estimated with some accuracy, but
       may or may not be more severe than sustained symptomatology. The subtlety of
       depending upon indirect evidence from experimental animals to predict QALYs that
       depend heavily on human perception is a conceptual problem that was not directly
       addressed.

       3.2.5  Step 4: Cost Evaluation
             The final step is to evaluate the financial costs of the treatment
             technologies and the medical costs and estimate a cost effectiveness ratio.
             Is this an appropriate final step in the methodology? Are there other interim
             steps that should be evaluated?

       The final step, evaluating the financial costs of the treatment technologies and
estimating a cost-effectiveness ratio, is appropriate. Estimating a cost-effectiveness ratio is
useful in evaluating financial costs of the treatment technologies relative to QALYs. It is
important to recognize that the outcome of this analysis can be biased in either direction by
including or excluding certain costs in the assessment. The advantage of the proposed
approach is that, if properly done, it will make the costs included in any assessment quite
explicit. This should be beneficial to those interested in Agency decision making.

                                         26

-------
       Several obvious costs seem to have been omitted from the Agency's case study
(apparently to simplify the presentation).  Some of these will have to be included as the
applications of the methodology are expanded.  Many of these are discussed more fully in the
portions of the charge related to the case study. However, the DWC thinks that several need
to be discussed in the method itself:

       a) A decision was made to include only QALYs among the losses occasioned by
       disease, omitting the cost of health care  and lost output. This was done to  help keep
       the size of the document manageable, however, the result of omitting these costs  is to
       cause the costs of the two technologies to seem to be the same.  Given the high cost
       of cancer treatment, however, that assumption is questionable.

       b) Although the point is subtle, the current case study does not consider the economic
       costs associated with health care and lost productivity that propagate through the
       economy from the costs noted in "a)" above. In our meeting, EPA personnel indicated
       that future analyses would take this  into account. It is also important that issues of
       access to/availability of health care be considered as part of these costs.

       c) The document does not consider the  regulatory oversight costs of the Agency itself
       that are associated with alternative policy choices. There are likely different behavioral
       implications and therefore different costs associated with alternative regulatory
       prescriptions. In some cases these  could be important additions to the cost evaluation.

                                4. THE CASE STUDY

       This section of the report contains the Committee's general comments on the case
study, including remarks on its limitations and on the appropriate handling of uncertainties
associated with the method and the study.  As noted in  Section 2 of this report, the Agency
directed a number of questions to the Committee on the case study. Although the  Committee
considers these questions to be of consequence and important to its review, it has placed
them later in the document in Appendix A.  There, detailed comments are directed  at specific
case study charge questions. The Committee has chosen this approach  in order to give
emphasis to its comments on the method. Just as the case study was conducted to illustrate
the CRFM method, the Committee's  review of the case study was conducted to determine how
the procedures contained in the CRFM are  implemented.  In many instances, the results from
the Committee's consideration of the study  are generalized in its comments on the
methodology itself which are contained in Section 3 of this report.  Many of the comments in
Section 4, and in Appendix A focus on shortcomings of the case study, therefore, they may
appear to carry a more negative connotation than intended. Addressing the issues contained
in the Committee's critique will be critical to further developing the methodology and to
determining how generally the approach can be applied to the microbial and disinfectant
byproduct area.

4.1  General Issues for the Case Study

       4.1.1 Is the Case Study an Appropriate Application of the Framework
       Methodology?
Is this case study an appropriate application of this methodology? Are these initial

                                         27

-------
assumptions plausible for a hypothetical site? Are there other factors that should be
considered within the scope of the case study?

       The case study provided insights into how the Comparative Risk Framework
Methodology (CRFM) would be applied, thus helping to demonstrate the potential of the
methodology in formulating, structuring, and describing complex environmental problems for
evaluation. The structure provided is more valuable than the calculated estimates generated
because that framework makes explicit the factors that have been considered in assessing the
risks and benefits associated with alternative drinking water treatment approaches. Thus, the
study suggests that the CRFM could provide a good structural  mechanism for improving
discussions of complex environmental problems within the community of persons responsible
for and interested in regulatory decision making.

       As indicated in the charge, the case study was  built around a set of circumstances that
is likely to represent a particular local water system rather than a general set of circumstances.
Consequently, the application to a very specific circumstance was more easily visualized than
applications at the national level. However, it is unclear whether the intent was simply to
model a single site-if so the assumptions generally seemed appropriate; to model sites more
broadly-if so, this case may not be typical of water supplies in  the nation as a whole; or to
model a community where the issues entertained would be at the forefront of a regulatory
decision-if so, the bromide contaminant may not be sufficiently represented in the scenarios
presented to the Committee. The point is that a major  variable in this type of analysis will be
variability in the source water.  The case study would have benefitted from some
consideration of variations in source, because the effectiveness of competing technological
fixes can be considerably different with variations in the water  quality.

       The case study ignores a number of important behavioral aspects associated with risk
that should be part of the methodology.  Several examples come to mind. The possibility of
home interventions (e.g., the provision of point-of-use  filters) raises behavioral issues that
don't come up when dealing with the more customary centralized water treatment
technologies.  Some issues were mentioned at the meeting, including  the possibility that the
use of bottled water among AIDS sufferers is already very high possibly lessening the
incremental gains estimated for home filtration; the incentive to return  to tap water for those
now using bottled water if filters are to be provided gratis; and  the possible strong local
pressure to expand use of home filtration for other sensitive populations  (having sensitivities
across a broad range) if the filters are regarded as a success.

       EPA representatives at the February, 1999 DWC meeting indicated that the next step
would be to develop an applications document that will include a collection of case studies.
The Committee endorses the conduct of additional cases studies and  suggests that certain
studies might be especially useful. For example, an evaluation of the  cost-effectiveness  of
conventional water treatment with filtration versus  no filtration,  or conventional water treatment
with filtration vs. ozonation with no filtration, for large municipalities would be very useful.  As
one expands the scale from the local (as in this case study) to  the national scale, the cost-
effectiveness of adding filtration to those systems that  now only disinfect would also be
important to add to the evaluation.

       4.1.2 Statistical Approaches and Uncertainty

                                          28

-------
       Are the statistical methods and assumptions used to develop parameter
       distributions (Chapter 5) reasonable and correctly applied? Are the simulation
       procedures and sensitivity analysis appropriate for these data (Chapter 6)?  Are
       there other statistical techniques that could be applied to improve the CRFM?

       An important aspect of the CRFM is the approach used to characterize uncertainty.
Uncertainty analysis can focus on "parameter uncertainty" as the case study does, or it could
be expanded to deal with "model uncertainty."  While the SAB understands the tendency to
focus on parameter uncertainty, because it is relatively simple to handle in an "objective" and
defensible manner, it is possible that failure to deal with  model uncertainty could prove to be a
critical flaw. In many cases, model uncertainty will dominate parameter uncertainty and efforts
which ignore it could lead to a false sense of confidence in the results of the cost-effectiveness
analysis, and serious distortions of research priorities. For these reasons, the SAB
recommends that model uncertainties  be clearly identified and addressed. As a minimum, the
robustness of conclusions to different plausible "models" could be explored in sensitivity
analyses.  Also, formal elicitation of expert judgment could be used to begin to understand the
extent of legitimate scientific debate and disagreement on key issues and to the impact of this
on the cost-effectiveness of alternative control strategies.

       In general, the members of the Committee felt that the assessment of parameter
uncertainties for the comparative risk framework methodology appeared to be, for the  most
part, on the right track.  However, the DWC identified some aspects of uncertainty and
sensitivity analyses that were not identified or sufficiently discussed in the case study.  The
aspects that should be considered in the further development of this particular application are
provided below:

       a) Uncertainty Regarding the Methodology

       i) The case study should further explain the process used to propagate the uncertainty
       and variability of model parameters to approximate the quantitative uncertainty of the
       model output [the cost-effectiveness (CE) ratio in the case study].  The draft report
       does not discuss the methodology in sufficient detail to permit the reader to fully
       understand what was done and why it was done.

       ii) The Committee recommends that the report be revised to include at least a
       qualitative discussion of the possible magnitude  of model uncertainty, and if resources
       permit, a quantitative assessment of model uncertainty should be described and
       implemented in the  case study.

       iii)  When conducting uncertainty analyses it is important to decide up front whether the
       assessment endpoint (the CE ratio in the case study) is considered to be some
       measure of the average for a specified population or a distribution over a population.  If
       a distribution is being estimated, then both variability (over the population) and
       uncertainty of parameters are present.  In that case the uncertainty analysis must be
       conducted in ways such that the uncertainty and  variability components are separated
       and properly propagated.  It appears that this separation of uncertainty and variability
       was not done  in the case study. The DWC believes this problem should be addressed
       in future revisions of the case study and in other  case studies that may be conducted.

                                         29

-------
The methods used for this purpose should be fully explained and consistent with the
full description of the propagation methodology laid out in the general methodology
section of the report.

iv) The credibility of uncertainty/sensitivity analysis results depends in large part on the
credibility of the distributions used (selected) for the parameters.  The rationale for
some of the selected distributions was not provided in the case study. This omission
should be corrected. Also, when it is difficult to see which distribution is most
appropriate, the sensitivity of the predictive model results should be obtained by using
alternative input distributions.

v) The Cryptosporidium data used in the case study provided an interesting example,
but the analysis of the data was complicated by the fact that the data sets were
censored, i.e., many samples were reported as being less than the detection limit (see
Table 5-15  on page  5-35 in the EPA document). It appears from this table that the
mean of the total oocysts present was computed using only the samples for which the
counts were above the detection limit. This approach will tend to give a computed
mean that is too large since the non-detects, presumably, have lower counts than the
detects.  Future revisions of the case study should consider using statistical methods
that have been developed for estimating the mean of censored data sets.  A paper that
discusses some of these methods is: Helsel (1990).

vi) The uncertainty analysis in the case study did not consider the possibility of
correlations among model parameters. The DWC recommends that any future
revisions of the case study should incorporate methods for propagating  the uncertainty
of both uncorrelated and correlated parameters.

vii) The Committee believes that the case study should describe why the propagated
uncertainty  information about the model output (CE ratio) is useful and how it should  be
used by decision makers.

viii) Apparently, the  case study used simple random sampling to  generate multiple
values for the uncertainty and variable parameters in the model.  However, other
methods of generating multiple values, such as Latin Hypercube sampling, can also be
used. Latin Hypercube sampling is known to generate data sets that are more
representative of the underlying distribution being sampled than can be  achieved by
simple random sampling with the same number of realizations. The authors of the case
study should consider if Latin Hypercube sampling should be used instead of simple
random sampling. The rationale for the selected method should be provided.

b) Uncertainties in the Public Health Outcomes of Alternative Treatments

i) The CRFM attempts to compare the different public health outcomes  of particular
treatments.  In addition to uncertainties inherent with predicting the frequency of
adverse health outcomes (e.g. dose-response relationships,  extrapolations to low
exposures,  and secondary spread of infection), the present case  study is complicated
by a  large number of factors that affect exposure. Central to this are uncertainties in
the outcomes of treatment as they are compounded by uncertainties in the

                                   30

-------
       characterization of the source water, treatment efficacy, robustness of the treatment
       train to process upset, and management of the treatment processes.  The methods
       used to propagate these types of uncertainties into the final analysis should be clearly
       described in the case study.

       ii) Subsequent case studies should consider the role of secondary transmission of
       waterborne infectious agents because for most microbial pathogens it is responsible for
       significant morbidity.  Data can be obtained from outbreak investigations and CDC
       investigators.

       iii) The uncertainty associated with extrapolating high-dose animal toxicity data to low-
       dose chronic exposures in humans is really not dealt with in the document. Doing so
       might be expecting too much of the current case study.  The case study combines
       "loose" indicators of disease outcome that have uncertain implications for the
       probability of disease with individual byproducts having some available test results
       (problem is generally associated with effects on reproductive and developmental
       toxicities).  A tendency to combine associations into assumptions that are applied to a
       group of chemicals injects a series of conservative assumptions one on top of another.
       That could tend to magnify the chemical risks to a point that the outcomes are
       unreasonable.  In the present case study extrapolation was not a problem because the
       microbial risks were so overwhelming as to make the chemical  risks relatively
       unimportant.  However, circumstances can be envisioned where inappropriate
       extrapolation could lead to an inappropriate public health decision.

       4.1.3 Case Study Limitations
       Are the limitations of the case study appropriate and clearly  presented? Are there
       other significant limitations not identified?

       Many limitations in the case study were clearly presented in the Agency's document.
The case study did raise a lot of issues about the general approach that need future
consideration.  In addition, the limitations of some of the parameters selected were not noted
in the document.

       An important limitation is the use of "active" Cryptosporidium oocyst concentrations in
water.  This use may lead to an underestimation of risk. If drinking water is being evaluated,
then the presence of any oocysts in the treated waters could mean that oocysts have in fact
broken through the treatment system's barriers. It may be more realistic to use the total
number of oocysts when evaluating such risk.

       In some respects the case study would have been more interesting if it had focused on
contaminant levels where real risk management decisions might have  to be made. For
example, within the case study a comparison was made that indicated that central treatment
might have been as cost-effective for  protecting the  vulnerable population as employment of
point of use devices.  While the analysis found a difference in cost-effectiveness that favored
point-of-use devices,  this difference was well within the boundaries of  uncertainty.  There were
a variety  of very unrealistic assumptions about the reliability of point-of-use devices and
behaviors that affect their reliability. If these other factors had been taken into account, there
might be  a strong reason for favoring central treatment. Since central  treatment would also

                                          31

-------
provide ancillary benefits in improved drinking water quality, the risk manager might find this an
easier scenario to advance.

      As the methodology evolves, attention should be directed to the question of scale. This
local case study focusing on a relatively simple situation required many assumptions.  These
assumptions had little effect on the outcome of the analysis, because of the large differences
in risks that could be attributable to Cryptosporidium and disinfectant byproducts under the
scenarios defined. Some concern must be expressed that some of these assumptions may
dampen the significance of important variables as the analyses progress to more diverse
scenarios, focus on multiple  locations, or attempt to deal with problems aggregated at the
national level. It seems reasonably safe to conclude that the CRFM would be useful at the
local level if the expertise necessary to conduct the analysis and correctly interpret the results
is broadly available (especially in light of the assumptions  made).  At a regional level, there
may be sufficient commonality in major variables (primarily source water issues) that a
relatively small number of variations in the analysis could still make the approach practical.
However, at the national level the application becomes much more abstract and, consequently,
it is difficult  to determine whether the approach would be effective.

                                    5. RESULTS

5.1  Are the results presented appropriate (Chapter 6) and the interpretations of the
results (Chapter 7) appropriate? Are there additional discussion items that should be
presented in either Chapters 6 or 7.

      The  results section was appropriately structured. The additional data that go into the
final analysis is presented in a way that is more trackable than presented in Chapter 5.

      The  DWC has made  numerous comments about some of the assumptions that underlie
these final calculations throughout this report. Some of these assumptions can have major
impacts on  the results of the case study.

      Chapter 7 provided a generally good overview of what the  case study might say to a
local decision maker or a risk manager responsible for making decisions on MCLs. The DWC
generally agrees with the potential utility of the CRFM at the local  level. The DWC found the
discussion on how this might be used by the national risk manager somewhat naive. All the
case study  really addressed  was which of two interventions might be most cost-effective in
preventing harm from waterborne Cryptosporidium to the general population and to a sensitive
population.  This analysis is very limited in that it depends heavily  on local conditions (i.e. are
there viable Cryptosporidium oocyts in the source water).  DBP risks were unimportant in the
analysis of the case because they were overwhelmed by the microbial risk. As  a
consequence, it is difficult to determine how this case study should influence the development
of MCLs for DBFs or for microorganism.  The first consideration that must be addressed in
developing  an MCL for an individual agent is whether the agent presents an unacceptable risk.
Such determinations are more or less independent of other risks in the water. A second level
consideration is how one might implement the achievement of that goal. As indicated in the
introduction, the Office of Groundwater and Drinking Water must implement their program in
the context  of the Safe Drinking Water Act and the MCLGs and MCLs are the mechanism
provided.

                                         32

-------
       The statement made on page 7-14 implies that microbial and DBP threats to health
from drinking water are competing risks. A more accurate statement is that microbial risks are
large if water is from a vulnerable source. The first obligation of a national risk manager is that
he/she cannot promulgate regulations that would compromise the delivery of water that is free
of waterborne infectious agents. That protection must be stacked up against accurate
assessments of probable risk from agents that are not as effectively dealt with by conventional
drinking water treatment or source protection.

       The next questions is:  "Given that there  is a suite of treatment processes that can
eliminate or at least minimize waterborne infectious disease, which of these  methods produces
the lowest risk from DBFs at equivalent levels of utility?" At this stage it is essential to
understand the local source water conditions in even more detail.  However, it is improbable
that a local water utility will be in a position to evaluate these risks for all possible DBPs.
Therefore, there needs to be a set of standards  that state  unequivocally the  levels above
which no individual DBP should be allowed to occur.  The real question that arises in this case
is the burden of proof that such occurrence is actually harmful at the projected levels.

       The utility of the CRFM at the national level may be in its ability to order information.
One application is to identify research needs as  suggested in section 7.4 of  NCEA's document.
An equally important application would be to help the national risk manager  view the types of
behaviors that an MCL might trigger on the part  of the regulated community. For example, will
the need to meet this MCL force a utility in the direction of an untested or poorly evaluated
treatment alternative. The practical outcome of  the first regulation of trihalomethanes in  the
U.S. was to increase the use of chloramines for  disinfection. In retrospect, this  probably
caused little harm, but it caused the Agency to recognize that regulations could influence
existing disinfection.  That recognition was a  major stimulus for research and is the technical
basis behind much of the research that has been identified on the microbial  and disinfectant
byproduct problem in the last decade. Consequently, the  DWC encourages NCEA to press
forward with development of the CRFM as a valuable tool  for displaying all the relevant factors
that need to be considered in the development of regulations.  Perhaps the most important
refinement that could be made with respect to the microbial and disinfectant byproduct world
would be to incorporate consideration of how socioeconomic factors will influence the
response of the regulated community to a new regulation.

       A specific problem in this section is the calculation  of cost effectiveness  ratios for ozone
pretreatment for the AIDS subpopulation. The calculation  of "average cost-effectiveness" of
ozone pretreatment in the AIDS subpopulation (p. 6-41) is given as $2.27  per QALY. This
calculation is both irrelevant and misleading and should be removed. Cost-effectiveness is an
incremental concept;  it compares the effects and cost with and without an intervention.
Sometimes the increments are very large indeed - an entire water treatment plant,  for
example.  The comparisons here are not appropriate because there is no way to supply
pretreatment to the AIDS subpopulation at $23.37 per persons without also supplying it to
everyone else  in the service area at the same unit cost.
                                6.  RESEARCH NEEDS
                                          33

-------
6.1  Are the research needs clearly defined? Are the research needs that are highlighted
appropriate given the information in the Case Study? Are there other research needs that
should be identified?

      The sweeping nature of some of the assumptions made in the case study highlight the
need to  conduct research that can provide both qualitative and quantitative information for the
assessments that are essential components of the CRFM.  It is clear that the analysis of both
the microbial and disinfectant byproduct risks were controlled by the assumptions made rather
than by  any quantitative data.  As a consequence a valuable outgrowth of applying the CRFM
to this problem is the critical research needs that it helped to identify.  It also helps to make
clear the great responsibility that accrues to analysts to ensure that the host of simplifying
assumptions does not obscure the uncertainties in the analyses. With this in mind, a number
of research needs are apparent.  One category of need is specific to the continued
development of the CRFM and another focuses on research specific to the problem being
evaluated (for this case, research on microbial and disinfection byproducts).

      It is essential that these research needs are seen as distinct from the basic science
research in environmental microbiology, toxicology and epidemiology of waterborne diseases
from disinfection byproducts and infectious agents. While it is certainly true that advances in
the basic sciences may reduce uncertainty in estimation of health risks and in the identification
of cost-effective treatment technologies, without a solid basis for comparative risk assessment
much of this basic research will not achieve its true potential for affecting policy determinations
and improving  public health.

6.2 Research  Needs to Support Continued Development of Comparative Risk Framework
Methodology

      a) Much of the background for the application of QALYs depends upon issues
      involving trade-offs in medical practice or preventive medicine that potentially affect the
      same individual. Research should be performed that confirms valuations that the
      general public would assign to QALYs related to infectious disease morbidity and
      mortality relevant to environmental risk management.

      b) There is a need for  a better method to handle acute nonfatal disease.
      The assignment of QALY costs to health outcomes in Chapters 5 and 6 required a
      great many assumptions, combining results from a number of different studies and
      adapting results derived in one situation to a quite different situation.  For example, the
      study used the QALY cost of a year in severe pain derived in the Hamilton, Ontario
      study to estimate the QALY cost of a bout of severe acute illness (such as diarrhea or
      vomiting), by dividing the  one-year cost by 365 to get a one-day figure, and then
      multiplying by 14 for a two-week illness.  This procedure requires the assumption that
      pain produces "disutility" for a person at a constant rate.  There is no empirical or
      theoretical basis for assuming proportionality, or for that matter, for assuming any
      alternative form such as marginal increase or decrease. This assumption is a matter
      for empirical investigation.

      c) It appeared  to some members of the DWC that the values assigned to disease
      morbidity and mortality might depend on the individuals place in life and the extent to

                                         34

-------
       which they have dealt with serious medical problems in the past. Some attempt should
       be made to validate the QALYs that are assigned for microbial and chemically-induced
       disease outcomes from several disparate points of view that could be represented in
       the U.S. population and for the population represented by the case study.

       d) Methods for addressing model uncertainty in policy analysis.

6.3 Research Specific to the Problem Evaluated

       a)  Research is needed to better characterize risks that might arise from low level
       exposure to certain organisms and disinfectant byproducts. These research activities
       should first focus on those contaminants most likely to induce effects seen in
       epidemiological studies or that are particularly important in the decision process (e.g.
       bromate, viability of Cryptosporidium oocysts).  This research should include efforts to
       clarify the relationships between disease outcomes in animals and humans.

       b)  The hazards related to Cryptosporidium exposure are well identified, and the
       authors make it a point to emphasize the  idea that more data are needed within the
       context of the present case study. To deal with the total problem, data are needed not
       only for Cryptosporidium but for other microorganisms as well.

       c)  The possible survival and colonization of the distribution system by bacterial
       pathogens need to be studied if pathogens other than Cryptosporidium (which neither
       replicates in nor colonizes the distribution system) are to be included in future case
       studies.  That possibility is one point in the model that was not taken into consideration,
       since only  Cryptosporidium was being evaluated and Cryptosporidium will not grow or
       colonize.

       d)  The estimated cancer risks associated with disinfectant byproducts are not small
       compared  to risks that have precipitated other regulatory actions that have been taken
       by the Agency. Consequently, as long as one can be certain that an option does not
       sacrifice microbial risks, the actual tradeoffs with different disinfectant strategies may
       continue to depend on comparative  risks associated with disinfectant byproducts
       produced by each method into the foreseeable future. There must be  a way to more
       efficiently guide research in this area in the future and the CRFM can help in this
       regard. The DWC suggests that this application of the CRFM should be  pursued
       aggressively in the near term.

       e)  The research needs identified in Appendix A.4.9 may appear to be the general
       types of needs that are always expressed and are difficult to challenge.  However, the
       needs are  not specific and draw on a limited  segment of the  literature that is available
       on the toxicological effects of disinfectant byproducts.

             i) The needs articulated continue  to pursue questions that attempt to aggregate
             risks from disinfectant byproducts  by class. In the case of the trihalomethanes,
             this issue has been answered in sufficient detail to know that this aggregation is
             not an appropriate approach. Toxics  mechanisms of brominated
             trihalomethanes are distinct from those of chloroform. These disparate

                                          35

-------
             mechanisms suggest different approaches for risk assessment based under the
             auspices of the proposed new guidelines.  Similar dichotomization of
             mechanisms are surfacing with the haloacetic acids.  It is time to conduct
             research on the properties which are conferred to disinfectant byproducts by
             bromine substitution.

             ii)  The research needs should reflect health concerns for potential sensitive
             populations in much more specific terms, both for microbes and DBPs. For
             example, several potential problems can be identified with  the dihaloacetates.
             Given to metabolically competent animals, they are very rapidly metabolized
             and therefore relatively weak toxicants when dose is  expressed as external
             dose.  However, they are very potent systemically.   If a significant fraction of
             the population does not express the glutathione-S-transferase zeta that is
             responsible for a very large fraction of dichloroacetate metabolism, segments of
             the population could be very sensitive to these compounds. Moreover, it is
             probable that these concerns extend to other halogenated  organic acids
             produced in the chlorination of drinking water.

       f) The Agency has been conducting research on the effects of exposure to mixtures of
       toxic chemicals for the last several years.  However, it is also important to conduct
       research on the effects of exposure to mixtures of microorganisms. In contaminated
       water, it is likely that more than one type of pathogen will be present.  It is not known
       whether simultaneous exposure to different pathogens would increase susceptibility to
       infection or whether the effects would simply be the sum  of the risks for the individual
       organisms.

                                    7. GENERAL

7.1 Do you have  other specific recommendations for the improvement  of this case
study?

       The DWC recommendations  for improving the case study are found within the
responses to questions  in Appendix  A.  It is not necessary to recount those suggestions  here;
however, a few general  recommendations are emphasized below:

       a)  NCEA  should develop more succinct guidelines for translating  adverse health
       effects from animal experiments to human conditions. This guidance should also
       identify how mechanism or mode of action is to be used in these translations.  These
       issues will be daunting if they must be revisited with each case study.

       b)  The criteria used to determine if there is an incremental exposure due to a
       disinfectant treatment need to be much better established.  The DWC  is skeptical  of the
       exposure  information on several grounds.  What was the basis for determining
       something was a disinfectant byproduct? How well has the distribution of DBP
       concentrations over time and space within a water system been characterized? How is
       that information to be aggregated as it is scaled up to the national  level? Clearly,
       NCEA must take an active role in advising the Office of Water on how the data in  the
       Information Collection Rule might be better formatted for  purposes of risk assessment.

                                         36

-------
The historical data utilized in this case study are of limited quality and utility.

c) The CRFM must begin to consider source water characteristics as a major variable
in these analyses if the methodology is to be expanded beyond consideration of a
single water supply. Source water characteristics are important even where a single
water source is being considered, because that has major impacts on the efficacy of
various treatment technologies.

d) Microbial  risks need to be considered in significantly greater depth. Cryptosporidium
is not a good prototype for other waterborne pathogens for many reasons, including its
resistance to disinfection, its behavior in the distribution system and minimal risk for
secondary spread.  It also is important to recognize that different classes of microbes
have differing sensitivities to different interventions. There were too many
assumptions made in the current case studies that utilized behaviors of infectious
agents from  completely different classes.

e) Individual susceptibility must start to play a bigger role in evaluations of DBFs as
well as with  microorganisms. Potential and real genetic susceptibilities must be
considered as well as loosely grouped and  not altogether generalized susceptibilities
based on age, for example. The Committee recognized that the lack of data has made
it impossible to consider these variables in any depth.  However, NIEHS and NCI are
undertaking  efforts to examine the issues of susceptibility with much greater specificity.
EPA needs to be ready to make use of that information as it becomes available.
                                   37

-------
                                    APPENDIX A

                        Specific Comments on the Case Study

       The Drinking Water Committee reviewed and has prepared comments on the case
study both in response to charge questions from the Agency and because of the importance of
its focal topic. Many of the comments go beyond the charge as it related to the limited case
study presented. These added comments made by the Committee identify problems and
issues that should be addressed when the application of the methodology has been expanded
to consider the issues of microbial and chemical risk more broadly.  The Committee felt it
would be most useful to keep the expanded comments that related to particular questions
within the section.

1. Engineering/Water Treatment

       a.  Treatment Scenarios and Cost
       Are the choices of water treatment scenarios for evaluation relative to the "real
       world" scenarios appropriate? Do the construction and operation cost estimates
       adequately reflect real-world facilities and point-of-use device costs?

       It  is likely that the treatment scenarios in the case study are too limited to adequately
reflect "real world" situations. EPA's analytical focus on  the median performance of treatment
processes, and  on comparing the  most probable risk of chronic illness from exposure to
chemicals and endemic levels of disease may be inappropriate. The approach being
evaluated may be simply "doing the wrong thing" because it omits consideration of the
occurrence of outbreaks.  Historically, the technical literature has shown that  most waterborne
disease outbreaks have been caused by process upset/failure. Endemic disease is a
reasonable hypothesis that may be difficult to prove. The distinction is important because
reducing  endemic disease leads one to promote treatment practices that improve removal
whereas  reducing outbreaks leads one to promote treatment practices that improve reliability.
While these two objectives are not mutually exclusive, pursuing one does not  necessarily
accomplish the other. Therefore, the case study, and the general method for  evaluating
treatment alternatives should include the probability of system failure (e.g., the probability of
failure of  chlorine versus ozone; the failure probability of central treatment versus home
treatment). The treatment failure issue can be illustrated by a number of situations. For
example, ozone is seemingly a more complicated system requiring air or oxygen clean up prior
to going into the ozonator. It would likely have a higher probability of failure than a gaseous
and/or liquid chlorine feeding system.

       Also the effect of colonization of reverse osmosis (RO) units by heterotrophic bacteria
and the risk of opportunistic pathogens for immunocompromised individuals should be
considered.  Payment (1991) reported that families with RO units with high levels of bacterial
growth were at increased risk of gastrointestinal symptoms.

       Traditionally, the most commonly accepted approach to achieving reliability is by using
multiple treatment barriers. For example, deep, protected groundwater supplies, which have
little risk of contamination, are often treated with chlorination alone, conversely, for several
decades  it has been accepted practice to use both filtration and disinfection for any water

                                         A-1

-------
supply exposed to significant sewage contamination. If greater removal of bacteria and
viruses were the only goal, the application of greater amounts of chlorine might suffice, but the
inclusion of two independent treatment barriers reduces risk in two ways: first it greatly
increases the probability that at least one barrier will be operating at all times, and second it
does the job of reducing the pathogens both by removal and inactivation, making the treatment
plant's performance more robust.

      The Committee was asked to comment on the cost estimates included in the case
study. Some informal comments were offered at the two meetings during which the DWC
discussed the CRFM  with NCEA staff. However, the DWC does not feel that it is the
appropriate group to provide further comments on Agency assumptions in this area.
Determining the adequacy of these analyses would require detailed engineering and cost
estimation that the Committee judged to be beyond the scope of a DWC review.

      b. Treatment System Efficacy
      The Agency asked the Committee to evaluate the following:
             i. The  accuracy of water treatment system efficacy estimates, including the
             assumed distribution, and
             ii. The appropriateness and reasonableness of water treatment system
             assumptions concerning DBF concentrations in finished drinking water.

      This aspect of the case study is difficult to answer directly. The Committee primarily
used the case study to identify shortcomings of the general methodology. While it is easy to
identify shortcomings in this case study,  it is not as clear how these limitations might reflect on
a broader application  of the methodology. The following comments simply identify some
limitations that the Committee thinks would be problematic if they were carried forward in a real
world analysis.  In general they reflect elements that were left out of the case study, and which
should not be omitted if a real world scenario were evaluated using the methodology.

      i) The case study is limited to, "...the comparison of alternative water treatment
      technologies and not to comparing changes in technological applications (e.g.  changes
      in the  levels of chlorination)." This limitation is important because the degree to which a
      particular technology is applied can have profound effects on both its ability to reduce
      pathogens and its ability to produce chemical byproducts.  Besides chlorine, there are
      other examples. At low doses ozone does not remove Cryptosporidium, but at high
      doses it is more likely to produce unacceptable levels of bromate.

      ii) No attempt was made to consider the effect of the distribution system on either
      microbial contaminants or disinfectant byproducts.  Indeed some of the most important
      decisions in protecting public health in the next few years will have to focus on the
      distribution system as a variable. Do we use chloramines to maintain better residuals
      and lower coliform levels or do we use free  chlorine to provide better protection to the
      intrusion of pathogens at the risk of higher levels of known chemical byproducts? In
      the future it is  conceivable that primary pathogen reduction might be accomplished by
      either membrane filtration or granular media filtration and UV irradiation.  If the water
      produced is still high in  DBP  formation  potential, choices in the distribution system may
      have huge impacts on chemical risks from DBP levels in the distribution system.
                                         A-2

-------
iii) The case study assumed that no malfunctions occur that cause performance to
deviate from system specifications - such malfunctions are one of the major risk-factors
where outbreaks are concerned. Moreover this assumption is connected to the
decision to use Cryptosporidium as the target organism. Existing systems, if they work
perfectly, are expected to control the other pathogens and yet the discussion of
outbreaks makes it clear that Cryptosporidium is not the only organism implicated.

iv) There are concerns that relate to the actual application of the  CRFM. It could be a
tool that is utilized for evaluating individual water supplies. On the other hand,  it may
be intended for broader national analyses of the impacts of various regulatory
strategies.  If the latter case is intended, the large and variable impact that source
waters have is not taken into account in the case study.  The Total Organic Carbon
(TOC)  in the water treated in the case  study pilot plant is low, 1.76 mg/L during the
ozonation study and 1.98 mg/L during  the chlorination study. As  a result, the DBFs
formed in the pilot plant and used as inputs in the case study are  low compared to
many plants in the U.S.A. Bromide is also low, so that bromate production by ozone is
not particularly high. As a result, the QALYs associated with DBFs are low in the study.
Because the TOC is low, the applied ozone dose is low [the water has a low "ozone
demand"]. This low demand results in lower costs for the ozonation process. The low
TOC and low ozone dose also should result in a lower than typical production of
biodegradable organic matter in the plant, simplifying the design and operation of the
biofilter plant following the "pre-ozone" application. These may be "key parameters" for
which a sensitivity analyses might provide some insight into implications they may have
on a larger scale.

v) Baseline Technology - The conventional filtration plant is given credit for two logs of
removal of Cryptosporidium, in accordance with the newly revised Surface Water
Treatment Rule  (SWTR) but less than  suggested by the results observed in the pilot
plant used for the case study (Table A.1-2, p. A-16).  In contrast,  ozonation is given
credit for 0.5 to 1.5 logs of removal of Cryptosporidium, even though the pilot plant data
only  support 0.5 logs of removal [note  also that the pilot plant operated at about 27
degrees C, a temperature that should favor inactivation].  These data  indicate that the
effectiveness of Cryptosporidium removal by the conventional plant was probably
underestimated  while the effectiveness of Cryptosporidium inactivation by ozonation
was  probably overestimated. This assumption could overemphasize the value of
ozonation and lead to overestimates of the exposure of the general public and the
AIDS population to Cryptosporidium when ozone is not used.  Presumably, the QALYs
would also change. These baseline assumptions may be "key parameters" and should
have been subjected to a sensitivity analysis.

vi) Point of entry/use treatment devices are assumed to achieve  100% removal of
Cryptosporidium all of the time.  This assumption is not reasonable. Units will leak
organisms depending on the quality of their materials, the assembly design and the
quality of assembly and testing. Very small leaks around seals and fittings can result in
significant contamination. More importantly, these units are not operated with
supervision and indications are that they will often remain in place for some time after
they fail.  Thus, the failure rate is likely to be much higher than for central treatment.
                                  A-3

-------
      vii) The case study pointed out that the home filtration technology is cost-effective
      even after implementation of the pre-ozonation technology. It is not clear why home
      filtration is considered to be the "marginal" technology in this analysis.  The usual rule is
      to apply interventions in the order of their cost-effectiveness, which in this case would
      require the implementation of home filtration first followed by pre-ozonation if it could
      pass the cost-effectiveness test. That is, based on the costs and benefits as given, the
      two technologies are implemented in the wrong order.

2.  Risk Characterization

      a.  Cancer Risks
      Overall, are the cancer risk estimates based on exposures to mixtures of DBFs
      plausible and reasonable given the goals and limitations of the case study? Are
      the uncertainties presented in the assessments of DBF mixtures adequately
      described and characterized? Are there significant uncertainties that are not
      identified? Are the presentations and discussions of variability appropriate?

      The report described risk characterization as the evaluation and integration of major
scientific evidence and "bottom-line" results from hazard identification, dose-response
assessment and exposure assessment. Toxicological data were utilized in a more or less
conventional manner for those chemicals for which data actually existed.  However, by far the
greater number of compounds considered were evaluated were chemicals for which no
toxicological data existed for important endpoints. These  predictions for chemicals for which
no data exist were based on predictions by a computer model known as TOPKAT.  While
TOPKAT is an established program, one can gain very little mechanistic insight as to why
some chemicals were considered to be positive and others were considered negative for
carcinogenic potential.  Recent evaluations of TOPKAT and similar programs have shown that
expert judgment appears to be more dependable than their modeled predictions once the
chemicals are actually tested (Ashby and Tennant, 1994).

      The document did not identify the evidence supporting inclusion of some case study
chemicals as DBPs,  except for noting that the Office of Water identified them as such.  Some
of these chemicals are not familiar to those on the panel who are, themselves, familiar with
DBPs. Moreover,  it is virtually  certain that meaningful exposure data do not exist for a large
majority of these compounds.  As a consequence a significant portion of the toxicological risks
projected are based upon chemicals with an uncertain status as DBPs, virtually no specific
toxicological information, and little available exposure data. The document is unclear on how
much these compounds influenced the overall toxicological risk estimates.

      Further, the predicted carcinogenic risk from these analyses were summed using a
response addition  model and all were assumed to increase the incidence of bladder, colon
and colorectal cancer in the human population. These were the target organs identified in
epidemiological investigations; however, they were not the endpoints seen in the animal
toxicology studies. The epidemiological studies were discounted as providing sufficient
information to project risks; however, it was apparently used to distinguish cancer illness from
cancer death estimates.

      Essentially the same approach was used for estimating the aggregate risks for non-

                                         A-4

-------
cancer endpoints. Several aspects of this approach vary from traditional EPA evaluation
approaches. The body-weight to surface area correction was used in cancer risk assessment
but not the non-cancer assessment and the response addition model was used to sum the
activities of chemicals for which thresholds could not be excluded.  Since these are changes
from the usual practice, some justification for these departures is appropriate.

       The case study needs a consolidated section on risk characterization and that section
should discuss the limitations and uncertainties.  In general, mention of the uncertainties in the
method were scattered throughout the text of the document and some of the more subtle
departures from usual procedures described above, were not discussed at all. There was a
very general discussion of uncertainty on page xxvii of the executive summary. However, it
was very difficult to appreciate specific uncertainties across the different sections of the case
study itself. The DWC strongly suggests that these are critical issues for the risk manager and
must be brought together in a more easily identifiable way, perhaps in a separate section. In
that section, a table listing the uncertainties should distinguish between uncertainties due to
the lack of data, uncertainties in prediction (e.g. structure activity analyses), and uncertainties
that relate to quantification of risks.  The latter issue is usually grounded in science, but
virtually always steps into the policy arena. Some sense of this transition is provided in
Appendix A, but it is not explicit. Further, a discussion of how these limitations and
uncertainties might affect the estimates of risk should be included in the text associated with
the list.

       The approach taken in the case study did illustrate some tools that could be applied in
a comparative risk analysis; however, the DWC has significant concerns about the validity of
cancer risk estimates that arose from the analysis.  The reasons for this concern derive largely
from the uncertainty of the data base and the assumptions that were needed to use it in the
case study.

        In each analysis, there will be a need  to derive a best estimate with "confidence
intervals" that capture the uncertainties in the estimate (both qualitative and quantitative).
Discussions of variability in the DBP cancer risk estimates were not included as a separate
issue.  There were general discussions of the possibility that children or the aged might be
more susceptible. A more specific discussion of variability was included for the exposure
variable. A major issue of variability that was not discussed adequately is the variation in
susceptibility in human populations. There are sufficient data for some of the DBFs to project
some determinants of that susceptibility.  Of course, there is no basis for discussing variable
responses for the vast majority of chemicals considered.

       b.  Microbiological Risks

             i)  Overall, are the risk estimates based on exposures to the pathogen
             plausible and reasonable given the goals and limitations of the case study?
             Risks from only a single type of microorganism were assessed in the case
             study; was this an appropriate selection. If so, are there other organisms
             which should be evaluated with this approach in future applications of the
             CRFM and why?

             Are the uncertainties presented in the assessments of pathogens

                                          A-5

-------
             adequately described and characterized? Are there significant
             uncertainties that are not identified? Are the presentations and
             discussions of variability appropriate?

       The selection of Cryptosporidium for this first case study of the CRFM was an obvious
choice.  This organism is an important pathogen that is well studied, has been shown to be
transmitted by water, causes disease in both special populations and the general public, does
not replicate in the environment, and is very resistant to removal and inactivation by current
treatment techniques.

       The case study assumes that an individual can be reinfected with Cryptosporidium
every twelve weeks. This assumption corresponds to no immunity.  Further, during the
interaction with the Agency,  NCEA representatives noted that this assumption was based on
studies of Schistosomiasis.  This should be explained more fully in the document. In fact,
studies by Chappell  (1998) indicate that there may be some protective effect as a result of
exposure to Cryptosporidium. This study is further discussed in subsection b.ii. below.

       Eventually, the CRFM will have to include consideration of other pathogens where the
variables will be more numerous and complex. A major issue will be the fate of pathogens  in
the distribution system.  Most pathogens can be inactivated during treatment, but some can
remain infective and reach the distribution system. Some bacterial pathogens can even
colonize the distribution system. This was not considered in the case study, and it is not an
issue as long as Cryptosporidium is the only pathogen considered because it will not colonize
the distribution system.  However, this factor should be considered in future case studies,
especially if bacterial pathogens are considered in the analysis.  As application of the method
is expanded, it will be necessary to consider that there is not a consistent ranking in the
susceptibilities of waterborne pathogens (e.g., viruses, bacteria, other protozoa) to disinfection
strategies (e.g., chlorine, ozone, UV light).  This will also have to be taken into consideration in
future case studies.

       Weighing the benefits and costs of two interventions when multiple waterborne
pathogen are being considered will require a more sophisticated data base.  If there is
evidence that a specific intervention, such as ozonation, will reduce the  risks from several
waterborne pathogens, then all such reductions should be included in the overall consideration
of the intervention.  Not to do so leads to an underestimate  of the effect of the intervention.

             ii) Are the identified hazards related to exposure to the protozoan parasite
             Cryptosporidium parvum appropriate?
             -Are the probabilities of infection and disease conditional on exposure
             appropriate for the two populations considered (i.e., the general population
             and the AIDS subpopulation)? Have all of the possible outcome categories
             been  identified?
             -Is the scientific basis of the dose-response model for predicting pathogen
             risks well-founded and adequately described and supported?
             -Should other subpopulations (e.g., the elderly) be evaluated in future
             applications of the CRFM?
                                         A-6

-------
       Hazards from Cryptosporidium exposure are discussed and the authors emphasize the
need for more data in this regard.  However, the case study and the methodology focus
entirely on endemic hazards from Cryptosporidium (by implication other microbial contaminants
would be handled in the same manner). There is little recognition in the document that
indicates that endemic hazard is less well established in drinking water than is the occurrence
of sporadic outbreak of infections.

       The distinction between sporadic outbreaks and endemic disease is important because
the treatment strategies appropriate to these two circumstances differ significantly.  This is
also not recognized in the case study. While there may be some reason to be concerned with
endemic disease, it is not appropriate to ignore established hazards that occur as outbreaks.

       The dose-response model for the prediction of risks associated with Cryptosporidium is
based  upon studies in healthy volunteers.  There are other populations that need to be
considered in future case studies.  It appears that  other subpopulations, such as the elderly,
may be at increased risk from Cryptosporidium.  It has been suggested that the mortality ratio
for individuals in nursing homes may be as high as 50%. If, indeed, this mortality risk is
present, it would be a significant consideration in the risk assessment calculations.

       Infectivity of Cryptosporidium is another important factor for this risk (as is shown in the
sensitivity analysis).  Significant sources of uncertainty exist for these estimates and the
following issues should  be considered:

       aa) Estimates of the infectivity of Cryptosporidium should be based directly on analysis
       of the human challenge studies conducted by Chappell (1998). Do not rely on the
       interpretation by Perz et al (1998). There appears to be a large range of dose-
       response associated with various strains of Cryptosporidium.  Dr. Chappell's dose-
       response research indicates that the ID50 ranged from around 9 oocysts for the TAMU
       strain to 132 for  the Iowa strain to 1100 oocysts for the UCP strain. To date, they have
       only tested genotype 2 Cryptosporidium strains (animal strains) and not genotype 1
       (human) strains.

       bb) What is the effect of pre-existing anti-Crypto antibodies on host susceptibility to
       infection and what proportion of the population (stratified by age) have antibodies to
       Cryptosporidium? Dr. Chappell observed that antibody positive volunteers had an ID50
       of 1880 oocysts  (Iowa strain), which suggests that there may be some protective effect
       that is overcome at high doses. Antibody positive subjects had more severe illness and
       longer duration of illness, but fewer of these subjects shed oocysts at detectable levels,
       and those who did shed oocysts did so at lower concentrations. The implications of this
       for secondary transmission should be considered.

       cc) What are the probabilities of reinfection? The case study assumes that an
       individual can be reinfected as frequently as every 12 weeks (page 5-47). The
       reference for this is Hurst et al. (1996) but these authors are not infectious disease
       clinicians or epidemiologists. The primary source of this information should be
       explored.  EPA staff explained at the February 1999 meeting that data on
       schistosomiasis  was used to estimate probability of reinfection of Cryptosporidium. It is
       not appropriate to extrapolate from a helminth infection with a very different life cycle

                                         A-7

-------
      and pathogenesis to a protozoan infection. The data from Dr. Chappell's rechallenge
      experiments should be examined, and the opinion of Centers for Disease Control
      (CDC) experts in the Parasitic Diseases Branch should be solicited.  Because there are
      very limited data on this issue, it should be treated as a source of uncertainty.

      dd) In calculating risk, EPA assumed that the probability of infection depends on the
      number of organisms consumed in a 12-week period. In doing this, EPA assumed that
      the probability of infection is the same whether one is exposed to X organisms in a
      single dose or in X/84 organisms in 84 daily doses. There are no data to support this
      assumption. The resulting risk may  be an over- or under-estimation of the actual risk.

      ee) What are the probabilities of mild illness, moderate to severe illness and death
      given infection for immunocompetent and immunocompromised hosts? Again, it is
      important not to rely solely on the analyses of Perz et al. (1998).  Investigators at  CDC
      should be contacted regarding what they have observed in outbreak investigations and
      studies of Cryptosporidium infections in immunocompromised populations.  CDC
      should also have information about the infectious dose for immunocompromised
      individuals compared to immunocompetent individuals.  Infectious disease clinicians
      differ in their opinion on this matter.  Some believe that,  if stomach acidity is in a normal
      range, then the infectious dose would be the same for immunocompromised and
      immunocompetent individuals. Others clinicians believe that the infectious dose is
      lower for AIDS patients.

3.  Chemical Dose-Response Assessment

      The goal of both DBF and microbial risk estimation is to provide realistic, central
      tendency risk estimates that can be used to compare the treatment technologies
      in question. EPA chose to use a response addition model (as described in the
      1986 US EPA Mixtures Guidelines); see also Hertzberg et al. (in press) based on
      the lack of data on the mechanism of action of the DBP. This  is a mixtures
      approach to addressing DBP toxicity. Other approaches, for example, another
      mixtures approach such as dose addition or an assessment of individual chemical
      constituents of the mixture, could have been developed. It should be noted that,
      under the conditions set forth in the case study, DBP risk is significantly smaller
      than the risks posed by exposure to  Cryptosporidium.

      a. Is the discussion of mechanistic toxicity data for the individual chemicals
      and/or mixtures appropriate?

      The Committee did not believe that reproductive and developmental toxicity data were
handled correctly in the case study. Part of the Committee's concerns were expressed in
section 3.2.3.  In addition, the use of the response addition model, with its assumption of  no
threshold, is not consistent with Agency policy for the assessment of these endpoints. As
noted, this makes little difference for the present case study. However, once this type of
simplification has been used, it may set a precedent that is likely not to be acceptable where
these hazards are more critical.
                                        A-8

-------
       The current document provides relatively little description, analysis, or summary
defining the mechanism of action for the individual DBFs or for DBP mixtures. The substantial
expansion of discussion to include defining the mechanism of action would probably serve little
useful purpose in this case study, as knowledge of mechanism of action did not appear to
influence the model inputs or the interpretation of the model results (except for chloroform).

       However, for a variety of reasons, the Committee felt that it would have been useful to
group DBFs by presumed modes of action in a table (including an "unknown" category).  First,
as discussed in the next section, this information should underpin the model(s) used for
estimating risks from mixtures.  Second, the Agency is currently developing a document,
Guidance for Identifying Pesticides and Other Chemicals That Have a Common Mechanism of
Toxicity. To include this information would provide some consistency with other Agency efforts.
Third, such a summary might provide a useful starting point for a more explicit rationale for
using Quantitative Structure Activity Relationship (QSAR) predictions of the activity of
compounds for which there were no toxicological data.  Essentially, these chemicals all
appeared to treated as non-threshold toxicants, whereas chemicals with data frequently
appeared to treated as if there was a threshold [e.g. bromodichloromethane (BDCM) effects on
developmental toxicology].

       b. Dose-response analysis of individual DBFs (point added by the DWC)

       The individual DBP dose-responses were modeled by a linearized multi-stage model
with a threshold parameter using toxicity data  from the Integrated Risk Information System
(IRIS).  The use of this model and the IRIS data is appropriate and justified, as a starting point.
However, IRIS typically has  not contained sufficient information to allow uncertainty to be
evaluated and propagated by the model.  The case study failed to provide an adequate
assessment of the strengths and weaknesses of the studies used. For example, the route of
DBP exposure used for various studies, corn oil gavage vs drinking water, does not appear to
have entered into the evaluation of the data. In addition, several data sets failed to converge.
The case study contained little evaluation or analysis of the adequacy of the linearized multi-
stage model to adequately fit the IRIS data sets, nor was there an evaluation or analysis of the
adequacy of the IRIS data for the dose-response model used. In this sense, the general
methodology and the case study do not provide adequate  direction  to ensure consistency in
application  of the approach to new case studies. The DWC also is concerned that the IRIS
data set is incomplete, may  lack quality control, or be comprised of  obsolete data. Thus the
use of IRIS data should be supplanted or supplemented by primary data when appropriate.

       c. Appropriateness  of assumptions, techniques and models used in the
       estimation of risks posed by DBP mixtures, including both statistical and
       biological  considerations.

       The choice of a response additivity model to predict the health effects of the DBPs is
an appropriate approach for some types of toxicological effects. The response addition
approach is appropriate for summing risks for certain mechanisms of carcinogenesis, for
example. Clearly NCEA understood this because it chose not to treat chloroform risk as
response additive.  To be  consistent, the  possibility of other sublinear dose-response
relationships should have  been entertained more explicitly. The selection of the response
addition model for reproductive and developmental toxicities is inconsistent with previous

                                         A-9

-------
Agency policy in the drinking water area.  For chemicals where thresholds are usually assumed
the dose-addition model should be used.  Applying the response addition model in this
"exceptional way" should be specifically justified.

       In the context of this case study, the selection of response addition vs. dose addition
has little impact on the outcome. However, situations defined by future case studies or
applications of the methodology may include chemical exposure scenarios for which
experimental or epidemiological data support interactive toxicity (synergism or antagonism).
Neither the general methods, nor the case study section, defined criteria or approaches for the
inclusion or exclusion of data that support such interactive effects. The assumption of
independent action as a requirement for using the response addition model is not one that
deals effectively with the biology of interactive effects.  There must be a methodology for
integrating such effects into the process, even if they are relatively rare at low doses.

       Other assumptions made in the case study are probably not appropriate for the
analysis. For example, lumping all developmental toxicities into the highly dependent category.
Clearly, the degree to which this lumping is valid depends  on the nature of the birth defect. It is
not established that  minor effects, such as decreased crown-to-rump length, actually lead to
some level  of dependence. Such measures are useful screening assays, but they are not
predictive for more serious effects, and make the  transition to QALYs very difficult if not
impossible. An assumption of dependency might be more reasonable  for some other effects
such as valvular defects in the heart that have been detected with many DBPs. More specific
criteria should be developed to provide guidance for extrapolation of developmental toxicities
than simply conversion from a continuous to a stochastic response.

       The General  Methodology section should also consider defining strategies to
conceptually address the cumulative  risk of exposure to multiple pathogens.

       d. Usefulness of the mixtures risk assessment approach for  comparing health
       risks across drinking water treatment scenarios and levels of DBPs.

       The use of the mixtures risk assessment approach  was  valuable in that it provided an
illustration of how the framework might be applied. This methodology has its roots in statistical
rather than mechanistic considerations and as a result provides a somewhat distorted view of
the possibilities that  exist for interactions.  Independence of action is an assumption for
response addition, primarily because it is assumed that differing effects are not related and at
low dose will be randomly expressed in the population. Synergy arises from independent
action as well, but affect function through  converging and perhaps redundant pathways.  This
requires distinct mechanisms of action, but sets up the situation for an  amplified severity of
effects and observation of effects where none were previously observed.

       Consequently, it is important that the underlying mechanisms by which effects are
produced be explicitly identified whenever possible. This identification  becomes even more
important as QSAR is used to predict toxicological responses.  Inherently, QSAR predictions
are based in large part on common mechanisms.  A difficulty encountered in the report is that
in many cases it was tacitly accepted that chemicals in the same classes acted by the same
mechanism. There is certainly more  than one mechanism  involved in trihalomethane  and
haloacetic acid's effects.  There has been at least one report of synergy between low doses of

                                        A-10

-------
dichloroacetate and trichloroacetate where the effects of high doses of both byproducts results
in inhibition of one another's carcinogenic effects (Pereira and Phelps, 1996) clearly the two
compounds produce liver cancer by distinct mechanisms (Stauber et al., 1998). It is extremely
important that these differences in mechanism be kept track of in comparative risk scenarios
where projections of effects are extended over a large number of chemicals for which
descriptive toxicological data are not even available.

      e. Appropriateness of assessment of risks posed by the unidentified halogenated
      fraction (TOX).

      The DWC has reservations about extrapolating risks associated with those DBFs that
have been extensively studied to the much greater number of DBFs that have not been
studied. First,  the compounds that have been studied do not even come from the same
chemical classes as those that have not been studied.  Second, within the classes that have
been studied there  are significantly different potencies that seem to be related to differences in
the mechanisms of metabolic activation (e.g., brominated versus chlorinated trihalomethanes).
Third there are diverse types of effects produced (e.g., cancer, reproductive toxicity,
developmental effects) which may or may not have common mechanisms. Fourth, within the
context of  DBFs that have been identified as carcinogens, neither the target organs, nor the
potencies  predicted, approach those projected by analyses based on  epidemiology studies
that have been published by a number of investigators.  Part of this difference might be
explained  by the relative lack of data on most DBPs,  but there are also other potential
explanations.  The epidemiology data may be in error, there may be interactions that have not
been appropriately  identified, or humans may simply  respond  differently. These all must be
treated as  uncertainties.  While the inclusion of this projection provides some conservativeness
to the project, it does not diminish the level of uncertainty.  The CRFM does not provide
guidance on how to handle the uncertainties associated with these projections.

      Having  made the general point above, the Committee  notes that extrapolation of risk
from known DBPs to the unidentified halogenated fraction  (TOX) was not a critical issue  in the
context of  the present case study because the health impacts were dominated by the
overwhelming contribution of Cryptosporidium. That is to say,  disinfection byproducts, in
general, had little impact on the method's results and conclusion. However, analysis of other
cases will  require site-specific estimates of the likely  make-up of the byproducts that will be
formed.  The byproducts produced by ozone and chlorine dioxide are  primarily non-
halogenated or are  not organic and are not captured  within the TOX.  Moreover, source water
quality makes major differences in both the type and  extent of byproduct formation.  Thus, the
extrapolation of the health risk from  DBPs to the TOX in this case scenario must be recognized
as an exercise that is specific to circumstances of this particular case  and, therefore, has
limited, if any, application to other case studies.

      f. Have the  appropriate DBPs been addressed? Cancer that results from
      chloroform  was not addressed in the analysis because of the evidence (Drinking
      Water NODA, EPA, 1998) indicating that this chemical's mechanism of action
      exhibits a threshold. Is this  an appropriate decision?

      One must start with the disinfectant byproducts for which sufficient toxicological
information is available to make an approximate estimate of cancer risk. The analysis

                                        A-11

-------
excluded chloroform because of conclusions that were published in the 1998 EPA Notice Of
Data Availability (NODA). On the basis of that conclusion, it was appropriate to drop it from
the analysis because the concentrations seen in the water consumed were below the MCLG
put forward in that NODA.  However, since that time, the Agency has modified its opinion on
how chloroform will be handled in its regulatory rule making and the SAB must reserve
judgment on specific chloroform issues until it conducts the review discussed in the EPA final
Stage 1 rulemaking for DBPs.

       The removal of chloroform from consideration in the case study raises an issue of how
confined the CRFM would be by Agency policy.  As indicated in previous sections, there are
data that suggest modes of action among various carcinogenic DBPs that might warrant similar
treatment under the proposed new cancer risk assessment guidelines. Therefore, one has the
dilemma of whether the analysis can go ahead on the basis of current science or whether it
must await wider policy decisions to be made, a process that can be very slow. It would be
inappropriate for the DWC to prejudge Agency policy in this kind of exercise. However, we
can suggest that it would have been useful to illustrate the relative impact of linear and non-
linear extrapolations to low dose in this case study. As it turned out, of course, the disinfectant
byproducts contributed little to the risk trade-off in this illustrative case study and such an effort
would have been only an academic exercise.  In a case study aimed more specifically at the
relative risks associated with byproducts from different disinfectants, these questions would
have been critical.

       g. For future applications of the CRFM there is a potential interest in using the
       results of epidemiologic studies, please provide suggestions for the use of the
       epidemiologic data in the future applications.

       Epidemiological evidence that is consistent across multiple  studies can provide a basis
for estimating the health risks associated with waterborne disease. In these circumstances,
the data should be used to the extent that they are dependable (e.g. data may provide strong
evidence for hazard identification, but not be appropriate for quantitative estimates of risk)
When epidemiological data are uncertain, it is important to rely on toxicological information. A
strong case can be made if the toxicological and epidemiological data present a consistent
picture from a qualitative and quantitative point of view. As noted,  however, a level of
consistency from the two data sources has yet to be achieved.

       There are some shortcomings of epidemiology that need to be recognized. Some
epidemiological data support the hypothesis that endemic risks from microbes in drinking water
exist.  However, the data do not indicate if the risks are governed by the source water or by the
distribution system.  Without more refined information, it is difficult  to translate these data into
a strategy for mitigation. Clearly, more quantitative approaches are needed to make these
data useful within the context of the methodology.

       In a similar vein, the epidemiological evidence associating disinfected drinking water
with cancer and spontaneous abortions is interesting.  The findings suggest that there are as
yet unresolved issues about  disinfectants and their byproducts. To accept these data as
indicative of a problem raises the question of which treatment strategies would mitigate such
risks.  These risks were associated with chlorination; however, it cannot be stated with
confidence that these risks would not be found with alternative disinfection strategies. Thus,

                                         A-12

-------
these data serve as an incentive to better characterize the complex byproduct mixtures of
chemicals resulting from chlorination in subsequent epidemiological studies. Also,
toxicological studies that are designed to identify the cause(s) of the increased risk are
needed.  The CRFM can be a powerful tool for identifying shortcomings of this type in the
available data. That would provide the Agency with a much better appreciation of how it
should spend its  research dollars to resolve issues with more specificity than has been
possible in the past.

       It is also important to recognize that epidemiological  investigations provide other useful
data for use in the CRFM.  Specifically, waterborne disease outbreak investigations provide
useful information on risk factors associated with outbreaks, susceptible host populations,
incubation period, clinical characteristics of infection, secondary transmission of infectious
agents, effective  interventions and the ability of disease surveillance systems to recognize
outbreaks.

4.  Exposure

       a.  General issues related to the evaluation of exposures

       The accuracy of exposure assessment in this case study depends on information on
the occurrence of the microorganisms or disinfection byproducts of interest and on estimates
of individual exposure to tapwater via ingestion, inhalation and dermal contact. For both
disinfection byproducts and microbial pathogens, surrogate  parameters are often measured
instead of the occurrence of the actual pathogenic organism or hazardous DBP. Assumptions
must be made about the accuracy of the measured water quality parameter (sensitivity and
specificity) and how well this parameter serves as an indicator for the presence of the actual
hazardous waterborne substance.

       In general, the assumption of steady state conditions in source water and treatment
efficacy (e.g., average annual exposures) departs significantly from real conditions in many
treatment plants  (especially surface water sources) for both  microbial contamination and
disinfection byproducts. Source water variation and variation within the water distribution
system can significantly affect concentrations and types of microbial pathogens and
disinfection byproducts. Most microbial waterborne disease outbreaks have been associated
with a spike of microorganisms in the distribution system because of spikes of contaminants
entering the source water (spills, spring melts, heavy rainfall, etc.), failures at treatment plants,
or failures in the  distribution system.   Potential failure of the reverse osmosis units and
problems with monitoring their performance also need to be considered in the case study. It
may be more realistic for the model to identify distributions of exposure that take all of these
variables into account rather than average annual exposures.

       Concentration data for individual DBFs in the case study were recalculated from
sampling data assuming a  log normal probability distribution function and substituting half the
detection limit for non-detects instead of zero. The Committee discussed "The Use of
Censored Data" in section 6.2 in its July 19,  1995 SAB "Review of Issues Related to the
Regulation of Arsenic in Drinking  Water." The comments  made then are applicable to
exposure calculations with  censored DBP data as well.
                                         A-13

-------
       Source water variations can result in significant changes from one treatment plant to
another in identifying the potential concentration of Cryptosporidium oocysts in the treated
water.  The most recent meeting conducted by EPA with a panel of statistical experts on
clearly very preliminary data from the Information Collection Rule showed at best a 20%
detection of oocysts in the treated water of those sites analyzed, with as low as 5% of the sites
showing positive oocysts in their treated water. Site variability is extremely critical as  is being
able to evaluate the risks of ozone and/or membrane filters at the individual home.

       The case study used recent Cryptosporidium occurrence data from LeChevallier in
Trenton, NJ. The document uses the term "active" oocyst, which is confusing. It is almost a
contradiction in terms.  The determination of whether oocysts are likely to be infective is dealt
with in operational rather than absolute terms. The terms conventionally used to describe such
oocysts are: total, viable (based on dye permeability?) and infective (based on cell culture or
mouse assay). These definitions should be clarified in future case studies and discussions of
the method itself.

       The Committee suggests that an upper bound on risks might be  estimated by  using
total number of oocysts instead of total number of "active" or "viable" oocysts. Given  that the
methods to detect Cryptosporidium in drinking water are insensitive, any detection of  oocysts
in treated water should be used as an indication of the possible presence of infectious oocysts
in the water and a potential threat to human health. However, it must be recognized that the
most probable estimate of risk depends upon the viability of oocysts and their infectivity to
humans.

       The effect of colonization of reverse osmosis units by heterotrophic bacteria and the
risk of opportunistic pathogens for immunocompromised individuals should be considered.
See the paper by Payment et al. (1991). Payment reported that families with RO units with
high levels of bacterial growth were at increased risk of gastrointestinal symptoms. How do
you monitor bacterial colonization and water quality from these units?

       b.  Is the distribution of drinking water consumption rates appropriate?

       The DWC was unable to determine the exact distribution of drinking water consumption
rates used, although there was a significant amount of discussion related to the topic in the
case study. The report (Section 5.2) cited the US EPA Exposure Factors Handbook  (1997) in
providing tap water consumption rates for the  general population.  The source of the data was
Ershow and Cantor (1991). The rates were listed according to age groups. Then the report
provided age-weighted averages of the values to approximate consumption by 5-yr
increments. The data from Canadian Minster of Health and Welfare (1981) were used to
correct for the unheated tap water consumption rates.  The Perz et al. (1998) study was used
to derive the percentage (70%) of unheated tapwater for the AIDS population compared to the
general population, while total consumption was assumed to be the same. The Ershow et al.
(1991) data were used as a basis for having no adjustment made to reflect any potential
differences between pregnant women and the general  population.

       At this point in the Agency report (end of Section 5.2), before moving on to the
discussion of DBPs, or in Section 5.3.2, it would be useful to clearly specify the exact
consumption values used for deriving the risk estimates in the later part of the report,  along

                                        A-14

-------
with information on the associated parameters (e.g., age-weighted adjusted, percentile, total
water consumption).  In Section 5.3.1, tap water consumption is represented by Y in the
equation which described the response addition model used in the case study.   Section 5.3.2
described the assumptions made for tap water consumption.  But it seems that there is no
clear indication as to the final values used.  For example, if the values are from Table 5-2 as
noted in Section 5.3.5 then it should be indicated right after the discussion of the data sources
for consumption rates.  The correct values are needed for verification of the results.

      A related question is whether the water consumption rates used in the case study are
the same consumption rates used by U.S. EPA in its programs and are they consistent across
EPA programs? The Office of Groundwater and  Drinking Water is developing statistical
treatments of data gathered from the US Department of Agriculture's  Continuing Survey of
Food  Intake by Individuals that may be of use in future application of the CRFM to drinking
water problems.

      c. Is the identification of the fraction of unheated drinking water consumed
      appropriate?

      The identification of the unheated portion  of the water seemed to have appropriate
application to determining exposure to microbes.  In the case of DBPs, NCEA decided to utilize
total water consumption rather than a heated or unheated fraction (p.5-16). There are many
reasons to question how this modifies the accuracy of the risk assessments for DBPs.  Heating
would clearly increase volatilization of some DBPs,  but it may well accelerate the formation or
degradation of others.  Therefore, the assumption is justified, but the  uncertainties should be
noted.

      d. Assess the validity of the assumption that the pathogen would not survive the
      pathways (referred to as a "preparation methods") from tap to consumption
      identified.

      The question of "pathways" and "preparation methods" was difficult to identify
specifically in the document. There was a discussion of the effect of heating water on
pathogen survival, but that was all that could be identified.  The assumption that
Cryptosporidium would not survive in heated water may be correct. The extent of heating is a
behavioral variable for different uses (especially for beverage preparation) that was not
captured and how this affects the validity of this assumption was not discussed in the
document.

      e. Assess the validity and significance of the assumption that DBF concentrations
      do not change as a result of transport through the distribution system and many
      pathways through which  water is consumed (e.g., boiling to prepare tea).

      The assumption that DBPs do not change in distribution systems may be an acceptable
assumption in this particular case study, but not an acceptable assumption generally. Making
such an assumption is equivalent to assuming that free chlorine is always the primary
disinfectant or that all source waters are the same.

      f. Should other routes of exposure be included such as dermal  and inhalation

                                        A-15

-------
       routes? Would the contribution of these be significant?

       There is no doubt that routes of exposure in addition to ingestion need to be accounted
for in applications of the methodology to specific cases.  However, the contribution of these
routes  need to be considered in the context of the integrated distribution of total exposure.  If
one adds the 95th percentile exposure via ingestion, to the 95th percentile of inhalation
exposure to the 95th percentile exposure through the skin one will not arrive at the 95th
percentile of total exposure.  Clearly, the importance of various routes will vary depending
upon the nature of the contaminant.  Inhalation of aerosols has not been seriously considered,
but as  illustrated  by outbreaks of Legionella several decades ago contaminants need not be
volatile to present a significant inhalation hazard.

       There is no doubt that under certain conditions dermal and inhalation exposure can
make significant  contributions to the overall exposure to disinfectants, their byproducts and
microbial agents. Clearly some microbial agents in water are effectively spread through the
inhalation route (i.e. Legionella), and certainly inhalation of DBFs during showering is well
established. However, the method for treating such exposures needs careful consideration.
EPA has indicated that in order to account for volatile chemicals, an extra 2 liters/day is
considered (EPA, 1996).  This value seems to be derived from studies that emphasized
extremes in water usage rather than on how these various  routes contribute to exposure in a
population. If the Committee has an appropriate understanding of the issue, the question is an
overall distribution of effective exposures to individual agents that need to  be considered.
Clearly, the effective dose will vary depending upon the physical-chemical  properties of the
byproduct and the relative effectiveness of inhalation versus ingestion as a mode of
transmitting infections by different microbial agents.  Therefore, it is not appropriate to use the
approach suggested in the literature which treats inhalation and ingestion routes as some set
fraction of overall exposure.  The exposure estimates should be the population integral of
these various routes with their associated confidence intervals.  In any final analysis, these
integrated exposures should include consideration of the pharmacokinetic  variables of different
routes  of exposure to low doses of these chemicals in humans. Risks from DBPs  should
consider dermal  and inhalation  routes.

5.  Health Conditions

       a. Are the assignment  and definition of health conditions appropriate for the
       progression of Cryptosporidiosi&

       The DWC consolidated  its comments on this  point in section 2.b.ii of this Appendix.

       b. Are the assignments and definition of health conditions for developmental,
       reproductive and cancer risks appropriate? Are latency periods and reversibility
       issues handled reasonably?

       The DWC viewed the case study as an illustrative exercise, not a final analysis of the
issue.  Moreover, the results of the case study are largely trivial because of the overwhelming
risks that are assigned to risk from Cryptosporidium.  Nevertheless, there were methodological
issues  that should be addressed before the methodology is applied in a more definitive
analysis.

                                         A-16

-------
       There are significant difficulties in the way these hazards were defined and handled.
Mostly, the draft document reflected a superficial treatment of the toxicological literature of
most disinfectant byproduct classes and culminated in the assignment of toxicological
properties to the rest of the total organic halogen present in the water across species and to
lower doses.  This is an unprecedented level of extrapolation beyond the available data.  The
diverse set of mechanisms (or modes of action) and the potential for sensitive populations to
certain byproducts were not considered or identified as sources of uncertainty.  Scattered
throughout the Committee's comments are questions of whether some of the translation of
information into human conditions are appropriate.  These will not be repeated here. Of most
concern in the context of this question is to make sure that it is clearly recognized that
definition of health conditions at the human level based on animal data can raise serious
concerns about the validity of the overall analysis when they are eventually converted into
QALYs. Extreme care must be taken to ensure that conservatism on one side of the equation
is not propagated in such a way that it distorts the final comparisons.  A specific example of
how this can become a problem is outlined in subsection c below.

       For future case studies and further development of the method, the Agency will need to
become considerably more invested  in understanding the complex literature that is developing
in this area than was apparent in this first attempt.

       c. Are the uncertainties in this process identified?

       Uncertainties in this process are not identified with sufficient specificity to drive
research that could resolve these uncertainties. Several prior sections of this report address
specific problems in this area and will not be repeated  here (see Section 3 of this report).

       A substantive problem was identified  in the case study through discussions between
Committee and National Center for Environmental Assessment (NCEA) personnel  at the
February,  1999 meeting that needs to be noted.  Essentially,  the problem was that data
obtained in animal studies were used to characterize the risk "quantitatively" that is associated
with disinfection byproducts.  However, human data were used to estimate the resulting
QALYs. This introduced a problem that was not explicitly identified in  the analysis, in part
because it was not apparent that this manipulation of the data was actually performed. The
specific problem that could arise comes from the fact that the "risks" one estimates from the
epidemiology data outstrip the toxicology data by 1-3 orders of magnitude, depending upon
how the risks are actually calculated. Second, the epidemiology data  introduce cancer sites
that have not frequently been found in toxicological studies.  This leads to a significantly
broader band of uncertainty in the results of the analysis than is currently portrayed.  These
assumptions strike at the very heart of the analysis  because they directly impact the QALYs
assigned to the carcinogenic endpoints. In addition to the impact on the quantitation of the
risks, it is well known that cancers at different sites carry significantly different implications for
the quality of life as well as life expectancy once diagnosed.

       This problem has other practical consequences in the case study. In all likelihood the
reason this crept into the example was because there is no causal agent that can be
associated with the bladder and colorectal cancer sites seen  in humans. Consequently, there
is no way to determine the impact of changing water treatment.  However,  if these  risks are

                                         A-17

-------
real, they would completely drown out the risks that would be calculated from the DBFs that
have been characterized in animal bioassays for cancer. The Committee recognizes the need
to be pragmatic in these analyses, however, the analyst must be careful not to gloss over or
"forget" about these shortcuts because the result could be an analysis that does not provide a
true picture of the "state of the science".  Consequently, it is very important that these
differences be captured in the uncertainty analysis.  If this was the case, it was not made
apparent in the document.

6. Common Health Metric

      a. Is the use of Quality Adjusted Life Years (QALYs) appropriate?

      The QALY concept seems to result in an appropriate common health metric for this
case study. As indicated above, however, more attention must be paid to the string of
assumptions that are inserted into deriving the QALY from experimental data. NCEA did not
entertain alternatives.  Now that they have had some experience with this metric, it might be
appropriate to step back and assess alternatives.

      In the context of water related problems, the Committee suggests that NCEA consider
how acute outbreaks fit into this picture more specifically.  If outbreaks are included in the
analysis the number of "QALY's" they effect may not be great enough to justify their inclusion
in the analysis because they don't occur that often. This would be a mistake.  The morbidity
and mortality associated with outbreaks is of greater impact than the morbidity and mortality
resulting from estimates of endemic microbial infection and chronic disease from exposure to
chemicals.  This is because, while the morbidity and mortality in these epidemics is real and
measurable, the endemic and chronic estimates are just that, "estimates", extrapolations based
on assumptions of exposure, dose response, and, in the case of chemical exposure, cross-
species  extrapolation.  The proposed CRFM may not be suitable for directly comparing
outcomes that are so different in their quality.

      b. Is the assignment of QALYs given the health conditions appropriate?

      The assignment of QALY costs to health outcomes in Chapters  5 and 6 required a
great many assumptions, combining results from a number of different studies and adapting
results derived in one situation to a quite different situation.  For example, the study used the
QALY cost of a year in severe pain derived in the Hamilton, Ontario study to estimate the
QALY cost of a bout of severe acute illness, by dividing the one-year cost by 365 to get a one-
day figure, and then  multiplying by 14 for a two-week illness.  These assumptions embedded in
this procedure are obviously heroic. On the other hand, the authors really had little choice, for
attempts to estimate QALYs directly for acute illness have not been very successful.

      It would be useful to do a survey that directly pits against each other the two risks
examined in the case study (acute Gl disease and cancer after some period of latency).  The
survey results could  be used to estimate directly the QALY cost of the various outcomes of
these diseases, which  could then be compared to the QALY costs estimated in the current
study. A close correspondence between the survey-based QALY costs and those estimated in
the case study would help build confidence in the method used to calculate QALYs.  If there
were no close correspondence, then survey results would nonetheless  help analysts develop

                                        A-18

-------
better estimates of the QALY cost of acute nonfatal disease.

       There does appear to be a problem with the estimate of the QALYs for the two
alternative technologies for the AIDs subpopulation.  The number of QALYs that the point-of-
use filtration system would save appears to be about 22 QALYs per AIDS patient (11,636
QALYs for 429 persons), which is much greater than could reasonably be expected from the
elimination of cryptosporidiosis in this population, given the short expected lifespan of these
individuals (about two years, given the conditional survival probability of 0.53). As the
document points out (p. 6-41) this QALY estimate is actually shared by a number of AIDS
victims over the 20-year planning horizon; as patients die, others contract the disease and take
their place.   It still seems high on an individual basis, but impossible to say for sure without
further information. Is the analysis supported by an explicit dynamic model of AIDS
encompassing both survivorship and new cases, one that takes into account existing and
anticipated future patterns of AIDS infectivity and survivability? Or is a simple "steady-state"
assumption made, one that assumes that there are now X cases of AIDS and that in future
years there will also be X cases? (If a steady-state is being assumed in the base case, it would
seem that conditions would no longer be in steady-state after intervention, for AIDS patients
would have a higher life expectancy while new AIDS cases would be added to the population
at the same  rate as the base case.)  If there is a dynamic analytical model, it would be useful
to include more information in the report.  If there is no such model, it would be an important
and useful addition the analysis. Considering the large share of the benefits that accrue to
AIDS patients,  it seems important to examine the dynamics  of this population carefully.
                                        A-19

-------
                                  APPENDIX B

                           Acronyms and Abbreviations
BCA        Benefit-Cost Analysis
CDW        Centrally Distributed Water
CE         Cost-Effectiveness
CRFM       Comparative Risk Framework Methodology
CSFII        Continuing Survey of Food Intake of Individuals
DBP        Disinfection Byproducts
DWC        Drinking Water Committee
IRIS        Integrated Risk Information System
MCLG       Maximum Contaminant Level Goal
MCL        Maximum Contaminant Level
MHI         Median Household Income
NAS        National Academy of Science
NCEA       National Center for Environmental Assessment
NODA       Notice of Data Availability
PDF        Probability Distribution Function
QALY       Quality Adjusted Life Year
QSAR       Quantitative Structure Activity Relationship
RO         Reverse Osmosis
SAB        U.S. EPA Science Advisory Board
SDWA       Safe Drinking Water Act Amendments of 1996
THM        Trihalomethanes
TOC        Total Organic Carbon
TOX        Unidentified Halogenated Fraction
WTP        Willingness to Pay
                                       B-i

-------
                                   REFERENCES

Ashby, J. and R.W. Tennant. (1994) Prediction of rodent carcinogenicity for 44 chemicals:
       Results.  Mutagenesis, 9:7-15.

Bull, R.J. and F.C. Kopfler. (1991) Health effects of disinfectants and disinfection byproducts.
       AWWA Research Foundation.

Canada Ministry of Health and Welfare. (1981). Tapwater Consumption in Canada.
       Document No. 82-EHD-80.  Public Affairs Directorate, Dept. of National Health and
       Welfare. Ottawa, Canada.

Chappell, C. L, C.R. Sterling, and W. Jakubowski. (1998) Cryptosporidium parvum volunteer
       study: Infectivity, Illness,  and Immunity. In Proceedings of the 1998 Drinking Water
       Progress Review Workshop for the 1995/1998 Science to Achieve Results (STAR)
       Grants. December 8-9, 1998. Arlington, VA. US EPA600/R-98/162. p. 16-17.

Erschow, A.G., L.M. Brown, and K.P. Cantor. 1991. Intake of tapwater and total water by
       pregnant and lactating women.  Am. J. of Public Health.  81:328-334.

Helsel, D.R. (1990) "Less than obvious statistical treatment of data below the detection limit,"
       Environmental Science and Technology. 24(12): 1766-1774.

Hertzberg, R.C., G. Rice, and L. Teuschler. (1998).  Methods for health risk assessment of
       combustion  mixtures. In Hazardous Waste Incineration: Evaluating the Human Health
       and Environmental Risks. S.  Roberts, C. leaf, and J. Bean, Eds. (In Press).

Hurst, C.J., R. Clark, and S.E. Regli. (1996).  Estimating the risk of acquiring  infectious
       disease from ingestion of water. Ch. 4. In Modeling Disease Transmission and Its
       Prevention by Disinfection. C.J. Hurst, Ed. Cambridge Univ. Press. Cambridge, UK. P.
       99-199.

I LSI Working Group. (1996). A Conceptual framework to assess the risks of  human disease
       following exposure to pathogens. ILSI Working Group; Risk Analysis, 16(6):841-47.

Kodell, R.L., J.J. Chen, and D.W. Gaylor. (1995)  Neurotoxicity modeling for risk assessment.
       Reg. Toxicology and Pharmacol. 22:21-29.

LeChevallier, M.W., W.D. Norton, M. Abbaszadegan, and T.B. Atherholt. (1998).  Short term
       variability of Giardia cyst and Cryptosporidium oocyst concentrations in a surface water
       source used for potable water.  New Jersey Dept. of Envir. Protection, Div. of Science
       and Research. April.

Morgenstern, R.  (1998). Regulatory Impact Analysis at the US Environmental Protection
       Agency.  Resources for the Future.  Washington, DC.
                                         R-1

-------
Morris, R.D., A-M. Audet, I.F. Angelillo, T.C. Chalmers, and F. Mosteller.  (1992) Chlorination,
       chlorination byproducts, and cancer: A meta-analysis. Am. J. of Public Health. 82:955-
       963.

National Research Council. (1983) Risk Assessment in the Federal Government: Managing
       the Process.  National Academy of Science, Commission on Life Sciences, Committee
       on the Institutional Means for Assessment of Risks to Public Health.  National Academy
       Press. Washington, DC.  191pp.

Payment et al. (1991).  Gastrointestinal health effects associated with the consumption of
       drinking water produced by point-of-use domestic reverse-osmosis filtration units.
       Applied Environmental Micro. 57:945-948.

Payment, P., J. Siemiatycki, G. Richardson, G. Renaud, E. Franco, and M.A. Prevost. (1997).
       Prospective epidemiological study of gastrointestinal health effects due to the
       consumption of drinking water.  Int. J. Environ. Health Research.  7:5-31.

Pereira, M.A. and J.B. Phelps. (1996) Promotion  by dichloroacetic acid and trichloroacetic acid
       of N-methyl-N-nitrosurea-initiated cancer in the liver of B6C3F1 mice.  Cancer Letters.
       102:133-141.

Perz, J.F., F.K. Ennever, and S.M. LeBlancq.  (1998) Cryptosporidium in  tap water.
       Comparison of predicted risks with observed levels of disease.  Am. J. Epidemiology.
       147:289-301.

Sobsey, M., A. Dufour, C. Gerba, M. LeChevallier and P.J. Payment (199) Using a conceptual
       framework for assessing risks to health from microbes in drinking  water. J. of the
       American Water Works Association. 85:44-48.

Stauber, A.J., R.J. Bull, and B.D. Thrall. (1998) Dichloroacetate and trichloroacetate promote
       clonal expansion of anchorage-independent hepatocytes in vivo and in vitro.   Toxicol.
       And Applied Phramacol.  150:287-294.

US Court of Appeals for the District of Columbia Circuit. (1999). May 14, 1999 opinion in the
       case of the American Trucking Associations, Inc., et al., versus US Environmental
       Protection Agency (Opinion  No.97-1440).

US EPA Science Advisory Board. (1995). Science Advisory Board (SAB) Review of Issues
       Related to the Regulation of Arsenic in Drinking Water. US EPA SAB, Washington,
       DC. EPA-SAB-DWC-95-015.

US EPA. (1997). Exposure Factors Handbook.  Vol.1. General Factors. Office of Research
       and Development. National Center for Environmental Assessment. EPA/600/P-
       95/002Fa. August.

US EPA (1998).  National Primary Drinking Water Regulations: Disinfectants and Disinfection
       Byproducts. Federal Register.  63(241):69398.
                                         R-2

-------