Report
of the
CCL Classification Process
Work Group
to the
National Drinking Water Advisory Council
April 30,2004
-------
CCL CP Work Group Report
TABLE OF CONTENTS
EXECUTIVE SUMMARY ES-1
ES.l Introduction. 1
1.1 Background and Purpose 1
1.2 Charge to the NDWAC Work Group 2
1.3 Guiding Principles 2
ES.2 Overview of Recommended CCL Classification Process and Overarching Issues 3
2.1 Building on the NRC Approach 3
2.2 Transparency and Public Participation 3
2.3 Overview of the CCL Process 4
2.4 Overarching Issues 6
ES.3 CCL Classification Approach for Microbial Contaminants 7
ES.4 CCL Classification Approach for Chemical Contaminants 8
4.1 Buildingthe Chemical Universe 9
4.2 Screening Contaminants from the Universe to the PCCL 9
ES.5 Moving from the PCCL onto the CCL 10
5.1 Quantifying Attributes for Use as Inputs to Classification Models 11
5.2 Overview of Classification Approaches and Work Group Recommendations 12
5.3 Development of a Training Data Set 12
ES.6 Summary , 13
CHAPTER 1: INTRODUCTION 1-1
1.1 Background on the Contaminant Candidate List and the
National Research Council Recommendations 1
1.1.1 The NRC Recommendations 1
1,1,2 Charge to NDWAC Work Group 3
1.2 Convening and Membership of the NDWAC CCL Classification Work Group 3
1J NDWAC CCL Classification Process Work Group Guiding Principles, 5
1.4 Summary of the NDWAC CCL Work Group Deliberation Process 5
1.5 Role of the CCL in Protecting Public Health and Implications of .Inclusion
on the PCCL or CCL... 6
CHAPTER 2: OVERVIEW OF PROCESS AND OVERARCHING ISSUES 2-1
2.1 Transparency and Public Participation 1
2.1.1 Why Transparency is Important for the CCL 1
2.1.2 Public Participation. 3
2.2 Overview of Recommended CCL Classification Process 3
2.2.1 Building on the NRC Approach 4
2.2.2 Parallel Processes for Chemical and Microbial Contaminants. 5
22.2.1 Identifying fa CCL Universe 7
2.2.2.2 Screening from 0ie Universe to the PCCL 7
2.2.2.3 ChanicterizingtheFCCLConlaminanis 8
-------
CCL CP Work Group Report
2.2.3 Developing a Prototype Classification Approach 9
2.2.4 Incorporating Genomic Information in the CCL Process 10
2,3 Overarching Issues 11
2.3.1 Integrating Expert Judgment into the Process. 11
2.3.2 Implementation of an Active Surveillance Process for New and Emerging Agents. 12
2.3.2.1 Surveillance Activities 13
2.3.2.2 Primary Source Literature Review 15
2.3.2.3 Additional Surveillance Activities and Recommendations. 15
2.3.3 Implementation of a Nomination and Evaluation Process for New and Emerging Agents. 16
2.3.3.1 Additional Considerations for the Nomination Process 17
2.3.3.2 Accelerated Listing Process 18
2.3.4 Information Quality Considerations 18
2.3.4.1 NRC Discussion and Recommendations 18
2.3.4.2 WorkGroup Considerations and Recommendations on Information Quality 18
2.3.5 Use of Quantitative Structure Activity Relationships (QSARs) 21
2.3.5.1 Introduction 21
2.3.5.2 Background on QSARs 21
2.3.5.3 Conclusions, Recommendations, and Rationale 22
2.3.6 Use of an Adaptive Management Approach to Implementation 23
CHAPTERS: CCL CLASSIFICATION APPROACH FOR MICROBIALCONTAMINANTS 3-1
3.1 Identifying the Microbial CCL Universe 3
3.1.1 NRC Recommendations for the Microbial CCL Universe. 3
3.1.2 Defining the Microbial CCL Universe 4
3.1.2.1 Human Pathogens as the Basis for the Microbial CCL Universe. 6
3.1.2.2 EnsuringIndusiveness oj'theMicrobial CCL Universe. 7
3.2 Microbial CCL Universe to PCCL 7
3.2.1 NRC Recommendations for the PCCL 7
3.2.2 Screening Microbes for the PCCL 8
3.2.3 Screening Based Upon Biological Properties 9
3.2.4 Pathogens Associated with Opportunistic Infections 10
3.2.5 Alternative Pathways for Adding Pathogens to the Microbial CCL Universe and the PCCL
(Surveillance and Nomination) 10
3.3 Use of Attributes to Classify Microbial Contaminants 12
3.3.1 NRC Recommendations for Classifying Microbial Contaminants to the CCL 12
3.3.2 Use of Attributes for Characterizing and Ranking PCCL Microbes 12
3.3.3 Developing Draft Protocols to Quantify Attributes 13
3.4 Applications of Genomics to the CCL Classification Process 17
3.4.1 NRC Recommendation on Genomics 17
3.4.2 Potential Applications of Genomics 18
3.4.3 Challenges to Use of Genomics 18
3.4.4 Pilot Projects 19
3.4.5 Recommendations for the Use of Genomics in the CCL Process. 20
CHAPTER 4: CCL CLASSIFICATION APPROACH FOR CHEMICAL CONTAMINANTS 4-1
4.1 Building the Chemical CCL Universe 1
4.1.1 Summary of NRC Recommendations 1
4.1.2 Overall Recommendations for Identifying the Chemical CCL Universe 2
4.1.3 Specific Work Group Recommendations 3
-------
CCL CP Work Group Report
4.1.3.1 DataSourceCompilationApproach 3
4.1.3.2 Supplemental Surveillance and Nomination Processes 6
4.1.3.3 An Integrated Process for Addressing Known, New,andEmergin%Agents. 6
4.1.3.4 Chemical CCL Universe Identification Process for Retrieving Information and Data. 8
4.1.3.5 An Approach to Retrieving Data and Evaluating Data Sources 11
4.1.3.6 Data Quality Principles Compatible wifftJnclusionary Principles., 12
4.2 Process and Criteria for Screening Agents from the Chemical CCL Universe
tothePCCL 12
4.2.1 Summary of the NRC Recommendations 13
4.2.2 Principles for Selecting Agents for a PCCL from the Chemical CCL Universe 14
4.2.3 Workable Approach to Screening Using Widely Available Data Elements 15
4.2.3.1 Data Element* for Potency 16
4.2.3.2 Data Elements for Occurrence 18
4.2.4 Screening for Both Health Effects and Occurrence 20
4.2.5 Tagging Sources of Values for Data Elements and Implications 21
4.2.6 Approaches to Classifying Agents on the Chemical CCL Universe to the PCCL 22
4.2.6.1 Assigning Specific Values to Data Elements Used in the Screening process 23
4.2.6.2 Basis for Establishing the Screening Criteria/Decision Rules 23
4.3 Use of Attributes to Classify Chemical Contaminants 26
4.3.1 Introduction 26
4.3.2 Use of Data Elements to Quantify Chemical Attributes 27
4.3.3 NDWAC Work Group Recommendations 27
CHAPTERS: MOVING FROM THE PCCL ONTO THE CCL , 5-1
5.1 Quantifying Attributes for Use as Inputs to Classification Modids 2
5.1.1 The Alternatives: Using Actual Data Values versus Attribute Scoring 2
5.1.1.1 Summary ofNRC Recommendations on Quantifying Attributes. 4
5.1.1.2 NDWAC WorkGroup Evaluation of Attributes. 4
5.1.2 NDWAC Work Group Recommendations 5
5.2 Overview of Classification Approaches. 7
5.3 Recommended Approach to Selecting the CCL ....8
5.3.1 NRC Recommendations 8
5.3.2 NDWAC Work Group Recommendations 8
5.4 Training Data Set 11
5.4.1 NRC Recommendations on Training Data Set 11
5.4.2 NDWAC Work Group Recommendations 12
TABLE OF FIGURES, EXHIBITS, AND TABLES
Figure ES.l Overview of CCL Process Recommended by the NDWAC Work Group ES-5
Figure 1.1 -NRC Proposed "Two-Step" CCL Process 1-2
Figure 1.2- Overview of the Regulatory Process 1-7
Figure 2.1 -Overview of NDWAC Work Group Recommended CCL Process 2-6
Exhibit 2.1 - Health Effects and Occurrence Attributes 2-9
Exhibit 2.2 - EPA Activities Relevant to Hie Surveillance Process 2-13
-------
CCL CP Work Group Report
Figure 2.2- Diagram Schematic of an Adaptive Management Process 2-24
Figure 3.1- AMicrobial CCL Classification Process 3-2
Table 3.1 - Categories and Examples of the NRC-Proposed Microbia] CCL Universe 3-4
Fig. 3.2 - Microbial CCL Universe 3-4
Figure 3.3- Screening Contaminants from the Microbial CCL Universe to the PCCL 3-8
Table 3.2 - Proposed Screening Principles to Exclude Pathogens from the PCCL 3-9
Figure 3.4 - Alternative Pathways for Introducing Pathogens to the CCL Classification Process 3-11
Figure 4.1-Detailed Overview of Step 1 (Chemical CCL Universe) 4-2
Table 4.1- Advantages and Disadvantages of the Data Source Compilation Approach 4-4
Table 4.2 - Advantages and Disadvantages of the Reducing Data Sources Approach 4-5
Table 4.3 -Examples of Occurrence and Health Effects Data Sources 4-10
Figure 4.2-Selecting the PCCL from the Chemical CCL Universe 4-13
Figure 4.3 - NRC's Diagram of the CCL Universe 4-14
Table 4.4 - Possible Data Elements for Selecting Universe Contaminants for the PCCL 4-20
Figure 4.4-Examples of Alternative Forms of Screening Criteria /Decision Rules 4-25
Figure 5.1 Classifying and Selecting the CCL 5-1
Figure 5.2 - "Separated" Contaminants Poorly Define the Discriminant 5-13
Figiue 5.3 - A Discrirninant Function on the Basis of Two Attributes 5-14
Glossary Q-1
References R-1
Appendices
Appendix A - Summary of Recommendations from Classifying Drinking Water Contaminants
for Regulatory Consideration A-l
Appendix B - Summary of the NDWAC CCL Work Group Investigation of QSAR Models as
Sources of Data / Information for the CCL Development Process B-l
Appendix C - Draft Scoring Protocols Developed and Used for Trial Attribute Scoring Exercise
(Workshop) C-l
Appendix D - Proposed Attribute Scoring System for Microbes D-l
Appendix E - Prototype Classification Methods/Results of Initial Pilot Evaluation E-1
-------
CCL CP Work Group Report
Executive Summary
ES.1 Introduction
The Contaminant Candidate List (CCL) Classification Process Work < jroup (the Work Group)
was charged by die National Drinking Water Advisory Committee (NDWAC) with reviewing the
National Research Council (NRC) 2001 report, Classifying Drinking Water Contaminants for
Regulatory Consideration. The Work Group was asked to advise the NDWAC on development and
application of the classification approach suggested by the NRC, including evaluating proposed and
alternative methodologies, hi conducting its review, the Work Group considered the large and
growing number of agents that might become candidates for scrutiny in the CCL process, and the
rapid expansion of information on these agents. Based on this review, the Work Group drew the
following conclusions.
• There is merit in the premise of a three-step selection process as proposxt by NRC for both
chemical and microbial contaminants:
- Identify th e CCL Universe
- Screen the Universe to a Preliminary CCL (PCCL)
- Select the CCL from the PCCL
" The Agency should move forward with the NRC recommendation to develop and evaluate some
form of prototype classification approach.
• Expert judgment plays an important role throughout the three-step selection process, particularly
in reviewing the prototype model and the new classification approach
" Enhancement of surveillance and nomination processes are essential to assure a full consideration
of emerging chemical and microbial contaminants.
The Work Group also identified a number of practical limitations or difficulties in developing and
applying the recommended approach, and sought to advise the NDWAC on liow these might be
addressed.
1.1 Background and Purpose
The Safe Drinking Water Act Amendments of 1996 require the US Envil-onmental Protection
Agency (EPA, or the Agency) to publish every five years a list of chemical arid microbial
contaminants that are known or anticipated to occur in public water systems and thatmay have
adverse health effects, and that, at the time of publication, are not subject to ajiy proposed or
promulgated National Primary Drinking Water Standards. The first CCL was published in 1998 and
was categorized based on four priority areas in drinkng water research (occurrence, health effects,
treatment, and analytical methods). On a staggered, second five-year cycle (thiree-and-a-half years
after a CCL is required), EPA is required to evaluate this research together wi(h any already available
Executive Summary ES-1
-------
CCL CP Work Group Report
information and make a determination for at least five contaminants on whether or not to proceed with
the regulatory development process.
The first CCL was developed based upon a review by technical experts of readily available
information, and contained 50 chemical and 10 microbial contaminants/groups. EPA completed its
first regulatory determination process in July 2003. EPA recognized the need for a more robust and
transparent process for identifying and narrowing potential contaminants for future CCLs and
requested advice from the NRC on developing such a process, hi its 2001 report, the NRC proposed a
broader, more comprehensive screening process to assist the EPA drinking water program to identify
those contaminants for which further research - and ultimately a decision on whether or not to
proceed with a regulatory development process - would be appropriate.
1.2 Charge to the NDWAC Work Group
The NDWAC's Charge to the CCL Work Group is set forth below.
"Evaluate recommendations made by the National Research Council, including
methodologies, activities and analysis, and make recommendations for an expanded
approach to the CCL listing process for the purpose of protecting public health.
"This may include, but not be limited to, advice on developing and identifying:
i Overall implementation strategy
ii. Classification attributes and criteria (and methodology that ought to be used)
iii. Pilot projects to validate new classification approaches (including neural network
and other prototype classification approaches)
iv. Demonstration studies that explore the feasibility of the WAR approach
v. Risk communication issues
vi. Additional issues not addressed in the NRC Report"
1.3 Guiding Principles
The Work Group adopted the following principles to guide its work:
" As public health is the first and foremost consideration, development and maintenance of the
CCL should maximize protection of public health, including consideration of sensitive
subpopulations.
• The CCL process should be built on the best available science, consistent with the goal of
protection of public health and development of the CCL in a reasonable time frame.
• All aspects of the CCL process should reflect the important role of expert judgment in both
establishing procedures and reviewing the results of those procedires.
• All aspects of the CCL process should be systematic, open, accessible and available to
informed stakeholders, and well-documented so that a knowledgeable reader could
understand and reproduce the process of analysis leading to specific decisions made for the
PCCLandCCL
Executive Summary
ES-2
-------
CCL CP Work Group Report
• All aspects of the CCL process should apply equal rigor to chemical and microbial agents,
consistent with the data available for these two categories.
• There should be opportunities for public involvement at all key pnints in the CCLprocess,
with broad participation by affected parties.
ES.2 Overview of Recommended CCL Classification Process and Overarching
Issues
2.1 Building on the NRC Approach
In reviewing the NRC selection process, the Work Group focused on tie following:
• More completely addressing the scope of the CCL "Universe" as described by NRC - with
respect to both chemicals and microbes
" Identifying a robust and practical means of screening the Universe to a Preliminary CCL
(PCCL)
• Evaluating the application of a prototype classification algorithm tp select a CCL from the
PCCL
• Ensuring that both chemical and microbial contaminants are adequately and equally
considered by the CCL process
« More fully developing the role of expert judgment acknowledged by the NRC but not
devebped in its report
• Reviewing the NRC's call for transparency throughout the CCL process
• Expanding on the NRC model to explicitly allow for nomination of potential contaminants
for consideration
• Expanding on the NRC model by explicitly encouraging the Agencj' to maintain the CCL
process as an ongoing programmatic element, rather man as a protocol that is repeated every
five years (This expansion includes the concept of surveillance for duta to support the CCL
process.)
• Suggesting data and information "hierarchies" that might be used in the process
• Following through on the NRC's recommendation to incorporate consideration of data
quality into the CCL process
• Developing a framework For incorporating genomics and proteomics, including the NRC's
Virulence Factor Activity Relationship (WAR) concept, into the CCL process
2.2 Transparency and Public Participation
The CCL process will need to be explained so that the public can generally understand the
method used to develop the CCL. Ksy criteria, data, and assumptions that affect inclusion or exclusion
of contaminants ought to be noted, where possible, so that the reader can follow the logic regarding
why decisions are made. Decision-makers, stakeholders, and drinking water consumers need to be
able to understand why EPA has selected the CCL contaminants and why further research on these
Executive Summary ES-3
-------
CCL CP Work Group Report
contaminants is a good use of resources. The public will want to know why investment in the methods
used to select contaminants and investment in research on certain contaminants is an efficient and
effective use of resources that will lead to improved protection of public health. If EPA is transparent
in its decision-making, the public will have the rationale needed to understand how the method works
and why specific contaminants are or are not on the list.
The CCL Work Group agrees with die NRC that the EPA will need to garner public support to
implement the CCL method effectively and efficiently. The Work Group recommends that EPA
consider early and ongoing consultation with key stakeholders and outreach to the public as
implementation proceeds. Finally, the Work Group agrees with the NRC that the public involvement
program needs to be tailored to the public's needs and should start early in the process.
2.3 Overview of the CCL Process
Figure ES.l diagrams the CCL Classification Process recommended by the NDWAC Work
Group. It is a three-step process. The first step consists of the parallel development of the Microbial
CCL Universe and the Chemical CCL Universe (which together constitute the "Universe" of agents
identified as Step 1 in the diagram below). The second step consists of screening contaminants from
the Universe of identified agents to the Preliminary Contaminant Candidate List, or PCCL. The third
step is the classification of contaminants on the PCCL to produce the proposed CCL.
Selection of microbial and chemical contaminants through a single CCL process is mentioned in
the NRC recommendations. The Work Group found that, at this point in time, there are still systematic
differences in the strengths and weaknesses of the information available for chemical and microbial
contaminants.
-> The Work Group recommends that the procedure for screening and selecting CCL
contaminants consist of parallel processes for microbial and chemical contaminants that
meet in the formation of a single CCL, but that take best advantage of the information
available for each type of contaminant
For the third step, classifying contaminants on the PCCL to select contaminants for the CCL, the
NRC proposed five attributes, or characteristics of a contaminant that contribute to the likelihood that
it could occur in drinking water at levels and frequencies that pose a public health risk.
-> EPA should proceed initially with using the two health effects attributes and three
occurrence attributes described by the NRC as input for the PCCL-to-CCL classification
modeling for contaminants.
Executive Summary
ES-4
-------
CCL CP Work Group Report
Figure ES.1 Overview of CCL Process Recommended by the MDWAC Work Group1
Identifying the CCL
Universe
STEP1
Evaluation
Expert Review5
I Proposed CCL I
STEP 3
Notes:
1. Steps are sequential, as a£ components of each step, with the exception of surveillance and nomination. This
generalized process is applicable to both chemical and microbial contaminants, though the specific execution of
particular steps may differ in practice.
2. Surveillance and nomnation provide an alternative pathway for entry into the CCL, process for new and emerging
agents, in particular. Most agents would be nominated to the CCL Universe. Depending on the timing of the
nomination and the information available, a contaminant could move onto the PCC L or CCL, if justified.
3. Expertjudgment, possibly including external expert consultation, will be important throughout the process, but
particularly at key points, such as: reviewing the screening criteria and process fronl. the Universeto the PCCL;
assessing the training data set and classification algorithm performance during development of the PC CL to C CL
classification step.
4. After implementing the classification process, the prioritized list of contaminants would be evaluated by expets,
including a review of the quality of information.
5. The CCL classification process and draft CCL list would undergo a critical Expert Review by EPA and by outside
experts before the CCL is proposed.
Executive Summary
ES-5
-------
CCL CP Work Group Report
2.4 Overarching Issues
The Work Group identified several overarching issues that must be considered in developing the
CCL process. In addition to the need for transparency and public participation, these overarching
issues include:
• The integration of expert judgment throughout the CCL process
• Active surveillance and nomination/evaluation processes for new and emerging agents
• Information quality considerations
» The application of an adaptive management approach to implementing the CCL process
The approach to address these overarching issues is intended to be consistent with the Work
Group's guiding principles.
Integrating Expert Judgment into the Process. NRC recommendations include provisions for
"expert" and "scientific review" in the CCL process but provide little guidance as to what, how, and
when such review would be used. Like the NRC panel, the Work Group observed that expert
judgment is inherent throughout the development of the CCL process and in implementing that
process once it is developed. Critical reviews, involving various types of expert consultation and
collaboration, up to and including more formal expert reviews, will be useful at key points in the new,
evolving CCL process outlined in Figure ES-1.
-> There are several key milestones in the CCL process where a critical review would be
especially relevant:
• In Step 2, to review the screening criteria and their application to screen agents from the
CCL Universe to the PCCL;
• In Step 3, during development of the classification process from the PCCL to the CCL, to
assess the training data set(s), assess the performance of the classification algorithm(s)
tested, and to determine whether that performance is sufficient to justify immediate use of
the algorithm(s) or suggests the need for further development;
• After the classification process is implemented, to evaluate the prioritized list of
contaminants, including a review of the quality of information, to provide judgments on
the proposed draft listing;
• The CCL classification process and draft list should undergo a formal expert review,
including extema I experts, before the CCL list is proposed.
Surveillance and Nomination/Evaluation Processes. The Work Group believes that a
surveillance process will prove to be an important and necessary component to ensure timely
identification of information relevant to new and emerging agents.
-> The Work Group recommends that EPA establish an active surveillance process to provide
identification of new and emerging agents for the CCL.
Executive Summary
ES-6
-------
CCL CP Work Group Report
-> The Work Group recommends that EPA develop a nomination and evaluation process fir
new and emerging agent;, to enable agencies and interested stakeholders from the public
and private sectors to nominate agents for consideration in the C^L process.
It is envisioned that surveillance and nomination would be integral components of the CCL
process, providing an alternative pathway for entry into that process rather ithan a separate process.
Typically, agents identified by the surveillance process would be nominated for placement in die CCL
Universe, not on the CCL. However, depending on the timing of the identification of the new and
emerging agents (in relationship to CCL publication schedule), and the nature of the information
about diem, contaminants could move onto the PCCL, or even onto the proposed CCL through an
expert review process, or (if justified) through an accelerated Agency decision-making process.
Information Quality Considerations. It would be expected that for mamy of the agents initially
selected for consideration in the CCL process, the available data would consist of various types with
different characteristics and robustness. The Work Group also recognized that the data or information
used to select the CCL will be more detailed and comprehensive than the data used to identify the
CCL Universe. Additionally, the CCL process will apply more scrutiny to contaminants when
selecting the CCL than when screening the Universe of agents to identify- cojntaminants for the PCCL.
To address the variability of the disparate types of data, it is essential that thq nature of the data used to
support these steps be documented for review in the later steps of the CCL process. This
characterization should identify the data sources, the methods used to derive'the data, what quality
assurance procedures were in place during data gathering, processing, or analysis, and whether the
data characterize "demonstrated" or "potential" occurrence or health effects. In selecting the CCL, the
nature and type of information should be considered further and in a manner that is consistent with the
development of the prototype classification algorithm. The Work Group emphasized that it is
important for EPA to develop and document appropriate data quality approaches as part of the
adaptive management approach to implementation discussed below. EPA should establish data quality
approaches applicable throughout Steps 1 through 3 prior to identifying the CCL Universe.
Adaptive Management Approach. The Work Group proposes an adaptive management
approach. Adaptive management principles could be applied in the developmlait, implementation, and
refinement of die diree-step CCL method, particularly in me initial phases of implementation. This
process incorporates systematic and continual integration of design, management, and monitoring,
which would enable EPA to make nformed adjustments and adaptations. Thin process incorporates
systematic and continual integration of design, management, and monitoring, which would enable
EPA to make informed adjustments and adaptations, resulting in an improved metitod based on
experience from the outcomes of successive generations of implementing die Universe-to-CCL
approach.
These overarching issues are discussed in more detail in Chapter 2,
ES.3 CCL Classification Approach for Microbial Contaminants
The Work Group evaluated the differences in chemical and biological characteristics of
demonstrated and potential water contaminants. The conclusions suggest mat identifying die
Microbial CCL Universe and screening the set of biological agents to a PCCL «:an be consistent widi
the NDWAC's proposed principles for chemicals but will require different data sources and data
Executive Summary ES-7
-------
CCL CP Work Group Report
elements, and may require more involvement from experts than the approach described for chemical
agents and contaminants. The Work Group's recommendations for a classification approach to
microbial contaminants are summarized as follows.
-> The NDWAC Work Group recommends that the Microbial CCL Universe be based on the
evaluation of data sources and literature reviews that identify organisms known or
suspected to cause human disease.
-> The Work Group recommends that the selection of human pathogens for the PCCL start
with a Microbial CCL Universe of recognized human pathogens (e.g., the amended Taylor
et al. 2001 list), and that those pathogens known to be associated with source water,
recreational water, and drinking water be selected for inclusion into the PCCL.
-> The Work Group supports the following concepts for EPA's consideration as they develop
future CCLs:
• Biological characteristics should be recognized as legitimate criteria for screening
pathogens for the PCCL
• The list of pathogens inhabiting the Microbial CCL Universe should be screened for
biological characteristics promoting or mitigating against survival and transmission in
water.
-> The Work Group recommends that organisms associated with opportunistic infections be
excluded from the PCCL unless clinical, epidemiological, or similar information implicates
them as the cause of waterborne disease. The Work Group suggests that EPA increase
surveillance for infections caused by these organisms, especially in sensitive subpopulations.
-> EPA should review public health surveillance techniques, in conjunction with the Center
for Disease Control (CDC), with a view to making those techniques as proactive, robust,
and effective x possible in identifying the occurrence of waterborne or watershed disease
outbreaks and the organisms associated with those outbreaks.
The Work Group also evaluated the use and potential of VFARs for the CCL process. Genomics
and proteomics are recognized as powerful tools for the elucidation of pathogenic mechanisms but the
technology is yet largely unproven for CCL application.
-> The Work Group recommends that EPA should monitor the data and information that
emerge as genomics progresses and integrate them for consideration in the CCL process.
The process should be updated and maintained in a continuing process and verified against
expert opinion. The Work Group recommends that EPA monitor the progress of genomics
and the related technologies and integrate them into the CCL process, as feasible.
t
ES.4 CCL Classification Approach for Chemical Contaminants
The recommended process contains three distinct steps: (1) building the Universe of chemical
agents; (2) screening contaminants from the Chemical CCL Universe to the PCCL; and, (3) moving
contaminants from the PCCL to the CCL.
Executive Summary
ES-8
-------
CCL CP Work Group Report
4.1 Building the Chemical Universe
After review of NRC's recommendations, available data sources, and'consideration of the
potential scope of the Universe of known chemical agents, the Work Groub recommends EPA adopt a
principles-based approach, con sistent with mat described by the NRC.
-> EPA should use the incluisionary principles as the foundation for identifying the Chemical
CCL Universe. These principles are as follows:
• The Chemical CCL Universe should include those agents that have demonstrated or
potential occurrence in drinking water; or
« The Chemical CCL Universe should include those agents that have demonstrated or
potential adverse health effects.
The Work Group recommends a strategy of accessing discrete databases to retrieve various,
unique sets of records with multiple selection criteria, a process known as a "data source compilation
approach." The Work Group further recommends supplementing this iterative information retieval
process with surveillance and nomination processes mat provide altemative'pathways into the CCL
classification process.
4.2 Screening Contaminants from the Universe to the PCCL
The Work Group proposes EPA develop a screening process that relies on widely available data
elements that reflect certain aspects of both health effects and occurrence.
-> The Work Group recommends that the screening criteria and methods be:
• capable of assessing as many of the contaminants in the CCL Universe as possible, even
those with limited data;
• as insensitive as possible to data limitations;
• as simple as possible, to require fewer resources and less time;
« capable of identifying those contaminants of greatest significance for further
consideration; and,
« to the extent feasible in Ught of the significant differences in avaifabuity of data for
chemicals and microbes, as similar as possible to the microbial approach.
-> The Work Group recommends that a limited set of data elements th,at are widely available
and that represent important characteristics of health effects and occurrence be used as the
basis of the screening to select contaminants from the Chemical CCL Universe.
Chapter 4 (section 4.2.3) details the Work Group's analysis and recommendations for a
"workable approach" to screening the Chemical CCL Universe using widely Bailable data elements
for health effects (where die data element with the most "health-protective" value would ultimately be
used in the screening process) and for occurrence (where the data ebment with die greatest occurrence
frequencies and highest contamination levels would ultimately be used in the Screening process).
Executive Summary ES-9 ^.
-------
t
CCL CP Work Group Report
-> The Work Group recommends that the contaminants that are screened to the PCCL be
those for which values for data elements for both health effects and occurrence reach a level
of concern, based on the screening process, for inclusion on the PCCL. Generally, neither
alone would be sufficient under this screening process.
At the same time, the Work Group recognizes that there as likely to be contaminants that ana
highly toxic but have low potential for exposure or that have high potential for exposure but do not
appear to be highly toxic. The Work Group recommends that EPA use a supplemental assessment to
identify such agents that should be further investigated and perhaps should be included on the PCCL.
-> The Work Group recommends that EPA allow expert judgment to be used to correct
mistakes or oversights that will arise from this relatively simple process. It will likely be
appropriate to add some number of contaminants to the PCCL that pose a concern but
that do not fit the process outlined. The Work Group recognizes that unforeseen
circumstances will arise, and recommends that EPA allow for supplemental consideration
to address them.
The Work Group considered a number of other issues specific to the classification of chemical
contaminants.
-> The Work Group recommends that "tags" be used to retain information about the sources
of values used in the screening process and that this be done in such a way as to preserve
this information for later steps hi the process. The tags should identify values derived from
models such as QSARs. The tags should also identify what combination of "demonstrated"
and "potential" values for health effects and occurrence were used.
-> The Work Group recommends that, as the Agency develops approaches to screen chemical
agents from the Universe to the PCCL, it should consider a range of options both for using
data element values in the screening process and for establishing appropriate screening
criteria to select PCCL contaminants. The screening method developed should be practical
and transparent, and should efficiently screen the Universe to the PCCL. The method
should also employ a level of precision that appropriately characterizes the nature and type
of information used. While the Work Group discussed several options and identified their
advantages and disadvantages, it did not recommend a single approach.
ES.5 Moving from the PCCL onto the CCL
The Work Group discussed structured decision approaches to select the CCL from the
PCCL. Some of the structured decision approaches discussed, particularly classification
algorithms, require as inputs some specific measures of the attributes that characterize a
contaminant's known or potential health risks. These specific measures could be either the actual
values reported in the scientific literature (such as water concentration measurements or
Reference Dose values), or generated values or "scores" based on the actual values reported in
the literature to characterize the attributes. The Work Group discussed several methods to
quantify attributes. Each of these quantification methods presents a set of benefits and challenges
particular to the method. The Work Group did not develop specific recommendations for
quantifying attributes or a preferred structured decision approach. The Work Group does provide
Executive Summary
ES-10
-------
CCL CP Work Group Report
a series of general recommendations for the Agency to use as a framework to develop and
evaluate the attributes and classification process to select contaminant^ from the PCCL.
5.1 Quantifying Attributes for Use as Inputs to Classification Models
The Work Group considered two basic approaches to quantifying attributes.
1) Using the actual quantitative value cr measurement provided by the data element to quantify
the attribute.
2) Scoring attributes, using a set of rules to convert the data element values to either:
• a normalized numerical score with continuous values allowed within the given scoring range; or
• a limited set of categorical scores within a given range.
The NDWAC Work Group did not reach a conclusion regarding which! approach to quantifying
attributes is preferred, and therefore does not make a specific recommendation favoring one over the
other. Some of the recommendations that follow refer to aspects of attribute 'Scoring and are therefore
relevant where EPA determines that attribute scoring is the preferred approach.
-> Attribute scoring protocols for contaminants should accommodate multiple data sources
and a variety of data elements that may be available to score contariiinants on the PCCL.
-> Attribute scoring across different types of data elements for a given, attribute should be
consistent and allow for a meaningful comparison among scored PCCL contaminants.
-> EPA should systematically refine and improve upon the details of th^e attributes as more
experience is gained, including refinements and improvements in gathering and processing
the needed data and information to score the attributes and with respect to using the
attribute scores in the selected classification approach. Further refinements may include
reducing the number of attributes; other refinements may pertain to the data elements
used to score the attributes, the scoring protocols, and the actual attribute scoring process
itself.
-> EPA should generate and include, along with the actual values or the attribute scores that
are generated, descriptive "tags" that provide additional data quality information that may
be used by experts reviewing the data, attribute scores and/or the PGCL-to-CCL
classification modeling results.
If attribute scoring is used, the scoring system selected by EPA for eai* attribute should
enable discrimination among contaminants, and there should be sufficient number of
scoring categories so that information loss during characterization of 'contaminants is
limited. At the same time, the scoring categories should not be so numerous that they
convey a false sense of precision.
Executive Summary ES-11
-------
CCL CP Work Group Report
-> If attribute scoring is used, the scoring protocols should be transparent and
straightforward.
5.2 Overview of Classification Approaches and Work Group Recommendations
-> The Work Group recommends that EPA pursue development of a prototype classification
algorithm (aposteriori approach) for sefecting contaminants for the next CCL. The Work
Group recommends moving forward to develop and test one or more prototype models as
tools to be used with expert judgment for decisions on classifying contaminants for future
CCLs.
The Work Group did not have time to evaluate the alternatives and recommend a particular
prototype model. It may be useful to have several models that are used in concert to corroborate
results. Also, it may be necessary to develop separate models for chemical and microbial
contaminants, or models that differentiate chemicals and microbes within the model structure. The
development of any model should be an adaptive process, and should be reviewed by experts, with
consideration given to updating the training data set, with each successive CCL cycle.
-> The Work Group recommends that the entire model development process be as
transparent as possible. The development process should be viewed as iterative, and EPA
should involve experts and allow opportunities for meaningful public comment on the
evaluation.
-> EPA should use another approach for selecting CCL contaminants in the near term (i.e.,
for CCL3) if there are difficulties in the model development process that cannot be
overcome.
-> The Work Group recommen ds that experts should be involved throughout the process of
narrowing a PCCL to a CCL, specifically as advisors in the design of an approach,
development of a training set, scoring of contaminant attributes, evaluation of algorithm
results, and ultimate selection of CCL contaminants.
5.3 Development of a Training Data Set
There are several issues to consider in the selection of a "training data set" used to inform the
decision-making tool, or algorithm. With respect to training data sets, the Work Group makes the
following recommendations.
-> The training data set should consist of contaminants (and corresponding decisions to "list"
or "not list" each contaminant) that reflect technically sound, consistent judgments about
what should and should not be included on the CCL.
-> The training set should include contaminant attribute data that are distributed throughout
the attribute space, and the training set should be selected to define the discriminant
surface (the function that defines "include" and "exclude" decisions) as precisely as
possible.
Executive Summary
ES-12
-------
CCI CP Work Group Report
-> The Work Group recom mends that EPA maintain transparency and clarity when
developing the training data set To the extent feasible, EPA shoujld document training data
set development and communicate its rationale for assigning decisions to training set
contaminants.
-> The rationale for the number and distribution of training set contaminants should be
described. Quantitative rationale should be expressed for the prototype classification
approach.
These ate important considerations for determining if the training set aid models have been
adequately developed to begin processing PCCL contaminants. The rationale should include a
description of the methods used for calibration and validation, and measuieu used to assess goodness
of fit, such as mischssification rates.
ES.6 Summary
EPA should proceed with the development of prototype classification diethods. The NDWAC
Work Group identified several overarching principles that EPA should use in developing a CCL.
These include die use of experts at key steps to allow for technical checks oij the process, and
nomination and surveillance processes that provide an alternative pathway for contaminants to enter
the CCL classification process when new information surfaces.
To classify chemicals, die Work Group recommends a three-step process that includes defining
and building the Universe of chemical agents, screening from the Chemical CCL Universe to create a
PCCL, and developing a CCL from the PCCL. For microbes, a somewhat different but parallel
process is recommended, involving identifying a Universe from a list of kno\lm human pathogens,
reducing this list to a PCCL based on habitat and biological properties indicative of a pathogen's
ability to be transmitted via water, and developing a CCL from the PCCL.
In making its recommendations, the Work Group identified a number of practical limitations or
difficulties in developing a classification approach. These limitations are outlined in the report, and
will require additional work to resolve, however die NDWAC Work Group's iassessment concludes
that there is merit in the NRC-proposed process for classifying microbes and chemicals, and that the
Agency should move forward in pursuing the approach outlined in this report.
Executive Summary ES-13
-------
CCL CP Work Group Report
Chapter 1
Introduction
1.1 Background on the Contaminant Candidate List and the National Research
Council Recommendations
The Safe Drinking Water Act Amendments of 1996 require the US Environmental Protection
Agency (EPA) to publish every five years a list of chemical and microbial contaminants that are
known or anticipated to occur in public water systems and that may have adverse health effects, and
that, at the time of publication, are not subject to any proposed or promulgated National Primary
Drinking Water Standards. The first Contaminant Candidate List (CCL) was published in 1998 and
was categorized based on four priority areas in drinking water research (occurrence, health effects,
treatment, and analytical methods). On a staggered, second five-year cycle (three-and-a-half years
after a CCL is required), EPA is required to evaluate (his research together with any already available
information and make a determination for at least five contaminants on whether or not to proceed with
the regulatory development process.
The first CCL was developed based on the review by technical experts of readily available
information and contained 50 chemical and 10 microbial contaminants/groups. EPA recognized the
need for a more robust and transparent process for identifying and narrowing potential contaminants
for future CCLs and requested advice from the National Academy of Sciences National Research
Council (NRC) on developing such a process. In its 2001 report, Classifying Drinking Water
Contaminants for Regulatory Consideration, the NRC proposed a broader, more comprehensive
screening process to assist the EPA drinking water program to identify (hose contaminants for which
further research - and ultimately a decision on whether or not to proceed with a regulatory
development process - would be appropriate.
1.1.1 The NRC Recommendations
The NRC's major recommendations are summarized in the following excerpts from the
Executive Summary (pages 4-6) of the 2001 report.
"The committee continues to recommend that EPA develop and use a two-step
process for creating fature CCLs as illustrated in Figure ES-I [reproduced below
as Figure 1.1],"
Ch. 1 - Introduction
1-1
-------
CCL CP Work Group Report
Figure 1.1 - NRC Proposed "Two-Step" CCL Process1
The "universe" of potential
drinking water contaminants includes:
1. Naturally occurring substances
2. Water-associated microbial agents
3. Chemical agents
4. Products of environmental
transformation of chemical agents
S. Reaction by-products
6. Metabolites in the environment
7. Radionuclides
8. Biological toxins
9. Fibers
STEP ONE
Screening criteria +
expert judgment
PCCL
A PCCL includes:
1. Contaminants that are demonstrated
to occur in drinking water and
demonstrated, to cause adverse health
effects
2. Contaminants that are demonstrated
to occur in drinking water and have the
potential to eel use adverse health
effects
3. Contaminar ts that are demonstrated
to cause advefse health effects and
have the potential to occur in drinking
water
4. Contaminants that have the potential
to occur in drijiking water and the
potential to cause adverse health
.effects
STEP TWO
Classification tool + expert judgment
CCL
"The committee also continues to recommend that this two-stefi process be
repeated for each CCL development cycle to account for new data and potential
contaminants that inevitably arise over time...
The committee recommends that the process for selecting contaminants for future
CCL(s) be systematic, scientifically sound, and transparent. The development
and implementation of this process should involve sufficiently broad public
participation."
The NRC recommended that a broadly defined universe of potential drinking water contaminants
be identified, assessed and culled, to a preliminary CCL (PCCL) using simple screening criteria and
expert judgment. All the contaminants on the PCCL would then be assessed in more detail using a
prototype classification tool, in conjunction with expert judgment, to evaluate the likelihood that they
could occur in drinking water at levels and at frequencies that pose a public health risk and move onto
1 The two steps referred to in the NRC report are 1) screening the universe of potential contaminants to generate a preliminary CCL, or PCCL; and, 2)
refining UK PCCL to [reduce a CCL However, because the NDWAC Work Group elaborated further en the NRC's concept of a "Universe" of potential
contaminants and on how to identify its scope and uontoits, this report generally refers to (he NRC approach as a "dmo-step" process (except where quoting
directly from the NRC report)
Ch. I - Introduction
1-2
-------
CCL CP Work Group Report
the CCL, NRC recommendations associated with specific steps in the CCL process are discussed in
subsequent chapters of this report, along with the NDWAC Work Group's deliberations and
recommendations. A detailed listing of the NRC recommendations is provided in Appendix A.
The NDWAC Work Group deliberated these major recommendations and some of the issues
relevant to ensuring the use of best available science and assisting in the transparency and
communication of the CCL - issues that should be considered in the evaluations to move a
contaminant from a broad Universe of potential drinking water contaminants and onto the CCL.
1.1.2 Charge to NDWAC Work Group
With the NRC recommendations ii hand, the Office of Ground Water and Drinking Water turned
to the National Drinking Water Advisory Committee (NDWAC) to provide advice on different
aspects of the staged approach recommended in the NRC report and to work out how this could be
implemented The NDWAC formed a Work Group on the Contaminant Candidate List Classification
Process (Work Group) to evaluate die NRC recommendation and report back to the full NDWAC,
The Charge to the CCL Work Group is set forth below.
"[To] Evaluate recommendations made by the National Research Council,
including methodologies, activities and analysis, and making recommendations for an
expanded approach to the CCL listing process for the purpose of protecting public
health.
"This may include, but not be limited to, advice on developing and identifying:
i. Overall implementation strategy
ii. Classification attributes and criteria (and methodology that ought to be used)
iii. Pilot projects to validate new classification approaches (including neural network and
other prototype classification approaches)
iv. Demonstration studies that explore the feasibility of the WAR2 approach
v. Risk communication issues
vi. Additional issues not addressed in the NRC Report"
1.2 Convening and Membership of the NDWAC CCL Classification Work Group
On June 19,2002, the Federal Register published a notice announcing the formation of the
NDWAC CCL Classification Process Work Group and requesting nominations to the group. During
the convening process, several areas of expertise were identified as important for the Work Group,
including computer modeling, epidemiology, contaminant occurrence, statistics, toxicology,
chemistry, microbiology, risk analysis, risk communication, water system operation, and public
health. The convening process sought to identify candidates with expertise in these areas as well as
individuals to represent the views of several stakeholder groups, including the water industry,
environmentalists, the public health community, rural water systems, and local elected officials. From
1 Viralence-factor activity relationships
Ch. 1 - Introduction
1-3
-------
CCL CP Work Group Report
among the candidates identified, EPA and the chair of the National Drinking Water Advisoiy Council
selected individuals to serve as members of the Work Group. Part way through the process two
members resigned from the group because of changes in their work obligatrons. The final membership
of the Work Group was as follows:
Dr. Laura Anderko, Univercity of Wisconsin, Milwaukee
Dr. Richard Becker, American Chemistry Council
Dr. Douglas Crawford-Brown, University of North Carolina Chapel Hill
Dr. Michael Dourson, Toxicology Excellence for Risk Assessment
Dr. Alan Elzerman, Clemson University
Dr. Jeff Griffiths, Tufts University
Dr. Wendy Heiger-Bernays, Boston University School of Public Healtli
Mr. Buck Henderson, Texas Commission on Environmental Quality, Association of State
Drinking Water Administrators
Dr. Nancy Kim, New York State Department of Health
Mr. Ephraim King, U.S. Environmental Protection Agency
Ms. Carol Kocheisen, National League of Cities
Mr. Gary Lynch, Park Water Company
Mr. Ken Merry, Tacoma Water of Tacoma Public Utility
Mr. Brian Ramaley, Newport News Waterworks
Dr. Graciela Ramirez-Toro, Centra de Education, Conservation y Interpretation Ambiental,
Lnteramerican University, Puerto Rico
Dr. Craig A. Stow, University of South Carolina
Dr. O. Colin Stine, University of Maryland, Baltimore
Mr. Ed Thomas, National Rural Water Association
Ms. Lynn Thorp, Clean Water Action
Dr. Daniel Wartenberg, University of Medicine & Dentistry of New Jersey- Robert Wood
Johnson Medical School
The Work Group was supported by a team of technical consultants and EPA staff The technical
consultants included Amy D. Kyle, PhD MPH, Consulting scientist, health aiid environment; Doug
Owen, Malcolm Pirnie, Inc.; Jeff Rosen, Perot Systems Environmental Services; Paul Rochelle,
Metropolitan Water District of Southern California; and Steve Via, American Water Works
Association (AWWA). George Hallberg, JoAnne Shatkin, Frank Letkiewicz,'Nelson Moyer, and
other staff from the Cadmus Group, Inc. also served on the technical team as tontractors to EPA.
Facilitation was provided by Abby Arnold, Sara Lttke, and other staff from RIESOLVE, and the
document was edited by Susan Savitt Schwartz.
Ch. I - Introduction 1-4
t
-------
CCL CP Work Group Report
1.3 NDWAC CCL Classification Process Work Group Guiding Principles
Early in their deliberations, the Work Group adopted the following principles to guide their
process:
• Public health is the first and foremost consideration. Development and maintenance of the
CCL should, to the extent possible, maximize protection of public health. Full consideration
should be given to sensitive subpopulations.
• The CCL process should be built on a foundation of science, and explicitly state and explain
the rationale for adoption of assumptions and estimates when these are used in lieu of actual
data.
• All aspects of the CCL process should be systematic and scientifically sound and should
maximize transparency, -while acknowledging that expert judgment also will be necessary;
when expert judgment is used, it should be clearly identified.
• Ultimately, the CCL process should be described and documented to such an extent that a
knowledgeable reader could understand the rationale for why a contaminant would be on or
off the "Universe," PCCL, or CCL. Ultimately, it should be clear which decisions are based
on expert judgment, science, policy considerations, or other considerations.
• The CCL process should apply equal rigor to chemical and microbial contaminants from a
public health perspective.
« The CCL decision-making process must be open, accessible, and available to all informed
stakeholders, including the interested general public as well as the professional and
scientific community and all directly affected parties.
• Consistent with the authority under which the CCL Work Group was formed, the group
encourages the opportunity for public involvement throughout the entire process. Broad
participation that is representative of the range of affected and interested parties is to be
encouraged, thereby incorporating public values, viewpoints, and principles into the
process.
• As much as possible, the goals and objectives of the CCL process should guide information
and data collection, EPA should clearly communicate these goals so that the desired types of
data can be identified and developed for future CCLs by sources other than solely EPA. In
addition, EPA should articulate the types of data and data elements preferred for developing
the CCL and how those data may affect the selection of contaminants for the CCL.
1.4 Summary of the NDWAC CCL Work Group Deliberation Process
The Work Group met in plenary ten times in Washington, DC: September 18-19, 2002;
December 16-17, 2002; February 5-6,2003; March 27-28,2003; May 12-13,2003; July 16-17,2003;
September 17,2003; November 13-14,2003; January 22-23,2004; and March 4-5,2004. At the first
meeting the Work Group heard an overview of the NRC recommendations from NRC committee
members who had helped to develop the recommendations. Based on this overview the Work Group
began to identify issues to address and formed several activity groups to focus on specific aspects of
the NRC proposed CCL process. The Work Group agreed to follow a fairly detailed work plan
designed by Work Group members and the technical team. The work plan proposed to address various
Ch. I - Introduction 1-5
-------
CCL CP Work Group Report
issues in parallel. Detailed meeting summaries for each meeting are available on the EPA website
[http://vyww.cpa-fiov/safcwater/ndwacsum.htmll.
All Work Group meetings were open to the public and announced in the Federal Register. At
each plenary meeting as well as by conference call and in activity groups, the Work Group reviewed
the components of the proposed NRC approach and examined their feasibility through various
analyses. For each component of the proposed approach, the remaining chapters of this report
summarize the questions considered by the Work Group; the analyses conducted to explore the
questions; key points discussed:, and die Work Group's recommendations and rationale for the
recommendations.
The Work Group was not able to address all 23 recommendations included in the NRC report,
(see Appendix A of mis report), but rather focused on specific aspects of the NRC recommendations.
Also, within the Work Group's deliberations, some topics received extensiye discussion and analysis
while others were less extensively debated. This uneven consideration of the NRC recommendations
should not be construed either as an endorsement or as a refutation of the NRC recommendations that
are not specifically addressed. Likewise, very general recommendations tend to reflect one of the
following situations:
1) Resolution of some topics will require time and resources on a scale required for the
Agency's actual implementation of the CCL process (e.g., development of training data sets).
The Work Group's schedule did not allow this level of involvement.
2) Specific guidance was inappropriate, as the Agency's actions will in reality need to reflect the
success of intermediate actions toward reaching recommended objectives.
3) Given the information available to the Work Group and the schedule, detailed
recommendations were not developed for all the NRC recommendations.
1.5 Role of the CCL in Protecting Public Health and Implications of Inclusion on
the PCCL or CCL
To understand the process proposed in these recommendations, it is useful to consider the role of
the CCL in the protection of public health and what it means for a contaminant to be placed onto or
left off the PCCL or the CCL. The Safe Drinking Water Act (SDWA), as amended in 1996, requires
the EPA to publish a list of contaminants that are known or anticipated to ociair in public water
systems, and which may require regulation under the SDWA [section 1412(W1)]. The SDWA, as
amended, also specifies that EPA must publish this list of contaminants (Drinking Water Contaminant
Candidate List, or CCL) not later than 18 months after the date of enactment! (i.e., by February 1998),
and publish a new CCL every five years thereafter. The SDWA requires that the list of contaminants
include those which, at the time of publication, are not subject to any proposed or promulgated
national primary drinking water regulation (NPDWR). The list must be published after consultation
with the scientific community, including the Science Advisory Board, after Notice and opportunity for
public comment, and after consideration of the occurrence database established under section 1445(g).
The unregulated contaminants considered for the list must include, but not to limited to, substances
Ch. 1 - Introduction 1-6
-------
CCL CP Work Group Report
referred to in section 101(14) of the Comprehensive Environmental Response, Compensation, and
Liability Act of 1980 (CERCLA), and substances registered under the Federal Insecticide, Fungicide,
and Rodenticide Act (FIFRA),
Contaminants on the CCL are evaluated to determine what additional data are needed and to
identify the next steps for each contaminant. Contaminants requiring additional data on their
occurrence, health effects, treatment, or analytical methods become research priorities to develop
sufficient data to support a regulatory determination. When sufficient data are available, regulatory
determinations evaluate the extent of exposure and potential public health protection to populations
via drinking water and whether or not initiation of a regulatory process is appropriate. The Agency
may determine that an appropriate action is development of health advisories, NPDW regulations, or
no action. The precepts for guiding EPA in making regulatory determinations for a drinking water
contaminant are included in Section 1412(b)(lXA) of SDWA. This section of SDWA requires EPA to
consider the following three evaluation criteria prior to making a regulatory decision:
1) potential adverse health effects from the contaminant;
2) occurrence of the contaminant in public water supplies with a frequency and at levels of
public health concern; and
3) whether regulation of the contaminant would present a meaningful opportunity for health risk
reduction for persons served by public water supplies.
Figure 1.2 illustrates (he regulatory
process in its entirety and provides a schematic
of the CCL process for identifying potential
drinking water contaminants in relation to the
development of NPDWRs. Below the dashed
line are the components deliberated by the
Work Group. Further detailed consideration of
health effects and opportunities of actual
public health risk reduction are assessed
through the subsequent steps of the regulatory
development process and were not part of the
charge to the Work Group (i.e., regulatory
determinations, development of MCLGs, and
development of MCLs). The Work Group
focused on the immediate objective of the
CCL process.
Figure 1.2 - Overview of the Regulatory Process
1
3
O)
8.
CO
Detailed Analysis and
Additional Research
NPDWR Final Rule Developed
4
NPDWR Public Comment
4
NPDWR Proposal Development
4
Regulatory Determinations
NewCCL
Research
Priorities
Jj
ll
II
t
CCL
t
Preliminary CCL
t
Universe of Contaminants
Ch. I - Introduction
1-7
-------
CCL CP Work Group Report
We note finally that limit sd societal resources mean that a finite number of contaminants can
move through the CCL process, and that uncertainties generally may be l^rge in risk estimates prior to
the detailed process of regulatory development. The CCL should identify ^ set of priority
contaminants that pose risk to public health. Simply increasing the size of the CCL may not contribute
to improving public health. Listing a contaminant on the CCL means that the assembled data indicate
it has properties generally indicative of significant risk and/or suggestive of the need for future
research aimed at clarifying those suggestive properties. It also means thai, relative to other
contaminants on the PCCL, a contaminant on the CCL has evidence that is more indicative or
suggestive of risk, and therefore offers a greater possibility of improving public health through
allocation of resources to better understand and/or control that risk. The CCL should identify those
contaminants whose existing evidence indicates either: 1) that a subsequent risk calculation might
produce risks warranting regulatory attention; or, 2) that the contaminart h,as a high measure of either
occurrence or effects, and that subsequent research on the other componen: might be expected to yield
the information needed for an estimate of risk.
The PCCL aids in narrowing the pool of candidates for the CCL. Placing a coitaminant on the
PCCL means that there are aspects of existing data, not necessarily conclusively validated, that
suggest significant risk and warrant resources to clarify this suggestion. It ailso means that, relative to
other agents in the Universe, a contaminant on the PCCL has evidence thaj is more indicative or
suggestive of risk, and is more likely to retain these characteristics after available data and information
are assembled, and employed as part of the process leading to the CCL. Placing a contaminant on the
PCCL does not mean it is established to pose a significant risk or has characteristics that are fully
indicative of significant risk that would warrant concern or justify further research. The PCCL is
simply an intermediate resting place for contaminants that will be scrutinized in more detail.
Ch. {-Introduction 1-8
-------
CCL CP Work Group Report
Chapter 2
Overview of Process and Overarching Issues
This chapter presents a brief overview of the CCL Classification Process recommended in this
report. It emphasizes the importance of making that process transparent to the public, and highlights
the ways in which the NDWAC Work Group's recommendation builds upon that of the NRC. The
chapter also presents a discussion of overarching issues - issues that affect or apply to more than one
aspect of the CCL process.
• Transparency and public participation
• Integration of expert judgment into the
CCLprocess
• Active surveillance and nomination/
evaluation processes for new and emerging
agents
• Information quality considerations
" The use of quantitative structure activity
relationship (QSAR) models
" The application of an adaptive
management approach to implementing the
CCLprocess
The approach to address these overarching issues is
intended to be consistent with the Work Group's
guiding principles (discussed in Chapter 1).
2.1 Transparency and Public Participation
Chapter 2 of the NRC report makes clear that,
to achieve acceptance, the CCL classification
process adopted by the EPA "needs to be based on
sound science, risk perception, social equity, legal
mandates to consider the risks ofvulnerable
populations, and the proper role of transparency
and public perception."
Definitions
• agent: any physical, chemical, or biological
substance. *
• known agents: physical, chemical, or biological
substances that have been identified in the technical
literature and adequately characterized to enable a
judgment regarding their inclusion in the CCL
Universe. *
* emergingagents: a subset ofhtown physical,
chemical, or biological substances previously
evaluated as not requiring inclusion in the CCL
Universe, forwhich new information becomes
available which heightens concern and triggers re-
evaluation. *
• new agents: physical, chemical, or biological
substances that are or may be newly-discovered or
synthesized,for which little is known about their
potential occurrence or adverse healtti effects. *
• contaminant: contaminant is defined similar to agents,
as any physical, chemical, or biological substance in
water. For this report the Work Group used
contaminant to indicate any agent for which data exist
that suggests that the agent belongs on the PCCL.
• attributes: characteristics of a contaminant or
potential contaminant tfiat contribute to the likelihood
that a particular contaminant or related group of
contaminants could occur in drinking water at levels
and frequencies that pose a public health risk.
* See farther discussion in Chapter 4.1.3
t
2.1.1 Why Transparency is Important for the CCL
Like the NRC, the CCL Work Group believes that the credibility of the CCL methodology EPA
adopts depends on sound science. The method will need to withstand peer review or scientific
critique, in which scientists can take the same information and test conditions and achieve comparable
Ch. 2- Overview of Process and Overarching Issues
2-1
-------
CCL CP Work Group Report
results. Acceptance also will depend on how the method is developed and how transparent - i.e., how
clear - it is to the pubtic. The explanation of the CCL process will need to be expressed so that the
public can generally understand the method used. This does not necessarily mean the process will be
simple or easy to understand.
Decision-makers, stakeholders, and drinking water consumers need to'be able to understand why
EPA has selected the CCL contaminants and why further research on these contaminants is a good use
of resources. The public will want to know why investment in the methods'used to select
contaminants and investment in research on certain contaminants is an efficient and effective use of
resources mat will lead to improved protection of public health. If EPA is ti-ansparent b its decision-
making, the public will have the rationale needed to understand how the method works and why
specific contaminants are or are not on the list.
As recommended by the NRC, the CCL Work Group discussed the importance of noting
uncertainties in data or information used in the process, as well as uncertaii ties in the proposed CCL.
If EPA is clear about these uncertainties, it will provide decision makers anld the public with the tools
needed to determine whether they believe EPA has made appropriate contaminant determinations,
based upon protection of public health, good science, and occurrence in drinking watec
If successful, the CCL classification approach recommended in this report will generate a list of
contaminants that enables EPA to concentrate research and other activities on those contaminants that
occur or potentially occur in drinking water, and that pose the most concert] for public health. This
will result in research being targeted as wisely and effectively as possible tc support public health
protection while addressing a concern of stakeholders and ratepayers that limited resources be spent in
a cost-effective manner. By investing in this kind of process up-front, the contaminants of significant
concern will be singled out for further study in an open and transparent mariner. EPA should use
caution when developing this up-front process to assure that resources are wisely invested when
implementing the recommendations in this report. This will help EPA alloqate limited funds to the
contaminants that pose the greatest public health risk with input from staketiolders. The resulting
effort will assist in supporting a credible and open process so the public knqws the rationale for why
research is being recommended and can support appropriate listing decisions.
The CCL Work Group endorses the following steps proposed by the NRC to encourage
transparency of whatever method EPA adopts (pp. 64-66 of NRC report):
• One of EPA's major goals in developingjuture CCLs should be to explain the process
sufficiently so that the reader can understand the rationale behindincluding particular
contaminants on the CCL. To achieve this goal would require thai transparency be
incorporated into the method used in the decision-making process in addition to being an
integral component in communicating the details of the decision-making process to the
public (NRC 2001, p. 61).
• The use of a classification tool needs explanation or rationale.
• The method for designing and calibrating the decision-making process must be explained
If decision-making for including or excluding a certain contaminant on future CCLs ultimately
depends on a combination of the results of a classification tool and EPA judgment, then this
relationship must be fully articulated along with the background assumptions and underlyhg Agency
Ch. 2 - Overview of Process and Overarching Issues 2-2
t
-------
CCL CP Work Group Report
judgments. Key criteria, data, and assumptions that affect inclusion or exclusion in potentially
controversial cases ought to be noted, where possible, so that the reader can follow the logic regarding
why decisions were made.
2.1.2 Public Participation
As quoted by the NRC,"'public participation encompasses a group ofprocedures designed to
consult, involve, and inform the public to allow those affected by a decision to have an input into that
decision' (Rowe andFrewer, 2000) " (p. 66). The NRC also points out that "a central tenet of public
participation is that the public is, in principle, capable of making wise and prudent decisions" (p 66).
The CCL Work Group agrees with the NRC that EPA will need to gamer public support to
implement the CCL mediod effectively and efficiently. Without this, it will be difficult to obtain buy-
in from various stakeholder groups. The CCL Work Group's principles on public participation are as
follows:
• The CCL decision-making process must be open, accessible, and available to all
stakeholders who are interested
• The CCL Work Group encourages EPA to provide the opportunity for public involvement at
key steps along the way. Broad participation that is representative of the range of affected
and interested parties should be apriority, thereby considering pub lie health values,
viewpoints, and principles.
The NRC recommended an approach that would lead to scientifically sound policy decisions,
informed by technical expertise, that are responsive to stakeholder values and concerns. This Work
Group process is a first step in this direction. Because the prototype classification process for
developing the CCL is a new approach, the Work Group recommends that EPA develop an outreach
program to educate and inform stakeholders about its use. This approach may be a challenge for some
to understand. Therefore, in the future, the Work Group recommends that EPA consider early and
ongoing consultation with key stakeholders and outreach to the public as implementation proceeds.
Finally, the Work Group agrees with the NRC that the public involvement program needs to be
tailored to the public's needs and should start early in the process.
2.2 Overview of Recommended CCL Classification Process
In providing an overview of the CCL process, this section of the chapter describes the Work
Group's recommendations for:
1) building on the NRC's concept of a three-step3 CCL classification process;
3 In its report, the NRC refers to its recommended approach as a "two-step" process, where a Univene of potential contaminants is assumed to exist,
the first step is screening that Universe to generate a Preliminary CCL (PCCL) and the second step is refining the PCCL to produce a CCL. However,
because the NDWAC Work Group elaborated further on the NRC's "Universe" and how to identify its contents, this report generally refers to the NRC
approach as a "three-step" process (unless directly quoting from the NRC report)
Ch. 2 - Overview of Process and Overarching Issues
2-3
-------
CCL CP Work Group Report
2) developing parallel processes for building the microbial and chemical CCL Universes and for
classifying the agents that comprise those Universes-first to the Preliminary CCL (PCCL),
and then to the CCL;
3) approaching the develo pment of a prototype classification approach; and for
4) incorporating genomic information into the CCL classification process.
2.2.1 Building on the NRC Approach
Having accepted the premise of a three-step selection process as proposed by NRC, the Work
Group focused on achieving objectives inherent in the NRC approach. The NRC approach presents a
number of logistical and practical hurdles for the EPA. For example, the NRC recommended that the
Agency describe the Universe of potential contaminants very broadly. As a result, the information
management system and decision criteria employed in the early stages of tlie CCL process must
process tens of thousands of agents- often on the basis of very limited data! or information. The Work
Group was cognizant of such practical implementation issues and recommended modifications to the
NRC approach to address them. Specifically, the Work Group focused on fie following objectives.
1) More completely addressing the scope of the CCL "Universe" ^is described by NRC -
with respect to both chemicals and microbes
2) Identifying a robust and practical means of screening the Univeirse to a Preliminary CCL
(PCCL)
3) Evaluating the application of a prototype classification algorithm to select a CCL from the
PCCL
4) Ensuring that both chemical and microbial contaminants are adequately and equally
considered by the CCL process
5) More fully developing the role of expert judgment acknowledged by the NRC but not
developed in its report
6) Reviewing the NRC's call for transparency throughout the CCL process
7) Expanding on the NRC model to explicitly allow for nomination of potential contaminants
for consideration
8) Expanding on the NRC model by explicitly encouraging me Agjsncy to maintain the CCL
process as an ongoing programmatic element, rather than as a protocol that is repeated
every five years. (This expansion includes the concept of surveillance for data to support
the CCL process.)
9) Suggesting data and information "hierarchies" that might be used in the process
Ch. 2- Overview of Process and Overarching Issues
2-4
-------
CCL CP Work Group Report
10) Following through on the NRC 's recommendation to incorporate consideration of data
quality into the CCL process
11) Developing a framework for incorporating genomics and proteomics, including the
NEC's Virulence Factor Activity Relationship (WAR) concept, into the CCL process
Key areas where the Work Group expanded on the work of the NRC, going beyond the NRC's
report and recommendations, include: the rote of surveillance and nomination processes, the role of
expert input into the CCL process, the issue of data quality considerations, and the concept of an
adaptive management approach to implementing the CCL classification process. The Work Group's
contributions in each of these areas, along with a discussion of other overarching issues, are presented
in Section 2.3. First, however, it will be helpful to step through the recommended CCL process
illustrated in Figure 2.1.
2.2.2 Parallel Processes for Chemical and Microbial Contaminants
Selection of microbial and chemical contaminants through a single CCL process is mentioned in
the NRC recommendations. The Work Group found that, at this point in time, there are still systematic
differences in the strengths and weaknesses of the information available for chemical and microbial
contaminants.
-> The Work Group recommends that the procedure for screening and selecting the CCL
contaminants consist of two parallel processes which meet in the formation of a single CCL,
but which take best advantage of the information available for each type of contaminant.
The general framework of assessment, drawing on information concerning health effects and
occurrence, and allowing for consideration of the qualify of information availabb, should be used for
both classes of contaminants. The specific recommendations for the microbial and chemical
classification processes are discussed in Chapters 3 and 4, respectively.
Ch. 2 - Overview of Process and Overarching Issues 2-5
-------
CCL CP Work Group Report
Figure 2.1 - Overview of NDWAC Work Group Recommended CCL Process1
Identifying the CCL
Universe
STEP1
STEPS
Notes:
1. Steps are sequential, as are components of each step, with the exception of surveillance and nomination. This
generalized process is applicable to both chemical and microbial contaminants, though the specific execution of
particular steps may differ in practice.
2. Surveillance and nomination provide an alternative pathway for entry into the CCL process for new and emerging
agents, in particular. Most no nunations would be for agents to the CCL Universe.! Depending on the timing of the
nomination and the infoimatian available, a contaminant could move onto the PCCL or CCL, if justified.
3. Expert judgment, possibly including external expert consultation, will be important throughout the process, but
particularly at key points, such as: reviewingthe screening criteria and process from the Universe to the PCCL;
assessing the training data set and classification algorithm performance during development of the PCCL to CCL
classification step.
4. After implementing the classification process, the prioritized list of contaminants would be evaluated by experts,
including a review of the quality of information.
5. The CCL classification process and draft CCL list would undergo a critical Expert Review by EPA and by outside
experts before the CCL is proposed.
Ch. 2 - Overview of Process and Overarching Issues
2-6
-------
CCL CP Work Group Report
2.2.2.1 Identifying the CCL Universe
Identification of the CCL Universe should follow the principles described by NRC to be inclusive
of agents with demonstrated or potential occurrence in drinking water and/or demonstrated or
potential adverse health effects. The NDWAC Work Group recommends that the EPA consider
adopting a three-stage process to identify the Chemical CCL Universe.
• Identify and retrieve data (including lists of agents) from sources that have data and
information about occurrence of contaminants in drinking water or source water or about
health effects.
" Identify and retrieve data (including lists of agents) and information from sources that have a
link (pathway) to drinking water concerns.
• Use other information sources, such as chemical properties, and models (e.g. QSARs) and
surrogate information, to address data gaps.
The Work Group recommends that the Microbial CCL Universe be based on the evaluation of
data sources and literature reviews that identify pathogens, i.e., organisms known or suspected to
cause human disease. (Those pathogens from this CCL Universe that are known to be associated with
source water, recreational water, or drinking water would be selected for inclusion on the PCCL.)
Specific recommendations for microbial and chemical contaminants are described in detail in
Chapters 3 and 4, respectively.
A tag indicating the type of data source employed to identify and describe a microbial or
chemical agent should be tracked along with information about that agent as it is further processed in
(he CCL process, as described in section 2.3.4.
2.2.2.2 Screening from the Universe to the PCCL
The Work Group reviewed the NRC's conceptual approach to screening a Universe of agents to a
Preliminary Contaminant Candidate List (PCCL). Work Group members agreed with the NRC
recommendation that this screened list should receive a higher degree of scrutiny before contaminants
are moved to the CCL. To address this intermediate step, the Work Group assessed the likely
availability of data about occurrence and adverse health effects of contaminants that may be present in
drinking water and concluded that, as noted by the NRC, these data may be very limited. The Work
Group considered the advisability of giving priority to contaminants for which more data are available
against the interest hi taking an inclusive approach. The Work Group also noted that the screening
approach should err on the side of allowing contaminants to move forward at this step in the
classification process rather than omitting potential drinking water contaminants from further
consideration. The Work Group concluded that it would be important for EPA to develop an approach
and screening criteria that are:
• capable of assessing as many of the agents in the Universe as possible, even those with
limited data;
• ay insensitive as possible to data limitations and that treat contaminants with different
amounts of data available as similarly as possible;
• as simple as possible, to require fewer resources and less time;
Ch. 2 - Overview of Process and Overarching Issues
2-7
-------
CCL CP Work Group Report
• capable of identifying those contaminants of greatest significance for further consideration;
and,
" to the extent feasible in light of the significant differences in availability of data for
chemicals and microbes, as similar in approach as possible.
To develop an approach that would be as simple as possible and allow' for the assessment of as
many agents from the Universe as possible, the Work Group discussed how to identify the most
essential characteristics of agents of concern. The Work Group sought to limit these characteristics to
those mat are most necessary arid informative. These characteristics would become the basis for
conducting this initial stage of sheening, from the Universe to the PCCL
The Work Group decided, based on the information available at this tune, that a more limited set
of the data elements describing health effects and occurrence characteristics would be more effective
when defining criteria to select contaminants from the Universe for the PCGL. The attributes used to
characterize microbial and chemical contaminants and the data elements usfcd to describe or measure
those attributes are discussed in Section 2.2.2.3, below. Specific recommendations for screening
contaminants from the Microbial and Chemical CCL Universes are described in Chapters 5 and 6,
respectively.
2.2.2.3 Characterizing the PCCL Contaminants
In considering the NRC recommendations, the Work Group agreed thdt structured decision-
making models could be used by EPA, in conjunction with expert judgment, to determine which
chemical and microbiological contaminants are most appropriately moved forward from the PCCL to
the CCL based on their known or potential health risks. These various models require as inputs some
specific measures related to those risks. These specific measures would be Quantified attributes
developed from either the actual values reported in the scientific literature (tntch as water
concentration measurements or Reference Dose values) or the generated va| ues or "scores" based
upon the actual values reported in the literature to characterize attributes. Attributes- or more
specifically, either the actual values or scores generated from the actual valjies for data elements used
to characterize those attributes- can serve as the inputs for these models. Attributes are defined in the
context of the CCL classification process as characteristics of a contaminanj: that contribute to the
likelihood that it could occur in drinking water at levels and frequencies thai: pose a public health risk.
The various types of measures or descriptors that may be used as a means fijir quantifying the
attributes are referred to as data elements. The expression "quantifying the Attributes" used in this
report refers to either the use of actual values reported in the scientific literature or to the generation of
values (or "scores") based upon these actual values to characterize the attributes numerically so mat
they can be used in the classification modeling steps. These quantified values for the attributes can
then be applied as inputs to the structured-decision making tools described iji Chapter 5 to prioritize
the contaminants for moving them from the PCCL to the CCL. In using data elements to quantify
attributes, the emphasis at mis step in the process should be to have those contaminants posing the
greatest public health risk proceed to the CCL.
The NRC indicated that it had spent considerable time deliberating on ihe number and type of
attributes mat should be used in tiie CCL development approach the NRC envisioned. From those
deliberations, Ihe NRC developed a set of five specific attributes - two addressing health effects, three
addressing occurrence - that they beieved constituted a reasonable starting point for EPA to consider.
Ch. 2 - Overview of Process and Overarching Issues 2-8
-------
CCL CP Work Group Report
The NRC envisioned that these five attributes would be applicable to both chemical and microbial
contaminants, but recognized that the types of measures and information used to quantify the
attributes would differ for these two categories of contaminants. The specific attributes identified by
the NRC were:
" Potency and Severity as key predictive attributes for health effects
• Prevalence and Magnitude as key predictive attributes for occurrence
" And Persistence/Mobility, as characteristics that might predict possible occurrence if direct
measures ofprevalence and magnitude were not available
For chemical contaminants, the Work Group agreed that EPA should start with the two health
effects attributes (potency and severity) and the three occurrence attributes (magnitude, prevalence,
and persistence/mobility) generally described by the NRC as input for the PCCL-to-CCL
classification modeling. (See text box for general definitions for each of these attributes, as provided
by the NRC.)
Exhibit 2.1 - Health Effects and Occurrence Attributes
Potency indicates the amount of a contaminant required to cause an adverse health effect; a
relative scaling of a dose-response relationship.
Severity describes the clinical significance of the most sensitive health end-point; a measure
of "How bad is the effect?"
Magnitude reflects the concentration or expected concentration of a contaminant relative to a
level that causes a perceived health effect.
Prevalence describes how commonly the contaminant does or would occur in drinking water.
Persistence/Mobility reflects the likelihood that the contaminant would be found in the
aquatic environment based solely on physical properties of the contaminant.
For microbial contaminants, the same set of attributes are used, but different kinds and
combinations of data elements are required to quantify those attributes, as discussed in Chapter 3.
There are numerous details concerning how many attributes are needed and how they should be
characterized and quantified that must be developed in conjunction with the development of the
specific classification approaches) to be used in the process of moving contaminants from the PCCL
to the CCL. This is in keeping with the NRC observation that the five attributes discussed in its report
were meant to be illustrative and represented a reasonable starting point for EPA's consideration.
2.2.3 Developing a Prototype Classification Approach
The NRC prototype classification approach is a challenging one, both because of the number and
difficulty of preparatory steps required and because of the inter-related complexities of the attribute
and scoring process. The Work Group was unable, given available time and resources, to actually
develop and test a training data set based on the attribute scoring protocols developed by EPA in
Ch. 2 - Overview of Process and Overarching Issues 2-9
-------
CCL CP Work Group Report
support of the Work Group activities. Consequently, the Woik Group did not have the opportunity to
pilot the NRC recommendation regarding the prototype classification approach.
Despite these limitations, the Work Group did feel that it could offer EPA practical advice on
how to proceed in its evaluation of the prototype classification approach.
->' Specifically, EPA should proceed with the following steps.
« Evaluate a range of performance indicators for the classification approach.
• Proceed to construct (he necessary data systems to support a classification approach.
" Prepare training and validation data sets and test the performance of the algorithm
against the Agency's performance indicators for the algorithm.
• Employ expert processes to make adjustments in the event the classification approach
does not perform adequately.
• Once the classification approach is demonstrated, appfy the approach to obtain a draft
CCL for expert evaluation and refinement of both the product and the process.
2.2.4 Incorporating Genomic Information in the CCL Process
Genomics and proteomics are potentially powerful tools for elucidating the pathogenic
mechanisms of microorganisms, and thus for understanding individual and population exposure and
response to contaminants. At present, the use of these techniques for screeniing microbes or chemicals
in the CCL process is premature; however, the Woik Group found considerable merit in the NRC's
recommendation for long-term development of VFARs. (See Chapter 3.4 fd>r further discussion and
specific recommendations of the Work Group related to VFARs.)
For chemical contaminants, current research in toxicology involves gathering data on me
relationship of genomics and chemical mechanisms of action, a growing figld called toxicogenomics,
The field of toxicogenomics is rapidly developing information about gqne and protein activity in
response to chemical exposure. Biological responses following exposure to chemicals are studied at
several levels. Presently, lexicologists identify organ system and specific adverse effects by exposing
animal and cell models to specific chemicals. There are strain and species differences in these
responses. Likewise, in the hum an population, there are differences in responses. These differences
ate often related to the genetic m ake-up of the individuals. A great deal of effort is being focused in
mis area (e.g., the National Institute of Environmental Health Sciences' Natonal Center for
Toxicogenomics). As in microbiology studies, technologies such as DNA njicroarrays or high-
throughput nuclear magnetic resonance (NMR) and protein expression analysis are being used for the
assessment of die biological effects from chemicals. In ths future, researched will begin to understand
which genes are turned on (or off) in response to specific chemicals. These icsponses will provide
useful information on the degree to which populations are exposed and perhaps begin to identify those
populations who may be more susceptible to those exposures.
As these genomic and proteomic techniques are developed and refined, their use should be
considered for future CCL development for both microbiological and chemical contaminant
evaluation.
Ch. 2 - Overview of Process and Overarching Issues 2-10
-------
CCL CP Work Group Report
-> As noted in Chapter 3 of this report, the Work Group recommends that EPA should
monitor the progress of genomics and related technologies and integrate them into the CCL
process, as feasible.
2.3 Overarching Issues
The remainder of this chapter addresses issues thataffect many aspects of the CCL classification
process. These overarching issues include the following.
• Integrating expert judgment into the CCL process
• Implementing active surveillance for new and emerging agents
• Implementing nomination/evaluation processes for new and emerging agents
• Dealing with quality of information in the CCL process
• Use of quantitative structure activity relationships (QSARs)
• Use of an adaptive management approach to implementation
2.3.1 Integrating Expert Judgment into the Process
NRC recommendations include provisions for "expert" and "scientific review" in the CCL
process but provide little guidance as to what, how, and when such review would be used. Like the
NRC panel, the Work Group observed that expert judgment is inherent throughout the development of
the CCL process and in implementing that process once it is developed. Critical reviews, involving
various types of expert consultation and collaboration, up to and including more formal expert
reviews, will be useful at key points in the new, evolving CCL process, as outlined above in Section
2.2 and in Figure 2.1.
-> There are several key milestones in the CCL process where a critical review would be
especially relevant:
• In Step 2, to review the screening criteria and their application to screen agents from the
CCL Universe to the PCCL;
• In Step 3, during development of the classification process from the PCCL to the CCL, to
assess the training data set(s), assess the performance of the classification algorithm(s)
tested, and to determine whether that performance is sufficient to justify immediate use of
the algortihm(s) or suggests the need for further development;
• After the classification process is implemented, to evaluate the prioritized list of
contaminants, including a review of the quality of information, to provide judgments on
the proposed draft listing;
• The CCL classification process and draft list should undergo a formal expert review,
including external experts, before the CCL list is proposed.
Each of these reviews would benefit from a range of relevant expertise from both inside and
outside the Agency. There are however, significant time and resource constraints to consider. To best
utilize the Agency's limited resources, formal expert review is most critical in evaluating the
Ch. 2 - Overview of Process and Overarching Issues
2-11
-------
CCL CP Work Group Report
classification process and draft CCL. In the Work Group's opinion this formal expert review should
involve external experts. This review would consider the performance of u>e CCL classification
algorithm, considering not only what was listed, but also looking selectively at the PCCL to identify
inconsistencies or biases in the algorithm's performance, and the application of expert judgment to the
prioritized list.
In emphasizing mis final review in the CCL process, we do not intend.to diminish the importance
of expert or critical review in me earlier steps of the CCL process. Expert involvement can be
particularly valuable for the Agency as it develops and implements an entirely new approach for the
CCL classification process. Inclusion of expert review early in the CCL prccess offers assurance that
the final product, the proposed CCL, will be technically sound and scientifically defensible. This will
afford the Agency opportunities to spot problems early and make timely and efficient adjustments.
Another benefit may be increased credibility with the stakeholder communi,y, as expert review
provides technical checks on the process as it evolves, rather than solely relying on the comments
from stakeholders and interested parties during the proposal and final Fedenil Register publication at
the end of the CCL process.
As implied., critical review and expert involvement can take many form*, from reviews internal to
EPA, less formal technical consultation with experts external to the Agency ?md stakeholders, to
formal external review. In particular, the Work Group does not envision applications of the more
rigorous external peer-review type activity during the earlier stages of the CCL Classification Process.
It is also important to note mat the Work Group recognizes that these reviews should be integrated
into the overall CCL process so mat concurrent activities and overall progress toward proposal of the
CCL can occur in a timely fashion. Further, the Work Group noted that involvement of the same
experts in review of the various steps in the process may afford both logistical', and technical
advantages.
2.3.2 Implementation of an Active Surveillance Process for New and Emerging Agents
The Work Group recognized mat it can take considerable time for infonmtion to be generated
about contaminants before they appear in many data sources and can be captured in the mainstream
CCL process. In fact, it is likely that any such broad process will not, for examjple, be able to quickly
reflect outbreak investigations that may identify new and emerging contaminants. Hence, the Work
Group discussed the need for active surveillance and nomination processes to provide an alternative
pathway for entry into the CCL process. The Work Group believes that a surveillance process will
prove to be an important and necessary component to ensure timely identification of information
relevant to new and emerging contaminants. Such relevant information may include recent
epidemiological or lexicological studies, new information related to sensitive ^populations, or new
investigations of occurrence or exposures. The Work Group recognizes that the surveillance and
nomination processes are key areas where expert judgment would provide input to the CCL process.
-> The Work Group recommends that EPA establish an active surveillan ce process to provide
identification of new and emerging agents for the CCL.
This process of identification should be an integral part of EPA's CCL process. The burden of
identifying new and emerging problems should not be solely on the public. Whi]e the recommended
surveillance process has not been characterized in depth, the following aspects should be considered.
Ch. 2-Overview of Process and Over arching Issues 2-12
-------
CCL CP Work Group Report
• Implementation of a proactive process to survey or obtain information from institutions and
organizations that might be expected to observe or generate new information about
occurrence or health effects of potential agents or contaminants. These could include
federal, state and local health departments, environmental agencies, drinking water utilities,
and research institutions. This would include an ongoing process for communication with
these institutions.
» Identification of key published data sources (or criteria for their selection) based upon
consistency with the inclusionary principles (discussed in detail in Chapter 4), and updated
with adequate frequency to provide the most current information available on potential
agents.
• A means for identifying new information from recent updates of data sources to minimize
redundant searching.
• A review process that is technically sound and logistically practical.
• A means for documenting the process and any decisions reached (transparency).
Further discussion and specific recommendations related to surveillance and nominations of
microbial and chemical contaminants are presented in Chapters 3 and 4, respectively.
2.3.2.1 Surveillance Activities
The Work Group recognizes that EPA has considerable ongoing activity that potentially relates to
surveillance, ranging from ongoing literature reviews, to attendance at professional meetings,
sponsorship of special meetings and special sessions at meetings dealing with drinking water issues,
communications with researchers in the field, and liaisons with foreign institutions. EPA's Office of
Water maintains linkage among Offices within EPA (e.g., Office of Pesticide Programs) and with
other agencies and organizations that play a key role in the surveillance process (e.g., Centers for
Disease Control and Prevention, US Geological Survey). Many offices within EPA are being caled
upon to conduct surveillance activities, so coordination must be a key component. Examples of EPA
activities are noted below (see box) as types of activities that the Work Group recognizes as beneficial.
These activities may need to be expanded and their linkage to the CCL may need to be strengthened.
Exhibit 2.2- EPA Activities Relevant to the Surveillance Process
The Office of Ground Water and Drinking Water (OGWDW) is working with the Office of
Wetlands, Oceans, and Watersheds (OWOW) to strenghen linkage between ambient water
(i.e., source waters) contaminant concerns and criteria (Clean Water Act) and the drinking
water program.
The Office of Science and Technology (OST) ecological health (i.e., ambient/source waters)
team and its human health (i.e., drinking water) team collaborate with the Office of
Prevention, Pesticides, and Toxic Substances (OPPTS) to identify new and emerging
contaminants of concern.
OGWDW, OST, and Office of Pesticide Programs (in OPPTS) coordinate on cross-cutting
scientific issues related to pesticides (e.g., share data on health effects; coordinate activities
on risk analysis and occurrence studies).
Ch. 2 — Overview of Process and Overarching Issues
2-13
-------
CCL CP Work Group Report
OST maintains a relationship with State and Regional risk assessors through the Federal-
State Toxicology and Risk Analysis Committee (FSTRAC). It is qften through the
interactions with this group thai EPA becomes aware of local contamination problems and
emerging contaminants. FSTRA C meetings are managed by OST and are held twice per
year.
Office of Research and Development (ORD) staff conduct research and annual reviews of
drinking water issues, (For example: Richardson, S.D. 2003. wyter Analysis: Emerging
Contaminants and Current Issues. Analytical Chemistry. 75(12):2831-2857); Daughton, C.,
and Temes, T., Pharmaceuticals and Personal Care Products injhe Environment: Agents of
Subtle Change? Environmental Health Perspectives. Volume 107, Supplement 6, December
1999; Bimbaum L.S. mdD. F. Staskal 2003. Brominated Flamo Retardants: Cause for
Concern? doi: 10.l289/ehp.6559 (availableat htte://dx.doLors/ 17 October 2003).)
The EPA Drinking Water Hotline receives reports of contamination incidents that are
reviewed as part of the CCL process and that may not appear in other data sources.
The Centers for Disease Control and Prevention (CDC), Council of State and Territorial
Epidemiologists (CSTE), and EPA maintain a collaborative surveillance system for the
occurrence and causes ofwaterbome-disease outbreaks. "Surveillance for Waterborne-
Disease Outbreaks" is published biannually in the Morbidity and^Mortatity Weekly Report
(MMWR). This collaboration is clearly recognized as part of the process to identify possible
new or emerging waterbome microbial contaminants (see Chapter 5).
Various offices within EPA interact wiih other offices or programs within the National
Institute of Health, such as the National Toxicology Program (NT.P), National Cancer
Institute (NCI), and offices/centers of the National Institute of Environmental Health
Sciences (e.g., Center for the Evaluation of Risks to Human Reproduction).
EPA andATSDR assess the presence and nature of contaminants andhealth hazards at
Superfund sites and may conduct public health assessments atRCRA sites.
The USGS Water Resources programs have established formal liaison coordination with
OGWDW(andOPP and OWOW) for sharing information and program coordination for the
National Water Quality Assessment (NAWQA) program and the Toxics Substances
Hydrology Program, including the National Reconnaissance of Emerging Contaminants.
OGWDW/OWhas been working to strengthen interaction with and1 information review from
its foreign counterparts particularly in Canada, the EU (including WHO), Japan, and Latin
America.
OGWDW and other OW offices have official linkages or liaisons far information sharing
with various groups on ahe front lines of water quality issues such d s the Association of State
Drinking Water Administrators (ASDWA), Association of State and Interstate Water
Pollution Control Administrators (ASIWPCA), National Water Quality Monitoring Council
(NWQMC), Ground Water Protection Council (GWPC), the Association of State and
Territorial Health Officials (ASTHO), among others.
OGWDW and other OW staff participate in support and review of research on water
contaminant issues with the American Water Works Association Research Foundation
(AWWARF) and the Water Environment Research Foundation (WERF).
Ch. 2- Overview of Process and Overarching Issues
2-14
-------
CCL CP Work Group Report
EPA staff often directly participate in meetings with various groups, such as those mentioned
above, as well as American Water Works Association (e.g., Water Quality Technical
Conference), American Chemical Society (ACS), Society of Environmental Toxicology and
Chemistry (SETAC), Society for Risk Analysis (SRA), American Society for Microbiology
(ASM), American Water Resources Association (A WRA), National Ground Water
Association (NGWA), Society of Toxicology (SOT) and the American Public Health
Association (APHA).
2.3.2.2 Primary Source Literature Review
Another component of surveillance is review of the primary research literature to identify, and
provide information for new or emerging agents. Many "text" and bibliographic sources of
information were identified in the Work Group's efforts to identify databases/data sources for the
Universe. Work Group discussions recognized that bibliographic sources could not likely be part of
the more automated process of identifying the CCL Universe. As noted in reports to the Work Group,
EPA has begun to develop automated data extraction tools (software/programs) that would be able to
partly automate data collection from some important text sources (e.g., Developmental and
Reproductive Toxicology (DART), part of the National Library of Medicine). As noted in other
reviews, bibliographic sources may be used to fill in data gaps for contaminants identified in the CCL
process. Various search engines can be employed; past discussions have identified key sources such as
PubMed, TOXLINE, CCRIS (Chemical Carcinogenesis Research Information System), GENE-TOX,
DART/ET1C (Developmental and Reproductive Toxicology/Environmental Teratology Information
Center), and ISI Web of Science. These sources can readily be searched to locate possible information
to fill data gaps on identified agents; however, searching the literature for information on new and
emerging agents must be part of a surveillance process because of the largely manual effort that is
required for more detailed review to assess pertinent literature. For example, preliminary studies might
be identified in bibliographic sources (e.g., epidemiological studies) related to emerging issues or
agents of interest. Such studies will nearly always need to be evaluated using expert judgment to
assess if the results can be utilized based upon sound scientific information.
The most up-to-date, "emerging" information may well come from information presented at
professional conferences. Surveillance of meeting proceedings requires yet a different level of effort.
This might be most efficiently handled through enhanced relationships with professional societies and
organizations, as discussed below.
2.3.2.3 Additional Surveillance Activities and Recommendations
While there are many activities and mechanisms in place that can contribute to the surveillance
process, for the explicit needs of the CCL process these may need to be strengthened and new
activities may need to be initiated. In many of the cooperative efforts noted, EPA may need to
explicitly outline the needs for the CCL process to ensure that there is adequate consideration and
communication.
At least a few professional organizations have been forming committees and sponsoring forums
to focus on emerging water quality issues. The EPA CCL staff will need to ensure close links to such
groups, perhaps through joint sponsorship of regular meetings, workshops or conferences. In a similar
manner, EPA may need to help stimulate similar focus groups within other organizations. This can be
Ch. 2 - Overview of Process and Overarching Issues 2-15
-------
CCL CP Work Group Report
accomplished in part with formiil communications and requests to professional and interest
organizations. To facilitate appropriate interest, EPA might undertake other actions, as needed, such
as:
• EPA might designate a formal liaison with outside groups to coordinate efforts in emerging
drinking water quality issue:;, or at least specify an EPA point-of-contaflt for a specific
organization(s).
• EPA might set up workshop s, or more narrowly focused meetings with appropriate professional
organizations and stakeholders, on emerging or problematic contaminant groups such as
microbiological contaminants or personal care products and pharmaceuticals.
• EPA might strengthen communications and review of unique state leva programs that are
working to identify and even monitor for new and emerging contaminants, or conducting special
health studies related to wator-borne contaminants (e.g., California, lowja, New Jersey, and New
York, among others).
• EPA might woric with journals or publication groups to standardize key words to facilitate data
gathering (as noted in Chapter 3 for microbiological surveillance).
-> In particular, the Work Group recommends that EPA institute a regularly scheduled (e.g.,
biennial) conference on "Emerging Issues in Drinking Water" as part of their research for
the CCL process, where stakeholders and professional groups could present their findings
and concerns on emerging and new agents.
With the myriad groups that can be involved, the Work Group suggest; that this could be a
particularly efficient mechanism. This would also provide a particularly visible and transparent
component for gathering stakeholder input. EPA's sponsorship of such meetings and EPA's
presentation of research and data needs for CCL consideration would also apt to stimulate and
structure needed research both within EPA and by various interest groups for the future of the
program. All of these activities should serve to provide "nominations" of agents to add to the CCL
Universe.
2.3.3 Implementation of a Nomination and Evaluation Process for New and Emerging
Agents
It is envisioned that surveillance and nomination would be integral components of the CCL
process and not a separate process. As such, surveillance and nominations t/pically would provide an
alternative pathway for entry for an agent into the CCL evaluative process, in other words, agents
identified would typically be considered for placement in the CCL Universe, not on the CCL.
However, depending on the timing of the identification of the new and emerging agents (in
relationship to CCL publication schedule), and on die nature of the information about them,
nominated contaminants could move onto the PCCL - or even onto the proposed CCL through an
expert review process, or (if justified) through an accelerated Agency decisijjn-making process (as
further discussed below in section 2.3.3.2).
-> The Work Group recommends that EPA develop a nomination and evaluation process for
new and emerging agents, to enable agencies and interested stakeholders from the public
and private sectors to nominate agents for consideration in the CCL process.
Ch. 2- Overview of Process antl Overarching Issues
2-16
f
-------
t
CCL CP Work Group Report
As noted, all of the surveillance activities should serve to provide "nominations" of agents to add
to the CCL Universe. It is important to note mat nominations from the surveillance process can occur
both before and after the PCCL stage (see Figure 2.1). Where they occur after the classification
approach, nominations can still be considered as part of the expert review process prior to publishing a
proposed CCL. In addition, there is the opportunity as part of the formal process of public comment
and response on the proposed CCL that can include nominations. This existing process commences
when the Agency, in establishing a new CCL, issues a proposal in the Federal Register with request
for public comment. The request for public comment includes not only the opportunity to comment on
what the Agency has proposed (the CCL), but also the opportunity to nominate additional potential
contaminants to be considered for the CCL. In this process, EPA reviews comments and nominations,
and responds with its decision, as part of establishing the final fist.
However, the NDWAC Work Group recommends that additional opportunities to nominate
agents should be available during the CCL evaluative process in advance of, and distinct from, the
formal comment period on the Agency's proposed CCL. The NDWAC Work Group recommends
that throughout the CCL process (for example as part of an "Emerging Issues" conference as
discussed above), suggestions from stakeholders for agents to be considered would be provided to
EPA, and that, if appropriate, these nominees could be added to the CCL Universe. (As with all parts
of the CCL development, documentation of how and where the agent was identified would be part of
the process.) Additional components to be considered for the nomination process are outlined below.
2.3.3.1 Additional Considerations for the Nomination Process
The Work Group also suggests that EPA develop additional components to the nomination and
evaluation process.
-> Although the nomination and evaluation process would require further specification by
EPA, the Work Group recommends that the nomination process consider the following
elements:
• A communications strategy to identify and engage prospective stakeholders
• Recommendations for systematic communications with stakeholders
• Development of a consistent and transparent evaluation process by EPA, to include:
a) information and documentation requirements (i&, new and emerging agents or
potential contaminants should not just short-circuit the evaluative process);
b) an evaluation process that the nominated agents must undergo (ie.,for a new agent to
go directly to the PCCL or CCL), it must present appropriate occurrence and health
effects information as other PCCL or CCL contaminants;
c) a means for confirming that the information offered has not previously been
considered;
d) a process and criteria for taking appropriate action for those found to have merit, and;
e) a means for documenting the process and any decisions reached.
Ch. 2 - Overview of Process and Overarching Issues 2-17
-------
CCL CP Work Group Report
2.3.3.2 Accelerated Listing Process
As new agents are identified, or as new information becomes available, there may be justification
to accelerate their passage to the CCL Universe, from the Universe to the PGCL, or from the PCCL to
the CCL. EPA could, if the data warrant, consider these contaminants on an accelerated basis. The
Work Group recommends EPA develop a formal accelerated ("fast track") process and ensure that the
process is communicated before, or at the time the Agency requests nominations from the public. The
process should be open and transparent and be consistent with the overall C CL screening and
evaluation procedures. The accelerated process should also consider the elements outlined in section
2.3.3.1, above.
2.3.4 Information Quality Considerations
2.3.4.1 NRC Discussion and Recommendations
In its 2001 report on classifying drinking water contaminants for regulatory consideration, the
NRC addresses4 some of the difficulties and challenges that EPA will face i|i applying data and
information to any classification designed to sort a very large number of chemical and microbiological
contaminants into exclusive categories: On or Off the PCCL; On or Off the CCL. The NRC
recognized that EPA would likely encounter many challenges in implementing a classification scheme
where imperfect or incomplete data must be used to determine whether a specific chemical or
microbiological organism may cr may not pose an existing or potential threut to consumers of public
drinking water.
The NRC did not, however, make specific recommendations in its 2001 report as to how EPA
should address or resolve issues related to the quality of the information used in the CCL development
process. NRC refers to section 1412(b)(3)(A) of the SDWA Amendments v/hich addresses the use of
science in decision-making under this statute, specifying that EPA shall "usp the best available, peer-
reviewed science and supporting studies conducted in accordance with souijd and objective scientific
practices; and data collected by accepted methods or best available methods (if the reliability of the
method and the nature of die decision justifies the use of the data)."
2.3.4.2 Work Group Considerations and Recommendations on Information Quality
The Work Group also considered the issues related to ensuring the use of the best available
information and methods with respect to the data sources to be accessed, ttu: data elements to be
extracted from those sources, and the processes to be applied using those dajta elements to screen or
classify a very large number of contaminants in the CCL Universe to reduce it to the relatively smaller
numbers on the PCCL and then the CCL. The Work Group recognized that EPA's process should
explicitly address compliance with Agency data quality guidelines and the Information Quality Act.
The Work Group also recognize;;, however, that the Agency must have some flexibility in the data
quality guidelines to fully embrace the inclusionary principles. Work Group members noted that
contaminants considered in the early stages of the CCL process will not necessarily be robustly
* to the section titled "Tlie Nature of the Task."
Ch. 2 - Overview of Process and Overarching Issues
2-18
t
-------
CCL CP Work Group Report
characterized, and the data available for some of those contaminants wil consist of different types of
data The Work Group also recognized that the data or information used to select the CCL will be
more detailed and comprehensive than the data or information used to identity the CCL Universe.
To address the variability of the disparate types of data, and to ensure transparency, all steps in
die CCL process should document information about the data sources (e.g., what quality assurance
procedures were in place during data gathering, processing, or analysis). Additionally, the CCL
process should apply more scrutiny to contaminants when selecting the CCL than when screening
contaminants from the Universe to the PCCL. The nature of the data used to support these steps
should be documented for review in the later steps of the CCL process.
Different data quality approaches can be established commensurate with the purpose for which
the data will be used (e.g., screening from the CCL Universe to the PCCL versus classifying from the
PCCL to the CCL). This is a priority-setting process that does not require the same detailed analysts as
a rulemaking process, and therefore data quality considerations should recognize this difference. The
Work Group noted a related key consideration in (he CCL process should be mat, in general, false
negatives should be avoided when going from the Universe to the PCCL and false positives should be
avoided when going from the PCCL to die CCL. It is important for EPA to develop and document
appropriate data quality approaches as part of the process of implementing the adaptive management
approach discussed below (in section 2.3.6). EPA should establish data quality approaches for use in
each step of the CCL classification process prior to identifying the CCL Universe.
-> The Work Group, therefore, recommends that information quality be considered in the
CCL process.
This recommendation raises two questions.
1) How is information quality to be summarized at any stage in the process from building the
CCL Universe to selecting the CCL itself?
2) What are experts, or algorithms, to do with this information quality summary at each stage?
The answers to these questions must reflect that an assessment of information quality or
uncertainty about some "best estimate" of a numerical value (such as die exposure or potency for an
agent) can be resource-intensive, often requiring more resources than does determining the "best
estimate" value itself.
-> As an overall recommendation, the Work Group recommends that EPA collect and
consider the "best available" data sources and data elements without restrictions or
screening-out of information based on any minimum quality criteria developed in advance.
The Work Group also offers the following specific recommendations regarding the consideration
of information quality at the major stages of the CCL development process.
1) Establishing the CCL Universe: It will be possible to "tag" the agent with a reference to the
quality of the data source or other information used to assign mat agent to the Universe. This
indicator would refer to considerations of the quality of the information source, and not be
Ch. 2 - Overview of Process and Overarching Issues 2-19
-------
CCL CP Work Group Report
specific to information on the agent itself obtained from that source. Since the quality of
information on different agents in the same information source can, vary, it is recommended
that this "tag" not be used for screening agents out of the Universe] but only to provide an
indication of the general reliability of the source of information thai: should be considered at
later stages.
2) Going from the Universe to the PCCL: Even at this stage of the process, it will not be
feasible to perform an information quality analysis specific to a contaminant. It will be
possible, however, to provide a richer "tag" for each contaminant. For example, the "tag"
might document whether a measured value or a QSAR estimation 'vas used for a screening
element. Chapter 4 discusses this in further detail. Since the "tag" s:till does not reflect a full
analysis at this stage, it is not necessary to use the "tag" to screen contaminants off the PCCL.
3) Going from the PCCL to the CCL: The Work Group recommends that EPA (and expert
reviewers) consider the information quality "tag" more fully at this stage than at earlier stages
of the process. EPA should consider developing information quality tags for the PCCL entries
and using those tags explicitly in developing the classification algojithm and using it to create
the CCL. One possibility is to include information quality as a sixth candidate attribute in
developing the classification algorithm. If, however, a final algoritiun is selected that does not
include the information quality attribute, then explicit consideration of the "tags" should
occur during expert review after the algorithm has been applied to ihe PCCL, but before the
CCL is published.
Work Group members agreed mat the list of contaminants selected for the CCL should undergo
an expert review. Members noted mat documenting the nature and type of information by assigning a
"tag" for consideration at this step allows this information to be used in the final analysis for the listing
decision. By fully documenting the information used in the process, the review of the information
used, and the decisions made to develop (he CCL can be conducted in an o]>en and transparent
manner.
More specifically, the Work Group discussed using the "tag" as part of] the expert review process.
For example, the review process could allow contaminants to move from this PCCL to the CCL only if
the "tag" indicated sufficiently high reliability of the evidence supporting inclusion of a contaminant
on the CCL. This would preven: the CCL from being populated with a num.ber of contaminants that
would, upon further review, be rejected both for regulation and for further lesearch. The disadvantage
of this approach is that it could require resources to support a judgment of tjie qualitative, expert-based
judgment, of the quality of information for individual contaminants that are candidates for movement
from the PCCL to the CCL. Alternatively, considering the nature and type of information used to
select contaminants after the draft CCL listing may be useful in determining whether a contaminant
remains on the draft CCL and in establishing priorities for regulatory determination.
Ch. 2- Overview of Process and Overarching Issues
2-20
t
-------
CCL CP Work Group Report
2.3.5 Use of Quantitative Structure Activity Relationships (QSARs)
2.3.5.1 Introduction
As part of its consideration of options for including potential drinking water contaminants that
lack applicable empirical data on health effects or occurrence in the CCL process, the Work Group
was presented material on the use of Quantitative Structure Activity Relationship (QSAR) models and
the output of those models. The Work Group recognized that the health effects and occurrence-related
properties information generated by QSAR models could potentially be used both h screening
contaminants to develop the PCCL from the Universe, and as data elements for attributes to develop
the CCL from the PCCL. The focus of this section is whether or not the use of QSAR models appears
to be a reasonable approach to generating information about less-well characterized chemicals, that
could be used in either or both of these steps in the CCL process. The present discussion does not,
however, specifically address the use of QSAR-generated data in the screening or classification steps.
More specific consideration of the use of QSAR-generated data in those steps is addressed in Chapters
4 and 5.
The Work Group agreed that use of QSAR data for agents for which EPA does not have data is a
potential tool.
However, a few members raised questions about the proprietary software currently available to
develop QSAR data These Work Group members noted mat most of the widely-used models use
proprietary algorithms that are not available for independent review. For reasons related to
transparency, ethics and validity, these Work Group members could not recommend EPA use QSAR
models because of the proprietary nature of these models.
Other members suggested that the value of the use of these models is important. While
recognizing this concern about the need for transparency, they supported a recommendation that the
Agency should use proprietary QSAR models. These Work Group members also noted mat
proprietary QSAR applications or computational algorithms were independently reviewed in their
development, even though the proprietary models arc not available for subsequent reviews.
This section provides what the Work Group learned about QSAR models. The Work Group did
not reach consensus on a recommendation for QSAR models. Therefore two different
recommendations are included below.
2.3.5.2 Background on QSARs
The Work Group sought to learn enough about QSAR models, their requirements, and their
limitations to see if they could arrive at some general conclusions and recommendations regarding a
potential role for QSARs in the CCL process. The Work Group did not attempt to conduct a
comprehensive analysis of the suitability of all potentially applicable QSAR models or the output that
they can generate, but rather sought to develop sufficient information based on a limited number of
representative QSAR models to inform the assessment of their applicability for producing the type of
information that could be used in the CCL Classification process.
Ch. 2 - Overview of Process and Overarching Issues 2-21
-------
CCL CP Work Group Report
Specifically, the commercisilly available QSAR application called TOFKAT (The Open Practical
Knowledge Acquisition Toolkit) was used to predict thfc rat chronic oral Lojwest Observable Adverse
Effect Level (LOAEL), and the QSAR model package developed by EPA and Syracuse Research
Corporation called Estimation Program Interface Suite (EPI Suite) was usec to predict solubility and
aerobic biodegradation information. A set of approximately 700 chemicals was used to test these
models. Some of the chemicals nhat were evaluated also had empirical information for the properties
predicted by the QSAR models, and were used largely to get a sense of hov reliable the QSAR
predictions were. Other chemic&ls not having empirical data for the QSAR model outputs were
evaluated to provide some insight into the potential difficulties of applying the models to substances
that are less-well characterized by actual measurements.
Technical reports and presentations documenting the QSAR model evaluation process were
prepared by the technical team supporting the Work Group. The informatio:i presented, together with
discussions of that information ty the Work Group, led to the two different recommendations
presented in the following section. (See Appendix B for summary of technical reports on QSAR
model evaluation.)
2.3.5.3 Conclusions, Recommendations, and Rationale
-> The full Work Group recommends that EPA explore use of QSAR models to those (agents
or contaminants) for which EPA does not have data.
While QSARs might be a valuable tool, a few Work Group members qould not recommend use
of the QSAR models because they cannot be property tested and evaluated. These members suggest
that, if EPA chooses to use QSAR models, the Agency should develop and use only folly transparent
software that is available for independent review.
-> These members suggest, if EPA chooses to use QSAR models, from an ethical,
transparency and validity viewpoint, only fully transparent QSAK models that are
available for independent review should be used. If nonproprietarr software is not
available, QSAR models should not be used.
Other Work Group members offer a general recommendation (noted bdow) on the use of QSAR
models, along with several considerations to guide foe Agency.
-> These members suggest, lased upon a review of the historical use
-------
CCL CP Work Group Report
• Consistent with the overall quality assurance procedures that EPA should apply in the CCL
process, the Agency should limit its consideration of QSARs to generally accepted, peer-reviewed
QSAR models that are validated, adequately documented, and perform with well-described
precision and accuracy.
• The Agency should follow the Office of Research and Development (ORD) framework and
recognize that available QSAR model predictions reflect me limitations and biases of the data sets
and methods used to develop those models.
• In constructing a sound framework for integrating QSAR predictions into the CCL process, the
Agency should be careful to employ QSAR predictions in a scientifically rigorous manner
cognizant of the tool's limitations.
• The Drinking Water program should utilize internal Federal office expertise to the extent possible
in the selection and application of QSAR models.
• Additionally, outside experts with relevant expertise should review on an ongoing basis
development of the Agency's approach to QSAR models and rules by which QSAR predicted
values are applied in the CCL process.
• Therefore, the Work Group recommends that when reliable empirical observations are available,
QSAR generated values should not be used.
2.3.6 Use of an Adaptive Management Approach to Implementation
In development of the new CCL classification process, an adaptive management approach could
provide a method of evaluating progress at milestones during development, implementation and
review of the CCL process. By applying an adaptive management approach as the CCL process is
developed and applied, the Agency will be able to determine: a) whether decisions made in the
process have provided adequate results and the needed information; and, b) what modifications need
to be made to the CCL process in the current or successive CCL cycles. Adaptive management
principles could be applied in the development, implementation, and refinement of the three-step CCL
process, particularly in the initial phases of implementation, and provide a framework to refine the
process. As new information identified in the adaptive management framework becomes available, the
Agency could use that information to evaluate and refine the process for current or future CCLs. (See
Figure 2.2, below.)
Adaptive management is a well-established concept in environmental management. The idea
arose from the recognition that environmental management decisions will always be made with
uncertainty about the precise outcome of the alternative actions being considered. This inherent
uncertainty may partially be addressed by further research, but often additional study delays action,
and occurs on scales that incompletely capture the dynamics of the system that will be affected by die
management actions. Thus, the best way to reduce uncertainty is to take action, to treat such
management actions as an experiment, to monitor the outcome of such experiments, and thus to learn
by doing.
Under an adaptive management approach, reducing uncertainty is an important goal in
implementation of each generation of the method This process incorporates systematic and continual
integration of design, management, and monitoring, which would enable EPA to make informed
Ch. 2- Overview of Process and Overarching Issues 2-23
-------
CCL CP Work Group Report
adjustments and adaptations, resulting in an improved method based on experience from the outcomes
of successive generations of implementing the Universe-to-CCL approach.
Figure 2.2 - Diagram Schematic of an Adaptive Management Process
Elements each step in the
process should consider:
• Evaluative criteria for
each phase;
• Adaptive learning
process;
• Characterizing data
quality;
• Transparency;
• Use of expert judgment
/review.
Source: British Columbia Ministry of Forests, Forest Practices Branch, http'/^ww for.gov.bc.caWp/amho-ne/Amdefs.htm, August 9,2000.
While adaptive management stresses the need for practical action in th^ face of uncertainty, it
also emphasizes the need to tailor management decisions to the nature and Quality of information
available at any moment in the process. With little information, some policies with minimal potential
for negative consequences ("no regrets") may be in order. More or better information may justify
policies with (for example) greater economic costs. As information become s better established,
progressively more explicit decisions, with more serious consequences, are [justified.
Concepts of adaptive management are a consistent theme in both the MIC and NDWAC
recommendations. (See the bulbted list in text box.) Both reports stress the need to iterativeh/ test and
refine the CCL methods, rather than simply waiting until the methods are perfected before applying
them in decisions on the CCL. In this regard, the present report emphasizes (features that are well
described in the context of adaptive management: (1) identify an approach, (2) define evaluative
criteria (factors to evaluate), (3) iterativeh/ implement the approach, (4) transparently assess evaluative
criteria and (5) make changes to improve performance of the approach. Adoptive management also
recognizes the utility of comparing alternative approaches to the creation olj the CCL (e.g. differento
posteriori methods, or an approach rooted more in facilitated discourse thaii, in a posteriori methods),
and the need to select the approjich best suited to the quality of the information and performance
available. Perhaps most importantly, adaptive management integrates interim evaluations into the
overall approach so mat change can take place as information becomes available.
This type of management approach is similar to those used in businesses and complex
organizations dedicated to continuous improvement or high performance and should be familiar to
most modem managers. This application to environmental systems (in this case, contaminants to be
considered for further research and regulatory determinations) is only an extension or adaptation of
those design-measure-feedback-redesign business models.
Ch. 2 - Overview of Process and Overarching Issues
2-24
f
-------
CCL CP Work Group Report
Chapter 3
CCL Classification Approach for Microbial
Contaminants
This chapter identifies the challenges presented by the data and information available for
microbial agents and provides the rationale from the Work Group's discussion to address those
challenges. Section 3.1 addresses developing a Universe of microbial agents. Section 3.2 discusses
recommendations to screen the Microbial CCL Universe to the PCCL. Section 3.3 introduces a
discussion of protocols and considerations to evaluate microbial contaminants on the PCCL to select
the CCL microbes. Section 3.4 presents the Work Group's discussion and recommendations on the
use of genomics and proteomics, specifically virulence-factor activity relationships (VFARs), in the
CCL classification process. Each section reviews the NRC recommendations and presents the
NDWAC Work Group's recommendations for developing a CCL classification approach to microbial
contaminants.
Chemicals and microbes exert their lexicological or pathological effects following exposure via
ingestion, inhalation, or dermal contact, depending upon the specific agent- and host-dependent
variables. However, chemicals and microorganisms behave in markedly different ways in the
environment and within the human host. The methods and information used to characterize these two
types of agents also vary. Chemical agents tend to be characterized by lexicological and occurrence
data that, if not measured, can be modeled or estimated. The adverse health effects of microbial agents
tend to be characterized by clinical and epidemiological data. Estimating the occurrence or potential
occurrence of microbes can be based on the biological characteristics of the microorganism, but there
are few analytical methods available for making such assessments and the information used to
characterize microorganisms is not readily modeled or estimated. The differences in chemical and
biological characteristics of demonstrated and potential water contaminants suggest that, while
identifying the Microbial CCL Universe from the total microbialuniverse of microorganisms and
screening a subset of biological agents from the Microbial CCL Universe to a PCCL can be consistent
with the NDWAC's suggested principles for chemicals, at mis time the approach to microbial agents
and contaminants will require different data sources and data elements, and may require more
involvement from experts than the approach described for chemical agents and contaminants.
The Work Group has developed the following steps to select microbial contaminants for the
CCL.
• The total microbial universe may consist of all microorganisms.
• The Microbial CCL Universe may consist of all human pathogens (i.e., organisms known to
cause disease in humans).
• The PCCL may consist of all organisms in the Microbial CCL Universe that maypbusibly
occur in and be transmitted by drinking water.
Ch. 3- CCL Classification Approach for Microbial Contaminants 3-1
-------
CCL CP Work Group Report
• Microbes in the Microbial CCL Universe may be evaluated against screening criteria to
determine the plausibility for water-related transmission (occurrence). Pathogens that are
known to cause waterborne disease (health effects) are placed on^the PCCL.
» Surveillance and nomination provide an alternative pathway for entry into the CCL process
for new and emerging microbial agents.
" EPA should continue to develop the process for selecting organisms on (he PCCL for the
CCL
These steps are discussed in the following sections of this chapter. A schematic representation of the
recommended microbial CCL Classification Process is shown in Figure 3.1.
Figure 3.1 - A Microbial CCL Classification Process
Surveillance Nomination
Microbial CCL Universe
(all human pathogens)
PCCL
(v/aterborne
pathogens)
CCL
Total Microbial Universe
(all microorganisms)
Note that this process differs from NRC recommendations by defining the Microbial CCL Universe as
microorganisms known to cause human disease. Microorganisms demonstrating the potential to cause
human disease may be added to the Microbial CCL Universe when surveillance demonstrates adverse
health effects or by nomination, based upon available data and information.; The subset of human
pathogens that may plausibly survive in and be transmitted by drinking watiar comprise the PCCL, and
a subset of microorganisms on tie PCCL that meet attribute scoring criteria (see Appendix D) are
Ch. 3- CCL Classification Approach for Microbial Contaminants
3-2
-------
CCL CP Work Group Report
placed on the CCL. Whereas the NRC recommended including both contaminants that have the
potential to cause adverse health effects and those with the potential to occur in drinking water in the
Universe of Potential Drinking Water Contaminants (see Venn diagram in Chapter 4.2), the Work
Group decided upon a more stringent definition of the Microbial CCL Universe, since using the
NRC's criteria would place large numbers of microorganism in the Microbial CCL Universe whose
biological properties would prevent them from surviving in drinking water or causing human disease.
3.1 Identifying the Microbial CCL Universe
The Universe of known microorganisms includes bacteria, viruses, protozoa, algae, and fungi.
Some microbes from each of these categories are pathogenic to humans, or produce toxins causing
human disease. Pathogens that cause gastrointestinal disease (enteric pathogens) are shed in feces.
Examples include Salmonella, Shigella, Cryptosporidium, Giardia, and noroviruses. Enteric
pathogens of humans can be discharged into water by sewage treatment plants, septic tanks, storm
sewer flows, runoff events after a rainfall, and other processes. Runoff from animal feeding operations
and die fecal contribution of feral animals and migratory waterfowl also have the potential to
introduce microorganisms, including human pathogens, into the aqueous environment. Other
microbes are natural inhabitants of the soil and environmental waters, and are well-adapted to the low
nutrient level and cool water temperatures of the ambient environment. Some aquatic microbes may
cause disease in humans under certain circumstances, especially in individuals with a weakened
immune system or other major underlying conditions that facilitate infection resulting in disease.
Pathogens causing opportunistic infections include Pseudomonas aeruginosa, Legionella
pneumophila, and iheMycobacterium avium complex (MAC). Many of the microorganisms in these
two categories have not been identified. New microorganisms, including pathogens, are constantly
emerging via evolutionary processes.
3.1.1 NRC Recommendations for the Microbial CCL Universe
The NRC recommended general guidelines for defining the Microbial CCL Universe of potential
drinking water contaminants as those microorganisms that are known or have the potential to occur in
drinking water, and those microorganisms that are known or have the potential to cause human
disease from exposure to drinking water by ingestion, inhalation, or dermal contact. These guidelines
recognize that knowledge of microbial occurrence and health effects is incomplete, and they provide
latitude for inclusion of new and emerging pathogens as they are recognized. The NRC
recommendation did not limit the boundary of the Microbial CCL Universe, but suggested that
microorganisms could be added to that Universe based upon expert knowledge and new information
about their occurrence and health effects. (The NRC also recommended the inclusion of biological
toxins in the CCL Universe, but because these are produced and released in the ambient environment,
they are addressed as chemical - not microbial - agents and contaminants.)
The NRC (2001) view of the Microbial CCL Universe included agents that occur naturally in
water, agents associated with human feces, agents associated with human and animal feces, agents
associated with human and animal urine, and agents associated with water treatment systems and
distribution systems, together with biological toxins. Table 3.1 illustrates these categories and provides
examples of microorganisms for construction of the Microbial CCL Universe of potential drinking
water contaminants.
Ch. 3 - CCL Classification Approach for Microbial Contaminants
3-3
-------
CCL CP Work Group Report
Table 3.1 - Categories and E xamptes of the NRC-Proposed Microbial CCL Universe
bial
Gt&Btfttff ,__,__. v y
Naturally occurring agents in water
Agents associated with human feces
Agents associated with human and animal feces
Agents associated with human and animal urine
Agents associated with water treatment and
distribution systems
Biological toxins
B&NtAfriMt ' j •- , ,
Legionetla. toxigenic algae
Enteroviruses, coxsackie B viruses, rotavirus
Enteric protozoa and bacteria
Nanobacteria, microsporidia
Biofilmsorganisms, e.g. Mycobt'cteriumavium-intracelkilare
endotoxin, aflatoxin
1 Some examples can belong to me re than one category of contaminants.
3.1.2 Defining the Microbial CCL Universe
Figure 3.2 lists the actions the Work Group recommends EPA implement to identify Ihe
Microbial CCL Universe, and locates this step in the CCL process. The recommendations are
discussed below.
Fig. 3.2 - Microbial CCL Universe
STEP1
STEP 2
STEP 3
CCL r
Universe |
i
^
i
PCCL |
F
Proposed CCL 1
Ch. 3-CCL Classification Approach for Microbial Contaminants
3-4
-------
CCL CP Work Group Report
The Microbial CCL Universe may be framed after thoughtful consideration of all possible
occurrences of aquatic microorganisms that may be present in all source waters (ground water, surface
water, and marine waters where appropriate), water treatment plants, water distribution systems,
plumbing, and recreational water venues supplied by treated drinking water. Microorganisms of
primary concern in water treatment and delivery are those that cause human disease and are shed in
feces. Pathogens associated with septic waste and sewage may contaminate ground water and surface
waters, thereby posing a public health risk. Salmonella,Shigella, Cryptosporidium, Giardia,
noroviruses, hepatitis A virus, and enteroviruses are examples of pathogens that are shed in feces and
that may contaminate water, resulting in sporadic cases of illness or waterbome disease outbreaks.
Many microorganisms causing water-related diseases in humans are not of fecal origin, but occur as
natural inhabitants of the aquatic environment. Legionellapneumophila, Aeromonas hydrophila,
Pseudomonas aeruginosa, and many other microorganisms associated with water-related
opportunistic infections have their natural habitat in water.
Hundreds of microorganisms are known to be pathogenic, causing infectious diseases in humans,
while thousands of microorganisms that may be present in the environment have the potential to cause
infrequent opportunistic infections in humans under unusual circumstances of exposure and host
susceptibility. Conversely, thousands of microorganisms found in the aquatic environment or in
domestic water distribution systems are not known to cause adverse health effects in humans,
regardless of their number or route of exposure. The diversity of the Microbial CCL Universe, and the
wide range in host susceptibility of human populations, make it difficult to characterize microbial
occurrence and the potential for adverse health effects for research and possible regulation.
Health and occurrence data may be more readily available for chemicals than for microbial
agents. Existing health effects databases for microbes are based upon case reports from public health
surveillance programs and epidemiological investigation of water-borne disease outbreaks. Existing
microbial occurrence databases are based upon indicator monitoring, except for data acquired during
epidemiological investigation of water-borne disease outbreaks, limited academic research studies on
pathogen occurrence, and occasional regulatory information collection requirements (e.g., the
Information Collection Rule (ICR) that required selected public utilities to gather information on
Giardia, Cryptosporidium, enteroviruses, total coliforms and fecal coliforms). Limited sources of
tabular data on occurrence and health effects of microorganisms are available on the Internet or
elsewhere, however the content is frequently incomplete and the quality of data is variable. Because of
the lack of pathogen occurrence data in readily accessible form, keyword searches of bibliographic
databases of primary literature, conference proceedings, technical reports, monographs, and reference
books will be required to adequately populate the Microbial CCL Universe. NRC (199%) recognized
the limitation of existing occurrence and health effects information sources for microbial
contaminants, and suggested that expert judgment would remain an important component of the CCL
process. Until a unified database of microbial information is available, the process of initial selection
of microorganisms for the Microbial CCL Universe and subsequent iterations to move them through
the CCL Classification process to the CCL will rely heavily upon expert judgment.
The term "Microbial CCL Universe" implies a subset of contaminants from a universe of all
microorganisms. Because of the number of contaminants to be considered in the NRC
recommendations, the Work Group discussed building an inclusionary CCL Universe by following
the basic NRC principles and selectively combining data elements from data sources into a Microbial
Ch. 3 - CCL Classification Approach for Microbial Contaminants
3-5
-------
CCL CP Work Group Report
CCL Universe. (This "data sowce compilation" approach is described in Detail in Chapter 4.1.)
Construction of a Microbial CCL Universe is envisioned to entail selectivs compilation of existing
data sources into an inclusive ;md unified data set of known contaminant parameters, from many well-
characterized sources of data and information for contaminants recommended by NRC. While several
comprehensive sources of data and nformation have been developed for chemical occurrence and
health effects, few equivalent data sources exist for microbes. Thus, the approach for selection of
microbial contaminants for the Microbial CCL Universe may of necessity incorporate alternatives,
based upon NRC guidelines, by using information from a variety of qualitative sources, including
surrogate monitoring, modeling, primary literature review, and expert judgment.
-> The NDWAC Work Group recommends that the Microbial CCL Universe be based on the
evaluation of data sources and literature reviews that identify organisms known or
suspected to cause human disease.
A survey of the primary literature was conducted as an example of th:.s approach. Appendix A
from Taylor et al. 20015 was used as an illustrative starting point for the Mcrobial CCL Universe.
This list includes 1,415 recognized bacterial, viral, parasitic, and fungal pathogens. This article
represents an attempt to identify all the known human pathogens through 21 search of published
literature. However, some human pathogens do not appear on the list as a result of recent emergence
or taxonomy and nomenclature: changes. Additions to the Taylor list have 'been proposed, and the
Work Group suggested a mechanism for adding organisms to the Micrcbitd CCL Universe through
surveillance and literature review. Therefore, organisms that are known to cause water-related disease
would be included in die Microbial CCL Universe by definition.
Because construction of trie Microbial CCL Universe is constrained by limitations of readily
available data, the Work Group recognizes the need for a nomination process to provide a means of
adding new and emerging pathogens to the Microbial CCL Universe (see phapter 2.3.3). Advances in
genomics and proteomics offer the possibility that molecular techniques, sjjch as the WAR approach
discussed in Section 3.4, may one day provide objective screening capability for selection of microbes
for the Microbial CCL Universe.
3.1.2.1 Human Pathogens as the Basis for the Microbial CCL Universe
The Work Group discussed how the Microbial CCL Universe might te identified according to
the principle-based, iterative approach (see Chapter 4.1.2) -giving full consideration to the differences
between chemicals and microbes, and recognizing the limitations of equivjdent data. Members also
expressed concern over the blanket inclusion of all microbes with the potential to occur in water, or all
microbes with the potential to cause disease, based on the current limited State of microbial occurrence
and health data. Members believed that the biological properties of microorganisms controlling
population diversity and dynamics should be considered in defining the Miicrobial CCL Universe.
Admission of microorganisms to the Microbial CCL Universe is based upon a proven ability to cause
disease in humans; thus autotrophs, thermophiles or other environmental rHiicroorganisms that may
occur in water are excluded from the Microbial CCL Universe because thqir biological properties
make it implausible that they could cause human disease.
'Taylor, Latham and Wooftiousc 2001. Risk factors for human disease cnutgence (Appendix A). Phil Trans. R. Soc Land. B 256:983-98
Ch. 3 - CCL Classification Approach for Microbial Contaminants 3-6
-------
CCL CP Work Group Report
3.1.2.2 Ensuring Inclusiveness of the Microbial CCL Universe
Considering the scope and diversity of microorganisms in the universe of potential water
contaminants, and the relatively few comprehensive data sources on their occurrence and health
effects, identification of the Microbial CCL Universe will rely on expert knowledge and keyword
searches of available bibliographical databases to ensure inclusiveness, while maintaining perspective
on the practical likelihood of water contamination and disease transmission.
The Work Group believes that adopting a published list of known human pathogens as the basis
for the Microbial CCL Universe is practical and transparent in practice. However, the limitation of mis
approach in capturing suspected pathogens with the potential to occur in water requires development
of a process to identify new information on emerging pathogens. The surveillance and nomination
processes described in Section 2.3 ensure that the Microbial CCL Universe remains current.
3.2 Microbial CCL Universe to PCCL
3.2.1 NRC Recommendations for the PCCL
The NRC recommends screening the CCL Universe based on evaluations of occurrence and
adverse health effects. Occurrence information includes demonstrated or potential contaminants of
ambient or finished water. This concept is illustrated by the list of waters considered by the NRC and
the Work Group, e.g. tap water, distribution water, finished water, source water, and watersheds.
Having thus broadly defined the CCL Universe based on known or potential occurrence or known or
potential health effects, the NRC suggested using the characteristics of occurrence in water and
pathogenicity to selectively screen microbial contaminants in the Microbial CCL Universe for
inclusion in the PCCL. This screening process would be supplemented by expert judgment.
Ch. 3-CCL Classification Approach for Microbial Contaminants
3-7
-------
CCL CP Work Group Report
3.2.2 Screening Microbes for the PCCL
Figure 3.3 identifies the Work Group's recommendations for actions the EPA should develop to
screen the MicrobialCCL Universe.
Figure 3.3 - Screening Contaminants from the Microbial CCL Universe to the PCCL
STEP1
STEP 2
CCL I
Universe |
PC
/
PI 1
1
\
Proposed CCL I
The composition of the Microbial CCL Universe and PCCL recommqnded by theNDWAC
Work Group differs slightly from NRC recommendations. Only microorganBms demonstrated to
cause human disease would inhabit the Microbial CCL Universe. Because most human pathogens do
not occur in water, or lack biological characteristics that permit their survival in water, it is plausible to
Emit the Microbial PCCL to those human pathogens that may be transmitted by water, and only
human pathogens with the potential to occur in water would comprise the 1'CCL. (See Figure 3.1,
above.) A mechanism to prioritize and reduce this list of microorganism for evaluation and ranking is
needed, and the Work Group's rationale for accomplishing this selection process centered on two
questions:
» What are the biological characteristics of pathogens that determine the potential for their
occurrence in drinking water?
• How may a screening process be constructed to identify pathogens for evaluation and
possible addition to the PCCL?
Work Group members recognized that criteria are necessary to selectively identify pathogens to
include on the PCCL. The Work Group suggests adoption of a rule-based Selection process for
moving pathogens from the Microbial CCL Universe to the PCCL. These principles may not be
sufficient by themselves, and expert judgment may be needed.
Ch. 3-CCL Classification Approach for Microbial Contaminants
3-8
-------
CCL CP Work Group Report
-> The Work Group recommends that the selection of human pathogens for the PCCL start
with a Microbial CCL Universe of recognized human pathogens (e.g., the amended Taylor
et al. 2001 list), and that those pathogens known to be associated with source water,
recreational water, and drinking water be selected for inclusion into the PCCL.
The resulting Microbial PCCL should be based on natural habitat and biological characteristics
that indicate a pathogen's ability to be transmitted via water. The members identified a simple key to
identify organisms that should move to the PCCL (See 3.2.3). Microorganisms having the potential to
cause human disease, but not yet demonstrated to do so could be added to the Microbial CCL
Universe by identification of genomic or proteomic elements suggestive of virulence as this
technology develops (See 3.4). Newly recognized microbes associated with waterbome disease would
be added to the PCCL as a result of public health surveillance processes already in place, or by an
expert or stakeholder nomination process. This process results in a realistic Microbial CCL Universe
and PCCL. The NRC acknowledged that practical limitations (i.e., genomic and proteomic data
availability) would constrain the development process, and this Work Group proposal attempts to be
consistent with NRC principles while acknowledging those limitations and proposing reasonable and
creative solutions.
3.2.3 Screening Based Upon Biological Properties
The Work Group applied further selective principles to restrict the number of microbes on the
PCCL to those meeting plausibility criteria in addition to occurrence. Examples of the proposed
screening principles that would exclude pathogens from the PCCL are shown in the table below,
Table 3.2 - Proposed Screening Principles to Exclude Pathogens from the PCCL
Proposed Screening Principles
Obligate anaerobes (microorganisms that cannot survive in oxygenated environments)
Obligate intracellular pathogens (environmental survival in water implausible)
Pathogens transmitted exclusively by direct or indirect contact with blood or body fluids (including sexually transmitted
Pathogens transmitted exclusively by insect vectors
Normal human intestinal, skin, or mucous membrane flora (except when documented to cause water-related disease)
Pathogens transmitted exclusively by respiratory secretions
Pathogens transmitted exclusively by animal bites
Pathogens of animals that are not known to occur in humans (limited host range)
Pathogens causing rare occurrences of disease not associated with water-related transmission
Several pathogens are transmitted by multiple transmission routes, and they would have to be
evaluated individually for plausibility of drinking water transmission by ingestion, inhalation, or
dermal contact. For example, aerosol transmission of pathogens such as Mycobacteriiun spp. and
Legionella spp. places them on the PCCL. Respiratory pathogens must be evaluated individually for
Ch. 3-CCL Classification Approach for Microbial Contaminants 3-9
-------
CCL CP Work Group Report
plausibility for water transmission. Ten species of the, genus Bacillus appear in Taylor et al. Appendix
A, while only two species are characteristically associated with human illjiess, and neither of these
species represents a significant risk by drinking water transmission. These examples illustrate that, as
die Agency develops the CCL classification process, it should refine screening criteria to better
identity microbial contaminants that pose risk through drinking water transmission.
-> The Work Group supports the following concepts for EPA's consideration as they develop
future CCLs:
• Biological characteristics should be recognized as legitimate criteria for screening
pathogens for the PCCL
• The list of pathogens inhabiting the Microbial CCL Universe should be screened for
biological characteristics promoting or mitigating against survival and transmission in
water.
• Genera may be categorically excluded as long as provisions ate made for selective
exemption of single species of a genus, &g. Bacillus anthratis
3.2.4 Pathogens Associated with Opportunistic Infections
Many organisms in the Microbial CCL Universe might be included tecause of their implication
in a very few cases of disease, perhaps only a single case. Some of these organisms, such as
Erysipelothrix rhusiopathiae or Pantoea agglomerans, can be found in water.
-> The Work Group recommends that organisms associated with opportunistic infections be
excluded from the PCCL unless clinical, epidemiological, or sim lar other information
implicates diem as the cause of waterborne disease. The Work €
increase surveillance for infections caused by these organisms, e
subpopulations.
roup suggests that EPA
pecially in sensitive
This increase in surveillance, in the Work Group's view, should balance these organisms'
exclusion from the PCCL. These organisms would be selected for exclusion by a consensus of expert
opinion. Opportunistic pathogens that cause a higher incidence of disease and are normal inhabitants
of water (e.g., Mycobacterium avium complex and Pseudomonas aerugi^osa) would not be excluded
from the PCCL using this screening procedure.
3.2.5 Alternative Pathways for Adding Pathogens to the Microhial CCL Universe and the
PCCL (Surveillance and Nomination)
-> The NDWAC Work Group recommends the development of procedures to include some
microbes in the PCCL or CCL outside of the defined process foi microbes.
The dynamic nature of pathogen emergence necessitates use of surveillance data to develop a
Microbial CCL Universe that is identified from current information. OtNr mechanisms for placement
of organisms into the Microbial CCL Universe, onto the PCCL, or directly onto the CCL include
genomic evidence of pathogen potential, recognition of new water-related waterborne disease agents,
identification of a waterborne disease outbreak by an organism not previously known to cause such
outbreaks, or nomination of organisms by experts based upon epidemiological data.
Ch. 3 - CCL Classification Approach for Microbial Contaminants 3-10
-------
CCL CP Work Group Report
Figure 3.4 presents an example of an alternative pathway for expeditiously incorporating
emerging pathogens into the CCL classification process. This alternate pathway shows the different
types of information to be considered and how a pathogen may be incorporated at different stages in
die CCL process. This approach provides a means of integrating pathogens identified through public
health surveillance programs, identified by using VFARs, or nominated by experts or stakeholders
into the CCL process.
Figure 3.4
Alternative Pathways for Introducing Pathogens to the CCL Classification Process
Surveillance, Nomination, or Literature Review
Yes
Screening Criteria
Yes
Attribute Scoring
Emerging pathogens recognized through surveillance or nomination are evaluated for their
potential to cause waterbome diseases (WBDs), based upon their biological characteristics and
epidemiology. Pathogens recognized to cause waterborne disease are next evaluated for their
involvement in recognized waterborne disease outbreaks (WBDOs). Pathogens causing waterborne
disease outbreaks may be placed directly on to the CCL. Emerging pathogens with no evidence
suggesting their involvement as agents of waterborne disease are placed in the Microbial CCL
Universe for further observation. Pathogens causing WBDs but not recognized to cause waterbome
disease outbreaks are placed on the PCCL.
The selection of pathogens from the PCCL for inclusion on the CCL is accomplished by scoring
attributes as recommended by the NRC (see section 3.3, below).
Ch. 3 - CCL Classification Approach for Microbial Contaminants
3-11
-------
CCL CP Work Group Report
3.3 Use of Attributes to Classify Microbial Contaminants
3.3.1 NRC Recommendations for Classifying Microbial Contaminants to the CCL
This step in the CCL classification process is intended to reduce the Microbial PCCL to a list of
priority pathogens for the CCL. The Work Group concurs with the NRC and recommends that EPA
consider a "prototype classification" algorithm (discussed in Chapter 5) to classify contaminants using
attributes that characterize occurrence and adverse health effects. This step is dependent upon
identification of available data (i.e., what is known about occurrence and health effects of potential
contaminants) and quantifying the attributes for use in the prototype classification algorithm. Expert
judgment is considered an important component of this step, as it is in the overall process. NRC
further recommended that a single approach be developed for selecting chemical and microbial
contaminants, requiring the development of predictive measures of pathogen occurrence and
virulence. (This is further discussed in Chapter 5.)
Using a prototype algorithm to classify contaminants on the PCCL for consideration for inclusion
on the CCL, the NRC selected five attributes to represent the contaminant's ability to cause health
effects or potential to occur in water. The Work Group has adopted the health effects attributes of
potency and severity, and the occurrence attributes of prevalence, persistence/mobility, and
magnitude, as starting points for evaluating and ranking agents as recommended by the NRC. (See
Chapter 2, section 2.2.2.3, for the NRC's definitions of these attributes.)
3.3.2 Use of Attributes for Characterizing and Ranking PCCL Microbes
The Microbial PCCL consists of human pathogens that are documented to be or may be
transmitted by drinking water, however the occurrence and health effects of these pathogens range
from rare and life-threatening to common and self-limiting. One way to prioritize pathogens for
placement on the CCL is to evaluate them for attributes as described abcnje. To quantify attributes, it is
necessary to use data element; for occurrence and health affects that are (nost appropriate for
microbes. Components reviewed by the Work Group for constructing scaring protocols are shown in
Appendix D.
The terms potency, severity, prevalence and persistence/mobilitysind magnitude most clearly
relate to chemical risk assessment models and practices. The Work Group recognizes that these
attributes, the protocols used to characterize contaminants for each attribute, and the data and
information used will need to be considered in context to describe the multiplicity of factors involved
in infective processes in humans. Pathogen occurrence, infectivity, and hpst susceptibility and
immune response determine the outcome of the host-pathogen relationship. An understanding of these
terms in the context of host-pathogen relationship is a prerequisite to assessments of microbial health
effects.
For example, \hepotency attribute characterizes the amount of a contaminant required to cause an
adverse health effect: i.e., an infective dose for a susceptible host. A potential pathogenic
microorganism must be viable, infective, and virulent. The pathogen-host] relationships determine the
course of disease in the host, which relates most closely to the term severity from the recommended
attributes. The outcome of the pathogen-host interaction is manifested in ;i spectrum of disease
ranging from asymptomatic infection to death of the host. The pathogeniqity of the microbe, the mode
Ch. 3 - CCL Classification Approach for Microbial Contaminants 3-12
-------
CCL CP Work Group Report
of transmission, and the population susceptibility determine the magnitude of the health effects.
Magnitude, in a microbiological context, was defined as the extent to which the pathogen can cause
disease outbreaks or significant numbers of individual cases above the endemic burden of disease in
the population.
To support the CCL process, EPA assembled a Microbial Sub-group comprised of
microbiologists and risk assessors in the Office of Ground Water and Drinking Water, Office of
Science and Technology and the Office of Research and Development to develop draft attribute
scoring protocols based on the NRC recommendations. The draft protocols are provided in Appendix
D of mis report. The elements considered for each of the attribute protocols neededto take into
account the data that were available and the expert judgment required to score each attribute. As with
the development of the Microbial CCL Universe and screening criteria of microbes from the Universe
to the PCCL, the available information was found in the primary literature and not in developed
databases.
3.3.3 Developing Draft Protocols to Quantify Attributes
The Work Group had the opportunity during its deliberations to review and evaluate the draft set
of specific attribute protocols for microbial contaminants that were developed by EPA. In addition,
some Work Group and technical support team members worked closely with EPA staff during its
deliberations to develop some specific technical guidelines to quantify attributes for the
microbiological contaminants. The Work Group provides general recommendations to develop and
evaluate attributes in Chapter 5. The remainder of this section summarizes the discussion specific to
microbes and the five attributes. These discussions identified issues EPA should consider in refining
the attributes and methods to quantify attributes for microbes and the CCL CP, and are presented
below.
Potency. Health effects attributes include potency and severity. Potency is defined by the NRC as
the amount of a contaminant mat is needed to cause illness. For microbes the infective dose is the
most useful marker of potency, however the infective dose is not known for many pathogens.
Microbiologists frequently speak in terms of the minimum infective dose, but the terms LD50and
lethal dose apply only to animal studies or in vitro cell culture assays. Some pathogens cannot be
grown in the laboratory and their infective dose can only be estimated. In the future, quantitative
virulence-factor activity relationships may become available for determining the relative potency of a
pathogen. The draft attribute characterization for potency scoring is constructed in a manner to allow
for absent data elements, while admitting the use of available information. The data elements that
should be considered for potency include knowledge of water-related disease, the class of pathogen
(i.e., bacteria, viruses, protozoa), the burden of disease in the population, the infective dose of the
pathogen, the likelihood of fecal or urinary sheddiig in humans and animals, and the presence of
genomic sequences conferring virulence. The attribute score was derived by categorizing the
pathogens according to a hierarchical scheme that started with data likely to be known for all
organisms such as knowledge of waterborne disease, and then subcategorized using data less likely to
be known like morbidity and infective dose. Each layer of subcategorization provides increased
resolution for the score. However, even those with minimal amounts of data will receive a score
commensurate with what is known. A proposed system for scoring potency is shown in Appendix D,
Table D2.
Ch. 3 - CCL Classification Approach for Microbial Contaminants
3-13
-------
CCL CP Work Group Report
Severity* NRC defines severity as the seriousness of the health effect, and suggests severity be
based on "the most sensitive health endpoint for a particular contaminant^ and considering vulnerable
subpopulations;... [and] should be based, when feasible, on plausible exposures via drinking water."
For microbial agents, severity may be defined in terms of colonization, infection, immune response,
disease, sequella, or death. The host-pathogen relationship is variable and dynamic. This continuum
may be unrecognizable at various stages. The most sensitive endpoint indicative of host-pathogen
interaction is an immune response, however this is not a practical end point for assessment of health
effects, since immunodeficient populations may be infected without eliciting an immune response.
While chemical health effects may be immediate or cumulative, microbiological health effects may be
unapparent for an extended time, depending upon the incubation period o:* the pathogen, and the
manifestation of disease. The data elements for scoring severity include recognition of significant
morbidity and mortality, the location and intensity of infective processes, the extent of contagion, the
amount of time lost to illness, the extent to which medical intervention is required for recovery, and
chronic manifestations or disabilities associated with the disease.
A centra] issue with severity scoring is whether to score on acute matiifestations of disease in
normal populations, or to scon; the worst possible outcome in the most sensitive population. Because
most frank pathogens are capable of killing some segment of the population, using worst possible
outcome in die most sensitive host inflates and clusters scores. The initial severity scoring tables were
constructed to use median outcome in normal populations, with case fatality rate and patient
population classification and percentage of patients in the population classifications as weighting
factors. One such approach applied the attribute characteristics to the population for which the most
data and information were available, then recalculated scores to acknowledge special circumstances
and to apply additional stringency. This proposed system applied worst case scoring criteria for
healthy and sensitive sub-populations, thereby driving many pathogens to maximal scores.
In an effort to overcome the complexities and limitations of a scoring {System using case fatality
rates and population-based weighting factors, the Work Group proposed a series of questions carefully
constructed so that a 'yes' answer indicated significance while a 'no' answer did not This approach
was useful in sequentially determining the cumulative data elements contributing to severity of disease
for both normal and sensitive sub-populations. Examples of the questions include:
• Does the organism cause significant morbidity (> 1,000/year) in the U.S. ?
• Does the illness require medical intervention for resolution?
• Does the organism cause mild disease in normal populations, bit', severe disease in
individuals with predisposing conditions?
• Does the organism cause pneumonia, meningitis, hepatitis, encephalitis, endocarditis, or
other severe manifestations of illness?
The more questions answered affirmatively the higher (more severe) the sqore. (See Appendix D,
Table D3.)
Prevalence. For the occurrence attributes, NRC defines prevalence as, "How commonly does or
would a contaminant occur in drinking water?" Prevalence may be determined using six of the seven
measures proposed by NRC in the PCCL screening criteria for demonstrate or potential occurrence.
In order of preference, these are: demonstrated occurrence in (1) tap water, (2) distribution systems,
Ch. 3 - CCL Classification Approach for Microbial Contaminants 3-14
-------
CCL CP Work Group Report
(3) finished water of water treatment plants, and (4) source water used for supplying drinking water;
and, if no information is available to demonstrate occurrence in water, (5) observations in watersheds/
aquifers, or (6) historical contaminant release data. It should be emphasized that prevalence involves
the consideration of both geographical (spatial) and temporal ranges of occurrence. Most pathogen
occurrence data are based upon indicator monitoring, hence they become surrogate information, not
pathogen occurrence data. Pathogen occurrence data come from epidemiological investigations
following outbreaks, research studies on pathogen distribution, and detection method evaluations.
There is little pathogen outbreak-occurrence information and even less pathogen data regarding
environmental and drinking water occurrence.
The Work Group developed a conceptual framework for prevalence, based upon actual detection
in drinking water, actual detection in source water, potential for zoonotic transmission through water
contamination, and potential for zoonotic agents to infect humans (host range). As with potency, a
simple scheme of hierarchical categories was tested, with the first category dividing pathogens by
their presence or absence in drinking or source water and subcategonzation according to any known
estimate of the frequency. These hierarchical categories are the basis of Appendix D, Table D4.
Prevalence scoring using these criteria proved to be more straightforward man other attributes,
primarily because occurrence data are either available or not available, limiting the number of criteria
in the scoring system.
Persistence/mobility. NRC used a persistence/mobility attribute as a surrogate for potential
occurrence when information is unavailable for a contaminant regarding its demonstrated occurrence
in water. For microorganisms, the following three characteristics pertain to their persistence and/or
mobility: high potential for amplification under ambient conditions; sedimentation velocities and
absorption capabilities; and, death or the ability to produce non-culturable or resistant states (e.g.
spores and cysts). When a contaminant already has data on demonstrated occurrence in water, and
thus information for the prevalence and magnitude attributes, those attributes will take precedence
over persistence/mobility.
Persistence implies steady state occurrence or amplification of microorganisms in water. This
occurs in surface water by production of resistant forms such as spores, cysts, oocysts; by colonization
of other life forms serving as a reservoir, through symbiotic relationships with amoebae; by adsorption
to particles; or by production of quiescent forms such as viable but non-culturable bacteria. In water
treatment plants and distribution systems, persistence is associated with colonization of infrastructure,
e.g. production of biofilms. Organisms that amplify (grow and multiply) are given higher scores than
organisms mat produce resistant forms but do not amplify in water. This scoring scale may
overemphasize relatively innocuous organisms that produce biofilms but rarely or never cause disease
in humans.
Data elements for scoring persistence-mobility include survival time in water under ambient
conditions, ability to amplify, ability to produce resistant forms, relationship to particles, and potential
for symbiotic relationships enhancing survival. The persistence-mobility scoring table (Appendix D,
Table D5) emphasizes non-turbid waters (i.e. ground water and treated drinking water), but the Work
Group believes that all source water should be included in scoring. Amplification frequently occurs in
surface source water where mere are large amounts of available nutrients, whereas the assimilable
organic carbon is limited in ground water and treated water, slowing or restricting amplification.
Ch. 3-CCL Classification Approach for Microbial Contaminants
3-15
-------
CCL CP Work Group Report
Persistence of bacteria that amplify under environmental conditions is highly variable, and the extent
to which they persist and move is largely a function of their population density.
Persistence of bacteria in drinking water is frequently related to the Jibility of bacteria to produce
biofilms, which promote growth of heterogenous bacterial populations while protecting them from
disinfection. (Biofilms are dynamic populations of bacteria that slough and serve as a steady-state
source of bacteria) Persistence of bacteria in the environment is determined by physical conditions
such as temperature and pH, availability of nutrients, presence of predators, and the ability of
microorganisms to form capsules, slime layers, spores, cysts, or other resling forms. Mobility of
bacteria in water is passively dependent upon hydraulic flow, which may .suspend bacteria adsorbed to
particulate material and sheer tnicrocolonies from biofilms. Because of tit; number and
unpredictability of these variables, it may be inappropriate to equate persistence-mobility of organisms
in surface waters with persistence-mobility in non-turbid waters such as ground water or treated
drinking water.
Mobility is not limited to chemicals, since microorganisms move though the aqueous
environment and in distribution system water actively (motility) and passively (adsorbed to
particulates, in symbiotic relationship with amoebae, and by hydrostatic flpw). Organisms percolate
through soil layers to contaminate ground water. Viruses are particularly mobile because of their
extremely small size and their relatively long survival times in the environment. Because mobility is
associated with the hydrodynamics of distribution systems, presence of bicfilms, presence of
particulates, and opportunity for symbiotic relationships., it is considered together with persistence for
scoring purposes.
Magnitude. NRC defines magnitude as "the concentration or expected concentration of a
contaminant relative to a level that causes a perceived health effect" (NRC ,2001). For characterizing
the attribute of magnitude, ideally two data elements are needed: the concentration of a contaminant in
water, and the concentration associated with an adverse health effect. NRC1 recommended the use of a
median water concentration in combination with a measure of potency, if available. Magnitude, in a
microbiological context, implies delivery (persistence-mobility) of an infeqive dose (potency) to the
customer's tap with resulting illness. The Work Group scored magnitude according to the number and
frequency of waterbome disease outbreaks reported in the U. S. and around1 the world, pathogen
distribution, and biological properties determining pathogen distribution. A draft scoring table is
shown in Appendix D, Table D6.
The microbial contaminants considered for preliminary scoring exercises were drawn mostly
from the current CCL. A set of ssven microbes was used by the EPA Microbiology Sub-Group, and a
set of eleven microbes on the current CCL plus Pseudomonas aeruginosawas used for a scoring
workshop sponsored by AWWA in November 2003. These contaminants wsre not sufficiently
representative of the range of pathogens likely to occur on the PCCL to adequately test the validity of
the scoring algorithms. However they did provide participants with example^ to test the scoring
protocols and provide suggestions to refine the scoring protocols.
The Work Group noted that attribute scoring using the EPA attribute scoring algorithms requires
expert knowledge based upon text-based literature to assign scores, Preliminjjry scoring exercises
conducted by different individuals produced different scores, and the same individual scoring
organisms on different days may produce slightly different scores as a function of the basic
Ch. 3- CCL Classification Approach for Microbial Contaminants
3-16
-------
CCL CP Work Group Report
assumptions entertained at the time. This variability suggests that scoring exercises should be
conducted by several experts and the results combined to arrive at final scores and rankings. The
Work Group also noted mat the scoring will need to document assumptions and data or information
used to score the attributes. The present algorithms do not lend themselves to automated scoring and
will require expert judgment and interpretation of text based sources. Nevertheless, the scoring
algorithms while considering a broad range of available information are relatively simple, thus scoring
can be performed easily and updated as necessary. Even in the absence of many bits of information a
reasonable attribute score can be determined, and as additional data become available the scores can
be refined. The simplicity is appropriate given the triage nature of the CCL, and makes the process
readily transparent.
The scoring algorithms proposed result from a lack of tabular data in organized databases. They
are based upon premises relating to health effects and occurrence of pathogens, supported by text-
based resource materials and expert knowledge. While existing genomic databases may eventually
facilitate a more objective approach for selecting genomic sequences associated with virulence of
microbes, databases containing health effects and occurrence data elements are only now being
considered. It is unlikely that a unified, searchable database of relevant data elements will be available
for selection of microbes for the CCL for several years. Meanwhile, expert processes will be required
to conduct attribute scoring, to evaluate the validity of scoring results, and to determine the threshold
for placing agents on the CCL.
The Work Group recognizes that the preliminary exercises using this scoring approach have not
attempted to reconcile scores to produce a composite result for each pathogen in the test data set, thus
the plausibility of resulting pathogen rankings has not been evaluated fully. Likewise, no attempt has
been made to date to evaluate and rank combined chemical and microbial scores resulting from
attribute scoring exercises.
3.4 Applications of Genomics to the CCL Classification Process
The final section of this chapter summarizes the NRC recommendations regarding application of
virulence-factor activity relationships (VFARs) to the CCL Classification Process, describes potential
applications of functional genomics and proteomics in the context of the CCL process to interested
stakeholders, and outlines possible short- and long-term options for further deliberation.
3.4.1 NRC Recommendation on Genomics
The NRC recommended use of VFARs for predicting the virulence of waterbome organisms as a
companion approach to quantitative structure-activity relationships (QSARs) for chemicals. Rapid
advances in bioinformatics, functional genomics and proteomics, together with development of
powerful molecular analytical tools such as polymerase chain reaction (PCR) and microarrays (bio-
chips), provide the technology to screen microorganisms at the genetic level even when their genomes
have not been fully sequenced. Theoretically, genetic elements coding for surface proteins, toxins,
attachment factors, invasion factors, or other virulence descriptors that are shared by microbial
pathogens can be identified and related to behavioral traits mediating severity, potency and
persistence. Thus, VFARs may be used to detect potential pathogens, and to rank or score attributes
pertaining to occurrence and adverse health effects.
Ch. 3 - CCL Classification Approach for Microbial Contaminants
3-17
-------
CCL CP Work Group Report
3.4.2 Potential Applications of Genomics
Bacteria sense environmental conditions, and respond to host exposure by turning on genes that
enhance environmental survival or their ability to invade host cells and dause disease. Polysaccharides
contained in the bacterial eel1 envelope and elaborated into the immediate cell environment (capsule
and slime layers), attachment mechanisms, symbiotic relationships with bther microorganisms, and
induction of a quiescent state all facilitate environmental survival., while activation of adherence,
invasion, toxin production, and various secretory genes facilitate pathogenesis. Shared nucleic acid
sequences (conserved region:; of chromosomes) within gene clusters asspciated with virulence
(pathogenicity islands) may be related to the NRC attributes of severity ind potency. Likewise, shared
nucleic acid sequences within those gene clusters associated with survival! in the environment may be
related to the NRC attributes of prevalence and persistence.
The genetic basis of these responses to environmental and host stimuli can be targeted for
construction of VFAR gene databases to screen for the presence of VFA-i genes in other organisms
sharing virulence- or survival-related gene sequences. These VFAR sequpnces may be used to rank or
score the NRC attributes pertaining to occurrence and adverse health effects, and they may serve as
primer sequences for PCR for molecular detection of virulence genes in unrecognized pathogens, for
direct detection of pathogens in environmental samples, and for construction of microarrays
containing hundreds of WAR gene sequences for rapid screening of microorganisms for their
pathogenic potential. These genomic applications may eventually be sensitive enough for detection of
non-culturable microorganisms, and for the direct detection of pathogens in environmental samples.
For pathogens known to cause waterbome outbreaks, occurrence data alqne may be sufficient for
inclusion on the CCL.
It is theoretically possible to select large numbers of genes associated with virulence and to use
these genes to screen the microbial universe for selection of potential pathogens for the PCCL. By
using suites of functionally related genes associated with bacterial pathoaenicity islands, it may be
possible to further select and prioritize potential pathogens from the PCCL for inclusion on the CCL,
based upon genetic function to enhance survival, manifested as potency, |Uid mediating severity of
disease. A categorical scoring system might be constructed, based upon the number of VFAR genes
identified and upon possession of functional suites of genes assigned to each NRC attribute.
Microarray technology is developing rapidly, together with knowledge of the molecular basis of
virulence. Microarrays have bsen used to detect viruses in cell cultures and clinical specimens, thereby
demonstrating the feasibility of the technology for pathogen detection. Similarly, microarrays may be
constructed to screen the microbial universe for potential pathogens for inclusion on the PCCL, and to
further prioritize PCCL organisms for inclusion on the CCL. An extension of this technology would
be the application of proteomizs by constructing microarrays to detect gejie products associated with
persistence or pathogenesis.
3.4.3 Challenges to Use of Genomics
Microbial genomes exhibit considerable plasticity, with frequent acquisition and loss of genetic
elements. The presence of multiple mobile genetic elements (e.g. bacteriophage, plasmids,
transposons, insertion sequences, etc.), together with the relative frequenq/ of chromosomal
recombinations, results in highly dynamic genomes that confound predictability. Presence of virulence
Ch. 3-CCL Classification Approach for Microbial Contaminants
3-18
-------
CCL CP Work Group Report
factor genes does not automatically result in expression of gene products; indeed, some genes (toxins,
pili, etc.) are controlled by multiple transcription regulators. The presence of enzymes that hydrolyze
nucleic acids (nucleases) in the environment, and substances in environmental samples that interfere
with PCR reactions (PCR inhibitors resulting in matrix interference) mitigate against detection of
identifiable free nucleic acid sequences in environmental samples. Pathogens are typically present in
the environment at concentrations below the detection limit of PCR, and culture enrichment
techniques are necessary before PCR may proceed successfully. Finally, validation of PCR methods is
problematic, thereby restricting its application.
An inherent limitation of genomics and proteomics is that they only recognize known gene
functions. Spontaneous mutations cannot be predicted, and heretofore unrecognized gene functions
will not be included in virulence factor screens. Available genomic information on microorganisms is
variable. While many viral genomes have been sequenced, relatively few bacterial, and evenfewer
protozoan genomes are known. Sequences deposited in GenBank or other genomic databases are
frequently incomplete, and accompanying annotations describing gene function frequently are
speculative. Sequence quality is highly variable, and no mechanism exists to assure data quality.
Because genomic databases are constructed using known microorganisms, genes, and fragmentary
sequences, the WAR approach has limited predictive value for anticipation of pathogen emergence at
this time. Currently, the only means of recognizing emerging pathogens is after they have caused
outbreaks, significant morbidity in a population, or serious outcomes in a few cases.
3.4.4 Pilot Projects
Initial explorations conducted for the Work Group using genomic databases to identify VFAR
genes were based upon virulence mechanism keyword searches. These searches identified variable
numbers of sequences for potential waterbome pathogens, but the vast number of unrelated sequence
matches, and the scant number of whole bacterial genomes precluded use of these sequences for
screening purposes. Genomic database searches based upon known virulence gene sequences
published in peer-reviewed literature detected shared sequences among bacteria, but revealed little
information about gene regulation and expression. These gene sequences have potential in selection of
microorganisms from the universe for inclusion on the PCCL. Genomic database searches based upon
mobile genetic elements, e.g., plasmids and pathogenicity islands, revealed multiple virulence-
associated sequences that were widely shared and transferred among bacteria. These suites of genes
offer promise for selecting organisms from the PCCL to the CCL by using them to score NRC
attributes.
A pilot project constructed a web-based database compiling information on organisms, outbreaks,
and genomic data on waterbome pathogens (hat can be used to identify potential pathogens for
inclusion on the PCCL and prioritizau'on of pathogens on the basis of potential virulence for inclusion
on the CCL. This database relies upon sequences deposited in GenBank or other genomic databases,
together with occurrence and epidemiological data on individual pathogens. This web database is not
expected to have predictive value for emerging pathogens.
Another pilot project was devoted to whole genome alignments of viruses and bacteria, to
identify conserved sequences that may be used to screen potential pathogens for virulence potential.
The approach depends upon the availability of whole genome sequences of the pathogens of interest.
Once unique WAR gene screening sequences are identified, they could be used to screen omer
Ch. 3—CCL Classification Approach for Microbiat Contaminants
3-19
-------
CCL CP Work Group Report
potential pathogens by sequence alignment using custom databases, or by constructing microarrays
based upon genomic or prottxunic technologies.
3.4.5 Recommendation s for the Use of Genomics in the CCL Process
Genomics and proteomics represent powerful tools for elucidation (if pathogenic mechanisms of
microorganisms; however, there are serious h'mitations to this technology that affect its application to
the CCL Classification Process.
• The technology is largely unproven for the desired applications.
• The technology may not be available in a robust form for use in the next CCL.
Despite these limitations, the Work Group recommends two steps mat can be taken to apply genomics
to the CCLCP process.
-> Select known virulence genes of gastrointestinal pathogens to identify data for screening
unknown organisms.
These data are based upon published sequences deposited in genomic databases, and can be used
to screen potential pathogens as their sequences become known, to construct microarrays for
screening potential pathogens, and ultimately for construction of PCR enhanced microarrays for direct
detection of potential pathogens from environmental samples. These data may be used for selection of
potential pathogens from the universe for the PCCL.
-> Select clusters of known virulence genes contained within pathog;enicity islands of
chromosomes or contained in mobile genetic elements that code for major mechanisms of
pathogenicity, e.g. adhesion, invasion, toxin production, etc.
These suites of genes contain both core function and regulatory control for gene expression, and
they are known to confer virulence when transferred to previously non-pathogenic bacteria. Selected
pathogenicity islands including genes responsible for attachment pili. protein secretory systems, and
toxin production may be used to rank and score attributes to facilitate selection of PCCL organisms
for the CCL.
Both of these processes may be implemented in the near term using published literature and
genomic databases. However, ihe microarray applications may be delayed, until technical and financial
limitations are resolved.
Both genomics and proteomics are developing rapidly and it is probable that microarrays or other
evolving technology will be available to facilitate selection of potential pathogens for the PCCL and
the CCL by 2010. Meanwhile, expert judgment, based upon published literature, epidemiological
investigations, and public health surveillance, remains the key approach for selecting potential
pathogens for the CCL. Expertise in bioinformatics and the molecular mechanisms of microbial
pathogenesis are desirable additions to microbiology, epidemiology, and water treatment expertise of
expert review panels.
To use VFARs to identify pathogens for inclusion in the CCL process., a wide variety of
information needs to be integrated. These types of information usually are iiot found in a single
Ch. 3 -CCL Classification Approach for Microbial Contaminants
3-20
-------
CCL CP Work Group Report
database but rather in a number of databases. The components of the information that need to be
integrated are identified in the NRC report. In many cases the NRC report says these things can be
done, but it does not specify the detailed logistics of doing this. The logistics of locating and
integrating these different types of information should be explored in detail.
-> To ease the process of automating microbial CCL processes in the future, the Work Group
has identified a number of practical measures that EPA should implement in the short-
term:
• EPA should find or define an approach to evaluate data incrementally and in a manner
that will readily allow application of search and match techniques to approximate the
process QSARs use in eliciting structural similarity (and inferentiatty, similarity of effect)
between structures in known and unknown organisms or genomicfragments.
• EPA should find or provide a physical repository (ie, data warehouse) for that material
• EPA should monitor the progress of genomes and the related technologies and integrate
them into the CCL process.
• EPA should monitor the data and information that emerge as genomicsprogresses and
integrate them for consideration in the CCL process, using an automated process to the
extent possible. The process should be updated and maintained in a continuing process
and verified against expert opinion.
• EPA should review public health surveillance techniques, in conjunction with the Center
for Disease Control (CDC), with a view to making those techniques as proactive, robust
and effective as possible in identifying the occurrence of water borne or watershed disease
outbreaks and the organisms associated with those outbreaks.
Ch. 3- CCL Classification Approach for Microbial Contaminants
3-21
-------
CCL CP Work Group Report
Chapter 4
CCL Classification Approach for Chemical
Contaminants
The purpose of this chapter is to review the NRC recommendations, and present the NDWAC
Work Group recommendations for the CCL Classification Process for chemicals. Section 4.1 offers
principles and a process for identifying a Chemical CCL Universe that is as inclusive as possible with
respect to potential drinking water contaminants. Section 4.2 discusses recommendations for selecting
a PCCL from the Chemical CCL Universe and provides potential screening approaches. Section 4.3
addresses the attributes used 1o characterize chemical drinking water contaminants, and me use of
different types of information and data to quantify those attributes for fuf ther decision-making.
4.1 Building the Chemical CCL Universe
4.1.1 Summary of NRC Ftecommendations
NRC (1999b and 2001) noted that an "ideal CCL development process" would identify the entire
Universe of potential contaminants and use data-driven screening processes to reduce the agents under
consideration to a CCL listing only those contaminants with a high probability that they need to be
regulated. However, the NRC recognized that currently the process is far from ideal - no
comprehensive list of potential drinking water contaminants yet exists; hqalth effects, occurrence, and
other related data for the vast majority of potential contaminants are highly variable, poor or
nonexistent; and EPA's resources are constrained (NRC 2001).
The NRC focus on a Universe of agents with "demonstrated or potential occurrence and/or
demonstrated or potential human health effects" is illustrated by examples of the kinds and classes of
agents recommended for consideration (see Tables 3-1 and 3-5 of the NRC report). The NRC also
suggested various data sources that should be reviewed in identifying CCt candidates (Tables 3-2
through 3-4, NRC, 2001). The NRC examples can be grouped into five subject areas as follows:
1) chemical agent groupings (e.g., "pesticides," "gas additives," "military munitions,"
"Pharmaceuticals") and types of microbes ("agents associated with human and animal feces");
2) transformation products (e.g., "reaction and combustion byproducts");
3) naturally-occurring substances (geochemical contaminants, radionjjclides);
4) biologically-active agents (e.g., "enzyme inhibitors," "hormonally active compounds"); and
5) chemicals with potential to enter drinking water (e.g., "compounds'widely applied to land,"
"constituents found in a landfill leachate," industrial discharges).
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-1
-------
CCL CP Work Group Report
These five groupings may provide useful insights in extending the context of NRC's examples
when identifying a CCL Universe that is consistent with the NDWAC's mclusionary, principles-based
approach discussed below.
Figure 4.1 lists the recommended actions EPA should develop to identify the Chemical CCL
Universe, and locates this step within the three-step CCL classification process.
Figure 4.1 - Detailed Overview of Step 1 (Chemical CCL Universe)
STEP1
STEP 2
CCL
Universe
i
PCCL
F
Proposed CCL
ifH}j|lHlMtJlt JMJ
4.1.2 Overall Recommendations for Identifying the Chemical CCL Universe
-> EPA should adopt a principles-based approach consistent with that described by the NRC.
After review of NRC's recommendations, avaiable data sources, and consideration of the
potential scope of the Universe of known chemical agents, the Work Group recommends EPA adopt a
principles-based approach, consistent with that described by the NRC. The goal of this approach is to
be inclusive of agents with demonstrated or potential occurrence in drinking water and of agents with
demonstrated or potential health effects.
-> EPA should use the inclusionary principles as the foundation for identifying the Chemical
CCL Universe. These principles are as follows:
• The Chemical CCL Universe should include those agents that have demonstrated or
potential occurrence in drinking water; or
• The Chemical CCL Universe should include those agents that have demonstrated or
potential adverse health effects.
Ch. 4 - CCL Classification Approach for Chemical Contaminants
4-2
-------
CCL CP Work Group Report
This principles-based approach provides a process for defining the ,Chemical CCL Universe on
the basis of a set of fundamental premises regarding the nature of the agents that should be considered.
If an agent meets either of the principles identified above, it is sufficient to place the agent in the
Chemical CCL Universe.
The Work Group concluded that a principles-based approach would be most consistent with
NRC's recommendations, as it could: a) incorporate the NRC's recommendations for including agents
with demonstrated or potential occurrence in drinking water and those with demonstrated or potential
health effects; b) provide a framework to include (versus exclude) agentl; at the earliest stage of the
Chemical CCL Universe identification process; c) not limit the number md types of agents or data
sources that could be considered for inclusion in (he Chemical CCL Universe, now or in the future;
and d) be implemented using a data source compilation process.
4.1.3 Specific Work Group Recommendations
The approach recommended by the Work Group is inclusionarv with respect to agents that are
not robustly characterized, and considers them at an early stage in the COL selection or classification
process. The approach does not limit the number and types of agents or data sources that can be
considered for inclusion in the Chemical CCL Universe, and yet it acknowledges mat for new and
emerging agents, relevant data may not be readily available. Therefore, tike Work Group has also
included recommendations for surveillance and nomination processes as an integral part of the
recommended overall process to identify agents that may need additional research and data collection
to provide a means of characterizing potential harmful exposure in drinkijig water for these new and
emerging agents. (See also Chapter 2, sections 2.3.2 and 2.3.3.)
4.1.3.1 Data Source Compilation Approach
-> EPA should identify agents for consideration in the CCL Universe using a "data source
compilation" approach, which is the process of accessing discrete data sources to retrieve
various, unique sets of records with multiple selection criteria.
The Work Group conducted a review of the number and types of known agents and available
data sources, and identified alternative approaches as part of its deliberations during development of
recommendations for building the Chemical CCL Universe. In discussion papers that the Work Group
considered, the numbers and types of known, new and emerging agents wiae characterized within a
hierarchy of available data sources. (See for example, Discussion Drafts far the NDWAC CCL Work
Group: "Dimensioning the Chemical Universe," January 13, 2003; and "fop-Down Versus Bottom-
Up Database Approaches for Defining the CCL Universe," January 22,2033.) As a result of Work
Group discussions, two approaches were identified for further consideratiqn. One approach was the
process of reducing a large array of data sources to relevant subsets of recqrds ("reducing data sources
approach"). The second approach was the process of accessing discrete data sources to retrieve
various, unique sets of records with multiple selection criteria (the "data source compilation
approach"). The advantages and disadvantages of two alternative approaches were discussed by the
Work Group and are summarized in Tables 4.1 and 4.2, below.
Ch. 4-CCL Classification Approach for Chemical Contaminants
4-3
-------
CCL CP Work Group Report
Table 4.1 -Advantages and Disadvantages of the Data Source Compilation Approach
Advantages
Disadvantages
1. Relevance. Records are pre-screened for inclusion in
discrete databases on the basis of key attributes.
2, More robust search capabilities. Discrete databases
are typically designed for specialized searches.
3. More data per record. Economical
4. Logistical benefits. Potentially less cost per record,
for publicly available databases.
5. Modular approach possible; can merge orrecombine
multiple databases if elements are consistent
1. Biases. Screening criteria may not coincide with
user's goals.
2. Subjective interpretations of data elements may skew
results.
3. Compounds with known issues/data more likely to be
included than emerging contaminants.
4. Fewer records.
5. Synonyms, homologues and mixture difficulties.
Omissions and redundancies possible.
6. Certain discrete databases proprietary, accessible
only by subscription that could hinder transparency.
7. Database incompatibilities. Nomenclature and search
fields vary among databases.
8. Weak link issue. Recombined databases are only as
current and accurate as least robust sub-database
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-4
-------
CCL CP Work Group Report
Table 4.2 -Advantages and Disadvantages of the Reducing Data Sources Approach
Advantages
Disadvantages
•(.Comprehensive scope. Large databases
represent the most complete list of known universe
of chemicals.
2. Less bias Introduced. Elements are included
based on broader criteria.
3. Data currency and consistency. Large
databases such as CHEMLIST are expanded
frequently with new compounds, in a consistent
format
4. Unique substance identifiers. Can reduce
inconsistencies in nomenclature.
1. Logistical impracticality. High costs are involved in
searching large databases! with fees based on retrieval (e.g.,
$1 per substance retrievediin CAS; $3 with physical-
chemical properties search).
2. Fewer data. Generally, tie larger the database, the less
data elements per contaminant
3. Search constraints Large, general databases contain
fewer searchable fields than databases designed for
particular purposes.
4. Missing elements. Only Icnown compounds/microbes can
be listed; oversights still possible (e.g., emerging
contaminants, metabolites).
5. Lack of relevance. Large; databases may contain
elements not relevant to COL attributes (e.g., nudeotide
sequences, compounds in gcant volumes, insoluble
compounds).
6. Moving target Large database searches may not be
reproducible as data expand.
7. Large databases costly to,maintain, and to update
historical entries (e.g., compounds no longer in commercial
use, removed from regulatory lists, etc).
8. Cross-referencing hurdles; Unique identifiers (except
CASRN) may not be cornpat ble with those in other
databases.
9. Synonyms, homologues ahd mixture difficulties.
Omissions and redundancies possible.
The Work Group agreed that although the "reducing data sources" approach would include
emerging and new agents to some degree, it would present significant challenges in developing a
manageable Chemical CCL Universe of agents. The "reducing data sources" approach seemed more
difficult, because it would include very large numbers of agents with little relevance to the CCL (e.g.,
the Chemical Abstract Services lists more than 41 million protein and nucleic acid sequences). The
Work Group noted that such agents would likely have no health or occurrence data or information,
and the "reducing data sources'' approach would therefore likely require a significant effort to reMew
a considerable volume of irrelevant records.
The Work Group agreed that the "data source compilation approach" was logistically favorable
for identifying the Universe of known agents likely to affect drinking watej, even though such an
approach may have some disadvantages in identifying new and emerging Agents. Therefore,
consistent with the inclusionaiy principles, the Work Group agreed to recoinmend the "data source
compilation" approach coupled with surveillance and nomination processes (described in 4.1.3.2,
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-5
-------
CCL CP Work Group Report
below) to consider new and emerging agents as part of an integrated overall process for defining the
Chemical CCL Universe. The data source compilation with surveillance and nominations approach
should be more efficient at producing a Chemical CCL Universe relevant to the CCL. Furthermore,
this approach is more compatible with the Work Group recommendation that data sources for the
Chemical CCL Universe be identified on the basis of multiple selection criteria (further outlined in
sections 4.1.3.3 and 4.1.3.4, below). The overall approach is envisioned by the Work Group to have
sufficient breadth to include known as well as new and emerging agents, consistent with the NRC's
recommendations.
4.1.3.2 Supplemental Surveillance and Nomination Processes
-» The Work Group recommends the "data source compilation approach" be supplemented
with a combination of surveillance and nomination processes to provide timely
identification of new and emerging agents.
It is envisioned mat surveillance and nomination would be integral components of the CCL
process and not separate processes. As such, surveillance and nominations typically would provide an
alternative pathway for an agent to enter into the CCL evaluative process. This approach addresses the
inclusionary principles by identifying agents through the surveillance process that may be potential
drinking water contaminants, but have data gaps. These agents may be identified and placed in the
CCL Universe. Chapter 2 (sections 2.3.2 and 2.3.3) provides a morecomplete overview of the
surveillance and nomination processes and recommendations.
4.1.3.3 An Integrated Process for Addressing Known, New, and Emerging Agents
-> The Work Group recommends that EPA consider adopting a integrated process for
building the Chemical CCL Universe, to include the following:
• identification of a Chemical CCL Universe with known agents;
« implementation of a surveillance process for new and emerging agents;
• implementation of a nomination process for new and emerging agents; and,
• adoption of an accelerated process for agents as needed.
The Work Group recognized that, conceptually, the principles-based approach defined above has
adequate breadth to encompass the full range of known, new and emerging agents. For
implementation purposes, however, die approach must be tailored to each of the three types of
candidate agents/contaminants defined in Chapter 2. For these discussions, the three types of agents
are defined as follows.
• Known agents are physical, chemical, or biological substances that have been identified in
the technical literature and adequately characterized (e.g., occurrence or health effects) to
enable a judgment regarding their inclusion in the Chemical CCL Universe. These are CCL
candidates, -which, by definition, can be identified through analysis of existing data sources.
Potential data sources of known agents have been identified for consideration in the
Chemical CCL Universe, and the list continues to expand. (Analyses performed for the Work
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-6
-------
CCL CP Work Group Report
Group show how the data sources identified to date, numbering over 200, relate to the
examples cited in the NRC 's Tables 3-1 and3-5.)
" New agents are physical, chemical or biological substances taat are or may be newly
discovered or synthesized, for which little is known about then-potential occurrence or
adverse health effects. Identification of new agents is challenging in several respects. The
Work Group's analysis illustrated that the rate of synthesis and discovery of new agents is
prodigious. For example, an average of approximately 4,000 substances are assigned CAS
registry numbers daily. The majority of these substances have,little data beyond name and
structure, however. Most are composed of chemical sequences of biological macromolecules
and proteomic sequences, and are not true candidates for the Chemical CCL Universe. On
the other hand, some new agents do move into commercial production rapidly now, and
these may need to be identified as agents for the CCL because of their potential to
contaminate water in the future. The "data source compilation approach " alone, using data
sources of known agents, would not likely capture many of these substances.
• Emerging agents area subset of known physical, chemical, or, biological substances
previously evaluated as not requiring inclusion in the Chemicql CCL Universe, for which
information becomes available that heightens concern and triggers re-evaluation. This
group contains agents that were either: a) not included in the (^CL Universe; or b) agents
for which new information becomes available that may heighten concerns and trigger
additional review.
Identifying the Chemical CCL Universe with Known Agents. Data sources would be identified
that provide relevant information about known agents that may be potential drinking water
contaminants. Data from these sources would be accessed, using the datal source compilation
approach, to identify agents for the Chemical CCL Universe and to retriel/e novel sets of records with
multiple criteria. Recommendations 4.1.3.4 and 4.1.3.5, below, provide flirther discussion of
components of implementing this approach.
Surveillance Process for New and Emerging Agents. The Work Grpup recommends that EPA
establish a surveillance process to provide identification of new and emerging agents. For details of
these recommendations and the surveillance process refer to Chapter 2. Ii| short, EPA's surveillance
process should include: implementation of a proactive, ongoing process of communication with
stakeholder organizations to obtain information; enhanced coordhation wjithin EPA and with other
agencies; and strengthening the linkage of ongoing activities (ranging froija literature reviews to
liaisons with professional organizations) to the needs of the CCL process. In particular, the Work
Group recommends that EPA institute a regularly scheduled conference (e.g., biennial) on "Emerging
Issues in Drinking Water" as part of their research for the CCL process. Such a forum could provide
an efficient and transparent mechanism for stakeholders and professional groups to provide their
findings and concerns for emerging and new agents (see Chapter 2).
Nomination and Evaluation Process for New and Emerging Agent*. The Work Group also
recommends that EPA develop a nomination and evaluation process for now and emerging agents,to
enable agencies and interested stakeholders from public and private sectors: to nominate potential
contaminants for consideration in the CCL process. As noted, all of the surveillance activities should
serve to provide "nominations" of agents to add to the Chemical CCL Universe. However, as noted in
Chapter 2, the Work Group recommends that other opportunities for adding potential contaminants
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-7
-------
CCL CP Work Group Report
should be available during the CCL process. Key elements that would require further specification by
EPA include a proactive communications strategy (as noted for surveillance) and information and
documentation requirements for the evaluation process. (See Chapter 2 for more detailed discussion.)
Accelerated Process. As new agents are identified, or as new information becomes available,
there may be justification to accelerate their passage to the Chemical CCL Universe, from the
Universe to the PCCL, or from the PCCL to the CCL. EPA could, if me data warrant, consider these
contaminants on an accelerated basis. The Work Group recommends EPA develop a formal
accelerated ("fast track") process and ensure that the process is communicated before, or at the time
the Agency requests nominations from the public. The process should be open and transparent and be
consistent with the overall CCL screening and evaluation procedures (See Chapter 2.3.2.2 for
discussion.)
4.1.3.4 Chemical CCL Universe Identification Process for Retrieving Information and Data
-> The NDWAC Work Group recommends that the EPA adopt a three-stage process to
identify the Chemical CCL Universe:
• Identify and retrieve data (including lists of agents) from sources that have data and
information about occurrence of contaminants in drinking water or source water or about
health effects.
• Identify and retrieve information (including lists of agents) from sources that have a link
(pathway) to drinking water.
• Use other information sources, such as chemical properties, to address data gaps. Use
models or surrogate information to estimate potential occurrence or health effects.
To identify known agents it is necessary to apply the data inclusion principles to actual data
sources. The following guidelines are proposed for identifying data sources that would be used for
building the Chemical CCL Universe.
• If agents in a data source have a reasonable pathway (as identified by NRC) to drinking
water sources, the data source should be used for the Chemical CCL Universe.
• If a data source contains information in a medium (e. g., sediment) that has potential to
transport to water, the data source should be used for the Chemical CCL Universe.
• If there are multiple data sources for an agent, all the data sources will be used for the
Chemical CCL Universe, as needed
• // may be acceptable to model or estimate values for data elements that cannot be obtained
from data sources (i.e., to fill data gaps).
• For data sources that contain a mix of information about known agents, it is appropriate
and necessary to select only the information that meet the occurrence and health effects
principles appropriate for the Chemical CCL Universe (e.g., for data sources from OSHA or
NIOSH, data about ergonomic hazards would be screened out).
• Bibliographic data sources will primarily be used in the Surveillance and Nomination
Processes for new and emerging agents (and to fill data gaps for known contaminants).
(However, because of the lack of database-type sources for microbiological contaminants, it
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-8
-------
CCL CP Work Group Report
may be necessary to use bibliographic sources and primary li'erature to compile data and
information for microbes for the immediate future CCL needsi This could be true of other
specific issues, perhaps including studies documenting outbreaks. See Chapter 3 for microbe
discussion.)
The proposed process would start with sources that contain data about measured concentrations
or verified presence of known agents in drinking water or about human health effects. In addition,
other sources may also provide data or information to aid in the assessment. In this regard, the Work
Group recognized that EPA would need to use expert judgment to assess relevant information. (See
also Transparency discussion in Chapter 2.1.) The multi-step three-stage process is explained in more
detail below.
Stage (1) Identify and retrieve data (including lists of agents) from sources that have data and
information about occurrence of agents in drinking water or source water or about health effects.
These data sources would form the first entries to the list of agents in the Chemical CCL Universe.
Also, these sources would begin to popuhte the relevant data elements required for the process of
screening from the Chemical CCL Universe to the PCCL, for those agents. Additional data elements
would be retrieved, at the appropriate stage in the CCL process, to meet ,the requirements of the PCCL
to CCL classification process.
The Work Group recommends giving equal but separate consideration to occurrence and health
effects information. (See Table 4.3 for examples of occurrence and health effects data sources.)
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-9
-------
CCL CP Work Group Report
Table 4.3 - Examples of Occurrence and Health Effects Data Sources
Occurrence
• National Contaminant Occurrence Database (NCOD)
• EPA's (developing) Unregulated Contaminant Monitoring Regulation database (which will become part of the
NCOD)
• USGS's National Water Qualify Assessment program
Health Effects
• ATSDR Comprehensive Environmental Response, Compensation, and Liability Act (CERCLA) Priority Lists
• EPA's Health Advisory Tables
• EPA's Integrated Risk Information System
• Agency for Toxic Substances and Disease Registry (ATSDR) Minimal Risk Levels (MRLs)
• California's Office of Environmental Health Hazard Assessment (CA OEHHA) Toxicity Criteria Database (cancer
values only)
• World Health Organization (WHO) Drinking Water Quality Guidelines
* World Health Organization's Classification of Pesticides by Hazard (CPH) Database
• The International Agency for Research on Cancer (IARC) lists of carcinogens
• TERA's International Toxicity Estimates for Risk
• OSHAor NIOSH data on hazards of agents based on occupational exposures (chemical and microbial agents
only)
These sources may be redundant for identifying agents to add to the Chemical CCL Universe
(i.e., they will identify and contain data about many of the same contaminants), but each is expected to
add unique data elements. New agents encountered while compiling data from these sources will be
added to the Chemical CCL Universe. For example, in accessing 23 data rich sources to identify an
example Chemical CCL Universe and for evaluating a process to screen contaminants from that
Universe to the PCCL, about half (8,750) of the sum of contaminants from all sources (17,891) were
unique contaminants.
Stage (2) Identify and retrieve data (including lists of agents) and information from sources
that have a reasonable link (pathway) to drinking water concerns. This could include reviewing
sources such as the following:
• High Production Volume (HPV) Chemical Lists
• Toxics Release Inventory (TRI)
• High Production Volume Master Summary Table
• FDA 's Generally Recognized as Safe (GRAS) Notices
Ch. 4— CCL Classification Approach for Chemical Contaminants
4-10
-------
CCL CP Work Group Report
• National Sedimer, t Inventory (NSI). (Contaminants included in the NSI would be compared
with the list of contaminants comprising the Chemical CCL Universe. Any contaminants on
the NSI that were not already in the Chemical CCL Universe would then be added to the
listings in the Chemical CCL Universe. Additional data might be included if deemed relevant
to occurrence.)
• Lists ofpharmaceuticals and personal care products.
A similar approach might be used with air-deposition data sources. Some data sources with
purely ecological endpoints (e.g., AQUIRE) may or may not be appropriate and will require further
expert review to assess if relsvant information is available.
Stage (3) In a third stage, other information sources would be used to fill data gaps. For some
contaminants it may be necessary to use surrogate information, or to model or estimate potential
occurrence or health effect end points.
There are two types of data gaps; those for which information has not been generated and those
for which the information is available but has not been accessed. This distinction is important as EPA
endeavors to fill data gaps cost-efficiently through an iterative approach ,and considers available
options to address this need.
As an example, for some agents added from lists without chemical characteristics data, the
Chemical Abstract Service (CAS) Scientific and Technical Network (STN) databases could be used as
a supplementary source to fill information gaps for needed data elements^, such as solubility. QSAR
modeling is an example of information mat might be used to estimate waiter solubility.
EPA's Office for Prevention, Pesticides, and Toxic Substances (OP1TS) routinely uses
quantitative structure activity relationships (QSARs) (e.g., from models slich as EPIWIN) to fill data
gaps for new chemicals as part of the Pre-Manufacturing Notification (PfljlN) program (under the
Toxic Substances Control Act). (See Chapter 4.5 for further discussion of QSARs.)
The Chemical CCL Univsrse is not finished until all three stages are completed. In addition, the
Surveillance and Nomination processes may also add agents to the Chemical CCL Universe.
4.1.3.5 An Approach to Retrieving Data and Evaluating Data Sources
-> The Chemical CCL Universe should be identified using an adaptive approach to retrieving
data and evaluating data sources and data dements for use in th« screening and
classification steps.
The three-stage process recommended in Section 4,1.3.4 provides a smarting point for retrieving
data and information, evaluating data sources and data elements for use in the screening and
classification steps. This process can be repeated as needed to obtain additional information.
This injects an important measure of manageability into the identification of the Chemical CCL
Universe by combining the inclusionary principles with what can be accomplished given limited
resources. This will make it possible to efficiently and cost-effectively prioritize occurrence and health
effects data sources, avoid or remove redundancies in data, and provide intosrim evaluations to
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-11
-------
CCL CP Work Group Report
determine how the process is working. The Work Group recognizes that to proceed with the CCL in a
timely manner, the Agency should develop and implement the CCL schedule to include the optimal
degree of iteration in developing the Chemical CCL Universe, considering the time and effort required
to conduct all of the necessary steps to meet the overall CCL schedule.
4.1.3.6 Data Quality Principles Compatible with Inclusionary Principles
-> For the Chemical CCL Universe, the CCL NDWAC Work Group recommends that EPA
establish data quality approaches that do not require such a high threshold that they would
be contrary to the inclusionary principles.
Chapter 2.3.4 provides a detailed discussion of various issues related to data characterization and
quality. A few key points pertinent to the Chemical CCL Universe are reiterated here. Section
1412(b)(3)(A) of the Safe Drinking Water Act Amendments specifies that EPA shall '*use the best
available, peer-reviewed science and supporting studies conducted in accordance with sound and
objective scientific practices; and data collected by accepted methods or best available methods (if the
reliability of the method and the nature of the decision justifies the use of the data)." The Work Group
recognizes that there is a desire by all for use of the highest quality scientific data in developing and
implementing environmental policies. However, it is fundamental for the CCL process that, in
identifying the Chemical CCL Universe, a wide net be cast to comply with the inclusionary principles.
Therefore, on principle, me Work Group supports the use of high quality data, but recommends that
EPA establish data quality approaches that do not place too high a bar, which would be contrary to the
inclusionary principles.
At a minimum, a description of the origin of the data must be available, including nominal
information that reflects what is known about data quality, which could include:
• contact name;
• description of the data elements;
• how the data were obtained; and,
• meaning/illness and relevance of the data.
4.2 Process and Criteria for Screening Agents from the Chemical CCL Universe to
the PCCL
The previous section described how a Universe of chemical agents with potential or actual
occurrence in drinking water or potential or actual capacity to cause health effects in humans would be
constructed. This section presents the Work Group's recommendations on how to select from among
the agents included in the Chemical CCL Universe those that should be listed on the Preliminary
Contaminant Candidate List (PCCL). This intermediate step between the Universe and the CCL
would provide a much smaller set of contaminants for more thorough assessment for the CCL.
Figure 4.2 summarizes the actions EPA should implement to screen the Chemical CCL Universe
to select the PCCL, and locates these actions within the CCL process.
Ch. 4 - CCL Classification Approach for Chemical Contaminants
4-12
-------
CCL CP Work Group Report
Figure 4.2 •• Selecting the PCCLfrom the Chemical CCL Universe
STEP1
STEP 2
I CCL
Universe
'
J
1 PCCL
t
STEP 3 I Proposed CCL
4.2.1 Summary of the NRC Recommendations
The NRC recommended that a systematic, transparent, and scientifi(j-ally sound process-
combining expert judgment with well-conceived screening criteria that ctiiuld be rapidly and routinely
applied to a large Universe of agents - be developed to select contaminants from the Universe for the
PCCL. The NRC intended that the process be more inclusive than that ussd for the 1998 CCL and
specifically mat it avoid excluding contaminants simply because of a lack of data about their
occurrence in drinking water.
As shown by the shaded areas denoted by Roman numerals in Figur^ 4.3 (from page 82 of the
NRC report), the NRC recommended that the PCCL include contaminants with "demonstrated" or
"potential" occurrence in drinking water and "demonstrated" or "potential!" capacity to produce
adverse health effects in humans. The shaded areas of the diagram represent the NRC's priority
ranking of contaminants for inclusion on the PCCL.
Ch. 4- CCL Classification Approach for Chemical Contaminants
4-13
-------
CCL CP Work Group Report
Figure 4.3 - NRC's Diagram of the CCL Universe
1 Contaminants for which both health
effects and occurrence are
"demonstrated" (highest priority)
II Demonstrated health effects and
potential for occurrence
HI Demonstrated occurrence and
potential for health effects
IV Potential for health effects and
potential for occurrence (lowest
priority)
The Universe of Potential Drinking Water Contaminants
Contairfnantalhat Y Cortairinanttthat
ere
lo occur In drinking to cause advert*
tieaWi effects
Contaminants that haw
the pofentaf to occur In
drinking wate
Confaffllnanla that have
tha potanNf to cauaa
adverse ItaaMIi aflecta
The NRC concluded that screening criteria would need to distinguish between the effects that
would be considered to be "demonstrated" compared to "potential," but also that it would be
important to include contaminants for which data were limited. The NRC identified the need to
develop screening criteria as a key step. While the NRC did not develop such criteria, it did identify
data elements and metrics that could be used in the screening. The NRC discussed comparison of
values observed in water to values of concern for health effects and recommended consideration of
severity of health effects in the development of the PCCL. For potency, the data elements and metrics
identified by the NRC included data from human and animal studies and models. The NRC indicated
that data from human and whole animal studies should be considered to indicate demonstrated health
effects and mat data from other toxicological studies and experiments be considered to indicate
potential for health effects. For occurrence, the NRC identified observations in tap water,distribution
systems, or finished water to represent demonstrated occurrence. Data about source water and
watershed production or release of chemical agents, and physical properties (including persistence and
mobility in aquatic systems) would represent a potential for occurrence. The NRC also indicated that,
to screen initially for inclusion on the PCCL, aqueous solubility could be used as the sole metric.
4.2.2 Principles for Selecting Agents for a PCCL from the Chemical CCL Universe
The Work Group reviewed the NRC's proposed approach in light of the principles adopted for
the process as whole and findings by consultants to the Work Group about the likely availability of
data about demonstrated occurrence of chemical agents in drinking water.
An analysis presented to the Work Group in July 2003 found that, as predicted by the NRC, data
about demonstrated health effects and demonstrated occurrence were available for relatively few
t
Ch. 4 - CCL Classification Approach for Chemical Contaminants
4-14
-------
CCL CP Work Group Report
agents.6 To achieve the principles of inclusiveness and to develop a more systematic process for
assessment, the Work Group concluded that it would be important to develop an approach that would
be capable of assessing a large number of agents in the Chemical CCL Universe and that would treat
agents with different amounts of data available as similarly as possible. Th<; Work Group developed
principles to support this.
The Work Group proposes EPA develop a screening process that relies on widely available data
elements that reflect certain aspects of health effects and occurrence. The screening process is to be
designed so that values for data elements reflecting both health effects and occurrence would reach a
level of concern for an agent to be screened through to the PCCL. This is a key distinction from the
development of the Chemical CCL Universe, in which an agent is to be included if there are data
suggesting that either health effects OR occurrence may be of concern.
-> The Work Group recommends that the screening criteria and methods be:
• capable of assessing as many of the contaminants in the CCL Universe as possible, even
those with limited data;
• as insensitive as possible to data limitations;
• as simple as possible, to require fewer resources and less time;
• capable of identifying those contaminants of greatest significancefoffurther
consideration,- and,
• to the extent feasible in light of the significant differences in availability of data for
chemicals and microbes, as similar as possible to the microbial approach.
4.2.3 Workable Approach to Screening Using Widely Available Data Elements
The Work Group sought to develop a process for screening the large number of contaminants in
the Chemical CCL Universe for the PCCL.
To develop an approach that would be as simple as possible and allow for the assessment of the
largest possible number of agents from the Chemical CCL Universe, the Work Groiup discussed how
to identify the most essential characteristics to be used as the basis of selection of data elements for the
screening process. The Work Group concluded that it is not necessary to use data elements that reflect
all five of the attributes defined by the NRG The Work Group recognizes mat subsequent steps in the
process, including the classification from the PCCL to the CCL, are more likely to uye data elements
that reflect all five of the attributes.
During its investigations, and drawing upon work performed by its technical consultants, the
Work Group found that data about demonstrated occurrence of contaminants in drinking water would
be available for fewer than 3% of the agents likely to be included in the Universe (pxafrple CCL
Universe Data Set, July 2003 presentation to the Work Group). Consequent^, the Woik Group sought
* Example CCL Universe Data Set: Progress andReoomme,idaiioa. Presentation to NDWAC CCL Woik Group Washington, DC, Jufy 15,2003
Ch 4 - CCL Classification Approach to Chemical Contaminants 4-15
f
-------
t
CCL CP Work Group Report
to identify data elements that would be informative for screening but also available for as many of die
contaminants in the Universe as possible.
The Work Group also sought to identify data elements for health effects that would be most
informative for screening and available for as high a percentage of the agents in the Universe as
possible.
So, for both occurrence and health effects, the Work Group identified data elements that they
thought would reflect the most important characteristics of contaminants for screening and that would
be most widely available. In doing this, the Work Group did not intend to preclude or restrict EPA
from considering other types of data that may come to its attention or become available. The intent
was to provide a workable approach to screening that would not be hamstrung by foreseeable limits in
the availability of data.
-> The Work Group recommends that a limited set of data elements that are widely available
and that represent important characteristics of health effects and occurrence be used as the
basis of the screening to select contaminants from the Chemical CCL Unwerse.
When thinking about the characteristics related to health effects that would be most important to
consider in this first screening step, the Work Group concluded mat potency is the most important
attribute to consider when selecting contaminants from the Chemical CCL Universe for the PCCL.
The Work Group concluded that values for data elements for potency were likely to be available for
many agents. The Work Group did not concur with the NRC conclusion that it would be necessary to
also consider severity at this stage.
When thinking about the characteristics related to occurrence, the Work Group identified a set of
data elements for screening that might be described as representing the potential for exposure. The
data elements are intended to reflect two traits: (a) persistence in the environment or in drinking water
distribution systems and (b) the potential for contaminants to be present in drinking water. (The latter
has also been referred to as a "source screen.") The Work Group recognizes that the selected data
elements represent surrogates for the traits of interest and are proposed for use because they are
expected to be relatively widely available. Other attributes and characteristics were also discussed.
The attribute of "magnitude" was considered but not selected as a focus because it requires estimates
of concentrations in water that are not likely to be available.
-> The Work Group recommends that widely available data elements representing potency be
used to reflect health effects. The Work Group recommends that widely available data
elements for occurrence that reflect persistence and likelihood that agents will get into
drinking water be used to reflect occurrence.
4.2.3.1 Data Elements for Potency
-> For potency, the Work Group recommends that data elements reflecting chronic effects,
cancer, and acute effects be considered. The data elements that represent the lowest doses
at which adverse effects occur are recommended to be used as the basis of the screening for
potency. In addition, for carcinogens, data elements that reflect published cancer hazard
classification descriptors or cancer slope factors are recommended.
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-16
-------
CCL CP Work Group Report
In general, these would include lowest observed adverse effect levels, (LOAELs) for chemicals
tested for non-cancer effects. For cancer, equivalent metrics for cancer effects or cancer classifications
such as those adopted by the EPA or International Agency for Research or; Cancer (IARC) or the
National Toxicology Program (NTP) are recommended to be used for screening at this stage. For
acute effects, the lowest lethal closes or LD50s in chemicals tested for mortality may be appropriate.
One of the critical issues in developing an approach to screening chemicals from the Chemical
CCL Universe for the PCCL is to decide which data elements to use. Many data elements related to
potency were identified during the discussions of the Work Group (see July 2003 Attribute Element
Crosswalk).
The Work Group sought to select a set of data elements that would be as widely available as
possible and that could be estimated using models such as QSARs if values bused on experiments
were not available. This was in keeping with the overall principle of adopting approaches that allowed
for the assessment of as many contaminants as possible. The Work Group selected data elements, in
general, that reflect observed value:: as directly as possible and that do not reflect changes or
adjustments for uncertainty and othsr such factors.
The Work Group has sought to include data elements mat will be representative of the major
types of health effects of concern: acute effects, non-cancer chronic effects, and cancer. The Work
Group has identified possible data elements for each of these three categories. Wjhen more than one
value is available for these data elements, the Work Group recommends that the lowest dose value be
selected.
The LOAEL is a widely reported result for chemical contaminants that are evaluated for non-
cancer effects. The LOAEL is the lowest dose at which adverse effects are shown. It may also be
appropriate to use an LD» (a measure of acute effects) when LOAELs are not available. Both
LOAELs and LDsos can be estimated using QSAR methods, and this is another reason the Work
Group recommends the use of these data elements.
For carcinogens, the type of toxicity value typically available is not analogous to a LOAEL.
(This is because the toxicity of carcinogens typically is represented in terms of a unit risk or cancer
slope value that reflects how quickly risk increases with dose.) EPA will need to consider carefully
how to address mis. One option is to generate (based on a review of the literature) values comparable
to LOAELs for carcinogens. These would be the lowest doses that cause adverse cancer-related
effects. Another option is to use the cancer slope factors that are typically used to represent the
potency of carcinogens. A third option, which the Work Group specifically recommends, is to use the
cancer classifications generated by EPA or other organizations such as IARC or the NTP; those
chemicals that are listed using descriptors such as "known," "probable," "likely." or "possible"
carcinogen would be considered to have a health effects value of concern.
For acute effects, the lowest dose that causes mortality (LDlo) or the dose thatcauscs mortality in
50% of exposed animals (LDso) would be used for chemicals for which only tests for mortality are
available.
The Work Group has not recommended using any data elements drawn from the types of assays
that the NRC considered to represent the "potential" for health effects. However, estimate!? obtained
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-17
f
-------
t
CCL CP Work Group Report
from QSAR models would represent values that would be considered to represent the potential for
health effects.
-> The Work Group recommends that, for potency, only one data element would be selected
for screening each contaminant for health effects. It would be the single data element with
the value that is most likely to lead to inclusion of the contaminant on the PCCL. (This
might also be called the most health-conservative value.)
The data elements described measure different things. Consequently, each data element would be
assessed separately. The one data element most likely to result in the contaminant being placed on the
PCCL (the one that is of the great health concern) would be the one to be used in the screening
process.
4.2.3.2 Data Elements for Occurrence
-> As with the screening for health effects, The Work Group recommends that the EPA
develop an approach for assessing potential exposure that uses a limited set of widely
available data elements. The data elements that are thought to be most widely available
are: 1) those related to the tendency of an agent to persist in the environment or in the
water distribution system; and, 2) those that reflect an agent's potential for occurrence in
drinking water based on information characterizing its source(s). Information about
demonstrated occurrence of contaminants in drinking water should also be used, where
available.
Due to data limitations and the interest in assessing a large number of contaminants, (he Work
Group concluded that it is important to develop a workable way to screen contaminants with regard to
occurrence using data elements that might be viewed as surrogates for potential exposure. The Work
Group recommends that EPA develop a screen based primarily on data elements that reflect the
potential for agents to reach drinking water and the persistence of agents in the environment, including
drinking water systems.
Persistence is included as a key element because a compound that is persistent, if released, could
eventually contaminate drinking water. Even though the time scale may be long, the potential to
persist should be addressed in this screening process. Conversely, a compound that is not persistent is
not likely to remain available in drinking water long enough to pose a concern. The Work Group
considers persistence as a characteristic that is best represented at this time as either "persistent" or
"not persistent" and not through any kind of continuous metric.
The potential for a chemical to reach drinking water as a result of being produced or released to
the environment is important to consider along with persistence, because even a highly persistent
compound will not contaminate drinking water if it is never released to me environment. The potential
for inadvertent releases should also be considered as should the capacity or propensity of a compound
to migrate. A combination of data elements to assess the potential to reach drinking water might be
called a "source screen." The Work Group recommends that EPA investigate developing a source
screen that could readily be implemented using data sources such as production volumes, amounts
released, use in disinfection processes and other such information that would contribute to the
potential for an agent to occur in drinking water. If this is fully addressed in the process of assembling
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-18
-------
CCL CP Work Group Report
the Chemical CCL Universe, it may need less attention during the screening from the Universe to the
PCCL.
The NRC specifically recommended use of solubility in screening agents at this stage, and the
Work Group discussed this issue at some length. Solubility is the equilibrium concentration of a
compound in water, often expressed in the form of milligrams per liter. Solubility is available for
many compounds and can also be estimated using quantitative structure-activity relationship models
(QSARs).
The Work Group concluded that it would not be appropriate to use solubility as a screening
criterion at this stage because it may screen out contaminants that occur widely, but at low
concentrations, or that occur as particulates in suspension. Also, most chemicals occur in solution well
below their equilibrium solubility concentrations, so solubility is a poor suiTogate for occurrence.
However, under certain assumptions, solubility is an indicator of the upper limit of a compound mat is
likely to occur in solution and may be useful for priority setting at a later stage in the process.
Several other data element;; for chemicals also have been considered. Log K<,w was rejected
because the group felt that it would not add any new information important at this phase of the
screening. (K<,w is the octanol-water partition coefficient, a measure of the tendency of a dissolved
compound to move out of water into a nonpolar material, also used as an injiex of the tendency to
bioaccumulate). Henry's Law Constant was rejected because it may not be applicable for ground
water and may be captured in persistence. (Henry's Law Constant is a measure of how much of a
substance stays in the water compared to how much evaporates or volatilize; into the air.) In surface
waters, which are exposed to the air, compounds that volatilize will not be persistent, simply because
they evaporate into the air and are not found in the water. However, in ground water, the same
compounds may be persistent because they cannot evaporate. TCE, a common solvent, is a good
example.
The Work Group focused on use of contaminant characteristics because'they concluded that data
about such characteristics is what is likely to be available. However, the Work Group believes it also
fully appropriate for EPA to consider data about demonstrated occurrence of contaminants in drinking
water or in ambient water bodies in addition to the screening approach discussed. Measurement of
contaminants in water is more direct evidence of their occurrence man persistence or source
indicators. However, the Work Group does not want lack of such data to create a barrier to full
consideration of a contaminant for the PCCL and so has not emphasized such, data in mis screening
approach.
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-19
f
-------
CCL CP Work Group Report
Table 4.4 - Possible Data Elements for Selecting Universe Contaminants for the PCCL
Characteristic
Potency
Exposure
Data Elements
Details
LOAEL; LDIo; LDW; analogues Numeric value of potency:
for carcinogens; carcinogen mg/kg or mole/kg
classifications
Persistence
Measured occurrence
HalfJife or other measure; if not
available, then the contaminant is
assumed to be persistent
Actual measurements of a
contaminant in drinting water
Potential to reach drinking
water
Quantities produced or released;
on production or release lists
One additional consideration is that there may be contaminants that reach drinking water not by
being directly dissolved into water but by being adhered or adsorbed onto particles. Such compounds
tend not to be soluble. Such compounds should, however, be included on the PCCL. If there were
such compounds that were not identified through the data elements discussed here, it may be
appropriate for them to be added to the PCCL.
4.2.4 Screening for Both Health Effects and Occurrence
-> The Work Group recommends that the contaminants that are screened to the PCCL be
those for which values for data elements for both health effects and occurrence reach a level
of concern, based on the screening process, for inclusion on the PCCL. Generally, neither
alone would be sufficient under this screening process.
However, in keeping with (he recommendations of the NRC that contaminants on the PCCL be
those with either demonstrated or potential health effects and occurrence the Work Group recognizes
that there will also be cases for which a different approach is appropriate. This approach is not
intended to preclude the addition to the PCCL of groups of contaminants of particular concern where
expert judgment concludes that they should be included. This approach is intended, instead, to provide
a feasible way to screen compounds for which the recommended data elements may not be available.
For example, the Work Group recognizes that disinfection by-products are formed and water
treatment chemicals are introduced to the drinking water system and may be a concern even if they are
not highly persistent
-> If an "On or Off' approach is used in the screening, the Work Group recommends that
contaminants that have the highest values for data elements related io either health effects
or occurrence but that do not make it onto the PCCL, be subjected to further review to see
whether there is cause for concern in drinking water. This supplemental assessment should
be done for very high potency values that score too low on exposure, and for very high
exposure values that score too low on potency. Expert judgment may conclude that some of
these compounds belong on the PCCL even if they fail the criteria for the screening.
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-20
-------
CCL CP Work Group Report
The Work Group recognizes that there are likely to be contaminants; that are highly toxic but have
low potential for exposure or that have high potential for exposure but do not appear to be highly
toxic. Some of these contaminants may pose a concern even if they do not pass the screening process.
The Work Group recommends that EPA use a supplemental assessment tjo identify such agents that
should be further investigated and perhaps should be included on the PCCL.
-> The Work Group recommends that EPA allow expert judgment to be used to correct
mistakes or oversights that will arise from this relatively simple process. It will likely be
appropriate to add some number of contaminants to the PCCL that pose a concern but
that do not fit the process outlined. The Work Group recognizes -hat unforeseen
circumstances will arise, and recommends that EPA allow for supplemental consideration
to address them.
The Work Group concurs with the NRC's view that expert judgment willneed to be used in
conjunction with the screening. There are likely to be contaminants that do not fit the screening
criteria that should be included on the PCCL. EPA should provide for expert review and assessment to
allow for the inclusion of such additional compounds when warranted. (See Chapter 2.3.1.)
4.2.5 Tagging Sources of Values for Data Elements and Implications
-> The Work Group recommends that "tags" be used to retain information about the sources
of values used in the screening process and that this be done in sudd a way as to preserve
this information for later steps in the process. The tags should identify values derived from
models such as QSARs. Tide tags should also identify what combination of "demonstrated"
and "potential'' values for health effects and occurrence were used:
Under the process proposed by the Work Group, both measured and estimated values may be
used for the data elements ultimately selected for the PCCL.
Measurements refer to data obtained from experiments, studies, surveys, or from environmental
sampling and analysis. This might include, for example, measurements of th<: agent in water
(occurrence); me results of epidemiological studies relating the presence of aji agent in water and the
appearance of effects; measurements of a NOAEL or LOAEL (health effects!) by die oral route of
exposure.
Estimates may be generated by appropriate and credible models (including quantitative structure-
activity relationship models, or QSARs, if consistent with the policy for acceptable QSARs addressed
in Chapter 2.3.5). Estimates may be derived by analogy, in comparing compounds without data to
similar compounds with data, using expert judgment or some other estimation process.
The NRC concluded lhat the type of data used in the assessment of contaminants for the PCCL
should contribute to the priority of the contaminants on me PCCL and that contaminants for which
mere are demonstrated health effects and demonstrated occurrence would hav^ a higher priority than
those where health effects or occurrence (or both) are considered to be potentij J.
Some members of the NDWAC Work Group believe that measured values have higher quality
than estimates. Other members of the Work Group believe that this will not aft/ays be true and that
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-21
-------
CCL CP Work Group Report
estimates based on good models and robust inputs may be of better quality than measurements based
on a small or biased selection of values. Work Group members agreed, however, that tagging data
elements used in screening with regard to the source of the data, and recording critical information
about the sources of data, will allow appropriate decisions to be made later in the process. It is
important to include information about data elements such as whether the values are measured or
estimated.
In keeping with the inclusionary principles adopted by the Work Group, the Work Group does
not recommend prioritizing contaminants on the PCCL itself. However, the Work Group does see
value in retaining information about the types of data used in the screening process. It is important to
note that the Work Group's recommendation differs from that of the NRC with respect to prioritizing
contaminants on the PCCL. Where the numbered regions on the NRC's Venn diagram of the CCL
Universe correspond to priority levels for contaminants incbded in the PCCL, the Work Group
proposes using these tags not to prioritize, but to track origins of the data used.
The challenge of the proposed PCCL process is to develop a means to conduct the screening
process as efficiently as possible so that it can be applied to screen the large number of agents in the
Chemical CCL Universe using a manageable approach in all or most cases. Screening will be easiest
if it is possible to identify the acceptable data elements and apply clear criteria for movement to the
PCCL. It is likely that significant work will be needed on existing data sets to understand and
standardize these before an efficient search process can be applied.
The Work Group feels that tagging contaminants on the PCCL to indicate underlying categories
of data or information is also valuable for the subsequent process of moving from the PCCL to the
CCL.
4.2.6 Approaches to Classifying Agents on the Chemical CCL Universe to the PCCL
-» The Work Group recommends that, as the Agency develops approaches to screen chemical
agents from the Universe to the PCCL, it should consider a range of options both for using
data element values in the screening process and for establishing appropriate screening
criteria to select PCCL contaminants. The screening method developed should be practical
and transparent, and should efficiently screen the Universe to the PCCL. The method
should also employ a level of precision that appropriately characterizes the nature and type
of information used. While the Work Group discussed several options and identified their
advantages and disadvantages, it does not recommend a single approach.
The NDWAC Work Group examined several important technical issues relating to the specific
approaches for screening the Chemical CCL Universe to select those agents to be placed on the
PCCL. These technical issues fall into three categories:
1) Issues related to assigning specific values for the data elements used in the screening
process
2) Issues related to the basis for establishing the appropriate screening criteria / decision rules
3) Issues related to the form of the screening criteria / decision rules to be applied
Ch4- CCL Classification Approach to Chemical Contaminants
4-22
-------
CCL CP Work Group Report
42.6.1 Assigning Specific Values to Data Elements Used in the Screening Process
Sections 4.2.4 and 4.2.5 described the various data elements related to potency and occurrence
that the Work Group recommends that EPA consider for screening the Chemical CCL Universe.
These are also summarized in Table 4.4. These data elements encompass .several different types of
specific measurements that arc expressed in different ways.
Having data elements expressed in a variety of different ways raises some implementation
challenges. The Work Group considered the advantages and disadvantages of using the actual
measurements or estimates provided for each data element versus using categorical values based upon
the measurements or estimates for the data elements. The Work Group recognized that transforming
measurements and estimates into ordered categorical data will, in many caj:es, result in a substartial
"loss of information." Avoiding such "loss of information" is important to ivoid a potential decrease
in the sensitivity of the screening process to properly differentiate among agents. As discussed further
below, EPA will need to develop decision rubs for determining which agents on the universe should
go to the PCCL and which should not. Transforming the underlying data to categorical values could
result in grouping some agents together and treating them the same even the ugh the underlying data
for them indicate there are differences among them. Of particular concern would be the categorical
values that fall near the decision boundary line that could result in decisions ,to either include or
exclude agents from the PCCL mat would be different from decisions that might be made on the basis
of the actual data that show there are differences among those similarly categorized agents.
Using the actual measurements or estimates for the data elements in the screening process
whenever possible will avoid this "loss of information" problem and the implications as noted. At the
same time, the Work Group also recognizes that there will be some circumstances where the use of
categories for data elements may be necessary. Ideally, the Agency will develop and implement a
screening process that could accommodate both actual data and categorical data. Where it is necessary
to use categorical data, it should be approached in a manner that uses a sufficient number of categories
to minimize the loss of information, while also not incorporating an excessive number of categories
that would inappropriately imply more precision in the underlying data than is appropriate.
4.2.6.2 Basis for Establishing the Screening Criteria / Decision Rules
There are a variety of ways to defhe the screening criteria or decision rules for determining
which agents in the Chemical CCL Universe should be placed on the PCCL. An important set of
issues that the Work Group considered are those related to how those decision rules should be
established. There are two major options in this regard.
The first major option is to identify levels of concern for data elements used in the screening that
have been developed outside the PCCL process, using expert judgment. Such authoritative levels of
concern could come from standard references or similar sources.
The second major option is to use the observed values for the data elements in the selection of
thresholds for health effects and occurrence that will cause those contaminants most likely to be of
concern to move to the PCCL. The thresholds selected could reflect the number of contaminants that
are sought for the PCCL as well as the appropriate weighting of the values for data elements for health
effects and occurrence.
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-23
-------
CCL CP Work Group Report
The Work Group did not address how to select threshold values for either method.
There are advantages and limitations to these options. Establishing decision rules in advance is
transparent, and can be implemented easily by those with limited technical training once the rules
have been established. Moreover, using generally recognized criteria for the levels of concern may
add credibility. One potential limitation of this approach is the potential for arriving at either an
inappropriately small or large number of contaminants for the PCCL
The second alternative also is transparent, in that one can draw a simple graph showing how a
proposed approach would divide the data into two groups- one representing those that would be
included on die PCCL and the other those that would not. An example is shown in Figure 4.4. One
could use a variety of statistical methods to identify groups with similar characteristics, and then
define rules for thresholds mat would distinguish between them. It would also be relatively easy to
conduct a sensitivity analysis of the various thresholds that could be applied to the data to see how
they would differ in terms of the chemical compounds to be placed on the PCCL. It also is possible to
adjust thresholds to achieve a PCCL of an appropriate size.
Ch4 - CCL Classification Approach to Chemical Contaminants
4-24
-------
CCL CP Work Group Report
A
A A
A A
Health Effects
(a) The Data
A
A
A
A
A
A
^ A A
^ A A
&
A A *
A A
HeaRh Effects
(b) Thresholds
1
Health Effects
(e) Linear Rule
Health Effects
(f) Curvilinear Rule Emphasizing Extremes
A A
Health Effects
(c) Thresholds: Occurrence and Exposure
A A
A A &
Health Effects
(d) Thresholds: Occurrence or Exposure
Health Effects
(g) Curvilinear Rule Excluding Singular Extremes
Figure 4.4. Examples of Alternative Forms
of Screening Criteria / Decision Rules
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-25
9
-------
CCL CP Work Group Report
Figure 4.4 [a] shows a hypothetical data set, with several compounds plotted on two axes, health
effects (on the x-axis) and occurrence (on the y-axis).
Figure 4.4 [b-d] shows the application of relatively simple threshold decision criteria. As shown
in Figure 4.4 [b], the threshold for occurrence would be reflected by the horizontal line and the
threshold for potency by die vertical line. Based on die four different regions defined by the specific
levels of concern, one could determine which compounds get classified into the PCCL and which do
not. For example, one could establish that only those that have both high health concern and high
occurrence are classified onto the PCCL. Those compounds are in the upper right area of the graph
defined by the thick lines (see Figure 4.4[c]). Alternatively, one could say that compounds that have
either high occurrence (without consideration of potency) or high potency (without consideration of
occurrence), as well as those that are moderate to high in both occurrence and potency, should be
classified onto the PCCL (Figure 4.4[d]). For such a rule, those compounds that fall in the lower left
quadrant of the graph, defined by the thick lines, are the only ones that would not be classified in the
PCCL.
It is also possible to apply rule-based combinations of data elements that reflect more complex
sets of interactions between health effects and occurrence than are reflected in the simpler threshold
versions shown above. Figure 4.4[e] shows an example of a linear rule-based combination of data
elements. This example implies the inclusion of compounds with high health effects and low-to-
moderate occurrence, and those with high occurrence and low-to-moderate health effects, as well as
those with both moderate-to-high health effects and moderate-to-high occurrence. The slope of the
line determines the relative importance of health effects and potency interactions. Figures 4.4[fJ and
4.4[g] show the application of still more complex curvilinear rule-based combinations of data
elements. These are applied in a similar manner to the linear rule, but are more flexible with respect to
what combinations of data elements result in an agent being included or excluded from the PCCL.
4.3 Use of Attributes to Classify Chemical Contaminants
t
4.3.1 Introduction
Chapter 5 of this report considers the several types of structured decision-making models that
could be used by EPA, in conjunction with expert judgment, to determine which chemical
contaminants on the PCCL are most appropriately moved forward to the CCL based on their known
or potential health effects and on their known or potential occurrence in drinking water Specific
measures related to those health effects and occurrence indicators- that is, the actual values of the
data for the various contaminants, or attribute scores based on the actual values of the data - provide
the inputs to those decision-making models.
The NRC developed a set of five specific attributes- characteristics of a contaminant that
contribute to die likelihood it could occur in drinking water at levels and frequencies that pose a public
health risk - mat they believed constituted a reasonable starting point for EPA to consider
• Potency and Severity as key predictive attributes for health effects
• Prevalence and Magnitude as key predictive attributes for occurrence
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-26
-------
CCL CP Work Group Report
• And Persistence/Mobility, as characteristics that might predict possible occurrence if direct
measures of Prevalence and Magnitude were not available
As envisioned by the NRC, these five attributes were applicable to both chemical and microbial
contaminants, though die NRC recognized mat die types of measures and information used to quantify
the attributes would differ for these two categories of contaminants. (See Chapter 2, section 2.2.2.3 for
a more detailed description and discussion of these attributes.)
Chapter 5.1 provides a detailed discussion of the Work Group's consideration of the alternatives
of using actual values for data elements versus scores based on those values to quantify the attributes
so they can be used as inputs to the classification models. This section of the report addresses the
attributes for chemicals, focusing on the types of information and data that are expected to be used to
quantify them.
4.3.2 Use of Data Elements to Quantify Chemical Attributes
As mentioned in Chapter 2, the characteristic represented'by an attribute often can be measured
or described by more than one type of data element. For example, for a given chemical compound, the
attribute of potency may be measured by a Reference Dose, a Cancer Risk Factor, or an LD50. In
deliberating on me use of data elements to quantify chemical attributes, therefore, the Work Group
considered these questions:
• What data elements should be used as measures to quantify those attributes?
• What should the hierarchy (preferences) among data elements for a given attribute be?
• How should quantitative values for a given attribute obtained from using different data
elements be normalized for ensuring consistency in their use in the classification models?
During the Work Group deliberations, EPA staff explored the attribute scoring alternative and
developed draft protocols for scoring the five attributes for chemical contaminants, including rules
delineating the hierarchy of data elements that should be used for scoring each attribute and the rules
(or algorithms) for assigning a specific attribute score based on die specific data element or
information item used for scoring. (See Appendix C.) To further help the Work Group gam insights
into the practical aspects of using the five attributes and the draft scoring protocols for chemical
contaminants in the CCL process, an Attribute Scoring Workshop was held in October 2003. The
scoring protocols that were used, the results obtained from applying them in the workshop, and the
observations of workshop participants, were presented to and considered by the Work Group in
developing the recommendations presented here.
4.3.3 NDWAC Work Group Recommendations
-> If attribute scoring is conducted, the scoring protocols for chemical contaminants should
accommodate multiple data sources and a variety of data elements that may be available to
score contaminants on the PCCL.
Ch4- CCL Classification Approach to Chemical Contaminants
4-27
-------
f
CCL CP Work Group Report
-> If attribute scoring is conducted, the scoring across the different types of data elements for
a given attribute should be consistent and allow for a meaningful comparison among scored
PCCL contaminants.
As discussed in Chapter 3, the Work Group did not reach a conclusion regarding whether to
recommend the use of actual values of data ebments or scores based on those values to serve as inputs
to the classification models. If scoring of attributes is carried out, the EPA should develop clear,
pragmatic chemical attribute scoring protocols that can accommodate the anticipated variety of data
sources and elements to be used in this process for chemical contaminants. The Work Group
anticipates that there will be instances where more than one data source provides information on a
contaminant. There also will be instances where more than one type of data element is available for a
contaminant. These two components of attribute scoring, the data source hierarchy and data element
hierarchy, should be evaluated simultaneously for each attribute being scored.
Indeed, whether EPA uses the actual valies for data elements or attribute scores based on those
values, a data source hierarchy should be developed to provide a descending ranked list of data
sources that begins with the more trusted data sources based on pie-determined criteria such as the
standard of peer review that is conducted on data prior to submittal. Similarly, the data element
hierarchy should provide a descending ranked list of data elements that provides clear instruction
about which data element should be used to quantify a contaminant when more than one data element
is available. For example, when characterizing the potency of a contaminant the Reference Dose may
be preferred to an experimentally derived Lethal Dose. The data element hierarchy for occurrence
attributes may descend from measured concentrations or presence to production and release estimates.
Following mis logic, if the preferred data element according to the hierarchy is unavailable, then the
next available data element in the hierarchy should be used to quantify the attribute. There also will be
instances when a decision has to be made regarding the use of a preferred data element versus a
preferred data source.
When attribute scoring is carried out, the Work Group anticipates mat different contaminants'
attribute scores will be based on different data elements for the same attribute. The scoring protocols
should ensure that the scales for assigning scores could produce an attribute score that would carry die
contaminant to die CCL regardless of which data element was used to generate the score for that
contaminant. That is, the attribute score should be a function of each data element for the specific
contaminant's characteristics that suggests a level of scrutiny or concern, and not a function of each
data element's position in the hierarchy.
It is important that the range of scores that a contaminant could receive for a particular attribute is
the same regardless of which data element is used to generate that score. For example, a contaminant
mat is scored using a data element from the low end of the data element hierarchy should be able to
receive any score in the range- including the highest possible attribute score- if the specific
information provided by that data element warrant it. This is to ensure that even those contaminants
lacking data for one of the preferred data elements in the hierarchy still have an opportunity to receive
a high attribute score and move forward in the process based on information that is provided by a less
preferred data element in me hierarchy.
Ch 4 - CCL Classification Approach to Chemical Contaminants
4-28
-------
CCL CP Work Group Report
Chapter 5
Moving from the PCCL onto the CCL
The NRC Committee recommended a three-step approach7 for identifying the list of
possible contaminants for future CCLs. This chapter discusses considerations for the third step:
quantifying attributes to describe contaminant risk and the use of the selected attributes in a
structured decision approach to select the CCL from the PCCL. Section 5.1 discusses options and
recommendations for attribute scoring. Section 5.2 presents an overview of various classification
approaches and the Work Group's consideration of these approaches. Section 5.3 presents the
Work Group's recommended approach to selecting the CCL, using a structured decision-making
tool. Section 5.4 discusses issues to consider in the selection of a "training data set" used to
inform this decision-making tool, or algorithm. Section 5.5 concludes the report with an
overview of documentation required to support key components of the CCL Classification
Process.
Figure 5.1 summarizes the recommended actions that EPA will need to further develop to
classify the PCCL, conduct expert reviews, and select the CCL.
Figure 5.1 Classifying and Selecting the CCL
STEP 1
CCL
STEP
i
PCCL
STEPq Proposed |yX
Selecting the ecu
' The NRC rqwrt afas to a "two-s«q)" approach because it does not comt the identification of the "liiiverae" as a first step. The NDWAC Woik
Group, having elaborated on the processes for identifying boti the Chemical and Microbi&l CCL Universes, considers this & first step; hence the NRC's "two-
step" process is referred to in this report as a "three-step" proc ESS
Ch. 5-Movingfrom the PCCL onto the CCL
5-1
t
-------
CCL CP Work Group Report
5.1 Quantifying Attributes for Use as Inputs to Classification Models
In its report, the NRC suggested a set of five attributes- two addressing health effects, three
addressing occurrence. (General definitions provided by the NRC for each of these five attributes are
displayed in the text box in Chapter 2, Exhibit 2.1.) The NRC noted in its report that it was neither
explicitly nor implicitly recommending that these specific five attributes be used by EPA, nor that
there necessarily be exactly five attributes. These five were offered by NRC as a starting point for
developing the appropriate attributes to be used in the CCL process.
Attributes are also discussed in Chapters 3 (for microbes) and Chapter 4 (for chemicals). The
remainder of this section discusses alternatives for quantifying the attributes so that they can be used
as inputs to the classification models for the PCCL to CCL stage of the process.
5.1.1 The Alternatives: Using Actual Data Values versus Attribute Scoring
The latter portion of this chapter examines the several types of structured decision-making
models that could be used by EPA, in conjunction with expert judgment, to determine which chemical
and microbiological contaminants on the PCCL are most appropriately placed on the CCL based on
their known or potential health effects and on their known or potential occurrence in drinking water.
These various models require as inputs some specific measures related to those health effects and
occurrence indicators. Attributes - or more specifically, either the actual values or scores generated
from the actual values for data elements used to characterize those attributes- can serve as the inputs
for these models.
Attributes are defined in the context of the CCL process as characteristics of a contaminant that
contribute to the likelihood that it could occur in drinking water at levels and frequencies that pose a
public health risk. As noted in Chapter 2, there are various types of measures or descriptors that may
be used as a means for quantifying the attributes. These measures and descriptors are referred to as
data elements.
The NRC developed, implemented and presented results for some attribute scoring metrics for
the five attributes that have been discussed previously in this report (see Chapter 2). The NRC
indicated that me attribute scoring metrics it explored were to be viewed as illustrative only. In
particular, the consideration of attribute scoring in the NRC report involved the use only of categorical
scores. That is, attribute scores explored by the NRC were limited to specific integer values with a
specified range such as 0 through 10, 1 through 10, or I through 3 depending upon the particular
attribute.
The NDWAC Work Group considered both using the actual values directly to quantify the
attributes and two alternatives to generating attribute scores. The attribute scoring alternatives that
NDWAC considered were:
1) Use a set of rales or an algorithm to convert the quantitative value or measurement provided
by the data element to a normalized numerical score, with a specified range (e.g., 0-10) for
Ch. 5-Movingfrom the PCCL onto the CCL
5-2
-------
CCL CP Work Group Report
that attribute, allowing die scores to be continuous values within that range (e.g., 8.26) and not
just integer values (e.g., 8).
2) Use a set of rules or an algorithm to convert the quantitative value or measurement provided
by the data element to a categorical score, with a specified range (e.g., 0-10) for that
attribute, but limit the resulting scores to specific integer values only (e.g., 8). (This is
essentially the approach used in the assessment of attribute scoring conducted by the NRC.)
An important advantage of using actual values to quantify the attributes is that it is, arguably, the
most direct reflection of the underlying data with no "loss of information" that could occur with other
methods to develop attribute scores. It should be noted mat the direct use of the underlying data is
what is being recommended for the occurrence and health effects measures in the screening process
for going from the Universe to (he PCCL.
Some technical challenges can arise, however, with this approach. For example, the data element
available for characterizing the .attribute may not itself be a specific numerical value. This can occur,
for example, in the scoring of severity where me data elements reflect qualitative descriptors of the
type of adverse effect(s) caused by the contaminant; in the scoring of magnitude based on
production/import volumes expressed in gross terms of "greater than or less than" some poundage; or
in the case of persistence where biodegradauon rates may be reported only in broad terms such as
weeks, months, or as "recalcitrant." Directly using actual values in me classification modeling
processes would require that the models be developed to accommodate the various types of values or
units for the data elements. For example, an attribute quantified as a concentration value of "10 parts
per billion" based on a finished drinking water measurement would need to be treated differently in
the classification model than an attribute quantified as *1,500,000 pounds" based on estimates of the
amount of the contaminant produced each year.
The first alternative for assigning numerical scores to attributes addresses the potential challenge
of having to design the classification model to address a variety of different types of data as input
values for the same attribute. This is accomplished by first converting (he measurements within each
of the data elements that might be used for a given attribute to a unitless attribute score across some
specified range of values (for example, 0 through 10). This conversion procedure would involve an
algorithm that would reflect a direct relationship between the original data and the resulting attribute
score. An appropriate number of significant digits should be maintained in the attribute scores so that
information loss in the conversion process is avoided. This implies that, knowing the algorithm and
the data element used, one could determine the underlying actual values for the data element used to
produce die attribute score. In addition to the conversions within each data element, it would also be
necessary that the algorithms employed "normalize" the resulting scores across different data elements
used for a given attribute. For example, an attribute score for potency of "3.8" based on a particular
LDso value and an attribute scon; for potency of'3.8" based on a particular Reference Dose should
convey the same degree of concern regarding potency even though they are derived from different
data elements.
The second approach for assigning attribute scores is conceptualy very similar to the first
approach with respect to applying algorithms that normalize for the disparate data elements that might
be used for a given attribute, except that the algorithm would generate categorical scores to reflect
Ch. 5-Moving from the PCCL onto the CCL 5-3
-------
CCL CP Work Group Report
similar levels of concent for that attribute. These category scores would also be in some specified
range, for example 0 through 10, but unlike the values for the option above that could have any value
(i.e., any number of decimal places, such as 8.26) the scores in this third approach would be limited to
integer values only (such as 8). As a result, this approach would in most instances result in some loss
of information upon which the scores are based. While there may be some advantages to mis approach
in mat it groups "like things" together, it does preclude being able to make finer distinctions among
those "like things" without going back to the original underlying data. This approach also may allow
for other models to be used that accommodate categorical variables.
It should be recognized that there are attributes under consideration (for example, severity) and
some data elements as mentioned previously that do not involve quantified measurements or
estimates, hi those cases, there is likely no option other than to assign a categorical score, probably as
an integer value, using a protocol to reflect appropriately the level of concern implied by the
underlying data element.
The NRC did not specifically address the alternative of using the actual data values as input to the
classification models, but rather addressed only the approach of developing attribute scores (and in
this regard only the integer value categorical scores as discussed above). The NDWAC Work Group
did not reach a conclusion regarding which approach is preferred and therefore does not make a
specific recommendation favoring one over the other. Some of the recommendations that follow refer
to aspects of attribute scoring and are therefore relevant where EPA determines that attribute scoring is
the preferred approach.
5.1.1.1 Summary of NRC Recommendations on Quantifying Attributes
The NRC (2001) recommended that EPA develop and use a set of attributes to evaluate the
likelihood that any particular PCCL chemical or microbial contaminant could occur in drinking water
at levels and frequencies mat pose a public health risk. The NRC further recommended that attribute
scores be the input to a prototype classification approach, used in conjunction with expert judgment, to
help identify the highest priority PCCL contaminants for inclusion on a CCL. The NRC suggested that
attributes and a scoring process be used because various types of information (i.e. data elements)
would need to be used to score the attributes for different contaminants, and because of the widely
varying data availability for emerging contaminants. The attribute scoring process would be used to
put the different types of data elements for the same attribute on a common scale for evaluation and
use in a classification model.
5.1.1.2 NDWAC Work Group Evaluation of Attributes
The Work Group carefully reviewed the information presented in the NRC report on the
attributes proposed by the NRC for consideration by EPA. The Work Group also explored a number
of important questions regarding attributes.
• Which attributes appear to be most appropriate for use in the CCL Classification Process
and how should those attributes be defined?
» What data elements should be used as measures in quantifying those attributes?
• What should the hierarchy (preferences) among data elements for a given attribute be?
Ch. 5 - Moving from the PCCL onto the CCL 5-4
-------
CCL CP Work Group Report
» What are the practical constraints of obtaining and processing the data and information
needed to quantify attributes?
• How should values for a given attribute obtained from using different data elements be
normalized for ensuring consistency in scores?
" When scoring is used, do the scales (scoring ranges) need to be consistent across attributes?
" How should data quality concerns be considered in the process of quantifying attributes?
5.1.2 NDWAC Work Group Recommendations
-> EPA should proceed initially with using the two health effects attributes and three
occurrence attributes described by the NRC as input for the PCCL-to-CCL classification
modeling for contaminants.
The Work Group determined that (he concept of ushg attributes as part of the process for
selecting (hose contaminants on the PCCL that are likely to pose the greatest human health risks from
drinking water and moving them forward to the CCL is sound. A number of specific questions and
concerns, discussed further below, were raised with respect to details about defining attributes and
scoring them for use in a classification approach. These concerns generally go to certain specifics of
implementing the attribute scoring process, and except as noted in Chapters 3 and 4 for magnitude, do
not suggest the need for any major conceptual changes from the NRC recommendations.
-> EPA should systematically refine and improve upon the details of the attributes as more
experience is gained. These should include refinements and improvements in gathering and
processing the needed data and information to quantify the attributes and with respect to
using the attribute values in the selected classification approach. Further refinements and
improvements should include consideration of whether fewer than all five attributes are
needed, as well as of the data elements used to quantify the attributes, and, if scoring is
used, the scoring protocols, and the actual attribute scoring process itself.
The Work Group recognizes that there are numerous details concerning how many attributes are
needed and how they should be characterized and scored that must be developed in conjunction with
the development of the specific classification approach(es) to be used as part of the process of moving
contaminants from (he PCCL to the CCL. This is in keeping with the NRC observation that me five
attributes discussed in its report were illustrative and represented a reasonable starting point for EPA's
consideration.
Consistent with the adaptive management approach discussed in Chapter 2, the Work Group
recognizes that results obtained by EPA from the initial development and implementation of the
classification model, as well as from associated expert judgment processes, will result in additional
information about attributes. EFA should consider this information when deciding whether fewer than
the five attributes as currently described are needed to adequately prioritize agents on the PCCL for
placement on the CCL. hi that same vein, it is possible that these initial results will help EPA to
improve on the data elements needed and make the information gathering process more focused and
efficient Therefore, it is important for EPA to specifically include as part of its CCL efforts an
Ch. 5-Movingfrom the PCCL onto the CCL
5-5
-------
CCL CP Work Group Report
adaptive process to assess the attributes and make changes to them and the scoring protocols based on
experience gained. (This is an example of the adaptive management approach described in Chapter 2.)
-> Attribute scores, if used, should increase with concern. That is, contaminants that warrant
higher scrutiny in the CCL process should receive higher scores for attributes.
The Work Group recognizes that the classification models that ultimately will make use of the
attribute scores if they are used can be structured to allow for a different ordering of the numerical
scores (i.e., for some attributes a score of 10 on a scale of 1 to 10 could reflect greatest concern, while
for another attribute a score of 1 on that scale could reflect the greatest concern). Nevertheless, to
enhance the interpretation of the attribute scores themselves outside of their subsequent use as input to
a classification modeling effort, the Work Group recommends that EPA use a consistent order in the
scores across all attributes to reflect greater or lesser degrees of concern (i.e., contribution to potential
risk) indicated by that particular attribute.
-> The Work Group recommends that EPA explore the alternative approaches of using actual
values to quantify attributes and attribute scoring described above, taking into account the
requirements for implementing them - both in terms of the quantification process itself and
in terms of their use in the classification models- and the possible implications each
approach could have on the outcome of the classification modeling. EPA should also
consider using a combination of scoring approaches depending upon the particular
attribute rather than selecting one approach only for all attributes.
-> If attribute scoring is used, the scoring system selected by EPA for each attribute should
enable discrimination among contaminants. If scoring categories are used, there should be
a sufficient number of scoring categories so that information loss during characterization of
contaminants is limited. At the same time, the scoring categories should not be so numerous
that they convey a false sense of precision.
The major purpose of attribute scoring is to provide relative values for the attributes that can be
compared among contaminants to identify those contaminants that merit further consideration. An
attribute scoring system set up on two scores may not be sufficient to discriminate contaminants
accurately and the process could lose information because it does not provide enough separation of
data. Conversely, a scoring system set up on 20 scores may convey a false sense of precision such that
a score of 9 versus 10 may not be significant. Therefore, the available data used to score an attribute
should be evaluated to determine the number of scoring categories that provides sufficient separation
of the contaminants without implying a false precision.
-> EPA should generate and include, along with the actual values or the attribute scores that
are generated, descriptive "tags" that provide additional data quality information that may
be used by experts reviewing the data, the attribute scores and/or the PCCL-to-CCL
classification modeling results.
As discussed in Chapter 2, the Work Group recognizes that inherent in the use of the various
types of data and information that will be needed to generate attribute scores are issues and concerns
about data quality. The Work Group explored some options for how EPA might specifically address
data quality concerns as part of the attribute scoring component of the PCCL to CCL effort Included
Ch. 5-Movingfrom the PCCL onto the CCL
5-6
-------
CCL CP Work Group Report
among the options considered were: 1) integrating a score reflecting data quality into the attribute
score itself; and, 2) generating a separate quantitative score for data quality to pair with the actual
attribute score. Integrating data quality into the scoring process increases the complexity of attribute
scoring.
Another alternative the Work Group considered is for EPA to include along with each attribute
value or score a data quality "tag" to indicate the source and nature of the information used to generate
the score. These "tags" are intended to provide some descriptive data quality information that will be
suitable for use by experts reviewing the attribute scoring process or the final outcome of the PCCL-
to-CCL classification modeling effort. Those reviewers may, then, use the information provided by
the tag to determine whether aspects of the underlying data or information used to score the attribute
should be taken into account in arriving at the final CCL determinations.
As EPA gains more experience in implementing the procedures for moving contaminants from
the PCCL to the CCL, and in particular with quantifying the attributes, it is anticipated that the
approach to capturing, characterizing, and using data quality information as part of mat process may
be refined. It is also expected that the specific approach used by EPA for considering data quality
concerns in this step of the process may evolve from the currently recommended "tags" to some other
procedure for further consideration of the quality of the information.
-> If attribute scoring is used, the scoring protocols should be transparent and straightforward.
Attribute scoring protocols should be clear and easy to follow. A group of users of varying
expertise should be able to derive equivalent attribute scores for the same set of contaminants.
Evaluuation of the scoring protocols should take into consideration varying types of data format and
display, data element names, and data units.
5.2 Overview of Classification Approaches
The NRC discussed three general approaches for classifying PCCL contaminants: expert
judgment processes, a priori rule-based approaches, and a posteriori prototype classification
algorithms.
Expert judgment processes are consultations with experts on a given subject matter to elicit
opinions and possibly consensus for decision-making purposes. Expert judgment processes may occur
as workshops, as facilitated discourses, or by eliciting die opinions of individuals, and combining the
extracted information in a rigorous framework.
An apriori rule is one in which a set of rules for decision-making is constructed through an
expert judgment process prior to making me decision. For example, a group of experts might decide
that the relevant attributes for CCL listing are potency and magnitude and that potency is twice as
important as magnitude. Further, they might decide that any contaminant receiving a total score of
greater than 25 should advance to the CCL. Then each contaminant would receive a total score of 2
times the potency score plus the magnitude score. Those contaminants with total scores exceeding 25
would be placed on me CCL.
Ch. 5-Moving from the PCCL onto the CCL
5-7
-------
f
CCL CP Work Group Report
In contrast, prototype classification methods develop a posteriori rales for decision-making
based on decisions that have already been made. Rather than specifying the relative importance of die
attributes a priori, an expert process establishes a "training data set." The expert process decides
which contaminants in the training set should be included or not included on the CCL. Based on these
decisions a mathematical pattern-recognition algorithm establishes the relative importance of each
contaminant attribute in the decisions made by experts. This "trained" algorithm would men serve as a
tool to aid experts for future CCL decisions.
These approaches are not necessarily mutually exclusive. For example, a process could first use
structured discourse or an a priori rule-based approach to create an appropriate training set for the a
posteriori approach. In such a combined method, the first stage could have participants reflect on and
discuss the strengths and weaknesses of different forms of the algorithm and weights of attributes.
This discussion could focus on a set of candidate agents that are in some sense "representative" of the
range of candidates. The participants could then form judgments regarding which of these candidates
should be listed or not in the training set. The training set could then be used in an a posteriori
approach to develop the final algorithm form and attribute weightings that best explain these
judgments.
5.3 Recommended Approach to Selecting the CCL
5.3.1 NRC Recommendations
The NRC recommended prototype classification algorithms be considered over expert processes
and rule-based approaches, citing limitations of experts (time, knowledge base, bias) in the former,
and the complexity of the expert decision process relative to simple rules in the latter. However, NRC
emphasized that with whate\er alternative is considered, EPA must continue to rely on expert
judgment throughout the classification process, because the data and knowledge for selecting drinking
water contaminant candidates - particularly emerging contaminants - are admittedly incomplete.
NRC suggested EPA explore alternative model formulations, conducting sensitivity analysis to
validate any findings, and being cognizant of the dangers of over-fitting and loss of generalization in
model development. NRC recommended that EPA use a prototype classification approach in
conjunction with expert judgment, suggesting that this approach lessens the need for transparency of
the algorithm.
5.3.2 NDWAC Work Group Recommendations
The Work Group concurred with the NRC report that EPA should mow toward a prototype
classification approach The Work Group concluded that implementation of a classification algorithm
could improve the transparency of the decision process and would improve consistency in how the
CCL is developed over time - provided that this approach can be developed in accordance with the
specific recommendations below. The Work Group recognized that these models are not "objective,"
as they will try to mimic die decisions that are made for the training data set, but felt that quantifying
these decisions made them more explicit and transparent. Thus, careful contaminant selection and
attribute scoring for the training data set are imperative.
Ch. 5 - Moving from the PCCL onto the CCL 5-8
-------
CCL CP Work Group Report
The rationale for recommending a prototype algorithm is based in large part on the NRC's
deliberations. The Work Group agreed with the NRC that a formal approach could improve decision
making over past expert judgment processes, at the very least by increasing the number of
contaminants that can be considered in the CCL process.
The NRC trial analysis had compared a linear discriminant model with an artificial neural
network (ANN), and found the ANN to outperform the linear discriminant model. The Work Group
also conducted an exploratory analysis using four methods: 1) logistic regression, 2) ANN, 3)
classification and regression trees (CART), and 4) multivariate adaptive regression splines (MARS).
The Work Group did not furthe r consider the linear discriminant model, because its use is based on a
set of fairly restrictive assumptions regarding the structure of the input data. All five approaches are
briefly discussed in Appendix E.
The Work Group exploratory analysis used a training data set of 46 contaminants that was
based on prior CCL decisions. A "best" model was chosen from each of the four aforementioned
methods using appropriate techniques. (See Appendix E for a description of this analysis.) The
four "best" models (i.e. one from each method class) were then compared in a cross-validation
exercise using randomly chosen subsets of the original training data set. The average
misclassification rates were calculated for each model. In this comparison the MARS model had
the lowest average misclassification rate, the ANN had the highest misclassification rate, and the
logistic and CART models were in between. An additional insight from this analysis was that the
"best" model in each category usually incorporated only two or three of the five attributes, with
Magnitude and Prevalence consistently included. However, the Work Group does not consider
the results to be definitive because the training data set was assembled only for exploratory
analysis. To make definitive recommendations, a more extensive and thorough analysis would be
needed.
The specific recommendations are as follows.
-> The Work Group recommends that EPA pursue development of a prototype classification
algorithm (a posteriori approach) for selecting contaminants for the next CCL. The Work
Group recommends moving forward to develop and test one or more prototype models as
tools to be used with expert judgment for decisions on classifying contaminants for future
CCLs.
The Work Group did not have time to evaluate the alternatives and recommend a particular
prototype model. Several features of the CART models, including their graphical depiction, which
could aid transparency, and their ability to partially accommodate missing data, make them
particularly attractive. However, more definitive testing and validation of the candidate approaches is
required to make a definitive recommendation. The Work Group recognizes that it may be useful to
have several models that are used in concert to corroborate results. Additionally, it may be necessary
to develop separate models for chemical and microbiological contaminants, or models that
differentiate chemicals and microbes within the model structure. The development of any model
should be an adaptive process, and should be reviewed by experts, with consideration given to
updating the training data set, with each successive CCL cycle.
Ch. 5-Movingfrom the PCCL onto the CCL
5-9
-------
t
t
CCL CP Work Group Report
Implementation of this recommendation depends on a well-constructed, reasonably reliable,
training data set. Section 5.4 provides further recommendations for constructing a training set.
These models are tools to help prioritize contaminants for CCL, not the final decision of whether
a contaminant should be listed. Experts should make die final decisions by review of the available
data, including information regarding the quality and uncertainly of the data used in attribute scoring.
The rationale for this recommendation is to ensure that EPA conducts adequate evaluation of
models before deciding whether or which models to use. Time constraints prohibited the Work Group
from conducting detailed evaluations of the models to make more specific recommendations
regarding their use.
-> The Work Group recommends that the entire model development process be as transparent
as possible. The development process should be viewed as iterative, and EPA should involve
experts and allow opportunities for meaningful public comment on the evaluation.
The details of and justification for the decisions that are made should be carefully documented
and publicly available. Some issues to consider in comparing algorithms could include:
• How well algorithms predict CCL classification (misclassification rates)
• How algorithm output is affected by changes to individual training set decisions
• How algorithm output is affected by changes to the attribute scoring rules
• The importance of including all five attributes (reducing their number can reduce the labor of
gathering data and scoring attributes)
• The relative performance of the different competing algorithms
" The relative performance with different training data sets
" How much of the input information is used in the evaluation
• How well algorithms work with missing or incomplete input data
• How well the results can be communicated to a non-technical audience
The purpose of these recommendations is to assure transparency as well as provide guidance
and direction to EPA during the development of the model(s) and to provide a systematic
framework for the experts reviewing the models and their performance for a diverse range of
chemical and microbiological contaminants.
-> EPA should use another approach for selecting CCL contaminants in the near term (Le.,
for CCL3) if there are difficulties in the model development process that cannot be
overcome.
This approach may include expert processes and/or a priori approaches. The Work Group does
not recommend alternatives be developed in parallel; however, the Work Group wants to ensure that
EPA's schedule for algorithm testing, development, and review allow adequate time for
implementation of an alternative approach for the next CCL, including appropriate public
involvement.
Ch. 5-Movingfrom the PCCL onto the CCL
5-10
-------
CCL CP Work Group Report
-> The Work Group recommends that experts should be involved throughout the process of
narrowing a PCCL to a CCL, specifically as advisors in the design of an approach,
development of a training set, scoring of contaminant attributes, evaluation of algorithm
results, and ultimate selection of CCL contaminants.
The rationale for recommending expert involvement and review stems from a concern that EPA's
use of prototype classification algorithms or other models be as tools in conjunction with broad
participation by experts and others in their development and evaluation
5.4 Training Data Set
The discussion below applies to both microbial and chemical training data sets. The Work Group
recommends against combining these two kinds of contaminants into a single training set, at least for
the next round of CCL. Ultimately, EPA should work toward the construction of a unified system of
attribute scoring, training and classification, but for the moment, separate treatments will be required
for microbial and chemical contaminants.
Training data will play an important role in methods used to classify PCCL contaminants (as
being on or off the CCL). The goal is for the training set to be the template for subsequent CCL
decisions. Developing the trainmg data set will be a complex activity, requiring significant data
synthesis, attribute scoring, and decision-making components.
A training data set consists of numerous contaminants, together with their health effects
data/information, occurrence delta/information, scored attributes (if scoring is performed), and listing
status ("on" or "off' the CCL). Training contaminants and their supporting data/information must be
carefully considered because ths listing status of the training set contaminants will inform which
PCCL contaminants should be listed. Historically listed contaminants may warrant additional
evaluation as new information becomes available. In the process of selecting a classification algorithm
it will be necessary to compare the performance of the candidate algorithms and validate their
respective performances. It is likely that this will be accomplished, in part, by choosing random
subsets of the training data set, training the algorithms with the randomly chosen subset, then
evaluating the performance of the algorithms on the contaminants that were excuded from the
randomly chosen subset. Thus, the training data set should contain enough contaminants to facilitate
an informative validation and comparison procedure.
5.4.1 NRC Recommendations on Training Data Set
The NRC suggested that it would be a relatively straightforward exercise to construct a training
set suitable for "training" a prototype classification algorithm capable of differentiating between
contaminants that should be included on the CCL and those that should not. Specifically, the NRC
recommended using a training data set consisting of chemicals, microorganisms, and other types of
(potential) drinking water contaminants that clearly belong on the CCL (such as currently regulated
contaminants), and those that clearly do not (such as food additives generally recognized as safe by
the US Food and Drug Administration).
Ch. 5 - Moving from the PCCL onto the CCL 5-11
f
-------
CCL CP Work Group Report
The NRC also recommended that EPA include in the training data set contaminants for which
values of some of the attributes are unknown, and that EPA investigate the importance of different
attributes by leaving out certain attributes in the training data set and examining the effect on
classification of the training data. Finally, the NRC also recommended that EPA withhold some
contaminants from inclusion in the training set for use in validation testing to assess the predictive
accuracy of any classification algorithm developed. If similar results were achieved using different
training data sets, this would help ensure a robust classification process.
The NRC noted that a classification tool could be perceived as lacking transparency, in that there
is no obvious indication of how it is working. Though the model might be difficult for the public to
understand, the NRC indicated that the judgments embodied in the training data set would be tilings
the public would be able to understand and that this would be one reason that this process could be
more transparent than a rule-based approach. The NRC also indicated that such an approach can be
made relatively transparent by clear communication of the basis for the attribute scoring scheme, the
basis for the training data set, and the basis for evaluating the accuracy of the algorithms' predictions.
If these aspects of the process are perceived to be sound, the derived algorithms will be easier to
justify and defend.
5.4.2 NDWAC Work Group Recommendations
The Work Group considered the NRC recommendations and further investigated some of the
technical issues that they raise. In particular, the Work Group disagreed with the NRC's assessment
that training data set construction could be "relatively straightforward." The NRC indicated that a
usable training data set could be constructed using only contaminants that clearly belong or clearly
don't belong on the CCL and about which consensus would be readily reached. The investigations of
the Work Group suggest that this may not be the case and that a more extensive training set with a
number of contaminants not readily classified would be needed. This raises such questions and
concerns as how to develop an appropriate training set, and, given the need for a relatively large
number of contaminants of diverse status, how this can be made transparent to the public.
The Work Group recommends the following principles to guide training set development
-> The training data set should consist of contaminants (and corresponding decisions to "list"
or "not list" each contaminant) that reflect technically sound, consistent judgments about
what should and should not be included on the CCL.
Some of the decisions will be obvious, but others will be more complex or less easily
differentiated.
-> The training set should include contaminant attribute data that are distributed throughout
the attribute space, and the training set should be selected to define the discriminant
surface (the function that defines "include" and "exclude" decisions) as precisely as
possible.
Training an algorithm to make difficult decisions may be important and requires that "difficult"
contaminants (i.e., contaminants for which the correct decision as to whether to list or not list is not
obvious) be included in the training set. By failing to include such difficult contaminants, the decision-
Ch. 5-Movingfrom the PCCL onto the CCL
5-12
-------
CCL CP Work Group Report
maker may be left with too wide a range of possible algorithms, resulting in a poorly specified
algorithm. The result could be classification decisions mat would change if another of the possible
algorithms had been used. This is explained through the two figures below.
Figure 5.2 shows a set of iiontaminants, each characterized by two attributes (Occurrence and
Health Effects). The solid dots are contaminants clearly judged to be listed, and the open circles are
contaminants clearly judged to be non-listed. Because the analysis considered only clearly listed and
clearly non-listed contaminant:;, there are no contaminants found in a broad band between these two
groups. The solid line shows one possible "discriminant" or line separating "listed" and "non-listed"
contaminants. Either of die dashed lines, however, would also explain the decisions underlying this
training set The result is an inability to specify precisely where the discriminant should be placed to
separate the two groups. The result will be substantial uncertainty in classifying contaminants that
eventually are found to lie between the two dashed lines. Understanding this, and having little
indication of which PCCL contaminants were assigned with confidence, EPA would need to review
many listing decisions indicated by the algorithm, and this would reduce the benefit of utilizing the
prototype classification algorithm.
Figure 5.2 - "Separated" Contaminants Poorly Define the Discriminant
456
Occurrence
10
Ch. 5-Movingfrom the PCCL onto the CCL
5-13
-------
CCL CP Work Group Report
By contrast, Figure 5.3 shows a case where the training set includes contaminants in this "border"
region. The resulting discriminant is now better specified, lying somewhere between the two dashed
lines in that figure. Note die trade-off between die two figures. The discriminant in Figure 5.2 is
successful at classifying 100% of die contaminants in the training set, but is located imprecisely. The
discriminant in Figure 5.3 does not classify 100% of the contaminants correctly (look at the dots near
the discriminant), but places the discriminant much more precisely.
10
9-
a-
7-
6-
5-
4-
3
2-
1 -
o.
I List contaminants
* falling on this side of
the curve.
Do not list contaminants
falling on this side of the
curve.
23456
Occurrence
10
Figure 5.3 - A Discriminant Function on the Basis of Two Attributes
-> The Work Group recommends that EPA maintain transparency and clarity when
developing the training data set. To the extent feasible, EPA should document training data
set development and communicate its rationale for assigning decisions to training set
contaminants.
-> The rationale for the number and distribution of training set contaminants should be
described. Quantitative rationale should be expressed for the prototype classification
approach.
These are important considerations for determining if the training set and models have been
adequately developed to begin processing PCCL contaminants. The rationale should include a
description of the methods used for calibration and validation, and measures used to assess goodness
of fit such as misclassification rates.
Ch. 5-Movingfrom the PCCL onto the CCL
5-14
-------
CCL CP Work Group Report
Glossary of Terms
The purpose of this glossaiy is to define terms that may be used in the discussion of the 2001
National Research Council report, Classijying Drinking Water Contaminants for Regulatory
Consideration, National Academy Press, and terms used by Work Group members that may be
subject to interpretation. These are suggested definitions, in some cases summarized from the NRC
report, presented in alphabetical order and referenced.
Adaptive Management: A continuing process of action-based planning, monitoring, researching,
evaluating, and adjusting with the objective of improving implementation and achieving the goals of
the selected alternative (41).
NDWAC defines an adaptive management approach as a process that involves the following
steps: 1) identify an approach, 2) define evaluative criteria (factors to evaluate), 3) iteratively
implement the approach, 4) transparently assess based on obscured results and evaluative criteria and
5) make changes to improve performance of the approach (this report).
Adherence: The ability of microbes to stick (adhere) to surfaces (49,41).
Adhesin. Microbial surface antigens that frequently exist in the form of filamentous projections
(pili or fimbriae) and bind to specific receptors on epithelial cell membranes; usually classified
according to their ability to induce agglutination of erythrocytes from various species, their differential
attachment to epithelial cells of various origins, or their susceptibility to reversal of such binding
activities in the presence of mannose (49,41).
Aflatoxin: A fungal toxin that is a powerful liver carcinogen. A group of closely related toxic
metabolites are designated as mycotoxins. They are produced by Aspergillusflavus and Aspergillus
parasiticus. Members of the group incbde aflatoxin bl, aflatoxin b2, aflatoxin gl, aflatoxin g2,
aflatoxin ml, and aflatoxin m2 (41).
Agent: Any physical, chemical, or biological substance (this report).
Algorithm: A procedure for obtaining a result It can be applied to solving a mathematical
problem in a finite number of steps that frequently involves repetition of an operation; broadly : a
step-by-step procedure for solving a problem or accomplishing some end especially by a computer
(10).
a priori: A Latin phrase that refers to something formed or conceived before data or events are
reviewed (10).
a posteriori: A Latin phrase that refers to derived by reasoning front observed facts (10).
Glossary G-l
-------
CCL CP Work Group Report
Assessment: combination of analysis of facts and inference of possible consequences concerning
a particular object (2).
Attribute: Characteristics of contaminants or potential contaminants that contribute to the
likelihood that a particular contaminant or related group of contaminants could occur in drinking
water at levels and frequencies mat pose a public health risk (this report).
NRC identified five attributes to characterize PCCL contaminants for classification: severity,
potency, prevalence, magnitude, and persistence-mobility. Severity and potency describe health
effects and prevalence, magnitude, and persistence-mobility refer to occurrence (1).
Bayesian: Probabilistic inference that combines prior knowledge and newly acquired information
via Bayes Theorem (10).
Binning: an approach for sorting agents, by classifying mem into two or more "bins" or groups.
NDWAC has discussed the use of a two bin (on or off) approach for selecting PCCL contaminants
from the universe (mis report).
Bio film, a community of microorganisms growing on a surface in a matrix of polysaccharides
and glycoproteins (41).
Bioinformatics: An interdisciplinary approach to biology that combines elements of
mathematics, statistics, computer science, and information theory, with genetics, medicine
epidemiology, pharmacology, molecular biology, physiology, biochemistry and microbiobgy (5).
Capsule; Thick gel like material attached to the wall of gram-positive or gram negative bacteria,
giving colonies a smooth appearance. May contribute to pathogenicity by inhibiting phagocytosis.
Mostly composed of very hydrophilic acidic polysaccharide, but considerable diversity exists (49,41).
CART: Classification and Regression Tree analysis is a statistical technique yielding a class of
models called tree-based models. It is an exploratory technique for uncovering structure in data, and
produces a graphic of a branched tree indicating splits in the data The points at which the data are
split are called nodes, and the splitting of the data into groups occurs such that the homogeneity of
each group is maximized. Data are split by binary recursive partitioning into groups of increasing
homogeneity (18).
CAS (Chemical Abstracts Service): CAS is a team of scientists who create a digital information
environment for scientific research. CAS provides pathways to published research in the scientific
literature back to the beginning of the 20th century (13).
CASRN: Chemical Abstract Services Registry Number. Unique substance identification number
defined by Chemical Abstract Services. Also represented as CAS Reg. No. (13).
CCL: Drinking Water Contaminant Candidate List (1). The Safe Drinking Water Act (SDWA)
Amendments of 1996 require that the Environmental Protection Agency (EPA) publish a list of
unregulated chemical and microbial contaminants and contaminant groups every five years that are
known or anticipated to occur in drinking water systems and that may pose a public health risk in
drinking water and may require regulation (1).
Glossary
G-2
-------
CCL CP Work Group Report
CCL Universe: A list of identified, new, and emerging potential drinking water contaminants
used to develop the CCL. The CCL Universe includes contaminants that have demonstrated or
potential occurrence in drinking water or that have demonstrated or potential health effects. This is
the NRC definition of the CCL Universe, and does not apply to the microbial CCL Universe as it is
being proposed in this report (23).
Classification system: A system for the sorting of data into discrete groups (1). Three broad
types of systems have been considered for sorting potential contaminants including: expert judgment,
rule based systems, and prototype classification algorithms (1).
Colonization (factors): Formation of compact population groups of the same type of
microorganism, as the colonies that develop when a bacterial cell begins reproducing (41).
Comparative risk assessment: A process to attempt to evaluate the relative magnitude of risks
and set priorities among a wide range of environmental problems (4).
Conservation: Changes at a specific position of an amino acid or (less commonly, DNA)
sequence that preserve the physico-chemical properties of the original residue (48).
Contaminant. The Safe Drinking Water Act (SDWA) defines "contaminant" as any physical,
chemical., biological, or radiological substance or matter in water (41 USC Sec. 300f).
NDWAC defines contaminant as any physical, chemical, or biological substance in water. For
mis report, the Work Group used contaminant to indicate any agent for which data exist that suggests
mat the agent belongs on the PCCL (this report).
Continuous: A property of data. A variable is continuous if between any two possible values of the
variable, mere exists another possible value for the variable. This is in contrast to categorized data
Criteria: standards on which a judgment or decision may be based; or characterizing marks or
traits (10).
Cyanobacteria: A division of photosynthetic bacteria, formerly known as blue-green algae, that
can produce strong toxins (51).
Cyanotaxin: Toxin produced by cyanobacteria (51).
Cytotoxins. Substances elaborated by microorganisms, plants, or animals that are specifically
toxic to individual cells; they may be involved in immunity or may be contained in venoms (41).
Data: factual information (as measurements) used as a basis for reasoning, discussion, or
calculation (10).
Database: a collection of data organized especially for search and retrieval (as by a computer)
(10). A key feature of a database is the relation of one data element to the next by a unique identifier
for each entry.
Glossary G-3
-------
CCL CP Work Group Report
Data element: one of the necessary data or values on which calculations or conclusions are based
(10). hi this case: A readily identifiable descriptor that characterizes information about a contaminant
(e.g., its identity, form, properties, test conditions, and study endpoints).
Data-poor: a qualitative description of the rehtive lack of availability of information regarding
data elements for contaminant or group of contaminants.
Data-rich: a qualitative description indicating the availability of information regarding data
elements for contaminant or group of contaminants.
Data source: Generally refers to a database or other source of information. To date, 242 data
sources have been identified, and the list continues to expand. Because large numbers of contaminants
are anticipated to be used to produce the Contaminant Candidate List, electronic databases and data
sources encompass most of the list (23).
Disease: Any change from a state of good health or interruption in the normal functioning of the
body, an organ, or tissue (5).
Ecology: A branch of science concerned with the interrelationship of organisms and their
environments. The totality or pattern of relations between organisms and their environment (10).
Emerging agents: a subset of known physical, chemical, or biological substances previously
evaluated as not requiring inclusion in the CCL Universe, for which new information becomes
available which heightens concern and triggers revaluation (this report).
Endemic Disease: continued prevalence of a disease in a specific population or area (22).
Estimates: estimates are any evidence of potency or exposure (or both) in drinking water-which
may have been derived through a process of inference and/or judgment based on data that are clearly
relevant to, but not necessarily directly concerned with drinking water. Estimates may be generated by
appropriate and credible models (including quantitative structure-activity relationship models, or
QSARs, if consistent with the policy for acceptable QSARs addressed elsewhere). Estimates also may
be derived from arguments by analogy, from measures in media other than water, through expert
judgment or some other estimation process, (this report)
Expedited process: As new agents are identified, or as new information becomes available, there
may be justification to accelerate their passage to the CCL Universe, from the universe to the PCCL,
or from the PCCL to the CCL. A re-evaluation process based on key criteria may be considered to
allow contaminants of immediate concern to be expedited or "fast-tracked." (this report).
Expert: one who has special skill or knowledge derived from training or experience relevant to
the particular subject matter or technical analysis at hand (this report).
Expert Judgment: opinion of an expert(s) on a particular subject based upon relevant technical
analysis or garnered as a technical consensus based on available information (this report).
Expert Review: critical or deliberate examination of a decision or process by an expert(s). As
used in this report, expert reviews may involve various types of expert consultation and collaboration,
Glossary
G-4
-------
CCL CP Work Group Report
up to and including formal peer reviews. Expert reviewers are qualified individuals (or organizations)
who are independent of those who performed the work, but who are collectively equivalent in
technical expertise to those who performed the original work ("peers"). EPA uses such review for
enhancing a scientific or technical work product so that the decision or position taken by EPA, based
on that product, has a sound, credible basis (this report).
Exposure. Amount of a particular agent that reaches a target system. It is usually expressed in
numerical terms of a concentration (2).
Genbank: Database of all published DNA, RNA and protein sequences maintained by the
National Center for Biotechnology Information (NCBI) (5).
Gene: The functional unit of inherited information, often expressed as a single trait. Genes are
located on the chromosomes. Each gene is encoded by a specific sequence of nucleotides in the
nucleic acid of the organism (5j.
Genome: The complete set of genes carried by an individual or organism. The genome serves as
a master blueprint for all cellular structures and activities for the lifetime of the cell or organism (5).
Genomics: The study of genes and their function (34).
Hazard: inherent property of an agent or situation capable of having adverse effects on
something. Hence, the substance, agent, source of energy or situation having (hat property (2).
Health Advisory: An estimate of acceptable drinking water levels for a chemical substance based
on health effects information; a Health Advisory is not a legally enforceable Federal standard, but
serves as technical guidance to assist Federal, State, and Local officials (7).
Health effects, demonstrated: NRC defines contaminants with demonstrated health effects as
those mat are associated with (1):
1) Human health data shoving health effects; or
2) Toxicological studies on whole animals.
Health effects, potential: NRC defines contaminants with potential health effects as those that
are associated with any toxicologic data or data from experimental models that predict biological
activity (other than human health data, and toxicologic data on whole animals which indicates
demonstrated health effects) (1).
Host: A human or other living animal, including birds and arthropods, that affords subsistence or
lodgement to an infectious agent under natural conditions. Some protozoa and helminths pass
successive stages in alternate hosts of different species (5).
Immune response: Alteration in the reactivity of an organism's immune system in response to an
antigen, in vertebrates this may involve antibody production, induction of cell-mediated immunity,
complement activation or development of immunological tolerance (47).
Glossary G-5
-------
CCL CP Work Group Report
s
Immunocomprondsed persons: Immunocompromised persons have reduced immune
responsiveness due to infection, disease, malnutrition, immnosuppressive drug therapy, or other
factors (5).
Incubation period : The time from the moment of inoculation (exposure) to the development of
clinical manifestations of a particular infectious disease (47).
Infection: The entry and survival or multiplication of an infectious agent in the body of humans
and animals. The process by which a pathogen establishes itself in a host includes transmission,
invasion, and multiplication. The target organ (e.g., intestinal tract) must come in contact with
sufficient numbers of an agent, the agent must possess specific virulence factors, the virulence factors
must be expressed, and an immune response may be elicited. Infection may be asymptomatic or result
in disease (5).
Infective dose: The number of organisms required to produce an infection in humans or animals
(5).
Insertion sequence: Small (< 2.5 kb) variable genetic elements with simple genetic organization
lhat are capable of inserting at multiple sites in a target DNA molecule (49, 46).
Invasion: The attack or onset of a disease; the entrance of bacteria into the body or deposition in
the tissues, as distinguished from infection; die infiltration and active destruction of surrounding
tissue, a characteristic of malignant tumors (49, 50).
Known agent physical, chemical or biological substances that have been identified in the
technical literature and adequately characterized to enable a judgment regarding their inclusion in the
CCL Universe (this report).
LD50 (50% Lethal Dose): a dose that causes mortality in 50% of exposed animals for chemical
contaminants (26).
Lowest Lethal Concentration/Dose (LC/LDlo): The lowest concentration/dose to cause death in
test animals (28).
Lowest-Observed-Adverse-Effect Level (LOAEL): The lowest exposure level at which there are
biologically significant increases in frequency or severity of adverse effects between the exposed
population and its appropriate control group (44).
Lowest-Observed Effect Level (LOEL or LEL): In a study, the lowest dose or exposure level at
which a statistically or biologically significant effect is observed in the exposed population compared
with an appropriate unexposed control group (44).
Magnitude: An attribute, defined by NRC as. the concentration or expected concentration of a
contaminant relative to a level that causes a perceived health effect (1). see Attribute.
MCL (Maximum Contaminant Level): The highest level of a contaminant that is allowed in
drinking water under federal regulations, which is set as close to the MCLG as feasible using the best
available treatment (20).
Glossary
G-6
-------
CCL CP Work Group Report
MCLG: Maximum Contamination Level Goal The maximum level of a contaminant in drinking
water at which no known or anticipated adverse effect on the health of persons would occur, and
which allows an adequate margin of safety. Maximum contaminant level goals are non-enforceable
health goals (12).
Measurements: Measurements refer to data showing directly an agent of interest occurs in water,
or that produces health effects via drinking water exposure or both. This might include, for example,
measurements of an agent in water; measurements of a health effect by the oral route of exposure(this
report).
Microarray: A large number of nucleic acid probes (100s - > 10,000) immobilized on small glass
or nylon supports (5).
Mobility: An attribute defined by NRC to identify whether a contaminant is likely to be found in
water, suggested by NRC to be considered with persistence as an attribute, particularly when there are
no available data indicating demonstrated occurrence in water. Mobility refers to a biological or
chemical contaminant's ability to move in water, defined for chemicals as properties such as aqueous
solubility, octanol water partition coefficient, Henry's Law constant recalcitrance, and for microbes as
properties that affect transportability in water, such as sedimentation velocity, size and adsorption
capability (1).
Morbidity rate: Sickness rate, the number of people who become sick compared with the number
who are well in a defined group over a defined time period (47).
Mortality rate: The proportion of deaths in a population or a specifb subpopulation (47).
Neural network: A prototype classification system (i.e. one that uses prototypes rather than fixed
features), a neural network is a mathematical representation of a network of biological neurons. Input
data are fed into die network and the output from the network is computed based on the architecture of
the network and the operative mathematical functions (1).
New agent: physical, chemical, or biological substances that are or may be newly-discovered or
synthesized, for which little is known about their potential occurrence or adverse health effects (this
report).
No-Observed-Adverse-Eff.
-------
CCL CP Work Group Report
exonucleases, each of which may be specific for the ribonucleic aciis (ribonucleases) or
deoxyribonucleic acids (deoxyribonucleases) (49,50).
Nucleic Acid: Any of the numerous large acidic biological polymers that are found concentrated
in the nuclei of all living cells. Nucleic acids contain phosphoric acid, sugar, and purine and
pyrimidine bases. Two types are DNA and RNA (5).
Occurrence: The presence or prevalence of a contaminant in the environment.
Occurrence, demonstrated: demonstrated occurrence of a contaminant in drinking water is
indicated by (in the NRC-recommended hierarchical order of importance) (1):
1) observations in tap water;
2) observations in distribution systems;
3) observations in finished water of water treatment plants; and
4) observations in source water.
Occurrence, potential: potential occurrence of a contaminant in drinking water is defined by
NRC as indicated by (1):
1) observations in watersheds and aquifers;
2) historical contaminant release data; and
3) chemical production data.
Outbreak: The occurrence of two or more of cases of a disease in a short period of time
associated with a common exposure (47).
Pathogens: Microorganisms that can cause disease in humans, animals or plants. They may be
bacteria, viruses, protozoa, or parasitic worms and are found in sewage in runoff from animal farms or
rural areas populated with domestic and/or wild animals, and in water used for swimming (12).
Pathogenicity islands: Fitness islands that confer pathogenicity or virulence in the organism in
which they are found (49,43).
PCCL: Preliminary CCL (1). NRC suggests a broadly defined universe of potential drinking
water contaminants is identified, assessed, and culled to a preliminary CCL (PCCL) using simple
screening criteria and expert judgment NDWAC recommends the screening criteria for selecting
PCCL contaminants be based on health effects and occurrence (this report).
PCR (Pofymerase chain reaction): The in vitro exponential replication of a specific DNA
sequence. The resulting amplification to detectable levels facilitates qualitative or quantitative genetic
analysis (1).
Persistence: The ability of a biological or chemical contaminant to remain in the environment
over time (6). For microbes, the ability of an organism (e.g., pathogen) to survive or a compound to
exist; that is, to remain in die environment (e.g., water) or in the host for extended periods of time (5).
See Attribute.
Glossary
G-8
-------
CCL CP Work Group Report
Persistence-Mobility: An NRC attribute defined as the likelihood that a contaminant would be
found in the aquatic environment based solely on its physical properties (1). See Attribute,
Persistence, and Mobility.
Pili: Hair-like projection from surface of some bacteria. Involved in adhesion to surfaces (may
be important in virulence) and specialized sex-pili are involved in conjugation with other bacteria.
Major constituent is a protein, pilin (49, 45).
Plasmid: A small, independently-replicating, piece of cytoplasmic DNA that can be transferred
from one organism to another. (Circular DNA molecules capable of autonomous replication found
both in eukaryotes and prokaryotes. Widely used in genetic engineering as vectors of genes (cloning
vectors) (45).
Potency: An NRC attribute defined as the amount of a contaminant required to cause an adverse
health effect (1). Potency of a pathogen may refer to the number of organisms required to cause
disease, while potency of a chemical refers to the dose required to cause disease. For example, some
pathogens require relatively few ($higella); others require a large number of organisms ^Salmonella
typimurium). (5). See Attribute*
Potential Exposure: NDWAC defines potential exposure as any information that suggests
exposure to an agent could occur via drinking water (this report).
Prevalence: NRC defines prevalence as how commonly a contaminant is found in drinking water
(1). For microbes, prevalence is one of the two most common broad measures of frequency used in
epidemiology (i.e. incidence and prevalence). The proportion of individuals in a population who have
the disease at a specific instant; prevalence provides an estimate of the probability (risk) that an
individual will be ill at a point in time (34). See Attribute.
Proteomics: Proteomics a discipline within functional genomics, is the study of proteomes,
protein sets expressed when the genomic blueprint of an organism is translated into functional
molecules (5).
Prototype classification algorithm: One based on prior classification of examples or prototypes.
Prototype classification methods develop relationships among contaminant attributes based on past
decisions (aposteriori). Rather than specifying the relative importance of the attributes a priori, an
expert process establishes a "tndning data set" that is used to develop the algorithm (this report).
QSAR: Quantitative structure-activity relationship (1). Quantitative structure-activity
relationships comprise one class of techniques used to predict behavior of novel chemicals based upon
similarities to chemicals for which specific behaviors have been empirically determined (11). QSAR
models are used to estimate properties when empirical data are not available.
Radionuclides: An unstable isotope of an element that decays or disintegrates spontaneously,
emitting radiation (17).
Recombination events: Chromosomal recombination during reduction division h the formation
of sex cells, a major mechanism of eukaryotic genetic variability (43).
Glossary
G-9
t
-------
CCL CP Work Group Report
I
Reference Concentration /Dose (RfC/D): Term used for an estimate of air exposure
concentration / daily oral exposure dose to the human population (including sensitive subgroups) that
is likely to be without appreciable risk of deleterious effects during lifetime. RflC/Ds have been
derived for acute, subchronic, and chronic exposure scenarios (25,31).
Regression analysts: A statistical procedure for estimating unknown model parameter values,
and their uncertainty, based on available data (5).
Risk: The probability of realization of adverse consequences or events (12).
Rule-based system: A priori classification models that use various features or parameters
[attributes] of a contaminant and weigh and combine these features according to an algorithm that is
decided upon in advance-usually as a result of some expert judgment (1).
Sensitive populations: Groups of individuals who respond biologically at lower levels of
exposure to a contaminant in drinking water or who have more serious health consequences than the
general population. These groups may include infants, children, pregnant women, the elderly, or
individuals with a history of chronic illness (52).
Sequelae: Conditions following as a consequence of a disease (47).
Severity: An NRC attribute defined as the degree to which a potential contaminant can cause an
adverse health effect. (1). see Attribute,
Scaling: Changing the units of measurement, usually for the numerical stability of an
algorithm. (37).
Slime layer (polysaccharide): A diffused layer of polysaccharide exterior to the bacterial
cell wall (49, 41).
Slope Factor (SF): Value, in inverse concentration or dose units, derived from the slope of a
inhalation dose-response curve; in practice, limited to carcinogenic effects with the curve assumed to
be linear at low concentrations or doses. The product of the slope factor and the exposure is taken to
reflect the probability of producing the related effect (25).
Spatial prevalence: The proportion of locales in which the contaminant can be found (1).
Temporal prevalence: The average fraction of time that a contaminant is found in a given locale
(1).
Toxicogenomics: The collection, interpretation, and storage of information about gene and
protein activity in order to identify toxic substances in the environment, and to help treat people at the
greatest risk of diseases caused by environmental pollutants or toxicants and to set policies that will
protect sensitive populations (40).
Toxicological study endpoint: A data element representing a summary statistic or observation
from an empirical health effects study. Population response endpoints express chemical-induced effect
concentrations in relation to a specified level of response among the test population (e.g., and LD49).
Glossary
G-10
-------
CCL CP Work Group Report
Toxicity threshold endpomts express concentrations at the onset of an observed adverse effect,
irrespective of the level of response (Health Effects Commonalities, June 2003).
Toxin: For microorganisms, a noxious or poisonqus substance that is either (1) an integral part of
the cell or tissue, (2) an extracellular product (e.g. exotoxin), or (3) represents a combination of the
two situations formed or elaborated during the metabolism, death, or growth of certain
microorganisms (e.g. endotoxin). Toxin producers include Clostridium botulimm, E. coli serovar
such as 0157:H7, Shigella, Vibrio cholerae, and some cyanobacteria (49,5).
Transposon: Genetic element that can transpose (move) to a different position in a genome or to
another genome. Transposons can be divided into two classes based on their structure. Elements of
one class, known as compound or composite transposons, have copies of insertion elements (IS
elements) at each end, transposition of composite transposons requires transposases coded by one of
their terminal IS elements. Transponsons of the second class have terminal inverted repeats of about
30 base pairs and do not contain sequences from IS elements (49,42).
Training Data Set: A prototype algorithm is developed using a training data set. A training set
comprises information about the problem to be solved as input stimuli. The training sets are used in an
iterative process to allow the prototype to 'learn' how to weight inputs until cfassification by the
algorithm is adequate. (19). A training data set consists of numerous contaminants, together with their
health effects data/information, occurrence data / information, scored attributes, and listing status (on
or off die CCL) (this report).
Transparency: Transparency provides explicitness in the risk assessment process. It ensures that
any reader understands all the steps, logic, key assumptions, limitations, and decisions in the risk
assessment, and comprehends the supporting rationale that lead to the outcome. Transparencyachieves
full disclosure in terms of:
a) the assessment approach employed
b) die use of assumptions and their impact on the assessment
c) the use of extrapolations and their impact on the assessment
d) the use of models vs. measurements and their impact on the assessment
e) plausible alternatives and the choices made among those alternatives
f) the impacts of one choice vs. another on the assessment
g) significant data gaps and their implications for the assessment
h) the scientific conclusions identified separately from default assumptions and policy calls
i) the major risk conclusions and the assessor's confidence and uncertainties in them
Glossary
G-ll
t
-------
t
CCL CP Work Group Report
j) the relative strength of each risk assessment component and its impact on the overall
assessment (e.g., the case for the agent posing a hazard is strong, but the overall assessment of
risk is weak because the case for exposure is weak) (3).
Treatment Technique, TT: A required process intended to reduce (he level of a contaminant in
drinking water (21).
Validation: process of assessing whether the predictions or conclusions reached are correct (2).
Virulence: The degree of pathogenicity; the degree of intensity or severity of disease produced
by a pathogen. Severity of disease does not necessarily reflect severity of infection. See also
virulence factor activity relationships (VFARs) (5).
Virulence factors: The variability in the virulence of a pathogen may be characterized by one or
more its biological characteristics. These characteristics, sometimes referred to as virulence factors,
may include genetic elements, proteins, toxins, attachment and invasion mechanisms, metabolic
pathways, and/or other architectural and biological characteristics of the pathogen. See also virulence
factor activity relationships (VFARs) (5).
Virulence factor activity relationships (VFARs): a novel approach proposed for investigation to
identify emerging waterborne microorganisms for the CCL. The terminology VFAR was coined to
refer to a presumed or demonstrated linkage or relationship between the presence of identified
genomic sequences of a microorganism and the ability of the microorganism to cause harm in
humans. When the linkage or correlation of virulence factors with potency, pathogenicity, and/or the
intensity/severity of disease yields a consistent statistical relationship, the relationship (i.e., the
VFAR) for known pathogens may then be used as a predictive model for assessing the potency,
pathogenicity, and/or virulence properties of related microbes (5).
Zoonotic: transmissible from animals to humans under natural conditions; pertaining to or
constituting a zoonosis (50).
Glossary G-12
-------
CCICP Work Group Report
References
Discussion Draft for NDWAC CCL Workgroup. Dimensioning the Microbial Universe.
February 25, 2003.
Discussion Draft for NDWAC CCL Workgroup. Defining me Microbial Universe. Jury 7,
2003.
Taylor, Latham and Woolhouse. 2001. Risk factors for human disease emergence (Appendix
A). Phil. Trans. R. Soc. Lond. B 256:983-98.
Wang, D., L. Coscoy, M. Zylberberg, P. C. Avila, H. A. Boushey, D. Ganem, and J. L.
DeRisi. 2002. Microarray-based detection and genotyping of viral pathogens. PNAS
99(24): 15687-15692.
Glossary References
1) National Academy of Sciences, National Research Council. 2001. Classifying Drinking
Water Contaminants for Regulatory Consideration. National Academy Press.
Washington, DC.
2) Joint OECD/IPCS Project on the Harmonization of Hazard/Risk assessment Terminology,
available at: http ://vAvw.who.int/terminologyAer/PDF documents/tsh.pdf.
3) EPA, December 2000. Risk Characterization Handbook, EPA 100-B-00-002, available at:
http://wvw.epa.gov/osp/spc/2riskchr.htm
4) USEPA. August 1996. IVoposed Guidelines for Ecological Risk Assessment. Available on
the web at: http://www.epa.eov/ORD/WebPubs/ecorisk/giiide.pdf
5) EPA 2002. Glossary.(Craun, Stine, et al)
6) EPA 1999. Class V: Underground Injection Control Study. Appendix E: Contaminant
Persistence and Mobility factors. Available on the web at:
http://wvm.epa.gov/ogwdwOOO/mc/cl5studv.html
7) Glossary of Terms Used in ITER, Available athttp://iter.ctcnet.ne^ublicurl/glossarv.htm
8) Disaster Advise Glossary, Phoenix International Consultancy Ltd., available on the web at:
http://www.disasteradvice.co.uk/glossarv search.asp
9) New England Foundation for Research, Science and Technology Glossary, available at:
http://contamsites.limdcareresearch.co.nz/glossarv.htm
10) Merriam-Webster Dictionary Online, Available at: http://www.m-w.com/home.htai
11) Fundamentals of Aquatic Toxicology. Effects. Environmental Fate, and Risk Assessment
£led, Rand G~NC 1995, Taylor & Francis Publishers since 1798, 1101 Vermont
Ave., N.W., Suite 200, Washington, D.C. 20005-3511.
References
R-l
-------
CCL CP Work Group Report
I
12) U.S. EPA, Office of Ground Water and Drinking Water. 2002. A Dictionary of Technical
and Legal Terms Related to Drinking Water. Available at:
http://www.epa.gov/safewater/pubs/gloss2.html
13) Chemical Abstract Service (CAS) website, available at: www.cas.org/EO/reesvs.html
14) NSF (National Sanitation Foundation) International, Drinking Water Standards Set,
Document number: NSF/ANSIDWA Set NSF International, available for sale at:
http://www.techstreet.com/cgt-bin/detail7product id=100428
15) Threshold of Regulation Policy- Deciding Whether A Pesticide with a Food Use Pattern
Needs a Tolerance. EPA 1999. October 18,1999. Available at:
http://www.epa.gov/fedrgslr/EPA-PEST/1999/October/Dav-27/6Q41 .odf
16) Michael Dourson, personal communication.
17) U.S. Nuclear Regulatory Commission Glossary. Available at:
http://www.nrc.gov/reading-rm/basic-ref/glossarv.htmltfR
18) Qian, S.; and Anderson C., Exploring Factors Controlling the Variability of Pesticide
Concentrations in the Willamette River Basin Using Tree-Based Models.
Environmental Science and Technology. Vol. 33, No. 19,1999.
19) Envistat Data Products, Neural Network Glossary. Available at:
http://envisat.esa.int/dataproducts/meris/CNTR5-2-5.htm
20) National Primary Drinking Water Regulations: Consumer Confidence Reports. Federal
Register Documents. Available at: http://www.epa.gov/OGWDW/ccr/ccr-frne.html
21) 2002 Edition of the Health Advisories and Drinking Water Standards. Available at:
http://www.epa.gov/waterscience/drinking/standards/dwstandards.pdf
22) Stedman's Online Medical Dictionary. 27th Edition. 2003. Available at:
http://www.stedmans.com/section.cfin/44
23) Building the CCL Universe and Associated Issues, Discussion Draft, May 22,2003,
Available on USEPA National Drinking Water Advisory Council, Contaminant
Candidate List Classification Process Work Group Web page, under: Data
Workgroup, activity materials.
24) Draft Proposed Universe to PCCL Process, Discussion Draft, May 13,2003. Available on
USEPA National Drinking Water Advisory Council, Contaminant Candidate List
Classification Process Work Group Web page, under: Methods Workgroup, activity
materials.
25) Glossary for Chemists of Terms Used in Toxicology, Pure and Applied Chemistry, Vol.
65, No.9, pp. 2003-2122,1993. Available at:
www.sis.nlm.nih.eov/Glossary/main.html
26) Glossary of Terms Used in ATSDR Chemical Profiles. Available within each chemical
profile at: www.atsdr.cdc.eovAoxpro2.hmil
References
R-2
-------
CCL CP Work Group Report
27) Epidemiology in Medicine. Little, Brown and Company, Boston/Toronto. 1987.
28) The Pesticide Management Education Program at Cornell University, Extoxnet Glossary.
2002. Available at: pmep.cce.comell.edu/^rofiles/extoxnet/riB/extoxnetglossarv.htinl
29) Agency for Toxic Substances and Disease Registry (ATSDR), Minimal Risk Levels
(MRL's) for Hazardous Substances. Available at: w\w\-.atsdr.cdc.eov/mrls.html
30) U.S. Food and Drug Administration's Food Contact Substance Notification Program's
CEDI/ADI Database, available at: httD://w]ww.cfsan.fda.eov/~dms/opa-edi.html
31) Joint Meeting on Pesticide Residues - Inventory of Pesticide Evaluations; available at:
http://www.inchem.org/docuinents/inipr/impeval/impr2001.htai
32) EPA 2000. Benchmark Dose Technical Guidance Document, EPA/630/R-00/001,
External Review Draft, October. Available at
http://www.epa.goy/ncea/bnchrmVbmds peer.htm
33) Dictionary of Epidemic logy. University of Cambridge. Last updated: 1/1/2003. Available
on the Internet at: http:/Avww.albanv.net/~^ic/gloss96.html# O
34) Human Genome Project Information. Genome Glossary. DOE Human Genome Program.
Available on the Internet at:
http://www.oml.gov/rechResources/Human Genome/glossary/
35) INTERNATIONAL UNION OF PURE AND APPLIED CHEMISTRY, Clinical
Chemistry Division Commission on Toxicology. GLOSSARY FOR CHEMISTS OF
TERMS USED IN TOXICOLOGY. 1993. Available at:
http://sis.nhti.nih.gov/Glossarv/main.html
36) California Department of Health Services. Review of MCLs in Response to PHGs. Last
Update: June 9,2003. Available at:
http://www.dhs.cahwnet.eov/ps/ddwem/ch-emicals/PHGs/
37) Mathematical Programming Glossary. 1996-2004. Available at:
http://carbon.cudenver.eduA~hgreenbe/elossarv/tndex.php)
38). MathWorld: Wolfram Research. 1999-2004. Available at:
http://raathworld.wi3lfram.com/>
(39) US EPA. 2000 Peer Review Handbook 2nd Edition. EPA 100-B-OO-OOl.
40) MedicineNet. Available on the internet at:
http://www.medtenns.com/scripl/main/art,asp?articlekev=30715
41) Department of Medical Oncology, University of Newcastle upon Tyne. The On-line
Medical Dictionary. 1997-2002. Available on the Internet at:
http://cancerweb.ncl.ac.uk/omd
42) The Encyclopedia of Molecular Biology, 1995, Blackwell Science, Inc. 237 Main Street,
Cambridge, MA 02141. Editor in chief: Sir John Kendrew
References R-3
*
-------
t
CCL CP Work Group Report
43) Hacker, J; Carniel, E; Ecological fitness, genomic islands and bacterial pathogenicity: A
Darwinian view of the evolution of microbes, EMBO, 2 (5). 2001. pp this report6-371.
44) Integrated Risk Information System (IRIS). USEPA. 2004. Available on the web at:
http://www.epa.gov/iris/gloss8.htm
45) Lackie, J.M and J.A.T. Dow. The Dictionary of Cell and Molecular Biology. 1999.
London: Harcourt Brace and Company. Internet edition maintained by Julian Dow
and may be accessed at: http://www.mblab.gla.ac.uk/dictionary/
46) Mahillon, J; Chandler, M. Insertion Sequences. Microbiology and Molecular Biology
Reviews, Sept. 1988, p. 725-774.
47) Medine Plus Health Information, Cancerweb Dictionary, Available at:
http://www.nlm.nih.gov/medlineplus/dictionaries.html last updated 2002.
48) National Center for Biotechnology Information, BLAST info More Information Glossary,
revised June 7,2000, available at:
http://www.ncbi.nlm.nih.gov/Education/BLASTinfo/glossary2.htail
49) NDWAC WAR Subgroup Findings; Attachment 8. Glossary of GenBank Keywords and
Terms used in the December Findings Document; Dec 6,2002
50) Saunders, W.B.: Harcourt Health Sciences. Dorland's Illustrated Medical Dictionary.
2002. Available on the Internet at:
http://www.mercksource.com/pp/us/cns/cns health library frame.isp?pg=/pp/us/cns/
ens hi dorlands.isp?pg=/pp/us/common/dorlands/dorland/dmd a-b O0.htm&cd=3d
51) Sydney Catchment Authority (SCA) Annual Water Quality Monitoring Report 2000-
2001: Technical Terms. Available on the internet at:
http://www.sca.nsw.gov.au/awqr/glossarv-technical-pl35.html
52) US EPA. 2000, Report to Congress: EPA Studies on Sensitive Subpopulations and
Drinking Water Contaminants EPA 8I5-R-00-015. Office of Water and Office of
Research and Development. Washington DC 20460.
References R-4
-------
CCL CP Work Group Report
Appendix A
Summary of Recommendations from Classifying
Drinking Water Contaminants for Regulatory
Consideration
Recommendations of the VRC Committee on Drinking Water Contaminants (NRC,
2001)
Page
Number
Executive Summary
The committee recommends that EPA develop and use a twostep process for creating
future CCLs. In summary, a broadly defined universe of potential drinking water
contaminants is first identified, assessed, and culled to a preliminary CCL (PCCL) using
simple screening criteria and expert judgment. All PCCL contaminants are then
individually assessed using a "prototype" classification tool in conjunction with expert
judgement to evaluate the likelihood that they could occur in drinking water at levels and
frequencies that pose a public health risk to create the corresponding CCL.
The committee recommencs that this two-step process be repeated for each CCL
development cycle to account for new data and potential contaminants that inevitably
arise over time.
All contaminants that have not been regulated or removed from the existing CCL should
be automatically retained on each subsequent CCL.
The committee recommencs that the process for selecting contaminants for future CCLs
be systematic, scientifically sound, and transparent. The development and implementation
of the process should involve sufficiently broad public participation.
The committee recommends that the definition of vulnerable subpopulations should not
only comply with the amended language of the SDWA, it should also be sufficiently
broad to protect public health.
EPA should begin by considering a broad universe of chemical, microbial, and other types
of potential drinking water contaminants and contaminant groups.
EPA should rely on databases and lists that are currently available and under development
along with other readily available information to begin the identification of the universe of
potential contaminants that may be candidates for inclusion on the PCCL.
As an integral part of the development process for future PCCLs and CCLs, all
information used from existing or created databases or lists should be compiled in a
consolidated database to provide a consistent mechanism for recording and retrieving
information on the contaminants under consideration
To generally assist in the identification of the universe of potential contaminants and a
PCCL, the committee recommends EPA consider substances based on their commercial
use, environmental location, or physical characteristics.
The committee recommends the use of a Venn diagram approach to conceptually
distinguish a PCCL from the broader universe of potential drinking water contaminants.
4
4
4
6
6
7
7
8
8
8
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-i
*
-------
CCL CP Work Group Report
Executive Summary (continued)
Regarding the development of screening criteria for health effects, the committee
recommends that human data and data on whole animals be used as indicators of
demonstrated health effects and that other lexicological data and data from experimental
models that predict biological activity be used as indicators of potential health effects.
A variety of metrics could be used to develop screening criteria for occurrence of
contaminants in drinking water. These are identified in a hierarchical framework in the
committee's first report and include (1) observations in tap water, (2) observations in
distribution systems, (3) observations in finished water of water treatment plants, (4)
observations in source water, (5) observations in watersheds and aquifers, (6) historical
contaminant release data, and (7) chemical production data. The committee recommends
that the first four of these should be used as indicators of demonstrated occurrence, and
information that comes from items 5-1 should be used to determine potential occurrence.
10
Each PCCL should be published and thereby serve as a useful record of past PCCL and
CCL development and serve as a starting point for the development of future PCCLs.
10
Development of the first PCCL should begin as soon as possible to support the
development of the next (2003) CCL; each PCCL should be available for public and other
stakeholder input (especially through the Internet) and should undergo scientific review.
10
The committee recommends EPA develop and use a set of attributes to evaluate the
likelihood that any particular PCCL contaminant or g-oup of related contaminants could
occur in drinking water at levels and frequencies that pose a public health risk.
11
These contaminant attributes should be used in a prototype classification approach such as
described in Chapter 5 and in conjunction with expert judgment to help identify the
highest priority PCCL contaminants for inclusion on a CCL.
11
Should EPA choose to adopt a prototype classification approach for the development of
future CCLs, the committee recommends that options for developing aid scoring
contaminant attributes should be made available for public and other stakeholder input
and undergo scientific review.
12
The assessment of severity should be based, when feasible, on plausible exposures via
drinking water. The committee also recommends that EPA give consideration to different
severity metrics such as ranking through use of either quality adjusted or disability
adjusted life years lost from exposure to a contaminant.
12
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-2
-------
CCL CP Work Group Report
Executive Summary (continued)
Regarding the assessment cf contaminant prevalence, in some cases (particularly where
contaminants have been included on a PCCL on the basis of potential occurrence rather
than demonstrated occurrence), insufficient information will be available to directly assess
temporal or spatial prevalence (or both). Thus, EPA should consider the possibility of
including information on temporal and regional occurrence to help determine (score
PCCL) contaminant prevalence. The issue of changing (or incorporating) "thresholds" for
contaminant detection, rather than relying on continually decreasing detection limits, is
one that needs explicit attention and discussion by EPA and stakeholders.
12
As existing and readily available databases may not be sufficient to rapidly and
consistently score health effect and occurrence attributes for individual PCCL
contaminants, all information from existing or created databases or lists used in the
development of a CCL and PCCL should be compiled in a consolidated database (as
previously recommended).
12
Contaminant databases used in support of the development of future CCLs should report
summary statistics on all data collected, not only the quantifiable observations. In this
regard, EPA should formalize a process for reporting means and/or medians from data
with large numbers of non-idetect observations. In addition, EPA may want to consider
providing other measures of concentration in water supplies such as die 95h percentile of
contaminant concentration.
12
The committee presents two alternative models for use in the prototype classification
scheme—a linear model and a neural network. Although the neural network performed
better than the linear model, the committee cannot at this time make a firm
recommendation as to which model EPA should use as a prototype classification scheme
due to uncertainties in the training data set used by the committee. Thus, the committee
recommends that EPA explore alternative model formulations and be cognizant of the
dangers of overfilling and loss of generalization.
13
The committee strongly recommends that EPA greatly increase the size of the training
data set that was used illustratively in this report to improve predictive capacity.
14
EPA should accuratery and consistently assign attribute scores for all contaminants under
consideration, i.e., contaminants in the training data set as well as contaminants to which
the prototype classification algorithm will be applied for a classification determination.
To do this, EPA will need to collect and organize available data and research for each
PCCL contaminant and document the attribute scoring scheme used to help ensure a
transparent and defensible process.
14
EPA will need to withhold :ontamuiants from inclusion in the training data set to serve as
validation test cases that can assess the predictive accuracy of any classification algorithm
developed for use in the creation of future CCLs.
14
Appendix A - Summary oj Recommendations from Classifying Drinking Water
Contaminants
A-3
-------
t
CCL CP Work Group Report
t
Executive Summary (continued)
The committee recommends that EPA should use several training data sets to gauge the
sensitivity of the method as part of its analysis and documentation if a classification
approach is ultimately adopted and used to help create future CCLs.
The committee recommends the establishment of a scientific Virulence Factor Activity
Relationships (WAR) Working Group on bioinformatics, genomics, and proteomics, with
a charge to study these disciplines on an ongoing basis, and to periodically inform the
Agency as to how these disciplines can affect the identification and selection of drinking
water contaminants for future regulatory, monitoring, and research activities.
The committee recommends that the findings of this report, and especially that of the
Biotechnology Research Group (the Interagency Report on the Federal Investment in
Microbial Genomics), should be made available to a WAR Working Group at its
inception.
The WAR Working Group should be charged with the task of delineating specific steps
and related issues and timelines needed to take WARs beyond the conceptual framework
of this report to actual development and implementation by EPA.
With (he assistance of the Working Group, EPA should identify and fund pilot
bioinformatic projects that use genomics and proteomics to gain practical experience that
can be applied to the development of WARs while simultaneously dispatching its charges
outlined in the two previous recommendations.
EPA should employ and work with scientific personnel trained in the fields of
bioinformatics, genomics, and proteomics to assist the Agency in focusing efforts on
identifying and addressing emerging waterborne microorganisms.
EPA should participate fully in all ongoing and planned U.S. federal government efforts in
bioinformatics, genomics, and proteomics as potentially related to the identification and
selection of waterbome pathogens for regulatory consideration.
15
19
19
19
19
19
19
Appendix A -Summary of Recommendations from Classifying Drinking Water
Contaminants
A-4
-------
CCL CP Work Group Report
Chapter 1: Drinking Water Contaminant Candidate List: Past, Present, and Future
An ideal CCL development process would include the following features:
• Meet all statutory requirements of the SDWA Amendments of 1996, such as
requirements for consultation with the scientific community and opportunities
for public comment.
• Begin with identification of the entire universe of potential drinking water
contaminants prior to any attempts to rank or sort them.
• Address risks from all potential routes of exposure to water supplies, including
dermal contact ar d inhalation as well as ingestion.
• Use the same identification and selection process for chemical, microbial, and all
other types of potential drinking water contaminants.
• Use mechanisms for identifying similarities among contaminants and
contaminant classes to assess potential risks of individual contaminants-
especially emerging contaminants.
» Result in CCLs that contain only contaminants that when regulated would
reduce disease, disability, and death, and excludes contaminants that have few or
no adverse effects on human health (e.g., contaminants removed or detoxified
through conventional drinking water treatment methods).
As recommended in its second report titled Identifying Future Drinking Water
Contaminants, the committee continues to recommend that EPA develop and use a
two-step process for creating future CCLs. In summary, a broadly defined universe of
potential drinking water contaminants is first identified, assessed, and narrowed to a
preliminary CCL (PCCL) using simple screening criteria and expert judgment. All PCCL
contaminants are then individually assessed using a "prototype" classification tool in
conjunction with expert judgment to evaluate the likelihood that they could occur in
drinking water at levels and frequencies that pose a public health risk to create the
corresponding (and much smaller) CCL. The "universe" of potential drinking water
contaminants includes: (1) naturally occurring substances, (2) water-associated microbial
agents, (3) chemical agents, (4) products of environmental transformation of chemical
agents, (5) reaction byproducts, (6) metabolites in the environment, (7) radionuclides, (8)
biological toxins, and (9) fibers. PCCL includes: (1) contaminants that are demonstrated
to occur in drinking water iind demonstrated to cause adverse health effects, (2)
contaminants that are demonstrated to occur hi drinking water and have the potential to
cause adverse health effects, (3) contaminants that .are demonstrated to cause adverse
health effects and have the potential to occur in drinking water, and (4) contaminants that
have the potential to occur in drinking water and the potential to cause adverse health
effects
The committee also continues to recommend that this two-step process be repeated
for each CCL development cycle to account for new data and potential contaminants
that inevitably arise over time. In addition, all contaminants that have not been
regulated or removed from existing CCL should be automatically retained on each
subsequent CCL.
42
43
44
Appendix A -Summary of Recommendations from Classifying Drinking Water
Contaminants
A-5
-------
CCL CP Work Group Report
Chapter 2: Sociopolitical Considerations for Developing Future CCLs
The committee believes that public participation procedures should satisfy the criteria of
equity, fairness, and justice. General recommendations to facilitate public participation in
environmental programs are provided in The Model Plan for Public Participation,
developed for the EPA by the Public Participation and Accountability Subcommittee of
the National Environmental Justice Advisory Council. In addition, Hampton (1999)
provides the following recommendations:
• The public should be involved in defining the process of participation.
• Public involvements should be early in the process (e.g., at the time of agenda
setting or when value judgements become important to the process).
• Participants should have access to appropriate resources such as the information
that is necessary in order to participate fully in the process, access to scientists,
technical assistance, and sufficient time to prepare for the deliberations.
• Prior agreement should be reached with the participants as to how the output of
the procedure (e.g., recommendations, decisions) will be used and how it will
affect agency policy decisions.
68
The committee recognizes that the development of a PCCL from the universe of potential
drinking water contaminants, as well as contaminant movement from a PCCL to the
corresponding CCL, is a complex task requiring numerous difficult classification
judgements in a context where data are often uncertain or missing. In order to be
scientifically sound as well as publicly acceptable, the process for developing future CCLs
must depart considerably from the process used to develop the first (1998) CCL. The
committee recommends that the process for selecting contaminants for future CCLs
be systematic, scientifically sound, and transparent. The development and
implementation of the process should involve sufficiently broad public participation.
69
The ultimate goal of the contaminant selection process is the protection of public health
through the provision of safe drinking water to all consumers. To meet this goal, the
selection process must place high priority on the protection of vulnerable subpopulations.
69
The committee recommends that the list of vulnerable subpopulations described in
the amended SDWA should not be seen as a minimum list, but rather as several
examples of possible vulnerable subpopulations. A minimum list must go much further
than this. The definition of vulnerable subpopulations should not only comply with the
amended language of the SDWA, it should also be sufficiently broad to protect public
health and, in particular, EPA should consider including (in addition to those subgroups
mentioned as examples in the amended SDWA) all women of childbearing age, fetuses,
the immuno-compromised, people with acquired or inherited genetic disposition that
makes them more vulnerable to drinking water contaminants, people who are
exceptionally sensitive to an array of chemical contaminants, people with specific medical
conditions that make them more susceptible, people with poor nutrition, and people
experiencing socioeconomic hardships and racial/ethnic discrimination.
69
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-6
-------
CCL CP Work Group Report
Chapter 2: Sociopolitical Considerations for Developing Future CCLs (continued)
Transparency should be incorporated into the design and development of the classification
and decision-making process for future CCLs in addition to being an integral component
in communicating the details of the process to the public. Otherwise, the public may
perceive that the process is subject to manipulation to achieve or support desired results.
Therefore, sufficient information should be provided such that citizens can place
themselves in a similar position as decision-makers and arrive at their own reasonable and
informed judgements. This may require making available to the public the software and
databases used in the process.
The central tenet that the public is, in principle, capable of making wise and prudent
decisions should be recognized and reflected in the choice of public participation
procedures used to help create future CCLs. A "decide-announce-dcfend" strategy that
involves the public only after the deliberation process is over is not acceptable.
Substantive public involvement should occur throughout the design and implementation
of the process. EPA should strive to "get the right participation" (i.e., sufficiently broad
participation that includes the range of interested and affected parties) as well as "get the
participation right" (e.g., incorporating public values, viewpoints, and preferences into the
process).
Chapter 3: The Universe of Potential Contaminants to the Preliminary CCL
In general, greater consideration should be given to including substances on the PCCL
that cause serious, irreversible effects as opposed to those that cause less serious effects.
The committee is not suggesting that less serious health effects such as cholinesterase
inhibition should be ignored, however, it recognizes that health effects such as cancer or
birth defects may be given greater weight.
The committee believes that generally contaminant concentration alone should not be used
as a relevant metric for culling from the universe to the PCCL, although the committee
recognizes that some consideration of concentration may be needed as analytical
procedures continue to reduce detection limits. EPA may want to consider binary data,
such as found or not found in public water systems, for selecting chemicals for the PCCL
from the universe. Also, the committee believes th|at frequency over time should not be
used as the sole relevant metric for this step as this may place undue emphasis on
contaminants that are repeatedly found and eliminate those that may have a significant
impact but occur infrequently. The committee believes that prevalence at a large number
of public water systems or prevalence at systems that serve large numbers of people is an
important metric to determine inclusion into the demonstrated occurrence category.
Of the metrics that serve as indicators of potential occurrence, the committee recommends
that EPA use production 01 release data, combined with physical properties, to serve as
useful indicators of the potential for chemical occurrence in watersheds and aquifers.
70
70
85
86
86
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-7
-------
CCL CP Work Group Report
Chapter 3: The Universe of Potential Contaminants to the Preliminary CCL (continued)
For chemicals, a binary approach would serve to categorize the universe of chemicals
being produced commercially (i.e., would not include byproducts or chemicals formed in
the environment) into four bins for potential occurrence. The committee recommends that
if such an approach were used for commercial chemicals, all chemicals except those with
those with both low production volume and low water solubility should be considered for
inclusion on a PCCL.
87
EPA should review contaminants already included in the potential occurrence category
("ring") to determine if they have any mportant environmental degradation products,
production or reaction by-products or metabolites in the environment that should also be
considered for inclusion on the potential occurrence list.
88
EPA should also review naturally occurring substances and fibers to determine if any of
them need to be included on the potential occurrence list. EPA may also want to review
data for specific watersheds and aquifers to determine if any other contaminants should be
included on the potential occurrence list.
88
In keeping with its inclusive nature, the PCCL should not be expected to maintain a more-
or-less fixed number of potential drinking water contaminants.
88
EPA should begin by considering a broad universe of chemical, microbial, and other
types of potential drinking water contaminants and contaminant groups. The total
number of contaminants in this universe is likely to be on the order of tens of thousands of
substances and microorganisms, given that the Toxic Substances Control Act inventory of
commercial chemicals alone includes about 72,000 substances (NRC, 1999b). This
represents a dramatically larger set of substances to be initially considered in terms of
types and numbers of contaminants than used for the creations of the 1998 CCL.
89
EPA should rely on databases and lists that are currently available and under
development along with other readily available information to begin identifying the
universe of potential contaminants that my be candidates for inclusion on the PCCL.
For example, EPA should consider using the Endocrine Disrupter Priority-Setting
Database (EDSPD) database to help develop future PCCLs (and perhaps CCLs). While
relevant databases and lists exist for many "universe categories" of potential drinking
water contaminants, others have no lists or databases (e.g., products of environmental
degradation). Thus, EPA should initiate work on a strategy for filling the gaps and
updating the existing databases and lists of contaminants (e.g., through involvement
of the National Drinking Water Advisory Council or panels of experts) for future
CCLs. This strategy should be developed with public, stakeholder, and scientific
community input.
90
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-8
-------
CCL CP Work Group Report
Chapter 3: The Universe of Potential Contaminants to the Preliminary CCL (continued)
As an integral part of the development process for future PCCLs and CCLs, all
information from existing or created databases or lists used should be compiled in a
consolidated database that would provide a consistent mechanism for recording and
retrieving information on the contaminants under consideration. Such a database
could function as a "master list" that contains a detailed record of how the universe of
potential contaminants was: identified and how a particular PCCL and its corresponding
CCL were subsequently created. It would also serve as a powerful analytical tool for the
development of future PCCLs and CCLs. As a starting point, the committee
recommends that EPA review its developing EDSPD database to determine if it can
be expanded and used as this consolidation database or whether it can serve as a
model for the subsequent development of such a database Regardless, the (re)design,
creation, and implementation of such a database should be made in open cooperation with
the public, stakeholders, and the scientific commui|iity.
90
To generally assist in the identification of the universe of potential contaminants and
a PCCL, the committee recommends EPA consider substances based on their
commercial use, environmental location, or physical characteristics. EPA should be
as inclusive as possible in narrowing the universe pf potential drinking water
contaminants down to a PCCL. The committee envisions that a PCCL would contain on
the order of a few thousand individual substances and groups of related substances,
including microorganisms, for evaluation and prioritization to form a CCL. However,
preparation of a PCCL should not involve extensive analysis of data, nor should it directly
drive EPA's research or monitoring activities.
90
The committee recommends the use of a Venn diagram approach to conceptually
distinguish a PCCL from the universe of potential drinking water contaminants.
However, due to the extremely large size of the universe of potential drinking water
contaminants, well-conceived screening criteria remain to be developed that can be
rapidly and routinely applied by EPA in conjunction with expert judgement to cull the
universe to a much smaller PCCL. Thus, the PCCL should include those contaminants
that have a combination of characteristics indicating that they are likely to pose a public
health risk through their occurrence in drinking waiter. These characteristics are
demonstrated or potential occurrence in drinking water and demonstrated or potential
ability to cause adverse hesilth effects.
91
Regarding the development of screening criteria for health effects, the committee
recommends that human data and data on whole animals be used as indicators of
demonstrated health effects and that other toxioological data and data from
experimental models that predict biological activity be used as indicators of potential
health effects.
91
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-9
-------
CCL CP Work Group Report
Chapter 3: The Universe of Potential Contaminants to the Preliminary CCL (continued)
A variety of metrics could be used to develop screening criteria for occurroice of
contaminants in drinking water. These are identified in a hierarchical framework in the
committee's first report (NRC, 1999a) and include (1) observations in tap water, (2)
observations in distribution systems, (3) observations in finished water of water treatment
plants, (4) observations in source water, (5) observations in watersheds and aquifers, (6)
historical contaminant release data, and (7) chemical production data. The committee
recommends that the first four of these should be used as indicators of demonstrated
occurrence, and information that comes from items 5-7 should be used to determine
potential occurrence. For commercial chemicals, their potential for occurrence in
drinking water may be estimated using a combination of production volume information
and water solubility. Most likely occurrence would involve high production volume
chemicals with high water solubility.
91
A new PCCL should be generated for each CCL development cycle to account for
new data and emerging contaminants.
91
Each PCCL should be published and thereby serve as a useful record of past PCCL
and CCL development and serve as a starting point for development of future
PCCLs.
91
Development of the first PCCL should begin as soon as possible to support the
development of die next (2003) CCL; each PCCL should be available for public and
other stakeholder input (especially through the Internet) and should undergo
scientific review.
91
Chapter 4: PCCL to CCL: Attributes of Contaminants
To overcome the limitations of current chemical fate and persistence models, the
committee recommends consideration of three general characteristics of contaminants that
would foster their persistence and/or mobility in water systems:
• High potential for amplification by growth under ambient conditions (applies to
microbial contaminants and not to chemicals).
• High solubility in water (applies primarily to chemicals); though transportability
of microorganisms may be assessed through sedimentation velocities and size
and adsorption capabilities.
• Stability in water, i.e., resistence to degradation via mechanisms such as
hydrolysis, photolysis, or biodegradation in the case of chemicals; death or the
ability to produce non-culturable states or resistant states (e.g., spores and cysts)
in the case of microorganisms.
100
Expanding upon the Chapter 3 recommendation for EPA to review the EDSPD database
to determine if it can be used to help develop a PCCL and perhaps help select PCCL
contaminants for inclusion on a CCL, the committee also recommends that EPA
consider the possibility of including information on temporal and regional
occurrence.
103
Appendix A -Summary of Recommendations from Classifying Drinking Water
Contaminants A-10
-------
CCL CP Work Group Report
Chapter 4: PCCL to CCL: Attributes of Contaminants (continued)
The committee recommends EPA develop and use a set of attributes to evaluate the
likelihood that any particular PCCL contaminant or group of related contaminants
could occur in drinking water at levels and frequencies that pose a public health risk.
More specifically, these contaminant attributes should be used in a prototype classification
algorithm approach such as described in Chapter 5 and in conjunction with expert
judgement to help identify the highest priority PCCL contaminants for inclusion on a
CCL.
Should EPA choose to adopt a prototype classification approach for the development
of future CCLs, the committee recommends that options for developing and scoring
contaminant attributes should be made available for public and other stakeholder
input and undergo scientific review.
The assessment of severity should be based, when feasible, on plausible exposures via
drinking water. The committee also recommends that EPA give consideration to
different severity metrics such as ranking through use of either quality adjusted or
disability adjusted Hfe years lost from exposureto a contaminant
Regarding the assessment of contaminant prevalence, in some cases (particularly where
contaminants have been included on a PCCL on the basis of potential occurrence rather
than demonstrated occurrence), information will often be insufficient to directly assess
temporal or spatial prevalence (or both). Thus, EPA should consider the possibility of
including information on temporal and regional occurrence to help determine (score
PCCL) contaminant prevalence. When prevalence cannot be assessed, this attribute
must then go unscored and the attribute persistence/mobility used in its stead. The issue
of changing (or incorporating) "thresholds" for contaminant detection, rather than relying
on continually decreasing detection limits, is one that needs explicit attention and
discussion by EPA and stakeholders.
Existing and readily available databases may not be sufficient to rapidly and consistently
score health effect and occurrence attributes for individual PCCL contaminants for
promotion to a CCL. As recommended in Chapter 3, all information from existing or
created databases or lists used hi the development of a CCL and PCCL, should be
compiled in a consolidated database that would provide a consistent mechanism for
recording and retrieving information on the PCCL contaminants under
consideration. As a starting point and as recommended in Chapter 3, EPA should
review its developing EDSPD database to determmg if it can be expanded and used
(or served as a model for the development d) such a consolidated database and to
help develop future PCCU and CCLs.
Contaminant databases used in support of the development of future CCLs should report
summary statistics on all d,ita collected, not only the quantifiable observations. In this
regard, EPA should formalize a process for reporting means and/or medians from
data with large numbers of non-detect observations. In addition, EPA may want to
consider providing other measures of concentration in water supplies such as the 95*1
percentfle of contaminant concentration.
105
105
106
106
106
106
Appendix A -Summary of Recommendations from Classifying Drinking Water
Contaminants
A-ll
-------
CCL CP Work Group Report
Chapter 5: PCCL to CCL: Classification Algorithm
A ranking process that attempts to sort contaminants in a specific order is not appropriate
for the selection of drinking water contaminants already on a CCL. In the absence of
complete information, the output of the prioritization schemes was found to be uncertain.
A linear model and a neural network were discussed and demonstrated for potential
use in a prototype classification scheme. It is recommended that EPA give careful
consideration and experiment with developing a prototype classification approach using a
neural network or similar methods. Furthermore, EPA should use several training sets to
gauge the sensitivity or the adopted model. Neural networks provide the flexibility to
capture linear as well as nonlinear dependencies. While the neural network performed
better than the linear model (with respect to minimizing the number of misclassLfied
contaminants), at this time the committee cannot make a firm recommendation as to
which model EPA should use due to the aforementioned uncertainties in the training data
set. Thus, the committee recommends that EPA explore alternative model formulations
and be cognizant of the dangers of overfitting and loss of generalization.
140
To adopt and implement the recommended approach for the creation of future CCLs, EPA
will need to employ or work with persons knowledgeable of prototype classification
methods and devote appreciable time and resources to develop and maintah a
comprehensive training data set. In this regard, the committee strongly recommends
that EPA greatly increase the size of the training data set that was used illustratively
in this chapter to improve predictive capacity. One way that EPA can expand the
training data set and classification algorithm is to allow for the expected case of missing
data. That is, purposefully include in the training data set contaminants for which values
of some of the attributes are unknown and develop a scheme that allows prediction for
contaminants for which some of the attributes are unknown.
140
EPA will also have to accurately and consistently assign attribute scores for all
contaminants under consideration. To do this, EPA will need to collect and organize
available data and research for each PCCL contaminant and document the attribute
scoring scheme used to help ensure a transparent and defensible process, the
importance of which was discussed in Chapter 2. To implement this scheme, EPA
must purposefully include in the training data set contaminants for which values of some
of the attributes are unknown and develop a scheme that allows prediction for these
contaminants. As recommended in Chapter 3, the creation of a consolidated database that
would provide a consistent mechanism for recording and retrieving information on the
contaminants under consideration would be of benefit.
140
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-12
-------
CCL CP Work Group Report
Chapter 5: PCCL to CCL: Classification Algorithm (continued)
EPA will also need to withhold contaminants frbm inclusion in the trailing data set
to serve as validation test cases that can assess the predictive accuracy of any
classification algorithm developed While the committee was able to withhold 5
contaminants presumed worthy of regulatory consideration (T = 1) for this purpose, it had
insufficient numbers of contaminants presumed not worthy of regulatory consideration (T
= 0) to similarly withhold. All withheld validation contaminants were correctly classified
as belonging in the T = 1 category and such results provide (albeit limited) additional
supporting evidence of the validity of the classification algorithm approach. EPA should
make every effort to increase the number of both types of validation test cases (especially
for T = O contaminants) to more thoroughly assess the predictive accuracy of any
classification algorithm developed for use in the creation of future CCLs.
If neural networks are used for prototype classification, the transparency in understanding
which contaminant attributes determine the category of a contaminant will be less than
that of a linear model or more traditional rule-based scheme. However, if one
acknowledges that the undsrlying process that maps attributes into categorical outcomes is
very complex, then there is little hope that an accurate rule-based classification scheme
can be constructed. The fact that the nonlinear neural network performed better than the
linear classifier is a strong indicator that the underlying mapping process is complex and it
would be a difficult task fcr a panel of experts to accurately specify the rules and
conditions of this mapping Furthermore, the loss in transparency in using a neural
network is not inherent, but rather derives from the difficulty in elucidating the mapping.
The underlying mapping in a neural network classifier can be examined just as one would
conduct experiments to probe a physical system in a laboratory. Through numerical
experimentation, one can probe a neural network to determine the sensitivity of the output
to various changes in input data. While a sensitivity analysis was not conducted due to
time constraints, the committee recommends that EPA should use several training
data sets to gauge the sensitivity of the method as part of its analysis and
documentation if a classification approach is ultimately adopted and used to help
create future CCLs.
Finally, EPA should realize that the committee is recommending a prototype
classification scheme to be used in conjunction with expert judgment for the future
selection of PCCL contaminants for inclusion screening on a CCL. Thus,
transparency is less crucial (though no less desired[) at this juncture than when selecting
contaminants from the-CCL for regulatory activities as discussed in the committee's first
report.
Chapter 6: Virulence-Factor Activity Relationships
The committee believes thirt "virulence-factor activity relationships" or VFARs can be a
powerful approach for examining emerging waterbome pathogens, opportunistic
microorganisms, and other newly identified microorganisms, and for predicting the
virulence of these pathogens.
141
141
141
142
145
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-13
-------
CCL CP Work Group Report
Chapter 6: Virulence-Factor Activity Relationships (continued)
The committee defines WAR as the known or presumed linkage between the biological
characteristics of a microorganism, and its real or potential ability to cause harm. VFARs
are conceived as being the relationship that ties specific descriptors (genetic elements,
surface proteins, toxins, attachment factors, metabolic pathways, invasion factors, and
other possible virulence attributes) with outcomes of concern (virulence, potency, and
persistence).
148
The committee has concluded that methods other than culture must be used to fully
evaluate microbial contamination of drinking water (e.g., PCR-based methods).
161
The committee anticipates that, in a very short period of time, microarrays could be
developed that are labeled with all of the genes for a variety of virulence factors identified
within enteric bacteria, pathogenic viruses, opportunistic protozoa, and other (waterbome)
microorganisms. These gene chips could be used to assay environmental and drinking
water samples for the presence of genetic virulence factors of concern.
163
It is one of the committee's central assertions that the assessment of persistence (survival)
in the environment using molecular techniques may be superior to some of the older
methods.
172
The same prototype classification method developed to distill the PCCL to the CCL can
also be applied to VFARs. Training sets of descriptor and response variables could be
developed and used in conjunction with the prototype classification methods to help
derive VFARs.
180
Establish a scientific WAR Working Group on bioinformatics, genomics, and
proteomics, with a charge to study these disciplines on an ongoing basis, and to
periodically inform the Agency as to how these disciplines can affect the identification
and selection of drinking water contaminants for future regulatory, monitoring, and
research activities. The committee acknowledges the importance of several practical
considerations related to the formation of such a working group within EPA, including
how it should be administered, supported (e.g., logistically and financially), or where it
could be located. However, the committee did not have sufficient time in its meetings to
address these issues or make any related recommendations.
184
The findings of this report, and especially that of the Biotechnology Research Group (the
Interagency Report on the Federal Investment in Microbial Genomics), should be made
available to such a working group at its inception. The committee views the activities of a
WAR Working Group as a continuing process in which developments in the fields of
bioinformatics, genomics, and proteomics can be rapidly assessed and adopted for use in
EPA's drinking water program.
185
This Working Group should be charged with the task of delineating specific steps and
related issues and timelines needed to take WARS beyond the conceptual framework of
this report to actual development and implementation by EPA. All such efforts should be
made in open cooperation with the public, stakeholders, and the scientific community.
185
Appendix A -Summary of Recommendations from Classifying Drinking Water
Contaminants
A-14
-------
CCL CP Work Group Report
Chapter 6: Virulence-Factor Activity Relationships (continued)
With the assistance of the Working Group, EPA should identify and fund pilot
bioinformatic projects that use genomics and proteomics to gain practical experience that
can be applied to the development of VFARs while simultaneously dispatching its charges
outlined in the two previous recommendations.
EPA should employ and work with scientific personnel trained in the fields of
bioinformatics, genomics, sind proteomics to assist the Agency in focusing efforts on
identifying and addressing emerging waterbome microorganisms.
EPA should fully participale in all ongoing and planned U.S. federal government efforts in
bioinformatics, genomics, iind proteomics as potentially related to the identification and
selection of waterbome pathogens for regulatory consideration.
185
185
185
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-15
-------
CCL CP Work Group Report
Identifying Future Drinking Water Contaminants (NRC, 1999)
1999 Recommendations of the NRC Committee on Drinking Water Contaminants
Committee Report: A Conceptual Approach for the Development of Future
Drinking Water Contaminant Candidate Lists
Page
Number
An ideal CCL development process would include the following features:
• It would meet the statutory requirements of the SDWA Amendments of 1996,
including requirements for consultation with the scientific community and
opportunities for public comment.
• It would start by identifying the entire universe of potential drinking water
contaminants prior to any attempt to rank or sort them.
• It would consider risks from all potential routes of exposure to water supplies,
including dermal contact and inhalation as well as ingestion.
• It would use the same identification and selection process for microbial, chemical, and
all other types of potential drinking water contaminants.
• It would have mechanisms for identifying similarities among contaminants and
contaminant classes that can be used to assess potential risks of individual
contaminants.
" It would result in CCLs containing only contaminants that, when regulated, would
reduce disease, disability, aid death, and it would exclude contaminants that have few
or no adverse effects on human health (e.g.,contaminants entirely removed or
detoxified through conventional drinking water treatment methods).
However, EPA's resources are constrained, ...the committee believes that EPA can and
should develop and use a process that:
" starts broadly, using existing lists of potential drinking water contaminants,
supplemented by readily available information;
• considers microbiological, chemical and other types of potential contaminants in a
common selection process;
" takes advantage of structure-activity relationships to help overcome deficiencies in
health effects and occurrence data;
" expands the knowledge base over time;
• uses simple criteria, supplemented by expert judgement, to initially cull the candidates
to a preliminary list; and
" employs a prioritization scheme, again supplemented by expert judgement, to
identify final candidates for inclusion on a CCL.
EPA should develop a two-step process for creating future CCLs. In this process, a broad
universe of potential drinking water contaminants is examined and then narrowed to a
preliminary drinking water contaminant candidate list (PCCL) using simple screening
criteria and expert judgment. Then, the PCCL is narrowed to a CCL using a quantitative
screening tool in conjunction with expert judgment.
18
EPA should be as inclusive as possible in narrowing the universe of contaminants
(perhaps on the order of 100,000 substances) down to a PCCL. The committee envisions
that a PCCL would contain on the order of thousands of potential drinking water
contaminants of all types for subsequent evaluation, prioritization, and culling to a CCL.
18
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-16
-------
CCL CP Work Group Report
As a start, a PCCL should contain all substances and microbes that are knownto cause
significant adverse health effects (regardless of exposure route) and have the potential to
occur in drinking water anc those demonstrated to occur in drinking water supplies (unless
they are known not to pose a significant health risk). A PCCL should also include all
substances that may pose a drinking water risk based on their potential for occurrence and
health effects.
18
Preparation of a PCCL should not involve extensive collection or analysis of data, nor
should it drive research or monitoring activities. However, the committee recognizes that
it will be necessary to develop and use screening criteria (e.g., production values of
commercial chemicals) to shorten the list of contaminants for a PCCL.
18
Development of a PCCL should begin as soon as possible to support the development of
the next CCL; the PCCL should be available for public and other stakeholder input
(especially through the Internet) and should undergo scientific review.
18
A new PCCL should be generated for each CCL development cycle lo account for new
data and potential contaminants.
19
As an integral part of the CCL development process, the committee recommends the use
of a comprehensive database that provides a consistent mechanism for recording and
retrieving information on all the contaminants under consideration. A well-designed
relational database can function as a "master list" that contains a detailed record of how
the PCCL and CCL were developed, as well as providing a powerful analytical tool for
the development of future CCLs.
19
To help identify commercial chemicals that might pose risks in drinking water, EPA
should consider exercising its authority under the Toxic Substances Control Act (TSCA)
to collect production and import data on both organic and inorganic chemicals by use
category.
19
To assist in the evaluation of microbial pathogens, it also may be useful to identify
common mechanisms of pathogenicity among contaminants in order to include them on
future CCLs. An approach analogous to chemical structure-activity relationships (SARs)
for microorganisms does not currently exist, but EPA should develop such a prioritization
tool for microbial contaminants through use of gene data banks and with the cooperation
and support of other federal and state health organizations.
19
Preparation of a CCL from a PCCL will require collection and evaluation of all available
health effects and occurrence data for each substance on the PCCL. To cull a list of
thousands of potential drinking water contaminants of all types to approximately a
hundred for inclusion on the CCL, EPA must combine expert judgment equally with a
single prioritization tool that can be used to evaluate any type of PCCL contaminant.
19
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants
A-I7
-------
CCL CP Work Group Report
EPA should develop a prioritization tool to help narrow the PCCL to a CCL. The tool 19
should be kept as simple as possible and be developed with regular public and other
stakeholder input. Ensuring transparency throughout its development and avoiding
"black-box" decision-making are critical steps. The tool should be validated using
contaminants with extensive health effects and occurrence data and well-established risks.
The tool must be able to identify and effectively address data gaps for each contaminant.
Following a re-examination of 10 existing chemical hazard ranking schemes, the
committee concluded that none was directly suitable for developing a CCL from a PCCL.
However, 3 of the schemes (Cadmus, ITC, and WMPT) contain features that would be
suitable for CCL development and could conceivably be adapted for this purpose.
The committee strongly recommends that no factors or components (e.g. measures of 13
occurrence of health effects) of the tool should be weighted in any way (even through
expert judgement).
The committee further recommends that any prioritization tool should be subjected to 13
validation and scientific review prior to use.
EPA should reserve a number of contaminants or a percentage of the CCL for 19
contaminants that are listed based solely or primarily due to expert judgment. EPA must
also retain the ability to remove contaminants from inclusion on a CCL based on expert
judgment.
The CCL should consist of roughly equal numbers of contaminants ready for regulatory 19
decisions and those requiring further research to drive such efforts equally. This
recommendation is consistent with EPA's partitioning of the first CCL into equivalent
future action categories.
Regardless of what process is adopted by EPA to develop future CCLs, the committee 20
strongly recommends that all contaminants that have not been regulated or removed from
the existing CCL (and future CCLs) should be automatically retained on each subsequent
CCL for reevaluation.
As in the previous report, the committee recognizes that the need for policy judgments by 20
EPA cannot and should not be removed from any CCL development process. In making
these decisions, EPA should use common sense as a guide and err on the side of public
health protection.
Appendix A - Summary of Recommendations from Classifying Drinking Water
Contaminants A-18
-------
CCL CP Work Group Report
Appendix B
Summary of the NDWAC CCL Work Group Investigation
of QSAR Models as Sources of Data / Information for the
CCL Development Process
The NRC (2001) suggested that QSAR analytical methodology be used to identify
chemicals with potential occurrence in drinking water and potential adverse human health
effects. This summary discusses work conducted toward evaluating the feasibility of
applying specific QSAR models for future CCLs.
Background
A variety of QSAR models have been developed for human health endpoints and
"packaged" into user-friendly commercial or public -use programs. Human health effect
endpoints predicted by QSAR models include: mutagenicity, carcinogenicity,
teratogenicity, neurotoxicity, reproductive and developmental toxicity, skin/eye
sensitization and irritation, and systemic toxicity. QSAR development for other
endpoints is underway by a number of EPA Office of Research and Development (ORD)
organizations, other regulatory institutions and the private sector (Benigni and Richard,
1998).
The more popular commercial QSAR packages for human health include The Open
Practical Knowledge Acquisition Toolkit (TOPKAT)
(http://www .accelrv s.com/prwiucis/topkat/X MultiCase (MCASE)
(http://www.multicase.cora/). and the Deductive Estimation of Risk from Existing
Knowledge (DEREK) Qittp://wwAv.chem.leeds.ac1uk/LllK/dei-ek/'index.htrnQ. Two general
types of models can be distinguished: statistically-based models such as TOPKAT, and
rule-based models such as MCASE. Two issue papers developed by the technical team
for the NDWAC Work Group reviewed these models [Screening Models and Algorithms
(1/28/03) and Status and Feasibility of Using (Quantitative) Structure-Activity
Relationship ((Q)SAR) Models for CCL Development (8/7/03)|. Concise characterizations
of these and other QSAR packages appear on OECD's (Organisation for Economic
Cooperation and Development) web-site at
http:/Avebdoinino 1 .oecd.orR/com:net/env/inQdcls.nsf
A broader array of QSAR packages are available for endpoints related to chemical
occurrence than for health effects, and have much larger domains (i.e., are applicable to
more chemicals). For example, EPA's Assessment Tools for the Evaluation of Risk
(ASTER) (http ://www.cpa. "ov/'med/dalabases/aster .him) includes a database of more than
56,000 chemicals, and batch searches of the entire European Inventory of Existing
Commercial Chemical Substances (EINECS) directory of 166,000 chemicals have been
conducted by the Danish EPA. Notably, for homeland security reasons public access to
ASTER has been suspended; the status of other packages in this regard has not been
determined.
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-l
-------
CCL CP Work Group Report
The OECD web-site lists predictive models for environmental fate and exposure
pathways, including human health routes of exposure. One with possible application in
identifying chemicals with potential occurrence in drinking water is the physical
chemical predictive package, the Estimation Programs Interface for Windows (EPIWIN)
(as part of the EPI Suite) (http:/^'uay.epa.gov/oppt/newchenis/denver). Also, the Persistent,
Bioaccumulative and Toxic (PBT) Profiler is a versatile new QSAR package developed
through Office of Prevention, Pesticides, and Toxic Substances (OPPTS) Sustainable
Futures Program (www.pbtigofilcr.nct). It was recently released by OPPTS as a resource
for industries to voluntarily use for screening chemicals.
Persistence and biodegradation estimates from packages such as CATABOL
mechanistic understanding of enzyme-
mediated processes, similar to those available for predicting toxicity. Hence, QSAR
models for persistence and biodegradation are more limited in scope and accuracy than
packages that predict physical-chemical parameters.
Model Assessments
The technical support team for the NDWAC CCL Work Group conducted a limited
assessment of the performance of two QSAR models, TOPKAT for predicting LOAELs,
and EPI Suite for predicting chemical properties. The purpose of the evaluation was to
inform NDWAC Work Group members regarding the potential for using QSAR derived
data for future CCLs.
TOPKAT is a commercial computational toxicology package that uses chemical
structural information (2-D descriptors of structural fragments) and QSAR models to
estimate a range of human health lexicological and non-human ecological endpoints.
Predictions are made for untested chemicals by comparison with structural fragments
contained in the model's training set.
TOPKAT was selected for evaluation for several reasons. It includes the capability to
predict rat chronic oral LOAELs for a variety of chemicals. It is currently being used by
EPA ORD scientists in the National Center for Environmental Assessment (NCEA), who
maintain a current license for its use for EPA research. NCEA has compiled a substantial
historical database of LOAEL predictions for diverse chemicals, and made the database
available for this exercise. Through the cooperation and technical assistance of NCEA in
Cincinnati, OH, 525 compounds from each of three test groups of chemicals were run by
technical support staff on NCEA's computers in order to expand the existing data set.
Developers of the statistically-based TOPKAT model estimated that predictions of rat
oral LD50's fell within a factor of 5 from test results for 86-100 percent of the 4,000
chemicals in the model (Health Designs, Inc., 1997). An evaluation of TOPKAT by the
Danish EPA, using 1,840 chemicals not contained in the training data set, gave somewhat
poorer results: an R2 = 0.3 1 ; they concluded that 86 percent of QSAR estimates fell
within a factor of 10 from test results (Wedebye and Niemela, 2000). An evaluation by
an EPA QSAR researcher in ORD estimates that for approximately one-third of
compounds, the model indicates that no prediction is possible due to nonconfonnance
with the training set domain (parameters are outside the "Optimum Prediction Space").
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-2
-------
CCL CP Work Group Report
This means that the training sets used in the NCEA TOPKAT model are limited; a more
expanded training set would reduce nonconformance. For 60 percent of the remaining
two-thirds of substances, TOPKAT estimates of Lowest Observable Adverse Effects
Levels (LOAELs) may roughly be within a factor of two. Since the model was
developed, however, there have been additional chemicals assayed and not yet included
in the training set database.
The EPI (Estimation Program Interface) Suite™ is a Windows-based suite of
physical/chemical property and environmental fate estimation models1 developed by the
EPA's Office of Prevention, Pollution, and Toxics and Syracuse Research Corporation
(SRC). EPI Suite™ uses a single chemical identifier called SMILES notation (Simplified
Molecular Input Line Entoy System) as input for 14 regression models to predict a suite
of chemical parameters. EPI Suite includes algorithms for calculating aerobic
biodegradability and water solubility, two of the properties in this limited analysis.
BIOWIN™ was used to estimate aerobic biodegradability of organic chemicals and
WSKOWWIN™ was used to estimate chemical water solubility, EPI Suite™ is
routinely used by OPPT's Chemical Control Division for evaluation of new chemicals
and chemical uses, as required under the Toxics Substances Control Act
(Premanufacturing Notice, PMN).
EPI Suite™ is publicly available, user-friendly, and considered reasonably well-validated
for the modules used in this assessment. EPI Suite™ enables the user to simultaneously
run 10 estimation programs for a selected chemical.
The QSAR model evaluations were for three groups of chemicals, initially:
• 262 Draft CCL 1 chemicals plus 22 Deferred Potential Endocrine Disrupters
• 262 Non-CCLl chemicals with selected toxicity and environmental fate parameters
• 262 Non-CCLl chemicals with no available data for selected toxicity and environmental fate
endpoints.
The results for the first two groups of chemicals indicate how well the QSAR model predictions
of chemical properties and toxicity endpoints for human health compare with published empirical
data. The third data set provides some indication of the portion of chemicals lacking
empirical data for which the models were able to predict the needed chemical properties
and toxicity endpoints. EPA's National Center for Environmental Assessment (NCEA)
provided some LOAELs dready generated with TOPKAT, as well as measured
values for comparison. To this list, CCL1 chemicals were added, as were chemicals
with available data from the CCL Example Universe Data Set. Regulated contaminants
and those identified in model training data sets were excluded. The initial data set was culled
to a test data set of 695 chemicals.
1 See http://www.epa.gov/opptintr/exposure/docs/episuite.htm for more
information on EPI Suite1 M. The model, including underlying algorithms and training
sets, are available for free download from the U.S. Environmental Protection Agency at:
http://www.epa.gov/oppt/exposure/docs/episuitedl.htm.
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-3
-------
CCL CP Work Group Report
Data Gathering
Input data required to run both TOPKAT and EPl Suite™ are chemical structural
formulae represented by the SMILES notation. A database of over 103,000 SMILES
notations are included in EPI Suite™, retrieved using chemical CAS Registry Numbers.
Notations not identified in EPl Suite were developed manually.
To compare model predictions with available data, LOAELs were gathered from the
Registry of Toxic Effects of Chemical Substances (RTECS). The lowest LOAEL from a
rat or mouse oral LOAELs from studies of 28 days or longer (TDlo value) was used,
converted to a daily dose from the reported cumulative dose by dividing dose by study
length (mg/kg-day). LOAELs for two hundred and twenty seven (227) chemicals were
in RTECS.
Five data sources were used in a hierarchy to identify the chemical properties of
solubility and biodegradation: SRC Chemfate Database, Physical -Chemical Properties
and Environmental Fate Handbook, the Hazardous Substances Data Bank (HSDB), the
International Program on Chemical Safety, and the National Toxicology Program.
Biodegradation data was found for 100% of chemicals, and solubility measurements for
206 chemicals.
Estimated LOAELs were generated from TOPKAT, run on computers at NCEA in
Cincinnati, with oversight by EPA staff. Solubility and biodegradation estimates were
made with EPISuite™ models downloaded from EPA's web site. Two programs were
downloaded and run for this evaluation: WSKOWWIN for predicting water solubility,
and BIOWIN for predicting biodegradability. WSKOWWIN estimates water solubility
(mg/L) of an organic compound by regression of the octanol-water partition coefficient
(Kow). BIOWTN estimates the time required for a compound to biodegrade under
aerobic conditions with mixed cultures of microorganisms. The half-life for
biodegradation of a chemical in water is determined using the ultimate biodegradation
expert survey module of BIOWIN. This estimation program provides an indication of a
chemical's environmental biodegradation rate in relative terms such as hours, hours to
days, days-weeks, and so on. The rate is estimated from the chemical's "half-life," i.e.,
the time required for one-half of the chemical to "completely" degrade (i.e.,
mineralization to H2O and C02).
Findings
TOPKAT was able to predict LOAELs for 45% of 525 chemicals tested (QSAR
predicted LOAELs were provided by NCEA for 170 chemicals). TOPKAT identifie d
55% of the chemicals as outside of the predictive domain. This evaluation of TOPKAT
is comparable with preliminary results from NCEA, given study design differences. The
comparison of TOPKAT LOAEL predictions with empirical results was difficult because
of variability in empirical measurements. Using the lowest reported LOAEL resulted in
20 percent of model results within a factor of two of the empirical data, 53% within a
factor of 5, and about 65% of predictions within a factor of 10 of empirical values. These
are slightly lower than a comparative study by NCEA, using LOAELs from EPA's IRIS
database, as compared to the use of RTECS LOAELs by this study.
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-4
-------
CCL CP Work Group Report
WSKOWWIN predicted water solubility for organic chemicals within a factor of five for
54% of chemic als evaluated Relatively high variability was observed among 46
chemicals with multiple empirical values, due to methods diversity and variability.
WSKOWWIN performance was user-friendly, and appears to be transparent and
applicable to broad range of chemicals.
BIO WIN predictions broadly distinguished chemicals that degrade rapidly in the
environment from those mat degrade slowly. Empirical data from certain test procedures
limited comparisons of predictions of degradation rates (e.g., weeks, months) from
BIOWIN.
Conclusions
QSAR modeling using TOPKAT for health effects may require greater selectivity in
chemicals (coverage of training sets) and health; effects modeled. Comparison of QSAR
model results to empirical data was limited by missing and highly variable measurements
reported. This may generally limit the ability to fully evaluate QSAR model predictions
for chemicals outside their respective training sets. Based on this validation exercise,
TOPKAT appears to have: limited utility for estimating rat chronic oral LOAELs. First,
many compounds evaluated for future CCLs are likely to be outside the model domain
because of the limiting training sets; model performance for these compounds was,
therefore, less than optimal. Second, for compounds within the TOPKAT model domain,
model performance was modest at best, though performance was similar to prior model
evaluations. Of greatest concern was the apparent underestimation of toxicity for those
compounds that are the most toxic, and therefore likely to be of greatest potential concern
(see Figure 1). In addition to these two concerns, TOPKAT requires additional modules
to be run in batch modes , and the training data set is proprietary.
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process B~5
-------
CCL CP Work Group Report
Figure 1. Results of TOPKAT compared with empirical LOAEL data.
Minimum Duration = 28 d, No TOPKAT Warnings
0123
Measured Median LOAEL (log mg/kg/d)
The frequency with which error codes were reported significantly limited the breadth and
applicability of TOPKAT. Of 525 chemical queries conducted, 288 (55 percent) were
accompanied by error codes indicating that they were outside the predictive domain.
Most commonly, SI (poor fit training set data with queried chemical), TK (queried
chemical contains an element not included in the training set) and OPS (outside the
prediction space) codes were encountered. This is useful information to judge the utility
of TOPKAT, a key goal of this exercise, though it will limit the ability to assemble
quantitative data TOPKAT predicted LOAELs for half of the chemicals it was able to
evaluate within a factor of 5 of empirical data. However, me empirical data used for
comparison may be a contributor to the poor correlation.
Part of the difficulty is that a LOAEL, even when specified as a rat chronic oral LOAEL
as in TOPKAT, is not a specific health endpoint. That is, there are numerous
toxicological mechanisms and specific adverse effects that can result from chemical
exposure that are difficult to predict specifically based on chemical structure. It is
generally recognized that limitations in understanding the mechanism of toxicological
action for chemicals also limits the ability for developing QSAR models to predict these
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-6
-------
CCL CP Work Group Report
mechanisms and outcomes, and the exposure levels that cause them. EPA scientists and
others have suggested that QSAR models for health effects with less biologically
complicated outcomes, e.g., mortality as measured by the LD50.
Based on this limited evaluation, QSAR modeling appears to be a feasible approach for
estimating solubility of organic compounds that lack empirical data. The WSKOWWIN
module in the EPI Suite™ was able to provide reliable estimates of water solubility of
organic compounds based on this limited analysis of experimental and predicted values.
Further, this package is e;isily accessible, user friendly and the training data set is
publicly available. WSKOWWIN does not address ionizing chemicals in its water
solubility calculations.
The B1OW1N model of the EPI Suite™ package also appears to be useful for predicting
chemical biodegradability. BIO WIN provides semi-quantitative estimates of time to
complete mineralization, and these predictions have been shown to be reasonably reliable
in terms of identifying chemicals that biodegrade fast or slow. Another model, the
OASIS_Catabol software package, estimates percent biodegradation of chemicals in a 28
day inherent test, a different endpoint that could be a quantitative data element for
evaluating chemical persistence.
Model selection for this exercise was limited by several factors, including time and
staffing resources. Other QSAR models that predict cither endpoints should be considered
and tested to evaluate use of QSAR for CCL data. These may include QSAR models that
predict: other health effects endpoints, other potential exposure data, degradation rates
due to hydrolysis, photolysis or degradation mechanisms other than aerobic
biodegradation.
Appendix B: Work Group Investigation into QSAR Models as Sources of
Data/Information for the CCL Process
B-7
-------
CCL CL Work Group Report
Appendix C
Draft Scoring Protocols Developed and Used for Trial
Attribute Scoring Exercise
This appendix provides five draft attribute scoring protocols used in an Attribute Scoring
Exercise conducted as part of the NDWAC Work Group process. Protocols for Magnitude,
Persistence-Mobility, Potency/Severity, and Prevalence follow.
Magnitude Scoring
This document describes how to assign a numerical score for the attribute magnitude, one of the
five evaluated in the October 21, 2003 attribute scoring workshop.
NRC Definition of Magnitude
The National Research Council (NRC) defines magnitude as the concentration or expected
concentration of a contaminant relative to a level that causes a perceived health effect.1 NRC
recommended that magnitude be scored on the basis of data on concentration and potency.
Approach for Magnitude Scoring
In this document, we describe an approach to attribute scoring that relies on concentration data
alone.
Protocol for Magnitude Scoring based on Concentration Only
Step One: Identify highest-ranked data element
When more than one data element is available for a particular contaminant, use the hierarchy
below to select the preferred element. Exhibit 1 presents the hierarchy of data elements to be
used in the magnitude scoring process. Note that the Magnitude element should be correlated
with the value used to score the attribute Prevalence.
NRC 2001. Classifying Future Drinking tfater Contaminants for Regulatory Consideration. Washington, D.C.:
National Academy Press.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-l
-------
CCL CL Work Group Report
Exhibit 1. Hierarchy of Magnitude Data Elements
Rank
M1
M2
M3
M4
MS
M6a
M6b
M7
M8
Magnitude Data Element
Finished Drinking Water- Median of detected
concentrations from Public Water Systems with
detections
Median of detected concentrations from ambient / raw
source monitoring sites with detections
Median of detected concentrations from ambient / raw
/ source water samples with detections (Note: use
combined surface / ground water if available and
higher of SW/GW if not
Finished Drinking Wa"er - Median of detected
concentrations from F'ublic Water Systems with
detections
Median of detections from ambient / raw 1 source
water samples with detections
Environmental release data, total pounds or tons
reported as released (TRI)
Environmental release data (ATSDR HazDat)
Production or Use data
Manufactured Chemicals
Type of Data
National scale finished drinking water
occurrence data (NCOD, NIRS)
National scale ambient monitoring
data (NAWQA)
National scale / representative data
(NREC)
Individual state / small regional
finished drinking water data
Individual state / smalt regional data
From Toxics Release Inventory
From ATSDR HazDat
Various, From EPA list (e.g. HPV) or
actual production amount if available,
or other information about use (e.g.
consumer use insecticide), NCFAP.
Various. From lists (e.g. TSCA, HPV,
IUR)
Step Two: Use look-up table to find attribute score for value identified in Step One.
For each ranked data element, there is a corresponding look-up table, which contains a range of
data values assigned to a numerical magnitude score. Locate the look-up table associated with
the highest-ranking data element identified in step qne. Use the look-up table to determine the
numerical score associated with the data value for the chemical being scored.
Place the magnitude score in the scoring worksheet. To generate a Magnitude Score as defined
by NRC, multiply the Magnitude score by the Potency Score, take the square root, and record the
resulting score.
Appendix C - Draft Protocol's Developed and Used for Trial Scoring Exercise
C-2
-------
CCL CL Work Group Report
LOOK-UP TABLES
Look-Up Table M1:
Finished Drinking Water- Median Detections from Public Water Systems with Detections
Median Detection Concentration (mg/L)
Combined
(SW and GW)
<0.00020
0.00020 - <0.00050
0.00050 -O.00051
0.00051 -<0.00075
0.00075 -<0.001 00
0.00100 -<0.00101
0.00101 -<0.00125
0.00125 -<0.00160
0.00160 -<0.00250
> 0.00250
SW
<0.00022
0.00022 - <0.00044
0.00044 -O.00055
0.00055 -<0.00072
0.00072 -<0.001 00
0.00100 -<0.00101
0.00101 -<0.00135
0.001 35 -<0.00200
0.00200 -<0. 00400
* 0.00400
GW
<0.00020
0.00020 -<0.00050
0.00050 - <0.00051
0. 00051 -<0.00080
0.00080 -<0.001 00
0.00100 -<0.001 10
0.001 10 -<0.00130
0.00130 -<0.00160
0.00160 -<0.00220
*0.00220
Magnitude
Attribute
Score
1
2
3
4
5
6
7
8
9
10
If both surface water and ground water data are available, score the simple total of SW and GW
values. If only surface water or ground water data are availabb, score the value using the
corresponding surface water or ground water column.
Look-Up Table M2: Median Concentration Values tig/L) from
Drinking Water Sites with Detected Concentrations
Magnitude Range iig/L)
0.0-0.049
0.05 - 0.099
0.1 -0.5
0.51 -0.99
1.0-1.5
1.51 -3.0
3.01 -5.0
5.01 -9.99
10.0-50.0
50.01 +
Magnitude Attribute Score
1
2
3
4
5
6
7
8
9
10
Look-Up Table M3: Median of Source Water Samples with Detections (National Data)
Magnitude Range (xg/L)
0.0 - 0.049
0.05-0.099
0.1 -0.5
Magnitude Attribute Score
1
2
3
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-3
-------
CCL CL Work Group Report
Magnitude Range iig/L)
0.51 - 0.99
1.0-1.5
1.51 -30
3.01-5.0
5.01 -9.99
10.0-50.0
50.01 +
Magnitude Attribute Score
4
5;
6
7
8
9
10
Look-Up Table M4: Public Water Systems with Detections (Regional Data)
Median Defection Concentration (mg/L)
Combined
SW+GW
<0.00020
0.00020 -<0.0005
0.00050 -<0.00051
0.00051 -<0.00075
0.00075 -<0.00100
0.00100 -O.00101
0.001 01 -<0.00125
0.00125 -<0.00160
0.00160 -<0.00250
> 0.00250
SW
<0. 00022
0.00022 -<0.00044
0.00044 -<0.00055
0.00055 -<0.00072
0.00072 -<0.00100
0.00100 -<0.00101
0.00101 -<0.00135
0.001 35 -<0. 00200
0.00200 -<0. 00400
;> 0.00400
GW
<0.00020
0.00020 -<0. 00050
(l.00050-<0.00051
dl.00051 -<0.00080
0.00080 -<0. 00 100
0.00100 -<0. 001 10
0.001 10 -<0.00130
0.00130-<0.00160
0.001 60 -<0.00220
: * 0.00220
Magnitude
Attribute
Score
1
2
3
4
5
6
7
8
9
10
If both surface water and ground water data are available, score the simple total of SW and GW
values. If only surface water or ground water data are available, score die value using the
corresponding surface water or ground water column.
Currently, no data were located for this table (from the 41 data sources sought). It is anticipated
that these data will become available in the future, and when they do, will fall here in the
hierarchy.
Look-Up Table MS: Median of Source Water Samples with Detections (Regional Data)
Magnitude Range (ig/L)
0.0 - 0.049
0.05 - 0.099
0.1 -0.5
0.51 - 0.99
1.0-1.5
Magnitude Attribute Score
1
2
3
4
5
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-4
-------
CCL CL Work Group Report
Magnitude Range (ig/L)
1.51 -3.0
3.01 -5.0
5.01 -9.99
10.0-50.0
50.01 +
Magnitude Attribute Score
6
7
a
9
10
Look-Up Table M6a: Environmental Release Data (TRI)
Toxic Release Inventory
Total Reported Release in 2001
(Quantity in Pounds)
< 10 pounds
11-300
301-1,000
1,001 -10,000
10,001 -50,000
50,001 -300,000
300,001 - 1,000,000
1,000,001 -8,000,000
8,000,001 -40,000,000
> 40 million
Magnitude
Attribute Score
1
2
3
4
5
6
7
8
g
10
Look-Up Table M6b: Other Environmental Release Data (ATSDR HazDat)
Maximum Concentration
(mg/L) Con
<0.001
Magnitude attribute score
1
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-5
-------
CCI CL Work Group Report
sO.01
<0.03
sO.08
sO.17
sO.34
sO.80
<2.30
^10
*100
2
3
4
5
6
7
8
9
10
Look-Up Table M7: Data for Pesticides
Mass of Pesticides Applied or Used
Default for any pesticide for non-environmental use (restricted hospital or
indoor use)
Default for any pesticide in environmental use
k 100.000 Ibs
withjout data
;> 1.000.000 Ibs
> 2.000.000 Ibs
;> 20,000.000 Ibs
> 50,000.000 Ibs active ingredient applied
Magnitude
Attribute Score
3
5
6
7
8
9
10
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-6
-------
CCL CL Work Group Report
Look-Up Table M8: Mass Produced/Imported Annually
\
Mass Produced/Imported
Annually (TSCA, HPV)
no data available on production
from any source
data available from other
sources and < 10,000 Ibs
;> 10,000 Ibs (CUS/IUR)
i 10,000 Ibs and other factors
warrant a higher score: e.g.,
consistently reported at this
level since CUS/IUR reporting
began; or in routine/wide
commercial use; or other
sources indicate production >
100,000 Ibs
* 1,000,000 Ibs (CUS/IUR,
HPV)
> 1,000,000 Ibs and other
factors warrant a higher score:
e.g., consistently reported at
this level since CUS/IUR
reporting began; or in
routine/wide commercial use
:> 1,000,000,000 Ibs (CUS/IUR)
2 1,000,000,000 Ibs and other
factors warrant a higher score:
e.g., consistently reported at
this level since CUS/IUR
reporting began; or in
routine/wide commercial use
Corresponding Score
1
3
5
6
7
8
9
10
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-7
-------
CCL CL Work Group Report
Persistence - Mobility Scoring
This document describes the process for assigning a numerical score for the attribute persistence-
mobility, one of the five attributes to be tested in th^ October 21, 2003 attribute scoring
workshop. In the protocol, Persistence - Mobility may be scored as a fifth attribute, or as a
surrogate measure for the attribute Prevalence, as thie lowest element in the hierarchy.
NRC Definition of Magnitude
The National Research Council (NRC) defines persistence-mobility as a surrogate measure when
prevalence is unavailable, describing the likelihood that a contaminant would be found in the
aquatic environment based solely on it physical properties.2 NRC recommended that persistence-
mobility be scored on the bas:is of data on physical chemical properties such as solubility and
half-life.
Approach for Persistence-mobility Scoring
The approach for scoring includes assigning two scores, one for persistence and one for mobility,
on a numeric scale of 1 through 3, representing low, medium, and high. Using a hierarchy of
physical property data elements, each contaminant is scored for both persistence and mobility.
The average of these two scores is multiplied by 10/3 to obtain the persistence-mobility score.
Below are two tables that include a hierarchy of available properties for each data element
representing either persistence or mobility.
Protocol for Persistence-mobility Scoring
Step One: Identify and score highest-ranked data value for Persistence
When more than one data element value is available for a particular contaminant candidate, use
the hierarchy below to select the preferred element. Exhibit 1, below, describes the hierarchy of
data elements to be used in the Persistence scoring process. When several values for a physical
property are available, the highest scoring value should be used for scoring, unless that value is
not representative of environmental conditions in drinking water. Enter the element type, source,
and attribute score in the Persistence Mobility Worksheet. Also record any available supporting
information, e.g. test conditions, in the Notes column.
NRC 2001. Classifying Future Drinking Water Contaminants for Regulatory Consideration. Washington, D.C.:
National Academy Press.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-8
-------
CCL CL Work Group Report
\
Exhibit 1. Hierarchy of Persistence Data Elements
Hierarchy
P1
P2
P3
P4
Element
Half Life
(T1/2)
Stability
(abiotic and
biotic
degradation)
Biodeg rate
(measured)
Biodeg rate
(estimated)
1 (low)
< 1 week
measured or calculated
biotic or abiotic half-life in
environmental or
laboratory waters
(excluding abnormal
conditions like activated
sludge, extreme pH) is
less than one week
days
days-weeks
days
days-weeks
2 (medium)
>1week-< 4 weeks
measured or calculated
biotic or abiotic half-life in
environmental or
laboratory waters
(excluding abnormal
conditions like activated
sludge, extreme pH) is
less than one month.
weeks
weeks - months
weeks
weeks - months
3 (high)
> 4 weeks
measured or calculated
biotic or abiotic half-life in
environmental or
laboratory waters
(excluding abnormal
conditions like activated
sludge, extreme pH) is
one month or longer.
months
recalcitrant
months
recalcitrant
Step Two: Identify and Score Highest Ranking Value for Mobility
The hierarchy of physical properties for scoring mobility is in Exhibit 2. Select the left-most data
element available for scoring. When several values for a physical property are available, the
highest scoring value should be used for scoring, unless that value is not representative of
environmental conditions in drinking water.
Exhibit 2. Mobility Scoring Hierarchy
Hierarchy
M1
M2
M3
M4
MS
Element
Organic Carbon Partition Coefficient
(Koc)
Log of OctanoU/Vater Partition
Coefficient (Log Kow)
Dissociation Constant (Kd) (crrrVg)
Henry's Law Constant (HLC)
(atm m3/ mol)
Solubility (mg/L)
1 (Low)
>300
>4
<5
>10'3
<1
2 (Medium)
100-300
1-4
1-5
10-7-10'3
1-1,000
3 (High)
<100
<1
>5
<10'7
>1,000
Step Three: Multiply the average of the persistence and mobility scores by 10/3 for the
persistence-mobility score.
Step 3B. Alternately, use one of the two elements (multiply score by 10/3) if only one is
available.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise
C-9
-------
CCL CL Work Group Report
If using persistence-mobility as surrogate measures for prevalence, use the Persistence-Mobility
Score for Prevalence. This will be used in conjunction with the use of production data for scoring
Magnitude (see Prevalence and Magnitude Attribute Scoring Protocols for details).
Potency/Severity Scoring
This document describes the process for assigning a numerical score for potency and severity,
during the October 21, 2003 attribute scoring workshop.
Protocol for Potency Scoring
Step One: Open the spreadsheet for Potency and Severity Scoring
Step Two: Enter the name of the chemical in the column labeled contaminant
Step Three: Identify and score highest-ranked data element for potency using the following
hierarchy of values.
• RfD or equiviiIeiit>NOAEL>LOAEL>LD50
• Measured > Modeled
• EPA Rfl» ATSDR MRL (ChroniO Intermediate >Acute)> Cal EPA PHG
>WHO/EU/Hcalth Canada
• OPP> IRIS for Pesticides
Step Four: Enter the selected measure of potency into the appropriate column of the spread
sheet Make sure that the units are in mg/kg/day.
Step Five; Select a measure for cancer potency if one is available. The preferable measure
will be the E-4 risk concentration in drinking water in mg/L If the risk is expressed at levels
other than E-4, convert the value to the target risk (E-4). If the cancer potency measure is the
slope factor, calculate the E-4 risk concentration using the following equation:
E-4 Risk concentration = 10.000 x 35 ke/dav/L
Slope Factor (mg/kg/day)'1
Step Six: Choose the higher of the non-cancer or cancer potency score as the measure of
potency.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-10
-------
CCL CL Work Group Report
Protocol for Severity Scoring
Step 1: Enter the critical effect that goes with the potency score selected through the
Potency Protocol in the appropriate column of the Potency-Severity Spreadsheet,
• If the potency is based on a tumohgenic response enter cancer as the critical
effect. If information on tumor type is provided include that information.
• If the potency score is derived from a non-cancer parameter enter the critical
effect(s) that go with the RfD, LOAEL, or NOAEL (no observed effect).
• If the potency score is from an LD50 study (measured or modeled) enter death as
the critical effect.
• If the potency score is a modeled LOAEL examine the Health Effects Information
to determine if an appropriate critical effect can be determined.
Step Two: Use the Severity Scoring Sheets (A and B) to give the Critical Effect a Score and
enter that score on the Potency/Severity Scoring Sheet.
• Severity Score A should be selected from the nine point scale and Severity Score B from
the five point scale.
I
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-l 1
-------
CCL CL Work Group Report
t
Prevalence Scoring
This document describes how to assign a numerical .score for prevalence, one of the attributes
evaluated for the October 21, 2003 Attribute Scoring Workshop.
Definition of Prevalence
The National Research Council (NRC) defines prevalence as how commonly a contaminant
occurs, or would occur, in drinking water3. Prevalence ideally involves both spatial and
temporal occurrence, and should be scored based on seven measurements (in order of
preference): tap water, distribution systems, finished water of treatment plants, and source water
used for supplying drinking water. If no information is available to demonstrate occurrence in
drinking water, use: observations in watersheds/aquifers, historical contaminant release data, or
chemical production data.
Approach for Prevalence Scoring
We have followed the approach recommended by NRC when scoring candidates for prevalence.
A wide variety of data sources exist which could be used for this exercise. Some of these
sources are better than others, and every effort was made to use the best and most complete data
available.
Protocol for Prevalence Scoring
Step One: Identify highest-ranked data value
When more than one data value is available for a particular contaminant candidate, a pre-
established hierarchy ensures that scoring decisionsl are made consistently. Exhibit 1, below,
describes the hierarchy of data elements to be used in the prevalence scoring process. Each data
source has been given a rank, corresponding with Exhibit 1, with 1 being the top of the
hierarchy. Note that the data element used to score Prevalence should be correlated with the
value used to score the attribute Magnitude. That is, the attribute Magnitude should be score with
the element in the corresponding M rank, in the accompanying Magnitude Scoing Protocol.
I
NRC 2001. Classifying Fufjre Drinking Water Contaminants far Regulatory Consideration. Washington, D.C.:
National Academy Press.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-l 2
-------
CCL CL Work Group Report
Exhibit 1. Hierarchy of Prevalence Data Elements
t
Rank
P1
P2
P3
P4
P5
P6a
P6b
P7
P8
Prevalence Data Element
Finished Drinking Water -Percentage of Public Water
Systems (PWSs) with Detections
Percentage of Ambient/ Raw/Source Monitoring Sites
with Detections
Percentage of Ambient/Raw/Source Monitoring Samples
with Detections
Finished Drinking Water - Percentage of PWSs with
Detects
Percentage of Ambient/Raw/Source Monitoring Sites with
Detects
Environmental release data, number of states reporting
releases
Hazardous substance release data, number of states
Production or Use data for Pesticides
Persistence / Mobility data
Type of Data
National scale / representative
data (NCOD.NIRS)
National scale / representative
data (NAWQA)
National scale / representative
data (NREC)
Individual state/ small regional
data
Individual state / small regional
data
From Toxics Release Inventory
HazDat
Various. From EPA (e.g. HPV)
list or actual production amount
if available, or other information
about use (e.g. consumer use
insecticide)- NCFAP
Physical chemical properties
Data elements corresponding to rank PI should be looked for first. If this element exists, this is
the one to use, no other elements need to be considered. If there are no data for rank PI, data for
rank P2 should be sought, and so on down the list until the highest ranked element is located.
Step Two: Use look-up table to find attribute score for value identified in Step One.
For each rank there is a corresponding "look up table" which contains a range of data values
assigned to a numeric prevalence score between 1 and 10. Once a data value has been found for
a particular element, that value can be looked up on these tables to determine the prevalence
score. The lookup tables are listed below.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-13
-------
CCL CL Work Group Report
Look-Up Table PI: Finished Drinking Water - Percentage of PWSs with Detections (national
data)
Category
Score
1
2
3
4
5
6
7
8
9
10
Total (SW and
C3W)
% PWSs with
detections
<. 0.10
0.11 -0.16
0.17-0.25
0.26 - 0.44
0.45 - 0.61
0.62-1.00
1.01 -1.30
1.31 -2.50
2.51 - 10.00
> 10.00
SW Only
% PWSs with
detections
! <. 0.22
Oj23 - 0.37
0!3B - 0.63
0:64-1.00
1,01 -150
1.51-2.00
2.01-3.10
3,11 -6.00
6.01 - 15.00
> 15.00
GW Only
% PWSs with
detections
<; 0.07
0.08-0.12
0.13-0.18
0.19-0.29
0.30 - 0.45
0.46 - 0.71
0.72-120
1.21-2.50
2.51 - 10.00
> 10.00
Record the attribute scores of all three measures. Use the Total for the Prevalence Score to look
up the Magnitude Score. In addition, if both surface water and ground water data are available,
score both SW and GW values and identify which provides a higher category score for
prevalence. If there is no distinction between SW or GW values score with the value provided.
If only surface water or ground water data are available, score the value using the corresponding
surface water or ground water column. Data for use with Table PI can be found in the data
sources NCOD and NIRS.
Look-Up Table P2: Percentage of Ambient/Raw/Source Monitoring Sites (national data) with
Detections
t
Prevalence Range
(% sites w/ Detects)
0.0 - 0.05
>0.05-0.1
>0.1-0.5
>0.5-1.0
>1.0-2.0
>2.0 - 5.0
>5.0-10
>10-20
>20 - 40
>40 - 100
Corresponding Score
1
2
3
4
5
6
7
8
9
10
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-14
t
-------
CCL CL Work Group Report
Data for use with Table P2 may be found in NAWQA.
Look-Up Table P3: Percentage of Ambient/Raw/Source Monitoring Samples (national data) with
Detections
I
Prevalence Range
(% samples w/ Detects)
0.0-0.05
>0.05-0.1
>0.1-0.5
>0.5-1.0
>1.0-2.0
>2.0-5.0
>5.0-10
>10-20
>20 - 40
>40 - 100
Corresponding Score
1
2
3
4
5
6
7
8
9
10
Data for use with table P3 are found in NREC.
Look-Up Table P4: Finished Drinking Water - Percentage of PWSs (local/state data) with
Detects
Prevalence
Score
1
2
3
4
5
6
7
8
9
10
All PWSs
Weighted
Average
% PWSs with
detections
£0.10
0.11-0.16
0.17-0.25
0.26 - 0.44
0.45 - 0.61
0.62-1.00
1.01 -1.30
1.31 -2.50
2.51 -10.00
> 10.00
SW Only
% PWSs with
detections
s 0.22
0.23 - 0.37
0.38 - 0.63
0.64-1.00
1.01 -1.50
1.51 -2.00
2.01 -3.10
3.11 -6.00
6.01 - 15.00
> 15.00
GW Only
% PWSs with
detections
<; 0.07
0.08-0.12
0.13-0.18
0.19-0.29
0.30 - 0.45
0.46 - 0.71
0.72-1.20
1.21 -2.50
2.51 - 10.00
> 10.00
For the workshop, no data were located for this table (fron the 41 data sources). It is anticipated
that these data will become available in the future. When these data are available they will fit
into this level of the hierarchy.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-15
-------
CCL CL Work Group Report
Look-Up Table P5: Percentage of Ambient/Raw/Source Water Sites (local/state data) with
Detects
Prevalence Flange
(% sites w/ Detects)
0.0-0.05
>0.05-0.1
>0.1-0.5
>0.5-1.0
>1.0-2.0
>2.0-5.0
>5.0-10
>10-20
>20 - 40
>40-100
Corresponding Score
1
2
3
4
5
6
7
8
9
10
For the workshop, no data were located for this table (from the 41 data sources). It is anticipated
that these data will become available in the future. When these data are available they will fit
into this level of the hierarchy.
Look-Up Table P6a: Number of States Reporting TRI releases
t
Number of States reporting
d ischarges
1
2
3
4
5
6
7-10
11-15
16-25
>25
Corresponding Score
1
2
3
4
5
6
7
8
9
10
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-16
-------
CCL CL Work Group Report
t
Look-Up Table 6b: Number of States Reporting Contaminant
in ATSDR HazDat (database of hazardous waste sites)
Number of States reporting
discharges to surface
water
1
2
3
4
5
6
7-10
11-15
16-25
>25
Corresponding Score
1
2
3
4
5
6
7
8
9
10
Look-Up Table P7: Surrogate Data for Pesticides
Number of States in which Pesticide
was used in 1997
Default for any pesticide for non-
environmental use
Default for any pesticide in
environmental use without use data
<6 in database
6-10
11-15
16-25
>25
Corresponding Score
3
5
6
7
8
9
10
The available data for Table P7 are from the source NCFAP.
Look-Up Table P8: Persistence Mobility
Please refer to the Persistence-Mobility Scoring Protocol for using P8.
Appendix C - Draft Protocols Developed and Used for Trial Scoring Exercise C-17
-------
CCL CL Work Group Report
Appendix D
Microbial Protocols / Attribute Scoring
Exhibit Dl lists the data elements associated with health effect and pathogen occurrence that
constitute the basis for the development of the attribute scoring system proposed in this
appendix.
Exhibit P1 - Data Elements for Pathogen Scoring
Attribute
Health Effect
Pathogen Occurrence
Elements
Severity of disease (manifestations, duration, sequelae, etc.)
Susceptible populations (number, immune status, etc.)
Incidence of disease (all sources vs. water-borne)
Availability and efficacy of available treatment
Infectious dose
Presence in source water
Persistence-mobility (stability, growth potential in water)
Documented water-borne outbreaks reported
Route of transmission (ingestion, inhalation or dermal contact)
Size of exposed populatioh
Potency
Potency is defined as the amount of a contaminant that is needed to cause illness. For microbes
the infective dose is the most useful marker of potency, however the infective dose is not known
for many pathogens. Microbiologists frequently speak in terms of the minimum infective dose,
but the terms LDso and lethal dose apply only to animal studies or in vitro cell culture assays.
Some pathogens cannot be grown in the laboratory and their infective dose can only be
estimated. In the future, quantitative virulence-factor activity relationships may become available
for determining the relative potency of a pathogen.
The data elements for scoring potency include knowledge of water-related disease, the class of
pathogen, i.e. bacteria, viruses, protozoa, the burden of disease in the population, the infectious
does of the pathogen, the likelihood of fecal or urinary shedding in humans and animals, and the
presence of genomic sequences conferring virulence.
A proposed system for scoring potency is shown in Exhibit D2. Data elements providing answers
to Category 1 questions are readily available from reference sources, however data elements for
Category 2 and Category 3 questions are not available for many pathogens on the PCCL. For this
reason, the questions are constructed in a manner to allow for uncertainty or unavailability of
data, while admitting the use of available information.
I
Appendix D -Microbial Protocols /Attribute Scoring
D-l
t
-------
CCL CL Work Group Report
Exhibit D2 - Potency Scoring Protocol
Category 1 v
Causeswater-related
disease in otherwise
healthy individuals
No water-related
disease, but organism is
a primary or
opportunistic pathogen
Category 2*
Category 3
Morbidity rats high
Morbidity rate low or
uncertain
Published human ID so
value available
Published humanID M
value notavailable
Viruses or protozoa
Bacteria or fungi
Enteric
Non -enteric
ID5o<102
IDso>102
Viruses or protozoa
Bacteria or fungi
Enteric
Non -enteric
Animal pathogen shed in faces or urine
Genetic sequences are
available in searchable
databases
Organism has known
virulence genes or gene
products
Known pathogenicity islands
No known pathogenicity islands
Virulence genes not documented
Score4
11
10
9
8
7
5
7
6
5
4
3
2
1
'Assumes all microbes on PCCLare pathogens, or potential pathogens, and occur or may occur in water.
2For scoring potency of toxins (e.g., various cyanotoxins, aflatoxins), assume that a toxin is a pathogen.
Infective dose based on single dose exposure to healthy individuals.
4Use single highest score for each microbe.
The most obvious data point for potency scoring is infective dose, however infective dose
data are rarely available, and extremely variable due to strain and host variability. Biological
properties of pathogens may be used to estimate potency where infective dose data are
unavailable. Assumptions built into the algorithm presume that viruses and protozoa have a
lower infective dose than bacteria, hence they score higher. Preliminary scoring exercises on
CCL organisms revealed little separation among scores. A more inclusive test data set would
probably provide a range of scores useful in ranking potency of pathogens. The position of
VFARs in the algorithm is controversial since genomic sequences are available for few
pathogens on the CCL, however the Work Group determined that manifestation of disease
(genomic expression) suggested higher potency than genomic potential (presence of virulence
genes or pathogenicity islands), or stated another way, functional data carry more significance
than structural data.
Severity
NRC defines severity as the seriousness of the health effect, and suggests severity be based on
"the most sensitive health endpoint for a particular contaminant, and considering vulnerable
subpopulations;... [and] should be based, when feasible, on plausible exposures via drinking
water."
The risk assessment terminology applicable to chemicals becomes problematic in the
microbiological context of the host-pathogen relationship. For microbial agents, severity may be
defined in terms of colonization, infection, immune response, disease, sequella, or death. The
host-pathogen relationship is variable and dynamic. This continuum may be unrecognizable at
Appendix D - Microbial Protoco Is/A ttrib ule Scoring
D-2
-------
CCL CL Work Group Report
various stages. The most sensitive endpoint indicative of host-pathogen interaction is an immune
response, however this is not a practical end point for assessment of health effects, since
immunodeficient populations may be infected withojut eliciting an immune response. While
chemical health effects may bs immediate or cumulative, microbiological health effects may be
unapparent for an extended time, depending upon the incubation period of the pathogen, and the
manifestation of disease.
The data elements for scoring severity include recognition of significantmorbidity and
mortality, the location and intensity of infectious processes, the extent of contagion, the amount
of time lost to illness, the extent to which medical intervention is required for recovery, and
chronic manifestations or disabilities associated with the disease.
A central issue with severity scoring is whether to score on acute manifestations of
disease in normal populations, or to score the worstipossible outcome in the most sensitive
population. Because most frank pathogens are capable of killing some segment of the population,
using worst possible outcome in the most sensitive host inflates and clusters scores. The initial
severity scoring tables were constructed to use median outcome in normal populations, with case
fatality rate and patent population classification and percentage of patients in the population
classifications as weighting factors. This approach was criticized as overly complex, and
potentially contentious, and the Work Group sought alternative scoring approaches.
One such approach applied the attribute characteristics to the population for which the
most data and information were available, then recalculating scores to acknowledge special
circumstances and to apply additional stringency. This proposed system applied worst case
scoring criteria for healthy and sensitive sub-populations, thereby driving many pathogens to
maximal scores. In an effort to overcome the complexities and limitations of a scoring system
using case fatality rates and population-based weighting factors, the Work Group proposed a
series of questions carefully constructed so that a 'yes' answer signified significance while a 'no'
answer did not.
Twelve questions were constructed to capture progressively severe outcomes of disease,
and the sum of 'yes' answers constitutes the numerical severity score for a particular pathogen
(Exhibit D3). This binary scoring process was conducted for typical and worst case disease
outcomes in normal and sensitive sub-populations for the microbial contaminants on the current
CCL. The results provided reasonable spread for both patient populations, although scoring
'worst case' tended to cluster pathogens toward the upper end of the scale. While 'worst case'
scoring is believed to provide the highest level of public healh protection, it fails to consider
existence of other reservoirs und transmission routes for pathogens besides drinking water, and
places undue responsibility for prevention of infectious diseases on the EPA regulatory process.
By limiting the manifestations of disease to those related to infections acquired by ingestion,
inhalation, and dermal contact with drinking water, the binary scoring system produces plausible
results for severity of illness.
t
Appendix D - Microbial Protocols /A ttribute Scoring
D-3
-------
CCL CL Work Group Report
Exhibit D3 - Severity Scoring Protocol
Question
1
2
3
4
5
6
7
8
9
10
11
12
Total Score2
Yes1
Question
Does the organism cause a CDC notifiable disease?
Does the organism cause significant morbidity (> 1 ,000/year) in the U.S.?
Is diarrhea a symptom of illness?
Does the illness require medical intervention for resolution?
Does the organism disseminate from the gastrointestinal tract to other organs?
Does the organism cause mild disease in normal populations, but severe disease in
individuals with predisposing conditions?
Is illness associated with 3 or more days lost from school or work?
Is person-to-person spread a typical component of the disease syndrome?
Does illness usually require hospitalization?
Does the organism cause pneumonia, meningitis, hepatitis, encephalitis, endocarditis,
or other severe manifestations of illness?
Does illness result in long-term disability orsequella?
Does the organism cause significant mortality (> 1 /1 ,000 cases)?
1Enter 1 for each yes answer.
2Add the numbers in the column to arrive at a final score.
Prevalence
For the occurrence attributes, NRC defines prevalence as, "How commonly does or would a
contaminant occur in drinking water?" Prevalence may be determined via the seven measures
proposed by NRC in the PCCL screening criteria for demonstrated or potential occurrence (in
order of preference): (1) tap water, (2) distribution systems, (3) finished water of water treatment
plants, and (4) source water used for supplying drinking water. If no information is available to
demonstrate occurrence in water, NRC recommended evaluating the potential for occurrence in
water through: (5) observations in watersheds/aquifers, or (6) historical contaminant release data.
It should be emphasized that prevalence involves the consideration of both geographical (spatial)
and temporal ranges of occurrence.
Most pathogen occurrence data are based upon indicator monitoring, hence they become
surrogate information, not pathogen occurrence data. True pathogen occurrence data come from
epidemiological investigations following outbreaks, research studies on pathogen distribution,
and detection method evaluations. There is little pathogen information and less pathogen data
regarding environmental and drinking water occurrence.
The Work Group developed a conceptual framework for prevalence, based upon actual
detection in drinking water, actual detection in source water, potential for zoonotic transmission
through water contamination, and potential for zoonotic agents to infect humans (host range).
These factors are the basis of Exhibit D4. Prevalence scoring using these criteria proved
to be more straight forward that other attributes, primarily because occurrence data are either
available or not available, limiting the number of criteria in the scoring system.
Appendix D - Microbial Protocols /Attribute Scoring
D-4
-------
CCL CL Work Group Report
Exhibit D4 - Prevalence Scoring Protocol
Category 11
Category 2
Detected in drinking water
Not detected in drinking water, but
detected in source water
Not detected in drinking water or
source water
Documented WBD in sWimmersin the U.S.
Common in source water
Monitored but infrequently detected
Rarely monitored or never detected
Broad host range for animals and humans
Narrow host range limited primarily to humans
Score2
7
6
5
4
3
2
1
1 Based upon worldwide occurrence data.
2Selectthe single highest score fore ach organism.
Persistence-Mobility
NRC used a persistence/mobility attribute as a surrogate for potential occurrence when
information is unavailable for a contaminant regarding its demonstrated occurrence in water. For
microorganisms, the following three characteristics pertain to their persistence and/or mobility:
high potential for amplification under ambient conditions, sedimentation velocities and
absorption capabilities, and death or the ability to produce non-culturable or resistant states (e.g.
spores and cysts). When a contaminant already has data on demonstrated occurrence in water,
and thus information for the prevalence and magnitude attributes, those attributes will take
precedence over persistence/mobility.
Pathogenic microorganisms are genetically adapted to their hosts, and they do not typically
survive the rigors of the ambient environment. Some fastidious pathogens such as Treponema
pallidum and HIV are inactivated within seconds of exposure to the ambient environment. The
factors determining the persistence of microorganisms in aquatic systems include:
• ability to withstand ambient conditions of temperature, pH, ionic strength, radiation and
oxygenation
• ability to compete with other microorganisms for substrate
• ability to produce or sequester themselves in biofilms or adsorbed to particles
• ability to produce resting forms, e.g. cysts or spores
• ability to persist in viable but non cultivable state
• ability to enter into commensal or symbiotic relationships with other microorganisms
• ability to resist disinfectants
• presence of predators, e.g. amoeba, cilliates, etc.
Persistence implies steady state occurrence or amplification of microorganisms in water. This
occurs in surface water by production of resistant forms such as spores, cysts, oocysts, by
colonization of other life forms serving as a reservoir, through symbiotic relationships with
amoebae, by adsorption to particles, or production of quiescent forms such as viable but non-
culturable bacteria. In water treatment plants and distribution systems, persistence is associated
with colonization of infrastructure, e.g. production of biofihn. Organisms that amplify are given
higher scores than organisms that produce resistant forms but do not amplify in water. This
Appendix D - Microbial Protocols /A (tribute Scoring
D-5
-------
CCL CL Work Group Report
scoring scale may overemphasize relatively innocuous organisms that produce biofilms but
rarely or never cause disease in humans.
Data elements for scoring persistence-mobility include survival time in water under
ambient conditions, ability to amplify, ability to produce resistant forms, relationship to particles,
and potential for symbiotic relationships enhancing survival.
The persistence-mobility scoring table (Exhibit D5) emphasizes non-turbid waters, i.e.
groundwater and treated drinking water, but the Work Group believes that all source water
should be included in scoring. The rationale for excluding turbid waters was that organisms
adsorbed to particles persist considerably longer than organism in non-turbid waters.
Amplification frequently occurs in surface source water due to large amounts of available
nutrient, whereas the assimilable organic carbon is limited in groundwater and treated water,
slowing or restricting amplification. For example, Aeromonas hydrophila grows to population
densities in excess of 8 logs per mL in sewage, from 4-6 logs per mL in surface water, while
maximal levels in distribution water rarely exceed 2 logs per mL (typical levels range from 10"2
to 10 CFU/mL), and groundwater is typically less than 1 CFU/mL. Persistence of bacteria, which
amplify under environmental conditions is highly variable, and the extent to which they persist
and move is largely a function of their population density. It may be inappropriate to equate
persistence-mobility of organisms in surface waters with persistence-mobility in non-turbid
waters.
Mobility is not limited to chemicals, since microorganisms move though the aqueous
environment and in distribution system water actively (motility) and passively (adsorbed to
particulates, in symbiotic relationship with amoebae, and by hydrostatic flow). Organisms
percolate through soil layers to contaminate groundwater. Viruses are particularly mobile
because of their extremely small size and their relatively long survival times in the environment.
Because mobility is associated with the hydrodynamics of distribution systems, presence of
biofilms, presence of particulates, and opportunity for symbiotic relationships, it is considered
together with persistence for scoring purposes.
Exhibit PS - Persistence-Mobility Scoring Protocol
Stability in Non-Turbid Water1
Usually dies rapidly in water (days)
Stability uncertain, no amplification
Stable for weeks to months, no amplification3
Stable for weeks, months, or years, with amplification or
protection from symbiotic relationships
Score2
2
3
4
5
1Non -turbid water is defined as ground water or filtered surface water.
2Select the single highest score for each organism.
Development of endospores, cysts or oocysts
4Capsule or slime production, protection by amoebae, autofrophic metabolism
Appendix D - Microbial Protocols /Attribute Scoring D-6
-------
CCL CL Work Group Report
Magnitude
NRC defines magnitude as "the concentration or expected concentration of a contaminant
relative to a level that causes a perceived health effe'ct" (NRC 2001). For characterizing the
attribute of magnitude, ideally two data elements art needed: the concentration of a contaminant
in water, and the concentration associated with an adverse health effect. NRC recommended the
use of a median water concentration in combination with a measure of potency, if available.
Magnitude, in a microbiological context, implies delivery (persistence-mobility) of an
infective dose (potency) to the customer's tap with resulting illness. The Work Group proposes
to score magnitude according to the number and frequency of waterborne disease outbreaks
reported in the U. S. and around the world, pathogen distribution, and biological properties
determining pathogen distribution. A scoring table ijs shown in Exhibit D6. This algorithm has
not been adequately evaluated using a test data set.
Exhibit D6 -Magnitude Scoring Protocol
Category1'2
Has caused numerous recently documented WBDOs in the U jS. or other developed country
Rarely causes documented WBDOs in the U.S. or other developed country
Has not caused documented WBDO in the U.S. or other developing countries, but has caused
documented foodbome outbreaks
Has caused numerous recent documented WBDO in developing countries, but its biological
properties would mitigate against causing WBDO in the U.S.
Rarely causes documented WBDOs in developing countries, but its biological properties would
mitigate against causing WBDO in the U.S.
Has never caused WBDOs in any country, or its biological properties mitigate against causing
WBDOs in the U.S.
Score3
6
5
4
3
2
0
1U.S. is defined asthe 50 slates and territories.
\Vaterborne disease outbreaks (WBDOs) associated with drinking water, fresh water used for recreation, and
outbreaks associated with hot tubs, swimming pools, etc. where makeup water is drawn from potable water sources.
3Selectthe single highest score fe roach organism.
Work Group discussion on organisms known to cause waterborne disease outbreaks
concluded that such pathogen s could be placed directly on the CCL (known pathogens causing
WBDO are already on the CCL, with exclusions based upon treatment effbacy), and that
attribute scoring would not play a significant role in moving them from the CCL Universe to the
CCL. While priority would be given to domestic outbreaks, organisms causing WBDO in other
countries would be evaluated for their public health significance in the U.S.
A.
Microbial Data Elements for Attribute Scoring
By obtaining information on the attributes of each of these elements for each known or
prospective pathogen, it is possible to assess the relative risk and prioritize pathogens according
to their occurrence and health effects.
Elements Considered in Patiiogen Occurrence
Spatial distribution (clumping, particle-association, clustering)
Appendix D -Microbial Protocols /Attribute Scoring
D-7
-------
t
t
CCL CL Work Group Report
Concentrations in environmental vehicles and foods
Seasonality and climatic effects
Temporal distribution, duration, and frequency
Niche (potential to multiply or survive in specific media)
Amplification, die-off, persistence
Indicators/surrogates predictive of pathogens
B. Elements Considered in Exposure Analysis
Identification of water and other media
Unit of exposure
Temporal nature of exposure (single or multiple; intervals)
Route of exposure and transmission potential
Demographic of exposed population
Size of exposed population
Behavior of exposed population
C. Elements Considered in Pathogen Characterization
Virulence and pathogenicity of the microorganism
Pathological characteristics and diseases caused
Survival and multiplication of the microorganism
Resistance to environmental control measures
Host specificity
Infection mechanism and route; portal of entry
Potential for secondary spread
Taxonomy and strain variation
D. Elements Considered in Host Characterization
Demographics of the exposed population (age, density, etc.)
Immune status
Pregnancy
Concurrent illness or infirmity
Nutritional status
Genetic background
Behavioral and social factors
E. Elements Considered in Health Effects
Morbidity, mortality, sequelae of illness
Severity of illness
Duration of illness
Chronic or recurrent
Potential for secondary spread
t
Appendix D - Microbial Protocols /A ttribute Scoring D-8
-------
CCL CP Work Group Report
Appendix E
Prototype Classification Methods/ Results of Pilot
Demonstration
s
Following is a brief description of different model classes discussed by the NOW AC Work
Group that might be used for a prototype based classification approach. Four of these were
part of a demonstration exercise.
Linear Discriminant Analysis
Linear discriminant analysis can be thought of as a special case of linear regression analysis.
In linear regression analysis a response variable is described as a linear function of one or
more predictor variables:
Y=p0 + p,X, + e
where Y is the response variable, Po is an intercept term, Xi is a predictor variable, Pi is me slope
parameter and e is an error term which acknowledges that Y may not be perfectly predictable by
knowing XL Linear regression analysis refers to the procedure in which optimal values for Po and
pi are estimated, given a data set that consists of observations for Y and X] . A linear regression
model predicts me value of the response variable, given the values of a set of predictor variables,
while the linear discriminant model predicts which category the response variable is likely to
belong to, given the values of the predictor variables.
Logistic Regression
Generalized linear models are a class of models that have a linear model as their basis, but
the response variable is now a function of the linear model:
where f represents some mathematical function. Logistic regression is a special case of the
generalized linear model, in which the response variable is categorical, with two categories.
Artificial Neural Networks (ANN)
Artificial Neural Networks consist of a base node and several "hidden" nodes. The base node
is typically a logistic model and the hidden nodes further refine the results of the logistic
model. The number of hidden nodes included in the model is determined by evaluating the
improvement in model prediction with each additional node. Generally, adding hidden nodes
requires a large data set to ensure that the additional model structure is not just capturing the
idiosyncrasies of a small sample.
Classification and Regression Trees (CART)
A CART Model is analogous to a dichotomous key, used in biology to determine an animal
or plant species based on observable characteristics of the organism. The value or category of
the response variable is deteimined by evaluating the way in which the response variable is
related to the predictor variables as a set of "if- then" statements (i.e. if Xi is greater than a
value, and Xz is less than some value then Y belongs in a particular category). This model
Appendix E: Prototype Classification Methods /Results of Pilot Demonstration E-l
-------
t
CCL CP Work Group Report
can be visually depicted in a diagram that looks like a branching tree. Recent
implementations of CART include the ability to accommodate some missing data not only in
the training data set, but also among new observations.
Multivariate Adaptive Regression Splines (MARS)
Generalized Additive Models are a class of nonlinear models in which the relationship
between the response and predictor variables is not pre-specified by a particular
mathematical function. Instead the relationship is developed from the observed data. The
procedure divides the data into regions, similar to a moving window, and estimates a smooth
nonlinear relationship among the data within each region. Multivariate Adaptive Regression
Splines are a form of generalized additive model that allow the inclusion of interactions
among the predictor variables similar to interactions that might be included in various forms
of linear regression or analysis of variance models.
Model Pilot Demonstration
An exercise by the technical support team for the Work Group demonstrated how these
models work. First, 46 contaminants were selected to comprise a training set. Each
contaminant was scored for five attributes (severity, potency, prevalence, magnitude, and
persistence/mobility) according to a draft scoring protocol. Next, each contaminant was
assigned a decision: either "list" or "do not list." Finally, the data were used to inform the
Logistic Regression, ANN, CART, and MARS models. This was done in two steps. First,
the structure of a "best-fit" model was determined for each of the four model classes. In the
second step, the models from each class were compared to show how each class of model
performed with an example data set.
Model Selection within each of the four model classes
"Over-fitting" is a concern when selecting a best-fit model. Any of these four model types
could be made to fit a particular data set very well by making the model more complex (this
usually means estimating more model parameters). However, the addition of model
complexity can come at the cost of a loss of generality; the added complexity may capture
the idiosyncrasies of the specific training data set, and may not be representative of the
broader processes that generate the data. Several methods were used as guidance to avoid
over-fitting, depending on the specific model being fit. Cross-validation is a technique in
which the data set is repeatedly, randomly sub-divided and a model is fit to a subset of the
data, then used to predict the complementary subset that was "left-out" of the fitting process.
A second method is the Bayesian Information Criterion a combined measure of a model's
predictive capability and complexity. For the logistic regression, standard classical methods
of assessing "statistical significance" were used as guidance for the number of predictor
variables that should be included in the model.
Comparison of the pilot example models
Each of the four model classes was assessed using a "ten-fold" cross validation procedure.
The data set was randomly divided into 10 roughly equal sized groups. The four models
Appendix E: Prototype Classification Methods /Results of Pilot Demonstration E-2
-------
CCL CP Work Group Report
were fit 10 times using the selected models, each time setting aside one group of the data.
The fitted models were than applied to the left-out data set and the predictive
misclassification rates for each model were recorded. The exercise showed that the four
model classes could be compared by misclassification rates of a training data set.
Lessons Learned
Lessons learned were limited by the use of a small set of contaminants scored by a draft (not
final) attribute scoring protocols. Specific findings, such as estimated classification error
rates, could be inaccurate predictors of performance in the future, when the models are
informed by a complete training set, with well-justified "list'Vdo not list" decisions, and
final scoring protocols.
The major lessons learned in the exercise were:
o The major cost of running any model is development of the training data set (i.e.,
developing attribute scores, selecting the training data set, and scoring the training
data set contaminants). Once the training data set is available, computer processing
(training, cross-validation, diagnostics) is relatively quick . Additional time and
resources may be required to modify the training data set based on the results of the
diagnostic exercises.
o All four models can classify contaminants biased on complete training data sets.
o All models could deal with integer attribute scores.
o All models could deal with raw attribute data
o The CART model could deal with missing data/scores. Other models could
not, and therefore had smaller training data sets. This may also require
consideration in developing the attribute scoring protocols and selecting
training set contaminants.
o All four models provided diagnostic information:
o Estimated classification error rates can indicate whether the training set is of
adequate size.
o Estimated classification error rates may allow rejection of one or more
models.
o Comparing results across models provides information on the training set
contaminants:
o AH four models correctly classify most contaminants.
o Four contaminants were misclassified by all four models, suggesting that the
five attribute protocols scores may need to be refined to account for
characteristics of these contaminants.
Appendix E: Prototype Classification Methods / Results of Pilot Demonstration E-3
------- |