EPA-820-R-14-009
Microbiological Risk Assessment (MRA)
Tools, Methods, and Approaches for Water Media
December 2014
Office of Science and Technology
Office of Water
U.S. Environmental Protection Agency
Washington, DC 20460
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Notice
The tools and procedures set forth in this document are intended to describe the United States
(U.S.) Environmental Agency's (EPA) approach for conducting or revising microbial risk
assessments to protect human health from exposure to water-based media. They are also intended
to serve as guidance to EPA and EPA contractors for conducting microbial risk assessments.
This document has been reviewed in accordance with Agency policy and is approved for
publication and distribution.
Mention of commercial products, trade names, or services in this document or in the references
and/or footnotes cited in this document does not convey, and should not be interpreted as
conveying, official EPA approval, endorsement, or recommendation.
11
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Foreword
This document presents a series of tools, methods, and approaches for planning and conducting
microbial risk assessments in support of human health protection for water-based media. This
document provides guidance for microbial risk assessments conducted or revised by EPA or EPA
contractors and should not be considered regulatory.
The tools and approaches described herein focus on conducting risk assessment for water-related
media (such as microorganisms in treated drinking water, source water for drinking water,
recreational waters, shellfish waters, and biosolids), but are sufficiently general to help guide the
development of microbial risk assessments of pathogens that might be found in food, food
products, or other media.
Director
Office of Science and Technology
ill
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Acknowledgments
Project Leader
Stephen Schaub*
U.S. EPA Office of Science and Technology
U.S. EPA Technical Reviewers
Alfred Dufour
Stig Regli
Philip Berger
Sharon Nappier
U.S. EPA Office of Research and Development/National Exposure
Research Laboratory
U.S. EPA Office of Ground Water and Drinking Water
U.S. EPA Office of Ground Water and Drinking Water
U.S. EPA Office of Science and Technology
John Ravenscroft** U.S. EPA Office of Science and Technology
* Principal author
** Contact
Initial draft report developed by ICF International under U.S. EPA Contract 68-C-02-009; interim
draft developed by Tetra Tech Clancy Environmental, Soller Environmental, and ICF International
under U.S. EPA Contract EP-C-07-036; final report developed by ICF International and Soller
Environmental under U.S. EPA Contract EP-C-11-005.
Contributing authors: Audrey Ichida, Jeffrey Soller, William Mendez, Jonathan Cohen, Timothy
Bartrand, Mark Gibson, Jennifer Welham, Margaret McVey, Gunther Craun, Walter Jakubowski,
Cynthia McOliver, and Martha Embrey.
The International Life Sciences Institute Risk Science Institute (ILSI-RSI) under cooperative
agreements with the U.S. EPA Office of Water, Office of Science and Technology, Health and
Ecological Criteria Division, developed a report titled Revised Framework for Microbial Risk
Assessment to address critical areas of microbial risk assessment (ILSI, 2000). This MRA Tools
document evolved out of that report. The steering committee for the development of that report
included the following:
Stephen Schaub
Gunther F. Craun
Alfred Dufour
Charles Gerba
Charles Haas
Alan Roberson
Mark Sobsey
U.S. EPA Office of Science and Technology, lead
G.F. Craun & Associates
U.S. EPA Office of Research and Development
University of Arizona
Drexel University
American Water Works Association
University of North Carolina
IV
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
External Peer Review Workgroup
This document has been subject to external peer review following the Office of Management and
Budget's Peer Review Guidance (2004) and EPA's Peer Review Handbook.
Potential areas for conflict of interest have been investigated via direct inquiry with the potential
peer reviewers and review of their current and past affiliations. Reviewers did not have conflicts
of interest.
Peer reviewers included:
Margaret Coleman Syracuse Research Corporation (currently Coleman Scientific Consulting)
Joseph Eisenberg University of Michigan
Charles Gerba University of Arizona
A subsequent draft version of this document was reviewed by EPA's Science Advisory Board -
Drinking Water Committee. The committee met on September 21-22, 2009. This document has
been revised in response to their comments.
v
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
vi
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
Table of Contents
Notice ii
Foreword iii
Acknowledgments iv
External Peer Review Workgroup v
Tables, Figures, and Text Boxes ix
Abbreviations and Acronyms xi
Executive Summary xiii
1. Introduction 1
1.1. Purpose and Scope of this MRA Tools Document 1
1.2. Development of the MRA Tools Document 3
1.3. MRA Framework 7
1.4. General MRA Concepts 11
1.5. Microbial Risk Assessment for Decision-Making 12
1.6. Factors Unique to Microbial Risk Assessment as Compared to Chemical
Risk Assessment 13
1.6.1. Microbial Growth and Death 13
1.6.2. Detection Methodologies 14
1.6.3. Genetic Diversity of Pathogens 15
1.6.4. Host Immunity and Susceptibility 15
1.6.5. Dose-Response Range can be Broad 15
1.6.6. Secondary Transmission 15
1.6.7. Heterogeneous Spatial and Temporal Distribution 16
1.6.8. Zoonotic Potential 16
2. Planning and Scoping and Problem Formulation 17
2.1. Introduction to Planning and Scoping and Problem Formulation 17
2.2. Overall Problem Formulation and Planning and Scoping 20
2.2.1. Statement of Concern 21
2.2.2. Statement of Purpose and Objectives 21
2.2.3. History and Context within the Agency 23
2.2.4. Scope 23
2.2.5. Questions to be Addressed in the Risk Assessment 25
2.2.6. Conceptual Model and Narrative 26
2.2.7. Planning and Scoping: Analysis (Operational) Plan 28
2.3. Analytical Approaches 28
2.3.1. Representative Model Forms for MRA Risk Estimation 30
2.3.2. Data Representation in MRA Risk Estimation Models 35
2.4. Elements to Consider During Problem Formulation 40
2.4.1. Infectious Disease Hazard Characterization 40
2.4.2. Initial Host Characterization 46
2.5. Linkage between Problem Formulation and Other MRA Components 49
3. Exposure Assessment 51
3.1. Occurrence 53
3.1.1. When Do Pathogens Occur in the Water Body? 53
3.1.2. Where Do Pathogens Occur in the Water Body? 55
3.1.3. What is the Level of Pathogens in the Water Body? 56
3.1.4. Interpretation of Analytical Methods 59
vii
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.2. Exposure Analysis 60
3.2.1. Identification of Media 61
3.2.2. Routes of Exposure 61
3.2.3. Units of Exposure 62
3.2.4. Spatial Nature of Exposure 65
3.2.5. Behaviors of Exposed Population 66
3.3. Exposure Profile and Linkage between Exposure Assessment and Other
MRA Components 66
3.3.1. Exposure Estimation 66
3.3.2. Exposure Description 67
4. Human Health Effects and Dose-Response 69
4.1. Human Health Effects Overview 69
4.1.1. Duration of Illness 69
4.1.2. Severity of Illness 70
4.1.3. Morbidity, Mortality, and Sequelae 70
4.1.4. Extent of Secondary Transmission 82
4.2. Dose-Response Assessment Overview 82
4.2.1. Overview of Common Dose-Response Model Forms for Pathogens 85
4.2.2. Summary of Available Dose-Response Relationships for Waterborne Pathogens .. 94
4.3. Host-Pathogen Profile and Linkage between Human Health Effects
Assessment and Other MRA Components 98
5. Risk Characterization 101
5.1. Introduction to Risk Characterization 101
5.1.1. Historical Context 103
5.2. Risk Estimation and Risk Description 103
5.3. Uncertainty and Sensitivity Analysis 104
5.4. Representative Examples of MRAs 110
6. References 113
viii
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Tables, Figures, and Text Boxes
Table 1. Elements of Microbial Risk Assessment (Source: Adapted from ILSI, 2000;
U.S. EPA, 2003c, 2003f, and 2012a) 9
Table 2. Overview and Comparison of Static and Dynamic Risk Assessment Models 30
Table 3. Distributions used in Monte Carlo simulations conducted by Olivieri et al. (1999) 39
Table 4. Representative Tools for Modeling Pathogen Survival, and Multiplication 41
Table 5. Virulence of Three Cryptosporidium parvum Isolates in Healthy Adult Humans
(Source: Okhuysen et al., 1999) 46
Table 6. Sources and Scales of Variability in Pathogen Occurrence 54
Table 7. Tools and Databases for Evaluation of Occurrence 57
Table 8. Average Volume Water Swallowed (mL) per Swimming Event (Schets et al., 2011) . 64
Table 9. Typical Incubation Periods for Some Waterborne Pathogens 70
Table 10. Overview of Dose-Response Relationships and Health Effects for Waterborne
Pathogensa (Source: Adapted from McBride et al., 2002) 95
Table 11. Approaches to Sensitivity and Uncertainty Analysis Recommended in EPA's
Exposure Factors Handbook (Source: U.S. EPA, 1997a) 106
Table 12. Sensitivity Analysis Methods and Techniques (Adapted from Frey and Patil, 2002;
Frey et al., 2004) 108
Figure 1. Risk Analysis 2
Figure 2. WHO Water Quality Framework (Source: Adapted from WHO, 2001) 2
Figure 3. Framework for Human Health Risk Assessment to Inform Decision Making
(Source: U.S. EPA, 2012a) 8
Figure 4. Enhanced Problem Formulation Process Diagram (Source: Adapted from U.S.
EPA, 2003 c) 20
Figure 5. Example of an Overview (Top-Tier) Conceptual Model 28
Figure 6. Infectious Disease Model Features for Use in Model Selection 31
Figure 7. Static Risk Assessment Conceptual Model 32
Figure 8. Dynamic Risk Assessment Conceptual Model (Source: Soller and Eisenberg, 2008). 34
Figure 9. Two Versions of the Epi Triad (Source: CDC, 1992) 42
Text Box 1. Microbiological Risk Profile for Food (Source: Adapted from CAC, 2007) 22
Text Box 2. Information Used to Establish Risk Ranges and Representative Examples of
Risk Ranges Currently Employed by U.S. EPA 25
Text Box 3. Examples of Risk Management Questions that Could Motivate an MRA
Investigation 26
Text Box 4. Exposure Analysis for the Long Term 2 (LT2) Enhanced Surface Water
Treatment Rule (Source: U.S. EPA, 2006a) 60
Text Box 5. Examples of Time Units Associated with Exposure 64
Text Box 6. Dose-Dependency of Host-Pathogen Interactions 85
Text Box 7. Brief Summary of Cryptosporidium Feeding Studies 88
Text Box 8. Brief Summary of Challenge Studies to Investigate the Dose-Response and
Host-Immunity Factors Related to Norovirus Infection 89
Text Box 9. Summary of Use of Exponential Model (Source: Rose et al., 1991) 90
Text Box 10. Two-Dimensional Probabilistic Risk Analysis of Cryptosporidium in Public
Water Supplies, with Bayesian Approaches to Uncertainty Analysis (Source: from case
study #8 in U.S. EPA, 2009b) 109
IX
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
x
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Abbreviations and Acronyms
AGI acute gastrointestinal illness
AIDS acquired immune deficiency syndrome
ANOVA analysis of variance
ARS Agricultural Research Service (USDA)
ASM American Society for Microbiology
AWQC Ambient Water Quality Criteria
CAC Codex Alimentarius Commission (Codex)
CART classification and regression tree
CDC U.S. Centers for Disease Control and Prevention
Codex Codex Alimentarius Commission
CFU colony forming units
CWA Clean Water Act
DALY disability-adjusted life years
EMPACT Environmental Monitoring for Public Access and Community Tracking
EO Executive Order
EPA U.S. Environmental Protection Agency
FAO Food and Agricultural Organization (United Nations)
FDA U.S. Food and Drug Administration
FUT2 alpha (1,2) fucosyltransferase gene
GI gastrointestinal (tract)
HACCP Hazard Analysis and Critical Control Point
HIV human immunodeficiency virus
HHRA human health risk assessment
ICR Information Collection Rule
ID50 infectious dose for 50% of the exposed population
IgG immunoglobulin G
ILSI International Life Sciences Institute
L liter
LD50 lethal dose for 50% of the population
LT2 Long Term 2 Enhanced Surface Water Treatment Rule
MAC Mycobacterium avium complex
MCMC Markov Chain Monte Carlo method
mL milliliter
MLE maximum likelihood estimation
MRA microbial risk assessment
MRM microbial risk management
NAS National Academy of Sciences
NRC National Research Council
NV Norwalk Virus
PCR polymerase chain reaction
PFU plaque forming unit
PMP Pathogen Modeling Program
RA risk assessment
QALY quality-adjusted life years
QMRA quantitative microbial risk assessment
RSI Risk Science Institute (ILSI)
XI
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
RT-PCR
reverse transcriptase polymerase chain reaction
SDWIS
Safe Drinking Water Information System
SMV
Snow Mountain Agent Virus
TCCR
transparency, clarity, consistency, and reasonableness
TMDL
total maximum daily load
U.S.
United States
USD A
U.S. Department of Agriculture
VBNC
viable but non-culturable
WHO
World Health Organization (United Nations)
Xll
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Executive Summary
Exposure to waterborne pathogens has long been recognized as a potential source of illness in
humans. Managing and minimizing this public health threat is an important aspect of the United
States (U.S.) Environmental Protection Agency's (EPA) Office of Water regulatory activities and
policy development. Risk assessment is a science-based tool that can be used to help managers
explore the relative merits of various management alternatives, identify important gaps in
knowledge, and inform regulatory actions.
This Microbial Risk Assessment (MRA) tools, methods, and approaches ("MRA Tools") was
developed to assist U.S. EPA and others in conducting MRAs—including quantitative microbial
risk assessments (QMRAs1) focused on human health risks from exposure to pathogens. The
primary audience for this document is EPA staff and contractors who are responsible for
conducting and managing MRAs for pathogens that occur in water and water-related media. Thus,
this document is intended to summarize MRA methods and techniques for risk assessors and
scientists, and provide a compilation of information that is useful for conducting rigorous, well
documented, and scientifically defensible MRAs. It is not however, intended to be a
comprehensive treatise, a step-by-step protocol, nor a textbook on the topic of MRA. In addition,
the MRA tools document does not address deriving water quality values for microbial indicators
of fecal contamination (e.g., E. coli, enterococci, bacteriophage), that information is addressed by
EPA's recreational water quality criteria Technical Support Materials (U.S. EPA, 2014). Although
the principle medium of interest is water and water-related media (e.g., recreational waters,
drinking water sources, shellfish harvesting waters, biosolids), select resources for food safety risk
were also consulted in the development of this document.
This MRA Tools document should be considered flexible and amenable to modification where an
Office or other user has particular requirements that may not be precisely covered in the text.
Moreover, the tools, methods, and approaches described herein should be considered a modular
toolbox with a broad scope. It is expected that those modular aspects that are relevant to the MRA
being conducted can be used as deemed appropriate by the EPA Office conducting the assessment.
It is also expected that the various EPA Offices will have different needs in terms of how an MRA
is documented. For example, some Offices may have a preference that specific components of the
MRA be documented in a specific section of the MRA report that may be different from the manner
described herein. This MRA Tools document provides for such conditions and should be
considered amenable and flexible in this regard. This document does not evaluate ongoing state-
of-the-art research within the field of MRA.
Microbial risk assessments can be initiated for a variety of reasons, including but not limited to
the following:
• to assess the potential for human risk associated with exposure to a known pathogen;
• to determine critical points for control, such as watershed protection measures;
• to determine specific treatment processes to reduce, remove, or inactivate pathogens;
• to predict the consequences of various management options for reducing risk;
• to identify and prioritize research needs; and
• to assist in interpretation of epidemiological investigations.
1 For the purposes of this document, the term MRA also includes QMRAs.
Xlll
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Individual risk assessments for specific situations can differ significantly with respect to the
questions that are addressed, the information required to address those questions, and the nature of
data gaps.
This MRA Tools document is comprised of a combination of concepts from numerous published
risk assessment frameworks and workshop proceedings. Although many of these frameworks and
proceedings were originally developed for other applications such as food safety, there are many
principles that also apply to water-related risk assessments. This MRA Tools document employs
an expanded and enhanced version of the EPA-International Life Sciences Institute (ILSI)
Framework for MRA. For the purposes of this document, the EPA-ILSI structure has been
modified in the recognition that MRA practitioners and managers often desire flexibility in the
development of MRAs, and that the National Academy of Sciences (NAS), National Research
Council (NRC) chemical risk framework and the EPA human health risk assessment framework
are also compatible with MRA. A common theme among frameworks is the iterative nature of risk
assessment. The modeling steps in risk assessment might be repeated multiple times as the scope
of the assessment is refined or as risk management questions evolve. Additional data and
sensitivity analyses also require repeated iterations.
Chapter 1 provides an introduction to MRA and summarizes concepts that are used throughout
this document, including the purpose and scope of this document and background information. It
also includes an overview of appropriate frameworks for the conduct of MRAs, such as the
frameworks developed by the NRC in 1983 and 2009, and previous EPA guidance. Chapter 2
describes problem formulation and planning and scoping; it includes factors for consideration
during problem formulation and a description of how problem formulation can be used to track
the risk assessment progress and process. Hazard identification is also discussed in Chapter 2 as
one critical component of the problem formulation process.
Exposure assessment is discussed in Chapter 3. Subtopics within the characterization of exposure
include the occurrence of the infectious disease hazard, exposure assessment, and the exposure
profile (a summary of the results of the exposure characterization process). Human health effects
assessment, including the dose-response assessment, is the focus of Chapter 4. Subtopics within
human health effects assessment include description of health effects, dose-response relationship,
and the host-pathogen profile (summary of the results of the health effects assessment). Common
forms of dose-response models are also summarized. In many cases EPA risk assessment
documents consider the exposure assessment and human health effects assessment as two steps
within an analysis phase. The third step of the analysis phase is the risk estimation, which brings
together the exposure, and health effects assessments.
Chapter 5 discusses the risk characterization phase of MRA. The topics summarized include the
historical context of risk characterization within EPA. Uncertainty analysis and sensitivity analysis
are discussed within the context of risk characterization. Some risk assessment documentation
includes the risk estimation step (bringing together the exposure and health effects assessments)
within the risk characterization phase. The risk characterization phase can be interpreted as
including the risk estimation step of the analysis, or only including the evaluation and context of
the results presented in the analysis phase.
A total of three appendices (A-C) are also included with this MRA Tools document, to provide
interested readers with additional detail on topics that are included in this document.
xiv
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
1. Introduction
1.1. Purpose and Scope of this MRA Tools Document
The purpose of this document is to summarize microbial risk assessment (MRA) tools, methods,
and approaches for risk assessors and scientists. It also provides a compilation of information that
is useful for conducting rigorous and scientifically defensible MRAs for pathogens that occur in
water and water-related media. Although the principle medium of interest is water (e.g.,
recreational waters, drinking water sources, shellfish harvesting waters, and biosolids), MRA
resources for food safety risk were also consulted in the development of this document. This MRA
Tools document is designed to be flexible and amenable to modification where a user has particular
requirements that might not be precisely covered in the text. It is intended to provide a modular
toolbox with a broad scope and it is expected that those modular aspects that are relevant to the
MRA being conducted can be used as deemed appropriate. This document is designed for use by
individuals with technical expertise (e.g., microbiologists, risk assessment modelers, public health
practitioners) and risk managers.
This MRA Tools document is purposely limited in its discussion and evaluation of state-of-the-art
research that is ongoing within the field of MRA, and rather focuses on mature and established
practices. In some places, literature is cited for readers that are interested in exploring topics in
MRA's future development. Moreover, this document does not provide instructions for conducting
statistical or modeling analysis. This document provides a systematic approach for framing
information to be considered, and information about conducting and documenting risk assessment.
It is compatible with other well-known frameworks and information to help ensure risk
characterization that is helpful and relevant for the decision-makers.
This MRA Tools document focuses on MRA as it fits into the more comprehensive framework of
risk analysis—an overarching term used to describe the interaction of risk assessment, risk
management, and risk communication (CAC, 2004; Figure 1). Another complementary framework
is the World Health Organization (WHO) Water Quality Framework (Figure 2), which provides
for broad public health-based approaches for countries to assess options for meeting public health
goals.
Although the NAS, NRC developed frameworks for risk assessment, those frameworks have been
primarily focused on chemical risks (NRC, 1983, 2009). Microbial interactions with hosts and the
environment are different from chemical interactions, and thus microbial risk assessment differs
from chemical risk assessment (see Section 1.6). This MRA Tools document was developed to
accommodate those differences between chemicals and microbes, while maintaining compatibility
with the overall NRC frameworks.
This document focuses primarily on risk assessment and only addresses risk management and risk
communication activities to the extent that they overlap with risk assessment. However, it is
important to note that risk assessment is not an effective process unless risk management and risk
communication activities are also comprehensively pursued. The interagency Microbial Risk
Assessment Guideline includes more details on risk management and risk communication (U.S.
EPA/USD A, 2012).
1
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Risk Analysis
Risk
Management
Risk
Communication
Risk
Assessment
Figure 1. Risk Analysis
Acceptable Risk
Risk
Management
Assess
Environmental
Exposure
Assessment
of Risk
Health
Targets
Public
Health
Status
Figure 2. WHO Water Quality Framework (Source: Adapted from WHO, 2001)
One topic not addressed in this document is how EPA sets priorities for MRA or selects which
MRAs to conduct. An example of one approach to priority setting is EPA's drinking water
Contaminant Candidate List Classification Process.2 The U.S. Food and Drug Administration
(FDA) also has formal guidelines for determining how to set priorities for initiating MRAs (FDA,
2002).
2 http://www.lJ.S. EPA.gov/ogwdwOOO/ndwacsum.html#ccl cp
2
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Microbial risk assessments can be initiated for a variety of reasons, including but not limited to
the following:
• to assess the potential for human risk associated with exposure to a known pathogen;
• to determine critical points for control, such as watershed protection measures;
• to determine specific treatment processes to reduce, remove, or inactivate various
pathogens;
• to predict the consequences of various management options for reducing risk;
• to identify and prioritize research needs; and
• to assist in epidemiological investigations.
Individual risk assessments for specific situations can differ significantly with respect to the
questions that are addressed, the information required to address those questions, and the nature of
data gaps. MRAs can be conducted to characterize the risk associated with a particular combination
of a pathogen and route of exposure, "in reverse" to compute a density of a specific pathogen that
would correspond to a pre-specified level of risk, or to evaluate the relative ranking of
pathogen/exposure combinations. Examples of each of these approaches are referred to throughout
this document.
There are some long-term goals in the MRA field that cannot yet be adequately addressed by the
tools and methods that are currently available. As the field advances, this document can be
expanded or modified to include new tools once they have been tested and gain general acceptance.
In addition, some MRA goals have ambitious data requirements that cannot be adequately
addressed at this time. Development of methods to advance MRA capabilities is a general goal of
the MRA field. Some examples of possible long-term development goals for MRA are presented
in Appendix A.
1.2. Development of the MRA Tools Document
This MRA Tools document is comprised of a combination of concepts from numerous published
risk assessment frameworks and workshop proceedings. Although many of these frameworks and
proceedings were originally developed for other applications such as food safety, there are many
principles that also apply to water-related risk assessments. Some of the resources employed to
develop this document are briefly summarized below and include the following:
• NAS, NRC, Risk Assessment in the Federal Government: Managing the Process (NRC,
1983)
• NAS, NRC, Science and Decisions: Advancing Risk Assessment (NRC, 2009)
• EPA and U.S. Department of Agriculture (USD A), Microbial Risk Assessment Guideline:
Pathogenic Microorganisms with Focus on Food and Water (U.S. EPA/US DA, 2012)
• EPA Office of Water/ILSI RSI Revised Framework for Microbial Risk Assessment (ILSI,
2000)
• EPA Guidelines for Ecological Risk Assessment (U.S. EPA, 1998b)
• EPA MRA Workshop s:
o Microbiological Risk Assessment Framework Workshop: Tools, Methods, and
Approaches (August 2002) (U.S. EPA, 2002c)
3
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
o Microbiological Risk Assessment Framework: Problem Formulation Workshop
(July 2003) (U.S. EPA, 2003c)
• EPA Thesaurus of Terms Used in Microbiological Risk Assessment (U.S. EPA, 2007a)
• EPA Office of the Science Advisor Staff Paper Risk Assessment Principles and Practices
(U.S. EPA, 2004d)
• EPA Framework for Human Health Risk Assessment to Inform Decision-making,
External Review Draft (U.S. EPA, 2012a)
• Codex Alimentarius Commission, Principles and Guidelines for the Conduct of
Microbiological Risk Assessment (CAC, 1999) and Principles and Guidelines for the
Conduct of Microbial Risk Management (CAC, 2007)
• Food and Agricultural Organization (FAO) and World Health Organization (WHO)
Microbiological Risk Assessment Series, No. 3, Hazard Characterization for Pathogens
in Food and Water Guidelines (F AO/WHO, 2003).
• WHO Water Quality: Guidelines, Standards and Health, Assessment of Risk and Risk
Management for Water-Related Infectious Disease (WHO, 2001)
In 1983, in response to a request by the U.S. Congress, NAS, NRC (hereafter referred to as NRC
framework) developed a framework that addressed primarily chemicals (NRC, 1983). It was
developed by a committee of volunteer experts drawn from academia, government, and industry
that was charged to conduct a study of institutional approaches to risk assessment within the federal
government. The NRC committee's report underwent extensive peer review and continues to be
widely cited and used in the chemical risk assessment community. The framework has also served
as a template for the development of numerous subsequent risk assessments and risk assessment
frameworks. In 2009, the NRC Committee on Improving Risk Analysis Approaches Used by the
EPA issued a report that further developed the original 1983 framework by expanding on problem
formulation and risk-based decision-making (NRC, 2009). The 2009 NRC framework has the
following three phases:
• Phase I: Problem Formulation and Scoping
• Phase II: Planning and Conduct of Risk Assessment
o Stage 1: Planning
o Stage 2: Risk Assessment (per the original 1983 NRC framework)
o Stage 3: Confirmation of Utility
• Phase III: Risk Management
The 2009 NRC framework also recommends formal provisions for internal and external
stakeholder involvement at all stages of risk assessment. In this MRA Tools document, Chapter 2
corresponds to Phase I and Phase II (Stage 1) of the 2009 NRC framework. Chapters 3, 4, and 5
correspond to Phase II (Stage 2) of the 2009 NRC framework. Chapter 5 includes some of the
concepts from Phase II (Stage 3) and Phase III of the 2009 NRC framework.
The International Life Sciences Institute's Risk Science Institute (ILSI-RSI) and the EPA Office
of Water developed a conceptual framework for assessing the risks of human disease following
exposure to waterborne pathogens—EPA-ILSI Framework for Microbial Risk Assessment
(hereafter called the EPA-ILSI Framework) (ILSI, 1996, 2000). The EPA-ILSI Framework
follows the general structure of the EPA Guidelines for Ecological Risk Assessment (U.S. EPA,
4
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
1998b). This MRA Tools document is based on and refines earlier frameworks. It follows the basic
structure of EPA's Framework for Human Health Risk Assessment to Inform Decision-making
(hereafter called EPA's HHRA Framework) (U.S. EPA, 2012a). Elements of several international
frameworks for food and water microbial risk assessment have been integrated to increase the
harmonization with other approaches.
The EPA-ILSI Framework describes a generic approach to identifying scientific information that
should be considered in attempts to quantitatively or qualitatively assess the human health risks
associated with exposure to infectious agents in water. The process to develop the EPA-ILSI
Framework included three workshops held in 1995, 1996, and 1999; deliberations by a 30-member
working group of scientists from academia, industry, and government; and two case study
quantitative risk assessments (Soller et al., 1999; Teunis and Havelaar, 1999) to test the utility and
flexibility of the framework. Notably, the participants in the 1999 workshop suggested that the
framework could be further revised to include a number of additional capabilities. Two specific
suggestions that are integrated into this document are the inclusion of specific information on the
various types of mathematical models that have been used in MRAs and methods to address time-
dependent aspects of infectious disease and immunity (dynamic modeling).
To support the continued enhancement of the EPA-ILSI Framework, EPA convened two
workshops, Microbiological Risk Assessment Framework Workshop Tools, Methods, and
Approaches in 2002 (hereafter referred to as the tools workshop) (U.S. EPA, 2002c); and
Microbiological Risk Assessment Framework: Problem Formulation Workshop in 2003 (hereafter
referred to as the problem formulation workshop) (U.S. EPA, 2003c). The tools workshop
identified available analytical tools, methods, and approaches that could improve qualitative and
quantitative microbiological risk assessments conducted under the existing EPA-ILSI Framework.
Another important objective was to identify major issues that limit the successful application of
the existing framework for conducting risk assessments.
The problem formulation workshop further developed the problem formulation stage of the EPA-
ILSI Framework. Results from that workshop included elaboration of the roles of risk assessors,
risk managers, risk communicators, and stakeholders during the problem formulation stage;
guidance for development of conceptual models; and modification of the process diagram (flow
chart) for risk assessment.
One important conclusion of both workshops was that the EPA-ILSI Framework is applicable to
addressing a wide variety of public health issues related to water quality and food safety. In
addition to pathogen-specific analysis, risk assessments could be used to evaluate regulatory
actions, evaluate groups of pathogens (e.g., viruses), and evaluate surrogates (e.g., turbidity in
drinking water). However, the EPA-ILSI Framework does not specifically discuss these types of
risk assessments and did not provide examples. The discussion of problem formulation in the
problem formulation workshop overlapped with EPA's Science Policy Council and Office of the
Science Advisor's definition of planning and scoping (U.S. EPA, 2000b, 2002b, 2004d). Although
the problem formulation workshop participants envisioned problem formulation as encompassing
many of the aspects of planning and scoping, for the purpose of this document, problem
formulation has been defined as part of the overall planning and scoping to be consistent with other
EPA risk assessment documents such as EPA's HHRA Framework (U.S. EPA, 2012a).
5
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
A Thesaurus of Terms Used in Microbiological Risk Assessment (hereafter referred to as the
Thesaurus) was developed in parallel to this document (U.S. EPA, 2007a). The Thesaurus
compiles definitions of terms from EPA sources, other U.S. Federal agencies, international
guidelines, foreign governments, and several nongovernmental organizations concerned with risk
assessment. Definitions in the Thesaurus were evaluated for their potential to cause confusion,
such as when the same term has differing definitions depending on its application, or when similar
concepts are known by different names in different disciplines. Refer to the Thesaurus for detailed
definitions of specific microbial risk concepts.
Community-based cumulative risk assessment is of growing interest to EPA. For example, EPA's
Workshop on Research Needs for Community-Based Risk Assessments (October 2007) described
community-based risk assessment as follows:
Community-based risk assessment is a model that addresses the multiple chemical and non-
chemical stressors faced by a community, while incorporating a community-based
participatory research framework and a transparent process to instill confidence and trust
among community members. It has become clear that cumulative risk assessments should
include both chemical and non-chemical stressors, exposures from multiple routes, and
population factors that differentially affect exposure or toxicity, and in some cases, resiliency
to environmental contaminants.
Although the concepts and factors presented in this MRA Tools document could be used to
consider microbial risks in the context of community-based cumulative risk assessment, at present
there are no examples of cumulative MRA in the literature.
EPA's Office of the Science Advisor's Staff Paper on Risk Assessment Principles and Practices
reviews EPA's chemical risk assessment practices across the agency (U.S. EPA, 2004d). It
discusses general risk assessment topics such as conservatism, default assumptions, uncertainty,
variability, and information gaps. The discussion of general topics is also applicable to MRA.
The Codex Alimentarius Commission (Codex or CAC) was created by the United Nations/FAO
and WHO to develop food standards, guidelines, and related texts such as codes of practice under
the Joint FAO/WHO Food Standards Programme. Codex follows an eight step Elaboration
Procedure for drafting, amending, and adopting standards and guidelines. In the final step of the
elaboration procedure, documents are adopted by the Commission and sent to the governments of
the participating countries for acceptance. Codex adopted Principles and Guidelines for the
Conduct of Microbiological Risk Assessment (hereafter referred to as Codex MRA Guidelines)
(CAC, 1999) and a companion document, Principles and Guidelines for the Conduct of Microbial
Risk Management (hereafter referred to as Codex Microbial Risk Management [MRM]
Guidelines) (CAC, 2007). Although mainly applicable to food safety risk assessment, the Codex
MRA Guidelines, and MRM Guidelines, contain many principles that also apply to water safety
risk assessments because important waterborne pathogens can also contaminate foods and food
products.
The FAO/WHO Microbiological Risk Assessment Series, No. 3, Hazard Characterization for
Pathogens in Food and Water Guidelines (FAO/WHO, 2003) and the FAO/WHO Microbiological
6
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Risk Assessment Series, No. 17 Risk Characterization of Microbiological Hazards in Food:
Guidelines (FAO/WHO, 2009), are overall frameworks that include summaries of strengths and
limitations of outbreak investigations, surveillance and annual health statistics, volunteer feeding
studies, biomarkers, intervention studies, animal studies, in vitro studies, and expert elicitation.
Elements adapted from the EPA-ILSI Framework are discussed in detail.
The WHO document Water Quality: Guidelines, Standards and Health, Assessment of Risk and
Risk Management for Water-Related Infectious Disease (hereafter referred to as WHO Water
Quality Guidelines) (WHO, 2001), is intended to harmonize3 the process of development of
guidelines and standards for drinking water, wastewater used in agriculture and aquaculture, and
recreational water environments. The series of reviews in the WHO Water Quality Guidelines
address the principle issues of concern linking water and health to the establishment and
implementation of effective, affordable, and efficient guidelines and standards.
Concurrent to the development of this MRA tools document EPA and USDA developed and
published the interagency Microbial Risk Assessment Guideline: Pathogenic Microorganisms with
Focus on Food and Water (U.S. EPA/USDA, 2012). The interagency guideline is a useful
resource, in particular for chapters on risk management and risk communication, which are beyond
the scope of this document.
1.3. MRA Framework
This MRA Tools document considers the factors identified in the EPA-ILSI Framework for MRA
(ILSI, 2000). The basic framework illustrated in Figure 3 is the same as EPA's HHRA Framework
(U.S. EPA, 2012a) and is based on the NRC's Science and Decisions: Advancing Risk Assessment
(NRC, 2009). MRA practitioners and managers often desire flexibility in the development of
MRAs, so the other frameworks listed in Section 1.2 may also have helpful information and should
be considered compatible with this MRA tools document.
As shown in Figure 3, the initial stage in conducting risk assessment focuses on carefully
describing the task to be completed; it includes the planning and scoping and problem formulation
components. The risk assessment phase includes developing the exposure and effects
characterizations and integrating those results for presentation as part of the risk characterization.
A key aspect of the HHRA Framework, "fit for purpose," is consideration of the usefulness of the
assessment for its intended purpose, to ensure that the assessment produced is suitable and useful
for informing the needed decisions. Attention to this concept is intended to assure, through focused
planning and problem formulation and periodic reconfirmation during the process, that the
informational needs of the risk managers will be met by the information being generated by the
assessment. Rather than a separate step or final check in the process once the risk assessment is
completed, an emphasis on the utility of the risk assessment occurs throughout the process. This
begins with planning and scoping and includes evaluating the applicability of the risk assessment
for informing risk management decisions; these evaluations may take place in several points of the
iterative risk assessment process.
3 In international law, harmonization refers to the process by which different states adopt the same laws (Stone,
2006). In this context it refers to the adoption of similar protocols.
7
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Public,
Stakeholder
and
Community
Involvement
<—1+
<-4>
i—~
Planning and Scoping
and Problem Formulation
Conceptual
Analysis
Model
Plan
Risk Assessment
Exposure
Assessment
Effects Assessment
Hazard Identification
Dose Response
Risk Characterization
Informing Decisions
Figure 3. Framework for Human Health Risk Assessment to Inform Decision Making
(Source: U.S. EPA, 2012a)
Table 1 lists factors that can be considered during the development and conduct of a microbial risk
assessment. These factors are discussed in more depth throughout this document and are presented
here as an overview. Not all factors will be appropriate or relevant for all MRAs and the list should
not limit the addition of other factors to the risk assessment. It is helpful to provide justification
when a particular factor is excluded from an MRA. Note that the factors listed in Table 1 are also
referred to as "elements" or "components" in this document and can be represented by parameters
in an MRA model or can be incorporated into the risk assessment in some other fashion
(qualitatively). A brief summary of other risk frameworks that are consistent with this MRA Tools
document is provided in Appendix B.
8
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 1. Elements of Microbial Risk Assessment (Source: Adapted from ILSI, 2000; U.S.
EPA, 2003c, 2003f, and 2012a)
Chapter Elements/Factors
Define the concern driving the RA
Define the purpose, and objectives of the RA
Understand the history and context within the Agency
(5
Define the scope of analysis
0)
>
O
Agree on questions the RA should answer
c
Develop the conceptual model
O
+-»
rc
3
E
i_
0
Develop the analysis plan (i.e., operational plan) agree on participants, roles, responsibilities,
resources available, schedule, and deliverable products
Agree on analytical approaches
E
0)
-o
Survival and multiplication
n
0
(5
N
Resistance to control or treatment processes
Q_
¦O
I
0)
0
+-»
(5
N
"i_
0)
+-»
o
(5
Ecology and epidemiological triad
c
(5
U)
(5
0)
>
b
(A
3
0
Virulence and pathogenicity of microorganism
c
"q.
Pathologic characteristics/disease caused including host specificity
o
cn
0
+-»
cs
£
o
Infection mechanisms/route of infection/portals of entry
-o
c
(5
3
F
o
£
Potential for secondary spread
O)
c
i_
0
LL
Taxonomy/strain variation
c
(5
E
0)
C
Demographics (e.g. age, size of population)
Q_
.q
0
l_
Q_
0
13
0
Immune status
N
0)
If
Concurrent illness/medical treatment
o
(5
J2 o
3 O
Q. C
Genetic background
£
o
O 0
Q. O
Pregnancy
+-»
>
0
2
Nutritional status
Social/behavioral traits
Temporal distribution/frequency
+-»
0)
Density in environmental media
0
E
O
c
2
Spatial distribution (clumping, aggregation, particles, clustering)
8! 5
W Q.
< (5
3
O
Niche (ecology, non-human reservoirs, zoonotic potential)
o
O
Survival, persistence, amplification
0)
O)
Seasonality
0
Q.
X
£
+-»
(5
Meteorological and climatic events
LU
Q_
Presence of treatment or control processes
Indicators and surrogates and relationships
9
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 1. Elements of Microbial Risk Assessment (Source: Adapted from ILSI, 2000; U.S.
EPA, 2003c, 2003f, and 2012a) (continued)
Chapter Elements/Factors
Exposure Assessment
Chapter (continued)
Exposure Assessment
Identification of media (water and shellfish for Ambient Water Quality Criteria [AWQC])
Routes of exposure
Units of exposure (magnitude [e.g., # of pathogen units, per volume of ingested water],
duration [e.g., per day or per event], frequency [e.g., days or events compounded over a year
or a life time]
Temporal nature of exposure (whether single or multiple exposures)
Spatial nature of exposure
Behavior of exposed population
Effects Assessment Chapter
Health
Effects
Duration of illness
Severity of illness
Morbidity, mortality, sequelae of illness (including acute and chronic effects)
Extent or amount of secondary transmission
Dose-Response
Statistical model(s) to analyze or quantify dose-response relationships
Human and/or animal dose-response data
Source and preparation of challenge material or inoculum in the dose-response study
Outbreak or intervention data
Route of exposure or administration used in dose-response study
Equivalence of methods used (including organism type, strain, and method units) for
occurrence data and dose-response study
Characteristics of the exposed population in dose-response study (age, immune status, etc.)
Infection or disease endpoint for the dose-response relationship (e.g., pathogen shedding,
serological response, symptoms)
Risk Characterization Chapter
Risk Characterization
Evaluate health consequences of exposure scenario (risk description [event])
Estimate the magnitude of the risk
Conduct sensitivity analysis (evaluate most important variables and information needs)
Summarize key issues and conclusions
Characterize uncertainty/variability/confidence in estimates
Address items in problem formulation
Ensure transparency, clarity, consistency, and reasonableness (TCCR)
Summarize assumptions including explanation of use of default values and methods
Describe overall strengths and limitations
Discuss how a specific risk and its context compares with similar risks
The complexity of issues surrounding the design and implementation of a microbial risk
assessment requires the use of a flexible toolbox approach, in which a variety of readily available
tools, methods, resources, and approaches (collectively called tools) are identified for
consideration and use at different phases of the assessment. The use of a toolbox approach is
10
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
integral to using this document, although this document does not provide, nor should be inferred
to provide, a comprehensive list of tools available for use in microbial risk assessment.
1.4. General MRA Concepts
This document includes various concepts and processes that broadly apply to MRAs. The
following is a brief overview of some general MRA concepts.
Iterative nature of risk assessment: The risk assessment process is not linear, but flexible and
dynamic (ILSI, 2000; U.S. EPA, 2012a). During any phase of the MRA process the other phases
should be revisited and refined as new information and insights become available.
Transparency, clarity, consistency, and reasonableness (TCCR): Risk assessments should
fulfill specific TCCR criteria (U.S. EPA, 2000b, 2012a). The TCCR criteria are summarized
below.
• Transparency: For risk assessment to be transparent, methods and assumptions should be
clearly stated and understandable to the intended audience, whether it consists of informed
analysts in the field, risk managers, or the general public.
• Clarity refers to the manner in which the risk assessment is presented, such as writing style
and the use of graphic aids.
• Consistency provides a context for the reader, such as whether the conclusions are in
harmony with relevant Agency policy, procedural guidance, and scientific rationales, and
if not, how and why the conclusions differ.
• Reasonableness addresses the extent to which professional judgments and assumptions
are well founded, as confirmed by expert peer review. Risk characterizations should be
consistent in general format, but recognize the unique characteristics of each specific
situation.
Data quality: Data used in an EPA risk assessment must be consistent with EPA's Information
Quality Guidelines (U.S. EPA, 2002a). These Guidelines build upon ongoing efforts to improve
the quality of the data and analyses that support EPA's various policy and regulatory decisions and
programs. They create a mechanism that enables the public to seek and obtain, as appropriate,
correction of information disseminated by EPA.
Data representation: In assessing risk associated with infectious disease hazard exposures, it is
usually necessary to estimate a number of parameters (quantities) in the risk models (equations)
that yield numerical estimates of the probability of infection or illness. Depending on the data
quality, different statistical measures (e.g., mean, median, specific percentile values) of these
parameters might be appropriate.
Data variability and uncertainty: Uncertainty and variability can affect the quality and
interpretation of MRA model results. Understanding, accounting for, and communicating the
effects of these factors is critical in an MRA. The EPA Exposure Factors Handbook (U.S. EPA,
1997a, 2011) indicates that uncertainty represents a lack of knowledge about factors affecting
exposure or risk, whereas variability arises from true heterogeneity across people, places, or time.
11
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Model validation: Model validation and verification in risk assessment are general terms that are
sometimes used to refer to rigorous data driven evaluation of models. However, these terms are
often used interchangeably to refer to a less rigorous "reality check" that could have poorly defined
validation criteria. Because validation implies different criteria in different situations, any
discussion of validation should refer to how the validation was performed so that readers may
properly understand the degree of rigor that the validation effort entailed. For example, one method
that has been used to validate risk assessment findings is to compare the outputs to epidemiological
data to determine whether the risk estimates are consistent with that which has been observed.
Risk assessment team: Risk assessment teams are multidisciplinary and may include individuals
with expertise in diverse disciplines, including economics; law; engineering; the sciences (such as
microbiology, epidemiology, toxicology, chemistry, and medicine); statistics; mathematics;
software programming; website design; and technical writing. Although individuals may have
overlapping roles, it is important that conflicts of interest between risk assessors and risk managers
be avoided to maintain the scientific integrity of the process and stakeholder confidence. Risk
assessment and risk management roles for risk assessment team members should be clearly
defined. Note that in Figure 1 that the activities of risk assessment, risk management, and risk
communication overlap. The same person may have multiple roles. In the Federal government,
each agency or office will have unique considerations regarding the composition and organization
of the team. The principles outlined in this MRA Tools document should be broadly applicable for
a wide variety of organizational compositions.
Stakeholders: The term "stakeholders" refers to people and organizations that can shape the
process or will be (or perceive themselves to be) affected by the risk assessment. Stakeholders
should be involved in the Planning and Scoping in a meaningful way. At a minimum, they should
be informed about the risk assessment problem, how it is to be addressed, and have an opportunity
to provide comments. When stakeholders are directly affected by the proposed assessment,
stakeholder comments should be sought to help team members better understand and define the
problem. Stakeholders should also be informed periodically of any changes in the problem
formulation.
Peer review: The role of peer review is to enhance the quality and credibility of EPA decisions by
ensuring that the scientific and technical work products underlying these decisions receive
appropriate levels of peer review by independent scientific and technical experts. EPA's Peer
Review Handbook provides guidance on conduct of peer review (U.S. EPA, 2000a, 2006e, 2012b).
1.5. Microbial Risk Assessment for Decision-Making
Risk assessment is used by governments worldwide for supporting decision-making. At EPA,
ecological and human health risk assessments are used to support many types of management
actions, including the regulation of hazardous waste sites, industrial chemicals, and pesticides, or
the management of watersheds or other ecosystems affected by multiple nonchemical and
chemical stressors (U.S. EPA, 1998b, 2012a; NRC, 2009). This MRA Tools document is
harmonized with EPA's Guidelines for Ecological Risk Assessment and EPA's Framework for
Human Health Risk Assessment to Inform Decision-making (U.S. EPA, 1998b; 2012a).
12
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
MRA in the Federal government is always conducted in the context of supporting decision-
making. Each step in risk assessment is planned and conducted within the context of the risk
management issue. Risk assessment is an iterative process so that risk managers and risk assessors
can work together to craft a risk assessment that answers the questions that are important to
managers. In many cases the results of the first iterations of the risk assessment will inspire
additional risk management questions that can be incorporated into later iterations of the risk
assessment. Communication between risk assessors and risk managers is crucial for designing a
risk assessment that fulfills the risk manager's needs and provides clarity for the risk managers
regarding the uncertainties and caveats associated with the risk assessment.
The wide use and important advantages of risk assessments do not mean they are the sole
determinants of management decisions; risk managers consider many factors. For example,
decisions may be informed by a range of factors, evidence, and policy choices, such as the
following (U.S. EPA, 2012a):
• Laws and Regulatory Requirements—legal mandates, flexibility and constraints.
• Economic Factors—costs, benefits and impacts of potential actions.
• Sustainability—life cycle, multimedia and long-term impacts.
• Technological Factors—feasibility, impact and range of risk management options.
• Political Factors—interactions with different branches and levels of government and the
citizens that they represent.
• Public and Social Factors—susceptible population groups, nonchemical stressors and
cumulative risk assessment considerations.
Reducing risk to the lowest level may be too expensive or not technically feasible. Thus, although
risk assessments provide critical information to risk managers, they are only part of the
environmental decision-making process (U.S. EPA, 1998b).
1.6. Factors Unique to Microbial Risk Assessment as Compared to
Chemical Risk Assessment
Chemical risk assessment methods were examined for their applicability to microbial risk
assessment by an EPA Office of Water workgroup. Many of the concepts developed for chemical
risk assessments have parallels in MRA, but additional features have been developed to account
for the differences between chemicals and microbes. Microbial risk assessments from the early
1990s identified several areas where chemicals and microorganisms differ, as noted in the sections
that follow.
1.6.1. Microbial Growth and Death
Pathogens increase and decreases in number in the environment and in a host, and are variably
affected by environmental and treatment factors. Different species, and even different strains
within a pathogenic species, grow and die in unique patterns. In contrast, although chemicals can
bioaccumulate and bioconcentrate, they are not known to multiply in the environment or in hosts.
13
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Both chemicals and pathogens can decrease due to environmental factors, chemicals can be
transformed or degrade, and pathogens can die. Not all methods used to detect and quantify
microbes can distinguish between living and dead organisms; therefore, the assay method might
affect data analysis when combining or comparing studies. A further complication is that several
species of bacteria, including frank pathogens (e.g., Vibrio spp.), have been found to exist in a state
called "viable but non-culturable" (VBNC). This means that, although unable to multiply on agar-
medium culture plates or grow in liquid media, such cells remain functional and metabolically
active (NRC, 2004). Whether pathogens in the VBNC state are infectious has not been
conclusively determined (Bogosian and Bourneuf, 2001). In contrast, chemical quantification
methods are generally more reproducible and able to reflect the "active" concentration of toxic
agents. Microbial toxins can remain after the organism dies, and some enterotoxins are heat stable
and resistant to degradation. These toxins can cause many of the symptoms of gastrointestinal (GI)
tract illness.
1.6.2. Detection Methodologies
Generally, methods for detecting chemical pollutants are sufficiently sensitive to detect and
quantify concentrations well below the levels that are known to have human health effects. This is
not necessarily the case for pathogens. Theoretically, a single pathogenic organism can cause
infection (and lead to illness). Analytical methods for detecting low levels of pathogens (e.g., one
organism in 2 liters [L] of water) are not available in all cases. Although fecal indicator bacteria
are useful for detecting fecal contamination, indicator bacteria do not necessarily correlate with
the presence of human pathogens or public health risk (NRC, 2004). Therefore this document
focuses on estimating risk from exposure to pathogens. Microbes are subject to environmental
matrix effects that can cause uneven distribution that can result in consecutive measurements that
differ significantly. Matrix effects can also affect the precision and accuracy of the analytical
methods used to detect and quantify microbes in water.
As noted above, microorganisms in a VBNC state are also a concern for interpretation of
enumeration methods. The analytical methods are probably the biggest challenge and represent the
largest source of difference between chemical and microbial risk assessments. The microbial
methods include microscopic techniques that do not rely on the viability of the microbe (e.g.,
protozoan pathogens Cryptosporidium and Giardia). The approach to establishing minimum limits
of detection and practical quantification limits for microbial methods is unlike the approach taken
for analytical methods used to enumerate chemical concentrations. While the assumption of one
organism per volume of water sampled as a method detection limit may work for some microbial
assays, it is not valid for all assays (AWW A, 2006). For example, the highly variable recovery
rates for Giardia and Cryptosporidium cysts and oocysts, respectively, may be affected by the
amount of processing the sample goes through before the enumeration step. In turn, this affects
the reliability and reproducibility with which one oocyst or cyst can be enumerated in a sample
volume. Poor reproducibility contributes to increased uncertainty as the density approaches the
minimum detection limit. For these reasons, enumeration methods for microbes introduce a
sufficiently high level of uncertainty that the details of those methods need to be discussed in the
context of their effect on the risk assessment.
14
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
1.6.3. Genetic Diversity of Pathogens
Microorganisms are genetically diverse and allelic ratios in a population can change significantly
within a few generations. In addition, microbial genomes can evolve quickly (within days or
weeks) through mutation or horizontal gene transfer. Strains of the same species (e.g.,
Cryptosporidium parvum) can have multiple genotypes, potentially with different virulences for
human hosts (Morgan et al., 1999; Xiao et al., 2000). Some pathogens (e.g., Helicobacter pylori,
many viruses) behave like quasi-species, which are fluctuating populations of genetically distinct
variants that co-exist within a single host (Boerlijst et al., 1996; Covacci and Rappuoli, 1998).
Microbes can, thus represent a "moving target" because the distribution of strains and virulence
factors can fluctuate rapidly in a given water body (Loewe et al., 2003; NRC, 2004). Variation
found in the environment can also depend on different sources and types of microbial pollution.
In addition microbes can acquire antibiotic resistance, which affects the range of clinical
treatments that are possible and can render normally treatable illnesses life threatening.
1.6.4. Host Immunity and Susceptibility
Human hosts can have different susceptibilities to infection by particular pathogens, and levels of
immunity against different pathogen species and strains may differ widely (i.e., variability among
humans and variability among pathogens). Although body weight, age, and metabolic capacity
differences are considered in the development of chemical criteria, genetic and acquired
differences in susceptibility are not usually considered. Infection and illness due to pathogens are,
in some cases, highly dependent on the immune status of the individual, which can fluctuate based
on the time since the last exposure, presence of concurrent infections (e.g., human
immunodeficiency virus [HIV]), and a number of other factors (e.g., life stages, gender, genetics)
(Balbus et al., 2000; Parkin et al., 2003; Parkin and Balbus, 2000). For some pathogens, previous
exposure may provide additional protection from that pathogen as a result of increased host
immunity (Soller and Eisenberg, 2008).
1.6.5. Dose-Response Range can be Broad
The levels of pathogens required to cause infection and/or disease can vary substantially across
pathogen species. Even within a particular species, those levels can vary by orders of magnitude,
depending on the strain. The possible host responses may encompass asymptomatic infection,
symptomatic infection (illness or disease, including chronic sequelae), and even death.
Quantitative data on the exposed population's immunity and susceptibility to a pathogen and data
on pathogen strain infectivity in human subgroups with differing immunity would allow the
development of dose-response curves that represent a range of possible dose-response
relationships. However, these types of data are not readily available. For example, although human
dose-response data for six isolates of Cryptosporidium are available (e.g., Okhuysen et al., 2002),
the data only include responses from healthy adult volunteers (for ethical reasons).
1.6.6. Secondary Transmission
Microbial infections can be transmitted from an individual to other susceptible individuals, and
even to some animals. With the exception of the mother-fetus relationship, chemicals in tissues of
15
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
exposed individuals are not known to transmit to other individuals.4 For example, in one
investigation that studied person-to-person transmission of infection, the effect of rotavirus
transmission within households and on the risk of infection from outside of the household was
investigated through analyses of serum pairs (Koopman et al., 1989).5 The researchers found that
17 to 20% of rotavirus infections were acquired in the household and the remainder acquired in
the community. Some microbes can remain viable for days, weeks, or months, in the environment,
which increases the potential for transmission. For some pathogens, humans can become
asymptomatic chronic carriers and thus can infect others and contaminate food and water sources
without displaying symptoms themselves for prolonged periods.
1.6.7. Heterogeneous Spatial and Temporal Distribution
Pathogens are typically heterogeneous in environmental matrices. Whereas most soluble
chemicals diffuse evenly in water matrices, pathogens may clump or may be embedded in or
attached to organic and inorganic particulate debris, making density determinations difficult.
Although density in pipe scale and biofilms is also a problem for chemical contaminants, some
pathogens can grow and/or be protected in these environments (NRC, 2004). Also, many types of
pathogens occur only episodically in drinking and source waters (and in ambient waters as well)
and typically can be found only during short-lived disease outbreaks (i.e., epidemics) in a
community. Seasonal increases in the environment cause water or wastewater to be contaminated
episodically, through breakdowns in wastewater management or water contamination controls.
Therefore, contamination sources may be different for each contamination event. Seasonal
fluctuations are thought to occur due to fluctuations in factors such as precipitation, temperature,
nutrient availability, human activity, and livestock events (e.g., birthing season). The episodic
nature of contamination makes calculation of relative sources of microbial contamination less
useful than relative source contribution for chemicals.
1.6.8. Zoonotic Potential
Many, but not all pathogens also infect and amplify in animals. There is evidence that these
zoonotic pathogens may change in infectivity, virulence, and the severity of disease caused in
humans depending on their previous host environment. There is also evidence that some of these
host-factor changes can influence subsequent infection cycles in exposed hosts (U.S. EPA, 2009a;
WHO, 2004). There are six key waterborne zoonotic pathogens in the United States, Salmonella,
Campylobacter, pathogenies, coli, Leptospira, Cryptosporidium, and Giardia (U.S. EPA, 2009a).
4 Chemicals that are on exposed individuals' clothing or skin can be transferred to household and other contacts.
5 Serum antibodies, which are specific to different pathogen strains, indicate an immune response in an individual
and are interpreted as an indicator of exposure to the specific pathogen strain for which antibodies are present.
16
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
2. Planning and Scoping and Problem Formulation
This chapter addresses the planning and scoping and problem formulation aspects of MRA. During
planning and scoping the purpose of the risk assessment is defined through a dialogue between
risk assessors, risk managers, risk communicators, and stakeholders. To be consistent with EPA's
Science Policy Council and Office of the Science Advisor's documents on human health risk
assessment, planning and scoping is considered as the broad set of activities necessary for
successfully initiating a risk assessment. The overall planning and scoping considers the risk
assessment within the context of overall agency resources (U.S. EPA, 2000b, 2002b, 2004d,
2012a).
Problem formulation falls within planning and scoping and can continue iteratively throughout the
risk assessment process (U.S. EPA, 2000b, 2002b, 2004d, 2012a). The purpose of the problem
formulation process6 is to develop the scope of the risk assessment, taking into account
management needs, Agency risk assessment policies, risk assessment tool availability, data
constraints, and the nature of the decisions to be supported. At any phase in the risk assessment
process, the problem formulation may be revisited.
For human health risk assessment, EPA considers the overall planning and scoping steps to be as
follows (adapted from U.S. EPA, 2003f, 2003c; also see Table 1):
• defining the concern driving the risk assessment;
• defining the purpose, and objectives of the risk assessment;
• understanding the history and context within the Agency;
• defining the scope of analysis;
• agreeing on questions the risk assessment should answer;
• developing the conceptual model;
• developing the analysis plan (i.e., operational plan) agreeing on participants, roles,
responsibilities, resources available, schedule, and deliverable products; and
• agreeing on analytical approaches.
It is not necessary to rigidly delineate various activities as part of planning and scoping versus
problem formulation. It is sufficient to understand that problem formulation includes discussion
of scientific and science policy choices related to the conduct of risk assessment while planning
and scoping includes problem formulation and the operational, logistical, and budgetary planning
necessary to successfully conduct the risk assessment.
2.1. Introduction to Planning and Scoping and Problem
Formulation
Tasks for problem formulation include describing specific risk management questions,
determining data and resource needs, performing preliminary exposure and health effects
assessments, developing a conceptual model, and defining key assumptions. Forming an
6 Note, Codex refers to this stage as "risk profile."
17
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
operational plan for conducting the risk assessment should also be accomplished during planning
and scoping. If it is determined that a full risk assessment is not needed or is infeasible, information
gleaned from the problem formulation stage can be used as a qualitative risk assessment or even a
semi-quantitative risk assessment, and the process can, in fact, stop after the problem formulation
stage. This stepwise approach can be a means of prioritizing resources and defining the scope of
the overall risk assessment and to determine whether sufficient information is available to conduct
a comprehensive quantitative risk assessment, if in fact, the risk management questions require a
comprehensive assessment. Wooldridge and Schaffner (2008) provide guidance on qualitative risk
assessment.
Identification of the nature of required inputs and outputs is necessary during problem formulation.
Two general risk assessment approaches are consistent with this MRA Tools document. In the first
approach, pathogen occurrence, exposure assessment, and dose-response assessment are combined
to arrive at an estimated risk level. This first approach would be used, for example, to characterize
the risk associated with a specific pathogen through specific route of exposure. In the second
approach, which can be useful for regulatory purposes, dose-response assessment, exposure
assessment, and a target risk level or risk range7 are combined to determine a pathogen density
that would provide a pre-specified level of public health protection. In the first approach, the
estimated risk (e.g., event, daily, or annual risk of infection or illness) is the output; in the second
approach, the pathogen density for a given exposure scenario is the output. There are also other
types of risk assessments that may be consistent with this document, including the following:
• Product/Pathogen Pathway Analysis—used mainly for microbial risks in a specific food;
the risk assessment models the temporal/spatial pathway a product follows through
production to consumption;
• Risk Ranking—ranks risks of the same pathogen from multiple sources, or ranks risks of
multiple pathogens from one source; for example see FDA-U.S. Department of Agriculture
(USDA) Listeria risk assessment (FDA/USDA, 2003);
• Risk/Risk Analysis—compares risks between different scenarios, usually management
options); and
• Geographical Introduction Analysis—used to estimate risk of introduction of disease
agents through food animals or animal products (e.g., intentionally as in bioterrorism or
unintentionally) to a region; for example, the risk of bovine spongiform encephalopathy
("mad-cow disease") occurring is U.S. herds due to importation of livestock from other
countries.
All of these types of risk assessments may have different types of outputs and require different
inputs. The information presented in this MRA Tools document should be evaluated within the
context of the scope of a given risk assessment.
The WHO Water Quality Guidelines (WHO, 2001) include methods for risk assessments that have
health targets, water quality targets, or a performance target that includes engineering technology
(including technological approaches for small communities). In this context MRA can be used to
7 "Target risk range" is similar to "appropriate level of protection", which is used in the World Trade Organization
"Agreement on the Application of Sanitary and Phytosanitary Measures" and the Codex MRM Guidelines (CAC,
2007).
18
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
(1) provide estimates of the burden of disease, (2) establish norms and standards such as water
quality, (3) assess the safety of a system against performance standards, and (4) assess health
impacts. The WHO Water Quality Guidelines methodology is similar to the Codex MRM approach
(CAC, 2007) in that it relates public health goals to "food safety objectives," "performance
objectives," "performance criteria," and "microbiological criteria".
The WHO and many countries have adopted Water Safety Plans, which use the concept of Hazard
Analysis and Critical Control Point (HACCP) (Bartram et al., 2009). Seven basic principles are
employed in the development of HACCP plans that meet a stated goal, and include hazard analysis,
critical control point identification, establishing critical limits, monitoring procedures, corrective
actions, verification procedures, and record-keeping and documentation (NACMCF, 1997).
During the problem formulation stage, the above concepts can be discussed in the text of one or
more of the suggested problem formulation components. These components, which are discussed
below, include the statement of concern, statement of purpose, questions the risk assessment
should address, and conceptual model narrative. For example, a risk assessment that estimates the
burden of disease can compare water treatment processes, which is a technology-based
performance perspective.
The problem formulation process diagram is shown in Figure 4. Note that this diagram does not
include specifics about what questions can be asked or how the conceptual model should be built.
However, it does show the types of information that should be collected to determine the feasibility
of conducting a MRA8.
The diagram is roughly chronological. Initially, a concern or set of concerns is identified. Those
concerns can come to the attention of the Agency through various routes. The statement of concern,
statement of purpose, and questions to be considered evolve throughout the problem formulation
stage. These can be integrated into a risk assessment "charge." Risk managers are responsible for
ensuring that appropriate problem formulation documentation is developed so that it is sufficient
for the particular problem at hand. With initial information regarding the scope and questions for
the risk assessment, risk assessors determine the feasibility of carrying out those plans given the
available data, risk assessment tools, and time and resources. A concise conceptual model,
narrative, and analysis plan are developed. A screening-level risk assessment may first be
performed to determine if the risk assessment questions can be addressed without an extensive
formal quantitative risk assessment. In some cases, a screening level risk assessment may be
adequate for decision-making. If a formal quantitative risk assessment is desired and feasible, a
more detailed conceptual model/narrative and analysis plan are developed. The problem
formulation documentation can be used to assist risk managers with policy decisions that are
needed to define the scope of the risk assessment. For example, risk assessors can outline options
for risk managers to consider. Because risk assessment is iterative by nature, aspects considered
during the problem formulation may need to be revisited multiple times as new information and/or
data become available. During problem formulation, the risk assessment options that are
considered, the options that are chosen, and the justification for those decisions, should be carefully
tracked and documented.
8 For the purposes of this document, the term MRA also includes QMRAs.
19
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Problem Formulation
NO
YES
Concerns
Conduct risk assessment
Screening level
risk assessment
Risk assessment
completed
Questions the RA
should answer
Statement of concern
(Problem Statement)
Statement of purpose
(Objectives)
Risk assessment feasibility
preliminary analysis
Brief Conceptual Model,
Narrative, and Analysis Plan
Revise Conceptual Model,
Narrative, and Analysis
Plan
Does the risk assessment
adequately address the
questions?
Risk Management Activities
Figure 4. Enhanced Problem Formulation Process Diagram
(Source: Adapted from U.S. EPA, 2003c)
It should also be noted that risk assessments can be developed in phases. As indicated previously,
a screening level risk assessment may be the initial step that later leads to an enhanced fully
quantitative risk assessment. The complexity of the risk assessment may be incrementally
increased by adding new models or parameters or by more rigorously characterizing parameter
values (e.g., from point estimate values to a statistical distribution, as described in Section 5.3). In
many cases, sensitivity analysis can guide prioritization regarding further data gathering or
refinement of parameter estimates. The iterative nature of the problem formulation process should
allow for further definition and refinement of possible phases of the risk assessment. If multiple
versions of the risk assessment are conducted as a result of this iterative process, the choices for
each version (also referred to as phase) of the risk assessment should be tracked and documented.
2.2. Overall Problem Formulation and Planning and Scoping
During the problem formulation process, the purpose of the risk assessment is defined through a
dialogue between risk assessors, risk managers, risk communicators, and if appropriate—
stakeholders. A valuable aspect of the process is documenting the problem formulation
development. The value is that it provides a written record of the justification for the decisions
regarding the scope, goals, and necessary documentation of the risk assessment. The form of this
documentation can vary depending on the needs of the EPA Office conducting the assessment.
20
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
The range of acceptable forms for this documentation ranges from a formal and stand-alone
problem formulation document to internal notes kept by the project (or work assignment) manager
for the EPA Office conducting the assessment. The final risk assessment report should include all
of the problem formulation information for the risk assessment iterations that are being published.
The concepts and in many cases the language used during development of the problem formulation
documentation can be used in the final risk assessment document. Depending on the form of the
problem formulation documentation, the statement of concern and the statement of purpose can be
included as part of an executive summary of the risk assessment document. The scope, questions
to be addressed, conceptual model, and data not included can be used in the problem formulation
chapter of the risk assessment document. Other planning and scoping documentation can be
summarized in the problem formulation chapter, a planning and scoping chapter or, if desired,
attached as an appendix. The analytical approaches including tools, data inventory, summary of
assumptions, and discussion of recommended factors are reiterated as appropriate in the exposure
and human health chapters of the risk assessment document. The summary of assumptions is
reiterated in the risk characterization chapter (5), which also includes the discussions of variability,
uncertainty, and identified gaps.
For comparison, the risk profile approach developed by Codex for microbiological food risk is
presented in Text Box 1. Note that in the Codex paradigm, the risk profile is similar to and serves
the same purpose as problem formulation described in this MRA Tools document.
2.2.1. Statement of Concern
A concise statement of concern should be developed during problem formulation to convey, in
simple terms, what hazard is being addressed and how it is thought to relate to human health for
an exposure scenario.
2.2.2. Statement of Purpose and Objectives
The purpose and/or objectives of the risk assessment should be stated in a concise paragraph.
Example language for risk assessments performed for the purpose of derivation of Ambient Water
Quality Criteria (AWQC) for a specific pathogen is provided below. Note, the designated use and
the national scope might be different in other cases.
This risk assessment is being performed to support the derivation of Ambient Water
Quality Criteria (AWQC) for [pathogen] under §304(a) of the Clean Water Act (CWA).
These will be nationally recommended AWQC for the protection of the [insert designated
use] designated use. It should not be implied that the AWQC will be protective of other
designated uses, such as [insert designated uses that are excluded]. As with other §304(a)
AWQC, the AWQC for [pathogen] are recommended for adoption by states to be used for
total maximum daily load (TMDL) determination. States also use AWQC to help assess
whether water bodies are threatened or impaired (§305b or §303d CWA) for the specified
designated use.
21
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 1. Microbiological Risk Profile for Food
(Source: Adapted from CAC, 2007)
A risk profile should present, to the extent possible, information on the following:
1. Hazard-food commodity combination(s) of concern:
• Hazard(s) of concern
• Description of the food or food product and/or condition of its use with which problems
(foodborne illness, trade restrictions) due to this hazard have been associated
Occurrence of the hazard in the food chain
2. Description of the public health problem:
• Description of the hazard including key attributes that are the focus of its public health
impact (e.g., virulence characteristics, thermal resistance, antimicrobial resistance)
• Characteristics of the disease, including:
o Susceptible populations
o Annual incidence rate in humans including, if possible, any differences between
age and sex
o Outcome of exposure
o Severity of clinical manifestations (e.g., case-fatality rate, rate of hospitalization)
o Nature and frequency of long-term complications
o Availability and nature of treatment
o Percentage of annual cases attributable to foodborne transmission
• Epidemiology of foodborne disease
o Etiology of foodborne diseases
o Characteristics of the foods implicated
o Food use and handling that influences transmission of the hazard
o Frequency and characteristics of foodborne sporadic cases
o Epidemiological data from outbreak investigations
• Regional, seasonal, and ethnic differences in the incidence of foodborne illness due to
the hazard
• Economic impact or burden of the disease if readily available
o Medical, hospital costs
o Working days lost due to illness, etc.
3. Food production, processing, distribution, and consumption:
• Characteristics of the commodity (commodities) that are involved and that may impact
on risk management
• Description of the farm to table continuum including factors which may impact the
microbiological safety of the commodity (i.e., primary production, processing, transport,
storage, consumer handling practices)
• What is currently known about the risk, how it arises with respect to the commodity's
production, processing, transport and consumer handling practices, and who it affects
• Summary of the extent and effectiveness of current risk management practices
including food safety production/processing control measures, educational programs,
and public health intervention programs (e.g., vaccines)
Identification of additional risk mitigation strategies that can be used to control the
hazard
22
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 1. Microbiological Risk Profile for Food
(Source: Adapted from CAC, 2007)
(continued)
4. Other risk profile elements:
• The extent of international trade of the food commodity
• Existence of regional/international trade agreements and how they may affect the public
health impact with respect to the specific hazard/commodity combination(s)
• Public perceptions of the problem and the risk
• Potential public health and economic consequences of establishing Codex Microbial
Risk Management (MRM) Guidance
5. Risk assessment needs and questions for the risk assessors:
• Initial assessments of the need and benefits to be gained from requesting an MRA, and
the feasibility that such an assessment can be accomplished within the required time
frame
• If a risk assessment is identified as being needed, recommended questions that should
be posed to the risk assessor
6. Available information and major knowledge gaps provide, to the extent possible, information
on the following:
• Existing national MRAs on the hazard/commodity combination(s)
• Other relevant scientific knowledge and data that would facilitate MRM activities
including, if warranted, the conduct of an MRA
• Existing Codex MRM Guidance and related documents (including existing Codes of
Hygienic Practice and/or Codes of Practice)
• International and/or national governmental and/or industry codes of hygienic practice
and related information (e.g., microbiological criteria) that can be considered in
developing a Codex MRM guidance document
• Sources (organizations, individual) of information and scientific expertise that can be
used in developing a final Codex MRM Guidance
• Areas where major absences of information exist that could hamper MRM activities
including, if warranted, the conduct of an MRA
2.2.3. History and Context within the Agency
Previous risk assessments addressing the same or similar hazards should be summarized to provide
context for the current risk assessment. In particular, if previous EPA risk assessments have been
conducted, then the relationship between the current and previous risk assessments should be
summarized. Relevant information for presenting updated MRAs may include new mandates,
policy developments, technical advancements, risk assessment method and tool advancements,
and new or enhanced data sets.
2.2.4. Scope
The scope section of problem formulation outlines the scenarios that the risk assessment will
address. It is often helpful to list several options for answering the questions listed below. Then,
managers and assessors can engage in a dialog to determine which options will be used. The scope
should summarize the following:
1. Which infectious disease hazard is being addressed (e.g., pathogen or pathogen strain[s])?
23
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
2. Which human populations will be included in the risk assessment (e.g., general population
or subpopulations, or geographically defined populations)? Describe which populations are
explicitly included in the risk assessment model, which will be accounted for implicitly,
and which populations may be excluded by the risk assessment model (e.g., most extreme
behaviors).
3. What health outcomes or endpoints are addressed by the risk assessment, including how
the health outcome is measured? Clearly defining the health endpoint is important for
transparency and also focuses the scope of the risk assessment (e.g., infection, disease
symptom/s, mortality).
4. What unit and routes of exposure are relevant and why (magnitude, duration, frequency
units)?
5. For risk assessments designed to derive criteria to set "safe" levels of microorganisms,
what level of protection (target risk or risk range) will be provided by the criteria, and what
is the technical or policy justification for those criteria?
6. What specific exposure scenarios will be modeled? List specific scenarios the risk
managers would like to model (varying the inputs), including desired spatial and temporal
features.
2.2.4.1. Risk Ranges
Currently, EPA does not have an Agency-wide policy for defining acceptable or tolerable levels
of health-based risk associated with pathogenic microorganisms. In fact, there are various
regulatory requirements that influence the degree to which MRAs conducted within the Agency
are driven by risk ranges. For example, the illness rates associated with the RWQC are 32 and 36
GI illnesses per 1000 primary contact recreators (U.S. EPA, 2012c). The current policy for
drinking water standards is to characterize the degree of protection without specific risk-based
targets. In this approach the protective ranges have been influenced mainly from feasibility of
measurement and application of control technology, taking costs into consideration. Furthermore,
semi-quantitative or qualitative MRAs may be necessary under some conditions, and these
assessments may still be meaningful for risk management decisions. For example, it may be
possible to evaluate the relative degree of protection from fecal contamination in drinking water
sources without quantitatively characterizing the risk associated with a specific health endpoint.
Although acceptable risk and target risk are both numeric values that are determined through
science policy decisions they are not necessarily always the same. There may be an expectation
among some stakeholders that a certain target risk range is acceptable. However, given that
different stakeholders may have different ideas about what is acceptable and what is not, it may be
misleading to label a risk range as "acceptable." Risk ranges are values that can be estimated
empirically from data. However, there may not be clear or convincing information to determine if
historically accepted risk ranges are considered acceptable to current stakeholders or not. When
risk ranges are used as a driving force or target for MRA conduct within EPA's Office of Water,
the risk range is defined along four dimensions, as described below:
1. Risk range is for a specified population (population can be defined in a variety of ways,
such as "general," highly exposed, or highly susceptible).
2. Risk range is associated with a defined health endpoint.
24
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3. Risk range covers a defined time span of exposure.
4. Risk range may also be linked to a specific exposure scenario.
Several representative EPA risk range examples employed currently are presented in Text Box 2.
Text Box 2. Information Used to Establish Risk Ranges and Representative Examples of
Risk Ranges Currently Employed by U.S. EPA
Pollutant
Specified
Population
Health Endpoint
Exposure
Duration3
Exposure
(Designated Use)
Carcinogen
General or
children
Cancer
Lifetime
All water uses
Carcinogen
Highly exposed
subgroups
Cancer
Lifetime
All water uses
Indicator bacteria
General
Gl illness
Per
Primary contact
(geometric mean)
recreational
event
recreation
Cryptosporidium
(average densities)
General
Cryptosporidiosis
Daily
Treated drinking
water consumption
a Exposure duration should not be confused with durations relevant for monitoring protocols.
Specific EPA examples:
• Currently, EPA's surface water program has derived AWQC for chemical carcinogens that
generally correspond to lifetime excess cancer risk level of 10 s (1 cancer in a million exposed
individuals); however, AWQC may correspond to a range from 10-7 (1 cancer in 10,000,000
exposed individuals) to 10-5 (1 cancer in 100,000 exposed individuals) (U.S. EPA, 2000c).
• EPA's recreational water quality criteria provide a specific level of public health protection. That
level of protection is a 30-day geometric mean illness level that is less than or equal to 32 or 36
Gl illnesses per thousand recreation events (depending on the fecal indicator bacteria level
selected). Although the duration of each exposure event is daily (or shorter), the duration
associated with the protection of the criteria is a 30-day period.
• Under EPA's drinking water program (Safe Drinking Water Act), the Long Term 2 Enhanced
Surface Water Treatment Rule (LT2) established source water categories (bins) for
Cryptosporidium. The level of public health protection that is provided by LT2 was driven by a
concern for misclassification of binning and cost feasibility for the number of samples that can be
monitored. Thus, the ranges of public health protection provided by LT2 are an outcome of this
risk management approach rather than a pre-specified target risk range (U.S. EPA, 2003a, b,
2006a).
Note that in these examples the health outcomes are different. Chemical exposures result in an endpoint
based on a health effect, whereas microbial exposures can result in infection that can then result in illness.
Not all infections result in illnesses. A morbidity factor can be used to convert infections to illness.
2.2.5. Questions to be Addressed in the Risk Assessment
Microbial risk assessments should be scientifically defensible and relevant to regulatory and public
health concerns. Therefore, the risk assessment should be framed within the context of Agency
policy. The nature and the specifics of the risk management options that need to be evaluated
should be developed during problem formulation so that the risk assessment design can address
any questions that the risk managers want answered. The questions are important for transparency
25
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
and communication between risk managers and risk assessors. Text Box 3 illustrates this point
with three examples of questions that risk managers can ask. There may be two types of questions,
(1) questions the risk assessment should be able to answer, and (2) questions that the risk assessors
need to have answered by the risk managers for appropriate design of the risk assessment. The
second point highlights the need for iterative interaction between risk assessors and risk managers.
Text Box 3. Examples of Risk Management Questions that Could Motivate an MRA
Investigation
• What effects have broad-based health programs or specific actions (e.g., health education
about disinfection) had on (1) the risk of a specific disease (e.g., cryptosporidiosis) and (2)
acute gastrointestinal illness risks among children?
• Which pathogens are associated with human health risks from a specified exposure scenario
(e.g., freshwater recreation activities)?
• Are there reduced risks to public health associated with implementation of specific water
treatment technologies?
2.2.6. Conceptual Model and Narrative
A conceptual model is a graphical representation of the real-world scenario that is being addressed
in a given risk assessment (U.S. EPA, 2002b). There should also be an accompanying narrative
that explains the conceptual model. The scope of the risk assessment should be consistent with the
conceptual model.
The EPA problem formulation workshop (U.S. EPA, 2003c) recommended that multi-tiered
conceptual models be constructed. The first (top) tier of the model should be relatively simple,
representing only the major components of the assessment. Sub-tier conceptual models can build
in more complexity and may require several iterations. The conceptual model should reflect the
uniqueness of the situation that is to be addressed. In some cases, a visual diagram that represents
how the risk assessment is assembled in the actual software code may serve as a useful sub-tier
conceptual model. Although useful for documenting the technical details of the risk assessment,
this type of software code map may not clearly communicate the concepts, so should not be solely
relied upon as a conceptual model. Collectively, the conceptual model(s) and its narrative should
do the following:
• illustrate the risk hypothesis (e.g., provide a flow chart of how risk is thought to occur
within the context of the risk assessment scope);
• outline the tools needed to assess the risk (statistical and other models);
• identify available databases that are needed;
• identify default assumptions;
• show what the risk assessment will or will not be able to do, including whether the
assessment is quantitative or qualitative;
• summarize data gaps and quality of data;
• consider the interactions between agent, host, and environment when evaluating risk;
• define key uncertainties;
26
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• identify nodes in the risk assessment, including a brief description of the node and what
can happen at the node9; and
• identify management actions and places where interventions can take place.
Some of the benefits of developing a conceptual model include the following (from U.S. EPA,
1998b):
• The process of creating a conceptual model is a powerful learning tool to inform the
conduct of the MRA.
• Conceptual models are easily modified as knowledge increases.
• Together with their narrative description, conceptual models highlight what is known and
not known and can be used to plan future work.
• Conceptual models can be a powerful communication tool; they provide an explicit
expression of the assumptions and understanding of a system for others to evaluate.
• Conceptual models provide a framework for prediction and are the template for generating
more risk hypotheses.
It is important that the conceptual model remain free of risk assessment process elements because
trying to reflect the risk assessment process in the conceptual model weakens the conceptual
model's ability to represent real-world scenarios. Therefore, the risk assessment processes
(allocation of Agency resources and deliverable schedule) should be represented separately from
the conceptual model. Because diagrams can be interpreted in different ways by different people,
it is essential that a narrative accompany the conceptual model diagram. Details about elements
should be included in the text and not clutter the diagram.
Although the concepts of problem formulation and risk assessment can be separated and discussed
in a linear manner, the actual process of problem formulation and risk assessment development is
an iterative process. The problem formulation stage should be revisited as the risk assessment takes
shape. Defining the scope of the risk assessment and choosing an appropriate model may require
several iterations, especially if the risk assessment addresses risks or scenarios that have not been
modeled previously.
During problem formulation and developing the first drafts of the risk assessment, it should be
possible to determine how complex the risk assessment model needs to be to address the questions
posed by risk managers. In some cases where the risk assessment questions are simple and limited
in scope, a qualitative risk assessment or a simple risk assessment model may be adequate—even
when robust data sets are available. As a general guideline, models should only be as complex as
they need to be to address the specific risk management questions. A useful model can help the
Agency allocate resources and develop a research agenda as well as provide transparency. A
simplified model may help the public better understand the process and should thus accompany a
very complex model. Within this context, the conceptual model can also be used by the Agency to
consider resource allocation and to develop a research agenda.
9 For example, rainfall, sunlight, and wind speed/direction could be separate nodes in a microbial risk assessment.
Relevance of nodes can be evaluated by performing sensitivity analysis.
27
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Figure 5 presents an example overview or top-tier conceptual model for a risk assessment. In this
example, the model summarizes how waterborne risk from Cryptosporidium is thought to occur.
The conceptual model diagram is a visual representation of the risk hypotheses. The risk
hypotheses are the proposed answers to risk assessment questions about how exposure occurs and
what endpoints are important for the human health hazard. It should be noted that risk hypotheses
are not equivalent to statistical testing of null and alternative hypotheses. However, predictions
generated from risk hypotheses can be tested in a variety of ways, including standard statistical
approaches (U.S. EPA, 1998b). The top tier model should clearly indicate how exposure occurs to
provide a conceptual understanding of the magnitude, duration, and frequency of exposure.
1) Point and non-point sources of
Cryptosporidium in the environment:
• Urban/Livestock run-off
• Wastewater treatment plants
• Wildlife
• Domestic pets
Infected host
shedding oocysts
*
\
6) Probability of
illness based on host
(e.g., immune status)
2) Cryptosporidium
oocysts enter surface
water bodies
I | Environmental Conditions
I | Exposure
U Human Health Response
\
\
\
\
\
~h
Oocyst survival
and die-off is based
on environmental factors^
4b) ingestion of surface
water during recreation
Asymptomatic \
shedding of
oocysts
\
\
\
\
\
5) Probability of infection
based on dose-response
3) Surface water is used as
source for public water supply
(drinking water treatment plant)
Treatment removes
and inactivates
some level of ocysts
4a) Excess viable Cryptosporidium oocysts
reach consumers through treated tap water
Figure 5. Example of an Overview (Top-Tier) Conceptual Model
2.2.7. Planning and Scoping: Analysis (Operational) Plan
The operational plan should include strategies for dealing with data needs, peer review plans, and
any other relevant logistical needs. Information such as lists of relevant experts (for consultation
or data contribution) and literature search strategies can be included. This plan may contain a risk
assessment process diagram that is a graphical representation of the operational plan that helps
explain the logistics of conducting the risk assessment. The plan can also outline proposed phases
for the risk assessment as well. Other essential management activities that are part of planning and
scoping include timelines, planned deliverables (e.g., status briefing memos, draft for peer review,
final draft), team assignments, and possibly budget details. Planning and scoping activities beyond
the core scientific issues of problem formulation may be referred to in the risk assessment if those
details help increase understanding and transparency.
2.3. Analytical Approaches
For the purposes of this MRA document, the concept of parsimony is encouraged; that is, models
should be as simple as possible, but no simpler. Within this context, more complex models should
be considered or used under conditions in which the added complexity may provide sufficient
28
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
additional insight that the additional complexity is warranted (King et al., 2008; Soller and
Eisenberg, 2008). In selecting a MRA model, caution must be taken to ensure that any simplifying
assumptions that are employed are in fact appropriate from an epidemiological perspective. Within
that context, and to the extent possible, MRAs should use epidemiological data as fundamental
components of the assessment and should demand higher quality input data and fewer simplifying
assumptions when seeking increased risk assessment accuracy and precision.
The problem formulation should include information on the following topics:
• Tools. The tools section of the problem formulation should indicate what software will be
used for the risk assessment and may include why the software was chosen. The tools list
should also include mathematical tools such as options for dose-response models. "Tools"
is also an appropriate section to describe the methods that will be used for dealing with
uncertainty. Other types of methodological tools can also be presented.
• Data Inventory. The data inventory should list publications that might be consulted during
the risk assessment process and sources of data that are being considered for the risk
assessment. The list does not need to be comprehensive in the beginning and can be
presented in an appendix of the problem formulation if it is overly long. The data inventory
may be a work in progress throughout the risk assessment. The data inventory can refer to
a literature search strategy that can be presented in an appendix. Literature search strategies
should identify which search engines and databases will be used, keywords, key authors,
language limitations, and timeframe for the search.
• Summary of Assumptions. The summary of assumptions can be organized in different
ways; however, listing assumptions that are related to essential risk assessment factors is a
systematic way to start. How assumptions limit the scope of the risk assessment and
contribute to uncertainty should be explained. The assumptions can be modified and
updated as the risk assessment develops.
• Sources of Variability and Uncertainty. The sources of variability and uncertainty should
be introduced in this section, which should also describe the degree to which variability
and uncertainty is or is not captured in the assessment. The iterative nature of problem
formulation allows this list to be modified as the risk assessment scope is defined.
• Factors and Data not Included and Explanation of Why. There may be information that
is not used or avenues not pursued in the risk assessment. The explanation for not including
that information should be presented, particularly if other related or similar types of risk
assessments have included the information.
• Identified Gaps in the Knowledge Base. Although gaps and data limitations may be noted
throughout problem formulation, they should also be summarized. Gaps can include a lack
of adequate analytical or statistical methods and/or appropriate data and data quality. The
summary of knowledge gaps can be useful for prioritizing future resource allocation (e.g.,
research and development needs) within the context of the results of the risk assessment.
Knowledge gaps and data limitations can also affect the number and type of assumptions
used in the risk assessment.
• Environmental Sampling Strategies and Analysis Methods. Any issues associated
with environmental sampling and analysis should be outlined during problem formulation
so they can be fully considered during risk characterization. For microbial enumeration,
issues may include percent recovery from different sample matrices and the ability of a
29
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
method to determine viability. The accuracy, precision, and biases should be included in
the description of the methods and protocols.
• Evaluation of Management Practices. Depending on the scope of the risk assessment, it
may be appropriate to identify which components in the risk assessment can influence or
be influenced by management actions. It may be desirable to incorporate scenarios in the
risk assessment that include evaluation of best management practices.
2.3.1. Representative Model Forms for MRA Risk Estimation
A variety of model forms can be employed in MRA. Regardless of the form of the model, these
models necessarily include exposure and health effects (dose-response) components. Thus, the
choices made during the problem formulation phase serve as critical components of the risk
assessment. Particular characteristics of each model form allow for the capture of different aspects
of the disease transmission system (U.S. EPA, 2004c). In the following sections, several of the
most commonly employed models are summarized and reviewed. Exclusion from the following
discussion should not preclude use of a particular model form; however, justification for use of a
particular model form should be included in the risk description. An overview of two commonly
employed classes of MRA models is provided in Table 2, and features of the models that may be
used in selecting the appropriate model for a particular application are shown in Figure 6. The
model forms summarized in Table 2 (Static and Dynamic) differ in that dynamic models
specifically account for the temporally changing effects of person-to-person transmission and
immunity in a population, whereas static models treat these innate characteristics as constant
modulators of population risk.
Table 2. Overview and Comparison of Static and Dynamic Risk Assessment Models
Static Risk Assessment Model Dynamic Risk Assessment Model
Number of susceptible individuals is time Number of susceptible individuals varies over time
invariant
Environment-to-person Environment-to-person, person-to-person, and
person-to-environment-to-person
Individual-based perspective Population-based perspective
Typically assumes that the potential for Typically account for the potential for secondary or
secondary transmission of infection or person-to-person transmission of infection or
disease is negligible or scales linearly with disease
the number of infections
Typically assumes that immunity to Exposed individuals may not be susceptible to
infection from microbial agents is negligible infection or disease because they may be infected
already or may be immune from infection due to
prior exposure
Dose-response function is the critical The dose-response function is important; however,
component in a quantitative risk person-to-person transmission and immunity may
assessment also be important
30
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Dynamic Models
Infectious Disease Models
Pros
Higher likelihood of
predictions similar to
outbreak data
Most easily applied
for smaller scale
investigations
Stochastic Dynamic Models
Significant data
requirements
Mathematical
complexity
Research Tool
Cons
Pros
Relatively easy to
execute and explain
Modest data needs
Static Models
Assumes immunity
and secondary
transmission are
relatively unimportant
Limited ability to
include susceptibility
va riability
Cons
Reasonable
computation demands
Appropriate for large-
scale risk analyses
Pros
More accurate
representation of
disease transmission
process
Deterministic Dynamic Models
Greater data
needs than static
models
Cons
Increasing: Complexity, Data Needs, Fidelity
Figure 6. Infectious Disease Model Features for Use in Model Selection
2.3.1.1. Static Models
Some infectious diseases are not readily transmitted from person-to-person but are acquired, to the
best of current knowledge, only by consumption of or contact with contaminated environmental
materials (e.g., Mycobacterium avium complex [MAC] infection from drinking water). In other
cases, although an agent may have the potential to be transmissible, the person-to-person
component is unknown or thought to be negligible.
Understanding the pattern of human infections from such pathogens or exposure scenarios may be
best achieved through the use of static models (parallel to those used for toxicological risk
assessments). The chemical risk assessment-based models are used to estimate risk at an individual
level and typically focus on estimating the probability of infection or disease to an individual as a
result of a single exposure event. With respect to microbial contaminants in water, a fundamental
simplifying assumption of static model-based analysis is that exposure events and infection/disease
are independent; that is, the outcome from one exposure event does not affect a subsequent
exposure, and one individual's outcome has no effect on any other individual's outcome. In most
static models, it is assumed that the population may be categorized into two epidemiological
states—a susceptible state and an infected or diseased state. In these models susceptible individuals
are exposed to the pathogen of interest and move into the infected/diseased state with a probability
that is governed by the dose of pathogen to which they are exposed and the infectivity (dose-
response relationship) of the pathogen.
A static model is appropriate in cases where the probability of infection or illness is not likely to
be substantially impacted by population-level factors such as person-to-person transmission. Such
models can handle complex details about the course of events that lead to exposure and infection
31
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
and can be analyzed by well-established statistical techniques that require fewer assumptions than
do dynamic models (discussed below). Static models are useful for analyzing situations where the
effect of an intervention directed to individuals (e.g., point-of-use remediation) is more important
than the effect on transmission throughout the population; they are not appropriate for measuring
indirect effects at the population level (e.g., the effect of water treatment interventions on risk due
to secondary transmission).
A representative conceptual model for a static MRA model is presented in Figure 7. As can be
seen, individuals who are exposed to pathogens from a specific source, move from a susceptible
state into an infected or diseased state with some probability that is governed by their exposure
and the dose-response relationship for that pathogen. Also note that previous exposures to the
pathogen, interactions with other (potentially infected) individuals, other routes of exposure, and
immune status are not included in this type of model. However, it is possible to use these models
to estimate the cumulative risk of recurring exposures, provided that those recurring exposures are
assumed to be independent (one such example is an estimated annual risk from daily ingestion of
drinking water).
Susceptible
Infected/
Individual
Prob(dose)
Diseased
Pathogen
from
Specific
Source
Figure 7. Static Risk Assessment Conceptual Model
2.3.1.2. Dynamic Models
Risk managers and regulators are often concerned with risk on a societal or population scale. For
a thorough evaluation of risks that are manifest at the population level, MRA methods must explore
the relative importance of secondary transmission and immunity, and thus capture and integrate
the dynamic interplay of hosts, agents, and environments.
Secondary cases (often represented in epidemiological studies by a secondary attack rate)
generally refer to cases that occur among contacts, within the incubation period of the pathogen,
and following exposure to a primary case. In some cases, direct person-to-person transmission
cannot be distinguished from contamination of the immediate environment (e.g., toddlers sharing
toys versus direct physical contact during play). Depending on the purpose of the assessment, it
may be appropriate that the definition of secondary transmission include infections that result from
propagation of the specific exposure of interest, but not encompass distant transmissions (separated
by time and/or space) that may be more appropriately considered to result as a function of person-
to-environment-to-person transmission. Temporal and spatial limitations can be specifically noted
in the definition of secondary transmission.
32
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
MRA models can be configured to account for secondary transmission and immunity in a
population through the use of a dynamic model (Anderson and May, 1991). These models, which
can take several forms (deterministic or stochastic), characterize the dynamic epidemiological
status of the population (e.g., susceptible to infection, symptomatic infection, immunity). Static
MRA models do not account for the dynamic nature of secondary transmission, although dose-
response parameters derived from static models may be incorporated into dynamic models.
Inclusion of secondary transmission in MRAs can provide non-intuitive results (Eisenberg et al.,
2008); therefore, if secondary transmission and other innate characteristics of infectious disease
transmission are not included in the assessment, a sound justification for this decision is a
suggested component of the risk assessment documentation.
The use of these transmission models in MRA has increased in the past 10-15 years and there are
numerous examples of such models in the literature. For example, Zelner et al. (2010) use a
transmission model to examine secondary spread through households after a point source
foodborne outbreak. Eisenberg et al. (2005) used transmission models to analyze the 1993
Cryptosporidium drinking water outbreak focusing on (1) disaggregating the risk associated with
direct exposure to the contaminated water and subsequent secondary spread; (2) assessing the role
that person-to-environment-to-person played in the outbreak, and (3) assessing the role that
immunity played in the outbreak. Sheng et al. (2009) provides a framework for examining
environmental infection transmission systems and Eisenberg et al. (2002) provides a policy
perspective for using transmission models in decision-making.
Dynamic MRA models take two main forms—deterministic and stochastic. "Deterministic" means
that the model output is strictly determined by the starting conditions and the values of the
parameters in the equations that define the system (i.e. point estimates). In stochastic models,
events are treated as stochastic (random) events within a distribution rather than deterministic ones.
Dynamic MRA methods have been used for numerous specific case studies in the United States
(Eisenberg et al., 1996, 1998; Koopman et al., 2002; Soller et al., 1999, 2003, 2006) and recently
to support regulatory decisions by EPA (U.S. EPA, 2006b). Stochastic MRA models are still
research tools that continue to develop.
Deterministic Dynamic MRA Models
Deterministic dynamic MRA models are suitable for large populations of individuals randomly
interacting with one another. In this form, the population is divided into one of the following
different epidemiological states: (1) susceptible, (2) diseased (infectious and symptomatic), (3)
carrier (infected but asymptomatic), and (4) immune (partial or complete). Only a portion of the
population is in a susceptible state at any point in time, and only those individuals in a susceptible
state can become infected through exposure to pathogens. The dynamic aspect of the model means
that members of the study population move between epidemiological states at different rates, and
thus, the number of individuals in each state changes over time.
Variables in the model track the number of individuals that are in each of the epidemiological
states at any given point in time (thus, these variables are called state variables). The sum of the
number of individuals in each of the epidemiological states equals the total population. A
representative conceptual model for this type of MRA model is presented in Figure 8.
33
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
T
Carrier
Psym
Post-
infection
Exposed
Diseased
Environmental
Pathogen
Sources
Susceptible
Susceptible - Susceptible to infection ~ Movement of individuals
Exposed - Infected, pre-symptoms, pre-infectious
Carrier - Infectious, asymptomatic !~ Movement of pathogens
Diseased - Infections, symptomatic
Post-infection - Protected from infection
Figure 8. Dynamic Risk Assessment Conceptual Model
(Source: Soller and Eisenberg, 2008)
Deterministic dynamic MRA models are expressed mathematically as a set of differential
equations. These equations describe the rate of change in the number (or density) of individuals in
a particular state (or compartment) over time and have defined parameters and starting conditions.
Rate parameters (i.e., the Greek letters in Figure 8) determine the population's movement from
one state to another. Factors affecting the population dynamics include the level and frequency of
exposure, the ability of individuals in infectious states to infect susceptible individuals, and the
temporal processes of the disease (e.g., incubation period, duration of disease, duration of
protective immunity). The rate parameters may be determined through literature review or through
site-specific data, if available and appropriate. Whether single or multiple exposures are
considered should also be discussed during the problem formulation stage.
Deterministic dynamic MRA models have a number of limitations. Modeling relatively small
populations can lead to misestimation of disease. These models also require appropriate parameter
values for transmission rates, and such information can be difficult to determine accurately. Lack
of knowledge and data as well as inherent biological variability suggest a need for uncertainty and
sensitivity analyses of parameter values.
Finally, comparison of static and deterministic dynamic models indicates that under a specific set
of assumptions, the two models are essentially equivalent (Soller and Eisenberg, 2008). The
conditions in which a static model would yield similar results to a deterministic dynamic model
are as follows:
• the background density of the pathogen (or equivalently, the endemic level of
infection/disease) in the population is zero or unimportant;
34
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• the duration of infection and disease approaches zero; and
• infection and/or disease do not confer immunity, or the duration of immunity approaches
zero.
Stochastic Dynamic MRA Models
In a stochastic form, dynamic models incorporate probabilities at an individual level and are
evaluated by an iterative process (e.g., susceptible person A has a probability of contacting
person B, who has a probability of being infectious). This type of model also uses states (or
compartments) for classifying the epidemiological status of the population and subpopulations
(e.g., HIV-positive individuals, individuals greater or less than 5 years of age) under study, but
differs from the deterministic dynamic MRA models in that the compartments contain discrete
individuals rather than the numbers or densities of persons.
In stochastic dynamic MRA models, events are treated as random (stochastic) events within a
distribution rather than deterministic ones. These models employ distributions of outcomes rather
than an average of outcomes as used in deterministic models. A stochastic model will produce
different results, within a range, each time it is run—even with the same starting conditions and
parameters due to the effects of chance. Stochastic forms are suitable for small populations and
heterogeneous mixing. In a small population, chance events, such as an infectious person
contacting only immune persons during the infectious period of illness, can have a substantial
effect on the transmission dynamics of the disease (U.S. EPA, 2004c).
These types of models have been used to investigate the stochastic effects of disease transmission
and localized exposure (U.S. EPA, 2004c). For example, King et al. (2008) used a nonlinear
stochastic model coupled with a new likelihood maximization procedure for model parameter
values to explain the dynamics of cholera infection in Bengal, the pathogen's endemic home.
2.3.2. Data Representation in MRA Risk Estimation Models
In assessing risk associated with infectious disease hazard exposures, it is necessary to estimate a
number of parameters in the risk models. Depending on the data quality, different representations
of these data (as discussed below) may be appropriate. For some chemical risk assessments, EPA
has made a policy decision that conservative (more protective) estimates of some exposure factors
should be used to assure the desired level of health protection for sensitive segments of the exposed
population (U.S. EPA, 2000c). The Agency has not developed a comprehensive policy regarding
how conservative parameter estimates should be in MRA. In fact, the use of multiple layers of
conservative estimates for microbial contaminants has been shown to result in risk estimates that
are not credible and that are overly protective (U.S. EPA, 1995a). Thus, the selection of values
used in the risk assessment and the respective data representation should be well documented in
the risk description. The following is an overview of the various ways that data can be represented
in an MRA.
2.3.2.1. Point Estimates
35
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
A "point estimate" is a single-valued estimate of a parameter used in risk assessment. Using point
estimates for all the parameters in a risk equation results in a single value (point estimate) of risk
that provides no information concerning the potential sources of variability or uncertainty or the
magnitude of that uncertainty associated with the risk estimate. Lack of information regarding
potential variability and uncertainty in the quantity being estimated is a fundamental weakness of
the point estimate approach. The strength of using point estimates is the relative ease of use and
simplified risk assessment output. In some cases, the point estimates themselves may be selected
taking the potential uncertainties in the parameter values into consideration. For example, EPA's
Methodology for Deriving Ambient Water Quality Criteria for the Protection of Human Health
(U.S. EPA, 2000c) recommends that 90th percentile estimates derived from national studies of
drinking water and fish consumption by adults be used in estimating criteria values for chemicals.
Confidence limits provide an indication of the degree of uncertainty associated with a statistic
(Snedecor and Cochran, 1989). Although they are usually derived for estimates of the arithmetic
mean, they can also be estimated for other statistics (e.g., median and percentiles). The narrower
the interval, the more precisely the statistic has been estimated. The magnitude of uncertainty is
expressed in the form of upper and lower confidence limits (collectively known as the confidence
interval); confidence limits always have an associated confidence level (e.g., 90%, 95%). The
confidence level reflects the estimated probability that the numeric statistic estimated, based on a
sample of a given population size, will fall within the specified confidence interval. Confidence
limits typically assume that the underlying distribution in the study population is "normal"
(Gaussian), but alternative assumptions can also be used. Confidence limits can also be derived
that make no assumptions (nonparametric) about distribution shape. An example of a confidence
limit can be found in EPA's Risk Assessment Guidance for Superfund (U.S. EPA, 1989), where
the 95%) upper confidence limit on the arithmetic mean soil density is recommended as the
appropriate point estimate for a screening level risk assessment at Superfund sites.10
2.3.2.2. Statistical Distributions
If adequate data are available, it may be possible to accurately characterize the statistical
distribution of a parameter used in risk assessment. That is, there may be enough data to select the
form of the distribution and to accurately estimate its parameters (e.g., mean, standard deviation,
percentile values for a normal or lognormal distribution). Where such data are available (examples
include national surveys of water intake and body weight), individual summary statistics can be
estimated very accurately (i.e., confidence limits are narrow).
2.3.2.3. Bayesian Methods
The same methods that are used in dose-response modeling can also be used to characterize the
uncertainty in model parameters through the generation of "uncertainty samples." These
uncertainty samples are particularly useful in MRA because they fully characterize the uncertainty
for a specific model parameter, given the available data. For example, Messner et al. (2001)
combined three Cryptosporidium isolates that were considered representative of a larger
population of human-infecting strains and determined that the risks of infection produced from
10 For up to date information on Superfund risk assessments visit:
http://www.epa.gov/oswer/riskassessment/risk superfund.htm
36
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
single oocyst doses for a mixture of the three isolates and for an oocyst selected at random from
the larger population of strains were 0.018 and 0.028, respectively. A related uncertainty analyses
was conducted for the MRA that supported the economic analysis of the Long Term 2 Enhanced
Surface Water Treatment Rule (LT2) drinking water regulation (U.S. EPA, 2006a).
Hierarchical Bayesian modeling is often used in MRAs to combine results from different studies
or isolates in a meta-analysis. For each study or isolate, the parameters can be randomly selected
from a distribution that depends on several additional parameters, called hyperparameters. For
example, the single oocyst infection probability for each Cryptosporidium isolate can be modeled
as being randomly drawn from a normal distribution with hyperparameters [i and o that represent
the variability of the isolate infectivity across the isolate population (Messner et al., 2001). Markov
Chain Monte Carlo method (MCMC) methods can accommodate hierarchical Bayesian models
(Section 4.2.1 and Appendix C).
For example, Gronewold et al. (2009) demonstrated that Bayesian techniques can be used for
quantifying and analyzing uncertainty in exposure model fate and transport parameters. In that
study, Bayesian methods for addressing uncertainty and developing models were compared with
regression techniques in which a model was assumed and uncertainty was assumed related to
confidence in the estimates of model parameters. In comparing approaches for estimating decay
rate parameters from microbial survival experiments, Gronewold et al. (2009) found that Bayesian
techniques, because they rely on fewer assumptions about parameter variability than alternative
techniques, provided higher estimates of variability in the parameters and likely reflect actual
conditions more accurately. Bayesian techniques also allowed these researchers to assess the forms
of models proposed for microbial inactivation and to assess alternative models of the process. The
work reported in Gronewold et al. (2009) is an extension of prior studies by the authors
(Gronewold et al., 2008) in which uncertainty in different enumeration processes was quantified
and related to assessment of water quality.
EPA anticipates that hierarchical modeling will be important in the future of microbial risk
assessment. Roles that Bayesian techniques may be expected to play include development of dose-
response models in the absence of human dose-response data, parameter estimation for sparse data
sets or for data sets exhibiting wide variability, or assessment of alternative models, particularly
in exposure assessment. Bayesian methods are further discussed in Section 4.2.1 with respect to
their use in dose-response modeling and are discussed more fully in Appendix C.
2.3.2.4. Probabilistic Simulations
Distributional data and/or Bayesian-based uncertainty samples can be used in probabilistic MRAs.
In these types of analyses, risk calculations (each of which yields a point estimate) are repeated
many times (typically thousands of times) using random or structured "draws" of values from the
distributions of each parameter value. The resulting distribution of risk provides information about
the expected precision of the estimate, given the distributions of and/or uncertainty associated with
the input parameters. The contributions of variability in individual parameters can also be
estimated and the correlations among parameters can be accommodated within a Monte Carlo
framework, described below.
37
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
EPA has developed guidelines for when probabilistic methods can and should be used in health
risk assessments (U.S. EPA, 1997b). The most common obstacle to the use of probabilistic
modeling is the lack of data to adequately characterize the variability and/or uncertainty in key
input parameters. One approach that has been used at EPA is a "tiered approach" to risk
assessment, whereby the first step is a set of screening calculations to determine if the risks being
estimated fall within the range of concern under a credible set of assumptions. If the results of the
screening level analysis warrant further evaluation, sensitivity analyses can be used to further
characterize the likely range of risks and to guide data gathering efforts for key parameters. If
sufficient data are available, and if more detailed information is needed or desired regarding the
decision being evaluated (e.g., setting a health-based criterion), then Monte Carlo modeling may
be useful as a subsequent tier.
A Monte Carlo simulation is a statistical technique for evaluating the range of possible outcomes
of multiple processes whose outcomes or inputs are random variables. The alternative to Monte
Carlo simulation—integration over the distribution of possible values for the random variables—
is often mathematically impossible. Additionally, many software packages or programming
languages make even large number of Monte Carlo simulations an easy operation on personal
computers. In MRA Monte Carlo simulations, random values for the variables in a mathematical
model for estimating risk are drawn from appropriate distributions. For each simulation, the risk
is calculated based on the mathematical model using the values drawn from their respective
distributions. The results of many simulations may be combined as a distribution of risks
associated with the system. This distribution of risks allows more comprehensive characterization
of risk than a simple point estimate. For example, risk management decisions might be made based
on the 95th percentile of the distribution resulting from the Monte Carlo simulation. The
distribution of outcomes permits a more informed selection of risk-based decisions considering
the full range of data inputs and the nature of the effects of concern. It can also support
consideration of costs and benefits arising from key decisions. Use of Monte Carlo simulations
requires significantly more information about the system being modeled than generation of point
estimates. Each random variable is characterized by a distribution (either parametric or non-
parametric) and the choice of the distributional form and estimation of distribution parameters both
require data or some other information that informs their selection.
Several illustrations of the Monte Carlo modeling technique within QMRAs of drinking water
exposures are provided below to show the importance of considering variability in processes
within a MRA and to illustrate the application of the technique. Olivieri et al. (1999) used Monte
Carlo simulations of alternative advanced water treatment (e.g., reverse osmosis) trains to compare
the efficiency of the entire treatment trains with respect to removal and inactivation of an enteric
virus surrogate. The data used in that study were generated in pilot studies of the unit operations.
The Monte Carlo simulation allowed exploration of interactions in performance of individual unit
operations prior to building a full-scale facility. Distributions for removal in each of the unit
operations were based on observed removal from challenge studies of pilot processes and are
presented in Table 3. In this study Monte Carlo techniques provided an opportunity for evaluation
of multi-component treatment processes and for providing quantitative information for processes
expected to reduce microorganism density below detectable limits.
38
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 3. Distributions used in Monte Carlo simulations conducted
by Olivieri et al. (1999)
Unit Process
Distribution used to characterize
removal of viruses
Justification
Influent
Log-normal
Maximum likelihood
estimation
Reverse osmosis
filtration
Weibull and gamma
Maximum likelihood
estimation
Ultrafiltration
Log-normal
Insufficient data for
maximum likelihood
estimation; distribution
assumed log-normal
Ozonation
Point estimate
Insufficient data to
characterize distribution
Micro filtration
Gamma
Maximum likelihood
estimation
Chlorine contactor
Uniform distribution
Based on CT (contact
time) operational range
Pouillot et al. (2004) used a second-order Monte Carlo simulation to estimate cryptosporidiosis
risk based on observed Cryptosporidium densities in a finished water reservoir and to initiate
discussion of monitoring strategies and alternative water quality standards for managing risk. This
study is reviewed here because of its separation of uncertainty and variability via the second order
Monte Carlo simulation. It is also notable in its explicit modeling of both immune-competent and
immune-compromised individuals. The second order Monte Carlo simulation entailed using a
Monte Carlo simulation to develop a distribution of a parameter making up the system model, then
sampling from the resulting distribution in a Monte Carlo simulation of the overall system. In this
case, the inner Monte Carlo simulation developed distributions of uncertain parameters and the
outer Monte Carlo sampled from those distributions systematically to develop a distribution for
risk. Other studies (Petterson et al., 2007; Signor et al., 2005) have noted the importance for
explicitly modeling parameter uncertainty within MRAs and proposed the use of second Monte
Carlo simulations in such analyses.
The decision whether to use probabilistic methods can be technically complex; thus, expert
statistical advice should be sought to support such decisions. When planning such assessments, it
is important to ensure that the approach taken to characterize uncertainty is consistent across the
models used in all stages of the risk assessment. An example of such an analysis can be found in
EPA's risk assessment in support of the LT2, which addresses Cryptosporidium contamination in
sources of drinking water (U.S. EPA, 2006a).
39
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
2.4. Elements to Consider During Problem Formulation
The problem formulation process should provide a working outline of the risk assessment.
Furthermore, information that is required for the exposure assessment and health assessment
phases of risk assessment is preliminarily gathered and reviewed during problem formulation. The
elements also have the potential to influence one another. Where appropriate, these influences
should be noted during problem formulation and included in the risk assessment documentation.
In this document, the infectious disease hazard characterization and host characterization are
initially considered as part of problem formulation because the resulting data and information are
important for building the conceptual model(s) and making the decision if adequate data are
available for the desired scope of a given risk assessment. It is appropriate to consider these steps
as overlapping with the risk assessment phase because the data gathered during problem
formulation are then used during the risk assessment.
2.4.1. Infectious Disease Hazard Characterization
For the purposes of this document, an infectious disease hazard is defined as a pathogenic
microorganism. If you wish to consider a surrogate for an infectious disease hazard, such as fecal
indicator bacteria, please refer to EPA's Technical Support Materials (U.S. EPA, 2014). Infectious
disease hazards can also include multiple pathogens simultaneously, such as reported by Westrell
et al. (2003), where risks due to failures in drinking water treatment systems were modeled for the
pathogens Cryptosporidium, rotavirus, and Campylobacter jejuni.
Factors related to infectious disease hazards that should be considered during problem formulation
are listed and briefly discussed below (also see Table 1):
Infectious disease hazard characterization elements include (adapted from ILSI, 2000):
• survival, multiplication, and accumulation;
• resistance to control or treatment processes; and
• ecology (including zoonotic potential, vectors, and epidemiological triangle).
Pathogen elements that overlap between exposure and human health effects include:
• virulence and pathogenicity of microorganism;
• pathologic characteristics/disease caused, including host specificity (including zoonotic
potential and vectors);
• infection mechanisms/route of infection/portals of entry;
• potential for secondary transmission; and
• taxonomy/strain variation.
Environmental Survival, Multiplication, and Accumulation
A pathogen may be able to survive in water but be unable to infect a host. Many molecular-based
microbial assays and some fluorescent antibody assays do not distinguish between live/dead or
40
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
infectious/noninfectious organisms (e.g., deoxyribonucleic acid amplification methods,
polymerase chain reaction [PCR]). For assays that require growth of the microorganisms under
laboratory conditions there is a concern that VBNC microorganisms will not be detected. Risk
assessors should be aware of, and report the caveats of, the assays used to quantify microorganisms
in the studies they use as data sources for an MRA.
Multiplication refers to the ability of some microorganisms to reproduce or grow in the
environment. The combination of survival, viability, infectivity, virulence, and multiplication may
be addressed through fate and transport modeling. Accumulation can occur in a variety of ways.
Some examples include accumulation in biofilms (in pipes or tanks), accumulation in sediments,
adsorption to particulate matter in water, and bioaccumulation in filter feeding aquatic organisms
(e.g., shellfish). These factors contribute to heterogeneity of microbes. Places in the risk scenario
where accumulation can occur should be noted.
Survival, multiplication, and accumulation of microorganisms are dependent on environmental
conditions such as temperature, nutrient availability, and other water quality parameters (NRC,
2004). Treatment processes can also influence survival and may alter virulence and pathogenicity.
Table 4 presents several representative tools for modeling pathogen survival, multiplication, and
accumulation. Environmental niches that can harbor pathogens should be considered, such as
biofilms and amoebae (e.g., Legionella can live inside amoebae; Brown and Barker, 1999). The
extent to which survival, multiplication, and accumulation will affect the risk assessment should
be considered and documented during problem formulation.
Table 4. Representative Tools for Modeling Pathogen Survival, and Multiplication
Tools
Reference
Survival and Transport of Viruses in the Subsurface: An Environmental
U.S. EPA, 2003d
Handbook. This issue paper discusses some of the conditions under
which viral contaminants may survive and be transported in the
subsurface, identifies sources as well as indicators of viral
contamination, outlines the effects of hydrogeologic settings on viral
movement, and introduces the reader to the current state of virus
transport modeling along with an example of modeling applications.
Continuous simulation
Recommended by TMDL
Protocol (U.S. EPA, 2001)
Monte Carlo simulation
Recommended by TMDL
Protocol (U.S. EPA, 2001)
Log-normal probability modeling
Recommended by TMDL
Protocol (U.S. EPA, 2001)
USDA/Agricultural Research Service (ARS) Pathogen Modeling Program
Version 7.0
(PMP) estimates the effects of multiple variables on the growth or
httD://www.arserrc.aov/mfs/
survival of foodborne pathogens
Download.htm
41
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
Table 4. Representative Tools for Modeling Pathogen Survival, and Multiplication
(continued)
Tools Reference
ComBase, also developed by USDA/ARS, is an on-line database of http://wvndmoor.arserrc.qov
predictive microbiology information collected from researchers, /combase/default.aspx
institutions, and published literature. ComBase may be searched based
on temperature, pH, water activity, condition, source (publication),
organism, and environment. Files are provided giving organism,
maximum rate, doubling time or D-value, source, conditions,
environment, temperature, pH, water activity, a table and chart for log
density versus time, and other available details. (Maximum rate is the
maximum slope of the "log [cell density] versus time" curve, in a given
environment.)
Resistance to Control or Treatment Processes
Microorganisms have varying degrees of resistance to water treatment and control processes. The
extent to which these control or treatment processes will affect the risk assessment should be
considered and documented during problem formulation. For example, data on how pathogens
respond to both wastewater treatment and public water supply treatment should be noted, as
appropriate. If the risk assessment is for a performance target, then the treatment and control
processes may be of central importance. For example, Cryptosporidium oocysts are very resistant
to conventional disinfection with chlorine, so chlorination in the absence of filtration may be
inadequate to protect public health if oocysts are present in source waters for drinking water.
Ecology
The epidemiological triangle (epi triad) is a recommended model for conceptualizing agent-host-
environment interactions and is a useful way to consider ecology (Figure 9). The epi triad can be
used to predict epidemiological outcomes and provides a tool to discuss parameters that influence
public health outcomes. The epi triad can capture how pathogen, host, and environment all affect
each other.
Pathogen
Pathogen
Environment
Environment
Figure 9. Two Versions of the Epi Triad (Source: CDC, 1992)
Physical properties of microorganisms that relate to their transport/mobility (e.g., hydrophobicity)
and data on pathogen survival and bacterial colonization under varying ecological conditions (e.g.,
stressors such as pH, nutrient availability, and temperature) should be considered and discussed
within the context of the scope of the risk assessment. Often, the ecology of pathogens can be
42
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
elucidated by examining their transport, fate, and survival in the environment—particularly how
they react in variable media. If sufficient data or information is available to prepare an ecology
summary for the pathogen, it should focus on the appropriate exposure source11 (water, food,
other) and may include environments that the microorganisms encounter before they enter the
media of concern. For example, MAC thrive in hot water and have been known to colonize hot
water systems in hospitals and buildings (Primm et al., 2004).
Ecological niches may also be provided by other microorganisms. For example, biofilms create
ecological niches that are important to consider because microbes often exhibit different properties
in communities compared to the same species living in suspension. For example, V. cholerae from
human stool has enhanced infectivity in a rabbit model relative to infectivity of dispersed
(planktonic) cells (Faruque et al., 2006). This is because V. cholerae from human stool are in
conditionally viable forms (biofilms and multicellular clumps).
Protozoa can also harbor bacteria within their cell membranes, thereby protecting the bacteria from
many environmental stresses. V. cholerae occur commensally in zooplankton. A single copepod
can carry up to 104 cells of V. cholerae, and human volunteer studies show that ~104 to 106 V.
cholerae can cause clinical cholera (Colwell et al., 2003). Legionella are known to reside within
at least 20 species of amoebae, two species of ciliated protozoa, and one species of slime mold
(Lau and Ashbolt, 2009). Species from the pathogens Vibrio, Mycobacterium, Helicobacter,
Afipia, Bosea, Pseudomonas, and mimiviruses are also associated with protozoa in the
environment.
Virulence, Pathogenicity, Pathological Characteristics, and Host Specificity
Virulence and pathogenicity refer to how easily and effectively a pathogen may cause disease in a
host. Virulence is "the degree of intensity of the disease produced by a microorganism as indicated
by its ability to invade the tissues of a host and the ensuing severity of illness." Pathogenicity is
"the property of an organism that determines the extent to which overt disease is produced in an
infected population, or the power of an organism to produce disease. It is also used to describe
comparable properties of toxic chemicals. Pathogenicity of infectious agents is measured by the
ratio of the number of persons developing clinical illness to the number exposed to infection" (U.S.
EPA, 2007a). Although both can be expressed numerically, the general definitions can be broader.
Pathological characteristics are a description of the disease symptoms that result from exposure
and infection by the pathogen (including strain variations). The known range of disease symptoms
should be briefly reviewed and the specific health endpoint that the risk assessment addresses
should be presented within the context of the broader range of health endpoints.
Host specificity is a pathogen characteristic that is related to host susceptibility. A species is not
considered a host if it cannot be infected by the pathogen. Note that a species can still be considered
a host even if no illness results from infection. Within a host species there is variability in
susceptibility. For example, mild illness can occur in immune-competent persons resulting from
11 Exposure sources can be the media through which the contaminant is delivered, such as water or shellfish, or
exposure sources can indicate the origin of the contaminant, such as point source, non-point sources, or naturally
occurring. Exposure route indicates the sites of body contact that are relevant for access to sensitive tissues and
organs, such as ingestion, inhalation, and dermal exposures.
43
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
exposure to a pathogen; whereas, severe illness may occur in immunocompromised persons.
However, host specificity most often refers to the range of species that are infected by the
pathogen.
Information that can facilitate the comparison of human response to a pathogen versus laboratory
animal models' response should be examined and is particularly important if data from animal
models will be used to characterize dose-response or symptomatology during the risk assessment.
Wild and domestic animals may also be prone to infection and disease (zoonotic potential) and
thus may be a source of pathogens for human exposure either directly or through transport in the
environment. Some pathogens have non-human carriers, also known as vectors, which are
important in the pathogen life cycle or serve as an environmental reservoir.12 The potential role of
susceptible animals, vectors, and environmental reservoirs in the risk scenario should be addressed,
which may include an explanation of how animals are contaminating the water sources of concern.
These factors are also evaluated in greater detail in the health effects section of the risk assessment.
Infection Mechanisms, Route of Infection, and Portals of Entry
Infection mechanisms, route of infection, and portals of entry emphasize the manner in which
pathogens interact with hosts. The exposure routes13 that will be included in the risk assessment
are defined during problem formulation. Part of that definition should include identification of
known routes that will not be part of the scope of the risk assessment. For example, in many
waterborne pathogen risk assessments, the ingestion route of exposure is investigated and other
routes of exposure (e.g., inhalation) are not included. In cases where a pathogen is not known to
be infectious through certain routes, such as the dermal route, discussion, including rationale, for
not including the dermal exposure route should be included in the risk assessment documentation,
particularly for risk assessments conducted to meet specific statutory requirements where reasons
for excluding a route must be justified. Additional discussion of how the choice of included routes
impacts the uncertainty, qualitative or quantitative, should be included in the risk characterization
section.
For an infection to occur, the host's target organ must come in contact with a sufficient number of
microorganisms; the microorganism must possess specific virulence factors; these virulence
factors must be expressed; and the defenses of the host and/or target organ systems (e.g., digestive
system, lung) must be overcome. With some microorganisms (e.g., Giardia, Cryptosporidium),
the interaction with the particular organ is so specific that infections are almost always confined
to that one organ site; with others (e.g., Salmonella, enteroviruses) the pathogen has the potential
to infect more than one target organ. When attempting to establish a health risk due to exposure to
pathogens through contact with food and drinking water, one must consider that the human GI
tract is a complex organ system with a variety of specific host defense mechanisms. It is only when
12 The term "vector" can mean "anything which transmits parasites" (for use in this MRA Tools document, a vector
can transmit bacteria, viruses, or parasites) (www.swintons.net/ionathan/Academic/glossarv.html') or can refer to
intermediate hosts that are required for life cycle completion. Environmental reservoirs include free-living amoeba
that can harbor bacteria intracellularly allowing the bacteria to survive in harsher environments than they could
normally survive (NRC, 2004).
13 Route of exposure refers to how the pathogen comes in contact with the vulnerable host receptor cells that support
infection (e.g., inhalation, dermal contact, oral), whereas source of exposure refers to the physical matrix that carries
the pathogen (e.g., air, water, food, soil).
44
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
the pathogen has particular virulence factors for sites in the GI tract, and the specific host defense
mechanisms in the GI tract are breached, does infection occur. Infection without symptoms and
the duration of infection are important attributes of the infection process because they contribute
to the potential for secondary transmission via the shedding of pathogens into the environment.
An MRA can also include exposures that have not previously been commonly considered. For
example, Feazel et al., (2009) analyzed ribosomal ribonucleic acid gene sequences from 45
showerhead sites around the United States. The majority of showerhead microbiota were
comprised of genus- or species-level groups that are commonly found in water and soil. The
showerhead environment strongly enriches for microbes that are known to form biofilms in water
systems, including Mycobacterium spp., Sphingomonas spp., Methylobacterium spp., and other
pathogens. The detection of significant loads of M. avium in showerhead biofilms identifies a
potential personal health concern. Legionella, which can cause Legionnaires disease and Pontiac
fever, has been linked epidemiologically to hot water systems, cooling towers, evaporative
condensers, humidifiers, whirlpool spas, respiratory devices, and decorative fountains. It has also
been isolated from those sources as well as water taps, hot tubs, showers, creeks, ponds, and soils
from the banks of water bodies (APHA, 2004). There may also be other opportunistic pathogens
that are of interest to risk managers and assessors.
Secondary Transmission
The potential for secondary transmission will also contribute to human exposure. Secondary
transmission refers to infection spreading from one infected person to another person. Secondary
cases (often represented by a secondary attack rate) generally refer to cases or an attack rate that
occurs among contacts, within the incubation period of the pathogen, and following exposure to a
primary case. In some cases, direct person-to-person transmission cannot be separate from
contamination of the immediate environment and subsequent transmission to another person (e.g.,
toddlers sharing toys versus direct physical contact during play). In most cases, it is appropriate
that the definition of secondary transmission include infections that result from propagation of the
specific exposure of interest, but not encompass distant transmissions (separated by time and/or
space) that may be more appropriately considered to result as a function of person-to-environment-
to-person transmission. Temporal and spatial limitations should be specifically noted in the
definition of secondary transmission for a given pathogen. Full discussion of the range of scenarios
that qualify as secondary transmission should be included where appropriate.
The above definition of secondary transmission is limited to avoid overlap with pathogen
occurrence in the environment (person-environment-person), although people are, of course, part
of the environment. However, the potential for re-introduction of the pathogen into the exposure
media can also be within the definition of secondary transmission. Dynamic MRA models can
characterize secondary cases that occur among contacts following exposure to a primary case,
whereas static MRA models usually consider secondary transmission to be negligible or include it
as a non-fluctuating multiplicative factor (e.g., secondary cases equal primary cases multiplied by
0.1; assuming a 10% secondary transmission rate). The problem formulation documentation
should indicate if and how secondary transmission is included in the assessment. If it is not
included, justification for this decision should be provided.
45
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Taxonomy and Strain Variation
Taxonomy and strain variation have a potentially large effect on risk assessment. The difference
in dose-response range between isolates (and strains) can be orders of magnitude (see Table 5).
Some strains may not be infective for humans. In addition, the ratio of different strains in the
environment can fluctuate. These factors make characterization of pathogen occurrence difficult
(Messner et al., 2001; Teunis et al., 2002). The extent to which strain variation is accounted for in
the risk assessment should be documented during problem formulation.
Table 5. Virulence of Three Cryptosporidium parvum Isolates
in Healthy Adult Humans (Source: Okhuysen et al., 1999)
Isolate
Isolate Characteristic
Iowa
UCP
TAMU
Infectious dose for 50%
of population (IDso)
87 oocysts
1042 oocysts
9 oocysts
Attack rate
86%
52%
59%
Duration of symptoms
64.2 hours
81.6 hours
94.5 hours
2.4.2. Initial Host Characterization
Host characterization involves an evaluation of the intrinsic and acquired traits that modify the risk
of infection or illness in a potentially exposed human population. It is also possible that host factors
may be important in determining the severity or outcome of an infection. For example, high-risk
groups may develop severe symptomatic illness, whereas, low-risk groups may develop
asymptomatic infections or mild illness.
The following populations are typically considered more susceptible14 than the general population:
pregnant women; neonates and children; people over 65 years of age; individuals residing in
nursing homes or related care facilities; and cancer, organ transplant, and acquired immune
deficiency syndrome (AIDS) patients (Haas et al., 1999). The Report to Congress, EPA Studies on
Sensitive Subpopulations and Drinking Water Contaminants (U.S. EPA, 2000d), summarizes
EPA's approach to identifying and characterizing susceptible subpopulations that may be at greater
risk from exposure to drinking water contaminants than the general population.
Host characteristics have the potential to influence both the exposure and the health effects
components of the risk assessment. These factors are often used to define potential subpopulations
of interest for a risk assessment because they can influence the assessment with respect to
14 Sensitive subgroups are "identifiable subsets of the general population that, due to differential exposure or
susceptibility, are at greater risk than the general population to the toxic effects of a specific air pollutant (e.g.,
depending on the pollutant and the exposure circumstances, these may be groups such as subsistence fishers, infants,
asthmatics, or the elderly)" (U.S. EPA, 2007a). Susceptible subgroups "may refer to life stages, for example,
children or the elderly, or to other segments of the population, for example, asthmatics or the immune-compromised,
but are likely to be somewhat chemical-specific and may not be consistently defined in all cases" (U.S. EPA,
2007a). Note that in the above definitions "susceptible" refers to host characteristics and "sensitive" refers to host
characteristics and exposure patterns. Another similar usage of terms is "susceptible" to refer to intrinsic factors and
behavioral and situational factors as extrinsic, with both intrinsic and extrinsic being grouped under "vulnerable."
46
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
susceptibility to infection and severity of illness. The extent to which these factors are considered
in the risk assessment should be described in the problem formulation documentation. Factors
related to host characterization that should be considered during problem formulation are briefly
discussed below (also see Table 1):
• Part of host characterization in problem formulation is defining demographics of the
populations of concern. The size of the exposed population refers to the number of people
who come in contact with the media of concern. The demographics and behavior of the
exposed population can conceptually include many possible subgroups. Defining the
subpopulations that will be considered is a key component of problem formulation.
Subgroup differentiation is not necessary unless there is evidence for relevant differences
between the subgroups.15 There should be scientific rationale presented for dividing
subgroups as well as data that directly pertain to that subgroup or can be adjusted to address
that subgroup. For example, it is unlikely that differentiating between 24 and 25 year-olds
would provide any additional useful information for risk managers.
• The young and the elderly generally have less resistance to infections16. Children,
especially malnourished children, may be more likely to exhibit severe effects of acute GI
illness (AGI) after exposure to some pathogens (e.g., pathogenic E. coli, some enteric
viruses). However, some pathogens (e.g., Hepatitis A, poliovirus) may cause less clinical
illness in children than in adults (Gerba et al., 1996a). Age can also contribute to different
exposure patterns due to behavior. For example, children may have higher levels of
incidental ingestion of water during swimming than adults (Dufour et al., 2006). Because
drinking water consumption increases with age, the elderly consume more drinking water
than adults or children (Roseberry and Burmaster, 1992).
• Populations that are considered immunocompromised or immunosuppressed due to recent
or concurrent illness or medical treatment may be defined as subpopulations that the risk
assessment will address (Effler et al., 2001). However, all definitions of subpopulations
included in the risk assessment should include the criteria used to classify individuals as
immunocompromised and may need to be limited to specific identifiable types of immune
defects. Extreme physical or emotional stress can lower immune competency. The host GI
environment can vary in ways that affect pathogens and innate immunity also plays a role
in infection dynamics.
• Previous exposure may confer limited and/or short-term protective immunity for some
pathogens (Frost et al., 2005). The converse of this may also be true; that is, when
individuals or populations that have not previously been exposed to particular pathogens,
infection and illness rates can be higher than would otherwise be anticipated. "Traveler's
diarrhea" is a well-known observed phenomenon that exemplifies this type of situation.
• Concurrent illness/medical treatment (physical and mental stressors may increase
susceptibility).
15 Note that risk assessments being performed as part of a statutory requirement may already have mandated
subgroups.
16 Although children are referred to in conjunction with subpopulations in this document, U.S. EPA acknowledges
that childhood represents a life-stage rather than a subpopulation, the distinction being that a subpopulation refers to
a portion of the population, whereas a life-stage is inclusive of the entire population (http://www.lJ. S.
EPA, gov/teach/index.html').
47
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• Genetic background can also affect immune status, but may play a larger role in mechanism
of infection and disease progress.
• Pregnancy may cause women to be more susceptible to a pathogen. For example, Hepatitis
E, which causes a self-limiting disease in most infected persons, can cause up to 20%
mortality in women in the third trimester of pregnancy (Jameel, 1999).
• Malnourished individuals tend to have weaker immune defenses than well-nourished
individuals.
• Social and behavioral traits primarily affect exposure patterns. For example, a relatively
small proportion of the population is responsible for consuming the majority of raw and
partially cooked shellfish (FDA, 2005). As mentioned above, age may also be related to
behaviors that affect pathogen exposure patterns.
Data for the above elements can be arranged into groups by stratification or multivariate analysis.
Alternatively, host characteristics can be considered by conducting a separate risk assessment for
each characteristic that is believed to have some importance. For example, in addition to a risk
assessment for the overall population, a separate risk assessment may be performed for each
subpopulation of interest (e.g., young children, the elderly, pregnant women,
immunocompromised persons) provided that sufficient data are available for valid statistical
interpretation. EPA's Risk Assessment Forum has developed Guidance on Selecting Age Groups
for Monitoring and Assessing Childhood Exposures to Environmental Contaminants (U.S. EPA,
2005b), which recommends subgroups that address anatomy/physiological development in the
following age groupings: birth to <1 month, 1 to <3 months, 3 to <6 months, 6 to <12 months, 1
to <3 years, 3 to <8 (female) or <9 (male) years, and 8 or 9 years to <16 (female) or <18 (male)
years.
In recognition that children have a special vulnerability to many toxic substances, the EPA
Administrator's October 1995 Policy on Evaluating Health Risks to Children directs the Agency
to explicitly and consistently take into account environmental health risks to infants and children
in all risk assessments, risk characterizations, and public health standards set for the United States.
In April 1997, President Clinton signed Executive Order (EO) 13045 Protection of Children from
Environmental Health Risks and Safety Risks, which assigned a high priority to addressing risks
to children (EO, 1997). In May 1997, EPA established the Office of Children's Health Protection
to ensure the implementation of the President's EO. EPA has increased efforts to ensure its
guidance and regulations take into account risks to children. In 2002, EPA published an interim
report on child-specific exposure factors (U.S. EPA, 2002d).
2.4.2.1. Environmental Justice
EO 1289, Federal Actions to Address Environmental Justice in Minority Populations and Low-
Income Populations (February 1994), ordered Federal agencies, including EPA, to "...make
achieving environmental justice part of its mission by identifying and addressing, as appropriate,
disproportionately high and adverse human health or environmental effects of its programs,
policies, and activities on minority populations and low-income populations..." (EO, 1994). EPA
responded to EO 12898 with The EPA's Environmental Justice Strategy (U.S. EPA, 1995b). In
2001, in a memo from the EPA Administrator environmental justice was defined as follows: "The
Agency defines environmental justice to mean the fair treatment of people of all races, cultures,
48
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
and incomes with respect to the development, implementation, and enforcement of environmental
laws and policies, and their meaningful involvement in the decision-making processes of the
government" (emphasis in original).17 EPA further defined meaningful involvement in EPA's
Public Involvement Policy (U.S. EPA, 2003g) as follows:
"Meaningful involvement"...means that: (1) potentially affected community residents
have an appropriate opportunity to participate in decisions about a proposed activity that
will affect their environment and/or health; (2) the public's contribution can influence the
regulatory agency's decision; (3) the concerns of all participants involved will be
considered in the decision-making process; and (4) the decision-makers seek out and
facilitate the involvement of those potentially affected.
Risk assessment documentation should provide clear descriptions of subpopulations and other
parameters that may help EPA evaluate whether there are potential environmental disparities that
could cause an environmental justice concern.
2.5. Linkage between Problem Formulation and Other MRA
Components
The planning and scoping and problem formulation process develops the scope of the risk
assessment, taking into account management needs, Agency risk assessment policies, risk
assessment tool availability, data constraints, and overall Agency resources. The output from this
process—documentation of the problem formulation development—provides the linkage to rest of
the MRA process.
A problem formulation lays out how everything fits together in the risk assessment. Well-
formulated problem formulation effectively communicates the scope, purpose, and methods that
will be used in the MRA. It provides clarity about the stressor and provides a conceptual model
for assessing risk. Finally it provides an analysis plan describing how exposure will be evaluated,
how the hazard will be evaluated, and how these will be integrated to develop the risk assessment.
17 http://earthl.EPA.gov/oswer/ei/html-doc/eimemo.htm
49
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
50
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3. Exposure Assessment
The characterization of exposure consists of the technical evaluation of data related to the potential
exposure to microbial contaminants in water media. The problem formulation phase of the MRA
precedes the exposure assessment and may partially address many of the issues to be evaluated in
the exposure assessment. However, the exposure assessment is more detailed and generally more
quantitative than the problem formulation phase.
In human health risk assessment, exposure is defined as human contact with a biological, physical,
or chemical agent—usually through ingestion, inhalation, or dermal contact. Risk assessment can
be performed for specific target populations or an individual target organism (a human with a
defined exposure pattern). Exposure assessment involves the determination or estimation
(qualitative or quantitative) of the magnitude, frequency, duration, and route(s) of exposure (U.S.
EPA, 1997a). A primary purpose of exposure estimation is to support dose estimation (U.S. EPA,
1992). Dose is the amount of a pathogen that enters or interacts with a host (ILSI, 2000).18
For nearly all MRA contexts, dose refers to potential dose (i.e., the number of pathogens ingested
in a specified period) because the actual numbers of pathogens that an individual is exposed to is
almost always unknown. In fact, most MRAs are performed without direct estimates of pathogen
dose. Doses are typically calculated as a function of pathogen density in the exposure medium
(e.g., drinking water, reclaimed water, biosolids) and the volume of that medium that is ingested
or inhaled. An important reason for calculating pathogen doses is that doing so allows the data
from one exposed population (e.g., the volunteers in a virulence study) to be applied to risk
assessments for other exposed groups, such as the general population.
Characterization of exposure involves an evaluation of the interaction between the pathogen, the
environment, and the human population (i.e., the classic epidemiological triad; Figure 9). The
infectious disease hazard characterization, occurrence, and exposure assessment sections are
brought together to develop an Exposure Profile that quantitatively or qualitatively summarizes
the magnitude, frequency, and pattern of human exposure for the scenario(s) under investigation.
Exposure is not limited to a pathogen-specific context. It can also be defined in terms of water
quality indicators such as the presence of coliforms, pathogen surrogates, or types of water sources
(e.g., ground water, impoundments, rivers) coupled with estimated efficiencies of treatment
18 U.S. EPA (U.S. EPA, 1997a, 2003e, 2004b, 2005a) has defined "dose as the amount of a substance available for
interactions with metabolic processes or biologically significant receptors after crossing the outer boundary of an
organism. The potential dose is the amount ingested, inhaled, or applied to the skin. The applied dose is the amount
presented to an absorption barrier and available for absorption (although not necessarily having yet crossed the outer
boundary of the organism). The absorbed dose is the amount crossing a specific absorption barrier (e.g., the exchange
boundaries of the skin, lung, and digestive tract) through uptake processes. Internal dose is a more general term
denoting the amount absorbed without respect to specific absorption barriers or exchange boundaries. The amount of
the chemical available for interaction by any particular organ or cell is termed the delivered or biologically effective
dose for that organ or cell." Note that these sub-definitions of dose are not used to describe pathogen interactions with
hosts. The term "minimum infectious dose" was intended to indicate the lowest dose that would cause infection in an
individual and assumed that there was a threshold dose. This term is generally considered to be obsolete because as
little as one microorganism is believed to be capable of causing infection in a susceptible individual. However, it
should be reiterated that infection does not imply symptomatic illness (U.S. EPA, 2007a).
51
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
technologies (e.g., filtered and disinfected water versus disinfected water only). In all cases where
indicators are used in risk assessment, it is important to document fully the basis for their use (i.e.,
the extent to which they are correlated with the pathogen or health effect of concern), and to specify
clearly the conditions under which the correlation is expected to be valid. QMRA for indicators
and surrogates is beyond the scope of this document. For more information on using this type of
information, please refer to EPA's Technical Support Materials (U.S. EPA, 2014).
Among elements of MRAs, the exposure assessment is the one with the greatest flexibility in its
formulation and often, the greatest data needs. Although some MRAs may be created to provide
large-scale or general estimates of microbial risk (e.g., nationwide estimates of GI illness due to
consumption of drinking water), most QMRAs estimate risk for specific venues and pathogens.
As noted above, sources, transport processes, and epidemic and episodic occurrences of pathogens
are highly variable both at a particular site and between sites. In exposure assessment, an exposure
profile is developed, inclusive of all the processes, variabilities, and uncertainties for the venue of
interest. Thus, the exposure assessment must assemble models and data that address the questions
such as following:
• What are the sources of the pathogens?
o What is the temporal variation of the occurrence and abundance of the pathogen of
interest in the source?
o How certain are measurements of the pathogen density in the source?
• What is the pathway from source to receptor over which the pathogen must survive and be
transported?
o What are the transport or treatment processes that determine the microorganism
density at the point of ingestion?
o How do those processes vary with time?
o How persistent is the microorganism in the water matrix and for the physical
conditions at the venue of interest?
• What is the route of exposure (inhalation, oral, dermal)?
o If oral, what volumes of water will be ingested and where will ingestion occur?
¦ Do ingestion rates vary among potentially affected population groups?
¦ How do sites where ingestion occurs differ?
• What are the magnitude, duration, and frequency of the exposures of interest?
The answers to these types of questions will be quite different for different hazards. Sites and
models describing exposure may include highly detailed flow models, as included in an assessment
of the benefits of additional wastewater treatment (Soller et al., 2003). They can be simple, but
effective, as in a study of cryptosporidiosis arising from drinking water consumption by Pouillot
et al. (2004). In all MRAs, exposure assessments should be sufficiently detailed to meet the MRA
objectives and be consistent with the conceptual models generated in problem formulation.
Exposure assessment data needs are determined by the MRA objectives and by the formulation of
the exposure model. As described in Section 5.3, there are advantages to characterizing some
parameters of the exposure model as statistical distributions. Selection and parameterization of
those distributions require data. The more data that are generated or gathered, the more certain the
distribution choice and parameters will be.
52
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.1. Occurrence
Occurrence refers to the conditions that lead to the presence of a pathogen or to the distribution
pattern of a pathogen in the environment and the media of concern. The EPA-ILSI Framework
(ILSI, 2000) identifies the following elements to consider when characterizing pathogen
occurrence during exposure assessment:
• temporal distribution/frequency;
• density in environmental media;
• spatial distribution (clumping, aggregation, particles, clustering);
• niche (ecology, non-human reservoirs);
• survival, persistence, amplification;
• seasonality;
• meteorological and climatic events; and
• presence and effectiveness of treatment or control processes.
Many of these factors are interrelated and as such cannot be discussed independently. There are
three basic questions that should be answered to describe pathogen occurrence in a water body—
when (including duration), where, and how much (level)? When information on a particular
pathogen species of interest is lacking, it may be necessary to use occurrence data for surrogate or
index species. The limitations and uncertainty associated with those data and their use should be
evaluated and discussed.
3.1.1. When Do Pathogens Occur in the Water Body?
Temporal distribution/frequency describes when pathogens occur. Fluctuations in microbial
densities can occur on almost any time scale (Boehm, 2007; Boehm et al., 2002) and over a wide
range of spatial scales. In many cases, pathogen occurrence can vary on wide-ranging time scales,
including hourly, daily, weekly, monthly, seasonally, or yearly fluctuations. Spatial variations
often relate to the position of a pathogen source relative to the location of receivers or to mixing
processes at a site of interest. Representative drivers for these fluctuations are listed along with
their associated length and time scales in Table 6.
Meteorological and climatic events such as storms, changes in wind direction and shifts in currents
may cause changes in pathogen occurrence. Seasonality is a factor that affects the temporal
frequency of many waterborne pathogens, such as Cryptosporidium. Seasonality also influences
microbial density in environmental waters—either because of dependence of persistence on
temperature or because of differences in loading between wet and dry seasons. Moreover, seasonal
animal-related events, such as calving or bearing young seem to be associated with some zoonotic
pathogens. Occurrence data that are linked to temporal events such as seasons may be useful for
predicting how pathogen levels may respond to future events. If wastewater treatment or control
processes are less efficient, as may occur during storm events (resulting in combined sewer
overflows and sanitary sewer overflows), there may be associated temporal fluctuations in
pathogen levels. Urban and agricultural runoff can also influence pathogen occurrence in surface
waters.
53
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 6. Sources and Scales of Variability in Pathogen Occurrence
Variability Type
Variability Driver
Length Scale
Time Scale
Seasonal changes
Regional
Months
Human or animal epidemics
Watershed or sub-watershed
Weeks to months
Hydrometerological events
Watershed
Days
Temporal
Treatment plant operational events
Watershed for WWTPs,
distribution system for WTPs
Hours to days
Animal life-cycles
Sub-watershed
Weeks to months
Tidal process
—
Hours
Solar cycle
—
Hours
Wave action or bottom scour
Meters
Minutes
Alignment of source with water of interest
Kilometers
—
Spatial
Hydraulics and mixing
Meters to kilometers
Minutes to days
Distribution of sources
Watershed
—
Disease epidemics in a community also can affect pathogen occurrence, particularly in wastewater
and wastewater-impacted water bodies. Consideration of fluctuations in endemic levels of disease
in a community has the potential to strongly affect the interpretation of a risk assessment; therefore,
the characterization of exposure should be explicit in terms of whether endemic or epidemic
conditions, average or peak flow events, or specific events are to be evaluated, and a justification
for that decision should be provided.
The need to measure pathogens at low densities and the episodic nature of pathogen loading are
limitations of direct pathogen monitoring or the exclusive use of pathogen data in MRA exposure
assessments. For some of the pathogens of interest in waterborne exposures, very low numbers of
organisms can result in a high probability of infection. Despite advances in microbial monitoring
and particularly the use of molecular methods, detection of pathogens at levels of interest may not
be possible for some of the pathogens of concern or in some of the water matrices of interest (NRC,
2004). Difficulties in direct measurement of pathogen densities are particularly acute in drinking
water treatment processes whose intended effect is reduction of pathogens to very low densities.
These difficulties suggest the following two strategies for estimating exposure to pathogens:
• using distributions of pathogen densities in circumstances when relatively high densities
are known to be present along with modeling to determine pathogen densities at a point of
interest, or
• basing exposure on a surrogate measure of microbiological quality.
Fecal indicator organisms may be used in in development of criteria and assessment of the safety
of waters used for recreation. Development of criteria values for fecal indicator organisms is
beyond the scope of this document. The following information is presented to show how fecal
indicators are used as surrogates for the occurrence of pathogens. Epidemiological studies have
been used to establish relationships between fecal indicator organism density and human health
risk for sites with known fecal pollution sources (Parkhurst et al., 2007). Here, it is accepted that
54
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
waters from specific fecal pollution sources pose a characteristic risk to exposed persons. Given
this assumption, the fecal indicator organism density indicates the extent to which fecal pollution
from the specific source is present and is related to the risk of illness or infection at the site. Health
effects relationships based on indicator densities may be used directly in MRA illness estimation
when the fecal pollution source at the site of interest is the same as that for which the health effects
relationship was developed and with the understanding that the health effects relationship depends
on the indicator density, not the ingested indicator dose. Alternatively, the indicator density and
expected adverse health effects may be used to estimate pathogen densities in a known fecal
pollution source, as in a "reverse QMRA" process (U.S. EPA, 2014; Soller et al., 2014; Soller et
al., 2010a).
The second potential role for indicator organisms in QMRA is in characterization of process
efficiency, as in the use of indicator organisms in assessment of drinking water treatment risks.
For example, total coliforms do not pose a health hazard and are not related to a specific fecal
pollution source; however, they are relatively abundant at stages of the drinking water treatment
process where pathogens, if present at all, may be present at non-detectable densities. Total
coliform removal in unit processes may be characterized by a distribution or a range of values.
Comparison of total coliform removal for a specific unit process at a given time to the range of
removals for the process provides an indication of how well the process is operating. Because
removal rates differ among microorganism, the removal of total coliforms cannot be assumed
similar to that of pathogens. Rather, they may indicate where, within a removal range for a given
pathogen, the unit process performance may lie or whether water treatment failure has occurred
and whether a revised performance range should be used.
3.1.2. Where Do Pathogens Occur in the Water Body?
Spatial distribution of pathogenic microorganisms can differ depending on the microorganism and
on the properties of the water matrix. If pathogen occurrence fluctuates over time, then the degree
of clumping, aggregation, and clustering may also change as water parameters change. Unlike
chemicals, pathogens are particulates and may stick to each other or to sediment and other particles
(Gerba et al., 1991). The size and nature of particles will influence suspension and settling in
different hydraulic conditions. Therefore, particles that carry pathogens may be distributed within
a water body in an uneven (heterogeneous) manner. For waterborne pathogens, niche is relevant
for "free living" species. Pathogens may thrive in open water, sediments, or other ecologically
defined spaces. Non-human reservoirs may also be an important part of the pathogen ecology of a
particular pathogen. If there is appreciable survival, persistence, or amplification in non-human
species, then those sources of contamination of water may need to be considered during the
occurrence assessment. This includes animals (wildlife and domestic) as well as other
microorganisms. Survival, persistence, and amplification can differ in different microclimates
within a water body, and should be considered factors that influence where pathogens occur.
Pathogens in sediments may be resuspended in the water column due to changes in flows
associated with precipitation, runoff, tides, and currents.
55
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.1.3. What is the Level of Pathogens in the Water Body?
Characterizing pathogen occurrence relies on measuring the density in the environment and
correlating density with spatial and temporal patterns in the environment, such as niches, seasons,
weather events, and human-related activities. There are several difficulties that are commonly
encountered when measuring pathogen levels in water samples. Because microorganisms tend to
clump and aggregate (heterogeneous distribution), replicate samples can yield measurements that
differ substantially (even by orders of magnitude) (see Section 3.1.4 below).
The ability for a pathogen to survive and also remain infectious in a water body is dependent on
both pathogen characteristics and environmental factors. Pathogen-specific characteristics include
but are not limited to, genetic strain variations, the growth conditions the pathogen experienced
before entering the water body, duration in the environment, protective states, and VBNC states.
Environmental factors include but are not limited to, temperature, pH, turbidity, nutrient levels,
osmotic conditions, ultraviolet light exposure, predation, and interactions with other living
organisms. For example, and as noted previously, amoebae may be reservoirs that may contribute
to the survival, persistence, and/or amplification of environmental pathogens. Fate and transport
modeling can provide plausible scenarios and estimates of how microbial densities can change
over time as they move through the aquatic environment.
Pathogen occurrence patterns will also be affected (but not necessarily in a similar manner) by the
presence of control strategies and treatment processes (either wastewater or drinking water
treatment depending on the context). Mitigation strategies may involve improving existing control
processes or adding new control measures, which can be modeled in the risk assessment.
Discussion of the sources of microbes may be helpful in characterizing occurrence patterns. Some
commonly considered sources include wastewater treatment plant effluent, some industrial
effluents, leaking septic tanks, urban runoff, agricultural runoff, animals (e.g., livestock, domestic,
wildlife), and environmental niches (e.g., sediments, aquatic plant life). Densities of pathogens
vary in untreated sewage based on the level of shedding in the contributing human population and
densities in treated sewage vary based on levels before treatment and efficacy and type of treatment
processes. Differences in contributing populations can result in orders of magnitude differences in
microbial levels in sewage (Gerba et al., 2008).
Pathogens occur at different densities in water and wastewater treatment processes, whose primary
function is reduction of the densities of pathogens and other contaminants. Microbial risk
assessment provides a framework in which the densities of pathogens, from source to exposure,
may be estimated. This ability of MRA is important, given the variability in occurrence and density
of pathogens in different source waters, the variability in performance of unit operations, the serial
nature of water and wastewater treatment, and the dependence of performance of unit operations
on prior operations.
The technical literature contains numerous studies characterizing the range of unit process
operation removals for different pathogens and different operations (e.g., Betancourt and Rose,
2004, for removal of Cryptosporidium and Giardia in drinking water treatment processes). Other
studies report the overall removal of pathogens or their surrogates for entire water treatment
processes (e.g., Castro-Hermida, 2008, for removal of Giardia and Cryptosporidium from multiple
56
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
drinking water treatment plants; or Kistemann et al., 2008, for removal of multiple pathogens or
their surrogates in drinking water treatment plants). These studies demonstrate both the potential
for generating general ranges of removals that might be expected in different unit operations and
the variability of the efficiencies between plants and for different water matrices.
Numerous QMRAs have been conducted of drinking water treatment processes, several of which
are used as illustrations in other sections of this report (Olivieri et al., 1999; Westrell et al., 2003;
Pouillot et al., 2004; Petterson et al., 2007). An additional study (Smeets et al., 2008) illustrates
the use of ranges or distributions of removals for unit process operations within a model capable
of predicting risk for a particular treatment train. In that study, yearly Campylobacter illness risks
from consumption of drinking water produced at a plant employing filtration and ozonation were
estimated. Several parameters in the model constructed for this process are variable, including raw
water Campylobacter density, removal efficiency for filtration, and removal efficiency for
ozonation. Smeets et al. (2008) evaluated multiple approaches for developing distributions to
characterize the removal processes. Removal data for filtration and ozonation were based either
on paired data (influent and effluent densities of the unit operations) taken at the same day, or data
paired based on their rank. Additionally, both parametric and non-parametric distributions were
used for removal efficiencies, with the distributions based on the removal efficiency data. The
authors evaluated Weibull, gamma, and log-normal distributions and determined that the raw water
Campylobacter density data, filtration removal efficiency data, and ozonation removal efficiency
data were all fit best by log-normal distributions. The authors concluded the following from their
study:
• the rank method provided much better agreement between predicted and observed
Campylobacter densities for the unit operations;
• use of parametric representations of the removal efficiencies of the unit operations is
superior to non-parametric representations in this instance because parametric models
allow for the occurrence of rare events, although use of non-parametric methods reduces
the effect of distribution choice on risk estimation; and
• use of QMRA in assessing drinking water risk complements pathogen monitoring,
especially given the low densities of pathogens occurring in parts of the water treatment
process.
The outcome of the "occurrence" section of the process is an evaluation of all relevant factors
pertaining to the occurrence and distribution of the pathogen. Several tools and databases for
evaluation of occurrence, which may be useful for MRA exposure scenarios, are summarized in
Table 7.
Table 7. Tools and Databases for Evaluation of Occurrence
Tools
Reference(s)
EPA STORET and WQX
http: //www. e pa. a ov/sto ret/a bout.html
The STORET Data Warehouse is EPA's repository of the water quality
monitoring data collected by water resource management groups across
the country. Data can then be re-used for analysis. WQX is the framework
by which organizations submit data to the Warehouse.
57
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 7. Tools and Databases for Evaluation of
Occurrence (continued)
EPA Basins Program
http://www.EPA.gov/OST/BASINS/
A multi-purpose environmental analysis system that integrates a
geographical information system, national watershed data, and state-of-the-
art environmental assessment and modeling tools into one convenient
package. Download at
http://www.EPA.gov/waterscience/basins/basinsv3.htm.
This release includes additional links to water quality models as well as a
new data user interface tool with access to national data layers.
EPA EMPACT
(Environmental Monitoring for
Public Access and
Community Tracking) Study
EPA Information Collection
Rule (ICR) and Supplemental
Surveys (provides
summarized data on
Cryptosporidium only)
General information is available at:
http://www.EPA.aov/nerl/news/forum2003/water/brenner poster.pdf and
http://www.EPA.gov/ORD/NRMRL/pubs/625rQ2017/beaches html/chapterl
html.
Location-specific EMPACT studies are available on the Internet.
Overview with links for further information is available at:
http://www.EPA.qov/enviro/html/icr/ and
http://www.EPA.gov/safewater/icr.html.
EPA's Unregulated
Contaminant Monitoring
Program
Safe Drinking Water
Information System (SDWIS)
Statistical method: Markov
Chain Monte Carlo (MCMC)
simulation for modeling
environmental pathogen
densities in natural waters
EPA uses the Unregulated Contaminant Monitoring program to collect data
for contaminants suspected to be present in drinking water, but that do not
have health-based standards set under the Safe Drinking Water Act
(SDWA).
http://water.epa.gov/lawsreqs/rulesreqs/sdwa/ucmr/
SDWIS—Federal (SDWIS/FED) version is U.S. EPA's national regulatory
compliance database for the drinking water program. It includes information
on the nation's 170,000 public water systems and violations of drinking
water regulations.
• Access Drinking Water Information Online (through summary pivot
tables, Envirofacts, or direct connection to the mainframe)
htt p ://www. E P A. g ov/safewate r/d ata/g etd ata. htm I.
• SDWIS/FED Website (information for users who work with the
database) http://www.EPA.gov/safewater/sdwisfed/sdwis.htm.
Crainiceanu et al. (2003) "Modeling the United States national distribution of
waterborne pathogen concentrations with application to Cryptosporidium
parvum."
AWWARF report on effects of http://awwarf.org/research/topicsandproiects/execSum/488.aspx.
meteorological events "Rainfall events and other watershed perturbations, especially those during
(Cryptosporidium only) spring runoff, pose the greatest risk for causing waterborne
cryptosporidiosis."
USDA agricultural runoff
models (hydromodels)
http://wmc.ar.nrcs.usda.gov/technical/WQ/modeldesc.html
Information and links on 15 Natural Resources Conservation Service
models related to water and agriculture.
USDA/ARS HYDRUS models http://www.ars.usda.gov/Services/docs. htm?docid=8910
Simulates water flow and solute transport in a two-dimensional variably-
saturated medium (graphical user interface).
Surfrider
The Blue Water Task Force is the Surfrider Foundation's volunteer-run
water testing program.
http: //www, s u rfri d e r. o rg/b I u e-wate r-tas k-fo rce
58
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.1.4. Interpretation of Analytical Methods
It is important to document measurement techniques and their capabilities and limitations carefully
so that scientifically defensible decisions can be made about integrating or not integrating results
from different studies. For example, there may be differences in the relative ratio of infective,
viable, and nonviable Cryptosporidium oocysts in the samples taken from source water versus
laboratory generated sources of oocysts used in human trails. Young, freshly harvested oocysts are
typically used for the human trials and are most often used for disinfection studies, yet when
conducting environmental sampling, the age of the oocysts in any given sample can vary widely,
as could the distribution of viable and infective oocysts. This difference leads to part of the
uncertainty in the interpretation of the data, and thus assumptions regarding how laboratory
generated data are applied to environmental scenarios need to be clearly stated in exposure
assessment.
For data sets that involve laboratory methods issues related to sensitivity, specificity, limit of
detection, sampling method, and sample size should be examined. Their effects on the risk
assessment assumptions should be discussed. Culture-based approaches rely on growing the
organism in question, isolating it in pure culture, and characterizing the morphological,
biochemical, physiological and other traits. However, not all microorganisms are culturable, such
as VBNC, and some parasites and viruses.
Molecular methods can allow serotyping and determination of virulence traits. The risk assessment
documentation should discuss any relevant information related to analytical methods, including
error bars if possible. For example, Dupont et al. (1995) included standard deviations for the actual
numbers of C. parvum oocysts given experimentally to subjects versus what the intended dose
was. For an intended dose of 30 oocysts, the actual dose was 34 ± 3, for 100 oocysts the actual
dose was 108 ± 22, and for 300 oocysts, the actual dose was 313 ± 24. When new methods are
introduced, such as cell culture techniques to enumerate the number of infectious oocysts in water
samples, they may allow refinement of the dose-response relationship (Slifko et al., 1997, 1999,
2002). This refinement can then be used to improve future exposure assessments.
EPA's LT2 has a discussion of EPA Methods 1622/23 performance and how level of performance
influences the occurrence data for the Information Collection Rule (ICR) (U.S. EPA, 2006a). To
estimate how these errors would affect the assignment of systems to categories (bins) with different
levels of occurrence, EPA constructed a Monte Carlo model that dealt with the error components
in the following manner:
• Finite volume assayed—the model defines the number of oocysts present in a 10 L volume
as a Poisson random variable, whose mean is the product of measurement recovery, volume
assayed, and density at the time of sampling.
• Finite number of samples—true density varies over time as a random variable. Density is
modeled to vary in such a way that its natural logarithm is normally distributed with
standard deviation 1.762. This value was selected based on Bayesian analysis of survey
data and on expert opinion that at any given site the Cryptosporidium density would vary
within a three order of magnitude density range 95% of the time (i.e., 2.5% of the time the
density would be less than X, and 2.5% of the time the density would exceed 1000X).
59
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• Variable recovery—based on laboratory performance in ICR Supplemental Survey, EPA
assumed for the model an average recovery among all laboratories of 40% with a relative
standard deviation of 50%. Recovery is modeled as a Beta random variable with parameters
(a, P) = (2, 3). Mean recovery is therefore a/(a + P) = 2/(2 + 3) = 0.4. The standard deviation
of recovery is 0.2, which is half the mean recovery.
3.2. Exposure Analysis
An exposure scenario summary can be a short narrative description of how an individual is exposed
to a hazard. A more formal exposure scenario provides additional detail about the range of
exposures that are considered in the risk assessment. The EPA Exposure Factors Handbook (U. S.
EPA, 1997a, 2011) is the Agency-wide resource for building exposure scenarios for chemical
hazards. It can also be consulted for data that may apply to infectious disease hazard exposures.
The interagency Microbial Risk Assessment Guideline has considerable detail on exposure models
(U.S. EPA/USD A, 2012). Elements that should be evaluated for inclusion in exposure assessment
and are used to define the exposure scenario scope are presented below (adapted from ILSI, 2000):
• identification of media;
• routes of exposure (including secondary transmission);
• units of exposure (period of relevancy to characterize dose, includes magnitude, duration
and frequency);
• temporal nature of exposure (whether single or multiple exposures);
• spatial nature of exposure; and
• behavior of exposed population.
A summary of parameters used in an example EPA exposure analysis is presented in Text Box 4.
Text Box 4. Exposure Analysis for the Long Term 2 (LT2) Enhanced Surface Water Treatment
Rule (Source: U.S. EPA, 2006a).
Media - the media considered was surface water used as a source for drinking water from filtered systems
and unfiltered systems.
Units of Exposure (magnitude, duration, frequency) -# of oocysts per ingestion volume (magnitude),
per day (duration), for an annual frequency of 21 days (frequency)
Ingestion Volume - individual consumption mean = 1.071 L per day; (LT2 page N-16)
Frequency of Exposure - annual days of exposure = 21 days (range 15-30 days); (LT2 page N-16)
Routes - only oral ingestion (dermal and inhalation not considered)
Occurrence - although seven sources of data are cited, the main source of data was the ICR. Information
in the survey included water quality parameters, such as turbidity and pH, along with process units in the
plant and their sequences. The ICR survey is the most comprehensive database available for large
systems. EPA identified an expected level of laboratory analytical method performance based on results
with EPA Methods 1622/23 in the Information Collection Rule Supplemental Surveys, and also established
the mean as the appropriate statistical measure to classify source water Cryptosporidium oocyst levels.
The use of the arithmetic mean is advantageous for several reasons. The mean can be estimated more
60
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 4. Exposure Analysis for the Long Term 2 (LT2) Enhanced Surface Water Treatment
Rule (Source: U.S. EPA, 2006a).
(continued)
reliably than other statistical measures. For example, with a limited number of samples, the confidence
interval around the mean is substantially narrower (i.e., less uncertain) than for a 90th percentile estimate.
Defining a treatment trigger based upon a maximum value would be much less reliable than basing it on
a computation involving multiple values, due to the uncertainty associated with any single sample
measurement. The mean density also directly relates to the average risk of the exposed population and,
therefore, provides a good measure for indicating relative risks from one site versus another (e.g., doubling
the source water average density corresponds to about a doubling of the risk, assuming the same level
of treatment at both sites). In contrast, the median would not be an informative or appropriate
characterization because of the large numbers of non-detection measurements expected to occur,
resulting in a large number of sites with median values equal to zero. The median would fail to distinguish
differences between sites that had half or more of their measurements as zero and positive values for the
remainder, and those that truly had measurements of zero. The input used to calculate infectivity included
the following parameters: Density of oocysts = 0.0995 (range of 0.067 to 0.132 oocysts/L); percent
infectious = 20% (range 15% to 25%)
Spatial temporal characterization - the ICR survey was conducted from 1997 through 1998. It consists
of 18 months of data collected from all large systems in the United States serving over 100,000 people.
Other characteristics of the exposed populations - EPA considered as examples, Portland, ME,
Portland, OR, Tacoma, WA, San Francisco, CA, and New York, NY populations, including the predicted
number of people living with AIDS. EPA also considered small (< 10,000), medium (10,001 to 100,000),
and large (>100,000) population size categories for analysis.
3.2.1. Identification of Media
In this document, identification of media refers to the specific water sources being considered in
the risk assessment. Numerous water-related exposures can be considered using the tools, methods,
and approaches described. For example, exposures through the ingestion of drinking water,
ingestion of water during recreational activities (swimming), water reuse, inhalation of aerosolized
biosolids particles, and ingestion of soil amended with biosolids are all exposure scenarios that are
compatible with the tools in this MRA Tools document. To illustrate this point, MRAs for
biosolids-related exposures have been conducted in the same manner as described herein for water-
related exposures (Eisenberg et al., 2004, 2008; Gale, 2003, 2005), and MRAs for recycled
wastewater in agricultural irrigation, swimming, and landscape irrigation practices have also been
reported (Asano et al., 1992; Tanaka et al., 1998).
3.2.2. Routes of Exposure
The primary route of exposure19 considered in a water-based MRA is usually ingestion, but can
include other routes of exposures such as inhalation and dermal contact. Inhalation exposures may
be significant for some microorganisms (spore-forming bacteria, Legionella, Mycobacteria, and
19 Route of exposure refers to how the pathogen comes in contact with the vulnerable host receptor cells that support
infection (e.g., inhalation, dermal contact, oral), whereas source of exposure refers to the physical matrix that carries
the pathogen (e.g., air, water, food, soil).
61
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
some viruses). Similarly, dermal exposures (through intact skin or, more frequently, open cuts and
scratches) may be important for some scenarios (e.g., Pseudomonas aerugenosia, V.
parahaemolyticus, V. vulnificus). If data for a route of exposure are not available, it would not be
possible to quantitatively evaluate within the risk analysis, however, it remains a major uncertainty
and would need to be evaluated qualitatively in the risk characterization. For example, quantitative
consideration of inhalation route would be possible for the enteric virus coxsackievirus, because
respiratory pathway dose-response data are available (Couch et al., 1965).
Many water-based MRAs focus on exposure through drinking water and recreational activities,
such as swimming and other activities, where ingestion of water is likely. However, there are also
other potentially important routes of exposure that can also be of interest for which less data are
currently available, such as exposures to recycled water, biosolids, as well as secondary and non-
contact activities such as boating and fishing.
Routes of exposure should be discussed in the risk assessment documentation, including which
routes are considered and which routes are not, as part of the scope of the risk assessment.
Uncertainty in the exposure assessment should also be discussed for the routes that are not modeled
or assessed.
3.2.3. Units of Exposure
The unit of exposure is generally a "dose", which is a specified number of pathogens that a person
or population is exposed to. It is reasonable and convenient to consider exposure in terms of a
magnitude, duration, and/or frequency. For example, the magnitude could be the number of plaque
forming unites (PFU) per L, the duration and frequency could be 2 L per day for 360 days. For
pathogens, exposure events are usually measured on a time scale of daily or shorter intervals.
Historically, exposure time frames for pathogens have been based on the assumption that short-
term (event-based) exposures are most relevant (e.g., per swimming event for recreational
activities; per day for drinking water uses) rather than lifetime exposures. In contrast to chemical
contaminants in water, the adverse health effects associated with human exposure to waterborne
pathogens have been best documented for event-related (short-term, single exposure) rather than
chronic exposure over extended periods of time. These short-term exposure timeframes have been
used because infection requires that one or more pathogens be ingested and that at least one of the
ingested pathogens establishes itself in or on cells somewhere within the GI tract of the host
(Teunis and Havelaar, 2000). If no organisms have been ingested or none of those ingested succeed
in passing all of the host barriers, infection does not occur. Note that short-term exposures do not
necessarily imply that only short-term or minor adverse human health effects occur; for example,
illnesses from some pathogens (e.g., E. coli 0157:H7) can be severe and/or long-term or produce
sequelae (Rangel et al., 2005).
Although there are chemicals for which a single exposure model is appropriate, such as teratogens
that cause developmental defects or nitrate that can cause infantile methemoglobinemia, many
cancer-causing chemicals exhibit increasing risk as duration of exposure lengthens (i.e., exposure
over multiple years). This is because some chemicals can accumulate in the human body, and even
for chemicals that can be purged from the body, the damage they cause may not be readily
repairable. Therefore, damage may accumulate with each subsequent exposure. Although
62
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
accumulating damage is not necessarily an outcome of infection by pathogens, there are pathogens
that generate toxins that behave similarly to chemicals in this respect. Therefore, it may be
important to consider the mechanism by which pathogens cause illness symptoms when
considering whether short-term, event-based exposures are the predominant relevant exposure
pathway. Reinfection may also increase the potential for development of autoimmune disease. For
example, the autoimmune disease reactive arthritis can be triggered by the following pathogens:
Chlamydia, Salmonella, Shigella, Yersinia, Brucella, Leptospira, Mycobacteria, Neisseria,
Staphylococcus, and Streptococcus (Girschick et al., 2008).
The exposure timeframe basic unit should be discussed within the context of the exposure
scenarios. It may be difficult to determine if recurring exposure events are completely independent
or not. For example, MRAs for drinking water commonly assume that all water consumed over
the course of a single day is considered to be one dose, and consumption on subsequent days are
considered to be independent events (U.S. EPA, 2006a). When exposures are considered to be
completely independent (e.g., consumption on different days) the cumulative risk can be calculated
as the result of independent repeated daily risk events. At the other end of the scale, when
exposures are considered to be completely dependent, the doses can simply be added and treated
as a single risk event (e.g., add total volume of water consumed through each serving of water over
the course of a day). However, little data are available to describe the mechanisms of pathogen
infection processes to support the assumption that all consumption within a specified period
constitutes a single exposure event. Instead, the 24-hour timeframe is used for convenience and
because it is a biologically reasonable timeframe for human digestive processes. The
interdependence of exposure events may be important for some pathogens and may vary depending
on characteristics of the host, the pathogen, and event specific conditions such as delivery matrix.
Although exposure events may have varying interdependence, the assumption of independent
exposure events as a default assumption is commonly used in MRA (Regli et al., 1991). Risk
assessment tools for considering multiple exposures to a given pathogen are currently not
sufficiently developed to recommend any specific tools.20 The uncertainties associated with
defining the unit of exposure should be discussed in the risk assessment.
Some examples illustrating how exposure can be computed are presented in Text Box 5 below.
Note that various other uses of water, such as ingestion of crops irrigated with recycled water, and
aquaculture, could have different units of exposure than those shown in Text Box 5. Soller et al.
(2007) summarize the available literature and data that can be used for exposure to pathogens from
three uses of recycled water, including body contact recreation, crop irrigation, and landscape
irrigation. In all of those cases, exposure is through an ingestion route of exposure and is specified
as volume ingested per day.
20 Methods for considering exposure to multiple pathogens are also lacking.
63
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 5. Examples of Time Units Associated with Exposure
• For recreational exposures, the time period associated with an exposure may be per
swimming day or hours spent in the water. For example, risks from recreational exposures
may be calculated using estimates of volume of ambient water incidentally (inadvertently)
ingested over a given length of time (e.g., 50 mL/hour).
• For drinking water, the time unit of exposure is typically a single day, for example, 3 L per
person-day (which represents the 95th percentile for consumption from EPA's Exposure
Factors Handbook) (U.S. EPA, 2011).
¦ For shellfish, the associated exposure unit can be a meal or serving (without including how
much constitutes a meal), number of shellfish consumed per meal, or weight of shellfish
consumed per meal.
3.2.3.1. Volume Ingested
Quantitative data have been developed to characterize the volume of water that individuals and/or
populations ingest through drinking water (U.S. EPA, 1997a, 2002d, 2006a, 2008, 2011) and
recreational activities (Dufour et al., 2006; Schets et al., 2011). The data that are available for
characterizing the volume of drinking water ingested has received the most attention and is by far
the most comprehensive. The available data for characterizing the volume of water ingested during
recreational activities derives from studies conducted in swimming pools. Data characterizing
other routes of exposure are much more limited (Dorevitch et al., 2011). In fact, the limited data
for assessing exposure (with the exception of that for drinking water) is currently an important
limitation in MRAs.
Dufour et al. (2006) found that during swimming events lasting at least 45 minutes, children (<18
years of age) ingested significantly more water (average = 37 milliliters [mL], range = 0-154 mL,
N = 41) than adults (average =16 mL, range = 0-53 mL, N = 12). The raw data provided by
study authors (Dufour, personal communication) was fit to a statistical distribution. The best-fit
distribution is lognormal with log mean (2.92) and log standard deviation (1.43). The median value
of this distribution is 18.6 mL. Schets et al. (2011) also evaluated ingestion during swimming
events. In their survey conducted in the Netherlands, 75,000 inhabitants representing the general
Dutch population were asked to report the volume of water they swallowed as an estimated number
of mouthfuls in four classes: (1) no water or only a few drops, (2) one to two mouthfuls, (3) three
to five mouthfuls, and, (4) six to eight mouthfuls. Table 8 shows the self-reported results for the
8,000 questionnaires that were competed.
Table 8. Average Volume Water Swallowed (mL) per Swimming Event
(Schets et al., 2011)
Setting
Age <15 years old
Age >15 years old
Men
Women
Swimming pool
51 mL
34 mL
23 mL
Freshwater
37 mL
27 mL
18 mL
Seawater
31 mL
27 mL
18 mL
64
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.2.3.2. Temporal Nature of Exposure
The temporal nature of exposure includes the duration and frequency of the exposure. The duration
is how long the exposure happens. For microbes duration is usually per event (e.g., swimming),
per serving (e.g., food), or per day (e.g., drinking water consumed over a day). Frequency may be
once, or compounded events over a year or a lifetime. For example The U.S. Census Bureau
estimates 20 million (46 %) people (Age: 7-17 years) swim more than 5 times per year, and that
63 million (21%) people (Age: >17) swim more than 5 times per year (U.S. Census Bureau, 201 la,
b).
Both endemic and episodic exposures are possible and MRA exposure assessments can be
developed such that both of these exposure types are adequately modeled. For example, in a
retrospective analysis of illnesses related to a water treatment system and distribution network,
Westrell et al. (2003) estimated and compared illness rates for Cryptosporidium, rotavirus, and
Campylobacter for customers of water from a conventional treatment plant using chemical
disinfection to a hypothetical plant employing membrane filtration. Treatment processes were
characterized by distributions (as opposed to use of non-parametric characterizations) and
incidents sporadically included in the system model were (1) prolonged filter ripening periods
every second day for one month each year (2) chlorine failure, (3) cross connections in the
distribution system, and (4) pollution intrusion during low pressure transients. Monte Carlo
simulation of the treatment and distribution process indicated that a major portion of potential
infections occurred during normal plant and distribution system operation. Illnesses occurring at
low rates and endemically may be difficult to detect and attribute to drinking water in public health
surveillance. A counterpoint to the importance of endemic exposure is presented in an analysis of
the 1993 Milwaukee cryptosporidiosis outbreak (Eisenberg et al., 2005). In that study, the
combination of a drinking water treatment plant failure and ease of disease transmission led to a
very large outbreak. The analyses demonstrate an apparent feedback between the production of
infectious oocysts and the generation of unsafe drinking water. Here, the importance of the
epidemic nature of the exposure is more important than endemic exposure. These examples
illustrate the importance of considering and modeling both episodic and routine exposures and the
potential importance of dynamic modeling as a component of MRAs.
3.2.4. Spatial Nature of Exposure
The spatial nature of the exposure for waterborne pathogens is suggested by the Clean Water Act
designated use. For example, exposure during recreation (swimming, surfing, etc.) occurs while
people are on or in the water and is geographically confined to where the water body is located.
Exposure through drinking water is limited to the area that the public water supply serves, unless
water is transported (e.g., transported by truck to a neighboring community to fill a special need).
People may also move between water districts throughout the day as they travel for work or other
reasons. Exposure through consumption of raw or partially cooked shellfish can also occur at
locations removed from the water body from which the shellfish originated. Complex spatial
distributions of exposure and the exposed people can make characterizing exposure patterns
difficult; however, those patterns should be analyzed for their impact on the exposure assessment.
65
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
3.2.5. Behaviors of Exposed Population
Subpopulations can be defined by their susceptibility (e.g., intrinsic factors such as immune status
or related factors) or by behaviors (extrinsic factors) that may cause them to be highly exposed
(e.g., lifeguards, surfers, tri-athletes, other competitive swimmers versus casual bathers). In
particular, plausible extreme behaviors should be noted, and the discussion should clarify to what
degree individuals exhibiting those behaviors are addressed by the exposure scenario. Behaviors
can also influence the routes of exposure and the spatial and temporal nature of exposure.
Specialized exposure scenarios, such as occupational exposures, can also be developed. This type
of exposure consideration would most likely require that the risk assessment include both scientific
and regulatory considerations. Risk assessments limited to occupational exposures to water that
have caused infectious disease outbreaks are not common. However, there may be some
occupations that have frequent water exposures in which a MRA may be of interest (e.g.,
lifeguards, wastewater treatment plant workers).
3.3. Exposure Profile and Linkage between Exposure Assessment
and Other MRA Components
The exposure profile is a distillation of the most important information and analyses that are
conducted during the exposure assessment. Each of the components of the exposure assessment
describes the data and information that are available on that specific topic (i.e., occurrence,
identification of media, units of exposure, routes of exposure, spatial and temporal nature of
exposure, and characterization of exposed population). The exposure profile is a compilation
summary of those data and analyses that will be used in conjunction with the human health
assessment for estimating risk.
In the same manner as the problem formulation documentation, the formality of the exposure
profile documentation can vary depending on the needs of the EPA Office conducting the
assessment.
Consistent with the recommendations from the EPA-ILSI framework (ILSI, 2000) regarding the
iterative and fluid nature of risk assessment, the exposure profile (as well as the host pathogen
profile, as described in Section 4.3) should be critically evaluated by the risk assessors and
managers to determine if the problem formulation component needs to be revisited and refined
based on the availability of relevant data presented in the exposure profile. The linkage between
the exposure assessment and the problem formulation phase of MRA is iterative.
3.3.1. Exposure Estimation
In the exposure profile the exposure estimate is presented. It serves as the critical linkage from the
exposure assessment to the health effects assessment. Although the quantity and quality of data
that will be available for any particular risk assessment will necessarily vary, the exposure profile
provides critical input for the risk estimation.
66
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
The exposure estimate can include, as appropriate, a qualitative and/or quantitative evaluation of
the magnitude, frequency, and patterns of exposure to a pathogen for each of the populations in
each of the scenario(s) of interest. The exposure profile should also identify the specific
assumptions that are made during the exposure assessment and uncertainties that are thought to be
important for the risk assessment. Assumptions made during risk assessment are based on
scientific judgment and consider the scope of the risk assessment as defined during problem
formulation and planning and scoping.
3.3.2. Exposure Description
A description of the uncertainty associated with each element of the exposure assessment should
be provided to the extent that it is reasonable and possible. Uncertainty analysis in drinking water
MRAs has shown that exposure assessment can be a primary factor driving the output risk
distributions. Thus, the description of uncertainty in the exposure profile is an important aspect of
MRA.
The description of the assumptions and uncertainties related to exposure should be sufficient to
provide an appropriate level of insight into the strengths and weaknesses of the assessment for
evaluation during risk characterization. For example, Teunis and Havelaar (1999) used the
exposure profile section of their Cryptosporidium in drinking water risk assessment to summarize
the quantitative information on density of oocysts in raw water, recovery efficiency of the detection
method, reduction by treatment, and amount of finished water that is consumed. The distribution
type (e.g., negative binomial, beta, 2 choice binomial, log-normal) selected for each parameter as
well as median and 95% range are presented in a table. A description of the Monte Carlo
calculations and graphical as well as narrative discussion of the Monte Carlo simulation is also
included. In this example, the exposure profile highlighted several important observations that
affected the subsequent aspects of the MRA, including the following:
• Correction of oocysts counts for viability had little effect on the distribution of the density
of oocysts in river water.
• The two distributions for river water and storage overlap, so that occasionally the treatment
plant will be confronted with relatively high oocysts loads, even after passage of three
reservoirs in series.
• Although treatment (physico-chemical) has a marked effect on oocysts densities
(frequency distribution shifted by 4-logs), there is still a small probability of high densities
of oocysts in treated water that is related to occasional reduced performance of the
treatment plant.
As another example, Soller et al. (1999) used the exposure profile section of their drinking water
risk assessment for rotavirus to summarize the exposure parameter assumptions. Below is a
summary of the salient assumptions that were used in that MRA:
• The exposure model assumes that there is no upstream contamination or upstream
contamination has been diluted to the point that the effects are negligible.
• The exposure model assumes that there are no animal (agricultural and grazing) sources of
human infectious rotavirus.
67
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• Acute-phase infected humans engaged in water recreation near the drinking water intake
could be a significant source of rotavirus. However, this was not considered significant
because site specific data that would be required to add this parameter is not available and
body contact water recreation is likely to be insignificant during winter months, which is
the time of year when rotavirus infections are most significant.
• Wastewater treatment plant effluent is the most important source of rotavirus and is
assumed to have undergone secondary treatment with chlorine, contribute 5% of river
volume, and contain 1 to 375 focus forming units/L of rotavirus.
• Rotavirus decay in source water results in 99% reduction between 3 and 30 days.
• Chlorine residual provides between 0 and 1.0-log reduction in rotavirus between the
drinking water treatment facility and the tap.
68
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
4. Human Health Effects and Dose-Response
The human health effects assessment consists of the technical evaluation of data related to host
characterization, evaluation of human health effects, and quantification of the dose-response
relationship for contaminants in water. The problem formulation phase of the MRA precedes the
human health effects assessment and may partially address many of the issues to be evaluated in
the effects assessment, but the effects assessment is generally more detailed and quantitative than
the problem formulation phase. The two components of the risk assessment, which may be
conducted in parallel, are the characterization of exposure and the characterization of human health
effects, including the dose-response relationship.
The output from the human health effects assessment is a host-pathogen profile that provides
qualitative and or quantitative descriptions of the nature of the illness (e.g., morbidity and mortality
statistics) and quantitative dose-response analyses for the scenario(s) developed during problem
formulation. This chapter is divided into general health effects and dose-response modeling.
4.1. Human Health Effects Overview
The important health effects elements that can be considered during risk assessment are
summarized below and discussed in the following sections, including (adapted from ILSI, 2000;
see Table 1):
• duration of illness;
• severity of illness;
• morbidity, mortality, sequelae (long-term effects) of illness (including acute and chronic
effects); and
• secondary transmission and immunity.
4.1.1. Duration of Illness
Duration of infectious disease illness is usually expressed in days. Duration can often be divided
into duration of incubation (incubation period), duration of infection, duration of infectiousness
(duration that host excretes the pathogen), and duration of disease symptoms. The scope of the risk
assessment will determine the extent to which detailed information is required for each of these
factors. If secondary transmission is expected to be significant, then the incubation period and
duration of infectiousness may be important determinants of the magnitude of disease occurrence.
The incubation period for a disease is the interval from a person's exposure to the pathogen to the
time they develop symptoms or clinical illness (or the period between the dose and some
measurable response, such as shedding of the pathogen or serological response). Different diseases
have different incubation periods, and this information can be used to help identify the pathogen
responsible for a particular outbreak. Chronic sequelae from infections include all persistent and
future effects on health (disability, recurrence of infection) and may extend for years after acute
infection (see Section 4.1.3 below). A brief summary of some incubation periods for several
waterborne diseases is presented in Table 9.
69
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
Table 9. Typical Incubation Periods for Some Waterborne Pathogens
Pathogen
Incubation Period and Reference3
Cryptosporidium parvum
2 to10 days (average 7 days)b
Giardia lamblia
1 to 2 weeks (average 7 days)0
Shigella spp.
16 to 72 hoursd
Campylobacter jejuni
2 to 5 days (Trachoo, 2003)
Escherichia coli 0157:H7
1 to 8 days (average 3 to 4 days) (Weir, 2000)
Norovirus
24 to 48 hours (APHA, 2004)
Rotaviruses
< 4 days (average 1 to 2 days) (Aitken and Jeffries, 2001)
a Defined as time from exposure to onset of first symptoms.
b http://wvwv.cdc.qov/ncidod/dpd/parasites/crvptosporidiosis/factsht cryptosporidiosis.htm.
c http://www.cdc.aov/ncidod/dpd/parasites/aiardiasis/default.htm.
d http://pathport.vbi.vt.edu/pathinfo/pathoaens/Shiaella.html.
4.1.2. Severity of Illness
The severity of illness, morbidity, mortality, and chronic sequelae of illness are all factors that
need to be considered in the choice of health endpoints considered in the risk assessment. Severity
of illness is often difficult to quantify because disease symptoms often include subjective
descriptions. Severity of illness can be measured by more objective parameters, such as T-cell
count or other biological markers (e.g., liver function). Number of physician visits,
hospitalizations, or emergency room visits may also be used to assess severity, but these measures
have the disadvantage that they depend on the availability of such services, and cultural and social
values related to the use of medical services, and costs. Severity of infection does not necessarily
equate to severity of illness. Individuals that are infected and are able to transmit the disease, but
do not exhibit symptoms, are known as asymptomatic carriers. The length of time an individual
remains in the carrier state can vary based on pathogen and host factors. Severity of illness and
severity of infection are usually used in reference to an individual (as opposed to a population).
An individual may also exhibit varying degree of infectiousness during the course of an infection
and infectiousness between individuals can be different. Severe illness may or may not be
accompanied by severe infectiousness.
4.1.3. Morbidity, Mortality, and Sequelae
Morbidity and mortality measures can also be used to characterize disease burdens within a
population. Morbidity is a measure of the proportion of people who are afflicted with a given
disease or who display a given symptom per unit of population (e.g., per 1000 people, per 100,000
people). Mortality is a measure of the number of deaths per unit population, or number of deaths
out of the diseased population. Both morbidity and mortality are most commonly expressed as
annual rates (or rates during an outbreak).
Sequelae of illness, which are more commonly referred to as "chronic sequelae," are conditions
that occur after infection has occurred. Because chronic symptoms may be removed in time from
the acute infection, it is often harder to demonstrate a correlation between infection and symptoms.
70
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Furthermore, the type of epidemiological study design that can detect chronic sequelae (i.e.,
retrospective cohort study design) is not commonly conducted for waterborne illnesses. Section
4.1.4 includes information on chronic sequelae from infection with several waterborne pathogens
of public health significance.
U.S. Centers for Disease Control and Prevention (CDC) periodically reports estimates of the
incidence of foodborne illness calculated from total estimated illness (Mead et al., 1999; Scallan
et al., 201 la,b) and reports on waterborne disease surveillance (CDC, 1993, 1996, 1998b, 2000,
2002, 2004b, 2006b, 2008). These resources are helpful for framing the human health effects of
pathogens.
4.1.3.1. Acute and Chronic Health Effects of Microbial Contaminants
Waterborne microbial contaminants cause a range of acute health effects, including but not limited
to gastrointestinal illness, eye infections, ear infections, respiratory infections and other illnesses
described below for a set of pathogens. Symptoms such as bloating, lethargy, general malaise,
aches, malabsorption, weight loss, anorexia, and dehydration can result following infections with
Giardia and Cryptosporidium. Other symptoms include dysentery (characteristic of Vibrio
species); influenza-like symptoms such as fever, chills, headache, myalgia (e.g., Legionella);
hepatitis (Hepatitis A and E); and meningitis, which has been associated with Enterovirus,
Aeromonas hydrophila, Campylobacter, and Yersenia enterocolitica. Chronic diarrhea is typically
associated with enteric protozoa such as Cryptosporidium and Giardia, but some bacteria, such as
some adherent enteropathogenic E. coli, can also result in this condition. Infection with pathogenic
E. coli can result in kidney disease, and hypertension. People who are immunocompromised are
more likely to experience chronic diarrhea. Chronic sequelae (long-term health effects) may also
develop following infections from waterborne infections. Overviews of the public health
significance, including chronic sequelae, of several waterborne viruses, bacteria, amoebae, and
protozoa are provided below. Other resources for pathogen information include, a review of
waterborne pathogens (Craun et al., 2010), the American Society for Microbiology's (ASM)
Manual of Clinical Microbiology (ASM, 2011), and FDA's "Bad Bug Book" (FDA, 2006).
VIRUSES
Adenovirus
Adenoviruses infections most commonly cause lower and upper respiratory disease, AGI
(especially due to serotypes 40 and 41), acute conjunctivitis, pharyngoconjunctival fever, and
urinary tract infections (Enriquez and Thurston-Enriquez, 2006). Children, infants, the elderly,
immunocompromised, and other sensitive populations can be especially affected (Jiang, 2006).
Infection is most often spread via direct contact with an infected individual, fecal-oral route, or
recreational water activities. Although not thought to be the most common route of transmission,
there have been two drinking water outbreaks reported in Europe in which enteric adenoviruses
may have been a cause of acute gastroenteritis (Mena and Gerba, 2009). Adenoviruses in drinking
water may also contribute to viral infections of unknown etiology (Ko et al., 2003).
Neurological sequelae from Reye's syndrome following infections by organisms such as
adenovirus have been reported. For example, studies of Reye's Syndrome survivors have shown
71
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
sequelae ranged from severe psychomotor retardation to mild specific perceptual and/or language
impairments, IQ deficits, and chronic behavioral deficits that lasted from 6 to 18 months after
infection (Davidson et al., 1978; Brunner et al., 1979; Shaywitz et al., 1982).
Astrovirus
Astroviruses are transmitted by person-to-person contact via the fecal-oral route, and especially
via the ingestion of contaminated food or water or contact with fomites (Abad et al., 1997, 2001;
Schwab, 2006). Globally, much of the endemic- and outbreak-related cases of gastroenteritis in
children are caused by astroviruses (Walter and Mitchell, 2000; Glass et al., 2001). Infection with
astroviruses characteristically results in GI illness, most commonly self-limiting diarrhea,
vomiting, and mild dehydration. However, more severe cases are possible. Astroviruses continue
to be linked to documented waterborne disease outbreaks both in the United States and abroad
(e.g., Guix et al., 2005; Smith et al., 2006).
There are limited data on the occurrence of sequelae following astrovirus infection. Chronic
diarrhea has been reported lasting over 1 month (Grohmann et al., 1993; Lin et al., 2008).
Calicivirus (including Norovirus)
Caliciviruses, which include norovirus, are highly contagious and transmitted primarily through
the fecal-oral route—most commonly by consumption of fecally contaminated food or water or
through person-to-person spread (Schwab and Hurst, 2006; CDC, 2009b). However,
environmental and fomite contamination may also act as an important source of transmission and
infection. Waterborne outbreaks of norovirus disease in community settings are often caused by
sewage contamination of wells and recreational water (CDC, 2009b) and are widely considered to
be the causative agents in many viral waterborne outbreaks of unknown etiology (Maunula et al.,
2005; U.S. EPA, 2006c). Although pre-symptomatic viral shedding may occur, shedding usually
begins with onset of symptoms and may continue for up to 2 weeks after recovery from illness
(CDC, 2009b). The incubation period for norovirus-associated AGI in humans is usually between
24 and 48 hours (median in outbreaks 33 to 36 hours), but cases can occur within as little as 12
hours of exposure.
Infection usually presents as acute-onset vomiting, watery non-bloody diarrhea with abdominal
cramps, and nausea. Low-grade fever may also occur, though dehydration is the most common
complication, especially among the young and elderly. Symptoms typically last 24 to 60 hours,
and patients recover completely. Although asymptomatic infection may occur in as many as 30%
of infections, it role in calicivirus/norovirus transmission remains poorly understood.
Enterovirus
Enteroviruses are associated with a variety of clinical syndromes, ranging from mild and self-
limiting to severe, potentially fatal conditions. Non-poliovirus enteroviruses likely cause 10 to 15
million symptomatic infections and 30,000 to 50,000 hospitalizations in the United States each
year (Kim et al., 2001; CDC, 2006a). Enterovirus infections can cause a range of illnesses
including AGI (fever, nausea, diarrhea, or vomiting), conjunctivitis, skin rashes, cold and flu-like
illnesses, muscle inflammation, arthritis, paralysis, respiratory disease, meningitis, myocarditis,
organ failure, and even death. Infection is most often spread via the fecal-oral route, contact with
72
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
feces or virus-contaminated objects, direct contact with an infected individual, or recreational
water activities. Enteroviruses were responsible for at least one waterborne disease outbreak in the
developed world (Borchardt et al., 2003) and have been detected in U.S. source waters and treated
drinking water (Borchardt et al., 2004; Vivier et al., 2004). Enteroviruses in drinking water may
also contribute to viral infections of unknown etiology and unknown levels of endemic disease
(U.S. EPA, 2006c).
Enteroviruses potentially play a role in the development of chronic diseases, such as juvenile
diabetes and chronic fatigue syndrome (Behan and Bakheit, 1991; Fohlman and Friman, 1993),
myalgic encephalomyelitis (Lloyd et al., 1988), and myocarditis (Kim et al., 2001). Some patients
who have paralysis or encephalitis do not fully recover, and persons who develop heart failure
from myocarditis require long-term care for their conditions (CDC, 2010).
Hepatitis A Virus
Hepatitis A virus caused an estimated 32,000 infections in the United States in 2006, although
numbers of infections decrease each year, presumably due to the increased vaccination of children
(CDC, 2009c). Only rarely fatal, hepatitis A infections typically cause acute liver disease lasting a
few weeks to several months. Symptoms include fever, tiredness, nausea, decreased appetite, and
abdominal discomfort, followed by jaundice (Sobsey, 2006). The disease is more severe in adults
than it is in children. In the United States, Hepatitis A infection is most often spread from person-
to-person (often within the household) via the fecal-oral route and contact with feces or virus-
contaminated objects, but also by ingesting contaminated food or water, or through recreational
water activities. Drinking-water related Hepatitis A outbreaks have occurred in the United States
(e.g., Hejkal et al., 1982; U.S. EPA, 2006c; Yoder et al., 2008).
Hepatitis A has been associated with arthritis (Willner et al., 1998; Fan et al., 2009), specifically
mimicking autoimmune disease (Sridharan et al., 2000).
Hepatitis E Virus
Hepatitis E virus infections typically cause acute liver disease lasting a couple weeks to a few
months (Gerba, 2006). Although rarely fatal for most individuals (the most commonly affected
age group is young to middle-aged adults), the disease has a high mortality rate (15 to 20%) in
pregnant women (Mushahwar, 2008). Major waterborne Hepatitis E virus epidemics have been
documented in developing countries, but the disease is rare in the United States (Arguin et al.,
2008). Hepatitis E virus is transmitted via the fecal-oral route and infection is spread primarily
through fecally-contaminated drinking water or food, such as shellfish (WHO, 2009). Human
strains of Hepatitis E virus have experimentally infected pigs, and porcine strains have
experimentally infected primates (WHO, 2004; Halbur et al., 2001).
Rotavirus
Rotavirus infection is a leading cause of acute severe gastroenteritis in young children and infants
and worldwide (Abbaszadegan, 2006). They are excreted in very large quantities in the feces of
infected persons, so are present in relatively high densities in wastewater and environmental water
(Ansari et al., 1991). Each year, rotavirus infections account for an estimated 55,000 to 70,000
hospitalizations in the United States (CDC, 2008). Symptoms of rotavirus infection include
73
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
diarrhea, fever, and vomiting, leading to dehydration—especially in young children and infants.
Since the introduction of vaccines against rotavirus in 2006, numbers of cases in the United States
have declined. Rotavirus is highly contagious and the most infectious of the enteric viruses (Gerba
et al., 1996b). Infection is most often spread via the fecal-oral route, contact with feces or virus-
contaminated objects, direct contact with an infected individual, or from contaminated food or
water. Although not common, drinking-water related rotavirus outbreaks have occurred in the
United States (e.g., Hopkins et al., 1984, 1985). Rotaviruses have been detected in treated drinking
water (Abbaszadegan et al., 1999) and nondisinfected well water (Borchardt et al., 2003).
Rotaviruses in drinking water can also contribute to waterborne outbreaks of unknown etiology
and unknown levels of endemic disease.
BACTERIA
Aeromonas hydrophila
Aeromonads can cause mild to severe GI illness following consumption of contaminated food or
water and can vary in severity from self-limiting watery diarrhea with or without fever to an acute,
cholera-like illness with profuse watery diarrhea (Moyer, 2006). Severe infections most commonly
occur in young children and immunocompromised persons. Aeromonas hydrophila and related
aeromonads can also cause a variety of soft-tissue infections within hours following dermal
penetration injuries, leading to fever, pain, swelling, erythema, and edema. A wide range of
potentially serious complications can follow soft-tissue infection including cellulitis, arthritis,
sepsis, meningitis, and pneumonia (Noonburg, 2005). Other complications associated with
Aeromonas infection include hemolytic uremic syndrome, septicemia, meningitis, peritonitis,
wound infections, respiratory tract infections, and ocular infections (U.S. EPA, 2006d).
Arcobacter butzleri
Infections from A. butzleri, which was not distinguished from Campylobacter until recently
(Snelling et al., 2006), can result from fecal-oral transmission and can cause AGI, including
diarrhea associated with abdominal pain, nausea, vomiting and fever, and which can become life-
threatening in immunocompromised persons and populations. It is considered to be an emerging
foodborne pathogen of major significance in humans (Lehner et al., 2005; Snelling et al., 2006;
Cervenka, 2007). Although little is known about the mechanisms of pathogenicity or potential
virulence factors of Arcobacter spp., there is increasing evidence that livestock animals may be a
significant reservoir and that it may be zoonotic.
Campylobacter jejuni
The most important Campylobacter species from a public health perspective is C. jejuni, which is
also a major zoonotic pathogen. The most common health effect of C. jejuni infection is acute self-
limited AGI, characterized by diarrhea, fever, and abdominal cramps (Butzler, 2004). The
incubation period is typically 2 to 5 days but can extend up to 10 days. In about 50% of patients,
diarrhea is preceded by a febrile period associated with malaise, myalgia, and abdominal pain;
fresh blood may also appear in the stools by the third day. Most C. jejuni infections are sporadic
(often outbreak-related) in nature, and in most cases, the source of infection is never determined
(Hanninen et al., 2003). In drinking water outbreaks attributed to C. jejuni (e.g., Jones and
Roworth, 1996; Holme, 2003) the drinking water source is usually shown to be fecally-
74
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
contaminated by runoff of surface water after rain or by leakage of a sewage pipe close to the
drinking water pipeline.
Reactive arthritis is a rare complication of infection, and is strongly associated with people who
have the human lymphocyte antigen B27 (FDA, 2009). The arthritis typically occurs within 2
weeks after the onset of GI illness, but onset time can range from 4 to 35 days (Anonymous, 1998).
The condition is characterized by infection at a distant site, whereby joint inflammation occurs
without typical evidence of sepsis at the affected joint(s) (Shirtliff and Mader, 2002).
Reiter's syndrome is an aseptic arthritis (Kim et al., 2007) that is triggered by an infectious agent
located outside the joint. It also has a strong association with human leukocyte antigen-B27. Pope
et al. (2007) reviewed the epidemiological literature on Campylobacter-associated reactive
arthritis and found that follow-up for long-term sequelae related to Campylobacter and Reiter's
syndrome was largely unknown.
Campylobacter jejuni is increasingly recognized as a risk factor for Guillain-Barre syndrome, a
common cause of neuromuscular dysfunction (Bunning et al., 1997) that occurs in one in 1,000
infections (Butzler, 2004). Guillain-Barre syndrome is the most serious complication of
Campylobacter infection (Kaldor and Speed, 1984; Nachamkin et al., 2000; Yuki, 2001; Ang et
al., 2002; Gilbert et al., 2004). Studies by McCarthy and Giesecke (2001) have shown that the risk
of developing Guillain-Barre syndrome during the 2 months following a symptomatic episode of
C. jejuni infection is approximately 100x higher than the risk in the general population.
Immunoproliferative small intestinal disease is an infection associated lymphoma disease (Al-
Saleem and Al-Mondhiry, 2005). Molecular and immunohistochemical studies have demonstrated
an association with C. jejuni (Lecuit et al., 2004). This disease involves the proximal small
intestine resulting in malabsorption, diarrhea, and abdominal pain, as well as weight loss, intestinal
obstructions, and abdominal masses (Gilinsky et al., 1987; Rambaud et al., 1990; Fine and Stone,
1999; Al-Saleem and Al-Mondhiry, 2005). Symptoms are often chronic, and patients may
experience mild symptoms for 5 to 10 years before developing higher-grade illness or
lymphoplasmcytic and immunoblastic lymphomas (Al-Saleem and Al-Mondhiry, 2005).
Escherichia coli 0157
E. coli 0157 has been found worldwide in fecally contaminated surface water, groundwater, non-
chlorinated or inadequately treated drinking water and swimming pools, soil and sediment, and a
wide variety of foods (see review by Muniesa et al., 2006). It has been documented to cause both
endemic and outbreaks of illness in humans through fecal-oral transmission, consumption of
fecally-contaminated food and water, and contact with infected persons and animals (Rangel et al.,
2005). The clinical symptoms of infection vary from non-bloody diarrhea to bloody diarrhea,
referred to as hemorrhagic colitis. Hemorrhagic colitis is a serious, life-threatening condition and
can lead to hemolytic uremic syndrome, which results in renal damage and possibly death.
Enterohemorrhagic E. coli infections can lead to long-term or permanent kidney damage and renal
disease. Persons who develop chronic kidney failure may require lifelong dialysis or a kidney
transplant. Garg et al. (2003) conducted a comprehensive review and meta-analysis of the current
literature along with expert consultations in order to quantify the long-term renal prognosis of
75
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
patients with diarrhea-associated hemolytic uremic syndrome. A higher severity of acute illness
was strongly associated with worse long-term prognosis. Studies with a higher proportion of
patients with central nervous system symptoms (coma, seizures, or stroke), had a higher proportion
of patients who died or developed permanent end stage renal disease at follow-up. Death or end
stage renal disease occurs in about 12% of patients with diarrhea-associated hemolytic uremic
syndrome, and 25% of survivors demonstrate long-term renal sequelae.
E. coli is also implicated in rheumatoid diseases (Locht and Krogfelt, 2002; Schiellerup et al.,
2008), including reactive arthritis and Reiter syndrome with similar pathologies as described
above.
Helicobacter pylori
This common infectious pathogen occurs worldwide and is associated with a variety of upper GI
conditions (Bellack et al., 2006), including gastritis and peptic and duodenal ulcer disease. At
present, the route of transmission to susceptible hosts remains largely unknown, although fecal-
oral and person-to-person contact have been proposed as possible routes of exposure (Brown,
2000; Bellack et al., 2006). In addition, fecally-contaminated drinking water may also be a source
of transmission and subsequent infection (McDaniels et al., 2005; Reavis, 2005; Azevedo et al.,
2006).
H. pylori causes chronic gastritis (Chey and Wong, 2007), with a small proportion of affected
patients (6 to 20%) proceeding to more severe clinical disease (Parsonnet et al., 1991), and can
lead to gastric cancer (Wang et al., 2007). It has been estimated that H. /^/orz-positive patients
have a 10 to 20% lifetime risk of developing ulcer disease and only a 1 to 2% risk of developing
gastric cancer (Kuipers et al., 1995; Kuipers, 1999; Ernst and Gold, 2000). H. pylori is classified
as a Class I carcinogen by the International Agency of Research on Cancer (IARC, 1994).
According to the CDC, over 90% of duodenal cancers, and 80% of gastric ulcers are caused by H.
pylori (CDC, 1998a).
Legionella pneumophila
Legionella pneumophila can cause respiratory disease in humans when a susceptible host inhales
aerosolized water containing the bacteria or aspirates water containing the bacteria (Fields et al.,
2002; Hall, 2006). Legionellosis classically presents as two distinct clinical entities, (1)
Legionnaires' disease, a severe multisystem pneumonia-like disease that includes fever,
nonproductive cough, headache, myalgias, rigors, dyspnea, diarrhea, and delirium (Fraser et al.,
1977); and (2) Pontiac fever, a self-limited flulike illness (Glick et al., 1978). However, many
persons who seroconvert to Legionella will remain entirely asymptomatic (Boshuizen et al., 2001).
It has been estimated that 8,000 to 18,000 persons are hospitalized with legionellosis annually in
the United States (Fields et al., 2002). The disease is a major concern of public health professionals,
and especially organizations and individuals involved with maintaining building water systems.
However, legionellosis is generally considered to be a preventable illness because controlling or
eliminating the bacterium in water reservoirs will typically prevent the disease.
Legionella infections are associated with long-term development of chronic inflammatory and
fibrotic reactions (pulmonary fibrosis) in the human lung (U.S. EPA, 1999b).
76
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Mycobacterium avium
People can be exposed to M. avium in water through drinking, swimming, and bathing activities
and through ingestion or inhalation of water vapor or droplets (Primm et al., 2004). As
opportunistic waterborne pathogens, exposure toM avium can infect the lungs and lead to cough,
fatigue, weight loss, fever, and night sweats, especially in immunocompromised persons and
populations (e.g., AIDS patients) (Primm et al., 2004; LeChevallier, 2006). Disseminated MAC
infections in individuals with AIDS accounts for the majority of MAC-related morbidity and
mortality in the United States.
MAC has also been associated with hypersensitivity pneumonitis (Lacasse et al., 2003; Fink et al.,
2005; Hanak et al., 2007). Symptoms may be acute with flu-like symptoms, fever, chills, malaise,
headache, and cough, or chronic, characterized by dyspnea (a cough that may be dry or productive)
and weight loss (Lacasse et al., 2003; Field et al., 2004).
Plesiomonas shigelloides
Infection by P. shigelloides most commonly results from ingestion of fecally-contaminated food
(especially seafood) and water—especially in immune-compromised persons (Wong et al., 2000;
CDC, 2009a). Extra-intestinal health effects, most commonly observed in sensitive populations,
include septicemia, meningitis in neonates, cellulitis, septic arthritis, and acute cholecystitis.
However, in immunocompetent persons, infected persons typically develop self-limiting watery
diarrhea that can last for 2 weeks or more (Huq and Islam, 1983; Wong et al., 2000; CDC, 2009a).
Salmonella enterica
Most human infections can be traced back to contaminated food products, although water-related
outbreaks, including drinking water outbreaks, continue to be reported (Schuster et al., 2005;
CDC/MMWR, 2006; Covert and Meckes, 2006; Craun et al., 2006). The incubation period ranges
from 18 to 48 hours after ingestion and illness is usually characterized by acute, self-limiting AGI
(though some infections can be severe), fever, and septicemia, that lasts from 2 to 5 days (Covert,
1999).
A broad range of chronic sequelae have been reported, such as reactive arthritis, Reiter's
syndrome, rheumatoid syndromes, pancreatitis, osteomyelitis, myocarditis, colitis, choleocystitis,
and meningitis have been reported as a consequence of Salmonella infection (Thomson et al.,
1995; Dworkin et al., 2001; Motarjemi, 2002).
Shigella sonnei
Shigellosis, commonly known as acute bacillary dysentery, is characterized by the passage of loose
stools mixed with blood and mucous and accompanied by fever, abdominal cramps, and tenesmus
(Sur et al., 2004; Moyer and Degnan, 2006). The incubation period of shigellosis is typically 1 to
4 days, which is usually followed by sudden onset of AGI symptoms. In mild cases, the disease
may be self-limiting, but severe disease requires appropriate medication. The disease is
communicable as long as an infected person continues to excrete the organisms in feces at high
levels, which can persist up to 4 weeks from the onset of illness. Thus, humans are the principal
reservoir of infection (Sur et al., 2004), and transmission usually occurs via fecally-contaminated
water and food or through person-to-person contact (Niyogi, 2005).
77
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
In addition to acute and chronic GI effects, shigellosis can cause central nervous disease such as
seizures and convulsions ((Barrett-Connor and Connor, 1970; Khan et al., 1999). Shigella has also
been associated with urinary tract infections (Ekwall et al., 1984). Other complications from
shigellosis include bacteremia, hemolytic uremic syndrome, toxic megacolon (reviewed by Gupta
et al., 2004), and encephalopathy (Goren et al., 1992; Ferrera et al., 1996). Both reactive arthritis
and Reiter's syndrome are documented long-term sequelae from shigellosis (Gupta et al., 2004),
and the pathologies of these conditions are similar to those of other enteric pathogens such as
Salmonella.
Vibrio cholerae
V. cholerae is the cause of cholera, a GI illness of global and historic importance. In nature, V.
cholera is often associated with protozoa (including Acanthamoeba), zooplankton, sediments, and
shellfish, and can become a normal component of the aquatic microbiota (especially saline waters)
is many parts of the world, including the United States. Although contaminated food remains the
predominant source of infection and illness, contaminated water remains a highly important source
of V. cholerae infections worldwide (Toranzos et al., 2006). Infections have been shown to vary
from completely asymptomatic to very severe, profuse watery diarrhea and vomiting that can lead
to rapid dehydration and death within one to five days. No outbreaks of cholera have been reported
in U.S. drinking water supplies in several decades (see Steinberg et al., 2001).
Yersinia enterocolitica
Most human infections result from fecal-oral transmission and can usually be traced to
contaminated food or water. Y enterocolitica can cause a variety of symptoms depending on the
age of the person infected, but most commonly GI, fever, and sometimes vomiting in children that
is typically self-limiting and can last for up to 2 weeks (Flicker, 2006). However, post-infectious
chronic sequelae include a form of reactive arthritis and a pseudoappendicitis syndrome (Sharma
et al., 2003).
Yersinia elicits rheumatoid disease, similar to many other enteric bacteria such as Shigella,
Campylobacter, and E. coli. Post-infectious sequelae of Yersinia infection include a form of
reactive arthritis and joint symptoms (Wolf et al., 1991; Luo et al., 1994; Sharma et al., 2003;
Schiellerup et al., 2008).
AMOEBAS
Acanthamoeba
Acanthamoeba spp. are opportunistic pathogens that produce granulomatous amebic
encephalitis—a chronic central nervous system disease of immunocompromised hosts—and
various other diseases, including keratitis and pneumonitis (Marshall et al., 1997; Visvesvara and
Moura, 2006). To date, keratitis is the only water-related syndrome caused by Acanthamoeba.
Acanthamoeba was determined to be the causative agent of keratitis in an U.S. outbreak involving
a contaminated municipal water supply (Meier et al., 1998; Karanis et al., 2007). Untreated
keratitis can lead to loss of visual acuity and blindness (Illingworth and Cook, 1998; Marciano-
Cabral et al., 2000; Marciano-Cabral and Cabral, 2003; Khan, 2006; Awwad et al., 2007;
78
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Thebpatiphat et al., 2007; Visvesvara et al., 2007). The infection is also known to cause chronic
encephalitis in immune-deficient individuals (U.S. EPA, 2003h).
Blastocystis hominis
B. hominis is transmitted via the fecal-oral route through ingestion of contaminated food and water
(Garcia, 2006a). When present in large numbers, B. hominis can cause GI illness, including
diarrhea, cramps, nausea, fever, vomiting, and abdominal pain that may require medical attention.
Although the severity and duration of these symptoms can be increased in immunocompromised
persons, in healthy adults their presence may be asymptomatic (Chen et al., 2003). Globally, B.
hominis has been attributed to four and possibly five documented waterborne disease outbreaks
(Karanis et al., 2007)—including a 2005 gastroenteritis outbreak involving an inadequately
disinfected public drinking water system in Turkey (Tuncay et al., 2008).
Entamoeba histolytica
Transmission is via the fecal-oral route, directly from person-to-person contact, or through
contaminated food, water, or fomites (Marshall et al., 1997; Keene, 2006). In this manner, persons
become infected and can become ill after ingesting cysts. The incubation period is highly variable,
ranging from a few days to several months. Although most E. histolytica infections are
asymptomatic, amebiasis can be a serious and even life-threatening disease, especially in
developing nations. It can also result in recurrent diarrhea of varying severity, often with blood or
mucous, fever, abdominal pain, and tenesmus. While drinking water-related outbreaks have been
documented outside of the United States (e.g., Chen et al., 2001; Karanis et al., 2007), there have
been no reports on the occurrence of E. histolytica in source or finished drinking water, or U.S.
waterborne disease outbreaks since 1980 (Marshall et al., 1997).
Naegleria fowleri
Primary amebic meningoencephalitis associated with exposure to Naegleria fowleri is a very rare
but usually fatal central nervous system illness that has led to several deaths in the United States
(e.g., Barnett et al., 1996; Gyori, 2003; Marciano-Cabral et al., 2003; Visvesvara and Moura,
2006). Once entering into the nostrils of swimmers and others engaging in water sports, N. fowleri
penetrates the mucosal layer and migrates along the olfactory nerve tracts and eventually reaches
the brain (Schuster and Visvesvara, 2004). The often fatal disease is typically acquired by immune-
competent children and young adults while swimming and diving in untreated warm freshwater
lakes and ponds during summer months (Barnett et al., 1996; Marciano-Cabral et al., 2003;
Schuster and Visvesvara, 2004).
PROTOZOA
Cryptosporidium
Cryptosporidium is a zoonotic waterborne pathogen of global public health importance. The CDC
estimates that there are 700,000 cases of Cryptosporidium infections a year in the United States
(Scallan et al., 2011a). Worldwide, the estimated number of annual cases of cryptosporidiosis
exceeds several million and has been documented in almost 100 countries and on every continent
except Antarctica (Casemore et al., 1997; Fayer et al., 1997; Caccio, 2005). Cryptosporidium is a
well-known cause of opportunistic infections among AIDS patients (Roy et al., 2004) and a
79
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
common cause of outbreaks of GI illness (CDC, 2008). Surveillance data indicate that infections
are common among immunocompetent individuals and that 90% of cases are not involved in
outbreaks (Dietz et al., 2000). Valid species of Cryptosporidium have been described in more than
155 mammalian species, >30 avian species, 57 reptilian species, 9 species of fish, and 2 amphibian
species (O'Donoghue, 1995; Fayer, 2004).
Most clinical cases of cryptosporidiosis involve infection by C. parvum or C. hominis. The most
common feature of cryptosporidiosis is profuse, watery diarrhea. Other clinical signs of infection
include dehydration, fever, anorexia, weight loss, weakness, and progressive loss of condition
(O'Donoghue, 1995). Recovery is usually spontaneous within 1 to 2 weeks of infection.
Developmental stages of the parasite are often seen within the small intestine and occasionally
elsewhere (stomach, colon, liver, lungs). In general, the development of cryptosporidiosis depends
on the species, age, and immune status of the host (Fayer et al., 1997). Younger persons with less
developed or compromised immune systems are generally more susceptible to severe infection
than healthy adults (O'Donoghue, 1995).
Several Human volunteer feeding studies have been conducted to determine the infectivity of C.
parvum and C. hominis in healthy adults in order to predict the likelihood of enteric infection
following exposure to contaminated drinking water (DuPont et al., 1995; Chappell et al., 1996;
Okhuysen et al., 1999, 2002). A summary of the dose-response data for all six of the tested isolates
has been published by EPA (U.S. EPA, 2006b).
Cyclospora
Endemic in much of the developing world, but most common in tropical and subtropical areas,
cyclosporiasis is considered an emerging GI disease in developed nations. It has been identified as
the cause of several outbreaks in North America and Europe and with traveler's diarrhea
(Herwaldt, 2000; Karanja et al., 2007). Cyclosporiasis transmission has primarily been linked to
fecally-contaminated foods and water and is associated with a wide variety of GI symptoms, such
as loose or watery diarrhea, nausea, vomiting, abdominal cramps, or appetite loss. It can also be
associated with fever; chills; muscle, joint, or generalized body aches; headache; or fatigue.
However, the exact mechanism of disease transmission remains unknown because freshly excreted
oocysts are not infective and require days to weeks to mature sufficiently to become infective upon
consumption; thus, direct person-to-person transmission of the disease is unlikely (Herwaldt,
2000). Although asymptomatic infections are known to occur, especially in immunocompetent
children, the onset of symptoms in naive populations observed in outbreaks is typically 1 to 14
days after exposure and is often accompanied by a characteristic waxing and waning of symptoms
over time. In endemic countries, symptoms begin approximately 5 to 8 days after exposure and
may persist for over a month, although watery diarrhea remains the most common health outcome.
Reports of C. cayetanensis sequelae such as biliary disease, aculculous cholecystitis, Guillain-
Barre syndrome and reactive arthritis syndrome following prolonged infection were found in the
literature (Sifuentes-Osornio et al., 1995; Richardson et al., 1998; Connor et al., 2001).
Giardia
Giardia intestinalis (also known as G. lamblia and G. duodenalis) is a zoonotic waterborne
pathogen of global public health concern. It is the most common intestinal parasite identified by
80
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
public health laboratories in the United States (Rose et al., 1991; Kappus et al., 1994). CDC
estimates that there are approximately 1.2 million to 2 million illnesses annually in the United
States due to Giardia (Mead et al., 1999; Scallan et al., 201 la). Thirty percent of the U.S. general
population has seropositivity for Giardia (indicates current or past infection) and 7% of small
children are asymptomatically infected with Giardia (Frost and Craun, 1998). More than 100
waterborne giardiasis outbreaks have been reported worldwide from the beginning of the previous
century till 2004 (Plutzer et al., 2010). Giardia also infects a wide variety of domestic and wild
mammals (e.g., cats, dogs, cattle, deer, and beavers) (Thompson, 2000; Ballweber et al., 2010).
High risk groups for giardiasis include infants and young children, travelers to developing
countries, the immunocompromised, and persons who consume untreated water from lakes,
streams, and shallow wells (U.S. EPA 1998a; CDC, 2008). A wide spectrum of symptoms are
associated with giardiasis, which range from asymptomatic infection and acute self-limiting AGI
to persistent chronic diarrhea, which sometimes fails to respond to treatment. Asymptomatic
infection is very common (50 to 75% of infected persons are symptomatic) (Mintz et al., 1993;
U.S. EPA 1998a). Symptoms of giardiasis include diarrhea, abdominal cramps, bloating, weight
loss, and malabsorption (Rodrigquez-Hernandez et al., 1996; Hellard et al., 2000; Thompson,
2000). Case reports also indicate that giardiasis might be associated with the development of
reactive arthritis (Tupchong et al., 1999). Giardia infection is frequently self-limited, but
immunocompromised persons may have more serious and prolonged infection (Benenson, 1995).
Hospitalizations and deaths due to giardiasis are relatively rare; CDC estimates that giardiasis
causes approximately 10-30 deaths and 3,500- 5,000 hospitalizations annually in the United States
(Mead et al., 1999; Scallan et al., 2011a).
In some patients, symptoms last for only three or four days, while in others the symptoms last for
months. Generally, patients commonly resolve their infections spontaneously, with acute disease
lasting from one to four weeks (Smith and Wolfe, 1980). However, in some patients, the acute
stage may persist for months (Wolfe, 1990). The period of communicability lasts for the entire
duration of infection; however, the shedding of cysts can be intermittent (Benenson, 1995).
Clinical data suggest that Giardia cysts are highly infective for humans (Rendtorff 1954a,b,
Rendtorff and Holt 1954a,b; Rose et al., 1991; Teunis et al., 1996).
Isospora belli
Transmission is through fecal contamination of water and food from a human source, especially
in developing nations and in tropical regions (Marshall et al., 1997). Clinical manifestations of I.
belli infection, isosporidiosis, are most commonly chronic to severe diarrhea, which can persist for
months to years, causing weight loss, abdominal colic, and fever (Garcia, 2006b). These health
outcomes are generally increased in severity and duration in immunocompromised persons.
Worldwide, I. belli have been attributed to at least three documented waterborne disease outbreaks
(Karanis et al., 2007)
Microsporidia
Microsporidiosis is an emerging and opportunistic infection and is associated with a wide range
of clinical, often organ-specific syndromes in humans, including diarrhea, hepatitis,
81
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
keratoconjunctivitis, renal failure, and even blindness (Didier et al., 2004; Cali, 2006; Didier and
Weiss, 2006). However, persistent or self-limiting diarrhea are the most common symptoms
associated with microsporidiosis in immune-competent or immune-deficient individuals,
respectively (Didier et al., 2004). Microsporidia have been attributed to at least one recent
waterborne disease outbreak (Karanis et al., 2007).
Toxoplasma gondii
In humans, T. gondii is normally transmitted by ingestion of food or water contaminated with
oocysts or by ingesting infective animal tissues that contain tissue cysts (Dubey, 2004, 2006). In
pregnant infected hosts, T. gondii can multiply in the placenta and spread to fetal tissues, especially
during the first half of gestation, causing mental retardation, loss of vision, hearing impairment,
and mortality in congenitally infected children. Although toxoplasmosis is mostly an
asymptomatic infection in adults, it can cause serious disease morbidity and mortality in
immunocompromised persons (especially encephalitis in AIDS patients) (Artigas et al., 1994;
Sanchez et al., 2000).
4.1.4. Extent of Secondary Transmission
Whether secondary transmission is included in the analysis needs to be determined during problem
formulation. If the questions the risk assessment is asking requires secondary transmission to be
considered, then dynamic MRA models can characterize secondary cases that occur among
contacts following exposure to a primary case. Static MRA models usually consider secondary
transmission to be negligible or include it as a non-fluctuating multiplicative factor (e.g., secondary
cases equal primary cases multiplied by 0.1; assuming a 10% secondary transmission rate).
When secondary transmission is included the immune status of individuals becomes more
important for modeling. For example, individuals who are already infected with a particular
pathogen should not be considered susceptible to reinfection by the same pathogen while infected.
Previously infected but recovered individuals have decreasing immunity over time, which may
affect the design of the risk assessment model. Section 2.3.1.2 includes more information on
dynamic models and the inclusion of secondary transmission in risk assessment.
4.2. Dose-Response Assessment Overview
In the case of waterborne microbial contaminants, risk assessment generally involves estimating
the probability of illness or infection based on exposure estimates and dose-response relationships.
During dose-response analysis, data from human clinical studies, epidemiological studies, animal
studies, and/or outbreaks are used to develop a mathematical relationship between the intensity of
exposure and the subsequent occurrence of disease or infection.
Dose-response models are generally derived as the logical mathematical consequence of the
assumptions made about the infection process. For example, the exponential dose-response
relationship is derived by assuming that the distribution of organisms between doses is random
(i.e. Poisson), that each organism has an independent and identical survival probability, and that a
single organism can cause infection. The form of the relationship between exposure and response
is determined by (1) assumptions related to the biological processes leading to infection, and (2)
82
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
the "shape" of the relationship found in the data between exposure and the health outcome of
interest. Statistical techniques are often used to optimize the relationship's parameters to best fit
the available data.
A number of factors are addressed in the derivation of dose-response models and estimation of
their parameters for microbial risk assessment, including the following (adapted from ILSI, 2000):
Dose-response factors
• statistical model(s) to analyze or quantify dose-response relationships;
• human and/or animal dose-response data;
• source and/or preparation of challenge material or inoculums.
Factors that overlap with exposure assessment
• utilization of outbreak or intervention data (can also be used to build exposure scenarios);
• route of exposure or administration used in the dose-response study;
• equivalence of methods used (including organism type, strain, and method units) for
occurrence data and dose-response study;
Factors that overlap with health effects
• characteristics of the exposed population (age, immune status, etc.); and
• Infection or disease endpoint for the dose-response relationship (e.g., pathogen shedding,
serological response, symptoms).
The mathematical form of the dose-response model may vary with pathogen or strain, route of
administration, distribution of host statuses, and other factors. An overview of common dose-
response models for microbial based infections is provided below. Either human or animal data
may be used to derive dose-response estimates—although human data are generally preferred if
they are available. Pooling data from different animal models can enhance dose-response models
(Bartrand et al., 2008). Although pooling animal and human data should be conducted with caution
(FDA/USDA, 2003). Information from disease outbreaks may also provide useful information
about both primary infection risks (infections arising directly from exposure) and secondary
transmission (person-to-person) (Teunis et al., 2004, 2008a).
Knowledge of the conditions under which dose-response data were collected is essential both for
those developing dose-response models and for those evaluating dose-response models for use in
MRA. In particular, the strain of the pathogen, model form, enumeration method, and route of
inoculation can strongly influence the use of a reported dose-response relationship. Extrapolation
of dose-response relationships to conditions other than those for which data were collected should
be done only in conjunction with justification and with a full description of the conditions for
which the dose-response model was developed.
For example, dose-response models have been proposed based on data collected during
experiments with animal hosts whose response appears to differ substantially from that of humans.
83
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Comparison of murine response to Listeria monocytogenes (based on feeding studies) to that of
humans (based on epidemiological data) indicates a factor of ~106 difference in the lethal dose for
50 percent of the fetal population (LD50) between the two hosts (FDA/USDA, 2003). Numerous,
albeit dated, studies of pathogen-host combinations (e.g., Bell et al., 1955, for tularemia;
Holdenfried and Quan, 1956, for plague) have shown the potential for wide variation in response
between animal hosts and even among animal hosts of the same species but of different geographic
origin. Insights into the applicability of animals as models of human infection may sometimes be
drawn from pathology literature; in many cases animal models are selected for pathology
experiments based on similarities between their infection process and that of humans (Lyons and
Wu, 2007). Knowing and reporting these similarities may provide information for interpretation
of dose-response data. As another example, if a QMRA were to use rotavirus data collected via a
molecular method, such as reverse transcriptase (RT-)PCR methodology, it would be necessary to
harmonize those data with the median tissue culture infective dose data reported during the clinical
trial used as the basis for the commonly used dose-response relationship.
In summary, the use of dose-response models developed based on animal data for estimating
human response is not universally accepted within the risk community. Reasons for using such
models are that researchers select animal models based on similarity in response to that of humans,
because uncertainties related to extrapolating animal models to humans may be less than
uncertainties inherent in other techniques for developing human dose-response models (e.g., dose
uncertainties when outbreak data are used to generate dose-response models), and because, in some
cases, human data are not available at the low doses of concern and animal data can be obtained
whereas human data cannot. Current research on dose-response models may provide an avenue for
verifying the appropriateness of using animal dose-response models to estimate human response.
Different strains of a pathogen may have very different degrees of infectivity and the strain or
strains used to generate the experimental data may or may not be the most relevant to the
population that is exposed in the exposure scenario that is being modeled. For example, the
methods used to determine doses for dose-response relationships may vary in their sensitivity for
detection and hence introduce uncertainty into the accuracy of doses delivered (e.g., different cell
lines used for infectivity-based assays for viral titres vary in their susceptibility to infection).
Furthermore, and as noted previously, susceptibility in the general population may vary greatly
with age, general immune status, and pre-existing or acquired immunity to specific pathogens and
pathogen strains. The narrative accompanying the modeling should explain the potential impacts
of different strains (or other sub-classifications of pathogens such as serovar or single nucleotide
polymorphisms) on the outcome of the risk calculation. The potential uncertainty associated with
variation in infectivity is rarely included in MRA studies currently; however, it should be
calculated quantitatively if possible. Most MRAs treat severity of symptoms independently of dose
size because most of the established dose-response relationships are based on infection rather than
illness. However, there is some evidence for dose-dependence of severity of symptoms (Text
Box 6).
84
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 6. Dose-Dependency of Host-Pathogen Interactions
Most MRAs assume that severity of health endpoint is not influenced by magnitude of dose. For
example, with Cryptosporidium, whether an individual is exposed to 1 or 10,000 organisms, if they
become infected and ill, the health end points are assumed to be similar in severity. Thus, the
assumption is made that exposure to a larger Cryptosporidium dose will not result in worse symptoms
(U.S. EPA, 2006a). Although there is emerging evidence that this assumption may not be appropriate
for all pathogens at all doses, the data are generally insufficient to be included in quantitative risk
assessment. However, if there is evidence of dose-dependent severity of symptoms for the pathogen
of interest, then it should be discussed. For example, pathogenic E. coii feeding studies in human
volunteers demonstrated dose-dependency of disease severity and suggested that volume of liquid
stool can be used as a quantitative metric for illness severity (Bieber et al., 1998). Colwell et al. (2003)
reported that in Bangladeshi villages where sari or nylon cloth was used to filter surface water, the
number of cholera cases was reduced and the severity of disease was reduced compared to villages
that did not filter their surface water. Nauta et al., (2009) compared six Campylobacter risk
assessments and concluded that the most effective public health intervention measures for risks
associated with exposure to broiler chickens targeted Campylobacter density reductions, rather than
reducing its prevalence (note that this finding may or may not extend to environmental waters).
The duration of exposure, the number of exposures, and time between exposures may affect the
probability of an adverse health effect, as estimated by the dose-response relationship. As
discussed previously, determining the independence or lack of independence of exposure events
is complicated by host status as well as pathogen characteristics. For example, individuals who are
already infected with a particular pathogen should not be considered susceptible to reinfection by
the same pathogen while infected. Previously infected but recovered individuals have decreasing
immunity over time. However, the nature of the decrease in immunity depends on many conditions
including whether subsequent exposures boost immunity, host factors that relate to overall health
of the immune system, and pathogen factors such as rapidly evolving antigenic epitopes.
Because any given person's immunity fluctuates based on many host factors, and because different
pathogens elicit different immune responses, it is difficult to define a single exposure duration that
best describes all combinations of host-pathogen interactions. Given the variability and
complicated nature of capturing all appropriate exposure durations, most risk assessments choose
a default exposure event duration (e.g., all water consumed during 1 day).
4.2.1. Overview of Common Dose-Response Model Forms for Pathogens
This section provides a brief summary of the most commonly used dose-response models for
microbial pathogens. Namata et al. (2008) also provides a useful summary of dose-response
models for MRA. Although providing state-of-the-art guidance on deriving dose-response
relationships is beyond the scope of this MRA Tools document, there are several issues that risk
analysts and risk managers should be aware of when evaluating the dose-response literature.
Appendix C provides additional information on dose-response modeling.
The objective of the dose-response assessment is to develop a relationship between the number of
microbes a person or population has been exposed to and the likelihood of occurrence of an adverse
consequence (health outcome). In general, dose-response assessment would be relatively
straightforward if the level of microbial risk that was deemed acceptable was sufficiently high to
85
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
allow experimentation that would permit the directly observable assessment of risk (Haas et al.,
1999). However, the probability of infection (risk) from a single low-dose exposure event is often
sufficiently low that use of direct observation (or experimentation) is impractical. For example,
using a standard dose-response model (Medema et al., 1996), a dose of 30 Campylobacter jejuni
corresponds to an infection probability of -0.2, but that probability rises to 0.5 at a much higher
dose (-900). Thus, the use of parametric dose-response curves to facilitate extrapolation into the
low-dose range that matches the risk level of concern is necessary.
Dose-response models are mathematical functions that input the dose to which individuals or
populations are exposed and yield a probability (bounded by 0 and 1) of the particular adverse
health effect (Haas et al., 1999). These dose-response functions play a prominent role in risk
assessments for pathogens in water because they effectively translate exposures into risks. In real
world situations where large numbers of individuals may be exposed (e.g., public water supplies),
relatively low individual risk levels may be of concern from a public health perspective because
even low individual risks can translate into a large number of illnesses.
The two most commonly used dose-response models are the exponential and beta-Poisson models.
The use of exponential and beta-Poisson models is only valid, however, when their underlying
assumptions are met. More computationally intensive dose-response relations are also available
for conditions in which neither the exponential or beta-Poisson models are appropriate. Alternative
two-parameter models have been proposed for use in MRA assessment, including the log-normal,
log-logistic, extreme value models (Pinsky, 2000). Three-parameter models that have been
suggested for MRA include the Weibull gamma (Farber et al., 1996), exponential gamma, Weibull
exponential, and the shifted Weibull model (Kodell et al., 2002). Although three-parameter models
are more flexible than two-parameter models, they require data at four or more doses, which is not
available for many microbial pathogens. Research continues to be conducted on appropriate
methods for selection of models from among these and other candidate models (e.g., Moon et al.,
2004, 2005).
The models discussed in this section estimate risks for exposed individuals. Population-level risks
(i.e., the incidence of disease among a group of exposed individuals) are generally constructed by
combining individual risks with estimates of the distribution of doses to the exposed population.
To promote transparency and clarity in an MRA, the following points should be addressed for
each dose-response model chosen:
• a discussion of assumptions inherent in making extrapolations to doses lower than those
used in studies;
• a detailed description of dose-response and risk assessment modeling approaches,
including the applicability of the models for use in various exposure situations and for
various pathogens;
• methods used to assay doses and exposure, because they may not be equivalent, and a
summary of approaches taken to harmonize between methods;
• the models' key assumptions;
• the type of information that the various models are expected to provide;
• limitations of the models;
86
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• the use of likelihood methods to compare how well dose-response models fit the data;
• the biological rationale for the model selected;
• the strengths/weaknesses and advantages/disadvantages of the models, including a
comparison of the benefits and limitations of the chosen models versus other potential
models; and
• a discussion of the flexibility in approaches to the dose-response relationship depending on
the pathogen being considered and the assumption about a no-threshold effect (i.e., can it
be assumed that one organism is sufficient to produce infection in some portion of an
exposed population or subgroup?).
To take biological mechanisms into account, a dose-response model for microbes should account
for the heterogeneous distribution (random or clumping) of microbes in water (affecting exposure)
and a microbe's ability to reproduce in the human body (linked to pathogenicity) (Haas et al.,
1999). Laboratory dose-response studies are usually conducted under conditions in which the
microorganisms are randomly distributed in the administered dose. This is known as a Poisson
distribution. The framework for exponential models is based on well-studied mathematical
relationships; however, the model parameters use empirical data from clinical trials and
epidemiological studies that are organism-specific (e.g., an organism's infectious dose for 50% of
the exposed population [ID50]). A concern for environmental water samples regarding the Poisson
distribution is clumping, association with suspended solids, and other spatial distribution issues;
however, this phenomenon can be accounted for in dose-response modeling by incorporating
aggregation parameters into the dose- response model (see Teunis and Havelaar, 2000 and Teunis
et al., 2008b for further information).
The dose-response relationship that is defined by the equation is "fit" to experimental data using a
variety of statistical methods. If the model is a good fit, it will predict risks that are close to those
actually observed within the range of experimentally administered doses. However, the doses used
in volunteer studies may be higher than those typically encountered in the environment, so it is
necessary to extrapolate the risks associated with lower doses using the model derived from the
higher doses. In extrapolating to lower doses, risk assessors rely on the belief that the form of the
dose-response model is based on an accurate representation of the infection process that holds at
low doses as well as high doses. Text Boxes 7 and 8 illustrate how experimental data on
Cryptosporidium and noroviruses have been used to derive dose-response relationships for these
pathogens. Moreover, these dose-response relationships are generally based on clinical trials for
which only the average doses are known for each group. Recently, outbreak data have been used
to derive dose-response relationships for several waterborne pathogens. Teunis et al. (2004, 2008a,
2010) and Bollaerts et al. (2008) provide good examples of how outbreak data have been used to
derive dose-response relationships.
Several published studies (e.g., Coleman and Marks, 2000; Nauta et al., 2009) suggest that it might
not be advisable to extrapolate dose-response models based on clinical trials for waterborne
exposures, given the complexity of the pathology of illnesses and given the relatively low reported
incidence of illness and the relatively high daily exposure of humans to pathogens (Levin and
Antia, 2001). Although a critical evaluation of this perspective is difficult to provide, given the
limited data available for human response to exposure to pathogens of known dose and
87
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
characteristics, mechanistic modeling offers an avenue for development of improved models for
extrapolation to low doses.
Text Box 7. Brief Summary of Cryptosporidium Feeding Studies
Human feeding studies have been used for decades to systematically evaluate dose-response effects
for pathogens. Chappell et al. (Chappell et al., 1999, 2006; DuPont et al., 1995; Okhuysen et al.,
1999, 2002) have conducted volunteer feeding studies using five strains of Cryptosporidium parvum
that have formed the basis of dose-response parameters used in several MRAs. Students and
employees of the University of Texas Health Science Center and others in the surrounding area of
Houston were recruited as volunteers. Volunteers could not be caretakers of infants, elderly, or those
with chronic diseases or immunosuppression. Recruits were given extensive information on
cryptosporidiosis and had to score 100% on a written examination that tested the recruits on their
comprehension of the study, the fact that they could become ill, that there was no effective treatment
for the illness, and that the organisms could be spread to household contacts. The next stage of their
evaluation for inclusion in the study involved providing medical histories and passing extensive
medical tests.
In each of the three studies, the volunteers ingested a single known dose of viable C. parvum oocysts
of one of three isolates—IOWA, TAMU, and UCP. The subjects were given anywhere from 10 to
1,000,000 oocysts per dose.
Volunteers submitted stools passed after the challenge and completed daily diaries regarding their
stool passage and any symptoms; household contacts were also monitored for diarrheal illness. Blood
was collected from each of the volunteers at specified days post-challenge and tested for antibody
response.
To measure the challenge responses, the researcher considered two definitions of infection—
confirmed and presumed. A confirmed infection was based on oocysts detected in stools using direct
fluorescence assay. Some volunteers who had oocysts in their stools did not develop any symptoms.
In contrast, some volunteers had illness symptoms that were indistinguishable from those with
confirmed infection, but had no detectable oocysts in their stools. Because of the detection limit of the
assay methodology, these volunteers were presumed to be infected. Those volunteers who did not
have either Gl symptoms or fecal oocysts throughout the study were presumed to be uninfected.
Because the purpose of the studies was to develop dose-response curves for the different strains
based on infectious dose, it was important to be able to capture data on the median infectious dose.
The first study of the IOWA isolate was designed to cover a wide range of doses (30-1,000,000
oocysts) so to more effectively capture the median dose. The doses of the other two isolates were
adapted as the study progressed to narrow the range of doses; that is, the first group of volunteers
were challenged at a moderate dose, while the next group's dose level was altered, depending on
the outcome of the previous group. That way, the median infectious doses could be captured over a
smaller dose range. The entire time required for each dose-response study was 11 to 14 months.
The infectivity for each of the three isolates was estimated using the study data and then tested using
the exponential model (Messner et al., 2001). A comprehensive dose-response evaluation was
conducted by EPA during the development of the LT2 (U.S. EPA, 2003a,b, 2006a). Teunis (2009)
analyzed all five strains and found most likely IDso appears to be in the range of 30 to 50 oocysts.
88
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 8. Brief Summary of Challenge Studies to Investigate the Dose-Response and Host-
Immunity Factors Related to Norovirus Infection
The group of viruses called norovirus (previously known as Norwalk or Norwalk-like viruses) is the
most common cause of AGI outbreaks (e.g., through ingestion of contaminated food and water) in the
United States (cite Scallan et al., 2011a,b). Reports implicated noroviruses in 94% of U.S.
nonbacterial gastroenteritis outbreaks from 1996 to 1997 (Fankhauser et al., 1998). However, host
immunity to noroviruses remains poorly characterized. Although over 70% of U.S. adults have serum
antibodies to norovirus, the antibodies do not appear to confer any protection from reinfection
(Greenberg et al., 1979). Despite outbreak studies suggesting that norovirus has high infectivity and
high person-to-person transmissibility, certain exposed people never develop illness (i.e., remain
asymptomatic).
Lindesmith et al. (2003, 2005) conducted a series of human volunteer studies to examine the dose-
response characteristics of different strains of noroviruses (Norwalk [NV] and Snow Mountain Agent
Virus [SMV]) and the role of host immunity in the probability of (re)infection. For these studies, the
researchers recruited healthy adult volunteers with and without pre-existing serum immunoglobin G
(IgG) to NV. The first study included 31 volunteers and examined 3 low doses of NV. The second
study included 15 volunteers and examined 3 doses of SMV. The inoculum was diluted in sterile water
and ingested. The volunteers stayed 5 consecutive days/6 consecutive nights at a research center
for monitoring of Gl symptoms, then reported for follow-up visits for collection of stool, serum, and
saliva samples on days 8, 14, and 21 post-challenge.
Infection was defined as detection of viral shedding in stool by reverse transcriptase PCR (RT-PCR)
or seroconversion designated by a 4-fold or more rise in the specific IgG. Symptoms that defined Gl
illness were diarrhea (defined as more than 2 unformed stools within 24 hours), vomiting, abdominal
pain, muscle pain, fatigue, and chills—fever and headache were excluded.
Previous research has reported an association of a mutation in the alpha (1,2) fucosyltransferase
gene (FUT2) gene with immunity to NV infection (Marionneau et al., 2002). In that study, volunteers
with the FUT2 mutation remained healthy and had no significant increase in anti-NV salivary antibody
titers, even after high-dose exposure. Note that about 20% of the North American population has the
FUT2 mutation. Of the volunteers with fully functioning FUT2 genes, about half became infected. In
the remaining uninfected half of the group, salivary IgA levels showed mucosal immune response
post-challenge, suggesting that previous exposure had resulted in protective immunity.
Moe et al. (2002) found that infected subjects were generally older than uninfected subjects, and were
twice as likely to have NV-specific IgG in their baseline serum specimen. Consequently, the presence
of anti-NV serum IgG was not protective against infection. In other words, although these individuals
had been exposed previously to NV, perhaps multiple times, they continued to be susceptible to
reinfection. These studies provide important implications for microbial risk assessors—even with a
very low infectious dose in susceptible populations, susceptibility to Norwalk virus is multifactorial and
influenced by both acquired immunity and genetic traits.
In subsequent work, these researchers along with Teunis et al. (2008b) developed a dose-response
relationship for NV based on challenge study data and a new variant on the hit theory model of
microbial infection. This relationship accounts for variation in NV infectivity, as well as the degree of
virus aggregation. Moreover, the results indicate that passage through a human host does not change
NV infectivity and that NV is a highly infectious microorganism.
89
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
Exponential Model
The exponential model is the simplest model that is commonly used in MRA (Text Box 9); it is
based on the following assumptions (Haas et al., 1999):
• microorganisms are distributed in water randomly21 and thus, follow the Poisson
distribution;
• for infection to occur, at least one pathogen entity must survive within the host; and
• the probability of infection (in a person or animal model) per ingested or inhaled
99
organism is constant.
Text Box 9. Summary of Use of Exponential Model (Source: Rose et al., 1991)
Rose et al. (1991) used an exponential model to estimate the risk of infection after exposure to
treated water contaminated with Giardia cysts as shown by the following equation:
P, = 1 -exp(-rp V) ,
where P, is the probability of infection, r is the host-pathogen interaction probability, |j is the average
number of organisms, and \/is the volume of water consumed.
The parameter designating the infectivity of Giardia (r) in the exponential model was based on data
from studies in which volunteers were fed a range of 1 to 106 cysts and the response was measured
by the number of cysts excreted in the volunteer feces, not by clinical symptoms (Rendtorff, 1954a,b;
Rendtorffand Holt, 1954a,b). An average rvalue (the fraction of microorganisms that are ingested
that survive to initiate infection) was compared by determining the value of rat each dose. Based on
the results, the average r was calculated to be 0.01982.
Using the exponential model, the potential risk of infection was determined with varying levels of
Giardia cysts in drinking water. The model used 2 L of unboiled water a day as the consumption
parameter, V. The number of cysts (jj) was based on densities measured in source waters with 99.9,
99.99, and 99.999% estimated removal by treatment. The exposure was based on the numbers of
cysts per L multiplied by 2L. A maximum daily risk was estimated using the highest level of
contamination and a yearly risk was based on 365 days of exposure to the geometric mean density
of cysts. The model was checked for plausibility by entering the data from five waterborne giardiasis
outbreaks using the levels of Giardia cysts and the observed attack rates in the exposed population.
Under the exponential model, there is no minimum infectious dose, as a nonzero risk is predicted
with any non-zero dose. Assuming that a single organism is sufficient to cause infection, and that
the ingested organisms must pass through "multiple barriers" to survive long enough to cause
disease, yields the exponential risk model:
P.= 1 - e
-rD
[4-1]
21 As noted previously, because microbes are generally not thought to be distributed randomly in environmental
media, this assumption is considered to be a limitation of the exponential model unless adjustments are made as
discussed in Teunis and Havelaar (2000).
22 This assumption also introduces uncertainty because host variation is not considered.
90
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Where:
• Pr is the probability of an individual exhibiting the response specified by dose-response
parameter given exposure to the defined dose (unitless);
• D is the mean dose of microbes for a group in the clinical trial (number of microbes
ingested per event, which is often represented as daily water intake x density);
• r is the dose-response parameter that is "fit" to the data; higher value indicates higher risks
at lower doses23; and
• e is the base of the natural logarithm function (unitless).24
The "response" to exposure may be the development of clinical symptoms, and/or microbiological
or immunological evidence that microbes have persisted or multiplied in the body. For example,
oocyst shedding (regardless of whether illness symptoms are present) in stool is commonly used
to indicate Cryptosporidium infection.
Under this model the median infectious dose is N50 = ln(2)/r = 0.693/r (where 1 is the natural
logarithm).
Beta-Poisson Model
The beta-Poisson model is based on similar assumptions to the exponential model except that the
third assumption (that the probability of infection per ingested organism is constant) is relaxed.
This model allows the probability of infection per ingested or inhaled organism to vary within the
exposed population (Haas et al., 1999). In this model the probability of surviving and reaching a
host site (r in the exponential model) is beta distributed, and thus the model contains the two
parameters (a and P) of the beta distribution. Thus, the beta-Poisson accounts for differential
immunity in a population (but not specifically for differences between groups or subgroups in a
population). The exponential model generally provides a good fit to experimental data if the
infectivity of the administered organisms and the inherent susceptibility of the exposed population
(animal or human) are constant. However, when there is variability in the host-pathogen
interaction, diversity in the pathogen (as when multiple strains are present), or both, the dose-
response relation tends to be shallower than that of the exponential relation. The most commonly
used approximation to the beta-Poisson model is as follows:
Pr = l-(l+D/p)"a [4-2]
Where:
• Pr is the probability of an individual exhibiting the response specified by dose-response
parameter given exposure to the defined dose (unitless);
• D is the mean dose of microbes ingested (number of microbes ingested per event, which
is often represented as daily water intake times density);
23 For small values of r, the estimated individual risk is rxd (i.e., the model is linear at low doses).
24 Euler's number or Napier's constant is -2.71828.
91
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
• pis the location parameter; determines inflection point of dose-response curve (unitless);
and
• a is the shape parameter governing the steepness of the dose-response curve (unitless).
Unfortunately, in this approximation to the beta-Poisson model, a does not have an obvious
physical interpretation. What can be said is that it is a shape parameter governing the steepness of
the dose-response curve; the larger its value, the steeper the curve (McBride et al., 2002). The
derivation of the approximation to the beta-Poisson model, as shown above requires that J3»l, P
» a, and becomes a poorer approximation at small values of P or large values of D. In practice,
this condition is not always met, and caution is warranted, especially for uncertainty calculations
at low doses when this approximation is used (Teunis and Havelaar, 2000). This approximation to
the beta-Poisson is linear at low doses and the curve is always shallower than the exponential
model. However, as a approaches go, the approximate beta-Poisson model approaches the
exponential model (Haas et al., 1999).
Under this model the median infectious dose is Nso= P x (21/
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
5.3.1). Confidence intervals around the maximum likelihood estimates can be calculated using
approximations valid for large samples. For small sample sizes that are typical in MRA, bootstrap
confidence intervals are calculated by randomly resampling from the original dose-response data
and estimating the parameters for each of these bootstrap samples. However, if the sample sizes
are too small, bootstrapping becomes less desirable (Teunis and Havelaar, 2000).
Bayesian methods, which are discussed in more detail in Appendix C, exploit available subjective
and related information in addition to the numeric data. Ideally, the investigator expresses an initial
assessment of the unknown parameter distribution before examining the data by defining a prior
probability distribution for the parameters. The prior probability distribution is defined based on
subjective information and professional judgment. Recently published MRAs have used a "non-
informative" prior distribution to represent the lack of prior information. Using Bayes' rule, the
posterior probability distribution for the parameters given the data can be calculated. In a Bayesian
analysis, uncertainty intervals for the parameters and the dose-response function can be calculated
from the posterior distribution as "credible intervals"; for example, a 95% credible interval has a
95% probability of including the parameter value, given the data.
The MCMC method is often used to simulate values from a posterior probability distribution for
which direct analytical calculations are difficult, intractable, or inconvenient. Gilks et al. (1996)
provides a good description of these methods. Instead of being statistically independent, the
consecutive values form a Markov Chain so that the statistical distribution for one value depends
upon the previous value. After a sufficiently long "burn-in" period, every kth value is sampled,
giving an approximately random sample from the posterior distribution. It is unnecessary to know
the normalizing constant that makes the distribution integrate to one. A version of the Metropolis-
Hastings algorithm (Hastings, 1970; Gilks and Wild, 1992; Gilks et al., 1996) is used at each step
to simulate from the posterior distribution without knowing the normalizing constant.
An advantage of the Bayesian approach over the frequentist approach is the ability to incorporate
prior information. However, for the MRAs in the current literature this is not very useful because
the prior information is too limited and non-informative priors have been used. The subjective
nature of the choice of prior distribution is often thought to be a disadvantage of the Bayesian
approach. A more important advantage of the Bayesian approach is that, unlike frequentist
confidence intervals, the uncertainty intervals from a Bayesian analysis are easier to interpret and
are usually interpreted correctly. Furthermore, the Bayesian uncertainty estimates of dose-response
functions are generally easier to calculate and more exact than the frequentist confidence intervals.
Finally, Bayesian methods are well-suited to meta-analysis of multiple studies, pathogens, or
populations.
A predictive Bayesian dose-response function can be developed as follows. First, the parametric
form of the dose-response function is established by theoretical derivation and, if possible,
empirical confirmation. Then all available knowledge, other than the theoretical form of the
conditional distribution and empirical data already used for that purpose, is considered during
estimation of the parameters of the distribution. To do this, the parameters are recognized as
uncertain but subj ect to professional judgment, and thus, a prior probability distribution is assigned
to each parameter. Prior distributions are then refined with dose-response data, to obtain a posterior
distribution. Next, the predictive Bayesian dose-response function can be found by multiplying the
93
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
posterior by the conditional dose-response function and integrating over the parameter space
(Englehardt, 2004). As noted previously, MCMC methods can then be used to generate samples
from the joint posterior distribution (Messner et al., 2001).
Several researchers advocate for the combined use of Bayesian and frequentist (likelihood-based)
methods (Teunis and Havelaar, 2000; Messner et al., 2001). Often the frequentist approach is used
to provide maximum likelihood estimates of the dose-response function, and the Bayesian
approach is used to calculate uncertainty intervals (e.g., 80% or 95% credible intervals for the
parameters or the dose-response). Several papers use the Bayesian posterior mode to select the
dose-response function (Teunis et al., 2004, 2005, 2008a,b). The posterior mode is given by the
parameters that maximize the posterior probability, defined as the product of the prior and the
likelihood; thus, it is not necessary to calculate the normalizing constant for this calculation.
Other Dose-Response Methods
In addition to the dose-response models described above (exponential, beta Poisson), there are
other dose-response models either in use in QMRAs or that can be potentially incorporated into
QMRAs. These models include empirical dose-response models, threshold models, and
mechanistic models of varying resolution. The exponential and beta-Poisson models are
distinguished from empirical models because their derivation is based on a sequence of plausible
events, although this assessment is not universal (e.g., see Coleman and Marks, 1998). Threshold
models have in some cases provided significant improvements in fit over the exponential and beta-
Poisson models, but their use has been advocated on the basis of analysis of the infection process
and interpretation of epidemiological data. Mechanistic models are currently in development and
offer the potential for development of dose-response models for pathogens for which dose-
response data are unavailable of for the low-dose range. These models depict the pathogen-host
system in varying resolutions and may be stochastic, deterministic, or a combination. Given the
widespread use of the exponential and beta-Poisson dose-response models for waterborne
pathogens and the advantages these models offer (as described above), these alternative empirical,
threshold, and mechanistic models are not presented in the body of this report, but are described
and contrasted with the exponential and beta-Poisson model in Appendix C.
4.2.2. Summary of Available Dose-Response Relationships for Waterborne
Pathogens
An overview of the many of the currently available and generally accepted dose-response
relationships for waterborne pathogens is summarized in Table 10, which includes the pathogens
listed alphabetically, the resulting dose-response form and parameter values, and the
corresponding reference for that work. Note that in some cases reported relationships are also
provided for illness conditional on infection. Use of these conditional relationships is typically not
universal in MRA modeling, as often illness is modeled as variable independent of dose.
94
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 10. Overview of Dose-Response Relationships and Health Effects for Waterborne
Microorganism
Model
Parameters'3
Health Effects
Reference(s)
Adenovirus 4
Exponential
r = 0.4172°
Acute infectious
Crabtree et al., 1997
nonbacterial Gl illness,
Haas et al., 1999
acute febrile
APHA, 2004
respiratory disease,
acute hemorrhagic
conjunctivitis, phary-
noconjuctival fever
Campylobacter Beta-Poisson a = 0.145 (3 = 7.59
jejunihi
Infection: a = 0.024 (3 = 0.011
Hypergeometric
beta-Poisson (for
healthy adults)
Illness: Conditional
on infection (for R = 2.44x10s rp 3.63x10 s
children)9
Diarrhea, abdominal
pain, malaise, fever,
nausea, and vomiting
(typhoid-like
syndrome), febrile
convulsions, meningeal
arthritis, reactive
arthritis, Guillain-Barre
syndrome)
Medema et al., 1996
Teunis et al., 1996
Haas et al., 1999
APHA, 2004
Teunis et al., 2005
Coxsackievirus
Exponential
r= 0.0145
Vesicular pharyngitis
(acute self-limited, viral
disease characterized
by sudden onset, fever,
sore throat and small
pharyngeal lesions)
APHA, 2004
Haas et al., 1999
Cryptosporidiumd Exponential
r = 0.0042 Iowa isolate
r= 0.077
Cryptosporidiosis:
profuse watery
diarrhea, malaise,
fever, anorexia,
nausea, and vomiting
r = 0.0572 TAMU isolate
r = in the range 0.04 to
0.16 for unknown mixture
environmental isolates
Beta-Poisson
a = 0.27 (3 = 1.40 TU502
isolate
a = 0.114 (3 = 1.04
Moredun isolate
a = 0.145(3 = 1.52 UCP
isolate
APHA, 2004
Haas et al., 1996,
1999
Okhuysen et al.,
1999
http://wiki.camra.
msu.edu
U.S. EPA, 2006a,
Dupont et al. 1995,
Okhuysen et al.,
1999, 2002,
Chappell et al., 2006
httpV/wiki.camra.
msu.edu
Generalized beta-
Poisson for Illness
a = 0.060 (3 = 0.095
Englehardt and
Swartout, 2006
95
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 10. Overview of Dose-Response Relationships and Health Effects for Waterborne
Microorganism
Model
Parameters'3
Health Effects
Reference(s)
Echovirus 12j
Exponential
r= 0.0128
Acute Febrile
respiratory disease
Haas et al., 1999
APHA, 2004
Beta-Poisson
a = 0.401 (3 = 227.2
Teunis et al., 1996
a = 0.374 (3 = 186.7
Regli et al., 1991
Rose and Sobsey,
1993
a = 1.3 (3 = 75
Rose and Gerba,
1991
Endamoeba coli
Beta-Poisson
a = 0.1008 (3 = 0.3522
Not a human pathogen
Haas et al., 1999
Escherichia coli
(pathogenic
strains)
Beta-Poisson
a = 0.1778 (3 = 1.78x10®
Acute watery diarrhea
Haas et al., 1999
APHA, 2004
E. coli 0157:H7
Beta-Poissone
a = 0.248 (3 = 48.80
Diarrhea (bloody),
severe abdominal
APHA, 2004
Teunis et al., 2008a
Hypergeometric
beta-Poisson
a = 0.084 (3 = 1.44
(children)
a = 0.050 (3 = 1.001
(adults)
cramping, headache,
hemorrhagic colitis,
and hemolytic uremic
syndrome
Teunis et al., 2004
Giardia lamblia
Exponential
r= 0.0199
Giardiasis: diarrhea
(chronic); abdominal
cramps; bloating,
frequent loose, pale,
greasy stools; fatigue;
malabsorption
Regli et al., 1991
Rose and Gerba,
1991
Rose et al., 1991
Teunis et al., 1996
Haas et al., 1999
APHA 2004
Hepatitis A virus
Exponential
r = 0.5486f
Hepatitis: acute
inflammation of the
liver
Haas et al., 1999
Legionella
Exponential
r= 0.06
Legionellosis,
pneumonia,
Legionnaire's disease,
Pontiac fever
Armstrong and
Haas, 2008
Noro virus
Infection:
a = 0.040 (3 = 0.055
Usually self-limited,
APHA 2004
Hypergeometric
mild to moderate
Teunis et al., 2008b
function 1F1 (note:
disease with clinical
Messner et al., 2014
if aggregation is
symptoms of nausea,
different than
vomiting, diarrhea,
reported then 2F1
abdominal pain,
function is
myalgia, headache,
needed)
malaise, low grade
Illness: Conditional
fever, or a combination
on Infection9
r|= 2.55><10"3 r= 0.086
of these symptoms
96
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 10. Overview of Dose-Response Relationships and Health Effects for Waterborne
Microorganism
Model
Parameters'3
Health Effects
Reference(s)
Poliovirus III
Beta-Poisson
a = 0.409 (3 = 0.788
Non-specific fever,
acute flaccid paralysis
Rose and Sobsey,
1993
APHA, 2004
a = 0.409 (3 = 0.788
Regli et al., 1991
II
CO-
LO
O
II
ts
Rose and Gerba,
1991
Rotavirus
Beta-Poisson
a = 0.26 (3 = 0.42
Sporadic, seasonal,
often sever
gastroenteritis of
infants and young
children, characterized
Gerba et al., 1996b,
Haas et al., 1999
Regli et al., 1991
Rose and Sobsey,
1993
a = 0.232 (3 = 0.247
by vomiting, fever and
watery diarrhea
Rose and Gerba,
1991
APHA, 2004
Hypergeometric
beta-Poisson
a = 0.167 (3 = 0.191
Teunis and
Havelaar, 2000
Salmonella spp.
Beta-Poisson
a = 0.33 (3 = 139.9
Gastroenteritis (enteric
fever and septicemia)
Rose and Gerba,
1991
APHA, 2004
Gompertz log
(illness)
ln(a) in the range 29 to 50
b = 2.148
Coleman and Marks,
2000
Coleman et al., 2004
Soller et al., 2007
Salmonella (non-
typhoid)
Generalized linear /3o = 0.323 /3i = 5.616
mixed models and ^ = _8 462 p3 = .7 7q2
d2 = 0.780
fractional
polynomials of
dosek
Beta-Poisson a = 0.3126 (3 = 2885
Bollaerts et al., 2008
Haas et al., 1999
Bayesian mixed
model:
Infection:
Hypergeometric
function 2F1
Illness: Conditional
on Infection
Teunis et al. 2010
a = 0.00853 (3 = 3.14
(note 5000 samples of
model parameters
available from cited
author)
r|= 6.9*101 r= 8.23
Salmonella typhi
Fractional
polynomials
/3i = -18.1425
£2= 22.5300*105
Beta-Poisson
a = 0.1086 (3 = 6,097
a = 0.21 (3 = 5,531
Namata et al., 2008
Haas et al., 1999
Rose and Gerba,
1991
97
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 10. Overview of Dose-Response Relationships and Health Effects for Waterborne
Pathogensa (Source: Adapted from McBride et al., 2002) (continued)
Microorganism
Model
Parameters'3
Health Effects
Reference(s)
Shigella
Beta-Poisson
a = 0.21 (3 = 42.86
Shigellosis: acute
Haas et al., 1999
gastroenteritis,
dysentery, fever,
nausea, vomiting, and
cramps
Vibrio cholera
Beta-Poisson
a = 0.25 (3= 16.2
Profuse, watery
Haas et al., 1999,
diarrhea; vomiting
APHA 2004
a Calibrations based on available data that have used particular pathogen strains processed in particular ways. Where
more than one strain of an organism has been studied in clinical trials, a wide range of infectivities can be discovered.
Therefore it must be recognized that these calibrations can carry a substantial degree of uncertainty.
b For the exponential distribution Nso= 0.693/r; for the beta-Poisson distribution Nso= (3 * (21/a -1).
c Developed for inhalation exposure to adenovirus 4 aerosols.
dOkhuysen et al. (1999) Estimated based on IDso reported for the TAMU isolate. Teunis 2009 graphs 5 clinical trials
resulting in an estimated IDso of 30-50.
e Represents a meta-analysis of seven outbreaks and adjusting for heterogeneity. Alpha/beta pairs derived via MCMC
analyses are available from Dr. Teunis. Use of those pairs is preferred to the use of the values shown in this table
'Corresponding dose units are grams of feces.
g Dose-response relation forthe conditional probability of illness in infected subjects = 1 - (1+ r|CV) r, where r| Dand r
are shown in the table; CV is the dose (concentration * volume).
h An alternate dose-response model is proposed by Brynestad et al. (2008). That model is not included in Table 10;
however, it is described along with other empirical models in Appendix C.
' Coleman and Marks (2004) suggest the dose-response models for Campylobacter identified in this table do not
account for strain variability sufficiently and suggest the need for development of more detailed mechanistic models.
1 Note that the dose-response models for Echovirus 12 yield very different IDso values of -54 or -1053 depending on
whether or not a factor of 33 is applied to the dose range. Teunis et al. (1996) fitted a beta-Poisson model to the
echovirus 12 clinical trial data (Schiff et al., 1984) for 149 volunteers given doses of 0, 330, 1000, 3300, 10,000,
33,000, and 330,000 PFU. This model gives IDso « 1052. Haas (1983) fitted a simple exponential model forthe same
virus to a set of clinical trial data reported by Akin (1981), in which 60 volunteers were given doses of 10, 30, and 100
PFU. That model, gives IDso « 54 (Haas et al., 1999). The Akin data appears to be a preliminary subset of the Schiff
data and there was only one clinical trial.
k Derived based on a series of foodborne outbreaks; not necessarily valid for water matrices or exposures.
4.3. Host-Pathogen Profile and Linkage between Human Health
Effects Assessment and Other MRA Components
The host-pathogen profile is a distillation of the most important information and analyses that are
conducted during the human health effects assessment. The host-pathogen profile can provide,
depending on the available data, a qualitative and/or quantitative description of the human health
effects scenario (ILSI, 2000). An assessment of the assumptions made during the human health
effects assessment, and the uncertainty associated with the assessment because of lack of
knowledge about the scenario or insufficient experimental or epidemiological data, should be
presented. Any assumptions based on scientific judgment should be described and justified in the
host-pathogen profile. A summary of the quantitative or qualitative uncertainty analysis should
also be included.
Thus, the host-pathogen profile serves as the critical linkage from the human health effects
assessment to the exposure assessment. The iterative nature of risk assessment requires that the
host pathogen profile and the exposure profile be critically evaluated by the risk assessors and
98
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
managers to determine if the problem formulation component of the risk assessment needs to be
revisited and refined based on the availability of relevant data presented in these profiles.
99
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
100
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
5. Risk Characterization
As noted throughout the preceding chapters, risk assessment is an iterative process. During risk
characterization, the results of this iterative risk assessment process are integrated and
documented. Thus, risk characterization is the culmination of the MRA process, and the final
integrative step. The risk characterization needs to be complete, informative, and useful for
decision-makers. The agency's Science Policy Council Handbook: Risk Characterization (U.S.
EPA, 2000b) describes risk characterization as the step that "integrates information from the
preceding components of the risk assessment and synthesizes an overall conclusion about the risk
that is complete, informative, and useful for decision-makers" (U.S. EPA, 2000b, 2012a).
Risk characterization forms the starting point for formulating risk management considerations and
provides a foundation for (regulatory) decision-making. It characterizes both quantitative and
qualitative data in technical and non-technical terms, explaining the extent and weight-of-
evidence, results, and major points of interpretation and rationale. It also summarizes the strengths
and weaknesses of the evidence, conclusions, uncertainties, variability, potential impact of
alternative assumptions, and discusses scenario, model, parameter, and analysis options that may
deserve further consideration as the results from the assessment are subsequently used for decision-
making purposes.
EPA's policy statement on risk characterization (U.S. EPA, 2000b) is as follows:
Each risk assessment prepared in support of decision-making at EPA should include a risk
characterization that follows the principles and reflects the values outlined in this policy.
A risk characterization should be prepared in a manner that is clear, transparent, reasonable
and consistent with other risk characterizations of similar scope prepared across programs
in the Agency. Further, discussion of risk in all EPA reports, presentations, decision
packages, and other documents should be substantively consistent with the risk
characterization. The nature of the risk characterization will depend upon the information
available, the regulatory application of the risk information, and the resources (including
time) available. In all cases, however, the assessment should identify and discuss all the
major issues associated with determining the nature and extent of the risk and provide
commentary on any constraints limiting fuller exposition.
5.1. Introduction to Risk Characterization
The Agency's risk characterization policy (U.S. EPA, 2000b) calls for a transparent process and
documentation that is clear, consistent, and reasonable. TCCR is particularly relevant for risk
characterization because a risk assessment is often judged by the extent to which the risk
characterization achieves the principles of TCCR. This section provides a summary overview that
is intended to provide risk assessors, risk managers, and other decision-makers an introduction to
the goals and principles of risk characterization. More comprehensive documentation on the topic
of risk characterization has been prepared by the Agency and interested readers are referred to that
documentation (U.S. EPA 2000b). This document complements and extends that previous work
by discussing tools, methods, and issues specific to microbial contaminants in water and water-
related media.
101
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
The purpose of the risk characterization is to summarize the event of interest according to its
nature, severity, and consequences. The risk characterization also frames the risk assessment
results within the context of the problem formulation elements, specifically the nature of the
concern, the purpose and objectives, the history and context within the agency, and the questions
the risk assessment was designed to answer. For example, the risk management options defined
during problem formulation can be used to develop risk estimates with and without proposed
control measures. A discussion of the most sensitive variables (sensitivity analysis), or the
variables with the largest contribution to the overall uncertainty in the risk estimate, may provide
risk managers with insights that can be used for future resource allocation for developing risk
mitigation strategies. As new data become available or as risk managers ask new questions, the
problem formulation and risk assessment can be revisited and revised as needed and appropriate.
Discussions of variability, uncertainty, and identified gaps in the knowledge base should be
reiterated from the discussions presented in the problem formulation.
Information from the exposure and health effects components of the risk assessment should be
integrated to arrive at conclusions for the microbial risk assessment. The key issues that affect the
results should be summarized and put into context. For example, the risk characterization includes
a discussion and quantifications (to the extent possible) of (1) the uncertainties associated with the
analysis and key components; (2) the variability associated with key inputs to the model(s); (3) the
confidence in the resulting risk estimates through a weight-of-evidence discussion; (4) the
limitations of the analysis; and (5) the plausibility of the results. A candid and open discussion of
the uncertainty in the overall assessment, in each of its components, and related estimates of risk
is critical to a full characterization of risk. Uncertainty and sensitivity analyses are often conducted
to develop the information needed for this purpose (see Section 5.3). As the assumptions,
approaches, and conclusions of the risk assessment are presented, the strengths and limitations
should also be discussed.
The assessment of data quality should be part of a risk characterization. Whenever possible, the
data that are used should be both relevant and of high quality; however, it should be understood
that the quality of available information will vary substantially. A candid discussion of the quality
of the data employed should be provided, including how the data quality pertains to variability and
uncertainty. Sufficient detail should be provided so that the assessment can be duplicated by others.
A discussion (at least in a qualitative manner) of how a specific risk compares with similar risks
and discussion of the plausibility of the risk scenarios ("ground truthing") is valuable for TCCR.
This may be accomplished by comparisons with other pollutants or situations on which the Agency
has already decided to act or for other relevant situations. The discussion should highlight the
limitations of such comparisons as well as the relevance of the comparisons. Refer to Table 1 for
a concise summary of the elements described above.
Risk characterization is essential for managers who are evaluating public-health concerns,
considering regulatory and technologic decisions, and setting priorities for research and funding
(NRC, 2009). For example, the NRC (2009) identified the following questions as being central for
risk characterization: (1) What is the nature and magnitude of risk associated with existing
conditions? (2) What risk decreases (benefits) are associated with each of the options? (3) Are any
risks increased? and (4) What are the significant uncertainties?
102
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
5.1.1. Histo rical Context
The first significant reference to risk characterization is found in the 1983 NRC publication titled
Risk Assessment in the Federal Government: Managing the Process (commonly referred to as the
"Red Book"). In that seminal work, the NRC defined risk characterization as
.. .the process of estimating the incidence of a health effect under the various conditions of
human exposure described in exposure assessment. It is performed by combining the
exposure and dose-response assessments. The summary effects of the uncertainties in the
preceding steps are described in this step.
Since its publication, the concept of risk characterization evolved within EPA and also more
broadly within the U.S. Federal government. Concerns over adequately characterizing risk to
maintain the public's perception of and confidence in EPA's risk assessments resulted in a 1992
Agency-wide policy for risk characterization, which stated that "...scientific uncertainty is a fact
of life (and)...a balanced discussion of reliable conclusions and related uncertainties enhances,
rather than detracts, from the overall credibility of each assessment..(U.S. EPA, 1995a).
In 1997, the Presidential Commission on Risk Assessment and Risk Management noted that "risk
characterization is the primary vehicle for communicating health risk assessment findings," but
concluded that the difficulty in communicating risk effectively impedes the risk management
process (Omenn et al., 1997).
Risk characterization at EPA is considered to be a conscious and deliberate process to bring all
important considerations about risk (the likelihood of the risk and also the strengths and limitations
of the assessment) and a description of how others have assessed the risk into an integrated picture.
Based on the experiences across the Agency between 1995 and 2000, a single Agency-wide
document was determined to be needed. The Risk Characterization Handbook (U.S. EPA, 2002b)
was developed to respond to that need and remains current. However, the Risk Characterization
Handbook indicates that Agency offices may wish to prepare tailored guidance that meets their
individual needs to supplement and remain consistent with the information in the Handbook. This
MRA Tools document fills one such need as the field of MRA has evolved rapidly over recent
decades.
5.2. Risk Estimation and Risk Description
The risk estimation is the compilation of the types and magnitude of effects anticipated from
exposure to the microbe or medium and can be qualitative or quantitative depending on the data
and methods used. The risk description involves summarizing the event of interest according to its
nature, severity, and consequences. The risk description should also explicitly state whether and
how well both the statement of concern, statement of purpose, and objectives that were identified
in the problem formulation were addressed. It is this description that is the synthesis of all of the
previous components conducted within scope of the assessment.
The results from the exposure assessment (which may have involved a detailed exposure model)
can be expressed as the number of organisms to which an individual is exposed in a defined amount
103
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
of time and/or for a certain consumption rate, and can include single or repeated exposures. The
results from human health effects assessment (which may have involved a dose-response model)
can be expressed as the probability of individual infection or illness after a certain number of
organisms are ingested. The risk estimation can be expressed as an individual risk estimate (e.g.,
probability of illness = 0.001) or as a population level risk estimate (100 illnesses per year within
a population of 100,000 individuals). As described in further detail in the problem formulation,
the risk estimation can also be modeled to consider time-dependent elements such as secondary
(person-to-person) transmission, host immunity, and multiple routes of exposure (ILSI, 2000).
5.3. Uncertainty and Sensitivity Analysis
The terms sensitivity, uncertainty, and variability are terms of art in risk assessment, but may also
be used in the vernacular by risk managers. Risk assessors should be aware that risk managers
might be looking for a more qualitative answer when they ask "how sure are you?" Sensitivity
analysis is evaluating which parameter in the risk assessment has the most impact on the results.
In other words the results are sensitive to adjustments to the parameter. The parameters that impact
the results the most would usually be considered the more important parameters in the risk
assessment. For each parameter there is variability, which cannot be reduced by the collection of
more data, and uncertainty, which can be reduced by the collection of more data. It is not always
possible to separate variability and uncertainty. However, sensitivity analysis can capture both the
variability and uncertainty. Variability and uncertainty in the input parameters results in overall
uncertainty in the risk assessment output, which is often reflected by the risk estimate being
expressed as a range. Risk managers find discussions of uncertainty and sensitivity most useful
when they transparently describe the major sources of uncertainty and how significant those
sources of uncertainty are with respect to the results of the risk assessment. Risk managers are also
interested in which data gaps, if filled, would most improve the risk assessment.
Uncertainty analysis "is the computation of the total uncertainty induced in the output by
quantified uncertainties in the inputs and models..." (Morgan and Henrion, 1990). It is a key
concern for risk managers because uncertainty analysis provides information about the overall
reliability of the risk estimates. Measures of model uncertainty communicate to risk managers the
risk assessor's best judgment as to the overall quality of the numerical risk estimates generated by
the MRA. Confidence intervals, "credible ranges" developed through Monte-Carlo analyses,
Bayesian analyses, and other measures of dispersion in risk estimates, must be presented clearly,
and their meaning communicated clearly. Similarly, clear graphical or tabular presentations are
very useful. To the extent that intermediate calculations add value and understanding to the results,
they can also be included.
Key assumptions related to model selection,25 input data, and parameters should be provided and
discussed, as well as their implications for the model results and uncertainty. In many risk
assessments, assumptions and rough estimates for input values and/or uniform and triangular
distributions are used to account for uncertainties in input values that cannot be easily quantified.
Any conservative assumptions that are built into the model should be explained, and the effect of
using less conservative assumptions should be discussed.
25
One method that has been used to evaluate this source of uncertainty is model averaging (U.S. EPA, 2006a).
104
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Variability and uncertainty can be captured in risk assessments through the use of distributions
instead of point estimates. For example, if the input parameters are distributions of values, then
the risk assessment software can select a number randomly from each parameter's distribution and
calculate risk for that set of numbers, then the random selection of values and calculation of risk
is repeated thousands of times. The resulting set of numbers is a distribution itself, and can be
considered the risk estimate or output of the risk assessment.
Evaluating the effect of known sources of variability in model outputs can be done through one or
more forms of sensitivity analysis. Sensitivity analysis "is the computation of the effect of changes
in input values or assumptions (including boundaries and model functional form) on the outputs"
(Morgan and Henri on, 1990). These analyses provide an opportunity for assessing the form of the
models comprising the QMRA and can be used to identify the parameters to which the risk
estimates are most sensitive. Knowledge of the parameters driving the risk estimates provides
opportunities for effective risk management. Sensitivity analyses techniques range from simply
conducting a small number of additional model runs with different parameter values to performing
a fully probabilistic evaluation of the effects of variations in parameter values on model outputs
(e.g., using a Monte Carlo approach). The specific approach taken will depend on the nature of the
data and models supporting a given assessment.
While sensitivity analyses are useful for evaluating the effects of the variability in single
parameters on risk estimates, when multiple parameter values vary, the results of sensitivity
analyses must be interpreted cautiously (U.S. EPA, 1997a). If the variations in parameter values
are independent of one another, it is easy to overestimate the effect of varying more than one value,
because using upper or lower percentile values for more than one variable can yield point estimates
of risk that are overly conservative or insufficiently protective. If the variability in risk parameters
is correlated, the effect of their variations may not be easy to estimate using sensitivity analysis.
In such cases, a more detailed and comprehensive analysis may be required, usually employing
probabilistic approaches such as Monte Carlo or related simulation techniques. Where the
variability in model parameters can be partitioned into components mainly reflecting variability
and uncertainty, "two-dimensional" Monte Carlo analysis can be employed to estimate the relative
importance of these two components. (Refer to U.S. EPA, 2006 for an excellent example of a two-
dimensional Monte Carlo analysis.) Monte Carlo analysis and the usual "diagnostics" that it
generates can also be used both to estimate the overall precision in model outputs and to identify
those input parameters that contribute the most to the overall variability in the risk estimates (U.S.
EPA, 1997b; FAO/WHO, 2003; Frey et al., 2004). The EPA Exposures Factors Handbook (U.S.
EPA, 1997a, 2011) provides several approaches to quantitative uncertainty and sensitivity analysis
(see Table 11).
105
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 11. Approaches to Sensitivity and Uncertainty Analysis Recommended in EPA's
Exposure Factors Handbook (Source: U.S. EPA, 1997a)
Approach
Description
Example
Sensitivity analysis
Changing one input variable at a time while
leaving others constant to examine affect on
output
Fix each input at lower (then
upper) bound while holding
others at nominal values (e.g.,
medians)
Analytical uncertainty
propagation
Examining how uncertainty in individual
parameters affects the overall uncertainty of
the exposure assessment
Analytically or numerically
obtain a partial derivative of the
exposure equation with respect
to each input parameter
Probabilistic uncertainty
analysis
Varying each of the input variables over
various values of their respective probability
distributions
Assign probability density
function to each parameter;
randomly sample values from
each distribution and insert into
the exposure equation (Monte
Carlo simulation)
Classical statistical
methods
Estimating the population exposure distribution
directly, based on measured values from a
representative sample
Compute confidence interval
estimates for various percentiles
of the exposure distribution
In addition, Morgan and Henri on (1990) discuss in detail four techniques for sensitivity and
uncertainty analysis:
• deterministic: one-at-a-time analysis of each factor holding all others constant at nominal
values;
• deterministic joint analysis: changing the value of more than one factor at a time;
• parametric analysis: moving one or a few inputs across reasonably selected ranges such
as from low to high values in order to examine the shape of the response; and
• probabilistic analysis: using correlation, rank correlation, regression, or other means to
examine how much of the uncertainty in conclusions is attributable to which inputs.
EPA's Guiding Principles for Monte Carlo Analysis (U.S. EPA, 1997b) provides guidance on
selecting and developing the conceptual and mathematical models, selecting and evaluating input
data and distributions, evaluating variability and uncertainty, and presenting the results of Monte
Carlo analysis. In addition to a policy statement for the use of probabilistic analysis in risk
assessment at EPA, eight "conditions for acceptance," which are also reflected throughout this
MRA document, are outlined and reproduced below:
1. The purpose and scope of the assessment should be clearly articulated in a problem
formulation section that includes a full description of any highly exposed or highly
susceptible subpopulations evaluated (e.g., children, the elderly). The questions the
assessment attempts to answer are to be discussed and the assessment endpoints are
to be well defined.
2. The methods used for the analysis (including all models used, all data upon which
the assessment is based, and all assumptions that have a significant impact upon the
results) are to be documented and easily located in the report. This documentation
is to include a discussion of the degree to which data used are representative of the
106
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
population under study. Also, this documentation is to include the names of models
and software used to generate the analysis. Sufficient information is to be provided
to allow the result of the analysis to be independently reproduced.
3. The results of the sensitivity analysis are to be presented and discussed in the report.
Probabilistic techniques should be applied to the pathways and factors of
importance to the assessment, as determined by sensitivity analyses or other basic
requirements of the assessment.
4. The presence or absence of moderate to strong correlations or dependencies
between the input variables is to be discussed and accounted for in the analysis,
along with the effects these have on the output distribution.
5. Information for each input and output distribution is to be provided in the report.
This includes tabular and/or graphical representations of the distributions (e.g.,
probability density function and cumulative distribution function plots) that
indicate the location of any point estimate of interest (e.g., mean, median, 95th
percentile). The selection of distributions is to be explained and justified. For both
the input and output distributions, variability and uncertainty are to be differentiated
where possible.
6. The numerical stability of the central tendency and the higher end (i.e., tail) of the
output distributions are to be presented and discussed.
7. Calculations of exposures and risks using deterministic (e.g., point estimate)
methods are to be reported if possible and/or appropriate. Providing these values
will allow comparisons between the probabilistic analysis and past or screening
level risk assessments. Further, deterministic estimates may be used to answer
scenario specific questions and to facilitate risk communication. When
comparisons are made, it is important to explain similarities and differences in the
underlying data, assumptions, and models.
8. Since fixed exposure assumptions (e.g., exposure duration, body weight) are
sometimes embedded in the toxicity metrics (e.g., Reference Doses, Reference
Concentrations, unit cancer risk factors), the exposure estimates from the
probabilistic output distribution are to be aligned with the toxicity metric.[26]
The USDA (Frey et al., 2004) also identified several sensitivity analytical techniques useful for
MRA (Table 12). Although the USDA study focused on assessing microbial risks associated with
food processing, the general approaches summarized in Table 12 are also applicable to MRAs for
other media, including water and water-related media. The methods range from simple and
intuitive (varying input values across their observed ranges, scatter plots) to more complex
statistical procedures (e.g., classification and regression tree [CART]). For any given risk
assessment, it is likely that more than one of these methods will be useful for sensitivity analysis.
26 Note that acceptance of condition 8 is mainly relevant for chemical risk assessments and might not be relevant for
MRA because defaults that apply to all MRAs have not been developed.
107
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 12. Sensitivity Analysis Methods and Techniques (Adapted from Frey and Patil,
2002; Frey et al., 2004)
Sensitivity
Analysis Type
General
Description
Techniques
Description
Mathematical
Quantification of
the variation in
model output with
the range of
variation of an
input. Typically
involves systematic
variation of input
parameters,
evaluation of
model, and
assessment of the
influence of the
input parameters
on the model
output.
Nominal Range
Sensitivity
Analysis (NRSA)
Variation of individual inputs over their range while
holding all other inputs at their nominal values.
Sensitivity is assessed via comparison of model
outputs corresponding to the range of values. When
model output is probability, the difference in log odds
ratio (ALOR) method may be preferred.
Differential Variation of individual input in small range near
sensitivity central tendency values. Sensitivity is assessed
analysis (DSA) based on variation in model output in the range
around the central tendency.
Automatic
differentiation
(AD)
Difference in log
odds ratio
(ALOR)
This method is similar to DSA, except sensitivity is
assessed based on numerical partial derivatives for
the variation in model output with changes in input
parameters.
Similar to NRSA, except sensitivity is assessed via
the ALOR, where
/)(cvcnt | w ith changes in input)
ALOR = In
/>(Not event | w ith changes in input)
j , /)(cvcnt | w/out changes in input)
/)(not event | w/out changs in input)
Worst-case
determination
Similar to the ALOR approach, quantifies sensitivity to
a factor via a factor sensitivity ratio given by
FSk = log
Nk (extreme )
v N>, (average ) y
where k refers to the factor, N is the output (e.g.,
dose in the study conducted by Petterson et al.,
(2006)) and extreme and average refer to worst-case
and baseline values.
Break-point Search for values of inputs at which decision-makers
analysis would be indifferent between two or more risk
management options.
Statistical
Inputs to models
are assigned
probability
distributions and
sensitivity is
assessed via the
effect of variance of
the inputs on model
output. Inputs may
be varied using
Monte Carlo
simulation, Latin
hypercube
sampling, or other
methods.
Regression Linear models (either based on known relationships
techniques or analysis of scatter plots, etc.) are developed for the
(sample dependence model output on input variables,
regression or Regression is performed on a sample of data
rank regression) generated from the model (e.g., by Latin hypercube
sampling, as demonstrated by de Vos et al. (2006).
Sensitivity to input variables may be assessed via
comparison of standard errors of regression
coefficients or via application of stepwise regression
techniques.
Analysis of ANOVA is used to determine whether there is a
variance statistical relationship between input variables and
(ANOVA) model output; in contrast to regression techniques, no
functional form for the relationship is assumed and
data may be qualitative or quantitative.
108
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table 12. Sensitivity Analysis Methods and Techniques (Adapted from Frey and Patil,
2002; Frey et al., 2004) (continued)
Sensitivity
Analysis Type
General
Description
Techniques
Description
Sample
(Pearson)
Sample correlation measures the strength of linear
association between input variables and model
correlation or
rank (Spearman)
correlation
outputs. Rank correlation is a measure of the strength
of the monotonic relationship between two random
variables.
Classification
and regression
tree (CART)
Nonparametric technique that can select from among
a large number of variables those and their
interactions that are most important in determining
the whether an outcome variable reaches a criterion
value (Sollerand Eisenberg, 2008). Output variables
are divided into classes (e.g., above and below a
criterion) and a tree of events leading to the output
variable is developed and analyzed.
Graphical
Techniques for
visualizing the
change in model
Scatter plots
Plots providing information on the relationship
between input variables and model outputs are
constructed.
outputs with
changes in model
parameters.
Conditional
sensitivity
analysis (CSA)
Evaluating (usually graphically) the effect of changes
in a subset of model inputs while other inputs are held
at fixed values.
Additional EPA references relevant to uncertainty analysis include Report of the Workshop on
Selecting Input Distributions for Probabilistic Assessments (U.S. EPA, 1999c); Guidelines for
Preparing Economic Analyses (U.S. EPA, 2000e); Using Probabilistic Methods to Enhance the
Role of Risk Analysis in Decision-making with Case Study Examples (U.S. EPA, 2009b); and the
interagency Microbial Risk Assessment Guideline (U.S. EPA/USD A, 2012) (Text Box 10).
Text Box 10. Two-Dimensional Probabilistic Risk Analysis of Cryptosporidium in Public Water
Supplies, with Bayesian Approaches to Uncertainty Analysis (Source: from case study #8 in
U.S. EPA, 2009b)
Probabilistic assessment of the occurrence and health effects associated with Cryptosporidium
bacteria in public drinking water supplies was used to support the economic analysis of the final LT2
Enhanced Surface Water Treatment Rule. U.S. EPA's Office of Ground Water and Drinking Water
conducted this analysis and established a baseline disease burden attributable to Cryptosporidium in
Public Water supplies that use surface water sources. Next, it models the source water monitoring and
finished water improvements that will be realized as a result of the Rule. Post-LT2 Rule risk is es-
timated and the Rule's health benefit is the result of subtracting this from the baseline disease burden.
Probabilistic Risk Analysis. Probabilistic assessment was used for this analysis as a means of
addressing the variability in the occurrence of Cryptosporidium in raw water supplies, the variability in
the treatment efficiency, as well as the uncertainty in these inputs and in the dose-response
relationship for Cryptosporidium infection. This case study provides an example of a probabilistic risk
assessment that evaluates both variability and uncertainty at the same time and is referred to as a two-
dimensional probabilistic risk assessment. The analysis also included probabilistic treatments of
uncertain dose-response and occurrence parameters. Markov Chain Monte Carlo samples of
parameter sets filled this function. This Bayesian approach (treating the unknown parameters as
random variables) differs from classical treatments, which would regard the parameters as unknown,
but fixed. The risk assessment used existing datasets (e.g., occurrence of Cryptosporidium and
treatment efficacy) to inform the variability of these inputs. Uncertainty distributions were characterized
based on professional judgment or by analyzing data using Bayesian statistical techniques.
109
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Text Box 10. Two-Dimensional Probabilistic Risk Analysis of Cryptosporidium in Public Water
Supplies, with Bayesian Approaches to Uncertainty Analysis (Source: from case study #8 in
U.S. EPA, 2009b)
(continued)
(Results of Analysis. The risk assessment identified the Cryptosporidium dose-response relationship
as the most critical model parameters in the assessment, followed by the occurrence of the pathogen
and treatment efficiency. By simulating implementation of the Rule using imprecise, biased
measurement methods, the assessment provided estimates of the number of public water supply
systems that would require corrective action and the nature of the actions likely to be implemented. This
information afforded a realistic measure of the benefits (in reduced disease burden) expected with the
LT2 Rule. In response to U.S. EPA's Science Advisory Board comments, additional Cryptosporidium
dose-response models were added to more fully reflect uncertainty in this element of the assessment.
5.4. Representative Examples of MRAs
There are numerous examples in the literature of well-performed MRAs that have been conducted
for a wide range of purposes. For example, MRAs have been conducted by U.S. governmental
agencies, various governmental and non-governmental agencies outside of the United States,
including the WHO, as well as by researchers investigating local, regional, and national scale
issues. It is not feasible to describe all of these MRAs, however, several examples are highlighted
below (along with citations) for interested readers.
• EPA used MRA during the development of the Interim Enhanced Surface Water Treatment
Rule and the Long Term 2 Enhanced Surface Water Treatment Rule (U.S. EPA, 2002e,
2006a). In both cases, exposure to Cryptosporidium spp. through drinking water was the
focus of the assessment.
• EPA used MRA to quantify the benefits of the Ground Water Rule (U.S. EPA, 2006b). In
this case, the MRA was conducted on viral agents. A static risk model was used to quantify
the benefits of the rule and a dynamic model was used to evaluate the potential implications
of person-to-person transmission.
• FDA conducted a quantitative assessment of relative risk to public health from foodborne
Listeria monocytogenes among selected categories of ready-to-eat foods (FDA/USDA,
2003).
• FDA conducted a quantitative risk assessment on the public health effects of pathogenic
Vibrioparahaemolyticus in raw oysters (FDA, 2005).
• USDA conducted a quantitative risk assessment of Clostridiumperfringens in ready-to-eat
and partially cooked meat and poultry products. The purpose of the risk assessment was to
(1) evaluate the public health effect of changing the allowed maximal growth of C.
perfringens during manufacturing stabilization (cooling after the cooking step) of ready-
to-eat and partially cooked meat and poultry products; and (2) examine whether steps taken
to limit the growth of C. perfringens occurring in ready-to-eat and partially cooked foods
would be adequate to protect against growth of C. botulinum (Golden et al., 2009).
• USDA conducted a comprehensive risk assessment of Salmonella enterica serotype
Enteritidis {Salmonella Enteritidis) in December 1996 in response to an increasing number
of human illnesses associated with the consumption of shell eggs. The objectives of this
risk assessment were to establish the unmitigated risk of foodborne illness from Salmonella
110
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Enteritidis, identify and evaluate potential risk reduction strategies, identify data needs,
and prioritize future data collection efforts (USDA, 1998).
• MICRORISK is a collaborative research project with an objective to develop and evaluate
a harmonized framework for quantitative assessment of the microbiological safety of
drinking water in European Union Member States. MRA plays a central role in this
research effort. Numerous scientific publications have resulted from this effort
(http://www.microrisk.com/publish/cat index 25.shtml)
• Several MRAs have been conducted by researchers to inform management options for
Australian waters. For example, QMRA was used to estimate the reduction of risk
encountered by coastal bathers from the commissioning of deepwater ocean outfalls
(Ashbolt et al., 1997), and to identify conditions in which an urban freshwater recreational
lake should be closed and re-opened based on rainfall and lake level measurements as
surrogates for likely sewage impact and lake clearance of fecal contamination (Roser et al.,
2006).
• Similar to one of the Australian examples above, MRA was used by the Ministry for the
Environment in New Zealand to form the basis of their revised recreational water criteria
(NZ MFE, 2003). Researchers have also used MRA to evaluate the potential public health
benefits associated with alternative water and wastewater treatment processes (Weir et al.
2011, Soller et al., 2003), to understand the etiologic agents causing illness during
recreational water epidemiological results (Soller et al. 2010c), and more comprehensively
understand the causes and dynamics of waterborne and foodborne outbreaks that have
occurred in the United States (Eisenberg et al., 1998; Seto et al., 2007). EPA used MRA to
understand the potential human health-based implications of various sources of fecal
contamination to recreational waters (U.S. EPA, 2010).
In addition to the examples listed above, EPA's National Homeland Security Research Center
published a Compendium of Prior and Current Microbial Risk Assessment Methods for Use as a
Basis for the Selection, Development, and Testing of a Preliminary Microbial Risk Assessment
Framework. That literature review includes summaries of 135 studies published between 1994 and
2004— 44 related to exposure assessment (oral, inhalation, and dermal), 31 related to dose-
response, and 60 related to risk characterization (U.S. EPA, 2007b).
Ill
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
112
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
6. References
This list includes the references cited in this document, including appendices (A-C).
Abad, C., Pinto, R.M., Villena, C., Gajardo, R., and Bosch, A. 1997. Astrovirus survival in
drinking water. Applied and Environmental Microbiology 63(8):3119-3122.
Abad, C., Villena, C., Guix, S., Caballero, S., Pinto, R.M., and Bosch, A. 2001. The potential
role of fomites in the vehicular transmission of human astroviruses. Applied and Environmental
Microbiology 67:3904-3907.
Abbaszadegan, M., Stewart, P., and LeChevallier, M.A. 1999. Strategy for detection of viruses in
groundwater by PCR. Applied and Environmental Microbiology 65:444-449.
Abbaszadegan, M. 2006. Rotaviruses. Pp. 295-298 in Waterborne Pathogens. AWWA Manual
M48, 2nd Edition. Denver, CO: American Water Works Association.
Aitken, C., and Jeffries, D.J. 2001. Nosocomial spread of viral disease. Clinical Microbiology
Reviews 14(3):528-546.
Akin, E.W. 1981. Paper presented at the US EPA symposium on microbial health considerations
of soil disposal of domestic wastewaters.
Al-Saleem, T., and Al-Mondhiry, H. 2005. Immunoproliferative small intestinal disease (IPSID):
a model for mature B-cell neoplasms. Blood 105(6):2274-2280.
Allen, L.J.S., and Allen, E.A. 2003. A comparison of three different stochastic population
models with regard to persistence time. Theoretical Population Biology 64:439-449.
Anderson, R.M. and May, R. 1991. Infectious diseases of humans: Dynamics and Control. New
York: Oxford University Press.
Anderson, B.S., Sims, J.K., Liang, A.P., and Minette, H.P. 1988. Outbreak of eye and respiratory
irritation in Lahaina, Maui, possibly associated with Microcoleus lyngbyaceus. Journal of
Environmental Health 50(4):205-209.
Ang, C.W., Noordzij, P.G., de Klerk, M.A., Endtz, H.P., van Doom, P.A., and Laman, J.D. 2002.
Ganglioside mimicry of Campylobacter jejuni lipopolysaccharides determines antiganglioside
specificity in rabbits. Infection and Immunity 70(9):5081-5085.
Anonymous. 1998. Case Records of the Massachusetts General Hospital. Case 19-1998. New
England Journal of Medicine. 338:1830-1836.
Ansari, S.A., Springhorpe, V.S., and Sattar, S.A. 1991. Survival and vehicular spread of human
rotaviruses: possible relation to seasonality of outbreaks. Reviews of Infectious Diseases 13:448-
461.
113
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
APHA (American Public Health Association). 2004. Control of Communicable Diseases
Manual. 18th Edition. Washington, DC: APHA.
Arguin, P.M., Kozarsky, P.E., and Reed, C. 2008. Hepatitis, Viral, Type E. Chapter 4 in
Prevention of Specific Infectious Diseases, CDC Health Information for International Travel
2008 (Yellow Book). U.S. Centers for Disease Control and Prevention.
Armstrong,T.W., and Haas, C.N. 2008. Legionnaires' disease: evaluation of a quantitative
microbial risk assessment model. Journal of Water and Health 6(2): 149-166.
Artigas, J., Grosse, G., Niedobitek, F., Kassner, M., Risch, W., and Heise, W. 1994. Severe
toxoplasmic ventriculomeningoencephalomyelitis in two AIDS patients following treatment of
cerebral toxoplasmic granuloma. Clinical Neuropathology 13(3): 120-126.
Asano, T., Leong, L.Y.C., Rigby, M.G., and Sakaji, R.H. 1992. Evaluation of the California
wastewater reclamation criteria using enteric virus monitoring data. Water Science and
Technology 26(7-8):1513-1524.
Ashbolt, N.J., Reidy, C., and Haas, C.N. 1997. Microbial health risk at Sydney's coastal bathing
beaches. In: Proceeding of 17th Australian Water and Wastewater Association meeting. AWWA.
Melbourne, pp. 104-111.
ASM (American Society for Microbiology). 2011. Manual of Clinical Microbiology, 10th
Edition. Editor in Chief: James Versalovic. http://mcmlO.asmpress.org/
AWWA (American Water Works Association). 2006. Waterborne Pathogens, AWWA Manual
M48, Second Edition. American Water Works Association: Denver, CO.
Awwad, S.T., Petroll, W.M., McCulley, J.P., and Cavanagh, H.D. 2007. Updates in
Acanthamoeba keratitis. Eye Contact Lens 33(1): 1-8.
Azevedo, N.F., Pinto, A.R., Reis, N.M., Vieira, M.J., and Keevil, C.W. 2006. Shear stress,
temperature, and inoculation concentration influence the adhesion of water-stressed Helicobacter
pylori to stainless steel 304 and polypropylene. Applied and Environmental Microbiology
72(4):2936-2941.
Bailey, N.T.J. 1964. The Elements of Stochastic Processes with Application to the Natural
Sciences. New York: John Wiley and Sons.
Balbus, J., Parkin, R., and Embrey, M. 2000. Susceptibility in microbial risk assessment:
definitions and research needs. Environmental Health Perspectives 108:901-905.
Ballweber, L.R., Xiao, L., Bowman, D.D., Kahn, G., and Cama, V.A. 2010. Giardiasis in dogs
and cats: update on epidemiology and public health significance. Trends in Parasitology 26:180-
189.
114
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Barnett, N.D.P., Kaplan, A.M., Hopkin, R.J., Saubolle, M.A., and Rudinsky, M.F. 1996. Primary
amoebic meningoencephalitis with Naegleria fowlerr. clinical review. Pediatric Neurology 15(3):
230-234.
Barrett-Connor, E., and Connor, J.D. 1970. Extraintestinal manifestations of shigellosis.
American Journal of Gastroenterology 53(3):234-235.
Bartram, J., Corrales, L., Davison, A., Deere, D., Drury, D., Gordon, B., Howard, G., Rinehold,
A., and Stevens, M. 2009. Water Safety Plan Manual: Step-by-Step Risk Management for
Drinking-Water Suppliers. Geneva, Switzerland: WHO.
Bartrand, T.A., Weir, M.H., and Haas, C.N. 2008. Dose-Response Models for Inhalation of
Bacillus anthracis Spores: Interspecies Comparisons. Risk Analysis 28(4): 1115-1124.
Behan, P.O., andBakheit, A.M.O. 1991. Clinical spectrum of postviral fatigue syndrome. British
Medical Bulletin 47(4):793-808.
Bell, J.F., Owens, C.R., and Larson, C.L. 1955. Virulence of Bacterium tularense. I. A study of
the virulence of Bacterium tularense in mice, guinea pigs, and rabbits. Journal of Infectious
Diseases 97(2): 162-167.
Bellack, N.R., Koehoorn, M.W., MacNab, Y.C., and Morshed, M.G. 2006. A conceptual model
of water's role as a reservoir in Helicobacter pylori transmission: a review of the evidence.
Epidemiology and Infection 134(3):439-49.
Benenson A.S. 1995. Giardiasis. Control of Communicable Disease in Man. 16th edition,
American Public Health Association, Washington, DC.
Betancourt, W.Q., and Rose, J.B. 2004. Drinking water treatment processes for removal
of Cryptosporidium and Giardia. Veterniary Parasitology. 126:219-234.
Bieber, D., Ramer, S.W., Wu, C.Y., Murray, W.J., Tobe, T., Fernandez, R., and Schoolnik, G.K.
1998., Type IV pili, transient bacterial aggregates, and virulence of enteropathogenic
Escherichia coli. Science 280(5372):2114-2118.
Blaser, M.J., and Kirschner, D. 1999. Dynamics of Helicobacter pylori colonization in relation to
the host response. Proceedings of the National Academy of Sciences (USA) 96(15):8359-8364.
Blaser, M.J., and Kirschner, D. 2007. The equilibria that allow bacterial persistence in human
hosts. Nature 449(7164):843-849.
Boehm, A.B., Grant, S B., Kim, J.H., Mowbray, S.L., McGee, C D., Clark, C D., Foley, D M.,
and Wellman, D.E. 2002. Decadal and shorter period variability of surf zone water quality at
Huntington Beach, California. Environmental Science & Technology 36(18)3885-3892.
Boehm, A.B. 2007. Enterococci concentrations in diverse coastal environments exhibit extreme
variability. Environmental Science & Technology 41:8227-8232.
115
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Boerlijst, M.C., Bonhoeffer, S., and Nowak, M.A. 1996. Viral quasi-species and recombination.
Proceedings: Biological Sciences 263(1376): 1577-1584.
Bogosian, B.J., and Bourneuf, E.V. 2001. A matter of bacteria life and death. EMBO Reports
2(9):770-774.
Bollaerts, K., Aerts, M., Faes, C., Grijspeerdt, K., Dewulf, J., and Mintiens, K. 2008. Human
salmonellosis: estimation of dose-illness from outbreak data. Risk Analysis 28(2):427-440.
Borchardt, M.A., Bertz, P.D., Spencer, S.K., and Battigelli, D.A. 2003. Incidence of enteric
viruses in groundwater from household wells in Wisconsin. Applied and Environmental
Microbiology 69:1172-1180.
Borchardt, M. A., Haas, N.L., and Hunt, R.J. 2004. Vulnerability of drinking-water wells in La
Crosse, Wisconsin, to enteric-virus contamination from surface water contributions. Applied and
Environmental Microbiology 70:5937-5946.
Boshuizen, H.C., Neppelenbroek, S.E., van Vliet, H., Schellekens, J.F., den Boer, J.W., Peeters,
M.F., and Conyn-van Spaendonck, M.A. 2001. Subclinical Legionella infection in workers near
the source of a large outbreak of legionnaires disease. Journal of infectious Diseases 184:515-
518.
Brookmeyer, R., Johnson, E., and Barry, S. 2005. Modelling the incubation period of anthrax.
Statistics in Medicine 24:531-542.
Brown, L.M. 2000. Helicobacter pylori: epidemiology and routes of transmission.
Epidemiologic Reviews 22(2):283-297.
Brown, M.R., and Barker, J. 1999. Unexplored reservoirs of pathogenic bacteria: protozoa and
biofilms. Trends in Microbiology 7(l):46-50.
Brunner, R.L., O'Grady, D.J., Partin, J.C., Partin, J.S., and Schubert, W.K. 1979.
Neuropsychologic consequences ofReye syndrome. Journal of Pediatrics 95(5 Pt 1):706-711.
Brynestad, S., Braute, L., Luber, P., and Bartelt, E. 2008. Quantitative microbiological risk
assessment of campylobacteriosis cases in the German population due to consumption of chicken
prepared in homes. International Journal of Risk Assessment and Management 8(3): 194-213.
Buchanan, R.L., Smith, J.L., and Long, W. 2000. Microbial risk assessment: dose-response
relations and risk characterization. International Journal of Food Microbiology 58:159-172.
Bunning, V.K., Lindsay, J. A., and Archer, D.L. 1997. Chronic health effects of microbial
foodborne disease. World Health Statistics Quarterly 50(l-2):51-56.
Butzler, J.P. 2004. Campylobacter, from obscurity to celebrity. Clinical Microbiology and
Infection 10(10):868-876.
116
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
CAC (Codex Alimentarius Commission). 1999. Principles and Guidelines for the Conduct of
Microbiological Risk Assessment. (Step 8 of Codex elaboration) CAC/GL-30(1999).
CAC. 2004. Codex Alimentarius Commission - Procedural Manual Fourteenth Edition, Section
III - Working Principles for Risk Analysis for Application in the Framework of the Codex
Alimentarius.
CAC. 2007. Principles and Guidelines for the Conduct of Microbiological Risk Management.
(Step 8 of Codex elaboration) Alinorm 07/30/13. Report of 38th session of the Codex Committee
on Food Hygiene. CAC/GL 63-2007
Caccid, S. 2005. Molecular epidemiology of human cryptosporidiosis. Parassitologia 47:185-
192.
Cali, A. 2006. Microsporidia. Pp. 221-228 in Waterborne Pathogens. AWWA Manual M48, 2nd
Edition. Denver, CO: American Waterworks Association.
Casemore, D.P., Wright, S.E., and Coop, R.L. 1997. Cryptosporidiosis - human and animal
epidemiology. In: Cryptospoidium and Cryptosporidiosis, Fayer R (ed), CRC Press, New York.
Castro-Hermida, J.A., Garcia-Presedo, I., Almeida, A., Gonzalez-Warleta, M., Correia Da Costa,
J.M., and Mezo, M. 2008. Contribution of treated wastewater to the contamination of
recreational river areas with Cryptosporidium spp. and Giardia duodenalis. Water Research
42:3528-3538.
CDC (U.S. Centers for Disease Control and Prevention). 1992. Principles of Epidemiology, Self-
Study Course 3030-G, Second Edition.
http://www.uic.edu/sph/prepare/courses/ph490/resources/epiintro.pdf.
http://www.uic.edu/sph/prepare/courses/ph490/resources/epilesson01.pdf.
CDC. 1993. Surveillance for Waterborne Disease Outbreaks - United States, 1991-1992.
Morbidity and Mortality Weekly Report, 42: 1-22.
CDC. 1996. Surveillance for Waterborne-Disease Outbreaks - United States, 1993-1994.
Morbidity and Mortality Weekly Report, 45: 1-33.
CDC. 1998a. H. pylori: Fact Sheet for Health Care Providers.
http://www.cdc.gov/ulcer/files/hpfacts.PDF.
CDC. 1998b. Surveillance for Waterborne-Disease Outbreaks - United States, 1995-1996.
Morbidity and Mortality Weekly Report, 47: 1-33.
CDC. 2000. Surveillance for Waterborne Disease Outbreaks - United States, 1997-1998.
Morbidity and Mortality Weekly Report, 49: 1-35.
CDC. 2002. Surveillance for Waterborne-Disease Outbreaks - United States, 1999-2000.
Morbidity and Mortality Weekly Report, 51: 1-48.
117
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
CDC. 2004a. About Cyanobacteria. http://www.cdc.gov/hab/cvanobacteria/pdfs/about.pdf.
CDC. 2004b. Centers for Disease Control and Prevention Surveillance for Waterborne-Disease
Outbreaks associated with recreational water - United States, 2001-2002 and, Surveillance for
Waterborne-Disease Outbreaks associated with drinking water - United States, 2001-2002.
Surveillance Summaries, October 22, 2004. Morbidity and Mortality Weekly Reports 53(SS-8).
CDC. 2006a. Non-Polio Enterovirus Infections Website, Division of Viral Diseases.
http://www.cdc.gov/ncidod/dvrd/revb/enterovirus/non-polio entero.htm.
CDC. 2006b. Surveillance for Waterborne Disease and Outbreaks Associated with Recreational
Water - United States, 2003-2004. Morbidity and Mortality Weekly Report, 55: 1-30.
CDC. 2008. Surveillance for waterborne-disease and outbreaks associated with recreational
water use and other aquatic facilities - United States, 2005-2006. Morbidity and Mortality
Weekly Reports 57:1-72.
CDC. 2009a. Foodborne Pathogenic Microorganisms and Natural Toxins Handbook:
Plesiomonas shigelloides. http://www.foodsafetv. gov/~mow/chap 18.html.
CDC. 2009b. Norovirus: Technical Fact Sheet.
http://www.cdc.gov/ncidod/dvrd/revb/gastro/norovirus-factsheet.htm.
CDC. 2009c. Viral Hepatitis Website, Division of Viral Hepatitis.
http://www.cdc.gov/hepatitis/HepatitisA.htm.
CDC. 2010. Non-Polio Enterovirus Infections Website, Division of Viral Diseases.
http://www.cdc.gov/ncidod/dvrd/revb/enterovirus/non-polio entero.htm.
CDC/MMWR. 2006. Morbidity and Mortality Weekly Report 55 (SS12):31-58.
Cervenka, L. 2007. Survival and inactivation of Arcobacter spp., a current status and future
prospect. Critical Reviews in Microbiology 33(2): 101-8.
Chappell, C.L., Okhuysen, P.C., Sterling, C.R., and DuPont, H.L. 1996. Cryptosporidium
parvum: intensity of infection and oocyst excretion patterns in healthy volunteers. J. Infect. Dis.,
173:232-236.
Chappell, C.L., Okhuysen, P.C., Sterling, C.R., Wang, C., Jakubowski, W., and DuPont, H.L.
1999. Infectivity of Cryptosporidium parvum in healthy adults with pre-existing anti-C. parvum
serum immunoglobulin G. American Journal of Tropical Medicine and Hygiene 60(1): 157-164.
Chappell, C.L., Okhuysen, P.C., Langer-Curry. R., Widmer, G., Akiyoshi, D.E., Tanriverdi, S.,
and Tzipori, S. 2006. Cryptosporidium hominis: experimental challenge of healthy adults. The
American Journal of Tropical Medicine and Hygiene 75(5):851-7.
Chen, K.T., Chen ,C.J., and Chiu, J.P. 2001. A school waterborne outbreak involving both
Shigella sonnei and Entamoeba histolytica. Journal of Environmental Health 64(4):9-13, 26.
118
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Chen, T.L., Chan, C.C., Chen, H.P., Fung, C.P., Lin, C.P., Chan, W.L., and Liu, C.Y. 2003.
Clinical characteristics and endoscopic findings associated with Blastocystis hominis in healthy
adults. American Journal of Tropical Medicine and Hygiene 69:213-216.
Chey, W.D., and Wong, B.C.Y. 2007. American College of Gastroenterology guideline on the
management of Helicobacter pylori infection. American Journal of Gastroenterology 102:1808-
1825.
Codd, G.A. 2000. Cyanobacterial toxins, the perception of water quality, and the prioritisation of
eutrophication control. Ecological Engineering 16(l):51-60.
Codd, G.A., Ward, C.J., and Bell, S.G. 1997. Cyanobacterial toxins: occurrence, modes of
action, health effects and exposure routes. Archives of Toxicology Supplement 19:399-410.
Coleman, M., and Marks, H. 1998. Topics in dose-response modeling. Journal of Food
Protection 61:1550-1559.
Coleman, M., and Marks, H. 2000. Mechanistic modeling of salmonellosis. Quantitative
Microbiology 2:227-247.
Coleman, M.E., Marks, H.M., Golden, N.J., and Latimer, H.K. 2004. Discerning strain effects in
microbial dose-response data. Journal of Toxicology and Environmental Health-Part A -Current
Issues 67:667-685.
Colwell, R.R., Huq, A., Islam, M.S., Aziz, K.M.A., Yunus, M., Khan, N.H., Mahmud, A., Sack,
R.B., Nair, G.B., Chakraborty, J., Sack, D.A., and Russek-Cohen, E. 2003. Reduction of cholera
in Bangladeshi villages by simple filtration. Proceedings of the National Academy of Sciences
(USA) 100(3): 1051-1055.
Connor, B.A., Johnson, E.J., and Soave, R. 2001. Reiter syndrome following protracted
symptoms of Cyclospora infection. Emerging Infectious Diseases 7(3):453-454.
Couch, R.B., Cate, T.R., Gerone, P. J., Fleet, W.F., Lang, D.J., Griffith, W.R., and Knight, V.
1965. Production of illness with a small-particle aerosol of coxsackie A21. Journal of Clinical
Investigation 44:535-542.
Covacci, A., and Rappuoli, R. 1998. Helicobacter pylori: molecular evolution of a bacterial
quasi-species. Current Opinion in Microbiology 1(1):96-102.
Covert, T.C. 1999. Salmonella. In Waterborne Pathogens, AWWA manual M48, 1st Edition.
Denver, CO: American Waterworks Association.
Covert, T.C., and Meckes, M.C. 2006. Salmonella. Pp. 135-140 in Waterborne Pathogens.
AWWA Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Crabtree, K.D., Gerba, C.P., Rose, J.B., and Haas, C.N. 1997. Waterborne adenovirus: a risk
assessment. Water Science and Technology 35(11-12): 1-6.
119
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Crainiceanu, C.M., Stedinger, J.R., Ruppert, D., and Behr, C.T. 2003. Modeling the United
States national distribution of waterborne pathogen concentrations with application to
Cryptosporidiumparvum. Water Resources Research 39(9):SWC 2-1.
Craun, M.F., Craun, G.F., Calderon, R.L., and Beach, M.J. 2006. Waterborne outbreaks reported
in the United States. Journal of Water and Health 4(Suppl 2): 19-30.
Craun, G.F., Brunkard, J.M., Yoder, J.S., Roberts, V.A., Carpenter, J., Wade, T., Calderon, R.L.,
Roberts, J.M., Beach, M.J. and S.L. Roy. 2010. Causes of outbreaks associated with drinking
water in the United States from 1971 to 2006. Clinical Microbiological Reviews, 23(3):507-528.
Davidson, P.W., Otuama, L.A., Willoughby, R.H., and Swisher, C.N. 1978. Neurological and
intellectual sequelae of Reye's syndrome. American Journal of Mental Deficiency 82(6): 535-
541.
de Vos, C.J., Saatkamp, H.W., Nielen, M., and Huirnel, R.B.M. 2006. Sensitivity Analysis to
Evaluate the Impact of Uncertain Factors in a Scenario Tree Model for Classical Swine Fever
Introduction. Risk Analysis 26(5): 1311-1322.
Didier, E.S., and Weiss, L.M. 2006. Microsporidiosis: current status. Current Opinion in
Infectious Diseases 19(5):485-492.
Didier, E.S., Stovall, M.E., Green, L.C., Brindley, P.J., Sestak, K., and Didier, P.J. 2004.
Epidemiology of microsporidiosis: sources and modes of transmission. Veterniary Parasitology
126(1-2): 145-166.
Dietz, V., Vugia, D., Nelson, R., Wicklund, J., Nadle, J., McCombs, K.G., and Reddy, S. 2000.
Active, multisite, laboratory-based surveillance for Cryptosporidium parvum. American Journal
of Tropical Medicine and Hygiene 62:368-372.
Dorevitch, S., Panthi, S., Huang, Y., Li, H., Michalek, A.M., Pratap, P., Wroblewski, M., Liu, L.,
Scheff, P.A., and Li, A. 2011. Water ingestion during water recreation. Water Research 45(5):
2020-2028.
Dubey, J.P. 2004. Toxoplasmosis - a waterborne zoonosis. Veterinary Parasitology 126(l-2):57-
72.
Dubey, J.P. 2006. Toxoplasma gondii. Pp. 239-242 in Waterborne Pathogens. AWWA Manual
M48, 2nd Edition. Denver, CO: American Water Works Association.
Dufour, A., Evans, O., Behymer, T., and Cantu, R. 2006. Water ingestion during swimming
activities in a pool: a pilot study. Journal of Water and Health 4:425-430.
DuPont, H., Chappell, C., Sterling, C., Okhuysen, P., Rose, J., and Jakubowski, W. 1995. The
infectivity of Cryptosporidium parvum in healthy volunteers. New England Journal of Medicine
332:855-859.
120
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Dworkin, M.S., Shoemaker, P.C., Goldoft, M.J., and Kobayashi, J.M. 2001. Reactive arthritis
and Reiter's syndrome following an outbreak of gastroenteritis caused by Salmonella enteritidis.
Clinical Infectious Diseases 33(7): 1010-1014.
Effler, P., Ieong, M-C., Kimura, A., Nakata, M., Burr, R., Cremer, E., and Slutsker, L. 2001.
Sporadic Campylobacter jejuni infections in Hawaii: associations with prior antibiotic use and
commercially prepared chicken. Journal of Infectious Diseases 83:1152-1155.
Eisenberg, J.N., Seto, E.Y.W., Olivieri, A.W., and Spear, R.C. 1996. Quantifying water pathogen
risk in an epidemiological framework. Risk Analysis 16:549-563.
Eisenberg, J.N., Seto, E.Y.W., Colford, J.M., Olivieri, A.W., and Spear, R.C. 1998. An analysis
of the Milwaukee cryptosporidiosis outbreak based on a dynamic model of the infection process.
Epidemiology 9:255-263.
Eisenberg, J.N., Brookhart, M.A., Rice, G., Brown, M., and Colford, J.M., Jr. 2002. Disease
transmission models for public health decision making: analysis of epidemic and endemic
conditions caused by waterborne pathogens. Environmental Health Perspectives 110(8):783-790.
Eisenberg, J.N.S., Soller, J. A., Scott, J., Eisenberg, D.M., and Colford, J.M. 2004. A dynamic
model to assess microbial health risks associated with beneficial uses of biosolids. Risk Analysis
24:221-236.
Eisenberg, J.N.S., Lei, X., Hubbard, A.H., Brookhart, M.A., and Colford, J.M. 2005. The role of
disease transmission and conferred immunity in outbreaks: analysis of the 1993 Cryptosporidium
outbreak in Milwaukee, Wisconsin. American Journal of Epidemiology 16(l):62-72.
Eisenberg, J.N., Moore, K., Soller, J.A., Eisenberg, D., and Colford, J.M., Jr. 2008. Microbial
risk assessment framework for exposure to amended sludge projects. Environmental Health
Perspectives 116:727-733.
Ekwall, E., Ljungh, A., and Selander, B. 1984. Asymptomatic urinary tract infection caused by
Shigella sonnei. Scandinavian Journal of Infectious Diseases 16(1): 121-122.
Embrey, M.A. 2002. Cyanobacteria in Drinking Water. In Handbook of CCL Microbes in
Drinking Water (pp. 203-227). Denver, CO: American Water Works Association.
Englehardt, J.D. 2004. Predictive Bayesian dose-response assessment for appraising absolute
health risk from available information. Human and Ecological Risk Assessment 10(l):69-78.
Englehardt, J.D., and Swartout, J. 2004. Predictive population dose-response assessment for
Cryptosporidium parvum: infection endpoint. Journal of Toxicology and Environmental Health-
Part A-Current Issues 67(8-10):651-666.
Englehardt, J.D, and Swartout, J. 2006. Predictive Bayesian microbial dose-response assessment
based on suggested self-organization in primary illness response: Cryptosporidium parvum. Risk
Analysis 26(2):543-554.
121
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Englehardt, J.D., and Swartout, J. 2008. Development and Evaluation of Novel Dose-Response
Models for Use in Microbial Risk Assessment, Technical Report. EPA/600/R-08/033. National
Center for Environmental Assessment, Office of Research and Development, U.S.
Environmental Protection Agency.
Enriquez, C., and Thurston-Enriquez, J. 2006. Adenovirus. Pp. 253-248 in Waterborne
Pathogens. AWWA Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
EO (Executive Order). 1994. Executive Order 12898—Federal Actions to Address
Environmental Justice in Minority Populations and Low-Income Populations.
http://www.epa.gov/fedrgstr/eo/eol2898.htm.
EO. 1997. Executive Order 13045—Protection of Children from Environmental Health Risks
and Safety Risks. Federal Register 62(78): 19883-19888.
Ernst, P.B., and Gold, B.D. 2000. The disease spectrum of Helicobacter pylori: the
immunopathogenesis of gastroduodenal ulcer and gastric cancer. Annual Review of
Microbiology 54:615-640.
Fan, P.C., Chen, Y.C., Tian, Y.C., Chang, C.H., Fang, J.T., and Yang, C.W. 2009. Acute renal
failure associated with acute non-fulminant Hepatitis A: a case report and review of literature.
Renal Failure 31(8):756-764.
Fankhauser, R.L., Noel, J.S., Monroe, S.S., Ando, T., and Glass, R.I. 1998. Molecular
epidemiology of "Norwalk-like viruses" in outbreaks of gastroenteritis in the United States.
Journal of Infectious Diseases 178:1571-1578.
FAO/WHO (Food and Agriculture Organization of the United Nations/World Health
Organization). 2003. Microbiological Risk Assessment Series, No. 3: Hazard Characterization
for Pathogens in Food and Water, Guidelines.
http://whqlibdoc.who.int/publications/2003/9241562374.pdf.
FAO/WHO. 2009. Risk Characterization of Microbiological Hazards in Food: Guidelines, MRA
Series 17, http://www.who.int/foodsafetv/publications/micro/MRA17.pdf
Farber, J.M., Ross, W.H., Harwig, J. 1996. Health risk assessment of Listeria monocytogenes in
Canada. International Journal of Food Microbiology 30(1-2): 145-156.
Faruque, S.M., Biswas, K., Udden, S.M.N., Ahmad, Q.S., Sack, D.A., Nair, G.B., and
Mekalanos, J.J. 2006. Transmissibility of cholera: in vivo-formed biofilms and their relationship
to infectivity and persistence in the environment. Proceedings of the National Academy of
Sciences (USA) 103(16):6350-6355.
Fayer, R. 2004. Cryptosporidium: a waterborne zoonotic parasite. Veterinary Parasitology.
126:37-56.
Fayer, R., Speer, C.A., and Dubey, J.P. 1997. The General Biology of Cryptosporidium. In
Cryptosporidium and Cryptosporidiosis, ed. R. Fayer. New York: CRC Press.
122
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Fazil, A., Paoli, G., Lammerding, A.M., Davidson, V., Hrudey, S., Isaac-Renton, J., and
Griffiths, M. 2005. Microbial Risk Assessment as a Foundation for Informed Decision-Making:
A Needs, Gaps and Opportunities Assessment (NGOA) for Microbial Risk Assessment in Food
and Water. Public Health Agency of Canada.
http://www.uoguelph.ca/crifs/NGOA/Finalupdates/NGOAfinalreport.pdf.
FDA (U.S. Food and Drug Administration). 2002. Initiation and Conduct of all "Major" Risk
Assessments Within a Risk Analysis Framework. A Report by the Center for Food Safety and
Applied Nutrition Risk Analysis Working Group, http://www.cfsan.fda.gov/~dms/rafw-toc.html.
FDA/USDA. 2003. Quantitative Assessment of Relative Risk to Public Health from Foodborne
Listeria monocytogenes Among Selected Categories of Ready-to-Eat Foods. Rockville, MD.
http://www.foodsafetv.gov/~dms/lmr2-toc.html.
FDA. 2005. Quantitative Risk Assessment on the Public Health Impact of Vibrio
parahaemolyticus in Raw Oysters. Center for Food Safety and Applied Nutrition.
http://www.fda.gov/Food/ScienceResearch/ResearchAreas/RiskAssessmentSafetvAssessment/uc
m050421.htm.
FDA. 2006. The Bad Bug Book: Foodborne Pathogenic Organisms and Natural Toxins
Handbook. Washington, DC: FDA.
http://www.fda.gov/food/foodsafetv/foodborneillness/foodborneillnessfoodbornepathogensnatura
ltoxins/badbugbook/default.htm.
FDA. 2009. Bad Bug Book: Foodborne Pathogenic Microorganisms and Natural Toxins
Handbook Campylobacter jejuni.
http://www.fda.gov/Food/FoodSafetv/FoodborneIllness/FoodborneIllnessFoodbornePathogensN
aturalToxins/BadBugBook/ucm070024.htm.
Feazel, L.M., Baumgartnera, L.K., Petersona, K.L., Franka, D.N., Harris, J.K., and Pace, N.R.
2009. Opportunistic pathogens enriched in showerhead biofilms. Proceedings of the National
Academy of Sciences (USA) 106(38): 16393-16399.
Ferrera, P.C., Jeanjaquet, M.S., and Mayer, D.M. 1996. Shigella-induced encephalopathy in an
adult. The American Journal of Emergency Medicine 14(2): 173-5.
Field, S.K., Fisher, D., and Cowie, R.L. 2004. Mycobacterium avium complex pulmonary
disease in patients without HIV infection. Chest 126(2):566-581.
Fields, B.S., Benson, R.F., and Besser, R.E. 2002. Legionella and Legionnaires' disease: 25
years of investigation. Clinical Microbiology Reviews 15(3):506-526.
Fine, K.D., and Stone, M.J. 1999. Alpha-heavy chain disease, Mediterranean lymphoma, and
immunoproliferative small intestinal disease - a review of clinicopathological features,
pathogenesis, and differential diagnosis. American Journal of Gastroenterology 94(5): 1139-
1152.
123
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Fink, J.N., Ortega, H.G., Reynolds, H.Y., Cormier, Y.F., Fan, L.L., Franks, T.J., Kreiss, K.,
Kunkel, S., Lynch, D., Quirce, S., Rose, C., Schleimer, R.P., Schuyler, M.R., Selman, M., Trout,
D., Yoshizawa, Y. 2005. Needs and opportunities for research in hypersensitivity pneumonitis.
American Journal of Respiratory and Critical Care Medicine 171(7):792-798.
Fohlman, J., and Friman, G. 1993. Is juvenile diabetes a viral disease? Annals of Medicine
25(6):569-574.
Fraser, D.W., Tsai, T.R., Orenstein, W., Parkin, W.E., Beecham, H.J., Sharrar, R.G., Harris, J.,
Mallison, G.F., Martin, S.M., McDade, J.E., Shepard, C.C., and Brachman, P.S. 1977.
Legionnaires' disease: description of an epidemic of pneumonia. New England Journal of
Medicine 297:1189-1197.
Frey, H.C., Mokhtari, H., and Zheng, J. 2004. Recommended Practice Regarding Selection,
Application, and Interpretation of Sensitivity Analysis Methods Applied to Food Safety Process
Risk Models. Office of Risk Assessment and Cost-Benefit Analysis, U.S. Department of
Agriculture. http://www.ce.ncsu.edu/risk/Phase3Final.pdf.
Flicker, C.R. 2006. Yersinia. Pp. 157-160 in Waterborne Pathogens. AWWA Manual M48, 2nd
Edition. Denver, CO: American Waterworks Association.
Frost, F. and Craun, G.F. 1998. The Importance of Acquired Immunity in the Epidemiology of
Cryptosporidiosis and Giardiasis. EPA, OECD Workshop Molecular Methods for Safe Drinking
Water.
Frost, F.J., Roberts, M., Kunde, T.R., Craun, G., Tollestrup, K., Harter, L., and Muller, T. 2005.
How clean must our drinking water be: the importance of protective immunity. Journal of
Infectious Diseases 191:809-814.
Gale, P. 2003. Using event trees to quantify pathogen levels on root crops from land application
of treated sewage sludge. Journal of Applied Microbiology 94:35-47.
Gale, P. 2005. Land application of treated sewage sludge: quantifying pathogen risks from
consumption of crops. Journal of Applied Microbiology 98:380-396.
Garcia, L.S. 2006a. Blastocystis hominis. Pp. 189-192 in Waterborne Pathogens. AWWA
Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Garcia, L.S. 2006b. Isospora belli. Pp. 217-220 in Waterborne Pathogens. AWWA Manual M48,
2nd Edition. Denver, CO: American Water Works Association.
Garg, A.X., Suri, R.S., Barrowman, N., Rehman, F., Matsell, D., Rosas-Arellano, M.P.,
Salvadori, M., Haynes, R.B., and Clark W.F. 2003. Long-term renal prognosis of diarrhea-
associated hemolytic uremic syndrome - a systematic review, meta-analysis, and meta-
regression. Journal of the American Medical Association 290(10): 1360-1370.
Gerba, C.P. 2006. Hepatitis E Virus. Pp. 279-280 in Waterborne Pathogens. AWWA Manual
M48, 2nd Edition. Denver, CO: American Water Works Association.
124
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Gerba, C., Yates, M., Yates, S., and Hurst, C., eds. 1991. Quantitation of Factors Controlling
Viral and Microbial Transport in the Subsurface. In: Modeling the Environmental Fate of
Microorganisms, Washington, DC: American Society for Microbiology.
Gerba, C.P.; Rose, J.B., and Haas, C.N. 1996a. Sensitive populations: who is at the greatest risk?
International Journal of Food Microbiology 30(1-2): 113-123.
Gerba, C.P., Rose, J.B., Haas, C.N., and Crabtree, K.D. 1996b. Waterborne rotavirus: a risk
assessment. Water Research 30:2929-2940.
Gerba, C.P., Henze, M., Loosdrecht, C.M., Ekman, G.A., and Brdjanovic, D., eds. 2008.
Pathogen Removal. In: Biological Wastewater Treatment, Modeling and Design, London, UK:
IWA Publishing.
Gilbert, M., Godschalk, P.C., Karwaski, M.F., Ang, C.W., van Belkum, A., Li, J., Wakarchuk,
W.W., and Endtz, H.P. 2004. Evidence for acquisition of the lipopolysaccharide biosynthesis
locus in Campylobacter jejuni GB11, a strain isolated from a patient with Guillain-Barre
syndrome, by horizontal exchange. Infection and Immunity 72(2): 1162-1165.
Gilinsky, N.H., Novis, B.H., Wright, J.P., Dent, D.M., King, H., and Marks, I.N.S. 1987.
Immunoproliferative small-intestinal disease: clinical features and outcome in 30 cases.
Medicine 66(6):438-446.
Gilks, W., Richardson, S., and Spiegelhalter D.J. eds. 1996. Markov Chain Monte Carlo in
Practice. London, UK: Chapman and Hall.
Gilks, W.R. and Wild, P. 1992. Adaptive rejection sampling for Gibbs sampling. Applied
Statistics 41:337-348.
Girschick, H.J., Guilherme, L., Inman, R.D., Latsch, K, Rihl, M., Sherer, Y., Shoenfeld, Y.,
Zeidler, H., Arienti, S., and Doria, A. 2008. Bacterial triggers and autoimmune rheumatic
diseases. Clinical and Experimental Rheumatology 26(1 Suppl 48):S12-S17.
Glass, R. I., Bresee, J., Jiang, B. M., Gentsch, J., Ando, T., Fankhauser, R., Noel, J., Parashar,
U., Rosen, B., and Monroe, S.S. 2001. Gastroenteritis viruses: an overview. A'ovartis Foundation
Symposium 238:5-25.
Glick, T.H., Gregg, M.B., Berman, B., Mallison, G., Rhodes ,W.W., Jr., and Kassanoff, I. 1978.
Pontiac fever. An epidemic of unknown etiology in a health department. I. Clinical and
epidemiologic aspects. American Journal of Epidemiology 107:149-160.
Gold, M.R., Stevenson, D., and Fryback, D.G. 2002. HALYs and QALYs and DALYs, oh my:
similarities and differences in summary measures of population health. Annual Review of Public
Health 23:115-134.
Golden, N.J., Crouch, E.A., Latimer, H., Kadry, A., and Kause, J. 2009. Risk assessment for
Clostridium perfringens in ready-to-eat and partially cooked meat and poultry products. Journal
of Food Protection 72(7): 13 76-13 84.
125
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Goren, A., Freier, S., and Passwell, J.H. 1992. Lethal toxic encephalopathy due to childhood
shigellosis in a developed country. Pediatrics 89(6): 1189-1193.
Greenberg, H., Valdesuso, J., Kapikian, A., Chanock, R., Wyatt, R., Szmuness, W., Larrick, J.,
Kaplan, J., Gilman, R.H., and Sack, D.A. 1979. Prevalence of antibody to the Norwalk virus in
various countries. Infection and Immunity 26:270-273.
Grohmann, G.S., Glass, R.I., Pereira, H.G., Monroe, S.S., Hightower, A.W., Weber, R., and
Bryan, R.T. 1993. Enteric viruses and diarrhea in HIV-infected patients. Enteric Opportunistic
Infections Working Group. New England Journal of Medicine, 329(1): 14-20.
Gronewold, A.D., Borsuk, M.E., Wolpert, R.L., and Reckhow, K.H. 2008. An assessment of
fecal indicator bacteria-based water quality standards. Environmental Science & Technology
42(13):4676-4682.
Gronewold, A.D., Qian, S.S., Wolpert, R.L., and Reckhow, K.H. 2009. Calibrating and
validating bacterial water quality models: a Bayesian approach. Water Research 43:2688-2698.
Guix, S., Bosch, A., and Pinto, R.M. 2005. Human astrovirus diagnosis and typing: current and
future prospects. Letters in Applied Microbiology 41(2): 103-105.
Gupta, A., Polyak, C.S., Bishop, R.D., Sobel, J., and Mintz, E.D. 2004. Laboratory-confirmed
shigellosis in the United States, 1989-2002: epidemiologic trends and patterns. Clinical
Infectious Diseases 38(10): 1372-1377.
Gutting, B.W., Channel, S.R., Berger, A.E., Gearhart, J.M., Andrews, G.A., and Sherwood, R.L.
2008. Mathematically modeling inhalation anthrax. Microbe 3(2):78-85.
Gyori, E. 2003. December 2002: 19-year old male with febrile illness after jet ski accident. Brain
Pathology 13(2):237-239.
Haas, C.N. 1983. Effect of effluent disinfection on risks of viral disease transmission via
recreational water exposure. Journal - Water Pollution Control Federation 55:1111-1116.
Haas, C.N., Crockett, C.S., Rose, J.B., Gerba, C.P., andFazil, A.M. 1996. Assessing the risks
posed by oocysts in drinking water. Journal of the American Water Works Association
88(9): 131-136.
Haas, C.N., Rose, J., and Gerba, C.P. 1999. Quantitative Microbial Risk Assessment. New York:
Wiley.
Halbur, P.G., Kasorndorkbua, C., Gilbert, C., Guenette, D., Potters, M.B., Purcell, R.H.,
Emerson, S.U., Toth, T.E., and Meng, X.J. 2001. Comparative Pathogenesis of Infection of Pigs
with Hepatitis E Viruses Recovered from a Pig and a Human, Journal of Clinical Microbiology
39(3):918-923.
Hall, N.H. 2006. Legionella. Pp. 119-124 in Waterborne Pathogens. AWWA Manual M48, 2nd
Edition. Denver, CO: American Waterworks Association.
126
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Hanak, V., Golbin, J.M., and Ryu, J.H. 2007. Causes and presenting features in 85 consecutive
patients with hypersensitivity pneumonitis. Mayo Clinic Proceedings 82(7):812-816.
Hanninen, M.L., Haajanen, H., Pummi, T., Wermundsen, K., Katila, M.L., Sarkkinen, H.,
Miettinen, I., and Rautelin, H. 2003. Detection and typing of Campylobacter jejuni and
Campylobacter coli and analysis of indicator organisms in three waterborne outbreaks in
Finland. Applied and Environmental Microbiology 69(3): 1391-1396.
Hastings, W.K. 1970. Monte Carlo sampling methods using Markov Chains and their
applications. Biometrika 57:97-109.
Hejkal, T.W.,Keswick, B., LaBelle, R.L., Gerba, C.P., Sanchez, Y.,Dreesman, G., Hafkin, B.,
and Melnick, J.L. 1982. Viruses in a community water supply associated with an outbreak of
gastroenteritis and infectious hepatitis. Journal of the American Water Works Association
74:318-321.
Hellard, M.E., Sinclair, M.I., Hogg, G.G., and Fairley, C.K. 2000. Prevalence of enteric
pathogens among community based asymptomic individuals. Journal of Gastroenterology and
Hepatology 15:290-293.
Herath, G. 1995. The algal bloom problem in Australian waterways: an economic appraisal.
Review of Marketing and Agricultural Economics 63:77-86.
Herwaldt, B.L. 2000. Cyclospora cayetanensis: a review, focusing on the outbreaks of
cyclosporiasis in the 1990s. Clinical Infectious Disease 31(4): 1040-1057.
Hoeger, S. J., Hitzfeld, B.C., and Dietrich, D.R. 2005. Occurrence and elimination of
cyanobacterial toxins in drinking water treatment plants. Toxicology and Applied Pharmacology
203(3):231-242.
Holcomb, D.L., Smith, M.A., Ware, G.O., Hung, Y.C., Brackett, R.E., and Doyle, M.P. 1999.
Comparison of six dose-response models for use with food-borne pathogens. Risk Analysis
19(6): 1091-1100.
Holdenfried, R. and Quan, S.F. 1956. Susceptibility of New Mexico rodents to experimental
plague. Public Health Reports 71(10):979-984.
Holme, R. 2003. Drinking water contamination in Walkerton, Ontario: positive resolutions from
a tragic event. Water Science and Technology 47(3): 1-6.
Hopkins, R.S., Gaspard, G.B., Williams, F.P., Karlin, R.J., Cukor, K.G., and Blacklow, N.R.
1984. A community waterborne gastroenteritis outbreak: evidence for rotavirus as the agent.
American Journal of Public Health 74:263-265.
Hopkins, R.S., Shillam, P., Gaspard, B., Eisnach, L. and Karlin, R.J. 1985. Waterborne disease in
Colorado: three years' surveillance and 18 outbreaks. American Journal of Public Health
75(3):254-257.
127
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Hunter, P.R. 1992. Cyanobacteria and human health. Journal of Medical Microbiology 36(5):
301-302.
Huq, M.I., and Islam, M.R. 1983. Microbiological and clinical studies in diarrhoea due to
Plesiomonas shigelloides. Indian Journal of Medical Research 11.193-191.
IARC (International Agency for Research on Cancer). 1994. Shistosomes, Liver Flukes, and
Helicobacter pylori. IARC Working Group on the Evaluation of Carcinogenic Risks to Humans.
Lyon. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans 61:1-241.
Illingworth, C. D., and Cook, S.D. 1998. Acanthamoeba keratitis. Survey of Ophthalmology
42(6):493-508.
ILSI (International Life Sciences Institute). 1996. A conceptual framework for assessment of the
risks of human disease following exposure to waterborne pathogens. Risk Analysis 16:841-848.
ILSI. 2000. Revised Framework for Microbial Risk Assessment. Washington, DC.
http://www.ilsi.org/file/mrabook.pdf.
Jameel, S. 1999. Molecular biology and pathogenesis of hepatitis E virus. Expert Reviews in
Molecular Medicine (6 December): 1-16. http://www-ermm.cbcu.cam.ac.uk/99001271h.htm.
Jiang, S. 2006. Human adenoviruses in water: occurrence and health implications: a critical
review. Environmental Science & Technology 40:7132-7140.
Jones, I.G., and Roworth, M. 1996. An outbreak of Escherichia coli 0157 and
campylobacteriosis associated with contamination of a drinking water supply. Public Health
110(5):277-282.
Kaldor, J., and Speed, B.R. 1984. Guillain-Barre syndrome and Campylobacter jejuni', a
serological study. British Medical Journal 288(6434): 1867-1870.
Kappus K.D., Lundgren R.G., Juranek D.D., Roberts J.M., Spencer H.C. 1994. Intestinal
parasitism in the United States: update on a continuing problem. American Journal of Tropical
Medicine and Hygiene 50:705-713.
Karanis, P., Kourenti, C., and Smith, H. 2007. Waterborne transmission of protozoan parasites: a
worldwide review of outbreaks and lessons learnt. Journal of Water Health 5(1): 1-38.
Karanja, R.M., Gatei, W., and Wamae, N. 2007. Cyclosporiasis: an emerging public health
concern around the world and in Africa. African Health Sciences 7(2):62-67.
Keene, W. 2006 Entamoeba histolytica. Pp. 203-207 in Waterborne Pathogens. AWWA Manual
M48, 2nd Edition. Denver, CO: American Water Works Association.
Khan, N. A. 2006. Acanthamoeba: biology and increasing importance in human health. FEMS
Microbiology Reviews 30(4):564-595.
128
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Khan, W., Dhar, U., Salam, M., Griffiths, J., Rand, W., and Bennish, M. 1999. Central nervous
system manifestations of childhood shigellosis: prevalence, risk factors, and outcome. Pediatrics
103(2):E18.
Kim, K.S., Hufnagel, G., Chapman, N.M., and Tracy, S. 2001. The group B coxsackieviruses
and myocarditis. Reviews in Medical Virology 11:355-368.
Kim, S.K., An, J.Y., Park, M.S., and Kim, B.J. 2007. A case report of Reiter's syndrome with
progressive myelopathy. Journal of Clinical Neurology 3(4):215-218.
King, A.A., Ionides, E.L., Pascual, M., and Bouma, M.J. 2008. Inapparent infections and cholera
dynamics. Nature 454:877-880.
Kistemann, T., Rind, E., Rechenburg, A., Koch, C., Classen, T., Herbst, S., Wienand, I., and
Exner, M. 2008. A comparison of efficiencies of microbiological pollution removal in six
sewage treatment plants with different treatment systems. International Journal of Hygiene and
Environmental Health 211:534-545.
Ko, G., Cromeans, T. L.,and Sobsey, M.D. 2003. Detection of infectious adenovirus in cell
culture by mRNA reverse transcription-PCR. Applied and Environmental Microbiology
69(12):7377-7384.
Kodell, R.L., Kang, S-H., and Chen, J.J. 2002. Statistical models of health risk due to microbial
contamination of foods. Environmental and Ecological Statistics 9:259-271.
Koopman, J.S., Monto, A.S., and Longini, I.M., Jr. 1989. The Tecumseh Study XVI: Family and
community sources of rotavirus infection. American Journal of Epidemiology 130(4)760-768.
Koopman, J.S., Chick, S.E., Simon, C.P., Riolo, C.S., and Jacquez, G. 2002. Stochastic effects on
endemic infection levels of disseminating versus local contacts. Mathematical Biosciences 180:49-
71.
Kuipers, E.J. 1999. Exploring the link between Helicobacter pylori and gastric cancer.
Alimentary Pharmacology & Therapeutics 13(Suppl 1):3-11.
Kuipers, E.J., Thijs, J.C., and Festen, H.P. 1995. The prevalence of Helicobacter pylori in peptic
ulcer disease. Alimentary Pharmacology & Therapeutics 9(Suppl 2):59-69.
Labine, M. A., and Minuk, G.Y. 2009. Cyanobacterial toxins and liver disease. Canadian Journal
of Physiology and Pharmacology 87(10):773-788.
Lacasse, Y., Selman, M., Costabel, U., Dalphin, J.C., Ando, M., Morell, F., Erkinjuntti-
Pekkanen, R., Muller, N., Colby, T.V., Schuyler, M., and Cormier, Y. 2003. Clinical diagnosis of
hypersensitivity pneumonitis. American Journal of Respiratory and Critical Care Medicine
168(8): 952-958.
Lau, H.Y., and Ashbolt, N.J. 2009. The role of biofilms and protozoa in Legionella pathogenesis:
implications for drinking water. Journal of Applied Microbiology 107(2):368-378.
129
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
LeChevallier, M.W. 2006. Mycobacterium avium Complex. Pp. 125-130 in Waterborne
Pathogens. AWWA Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Lecuit, M., Abachin, E., Martin, A., Poyart, C., Pochart, P., Suarez, F., Bengoufa, D., Feuillard,
J., Lavergne, A., Gordon, J.I., Berche, P., Guillevin, L., and Lortholary, O. 2004.
Immunoproliferative small intestinal disease associated with Campylobacter jejuni. New
England Journal of Medicine 350(3):239-248.
Lehner, A., Tasara, T., and Stephan, R. 2005. Relevant aspects of Arcobacter spp. as potential
foodborne pathogen. International Journal of Food Microbiology 102(2): 127-135
Levin, B.R., and Antia, R. 2001. Why we don't get sick: the within-host population dynamics of
bacterial infections. Science 292(5519): 1112-1115.
Lin, H.C., Kao, C.L., Chang, L.Y., Hsieh, Y.C., Shao, P L., Lee, P.I., Lu, C.Y., Lee, C.Y., and
Huang, L.M. 2008. Astrovirus gastroenteritis in children in Taipei. Journal of the Formosan
Medical Association 107(4):295-303.
Lindesmith, L., Moe, C., Marionneau, S., Ruvoen, N., Jiang, X., Lindblad, L., Stewart, P.,
LePendu, J., and Baric, R. 2003. Human susceptibility and resistance to Norwalk virus infection.
Nature Medicine 9(5):548-553.
Lindesmith, L., Moe, C., Lependu, J., Frelinger, J.A., Treanor, J., and Baric, R.S. 2005. Cellular
and humoral immunity following Snow Mountain virus challenge. Journal of Virology 79:2900-
2909.
Lloyd, A.R., Wakefield, D., Boughton, C., and Dwyer, J. 1988. What is myalgic
encephalomyelitis? Lancet 1(8597): 1286-1287.
Locht, H. and Krogfelt, K. A. 2002. Comparison of rheumatological and gastrointestinal
symptoms after infection with Campylobacter jejuni/coli and enterotoxigenic Escherichia coli.
Annals of the Rheumatic Diseases 61:448-452
Loewe, L., Textor, V., and Scherer, S. 2003. High deleterious genomic mutation rate in
stationary phase of Escherichia coli. Science 302:1558-1559.
Luo, G., Seetharamaiah, G.S., Niesel, D.W., Zhang, H., Peterson, J.W., Prabhakar, B.S., and
Klimpel, G.R. 1994. Purification and characterization of Yersinia enterocolitica envelope
proteins which induce antibodies that react with human thyrotropin receptor. Journal of
Immunology 152(5):2555-2561.
Lyons, C.R., and Wu, T.H. 2007. Animal models of Francisella tularensis infection. Annals of
the New York Academy of Science 1105:238-265.
Marciano-Cabral, F., and Cabral, G. 2003. Acanthamoeba spp. as agents of disease in humans.
Clinical Microbiology Reviews 16(2):273-307.
130
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Marciano-Cabral, F., MacLean, R., Mensah, A., and LaPat-Polasko, L. 2003. Identification of
Naegleria fowleri in domestic water sources by nested PCR. Applied and Environmental
Microbiology 69(10):5864-5869.
Marciano-Cabral, F., Puffenbarger, R., and Cabral, G.A. 2000.The increasing importance of
Acanthamoeba infections. Journal of Eukaryotic Microbiology 47(1): 29-36.
Marionneau, S., Ruvoen, N., Le Moullac-Vaidye, B., Clement, M., Cailleau-Thomas, A., Ruiz-
Palacois, G., Huang, P., Jiang, X., and Le Pendu, J. 2002. Norwalk virus binds to histo-blood
group antigens present on gastroduodenal epithelial cells of secretor individuals.
Gastroenterology 122(7): 1967-1977.
Marshall, M.M., Naumovitz, D., Ortega, Y., and Sterling, C.R. 1997. Waterborne protozoan
pathogens. Clinical Microbiology Reviews 10(l):67-85.
Maunula, L., Miettinen, I.T., and von Bonsdorff ,C.H. 2005. Norovirus outbreaks from drinking
water. Emerging Infectious Diseases 11:1716-1721.
McBride, G.B., Till, D., Ryan, T., Ball, A., Lewis, G., Palmer, S., and Weinstein, P. 2002.
Freshwater Microbiology Research Programme. Pathogen Occurrence and Human Health Risk
Assessment Analysis. Technical Publication, 93 pp. Wellington, New Zealand: Ministry for the
Environment, http://www.mfe.govt.nz/publications/water/freshwater-microbiologv-
nov02/freshwater-microbiology-nov02.pdf.
McCarthy, N., and Giesecke, J. 2001. Incidence of Guillain-Barre syndrome following infection
with Campylobacter jejuni. American Journal of Epidemiology 153(6):610-614.
McDaniels, A.E., Wymer, L., Rankin C, and Haugland, R. 2005. Evaluation of quantitative real
time PCR for the measurement of Helicobacter pylori at low concentrations in drinking water.
Water Research 39:4808-4816.
Mead, P.S., Slutsker, L., Dietz,V., McCaig, L.F., Bresee, J.S., Shapiro, C., Griffin, P.M., and
Tauxe, R.V. 1999. Food-related illness and death in the United States. Emerging Infectious
Diseases 5(5):607-625.
Medema, G.J., Teunis, P.F., Havelaar, A.H., and Haas, C.N. 1996. Assessment of the dose-
response relationship of Campylobacter jejuni. International Journal of Food Microbiology
30:101-111.
Medema, G., and Smeets, P. 2004. The Interaction Between Quantitative Microbial Risk
Assessment and Risk Management in the Water Safety Plan. Kiwa Water Research/Delft
University.
http://217.77.141.80/clueadeau/microrisk/uploads/interaction in water safety plan.pdf.
Meier, P.A., Mathers, W.D., Sutphin, J.E., Folberg, R., Hwang, T., and Wenzel, R.P. 1998. An
epidemic of presumed Acanthamoeba keratitis that followed regional flooding. Results of a case-
control investigation. Archive of Ophthalmology 116:1090-1094.
131
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Mena, K.D., and Gerba, C.P. 2009. Waterborne adenovirus. Reviews of Environmental
Contamination and Toxicology 198:133-167.
Messner, M.J., Chappell, C.L., and Okhuysen, P.C. 2001. Water risk assessment for
Cryptosporidium-, a hierarchical Bayesian analysis of human dose response data. Water Research
35(16):3934-3940.
Messner, M.J., Berger, P., Nappier, S.P. 2014. Fractional Poisson - A Simple Dose-Response
Model for Human Norovirus. (in press)
Mintz, E.D., Hudson-Wragg, M., Mshar, P., Cartter, M.L., and Hadler, J.L. 1993. Foodborne
giardiasis in a corporate office setting. Journal of Infectious Diseases 167(l):250-253.
Moe, C.L., Frelinger, J. A., Heizer, W., and Stewart, P. 2002. Studies of the infectivity of
Norwalk and Norwalk-like viruses. Submitted to EPA National Center for Environmental
Research (#R826139). Washington, DC: EPA.
http://cfpub2.epa.gov/ncer abstracts/index.cfm/fuseaction/displav.abstractDetail/abstract/192/rep
ort/F
Moon, H., Chen, J. J., Gaylor, D.W., and Kodell, R.L. 2004. A comparison of microbial dose-
response models fitted to human data. Regulatory Toxicology and Pharmacology 40:177-184.
Moon, H., Kim, H-J., Chen, J. J., and Kodell, R.L. 2005. Model averaging using the Kullback
Information Criterion in estimating effective doses for microbial infection and illness. Risk
Analysis 25(5): 1147-1159.
Morgan, B. J.T., and Watts, S.A. 1980. On modelling microbial infections. Biometrics 36:317-
321.
Morgan, M.G., and Henrion, M. eds. 1990. Uncertainty: A Guide to Dealing with Uncertainty in
Quantitative Risk and Policy Analysis. New York: Cambridge University Press.
Morgan, U.M., Xiao, L., Sulaiman, I., Weber, R., Lai, A.A., Thomson, R.C., and Deplazes, P.
1999. Which genotypes/species of Cryptosporidium are humans susceptible to? Journal of
Eukaryotic Microbiology 46(5):42S-43S.
Motaijemi, Y. 2002. Chronic Sequelae of Foodborne Infections. In Foodborne Pathogens, eds. C.
d. Blackburn and P. J. McClure. Pp. 501-513. Boca Raton, FL: CRC Press.
Moyer, N.P. 2006. Aeromonas. Pp. 81-85 in Waterborne Pathogens. AWWA Manual M48, 2nd
Edition. Denver, CO: American Waterworks Association.
Moyer, N.P., and Degnan, A.J. 2006. Shigella. Pp. 145-148 in Waterborne Pathogens. AWWA
Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Muniesa, M., Jofre, J., Garcia-Aljaro, C., and Blanch, A.R. 2006. Occurrence of Escherichia coli
0157:H7 and other enterohemorrhagic Escherichia coli in the environment. Environmental
Science & Technology 40(23):7141-7149.
132
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Mushahwar, I.K. 2008. Hepatitis E virus: molecular virology, clinical features, diagnosis,
transmission, epidemiology, and prevention. Journal of Medical Virology 80:646-658.
Nachamkin, I., Alios, B.M., and Ho, T.W. 2000. Campylobacter jejuni infection and the
association with Guillain-Barre' syndrome. In Campylobacter (2nd edition). Eds. I. Nachamkin
and M.J. Blaser. Pp. 155-178. Washington, DC: American Society for Microbiology.
NACMCF (National Advisory Committee on Microbiological Criteria for Foods). 1997. Hazard
Analysis and Critical Control Point Principles and Application Guidelines. Adopted August 14,
1997.
http://www.fda.gov/Food/FoodSafetv/HazardAnalvsisCriticalControlPointsHACCP/HACCPPrin
ciplesApplicationGuidelines/default.htm.
Namata, H., Aerts, M., Faes, C., and Teunis P. 2008. Model averaging in microbial risk
assessment using fractional polynomials. Risk Analysis 28(4):891-905.
Nauta, M., Hill, A., Rosenquist, H., Brynestad, S., Fetsch, A., van der Logt, P., Fazil, A.,
Christensen, B., Katsma, E., Borck, B., and Havelaar, A. 2009. A comparison of risk assessments
on Campylobacter in broiler meat. International Journal of Food Microbiology 129(2): 107-123.
Niyogi, S.K. 2005. Shigellosis. Journal of Microbiology 43(2): 133-143.
Noonburg, G.E. 2005. Management of extremity trauma and related infections occurring in the
aquatic environment. Journal of the American Academy of Orthopaedic Surgeons 13(4):243-253.
NRC (National Research Council). 1983. Risk Assessment in the Federal Government:
Managing the Process. Washington, DC: National Academy Press.
NRC. 2004. Indicators for Waterborne Pathogens. Washington, DC: National Academies Press.
NRC. 2009. Science and Decisions: Advancing Risk Assessment. Washington, DC: National
Academies Press.
NZ MFE (Ministry for the Environment, New Zealand). 2003. Microbiological Water Quality
Guidelines for Marine and Fresh water Recreational Areas. Available
at:http://www.mfe.govt.nz/publications/water/microbiological-qualitv-iun03/microbiological-
qualitv-iun03.pdf
O'Donoghue, P. 1995. Cryptosporidium and cryptosporidiosis in man and animals. International
Journal for Parasitology 25:2:139-195.
Okhuysen, P.C., Chappell, C.L., Crabb, J.H., Sterling, C.R., and DuPont, H.L. 1999. Virulence
of three distinct Cryptosporidium parvum isolates for healthy adults. Journal of Infectious
Diseases 180(4): 1275-1281.
Okhuysen, P.C., Rich, S.M., Chappell, C.L., Grimes, K.A., Widmer, G., Feng, X., and Tzipori,
S. 2002. Infectivity of a Cryptosporidium parvum isolate of cervine origin for healthy adults and
interferon-gamma knockout mice. Journal of Infectious Diseases 185(9): 1320-1325.
133
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Olivieri, A.W., Eisenberg, D., Soller, J., Eisenberg, J., Trussell, R., and Gagliardo, P. 1999.
Estimation of pathogen removal in an advanced water treatment facility using Monte Carlo
simulation. Water Science and Technology 40:4-5.
OMB (Office of Management and Budget). 2004. Revised Information Quality Bulletin for Peer
Review, April 2004. http://www.whitehouse.gov/omb/inforeg/peer review041404.pdf.
Omenn, G.S., Kessler, A.C., Anderson, N.T., Chiu, P.Y., Doull, J., Goldstein B., Lederberg, J.,
McGuire, S., Rail, D., and Weldon, V.V. 1997. Framework for Environmental Health Risk
Management/Risk Assessment and Risk Management in Regulatory Decision-Making: Final
Report (2 vol.). Presidential/Congressional Commission on Risk Assessment and Risk
Management. http://www.riskworld.com/Nreports/1997/risk-rpt/pdf/epaian.pdf.
Parkhurst, D.F., Craun, G.F., and Soller, J. 2007. Conceptual Bases for Relating Illness Risk to
Indicator Concentrations. In Statistical Framework for Recreational Water Quality Criteria and
Monitoring. EPA, Office of Research and Development Monograph.
Parkin, R.T., and Balbus, J.M. 2000. Variations in concepts of "susceptibility" in risk
assessment. Risk Analysis 20:603-611.
Parkin, R.T., Soller, J.A., and Olivieri, A.W. 2003. Incorporating susceptible subpopulations in
microbial risk assessment: pediatric exposures to enteroviruses in river water. 13: 161-168.
Parsonnet, J., Vandersteen, D., Goates, J., Sibley, R.K., Pritikin, J., and Chang, Y. 1991.
Helicobacter pylori infection in intestinal- and diffuse-type gastric adenocarcinomas. Journal of
the National Cancer Institute 83(9):640-3.
Petterson, S., Signor, R., Ashbolt, N., and Roser, D. 2006. QMRA Methodology. In Quantitative
Microbial Risk Assessment in the Water Safety Plan. In Final Report on the EU MicroRisk
Project, Medema, G., Loret, J.-C., Stenstrom, T.A., and Ashbolt, N., eds (European Commission,
Brussels), pp. 3-64. Available at:
http://www.microrisk.com/uploads/microrisk_qmra_methodology.pdf.
Petterson, S.R., Signor, R.S., and Ashbolt, N.J. 2007. Incorporating method recovery
uncertainties in stochastic estimates of raw water protozoan concentrations for QMRA. Journal
of Water and Health 5(S1):51-65.
Pinsky, P.F. 2000. Assessment of risks from long term exposure to waterborne pathogens.
Environmental and Ecological Statistics 7:155-175.
Plutzer, J., Ongerth, J., and Karanis, P. 2010. Giardia taxonomy, phylogeny and epidemiology:
facts and open questions. International Journal of Hygiene and Environmental Health 213:321-
333.
Pope, J.E., Krizova, A., Garg, A.X., Thiessen-Philbrook, H., and Ouimet, J.A. 2007.
Campylobacter reactive arthritis: a systematic review. Seminars in Arthritis and Rheumatism
37:48-55.
134
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Pouillot, R., Beaudeau, P., Denis, J.-B., and Derouin, F. for the AFSSA Cryptosporidium Study
Group. 2004. A Quantitative Risk Assessment of Waterborne Cryptosporidiosis in France Using
Second-Order Monte Carlo Simulation. Risk Analysis 24(1): 1-17.
Primm, T.P., Lucero, C.A., and Falkinham, J.O. 3rd. 2004. Health impacts of environmental
Mycobacteria. Clinical Microbiology Reviews 17(1):98-106.
Rambaud, J.C., Halphen, M., Galian, A., and Tsapis, A. 1990. Immunoproliferative small
intestinal disease (IPSID): relationships with a-chain disease and "Mediterranean" lymphomas.
Springer Seminars in Immunopathology 12(2-3):239-250.
Rangel, J.M., Sparling, P.H., Crowe, C., Griffin, P.M., and Swerdlow, D.L. 2005. Epidemiology
of Escherichia coli 0157:H7 outbreaks, United States, 1982-2002. Emerging Infectious Diseases
11:603-609.
Reavis, C.J. 2005. Rural health alert: Helicobacter pylori in well water. American Academy of
Nurse Practitioners 17(7):283-299.
Regli, S., Rose, J.B., Haas, C.N., and Gerba, C.P. 1991. Modeling the risk from Giardia and
viruses in drinking-water. Journal of the American Water Works Association 83:76-84.
Rendtorff, R.C., 1954a. The experimental transmission of human intestinal protozoan parasites.
I. Endamoeba coli cysts given in capsules. American Journal of Hygiene 59:196-208.
Rendtorff, R.C., 1954b. The experimental transmission of human intestinal protozoan parasites.
II. Giardia lamblia cysts given in capsules. American Journal of Hygiene 59:209-220.
Rendtorff, R.C., and Holt, C.J. 1954c. The experimental transmission of human intestinal
protozoan parasites. III. Attempts to transmit Endamoeba coli and Giardia lamblia by flies. The
experimental transmission of human intestinal protozoan parasites. I. Endamoeba coli cysts
given in capsules. American Journal of Hygiene 60:320-326.
Rendtorff, R.C. and Holt, C.J. 1954d. The experimental transmission of human intestinal
protozoan parasites. IV. Attempts to transmit Endamoeba coli and Giardia lamblia by water.
American Journal of Hygiene 60:327-338.
Richardson, R.F., Jr., Remler, B.F., Katirji, B., and Murad, M.H. 1998. Guillain-Barre syndrome
after Cyclospora infection. Muscle Nerve 21(5):669-671.
Rodrigquez-Hernandez, J., Canut-Blasco, A., and Martin-Sanchez, A.M. 1996. Seasonal
relevance of Cryptosporidium and Giardia infections in children attending day care centers in
Salamanca (Spain) studied for a period of 15 months. European Journal of Epidemiology 12:291-
295.
Rose, J.B., and Gerba, C.P. 1991. Use of risk assessment for development of microbial
standards. Water Science and Technology 24:29-34.
135
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Rose, J.B., and Grimes, D.J. 2001. Reevaluation of Microbial Water Quality: Powerful New
Tools for Detection and Risk Assessment. Washington, DC: American Academy of
Microbiology.
Rose, J.B., Haas, C.N., and Regli, S. 1991. Risk assessment and control of waterborne giardiasis.
American Journal of Public Health 81:709-713.
Rose, J.B., and Sobsey, M.D. 1993. Quantitative risk assessment for viral contamination of
shellfish and coastal waters. Journal of Food Protection 56(12): 1043-1050.
Roseberry, A.M. and Burmaster, D.E. 1992. Lognormal distribution for water intake by children
and adults. Risk Analysis 12:99-104.
Roser, D.J., Davies, C.M., Ashbolt, N.J., and Morison, P. 2006. Microbial exposure assessment
of an urban recreational lake: a case study of the application of new risk-based guidelines. Water
Science and Technology 54:245-252.
Roy, S.L., DeLong, S.M., Stenzel, S.A., Shiferaw, B., Roberts, J.M., Khalakdina, A., Marcus, R.,
Segler, S.D., Shah, D.D., Thomas, S., Vugia, D.J., Zansky, S.M., Dietz, V., and Beach, M.J.
2004. Risk factors for sporadic cryptosporidiosis among immunocompetent persons in the United
States from 1999 to 2001. Journal of Clinical Microbiology 42:2944-2951.
Sanchez, J. F., Olmedo, M. C., Pascua, F. J., and Casado, I. 2000. Diabetes insipidus as a
manifestation of cerebral toxoplasmosis in an AIDS patient. Revista de Neurologia 30(10):939-
940.
Sartwell, P.E. 1950. The distribution of incubation periods of infectious disease. American
Journal of Hygiene 51:310-318.
Scallan, E., Hoekstra, R.M., Angulo, F.J., Tauxe, R.V., Widdowson, M.A., Roy, S.L., Jones,
J.L., and P.M. Griffin. 201 la. Foodborne illness acquired in the United States — major
pathogens. Emerging Infectious Diseases, 17:7-15.
Scallan, E., Griffin, P.M., Angulo, F.J., Tauxe, R.V., and R.M. Hoekstra. 2011b. Foodborne
illness acquired in the United States — unspecified agents. Emerging Infectious Diseases, 17:16-
22.
Schets, F.M., Schijven, J.F., and de Roda Husman, A.M. 2011. Exposure assessment for
swimmers in bathing waters and swimming pools. Water Research 45:2392-2400.
Schiellerup, P., Krogfelt, K.A., and Locht, H. 2008. A comparison of self-reported joint
symptoms following infection with different enteric pathogens: Effect of HLA-B27. Journal of
Rheumatology 35(3):480-487.
Schiff, G.M., Stefanovic, G.M., Young, E.C., Sander, D.S., Pennekamp, J.K., and Ward, R.L.
1984. Studies of echovirus-12 in volunteers: determination of minimal infectious dose and the
effect of previous infection on infectious dose. Journal of Infectious Disease 150(6): 858-866.
136
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Schuster, C.J., Ellis, A.G., Robertson, W.J., Charron, D.F., Aramini, J.J., Marshall, B.J., and
Medeiros, D.T. 2005. Infectious disease outbreaks related to drinking water in Canada, 1974-
2001. Canadian Journal Public Health 96(4):254-258.
Schuster, M.H., and Visvesvara, G.S. 2004. Pathogenic and opportunistic free-living amoebae:
Acanthamoeba spp., Balamuthia mandrillaris, Naegleria fowleri, and Sappinia diploidea. FEMS
Immunology and Medical Microbiology 50(1): 1-26.
Schwab, K.J. Astroviruses. 2006. Pp. 259-262 in Waterborne Pathogens. AWWA Manual M48,
2nd Edition. Denver, CO: American Water Works Association.
Schwab, K.J. and Hurst, C. 2006. Human Caliciviruses. Pp. 281-286 in Waterborne Pathogens.
AWWA Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Seto, E.Y., Soller, J.A., and Colford, J.M., Jr. 2007. Strategies to reduce person-to-person
transmission during widespread Escherichia coli 0157:H7 outbreak. Emerging Infectious
Diseases 13:860-866.
Sharma S., Sachdeva, P., and Virdi, J.S. 2003. Emerging water-borne pathogens. Applied
Microbiology and Biotechnology 61(5-6):424-428.
Shaywitz, S.E., Cohen, P.M., Cohen, D.J., Mikkelson, E., Morowitz, G., and Shaywitz, B.A.
1982. Long-term consequences of Reye syndrome: a sibling-matched, controlled study of
neurologic, cognitive, academic, and psychiatric function. Journal of Pediatrics 100(l):41-46.
Sheng, L., Eisenberg, J.N.S., Spiknall, I., Koopman, J.S. 2009. Dynamics and Control of
Infections Transmitted from Person to Person through the Environment. American Journal of
Epidemiology 170 (2): 257-265.
Shirtliff, M.E., and Mader, J.T. 2002. Acute septic arthritis. Clinical Microbiology Reviews
15(4):527-544.
Sifuentes-Osornio, J., Porras-Cortes, G., Bendall, R.P., Morales-Villarreal, F., Reyes-Teran, G.,
and Ruiz-Palacios, G.M. 1995. Cyclospora cayetanensis infection in patients with and without
AIDS: biliary disease as another clinical manifestation. Clinical Infectious Disease 21(5): 1092-
1097.
Signor, R.S., Roser, D.J., Ashbolt, N.J., and Ball, J.E. 2005. Quantifying the impact of runoff
events on microbiological contaminant concentrations entering surface drinking source waters.
Journal of Water and Health 3(4):453-68.
Slifko, T.R., Friedman, D., Rose, J.B., and Jankubowski, W. 1997. An in vitro method for
detecting infectious Crytosporidium oocysts with cell culture. Applied and Environmental
Microbiology 63(9):3669-3675.
Slifko, T.R., Huffman, D.E. and Rose, J.B. 1999. A most probable assay for enumeration of
infectious Cryptosporidium parvum oocycts. Applied and Environmental Microbiology
65(9):3936-3941.
137
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Slifko, T.R., Huffman, D.E., Bertrand, D., Owens, J.H., Jakubowski, W., Haas, C.N. and Rose,
J.B. 2002. Comparison of animal infectivity and cell culture systems for evaluation of
Cryptosporidiumparvum oocycts. Experimental Parasitology 101:97-106.
Smeets, P.W., Dullemont, Y.J., Van Gelder, P.H., Van Dijk, J.C., and Medema, G.J. 2008.
Improved methods for modelling drinking water treatment in quantitative microbial risk
assessment; a case study of Campylobacter reduction by filtration and ozonation. Journal of
Water and Health 6:301-314.
Smith, J.W., and Wolfe, M.S. 1980. Giardiasis. Annual Review of Medicine 31:373-383.
Smith, A., Reacher, M., Smerdon, W., Adak, G.K., Nichols, G.,and Chalmers, R.M. 2006.
Outbreaks of waterborne infectious intestinal disease in England and Wales, 1992-2003.
Epidemiology and Infection 134(6): 1141-1149.
Snedecor, G.W., and Cochran, W.G. 1989. Statistical Methods, 8th Edition, Iowa State
University Press.
Snelling, W.J., Matsuda, M., Moore, J.E., and Dooley, J.S. 2006. Under the microscope:
Arcobacter. Letters in Applied Microbiology 42:7-14.
Sobsey, M.D. 2006. Hepatitis A Virus. Pp. 273-279 in Waterborne Pathogens. AWWA Manual
M48, 2nd Edition. Denver, CO: American Water Works Association.
Soller, J. A., Eisenberg, J.N., and Olivieri, A.W. 1999. Evaluation of Pathogen Risk Assessment
Framework. Oakland, CA: Eisenberg, Olivieri and Associates.
Soller, J. A., Olivieri, A., Crook, J., Parkin, R., Spear, R., Tchobanoglous, G., and Eisenberg,
J.N.S. 2003. Risk-based approach to evaluate the public health benefit of additional wastewater
treatment. Environmental Science & Technology 37:1882-1891.
Soller, J.A., Eisenberg, J., DeGeorge, J., Cooper, R., Tchobanoglous, G., and Olivieri, A. 2006.
A public health evaluation of recreational water impairment. Journal of Water and Health 4:1-19.
Soller, J.A., Seto, E.Y., and Olivieri, A.W. 2007. Application of Microbial Risk Assessment
Techniques to Estimate Risk Due to Exposure to Reclaimed Waters. WateReuse Foundation,
Final Project Report WRF-04-011.
Soller, J. A., and Eisenberg, J.N.S. 2008. An evaluation of parsimony for microbial risk
assessment models. Environmetrics 19:61-78.
Soller, J. A. 2009. Potential implications of person-to-person transmission of viral infection to US
EPA's Groundwater Rule. Journal of Water and Health 7:208-223.
Soller, J.A., Schoen, M.E., Bartrand, T., Ravenscroft, J., and Ashbolt, N.J. 2010a. Estimated
human health risks from exposure to recreational waters impacted by human and non-human
sources of faecal contamination. Water Research: 44(16):4674-91.
138
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Soller, J.A., Bartrand, T., Ashbolt, N.J., Ravenscroft, J., and Wade, T.J. 2010c. Estimating the
primary etiologic agents in recreational freshwaters impacted by human sources of faecal
contamination. Water Research 44(16): 4736-4747.
Soller, J.A., Schoen, M.E., Varghese, A., Ichida, A.M., Boehm, A.B., Eftim, S., Ashbolt, N.J.,
Ravenscroft, J.E. 2014. Human Health Risk Implications of Multiple Sources of Faecal Indicator
Bacteria in a Recreational Waterbody. Water Research 66:254-264.
Sridharan, S., Mossad, S., and Hoffman, G. 2000.Hepatitis A infection mimicking adult onset
Still's disease. Journal of Rheumatology 27(7): 1792-1795.
Steinberg, E.B., Greene, K.D., Bopp, C.A., Cameron, D.N., Wells, J.G., and Mintz, E.D. 2001.
Cholera in the United States, 1995-2000: trends at the end of the Twentieth Century. Journal of
Infectious Diseases 184:799-802.
Stewart, I., Carmichael, W.W., Sadler, R., McGregor, G.B., Reardon, K., Eaglesham, G.K.,
Wickramasinghe, W.A., Seawright, A.A., and Shaw, G.R. 2009. Occupational and
environmental hazard assessments for the isolation, purification and toxicity testing of
cyanobacterial toxins. Environmental Health 8:52.
Stone, P. 2006. EU Private International Law: Harmonization of Laws (Elgar European Law
Series). Cheltenham, UK: Edwin Elgar Publishing Limited.
Sur, D., Ramamurthy, T., Deen, J., and Bhattacharya, S.K. 2004 Shigellosis: challenges and
management issues. Indian Journal of Medical Research 120:454-462.
Tanaka, H., Asano, T., Schroeder, E.D., and Tchobanoglous, G. 1998. Estimating the safety of
wastewater reclamation and reuse using enteric virus monitoring data. Water Environment
Research 70(1):39-51.
Teunis, P.F., van der Heijden, O.G., van der Giessen, J.W.B., and Havelaar, A.H. 1996. The
Dose-Response Relation in Human Volunteers for Gastro-intestinal Pathogens. Report No.
284550002. Bilthoven, The Netherlands: RIVM (National Institute of Public Health and the
Environment).
Teunis, P.F.M., and Havelaar, A.H. 1999. Cryptosporidium in Drinking Water: Evaluation of the
ILSI/RSI Quantitative Risk Assessment Framework. Report No. 284 550 006. Bilthoven, The
Netherlands: RIVM.
Teunis, P.F.M., and Havelaar, A. 2000. The beta-Poisson model is not a single hit model. Risk
Analysis 20(4):513-520.
Teunis, P.F., Chappell, C.L., and Okhuysen, P.C. 2002 Cryptosporidium dose response studies:
Variation between isolates. Risk Analysis 22(1): 175-183.
Teunis, P., Takumi, K., and Shinagawa, K. 2004. Dose response for infection by Escherichia coli
0157:H7 from outbreak data. Risk Analysis 24(2):401-407.
139
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Teunis, P.F.M., van den Brandhof, W., Nauta, M., Wagenaar, J., van den Kerkhof, H., and Van
Pelt, W. 2005. A reconsideration of the Campylobacter dose-response relation. Epidemiology
and Infection 133:583-592.
Teunis, P.F.M., Ogden, I.D., and Strachan, N.J.C. 2008a. Hierarchical dose response of E. coli
0157:H7 from human outbreaks incorporating heterogeneity in exposure. Epidemiology and
Infection 136(6):761-770.
Teunis, P.F.M., Moe, C.L., Liu, P., Miller, S.E., Lindesmith, L., Baric, R.S., Pendu, J.L., and
Calderon, R.L. 2008b. Norwalk virus: how infectious is it? Journal of Medical Virology
80(8): 1468-1476.
Teunis, P.F.M. 2009. Uncertainty in Dose Response from the Perspective of Microbial Dose. Ch.
6 in Uncertainty Modeling in Dose Response. Hoboken, NJ: Wiley.
Teunis, P.F., Kasuga, F., Fazil, A., Ogden, I.D., Rotariu, O., and Strachan, N.J. 2010. Dose-
response modeling of Salmonella using outbreak data. International Journal of Food
Microbiology 144(2): 243-249.
Thebpatiphat, N., Hammersmith, K.M., Rocha, F.N., Rapuano, C.J., Ayres, B.D., Laibson, P.R.,
Eagle, R.C., Jr., and Cohen, E.J. 2007. Acanthamoeba keratitis: a parasite on the rise. Cornea
26(6):701-706.
Thoeye, C., Van Eyck, K., Bixio, D., Weemaes, M., and De Gueldre, G. 2003. Methods Used for
Health Risk Assessment in State of the Art Report: Health Risks in Aquifer Recharge Using
Reclaimed Water. Geneva, Switzerland: WHO.
http://www.who.int/water sanitation health/wastewater/en/wsh0308chap4.pdf.
Thomson, G.T.D., Derubeis, D.A., Hodge, M.A., Rajanayagam, C., and Inman, R.D. 1995. Post-
Salmonella reactive arthritis: late clinical sequelae in a point source cohort. American Journal of
Medicine 98(1): 13-21.
Thompson R.C. 2000. Giardiasis as a re-emerging infectious disease and its zoonotic potential.
International Journal of Parasitology 30:1259-1267.
Toranzos, G.A., et al. (2006). Vibrio cholerae. Pp. 153-156 in Waterborne Pathogens. AWWA
Manual M48, 2nd Edition. Denver, CO: American Water Works Association.
Trachoo, N. 2003. Campylobacter jejuni: an emerging pathogen. Songklanakarin Journal of
Science and Technology 25(1): 141-157.
Tuncay, S., Deliba§, S., Inceboz, T., Over, L., Oral, A.M., Akisii, C., and Aksoy, U. 2008. An
outbreak of gastroenteritis associated with intestinal parasites. Tiirkiye parazitolojii dergisi
32(3):249-252.
Tupchong, M., Simor, A., and Dewar, C. 1999. Beaver fever—a rare cause of reactive arthritis.
Journal of Rheumatology 26:2701-2702.
140
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
USDA (U.S. Department of Agriculture). 1998. Salmonella Enteritidis Risk Assessment Shell
Eggs and Egg Products, http://www.fsis.usda.gov/ophs/risk/pdfriskl.pdf.
U.S. Census Bureau. 201 la. Table 11. Resident Population by Race, Hispanic Origin, and Single
Years of Age: 2009. U.S. Census Bureau, Statistical Abstract of the United States: 2011.
Available at: http://www.census.gov/compendia/statab/2011/tables/l ls0010.pdf. Accessed on
October 18, 2012.
U.S. Census Bureau. 2011b. Table 1248. Participation in selected sports activities: 2008. U.S.
Census Bureau, Statistical Abstract of the United States: 2011. Available at:
http://www.census.gov/compendia/statab/2011/tables/l lsl248.pdf. Accessed on October 18,
2012.
U.S. EPA. 1989. Risk Assessment Guidance for Superfund. EPA-540/1-89/002. Washington,
DC. http://www.epa.gov/oswer/riskassessment/ragsa/index.htm.
U.S. EPA. 1992. Framework for Ecological Risk Assessment. EPA/630/R-92/001. Washington,
DC
U.S. EPA. 1995a. Guidance for Risk Characterization. U.S. Environmental Protection Agency,
Science Policy Council. Washington, DC. http://www.epa.gov/OSA/spc/pdfs/rcguide.pdf
U.S. EPA. 1995b. The EPA's Environmental Justice Strategy.
http://www.epa.gov/compliance/resources/policies/ei/ei strategy 1995.pdf.
U.S. EPA. 1997a. Exposure Factors Handbook. EPA/600/P-95/002Fa. Office of Research and
Development, National Center for Environmental Assessment. Washington, DC.
U.S. EPA. 1997b. Guiding Principles for Monte Carlo Analysis. EPA/630/R-97/001.
U.S. EPA. 1998a. Giardia: Human Health Criteria Document. EPA-823-R-002.
U.S. EPA. 1998b. Guidelines for Ecological Risk Assessment. May 14, 1998, Federal Register
63(93):26846-26924. EPA/630/R-95/002.F.
U.S. EPA. 1999b. Legionella: Human Health Criteria Document. EPA-822-R-99-001.
Washington, DC.
U.S. EPA. 1999c. Report of the Workshop on Selecting Input Distributions for Probabilistic
Assessments. EPA/630/R-98/004. Risk Assessment Forum, Washington, DC.
U.S. EPA. 2000a. EPA Science Policy Council Peer Review Handbook. EPA-100-B-00-001.
Washington, DC.
U.S. EPA. 2000b. Science Policy Council Risk Characterization Handbook. EPA-100-B-00-002.
Washington, DC. http://www.epa.gov/osa/spc/pdfs/rchandbk.pdf.
141
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
U.S. EPA. 2000c. Methodology for Deriving Ambient Water Quality Criteria for the Protection
of Human Health. EPA-822-B-00-004.
http://www.epa.gov/waterscience/humanhealth/method/complete.pdf.
U.S. EPA. 2000d. Report to Congress EPA Studies on Sensitive Subpopulations and Drinking
Water Contaminants1. EPA 815-R-00-015.
http ://www. epa. gov//safewater/standard/rtc sensubpops .pdf.
U.S. EPA. 2000e. Guidelines for Preparing Economic Analyses. National Center for
Environmental Economics, EPA/240/R-00/003. Washington, DC.
U.S. EPA. 2001. Protocol for Developing Pathogen TMDLs. EPA 841-R-00-002.
http://www.epa.gov/owow/tmdl/pathogen all.pdf.
U.S. EPA. 2002a. Guidelines for Ensuring and Maximizing the Quality, Objectivity, Utility, and
Integrity, of Information Disseminated by the Environmental Protection Agency. EPA/260R-02-
008.
http://www.epa.gov/qualitv/informationguidelines/documents/EPA InfoOualitvGuidelines.pdf.
U.S. EPA. 2002b. Lessons Learned on Planning and Scoping for Environmental Risk
Assessments, EPA Science Policy Council. Washington, DC.
http://epa.gov/osp/spc/handbook.pdf.
U.S. EPA. 2002c. M/crobiological Risk Assessment Framework Workshop Tools, Methods, and
Approaches. Prepared by ICF Consulting, Inc.
U.S. EPA. 2002d. Child-Specific Exposure Factors Handbook. EPA-600-P-00-002B.
U.S. EPA. 2002e. National Primary Drinking Water Regulations: Long Term 1 Enhanced
Surface Water Treatment Rule. 40CFR Parts 9, 141 and 142, and Federal Register 67(9),
January 14, 2002. http ://www. epa. gov/safewater/mdbp/lt 1 eswtr. html.
U.S. EPA. 2003a. National Primary Drinking Water Regulations: Long Term 2 Enhanced
Surface Water Treatment Rule Proposal. Federal Register 68:154.
U.S. EPA. 2003b. Economic Analysis for the Long Term 2 Enhanced Surface Water Treatment
Rule Proposal.
U.S. EPA. 2003c. Microbiological Risk Assessment Framework: Problem Formulation
Workshop. Prepared by ICF Consulting, Inc.
U.S. EPA. 2003d. Movement and Longevity of Viruses in the Subsurface. National Risk
Management Research Laboratory. EPA/540/S-03/500.
http://www.epa.gov/ada/download/issue/540S03500.pdf.
U.S. EPA. 2003e. Integrated Risk Information, Glossary of IRIS Terms.
http ://www. epa. gov/iri s/gloss8. htm.
142
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
U.S. EPA. 2003f. Framework for Cumulative Risk Assessment. EPA/630/P-02/00IF. Office of
Research and Development, National Center for Environmental Assessment, Washington, DC.
http://www.epa.gov/raf/publications/pdfs/frmwrk cum risk assmnt.pdf.
U.S. EPA. 2003g. Public Involvement Policy of the U.S. Environmental Protection Agency. EPA
233-B-03-002. http://www.epa.gov/publicinvolvement/pdf/policv2003.pdf.
U.S. EPA. 2003h. Health Effects Support Document for Acanthamoeba. EPA-822-R-03-012.
Washington, DC.
http://www.epa.gov/safewater/ccl/pdfs/reg determine 1/support ccl acanthamoeba healtheffcts.
pdf.
U.S. EPA. 2004b. Air Toxics Risk Assessment Reference Library: Volume 1 Technical Resource
Manual. EPA-453-K-04-001A. http://www.epa.gov/ttn/fera/risk atra main.html.
U.S. EPA. 2004c. Developing Dynamic Infection Transmission Models for Microbial Risk
Assessment Applications. EPA-NCEA-C-1463.
U.S. EPA. 2004d. Risk Assessment Principles and Practices. EPA/100/B-04/001. Office of the
Science Advisor Staff Paper, http ://www. epa. gov/OS A/pdfs/ratf-final .pdf.
U.S. EPA. 2005a. Guidelines for Carcinogen Risk Assessment. EPA/630/P-03/001B.
http: //cfpub .epa. gov/ncea/cfm/recordi spl ay. cfm? dei d= 116283.
U.S. EPA. 2005b. Guidance on Selecting Age Groups for Monitoring and Assessing Childhood
Exposures to Environmental Contaminants. EPA/630/P-03/003F. National Center for
Environmental Assessment, Washington, DC.
U.S. EPA. 2006a. National Primary Drinking Water Regulations: Long Term 2 Enhanced
Surface Water Treatment Rule - Final. Federal Register 71(3).
http://www.epa.gov/safewater/disinfection/lt2/regulations.html.
U.S. EPA. 2006b. National Primary Drinking Water Regulations: Ground Water Rule. 40CFR
Parts 9, 141 and 142, Federal Register 71(216). http://www.epa.gov/safewater/disinfection/gwr/.
EPA. 2006c. Occurrence and Monitoring Document for Final Ground Water Rule. USEPA 815-
R-06-012. Office of Water, Washington DC.
U.S. EPA. 2006d. Aeromonas: Human Health Criteria Document. Washington, DC: Office of
Water.
U.S. EPA. 2006e. EPA Science Policy Council Peer Review Handbook. EPA/100/B-06/002.
Washington, DC.
U.S. EPA. 2007a. Thesaurus of Terms Used in Microbiological Risk Assessment. EPA Office of
Water, http://www.epa.gov/waterscience/criteria/humanhealth/microbial/thesaurus/index.html.
143
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
U.S. EPA. 2007b. Compendium of Prior and Current Microbial Risk Assessment Methods for
Use as a Basis for the Selection, Development, and Testing of a Preliminary Microbial Risk
Assessment Framework. EPA/600/R-07/129. National Homeland Security Research Center.
http ://www. epa. gov/NHSRC/pub s/600r07129.pdf.
U.S. EPA. 2008. Child-Specific Exposure Factors Handbook. Final Report. EPA/600/R-06/096F.
U.S. Environmental Protection Agency, Washington, DC.
U.S. EPA 2009a. Review of Zoonotic Pathogens in Ambient Waters. EPA 822-R-09-002.
http://water.epa.gov/scitech/swguidance/waterqualitv/standards/criteria/health/recreation/upload/
2009 07 16 criteria recreation zoonoticpathogensreview.pdf.
U.S. EPA. 2009b. Using Probabilistic Methods to Enhance the Role of Risk Analysis in Decision
Making With Case Study Examples (External Review Draft). EPA/100/R-09/001. Risk
Assessment Forum, Washington, DC.
U.S. EPA. 2010. Quantitative Mircobial Risk Assessment to Estimate Illness in Freshwater
Impacted by Agricultural Animal Sources of Fecal Contamination, EPA 822-R-10-005, Office of
Science and Technology, Washington, D.C.
U.S. EPA. 2011. Exposure Factors Handbook: 2011 Edition, EPA/600/R-090/052F.
http://www.epa.gov/ncea/efh/pdfs/efh-complete.pdf.
U.S. EPA. 2012a. Framework for Human Health Risk Assessment to Inform Decision Making.
EPA External Review Draft. 601-D12-001. http://www.epa.gov/raf/files/framework-document-7-
13-12.pdf.
U.S. EPA. 2012b. Peer Review Handbook, 3rd edition. EPA/100/B-06/002.
http://www.epa.gov/peerreview/pdfs/peer review handbook 2012.pdf.
U.S. EPA. 2012c. Recreational Water Quality Criteria. 820-F-12-058.
http://water.epa.gov/scitech/swguidance/standards/criteria/health/recreation/index.cfm
U.S. EPA/USDA. 2012. Microbial Risk Assessment Guideline Pathogenic Microorganisms with
Focus on Food and Water. EPA/100/J-12/001 and USDA/FSIS/2012-001.
http://www.epa.gov/raf/files/mra-guideline-iuly-final.pdf.
U.S. EPA. 2014. Overview of Technical Support Materials: A Guide to the Site-Specific
Alternative Criteria TSM documents. EPA-820-R-14-010.
Visvesvara, G.S., and Moura, H. 2006. Acanthamoeba spp. In Waterborne Pathogens. Pp. 141-
144. Denver, CO: American Waterworks Association.
Visvesvara, G. S., Moura, H., and Schuster, F. L. (2007). Pathogenic and opportunistic free-
living amoebae: Acanthamoeba spp., Balamuthia mandrillaris, Naegleria fowleri, and Sappinia
diploidea. FEMS Immunology and Medical Microbiology 50(1): 1-26.
Vivier, J.C., Ehlers, M.M., and Grabow, W.O. 2004. Detection of enteroviruses in treated
drinking water. Water Reearch 38:2699-2705.
144
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Wang, C., Yuan, Y., and Hunt, R.H. 2007. The association between Helicobacter pylori
infection and early gastric cancer: a meta-analysis. American Journal of Gastroenterology
102(8):1789-1798.
Walter, J.E., and Mitchell, D.K. 2000. Role of astroviruses in childhood diarrhea. Current
Opinion in Pediatrics 12(3):275-279.
Weir, E. 2000. Escherichia coli 0157:H7. Canadian Medical Association Journal 163(2):205.
Weir, M.H., Pepe Razzolini, M.T., Rose, J.B., and Masago, Y. 2011. Water reclamation redesign
for reducing Cryptosporidium risks at a recreational spray park using stochastic models. Water
Research 45(19): 6505-6514.
Westrell, T., Bergstedt, O., Stenstrom, T.A., and Ashbolt, N.J. 2003. A theoretical approach to
assess microbial risks due to failures in drinking-water systems. International Journal of
Environmental Health Research 13:181-197.
Westrell, T. 2004. Microbial Risk Assessment and its Implications for Risk Management in
Urban Water Systems. Doctoral Thesis, Linkoping University, Sweden: the Tema Institute,
Department of Water and Environmental Studies, http://www.diva-
portal.org/liu/abstract.xsql?dbid=4880.
WHO (World Health Organization). 2001. Water Quality: Guidelines, Standards and Health:
Assessment of Risk and Risk Management for Water-Related Infectious Disease. Eds. L.
Fewtrell, J. Bartram. Published on behalf of IWA Publishing, WHO and Swedish Institute for
Infectious Disease Control.
http://www.who.int/water sanitation health/dwq/whoiwa/en/index.html.
WHO. 2004. Waterborne Zoonoses: Identification, Causes and Control. World Health
Organization (WHO); Cotruvo, J.A., Dufour, A., Rees, G., Bartram, J., Carr, R., Cliver, D.O.,
Craun, G.F., Fayer, R., Gannon, V.P.J (eds). IWA Publishing: London, UK. ISBN: 1 84339 058
2. http://www.who.int/water sanitation health/diseases/zoonoses/en/.
WHO. 2009. Hepatitis E. Fact Sheet #280 (revised January 2005)
http://www.who.int/mediacentre/factsheets/fs280/en/index.html.
Williams, T. 1965. The basic birth-death model for microbial infections. Journal of the Royal
Statistical Society Part B 27(2):338-360.
Willner, I.R., Uhl, M.D., Howard, S.C., Williams, E.Q., Riely, C.A., and Waters, B. 1998.
Serious hepatitis A: an analysis of patients hospitalized during an urban epidemic in the United
States. Annals of Internal Medicine (2): 111-114.
Wolf, M.W., Misaki, T., Bech, K., Tvede, M., Silva, J. E., and Ingbar, S.H. 1991.
Immunoglobulins of patients recovering from Yersinia enterocolitica infections exhibit Graves'
disease-like activity in human thyroid membranes. Thyroid 1(4):315-320.
145
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Wolfe, M.S. 1990. Clinical Symptoms and Diagnosis by Traditional Methods. In Human
Parasitic Diseases, Vol. 3, Giardiasis, ed. E.A. Meyer, pp. 175-186, Amsterdam: Elsevier
Science Publishers.
Wong, T.Y., Tsui, H.Y., So, M.K., Lai, J.Y., Lai, S.T., Tse, C.W., and Ng T.K. 2000.
Plesiomonas shigelloides infection in Hong Kong: retrospective study of 167 laboratory-
confirmed cases. Hong Kong Medical Journal 6(4):375-380.
Wooldridge, M., and Schaffner, D., eds. 2008. Qualitative Risk Assessment. In Microbial Risk
Analysis of Foods. Washington, DC: ASM Press.
Xiao, L., Morgan, U.M., Fayer, R., Thompson, C., and Lai, A. A. 2000. Cryptosporidium
systematics and implications for public health. Parisitology Today 16(7):287-292.
Yoder, J., Roberts, V., Craun, G. F., Hill, V., Hicks, L. A., Alexander, N. T., Radke, V.,
Calderon, R.L., Hlavsa, M.C., Beach, M.J., and Roy, S.L. 2008. Surveillance for waterborne
disease and outbreaks associated with drinking water and water not intended for drinking—
United States, 2005-2006. MMWR Surveillance Summaries 57(9):39-62.
Yu, S.Z. 1989. Drinking Water and Primary Liver Cancer. Pp. 30-37 in Primary Liver Cancer,
eds. Y. Tang, M.C. Wu, S.S. Xia, New York: China Academic Publishers.
Yu, S.Z. 1995. Primary prevention of hepatocellular carcinoma. Journal of Gastroenterology and
Hepatology 10(6):674-682.
Yu, S., Zhao, N., and Zi, X. 2001. The relationship between cyanotoxin (microcystin, MC) in
pond-ditch water and primary liver cancer in China. Zhonghua Zhong Liu Za Zhi 23(2): 96-99.
Yuki, N. 2001. Infectious origins of, and molecular mimicry in Guillain-Barre and Fisher
syndromes. The Lancet Infectious Diseases l(l):29-37.
Zelner J., King, A.A., Moe C.L., and J.N.S. Eisenberg. 2010. How Infections Propagate After
Point Source Outbreaks: An Analysis of Secondary Norovirus Transmission. Epidemiology
21(5).
146
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Appendix A. Possible Future MRA Goals and Research Needs
Some examples of possible long-term development goals for microbial risk assessment include
the following:
A.l. Exposure Assessment
• Considering how animal reservoirs of disease might be incorporated into MRA.
• Developing better methods to account for the heterogeneous distribution of
microorganisms and the potential fluctuations in density of microorganisms in the
environment (spatial heterogeneity and temporal fluctuations).
• Developing methods to address relative source contribution for microbial risks; that is,
evaluating the relative contribution of drinking water and other pathways (such as food,
swimming/recreational, and other environmental exposures) to the total disease risk from
all sources. This can also include the development of microbial bioaccumulation factors
for organisms that can accumulate human pathogens and are eaten by humans raw or
partially cooked (e.g., shellfish). This can be based on an understanding of "disease
ecology" (e.g., consider all exposures that result in a given health endpoint) rather than on
common assumptions that tend to simplify that understanding (e.g., exposure via a single
pathway).
A.2. Human Health Effects Assessment and Dose-Response
• Developing dose-response models that consider situations where populations may be
repeatedly exposed to certain microbial pathogens over time (discrete versus continuous
dose and exposure). These models may include susceptibility and immunity variation and
life stages.
• Developing criteria for the use of animal model results for derivation of dose-response
models. Improved methods to extrapolate animal dose-response information to human
dose-response models should be pursued, as well as better ways to address the uncertainty
involved in such extrapolations (such as differences in health effects between humans and
animals).
• Exploring the issue of whether threshold or nonthreshold dose-response models are
most appropriate for various pathogen-host combinations.
• Developing biologically-based mechanistic models (such models are being developed but
are not yet available).
• Developing methods to investigate dose-response relationships for
immunocompromised and other more sensitive populations. This can include outbreak
related studies, epidemiological studies, or studies with immunocompromised animal
models.
A-l
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
A.3. Risk Estimation and Risk Characterization
• Developing methods to address cumulative risks from exposure to multiple pathogens
and to pathogens and certain chemicals.
• Developing additional methods for considering possible lifetime, cumulative risk from
exposure to one or more pathogens.
• Developing a framework for conducting community-based cumulative risk assessment
for microbial hazards.
• Developing methods for estimating risks of chronic or secondary sequelae.
• Developing methods for comparing risks among different pathogens and different
exposures (comparative risk). Common metrics that provide a basis for such comparisons
(e.g., to compare Vibrio vulnificus and E. coli 0157:H7) should be explored.
o The use of disability-adjusted life years (DALYs), which measure and compare the
effects of disease burden on a population, use morbidity (years lived with a
disability), mortality (years of lost life), and standardized life expectancy to
calculate a DALY value for a given disease for a defined population, is one method
for comparing risks (Gold et al., 2002).
o Quality-adjusted life years (QALYs) are a method for assigning a numerical value
for quality of life and translating that numerical value to a monetary measure
(WHO, 2001).
• Conducting research on the appropriate use of adjustment factors for microbial risk
assessment. The circumstances for using such factors and the criteria to determine the
magnitude of the factors and where they can be applied should be considered.
• Developing additional model validation methods to compare the results of the risk
assessment with "reality." If few data exist for this comparison, after the risk assessment
is conducted endpoints should be monitored so the model can be validated in the future.
• Further developing qualitative assessment methods, because quantitative data are not
always available.
• Improving the application of risk assessment as a predictive tool in developing
prevention strategies.
• Further developing methods and models for incorporating information on secondary
transmission.
• Further developing methods and models for incorporating information on immune
status. For some dose-response datasets where infection is the endpoint, this may be
difficult because immunity affects illness rather than infection (e.g., as observed for
Giardia).
A.4. General research needs to improve MRA include the following:
• more information on mechanisms of infection and virulence factors;
• data on variation among different hosts and pathogens;
• data on the effect of environment on pathogen growth, survival, and death;
• data from longer time frames in order to account for longer-term weather cycles (e.g., el
Nino);
• data on changing land use patterns advancement
A-2
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
improved sampling, detection, quantification methods, and viability/infectivity assays; and
continued development of a thesaurus or lexicon of risk assessment terms to facilitate the
evolution of terminology.
A-3
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
A-4
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Appendix B. Other Risk Frameworks that are Consistent with the
MRA Tools Framework
• The WHO State of the Art Report: Health Risks in Aquifer Recharge Using Reclaimed
Water has a chapter on methods for health risk assessment that includes a process diagram
for risk assessment (Thoeye et al., 2003).
• Quantitative microbial risk assessment (QMRA) approaches, along with summaries of six
research papers related to health risks from infectious microorganisms transmitted via
urban water and wastewater systems, are presented in this dissertation (Westrell, 2004).
Discussions of susceptibility and immunity, sensitive subpopulations, secondary
transmission, dynamic modeling, and health indices are also included.
• Rose and Grimes (2001) present a flow diagram for conducting a screening level risk
assessment (preliminary risk assessment) that advances users through nine questions to ask
during the planning of a screening level risk assessment. Molecular tools for characterizing
and identifying microorganisms are also reviewed.
• Medema and Smeets (2004) discuss the interaction between QMRA and the risk
management aspects of the WHO Water Safety Plan.
• The Canadian report, Microbial Risk Assessment as a Foundation for Informed Decision-
Making (Fazil et al., 2005), presents MRA in its larger context by discussing enabling
legislation, policy scrutiny, and international trade agreements and standards. The "current
status" as well as "the way ahead" is presented for prioritization and coordination; methods
and tool development; guidance documents (qualitative, technical, and methodology);
training for risk assessors; and risk-based decision-making, peer review, and integration of
risk communication.
• The aim of the MICRORISK Project (www.microrisk.com) is to develop a MRA process
that contributes to the decision-making process for risk management of drinking water. The
elements of the framework are the Quantitative Microbial Risk Assessment (QMRA) and
Hazard Analysis and Critical Control Points (HACCP). Funding entities include the
following: collaborative water utilities in the Netherlands (BTO), U.K. Water Industry
Research, and the Australian Commonwealth Government Department of Education
Science and Technology.
• The Center for Advancing Microbial Risk Assessment (CAMRA) is the Homeland
Security Center of Excellence jointly established with the U.S. EPA to develop scientific
knowledge on the fate and risk of potential bioterrorist and other high priority infectious
agents, (http://www.camra.msu.edu/).
• FDA hosts iRisk which is a web-based system designed to analyze data concerning
microbial and chemical hazards in food and return an estimate of the resulting health
burden on a population level.( https://irisk.foodrisk.org/)
The Interagency MRA Guideline was released after major development of this MRA Tools
document. These two documents are compatible. In addition, the external review draft of this MRA
Tools document was used in the development of the Interagency MRA Guideline (U.S.
EPA/USD A, 2012).
B-l
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
B-2
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Appendix C. Dose-Response Models
C.l. Alternate Dose-Response Models
C.l.l. Empirical Models
To date, the majority of studies in dose-response modeling and MRA for waterborne pathogens
have employed the exponential and beta-Poisson dose-response models. These models are
mechanistic (based on models of biologically-plausible processes), relatively simple, and have
provided good fits to many of the data sets for which they have been applied. Other models have
also been proposed and used as components of MRAs, particularly in the assessment of risks
associated with food. These alternative models are empirical (i.e., not derived based on
consideration of biological processes) and as such, their validity outside the data range for which
their parameters are estimated is unknown, and extrapolation with these models is not
recommended (Buchanan et al., 2000). In dose-response model selection, preference is given to
biologically plausible, mechanistic models such as the exponential and beta-Poisson over
empirical models. Other researchers (e.g., Coleman and Marks, 1998) have suggested that the
exponential and beta-Poisson models are not substantially different from empirical models.
However, this document does not adopt that position because exponential and beta-Poisson models
may be derived from basic considerations of the infection process, because the models may be
adapted to include other processes within the infection process, and because of the demonstrated
success in fitting the exponential and beta-Poisson models to available data.
Buchanan et al. (2000), Moon et al. (2004), Holcomb et al. (1999), and Haas et al. (1999) compared
the exponential and beta-Poisson models to various empirical dose-response models proposed for
use in dose-response modeling for foodborne pathogens. The empirical models assessed in those
studies are summarized in Table C-l. Presentation of these models in this document is intended to
provide a thorough description of models that have been used. In their comparison of models,
Moon et al. (2004) found that three-parameter models did not yield significant improvements in
fit over two-parameter models and that among two-parameter models, predictions in the low-dose
range were markedly different between models. Buchanan et al. (2000) suggest the development
of mechanistic models with consideration of factors important in the infection process as a route
to more accurate dose-response models that may be extrapolated outside the range of available
data. Holcomb et al. (1999) noted that among the empirical and mechanistic dose-response models
compared, only the three-parameter Weibull-gamma model provided goodness of fit for data sets
of dose-response data for four different pathogens and that different models predicted very
different low-dose-response. These observations lead the authors to suggest that continued dose-
response model development and evaluation is necessary.
C-l
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Table C-l. Empirical Dose-Response Models
Model
Equation
Parameters
Weibull-Gamma
P(d) = l-(l + db /
Three parameter model: b, a, p.
Weibull
P(d) = l - exp(-adh)
Two parameter model: a, b
Gompertz1
P(d)= l - exp [- exp (a + b f(d))\
Two parameter model: a, b f(d)
denotes a transformation (e.g., log)
Log-normal2
p(d)= vfc tSa)!P exP {~¥2)dt
Two parameter model: a, p
Log-logistic
P(d) = l/{l + exp [- (in d - a)l (5 ]}
Two parameter model: a, p
Exponential
-Gamma
P(d}=\ — exp(— yd)/{\ + db /(3j
Three parameter model: a, p, y
Weibull-exponential
P{d) = l - exp (- a dr )/(l + dr / /?)
Three parameter model: a, p, y
Shifted Weibull
f1-exp{r[(^-Q0/^!'} d^a
| 0 0^d^HS [C-l]
C-2
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
where HS is the Hill slope, given as:
logb/(lOO-g)]
log [ittDJIltDj
q is the chance of becoming ill (selected as 1%), IIID50 is the median infectious dose at which 50%
of the exposed population becomes ill, IllDq is the dose at which q% of the exposed population is
expected to become ill, and log refers to logio. The functional form of the sigmoidal relation is
such that the probability of illness rises sharply from a very small value at the dose expected to
produce illness in q% of the exposed population (1% in the model of Brynestead et al., 2008).
Based on published data, Brynestead et al. (2008) estimated that IIID50 and IllD\ were uniformly
distribution in the ranges 500 to 800 and 2000 to 6000 organisms, respectively. Risk estimates for
Campylobacter infection related to food preparation appear high for both the Hill slope model and
an alternative dose-response model, with the Hill slope model providing a lower, but unrealistic
estimate of the number of illnesses. This finding could indicate that current dose-response models
over-predict the incidence of illness or that exposure models over predict the incidence and
ingestion of Campylobacter in the food chain in Germany. Alternatively, the findings could be an
artifact of the high uncertainty inherent to the epidemiological data to which QMRA model results
were compared.
In their review of foodborne Campylobacter illness QMRAs, Nauta et al. (2009) compared
alternative Campylobacter dose-response models (illness endpoint) and observed that the
sigmoidal model predicts much lower illness probability at low dose than alternative published
models. This observation is consistent with the chosen form of the model. Based on the
assumptions of Brynestead et al. (2008), the sigmoidal illness incidence model reported by Nauta
et al. (2009) used a 1% illness incidence as the lower end of the illness dose-response relation,
arriving at the following dose-response model:
[C-3]
where IIID50 is the dose at which 50% of the exposed population becomes ill, d is the ingested
dose and a is given by
[ln(0.99)l
ln(0.0l)J rp Al
[c ]
Lln(Wq) .
The comparison of QMRAs of Campylobacter by Nauta et al. (2009) allowed the authors to
conclude that the Campylobacter dose-response model remains unknown, particularly given
potential variations in the ability of different strains of Campylobacter to initiate infection or
illness, influences that food matrix or other environmental factors may exert on the incidence of
illness, and the difference in response for different subpopulations.
C-3
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Selection of dose-response models requires comparison of fits of the models to data and
comparison of fits of more highly-parameterized models with those of models with fewer
parameters. When maximum likelihood estimation is used for determining the model parameters,
fits are compared on the basis of the deviances at the parameter values providing the best fit of the
model to the data. In general, models with more parameters are selected over models with fewer
parameters only when the improvement in fit of the model with more parameters over that of the
model with fewer parameters is statistically significant.
Threshold models (i.e., models that assume more than one organism is required to initiate
infection) can be derived under slightly different assumptions than those used to develop the
exponential and beta-Poisson dose-response models. Assuming pathogens in an ingested dose are
drawn from a homogeneous distribution (Poisson distribution) and each pathogen has an equal,
independent probability that it can initiate an infectious focus, the probability of infection by kmm
organisms is (Haas et al., 1999):
where r is a parameter of the distribution and T denotes the gamma cumulative probability
distribution function. This simple threshold model is a two parameter model whose parameters
may be determined via standard statistical techniques such as maximum likelihood estimation
(MLE). Deterministic models have also been used in evaluation of the potential that components
of the infection process can produce complete extinction of pathogens populations before a
systemic infection (e.g., establishment of a steady pathogen population in vivo) occurs (e.g., Blaser
and Kirschner, 1999; Coleman and Marks, 2000). These studies are described in the following
section.
C.1.3. Mechanistic and Physiologically-Based Models of Infection
Models of the infection process (i.e., mechanistic dose-response models) may be developed with
varying degrees of resolution. These models differ in the components of the infection process that
are explicitly modeled and whether they are deterministic or stochastic. Early attempts at
developing mechanistic dose-response models focused on stochastic pathogen birth and death
processes or on division of the infection process into stages that might be modeled separately.
Under the assumption that pathogens divide and are removed (via innate or active immune system
processes or other means) at constant rates, // and X, Bailey (1964) developed expressions for the
probability that an in vivo population of size N is realized at time I:
C.1.2. Threshold Models
/•(infection d) = T(kmm, rd)
[C-5]
mm(
,1 H (l-,!-/>')
[C-6]
and for the probability that the pathogen population reaches extinction at time t:
p0(t) = Ad
[C-7]
C-4
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
or as t ->oo:
P 0
t —> GO
In Equation C-6,
[C-8]
, ."I' ¦ I rC9,
A-jue
A [l _<>-*>]
A-jue'
5 = -J—rar [c-10]
Morgan and Watts (1980) used Equation C-6 to derive an expression for the probability that a
single pathogen (d= 1) achieves a threshold population at time l:
p(n >N,t) = {\- A)BN1 [C-ll]
Alternately, the probability that the incubation period, T, is less than a time, t, is:
p(T
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
Brookmeyer et al. (2005) developed a time-dependent dose-response model referred to elsewhere
as a competing risks model (Gutting et al., 2008). One of the authors' stated motivations in
developing a mechanistic model for Bacillus anthracis infection was utilization of available data
on infection by B. anthracis (spore germination rates, clearance rates, growth rates) in the absence
of detailed human dose-response data developed in experimental studies. Assuming a constant risk
of spore germination per unit time, ct>, and a constant risk per unit time of clearance from the lung,
k, and assuming that spore germination implies systemic infection, Brookmeyer et al. (2005)
showed that the cumulative attack probability for inhalation anthrax may be estimated as:
Inspection of Equation C-14 shows that the Brookmeyer competing-risks model yields the
exponential dose-response model in the limit t -> qo.
Blaser and Kirschner (1999) developed a deterministic model for in vivo pathogen growth,
including immune system response. In that study, stocks and flows of five quantities—mucus-
living Helicobacter pylori, H. pylori attached to epithelial cells, density of bacterial nutrients
released via inflammation, density of effector molecules, and host response—were included in a
system of ordinary differential equations describing the dynamics of these quantities. In mucus,
the conservation equation for H. pylori accounted for growth (first-order with respect to nutrient
availability), loss due to mucus shedding and migration, and gain due to emigration. The
conservation equation for H. pylori on epithelial cells included a growth term, a loss term related
to sloughing, and terms accounting for immigration and emigration. The authors used their model
to explore the importance of the parameters in their model of infection, determining that the
parameter that describes the ability of the immune system to respond was the most important
determinant of whether there would be extinction (all pathogens are removed from the system) or
whether sustained growth occurs and that the bacteria growth parameter had a limited effect on the
ability of pathogens to initiate infection but was the most important factor in determining the time
required for pathogens to reach a steady population in vivo. In subsequent modeling work, Blaser
and Kirschner (2007) used a deterministic model to explore infections with slow progression or
latent periods, during which there is equilibrium between host response and pathogen population
dynamics. Taken together, the two studies by Blaser and Kirschner (1999, 2007) demonstrate the
utility of deterministic models in exploring complex infection processes.
Coleman and Marks (2000) developed both stochastic and deterministic models of non-typhoid
salmonellosis and used the models to identify factors that influence the shape of the dose-response
curve in the low-dose region. In that study the important events occurring in the course of
Salmonella infection were posited to be survival of ingested bacteria to the target, colonization,
engulfment, intracellular survival, migration and multiplication, damage, and AGI. The authors
suggested stochastic models for each of these processes and presented an alternative formulation
based on a predator-prey framework. As pointed out by Coleman and Marks (2000) for infection
by non-typhoid Salmonella and also by Levin and Antia (2001) for infections in general, there may
be physiological and biological process that do not conform to the assumptions underlying the
beta-Poisson or exponential dose-response model, including clumping of pathogens in the ingested
F(t) = 1-exp -^-(l-e-(a,+K)t)
(d + K
[C-14]
C-6
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
dose, quorum sensing, and the possibility that organisms do not exhibit independent action. In the
context of MRA model results and results of feeding studies of healthy adult human volunteers,
the authors made a case for sub-linearity of response at low dose. The authors note that the potential
for sub-linear low-dose-response is likely differ between pathogen-host combinations and that
additional data such as in vitro studies may provide information for parameter selection for
mechanistic infection models. Development and validation of additional mechanistic models for
infection provides an avenue for evaluating low-dose-response.
The inherent variability of host-pathogen processes suggests use of stochastic models for
describing in vivo processes leading to infection. Allen and Allen (2003) describe Markov-chain
and stochastic differential equation models for estimating the pathogen burden in vivo as a function
of time. Their model of the infection process is relatively simplistic, comprised only of birth and
death processes in which birth and death rates may vary with time or pathogen density, but their
framework is amenable to inclusion of additional components (e.g., immune system components,
pathogens in different states). Based on evaluation of different models for a relatively simple case,
the authors concluded that combinations of deterministic and stochastic models offer the greatest
opportunity for including relevant features of the infection process in a computationally tractable
framework.
Recent dose-response modeling efforts have included development of highly-detailed,
physiologically-based models of the infection process. In their recent assessment of anthrax dose-
response models, Gutting et al. (2008) outlined the components of a hypothetical physiologically-
based biokinetic model of infection and response to aerosols of Bacillus anthracis. In the model,
the fate and transport of B. anthracis spores and vegetative cells is tracked in regions of the
respiratory system, in macrophages, in the blood and in lymph nodes. As done by Brookmeyer et
al. (2005) in their development of a competing risks model for inhalation anthrax, Gutting et al.
(2008) estimate model parameters for use in their biokinetic model using physiological and
microbiological data not collected in quantal dose-response studies or epidemiological
investigation. However, details of the techniques used for parameter estimation or of the model
were not provided in the study by Gutting et al. (2008).
C.2. Use of Bayesian Methods in Microbial Risk Assessment
Bayesian methods are being increasingly used by several researchers in microbial risk assessment
to estimate dose-response model parameters. In general, a dose-response function gives the
probability of illness or infection as a function of the dose and of several unknown parameters.
Experimental data are collected from subjects accidentally (in an outbreak) or deliberately (in a
controlled experiment with volunteer human subjects or with animal subjects) exposed to a
microbial dose that can be measured or estimated. The numbers of subjects that become infected
or ill for each dose level are observed, leading to a binomial likelihood. That is, the probability of
n "successes" out of N trials of dose level d, where "success" means illness or infection and the
success probability is given by the dose-response function. The "traditional" frequentist statistical
approach uses the binomial likelihood only, and chooses parameter values to maximize the
likelihood.
C-7
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
In general, if Ni subjects are exposed to a mean dose A, and //, of them developed infection, then
the likelihood for the full population (all dose groups) is given by the following:
The dose-response function is the function /'(Infected | D), which will depend upon the mean dose
D as well as the unknown parameters.
To estimate the uncertainty of the estimated dose-response parameters and dose-response function,
95% confidence intervals can be calculated using standard asymptotic theory, valid when the
sample sizes ("n") are large. The asymptotic theory uses the likelihood function (Equation C-15)
to derive an estimated standard error for each parameter, and the 95% confidence interval can then
be estimated as the maximum likelihood estimate plus or minus 1.96 standard errors. The 1.96th is
the 97.5th percentile of a standard normal distribution, which applies because for large samples the
estimated parameter approximately has a normal distribution.
Alternatively, and preferably for the small sample sizes usually available in microbial risk
assessment, a Monte Carlo bootstrap resampling method can be used to estimate the uncertainty
by randomly sampling with replacement from the original data and fitting the model to each of the
resampled data sets. Bayesian methods are preferable to bootstrapping with smaller sample sizes.
Bayesian methods exploit available subjective and related information in addition to the numeric
data from the experiment or outbreak. Ideally, the investigator expresses their initial assessment
of the unknown parameter distribution, prior to examining the data, by defining a prior probability
distribution for the parameters. The prior probability distribution is defined based on subjective
information and professional judgment.27 Using Bayes' rule, the posterior probability distribution
for the parameters given the data can be calculated. From Bayes' rule, the posterior distribution
equals the prior distribution for the parameters multiplied by the likelihood for the data (given the
parameters) and then divided by a normalizing constant. The normalizing constant is the integral
of the product of the prior and likelihood over all possible parameter
values.28 In a Bayesian analysis, uncertainty intervals for the parameters and the dose-response
function can be calculated from the posterior distribution as "credible intervals"; a 95% credible
interval has a 95% probability of including the parameter value, given the data.
27 Some Bayesian researchers use a more objective approach called the empirical Bayes method that is based on an
hierarchical model such that the likelihood depends upon parameters that have distributions depending upon other
parameters, called hyperparameters. A frequentist approach, such as maximum likelihood, is used to estimate the
hyperparameters and thus estimates the prior distribution without the use of subjective information. At this time, the
authors are not aware of any applications of empirical Bayes methods to microbial risk assessment.
28 Suppose 0 is the vector of unknown parameters, and has a prior distribution with probability density function f(0).
Suppose the data X has a likelihood given by g(X | 0), for example, Equation C-15. Then the posterior distribution
will have a probability density given by f(0) g(X | 0) / k(X), where k(X) is the normalizing constant. This is Bayes'
rule. The normalizing constant is the integral
I f(0) g(X | 0) d0, integrated over all possible values of 0.
The constant k(X) does not depend upon the parameters although it will depend upon the data X.
#doses
x [/"(infection | Di)]" x [l - /(infection | I)l)]
n
[C-15]
i=1
C-8
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
The choice of a suitable prior distribution is crucial and can be controversial. Recent published
MRAs have usually had very little subjective information to rely on for choosing a prior
distribution and the investigators have chosen a "non-informative" prior distribution to represent
the lack of prior information. The researchers have usually published their choice of non-
informative prior, but have not usually provided a rationale for their choice over other possible
non-informative priors.29 For example, Teunis and Havelaar (2000) used a beta-Poisson model,
described below, and chose the prior distribution for their parameters a and P such that their
logarithms (base 10) were assumed to have a wide uniform distribution from -12 to +6 and the
parameters were assumed independent. Englehardt (2004), using the same beta-Poisson model,
instead chose a joint uniform prior distribution for the parameters a and p. Teunis et al. (2004) also
used a beta-Poisson model, but used another non-informative prior, such that aJ(a+ P) is uniform
from 0 to 1 and logio(a+ P) is normally distributed with mean 0 and standard deviation 10. If the
non-informative prior distribution is wide then the posterior probability distribution should not be
sensitive to the choice of non-informative prior, which justifies the name "non-informative prior."
However, researchers have used different non-informative priors for the same model, which
suggests that the choice of the so-called non-informative prior can influence the results.
In the past, Bayesian researchers were much more limited in their choice of prior distributions
because they needed to choose a distribution to make the calculations tractable (a "conjugate"
prior), particularly the calculation of the normalizing constant. More recently, with MCMC
methods and fast computing methods, the calculations can be easily executed for a much wider
variety of prior distributions using Monte Carlo simulation methods.
The MCMC method describes a group of methods used to simulate values from a probability
distribution for which direct analytical calculations are difficult, intractable, or inconvenient. Gilks
et al. (1996) provide a good description of these methods. Well-validated software packages are
available to perform these calculations, including WinBUGS and Mathematica. For Bayesian
MCMC analyses, the simulated probability distribution is the joint posterior distribution of the
parameters given the data. Thus, at each step of the Markov chain, a vector of parameter values is
simulated, rather than a single parameter value. Furthermore, it is unnecessary to know the
normalizing constant for the posterior distribution, which is often the most difficult part of the
calculation. All that is needed are some constant multiples of the prior distribution and the
likelihood. The normalizing constants needed to make the prior and likelihood integrate to one are
not needed. A version of the Metropolis-Hastings algorithm (Hastings, 1970; Gilks et al., 1996) is
used at each step to simulate from the posterior distribution without knowing the normalizing
constant.30 Instead of being statistically independent, the consecutive values form a Markov chain,
29 Teunis et al. (2004, 2005, 2008a,b) transformed their parameters using logarithms and logit functions to avoid
high corrections between the parameters and thus improve the estimation. However, this does not really explain their
choice of non-informative prior for the transformed variables.
30 Suppose that the product of the prior and likelihood is equal to K x f(0), where 0 is the vector of all the unknown
parameters and K is an unknown normalizing constant that will depend upon the data values; that is, for MRA, the
numbers of illnesses or infections observed. To obtain the posterior distribution, K can, in principle, be calculated as
the reciprocal of the integral of f over the range of possible parameter values, but this calculation is often very
difficult analytically. Let q(0, cp) be any chosen proposal distribution, which is a probability density for the next
parameter vector cp that may depend upon the previous parameter vector 0. If the parameter vector at the previous
step is 0, one must first randomly sample a parameter vector from q(0, cp) to obtain a candidate vector cp*. With
probability a, one can accept the candidate vector, so that the vector at the next step of the Markov chain is cp*. With
C-9
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
so that the statistical distribution for one value depends upon the previous value. Using the MCMC
method, the Markov chain has a limiting, stationary distribution, so that after a sufficiently long
"burn-in" period the values have the desired probability distribution.31
The articles reviewed for this discussion do not specify the details of the Metropolis-Hastings
algorithms used. Many Bayesian analysts use the Gibbs sampler, which is a special version of the
Metropolis-Hastings algorithm that always has acceptance probability 1, so that a new parameter
vector is selected at each step. Instead of jointly updating all the parameters in a single step, the
Gibbs sampler simulates each of the parameters in turn.32 An algorithm such as Adaptive Rejection
Sampling (Gilks and Wild, 1992) is used to generate samples from the distribution of each
parameter without needing to calculate the normalizing constant.33
Bayesian modeling has been used by MRA researchers in various ways. Several authors have used
both Bayesian and frequentist (likelihood-based) methods (Teunis and Havelaar, 2000; Messner
et al., 2001). Often the frequentist approach is used to provide maximum likelihood estimates of
the dose-response function and the Bayesian approach is used to calculate uncertainty intervals
(e.g., 80 or 95% credible intervals for the parameters or the dose-response). The frequentist
likelihood ratio test is used to compare different dose-response models. Several approaches use
the mode of the Bayesian posterior distribution to select the dose-response function (Teunis et al.,
2004, 2005, 2008a,b). The posterior mode is given by the parameters that maximize the posterior
probability 1- a, one rejects the candidate vector, so that the vector at the next step of the Markov chain is, again, 0.
The probability ais calculated as min(f(cp*) q(cp*, 9)/{f(9) q(0, cp*)}, 1). Because f appears in both the numerator and
denominator, the unknown K cancels out and is not needed. A good choice of the proposal distribution will have
high acceptance rates and fast convergence to the stationary distribution.
31A burn-in period of about 5000 steps is usually sufficiently long that the stationary distribution has been reached;
various convergence tests can be used to assess convergence. Values generated during the burn-in period are
discarded, and one usually selects every klh value after the burn-in period for some suitably large k (e.g., 10, 20, 100)
so that the remaining "thinned" sequence of values are approximately independent. Thus, the thinned values after
the burn-in period can be treated as if they were a random sample from the given probability distribution.
32 Suppose that there are n unknown parameters in the posterior distribution. Instead of generating a new
multivariate vector of n parameters from a joint distribution, the Gibbs sampler generates each parameter in turn
from the univariate "full conditional" distribution of that parameter given the values of all the other parameters and
the data. Thus, each Markov chain step becomes a sequence of sub-steps where the n parameters are scanned in turn
and the mlh parameter value is randomly selected from the conditional distribution of the m"1 parameter given the
data and the most current values of the remaining n-1 parameters (i.e., the values of the first m-1 parameters from
the current scanning steps and the values of the last n-m parameters from the previous Markov chain vector). An
algorithm such as Adaptive Rejection Sampling (Gilks and Wild, 1996) is used to generate samples from each full
conditional distribution without needing to calculate the normalizing constant.
33 Suppose that the product of the prior and likelihood is equal to K(0.m) x g(0m, 0.m), where 0m is the m"1 unknown
parameter, 0.m is the vector of the other n-1 unknown parameters, and K(0.m) is an unknown normalizing constant
that will depend upon the data values and the values of the remaining n-1 parameters. To obtain the full conditional
distribution, K(0.m) can, in principle, be calculated as the reciprocal of the integral of g over the range of possible
values of 0m, treating the other parameters as constants, but this calculation is often very difficult analytically.
Adaptive rejection sampling (Gilks and Wild, 1992) randomly generates values of 0m from g, without knowing the
normalizing constant. The method requires that the function g is log-concave in 0m, which holds for many
distributions if the parameters are appropriately defined. The method may need to generate and reject several
random values until the final value is accepted, but at each rejection, more exact bounds for g are calculated so that
the probability of future rejections rapidly decreases. If the full conditional distribution is not log-concave, then the
Metropolis-Hastings algorithm can instead be used to generate values from the full conditional distribution without
knowing the normalizing constant.
C-10
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
probability, defined as the product of the prior and the likelihood; it is again not necessary to
calculate the normalizing constant.
Englehardt and Swartout have published several papers (Englehardt, 2004; Englehardt and
Swartout, 2004, 2006, 2008) advocating the use of the predictive Bayesian approach, which is the
unconditional dose-response probability, calculated as the integral of the posterior distribution
multiplied by the dose-response function, integrated over the parameter space. This can be thought
of as the dose-response function averaged over the uncertainty distribution. The predictive
Bayesian method has the advantage of producing an estimated dose-response function that is more
protective of public health than the maximum likelihood estimate, because at low doses the
estimated risk is generally higher. The method also has the advantage of avoiding the need to
specify a frequentist confidence level or a standard Bayesian prediction interval probability level,
which avoids potential inconsistencies when comparing risks from different health stressors; the
more risky stressor can depend upon the probability level chosen. On the other hand, upper bounds
of confidence or prediction intervals can be thought of as estimating the risk under a "worse-case"
scenario, and regulators may prefer a worse-case scenario approach to the predictive Bayesian
approach that represents the average scenario (averaging estimates over the parameter
uncertainty).
C.2.1. Comparison of Bayesian and Frequentist Methods
Before Bayesian methods were applied to MRA, risk assessors were generally limited to simpler
model formulations and approximate uncertainty estimates. Risk assessors also could not take
advantage of any available subjective information on the values of the unknown parameters. An
advantage of the Bayesian approach over the frequentist approach is the ability to incorporate prior
information, although for the MRAs in the current literature this is not very important because the
prior information is too limited and so non-informative priors have been used. A more important
advantage is that the uncertainty intervals from a Bayesian analysis are easier to interpret and are
usually not interpreted incorrectly—a Bayesian 95% credible interval for the dose-response is
interpreted as having a 95% probability of including the true probability of illness (or infection)
given the available data. The risk (probability of illness or infection) is treated as being random. A
frequentist 95% confidence interval is properly interpreted as having a 95% probability of
including the true probability of illness (or infection) in an identical future experiment, so that 95%
of a large number of identical future experiments will give confidence intervals that include the
true risk. The risk is treated as being an unknown constant. Lay persons will very often incorrectly
interpret the confidence interval as if it had the same meaning as the Bayesian credible interval.
Bayesian dose-response uncertainty calculations using MCMC also have the advantages of being
easier and more exact than the frequentist confidence intervals. Because the dose-response
function is a complicated function of multiple parameters, the confidence intervals are hard to
calculate or approximate analytically. The bootstrap or similar Monte Carlo resampling methods
can avoid these difficult analytical calculations but this often requires more computation than
MCMC. Furthermore, the large sample theory estimates of the confidence intervals are poor
approximations for the small samples typically found in MRA. While bootstrap estimates are more
reliable for small samples, they are also approximations to the true uncertainty distributions, even
if the number of bootstrap simulations is tending to the infinite limit. The MCMC uncertainty
C-ll
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
estimates are exactly correct for the posterior distribution assuming that the burn-in period is
sufficiently long that the chain can be considered stationary (ignoring the imperfect nature of
computer random number generation).
A further major advantage of the Bayesian method for MRA is the ability to use a hierarchical
Bayesian model to model cases where the host or pathogen response parameters vary over the
population of humans or organisms (e.g., see Messner et al., 2001, discussed below). This type of
meta-analysis is easier to apply in a Bayesian framework.
The major disadvantage of the Bayesian approach is the requirement for developing a prior
distribution that, in principle, is subjective and thus depends on the information available to the
investigator. Different investigators can choose different priors for the same model formulation,
even if the prior are "non-informative." The subjective nature of a prior distribution can be
disturbing. On the other hand, Bayesian statisticians often point out that the investigator's choice
of dose-response function or other mathematical model is also a subjective choice.
C.2.2. Applications of Bayesian Methods to Microbial Risk Assessment
In one of the earliest Bayesian analyses of microbial risks, Teunis and Havelaar (2000) modeled
rotavirus, Campylobacter, and Vibrio cholerae dose-response data using a series of models. In the
Poisson model, an individual is exposed to a number of organisms (e.g., colony forming units
[CFU]) that is assumed to have a Poisson distribution with a mean dose D equal to the volume
ingested multiplied by the average number of CFU per unit volume. Each single organism
independently has the same hit probability (r) of infecting the subj ect. It follows that the probability
of infection at dose D is exponential with parameter r < 1:
In the beta-Poisson model, r is assumed to vary among hosts or organisms with a beta distribution
with parameters a > 0 and P > 0. This gives the dose-response function as follows:
The function 1F1 is the Kummer confluent hypergeometric function.
The gamma-Poisson model is an approximation to the beta-Poisson model of the following form:
Prob (infected | D, r) = 1 - e"rD.
[C-16]
Prob (infected | D, a, P) = 1 - iFi(a, a + P, -D).
[C-17]
Prob (infected | D, a, P) = 1 - (1 + D / P)"a .
[C-18]
C-12
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
This model is also obtained if either r has a gamma distribution that includes all values r > 0 (even
though the hit probability r cannot exceed 1), or, more realistically, if the density has a gamma
distribution. The model in Equation C-18 was originally called the beta-Poisson model, because it
was derived as an approximation to the exact beta-Poisson model in Equation C-17. The same
three equations can be used to model the probability of illness, particularly for outbreaks or studies,
where the numbers of infection cases are not reported.
Teunis and Havelaar (2000) fitted all three models and estimated the parameters r, a, and P using
maximum likelihood (i.e., choosing values to maximize the probability of the data given the
parameters, which is a product of binomial probabilities). The likelihood is given by Equation C-l
above. Using the maximum likelihood estimates of the parameters in the dose-response function
(Equations C-l, C-2, or C-3) provides the maximum likelihood estimate of the dose-response
function. They also estimated approximate 95% confidence intervals and regions for the
parameters using the likelihood.
The authors also used a bootstrap resampling method to estimate the uncertainty of the dose-
response function by randomly sampling with replacement from the original data and fitting the
dose-response model to each of the resampled data sets. Thus, each bootstrap sample gives a
different dose-response function.
Teunis and Havelaar (2000) used a Bayesian MCMC approach primarily to more easily compute
the uncertainty estimates and to compare their maximum likelihood estimates with Bayesian
estimates. Their prior distribution for the parameters a and P assumes they are independent and
that their logarithms (base 10) have a uniform distribution from -12 to +6. The MCMC method
was used to generate pairs of parameter values a and P from the posterior distribution. They found
that the likelihood-based confidence regions for the parameters probability matched well to the
sampled Bayesian posterior distribution. For each parameter pair, the dose-response function was
calculated. For each dose D, a 95% credible interval for the probability of being infected is given
by the 2.5th to 97.5th percentiles of the set of dose-response functions evaluated at dose D.
Messner et al. (2001) used Bayesian methods to analyze the results of three human volunteer
studies, each using different isolates of Cryptosporidium—IOWA, TAMU, and UCP. For each
individual study, they fitted an exponential model (Equation C-2) by maximum likelihood and
then computed the maximum likelihood estimate of the dose-response function. They compared
their results with a Bayesian analysis based on the assumption that ln(r) has a uniform distribution
over the entire real line.34 The means and medians of the Bayes predictive distributions were very
similar to the maximum likelihood estimates. Given the assumed distribution for log(r), its
posterior density is proportional to the likelihood function.
To combine results from the three studies in a meta-analysis, Messner et al. (2001) used a
hierarchical Bayes model that had several groups of parameters. At the first level, the
hyperparameters are parameters with a prior distribution that does not depend on any other
parameter. At the second level, some parameters are assigned distributions that depend upon the
34 Such a prior is not a proper probability distribution because it cannot integrate to 1, but in many cases an
improper prior can be used to calculate a valid posterior distribution. This improper uniform prior can be regarded as
being the limit of a uniform distribution for log(r) over the range -M,M as M tends to infinity.
C-13
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
values of the hyperparameters. At the third level there are parameters that have distributions that
depend upon the first and/or second level parameters. The hierarchy can have multiple levels,
although most applications to MRA have at most two levels.
Messner et al. (2001) defined two hyperparameters [i and o. Their prior distributions were not
listed in their paper. The parameters r for each study were assumed to be independently drawn
from a normal distribution with mean |i and standard deviation o. This normal distribution
represents variability of r between isolates, i.e., the probability of infection from a dose of a single
organism depends upon the isolate. Thus, the model has a total of five parameters. The MCMC
method using the Gibbs sampler was used to generate samples of parameter vectors from the
posterior distribution given the data from all three studies. Eighty percent credible intervals (10th
to 90th percentile of the posterior distribution) were thus calculated for the parameters and for the
dose-response function at a dose of one oocyst.
Teunis et al. (2004) used Bayesian modeling to analyze data from an outbreak of E. coli 0157:H7.
They modeled the dose-response functions using the beta-Poisson model shown in Equation C-17.
A non-informative prior was selected such that u = a/(a+ P) is uniform from 0 to 1 and v = logio(a+
P) is normally distributed with mean zero and standard deviation 10. The transformed parameter u
is the mean of the beta distribution for r, and v is inversely related to the variance of the beta
distribution. The parameters u and v are assumed independent. The transformation improves the
parameter estimation since if there is only a single dose value, a and P are highly correlated. The
parameter values corresponding to the mode of the posterior distribution were calculated directly
by numerically maximizing the posterior probability, which is the same as maximizing the product
of the prior distribution and the likelihood. The posterior probability equals the prior multiplied by
the likelihood and divided by the normalizing constant, which does not depend upon a and p. Using
the posterior mode parameter values in Equation C-17 gives the posterior mode dose-response
equation. The uncertainty of this dose-response function was characterized using MCMC sampling
of parameter vectors. The dose-response function was calculated for each parameter vector and
the percentiles of the response probability for each dose were plotted. Frequentist likelihood ratio
tests were used to compare different dose-response models.
Teunis et al. (2005) analyzed Campylobacter jejuni dose-response using Bayesian methods. Data
from both a human volunteer study and two outbreaks caused by drinking raw milk (children and
teachers visiting farms; one in Holland and one in the UK) were combined in this analysis. The
model incorporated both the probability of infection and the conditional probability of illness given
infection for children. First, for the outbreak a certain probability of illness (pO) was assumed for
those who were unexposed to the raw milk but might have become ill due to an alternative route
of transmission. Second, a beta-Poisson model was used to model the probability of infection given
a mean dose (D). Third, a model for the conditional probability of illness, given that the individual
is infected and had mean dose D, was developed as follows:
Prob(ill | infected, D , r, rj) = 1 - (1 + r|D)"r. [C-19]
Non-informative prior distributions for the parameters were defined by assuming that all
parameters are independent and that logit(a/(a+ P)), logio(a+ P), logio(rr|), logio(r/ r|), and logit(pO)
are all normally distributed with mean 0 and standard deviation 10. By definition logit(x) =
C-14
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
log(x/(l-x)). The posterior mode parameter values were calculated by directly maximizing the
posterior probability. These values were used to compute the posterior mode dose-response
functions for the probability of infection (Equation C-17) and the probability of illness given
infection (Equation C-19). Uncertainty intervals for these dose-response functions were computed
by using MCMC to simulate vectors of parameter values.
Teunis et al. (2008a) analyzed data from eight outbreaks of E. coli 0157:H7 using a hierarchical
Bayes model. A homogeneous exposure model used a beta-Poisson dose-response function for the
probability of illness (Equation C-17). A heterogeneous version of the exposure model also
included known values of a dispersion parameter, treated as being the shape parameter for a gamma
distribution of the microbial densities. Using a different notation to that used in the paper, the
hyperparameters ml, m2, si, and s2 were assumed to be independent and have distributions such
that ml and m2 were normally distributed with mean -8 and standard deviation 8, and si and s2
were gamma (0.001,1000) distributed. For outbreak i, logit(a(i)/(a(i)+ P(i))) and logio(a(i)+ P(i))
were assumed independently normally distributed with means ml and m2 and standard deviations
si and s2. To obtain an overall group dose-response function, representing the dose-response
function for a future random outbreak, the a and P parameters of that outbreak were assumed to be
generated from the prior distribution of the hyperparameters; that is, logit(a/(a+ P)) and logio(a+
P) were assumed to be independently normally distributed with means ml and m2 and standard
deviations si and s2.
Teunis et al. (2008a) fitted these models using MCMC. The posterior mode dose-response function
for each outbreak was estimated by finding the sample parameter vector with the highest value of
the joint posterior (partial) probability, which is the product of the prior density for the
hyperparameters, the conditional density for a(i) and P(i) given the hyperparameters, and the
likelihood for the outbreak i. The overall estimates of the dose-response function for a future
outbreak were estimated by sampling a and P from the prior distribution of the hyperparameters
and computing the dose-response function for each pair. This gives a set of dose-response
functions. For each dose, the percentiles of the probability of being ill were computed and plotted
as a contour plot.
Teunis et al. (2008b) used Bayesian methods to analyze dose-response functions for the Norwalk
virus based on a volunteer study. Similar methods to the above studies were employed so the
details are not discussed here.
Englehardt (2004) compared maximum likelihood methods to a predictive Bayesian dose-response
approach. The method was applied to rotavirus data. First, he discussed the likelihood-based
Benchmark Dose Method, which computes a confidence interval for the dose at which a certain
change or percentage change in risk occurs and defines the benchmark dose as the lower
confidence value. He pointed out that for two different health stressors, it is possible that the dose-
response curves can intersect, in which case the more risky stressor (the stressor with the least
benchmark dose) depends upon the confidence level chosen. A similar issue arises with Bayesian
analyses using credible intervals to account for uncertainty; that is, the results depend upon the
assumed "confidence" level. Englehardt recommends averaging the dose-response function over
the posterior distribution of the parameters. In other words, the predictive Bayesian dose-response
model for dose D is calculated as the integral of the posterior distribution of the parameters given
C-15
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
the data multiplied by the dose-response function at dose D. This integral is over the entire
probability space. The predictive Bayesian dose-response is the unconditional dose-response
function and can be thought of as the dose-response function averaged over the uncertainty
distribution. This can be compared to frequentist or more standard Bayesian approaches for which
upper bounds of confidence or prediction intervals can be thought of as estimating the risk under
a "worse-case" scenario. Regulators may prefer a worse-case scenario approach to the predictive
Bayes approach, which represents the average scenario.
Englehardt (2004) applied the predictive Bayes approach to rotavirus data using the beta-Poisson
model. The maximum risk for any dose D is calculated using the exponential model with r = 1,
which assumes a hit probability equal to 1 so that infection is guaranteed (100%) if any organisms
are ingested. The minimum risk for any dose D is assumed to be obtained from the maximum
likelihood estimates of the parameters a and p. The predictive Bayes dose-response function has a
risk between the minimum and maximum risk. Englehardt (2004) points out that in general, if
enough data are available, then at low doses, the predictive Bayes risk will be lower than the
observed risk (proportion of illnesses), so that the approach is conservative (health-protective)
compared to maximum likelihood methods. To calculate the posterior distribution, Englehardt
used the MCMC method assuming an improper uniform prior for a and p. It is not clear from the
paper how the integral of the posterior multiplied by the dose-response function was calculated.
Although direct numerical integration is possible, in principle, a reasonable approach would be to
compute the dose-response function (probability of illness) for each dose D and each sampled pair
of parameter values, and then average the probability of illness over the entire sample. Note that
Englehardt defines the normalizing constant k as the constant that normalizes the likelihood
function. This is not correct in general since the normalizing constant k normalizes the posterior
distribution (i.e., the product of the prior and the likelihood).
Englehardt and Swartout (2004) applied the predictive Bayes approach to the Cryptosporidium
parvum data analyzed by Messner et al. (2001). However, these analyses separated out the results
for subjects with Ab+ and Ab- serum-antibody status. First, maximum likelihood estimates for the
beta-Poisson models were computed for each study (isolate) and Ab+ or Ab- status. Second, a
representative population of sensitive, Ab+, and Ab- subjects was simulated using the maximum
likelihood fitted dose-response functions; sensitive subjects were assumed to always respond at
the doses tested. Each simulated population is a parametric bootstrap sample. A beta-Poisson
model was fitted to each bootstrap sample using maximum likelihood. The set of maximum
likelihood estimates was used to compute 95% confidence intervals for the probability of infection
for each strain.
Englehardt and Swartout (2004) computed a predictive Bayes distribution for the r for a random
isolate and for the dose-response function. For a random isolate, r is assumed to have a beta
distribution with parameters a and P, assigned a joint uniform prior. The likelihood of the three
observed r values is given by the product of three beta distributions; each observed r value is the
mean for the bootstrap simulations of that isolate. Thus, the predictive Bayes distribution for r is
defined by multiplying the beta distribution for r by the posterior probability for a and P given the
three sampled values of r, and then integrating over the parameter space. This is the marginal
distribution of r. To obtain the predictive Bayes dose-response function, the marginal probability
C-16
-------
Microbial Risk Assessment Tools U.S. EPA Office of Water
for r was multiplied by the exponential dose-response function (Equation C-2) and integrated over
0
-------
Microbial Risk Assessment Tools
U.S. EPA Office of Water
------- |