United States
Environmental Protection
Agency
EPA/600/R-10/022 | July 2010 | www.epa.gov/ord
Workshop Report:
State-of-the-Science for
the Determination and Application
of Dose-Response Relationships in
Microbial Risk Assessment
(an i
EkIO". :
National Homeland Security Research Center Office of Research and Development
United States Environmental Protection Agency, Cincinnati, Ohio 45268

-------

-------
Workshop Report:
State-of-the-Science for
the Determination and Application
of Dose-Response Relationships in
Microbial Risk Assessment
APRIL 21 - 23, 2009
TOM HARKIN GLOBAL
COMMUNICATIONS CENTER
CENTERS FOR DISEASE CONTROL AND
PREVENTION
ATLANTA, GA
National Homeland Security Research Center Office of Research and Development
United States Environmental Protection Agency, Cincinnati, Ohio 45268

-------

-------
Disclaimer
This report was prepared as a summary of the presentations and discussions held at the U.S.
Environmental Protection Agency / Centers for Disease Control and Prevention Workshop, State-
of-the-Science for the Determination and Application of Dose-Response Relationships in Microbial
Risk Assessment (April 21-23, 2009). This report captures the main points and highlights of the
meeting; it is not a complete record of all detailed discussions, nor does it embellish, interpret, or
enlarge upon matters that were incomplete or unclear.
This text is a draft that has not been reviewed for technical accuracy or adherence to U.S.
Environmental Protection Agency or Centers for Disease Control and Prevention policy; do not
quote or cite. It does not necessarily reflect the Agencies' views. No official endorsement should be
inferred.
Questions concerning this document or its application should be addressed to:
Sarah Taft, PhD
U.S. Environmental Protection Agency
National Homeland Security Research Center
26 W. Martin Luther King Drive, MS NG16
Cincinnati, OH 45268
513-569-7037
Taft.Sarah@epa.gov
Microbial Risk Assessment Workshop Committee Members:
Sarah Taft, Ph.D., U.S. EPA, National Homeland Security Research Center
Tonya Nichols, Ph.D., U.S., EPA, National Homeland Security Research Center
Irwin Baumel, Ph.D., U.S. EPA, National Center for Environmental Research
Deborah McKean, Ph.D., U.S. EPA, National Homeland Security Research Center
Erin Silvestri, MPH, U.S. EPA, National Homeland Security Research Center
Stephen Morse, Ph.D., Centers for Disease Control and Prevention
If you have difficulty assessing these PDF documents, please contact Nickel.Katliy@epa.gov or
McCall.Amelia@epa.gov for assistance.

-------

-------
Contents
Disclaimer	iii
List of Figures	vi
Acronyms and Abbreviations	vii
Foreword	viii
Executive Summaiy	ix
Keynote Address	1
Presentation Sessions	3
Session 1: Federal Mission Needs for Microbial Risk Assessment	3
Microbial Risk Assessment for the Development of Cleanup Goals	3
Infection Transmission, Infection Control	4
Mission Needs for Dose-Response at FDA-CFSAN's
[Center for Food Safety and Applied Nutrition] Microbial Risk Assessment Program	4
Mission Needs at USD A	5
Session 2: Dose-Response Extrapolations	6
Accounting for Uncertainty and Variability with Mechanistic Knowledge
in Dose-Response Assessment	6
Let the Data Speak	7
NOAEL, LOAEL, and Dose-Response Curves: Lessons from Anthrax	7
Impact of Animal Models	8
Session 3: Pliysiological-Based Modeling	8
Physiologically-Based Modeling	8
Physiological-Based Modeling in Microbial Risk Assessment	9
Developing Mechanistic Models for Risk Assessment of Biothreat Agents	10
Session 4: Dose-Response Method Comparisons: Classical, Bayesian, Epidemiology,
and Benchmark Dose Modeling	11
Dose-Response Method Comparisons: Classical Studies	11
Dose-Response Comparisons: Bayesian Statistics	12
Modes of Action in Low-Dose Extrapolation	12
Microbial Dose-Response Methods Comparisons - Benchmark Dose Approach	13
Session 5: Dose-Response Applications for Vaccines and Therapeutics	14
Dose-Response Applications for Vaccines & Therapeutics	14
How are Biomarkers Utilized in Dose-Response Modeling of Infection and/or Disease? ..15
Dosc-Rcsponsc: Economics and Public Policy (or, the value of risk)	15
Group Discussion Summaries	17
Conclusions and Future Steps	23
Bibliography	25
Appendix A: Workshop Agenda	27
Appendix B: List of Parti pants	29
Appendix C: Slide Presentations	33

-------
List of Figures
Figure 1. The determination of risk-based target concentration is complicated by the
number of potential approaches and the assessment of minimum data requirements	3
Figure 2. Necessary steps for disease transmission	4
Figure 3. Microbial dose-response research needs and future directions of work
as identified by the FDA	5
Figure 4. Comparison of multiple fitted models to show different extrapolation
results in the low dose region of the curve	8
Figure 5. Considerations of interspecies differences that may reduce uncertainty
in animal-to-human extrapolations in microbial dose-response assessment for the
development of cleanup goals	9
Figure 6. Uncertainty in physiologicallv-based mechanistic modeling may be
reduced with the addition of increasing biological detail	10
Figure 7. When a population's resistance is distributed normally, the resulting
dose-response curve is a cumulative normal distribution (i.e., probit curve)	12
Figure 8. Risk of smallpox release relative to risk of smallpox infection for two
different exposed population sizes	15

-------
Acronyms and Abbreviations
AIC	Akaike's information criteria
BMD	benchmark dose
BMDL	benchmark dose 95% lower bound confidence limit
HMDS	Benchmark Dose Software
BMR	benchmark response
Cat Reg	U.S. Environmental Protection Agency's set of categorical regression models
also known as Cat Reg 2009 R Version
CDC	Centers for Disease Control and Prevention
CFSAN	Center for Food Safety and Applied Nutrition
EPA	U.S. Environmental Protection Agency
FDA	Food and Drug Administration
HACCP	Hazard Analysis and Critical Control Point
HIV	human immunodeficiency virus
ID50	dose that infected 50% of the test population
IgG	immunoglobulin G
IRAC	Interagency Risk Assessment Consortium
IRIS	Integrated Risk Information System
LD50	dose that caused death in 50% of the test population
LOAEL	lowest observable (or observed) adverse effect level
LOEL	lowest observable (or observed) effect level
LOTEL	lowest observable (or observed) tolerable effect level
MR A	microbial risk assessment
NOAEL	no observable (or observed) adverse effect level
PA	Bacillus anthracis protective antigen
PBBK	physiologically-based biokinetic
PBPK	physiologically-based pharmacokinetic
POD	point of departure
USAMRIID U.S. Army Medical Research Institute of Infectious Diseases
USD A	U.S. Department of Agriculture

-------
Foreword
Following the terrorist events of 2001, the U.S. Environmental Protection Agency's (EPA)
mission was expanded to account for critical needs related to homeland security. Presidential
Directives identified EPA as the primary federal agency responsible for the country's water
supplies and for decontamination following a chemical, biological, and/or radiological
attack. To provide scientific and technical support to help EPA meet this expanded role,
EPA's National Homeland Security Research Center (NHSRC) was established. The NHSRC
research program is focused on conducting research and delivering products that improve the
capability of the Agency to carry out its homeland security responsibilities.
As a part of its long term goals, one measure NHSRC has been charged with is deliveiy
of reports and databases with information 011 the health effects of contaminants by 2012.
Reliable dose-response data are critical to assessing the human health risks from exposure
to microorganisms originating from intentional and unintentional releases resulting in
contamination of buildings, drinking water systems, outdoor areas, or food. However, dose-
response data for biological threat agents in the low-dose range are very limited. To bridge
this critical data gap, advanced methods, animal studies, and other approaches arc required
to generate credible low-dose data to support the development of acceptable, scientifically -
defensible response and remediation actions.
The April 2009, State-of-the-Science for the Determination and Application of Dose-Response
Relationships in Microbial Risk Assessment workshop was held to discuss this and other
data gaps in dose-response relationships in microbial risk assessment (MRA). This effort
brought together many organizations across the country, including EPA's program offices,
federal government agencies and laboratories, academia, and the private sector. Participants
of the conference shared knowledge, explored differing opinions, and expanded overall
understanding in MR A dose-response relationships.
This report represents a summary of the presentations and discussions during the workshop.
We value your comments as we move one step closer to achieving our homeland security
mission and our overall mission of protecting human health and the environment.
Cynthia Sonkh-Mullin,
Acting Director National Homeland Security Research Center

-------
Executive Summary
Both the U.S. Environmental Protection Agency
(EPA) and Centers for Disease Control and Prevention
(CDC) are tasked with preventing and mitigating risks
presented by exposure to biological agents. These federal
agencies employ microbial risk assessment (MRA) to
inform the risk management decision making and risk
communication processes through credible scientific
data analyses. The practice of MRA has just recently-
expanded and. unlike the more formalized chemical
and radiological risk assessment processes, is not as
widely accepted or standardized. Therefore, various
agencies and organizations have individually determined
their own approaches, methods, and applications for
conducting MRA to fulfill the agencies' respective
missions.
EPA recognized these inconsistencies and the need
to provide a forum to present and discuss various
MRA methods and approaches employed by different
organizations. Therefore, EPA initiated the first
annual MRA Conference which was held in April
2008 in Rockville, MD. The conference had over 150
participants and included presentations on the mission-
directed applications of MRA by scientists representing
multiple federal agencies. Technical programs focused
on current data needs and research advances in MRA
hazard identification/characterization, exposure
assessment, dose-rcsponse, risk characterization, and
risk perception and communication. This innovative
meeting allowed microbial risk assessors to showcase
their research and collaborate with other scientists, risk
managers, and stakeholders from academia, private-
sector organizations, and federal agencies.
The success of and overwhelming participation in the
first MRA Conference prompted a second annual EPA
MRA Dose-Response Workshop in collaboration with
the CDC, which was held 21-23 April 2009, in Atlanta,
GA. As follow-on to the broader MRA presentations
and discussions that occurred during at the first MRA
Conference, this second workshop focused specifically
on the dose-response relationships in MRA. The dose-
response estimate describes the relationship between the
exposure dose of a biological agent and the probability
of adverse health effects. Reliable dose-response data are
critical to assessing the risks from exposure; however,
applicable and credible dose-response data for many
biological agents are veiy limited.
This report developed from the 2009 workshop, State-
of-the-Science for the Determination and Application
of Dose-Response Relationships in Microbial Risk
Assessment, summarizes and highlights the presentations
and discussions convened during the two and a half
days. The primary goal of the conference was to
share knowledge, explore differing opinions, and
expand overall understanding in MRA dose-response
relationships. Sixty-two workshop participants/subject
matter experts represented federal government agencies
and laboratories, academia, and the private sector. Dr.
Cynthia Chappell, from the University of Texas School
of Public Health, served as the keynote speaker. The
remaining conference agenda consisted of 19 speaker
presentations organized into five sessions:
•	"Federal Mission Needs for Dose-Response"
•	"'Dose-Response Extrapolations"
•	"Physiological-Based Modeling"
•	"Dose-Response Methods Comparisons"
•	"Dose-Response Applications for Vaccines and
Therapeutics"
Following each presentation session, there were
lengthy participant and presenter discussion periods.
The technical content of this Report is based entirely
on presentation information and discussions held at the
workshop.
The objectives of the Dose-Response Workshop were to:
•	Address the technical and scientific issues/
challenges in MRA dose-response
•	Discuss how to bridge critical data gaps using
advanced methods to generate microbial dose-
response data,
•	Examine novel approaches for the application
of MRA dose-response data to predict human
consequences
•	Share knowledge, improve understanding, and
identify data gaps for future research planning
through strong participation by subject matter
experts
Because of the diversity of attendees' disciplines,
the different inputs and decision making required to
support each organization's mission, and the limited
timeframe, our aim was not to have participants arrive at
a consensus on the best MRA dose-response approaches,
methods, and data. Instead, the primary goal of the
conference was to share knowledge, explore differing

-------
opinions, and expand overall understanding in MRA
dose-response relationships. The following are brief
summaries of some notable discussion highlights:
•	The key to the dose-response assessment is to
determine what can be done to make the models
useful and to recognize that the models do not need
to be perfect as long as the uncertainties encountered
in modeling projections are recognized and adjusted
for. Multiple models will most likely be necessary
to meet the challenges and required decisions. The
critical focus for model selection is who is making
what decision and why, the model should be built
and utilized to inform the decision.
•	The model needs to have the "right" complexity;
very often there arc too many parameters in the
models. There must be a deliberate effort to choose
the appropriate number of parameters in the dose-
response model while avoiding confronting the issue
of forcing the data to fit.
•	MRA requires evaluations that indicate the
adequacy of a dose-response model for the
particular assessment being performed. This, in turn,
requires clarification of assumptions being made
regarding the processes involved in generating the
data and determining outcomes. A model where
these assumptions are obscure is insufficiently
mechanistic.
•	The key processes or steps of the microbial infection
cycle (invasion, infection, illness) each have the
potential for a dose-response threshold. Within
each of these steps, there are potential barriers that
can influence the threshold dose required to reach
the next step. It remains difficult to discriminate
between these key processes or steps and to identify
and isolate the appropriate endpoints of invasion,
infection, and/or illness caused by a single pathogen.
•	Completely separating the exposure assumptions
from dose-response modeling is difficult. A thorough
understanding of the role of the environment in
exposure is necessary to better define the contexts in
which a biological agent exhibits pathogenicity and
therefore the dose-response relationship.
•	Pooling various dose-response data could allow
more information to be gathered to enable stronger
inferences as long as the differences in data sets can
accurately be reflected and adjusted for.
•	One of the greatest challenges with microbial dose-
response modeling is the very limited availability of
human dose-response data. As a result, the majority
of dose-response estimates rise from experimental
animal studies. To more accurately decrease the
uncertainty arising from the animal-to-human
extrapolations of dose-response data, microbial risk
assessors can utilize species-specific physiological-
based models.
•	Another challenge with modeling microbial data
from dose-response studies is that in most historical
studies extremely high doses were administered
to achieve effects, and therefore, the data require
extrapolation from high-to-low doses to predict
potential human responses at low doses. There can
be large orders of magnitude differences in dose-
response curve estimates in the resulting low dose
extrapolations depending on the dose-response
model utilized and the type and amount of data
being modeled.
•	Most dose-response models and data assume a
homogenous human population and generally
do not account for disease impact on sensitive
subpopulations. Outbreak data can be particularly
helpful in comparing the responses of healthy
populations with potentially sensitive subpopulation
(e.g., children, elderly).
•	It is best to combine the risk communication process
with the risk assessment/risk management processes
early in the planning and to do so often during the
entire process. Dose-response modelers need to
communicate the assumptions, and strengths and
weaknesses of their models up front so that decision
makers and risk managers can interpret and apply
the models correctly.
•	Developing standardized microbial risk assessment
terminology would be very valuable and would
facilitate successful collaborations across
disciplines. Communicating methods and results
between disciplines has been difficult at times as the
various disciplines sometimes apply different terms
to define the same approach or the same term to
define different approaches.

-------
Keynote Address
Microbial Risk Assessment Dose-Response
Challenges
Cynthia ChappellPh.D.
University of Texas School of Public Health
The keynote address, presented by Dr. Cynthia Chappell,
focused on the advantages and challenges of conducting
human dose-response studies. Dr. Chappell pointed out
that one of the greatest challenges in assessing human
health risk of pathogens is the lack of dose-response
data. For most pathogens, data are absent or limited to
studies conducted with surrogates and/or with laboratory
animals that may not mimic human disease. Additionally,
dose-response studies are generally conducted at high
doses to ensure disease effects will be observed; such
high doses may not be representative of doses relevant to
human exposures.
Dr. Chappell presented an overview of the feasibility
and ethics associated with collecting and analyzing dose-
response data from human studies. Her presentation
focused largely 011 her experiences with human
exposure studies to different Cryptosporidium species
and dosages. In this case, the dosing studies done in
healthy adults were appropriate since the infection is
self-limiting. The Cryptosporidium challenge studies
took place over 11 years (1993-2004) and involved
186 individuals. The objectives of the studies were to
describe the natural history of infection, identify the dose
that infected 50% of test population (ID50), calculate
illness attack rates, and evaluate immune responses to
infection.
The logistics of conducting human dosing studies require
considerable planning to address the ethical and safety
issues associated with infecting healthy people with a
pathogen, to ensure that volunteers are mentally and
physically qualified to participate in the study and are
available when the pretested inoculum is ready for use.
Unexpected and adverse events require special attention
during such studies and should be part of the planning
process.
Dr. Chappell summarized that, along with more
carefully-collected dose-response data, there is a critical
need for increasing knowledge regarding virulence
factors and for better data integration from multiple
animal and epidemiological outbreak studies.

-------

-------
Presentation Sessions
Session 1: Federal Mission Needs for
Microbial Risk Assessment
Representatives from the U.S. Environmental Protection
Agency (EPA), Centers for Disease Control and
Prevention (CDC), Food and Drug Administration
(FDA), and the U.S. Department of Agriculture (USD A)
presented the mission requirements for microbial risk
assessment (MRA). Specifically, each presenter was
asked to address: "How is microbial dose-response
information utilized in your agency's decision making?"
Microbial Risk Assessment for the Development
of Cleanup Goals
Tonya Nichols, Ph.D.
EPA
EPA conducts risk assessment of environmental
contaminants to inform risk management decisions.
To this end, detection capability must be in place to
determine that a release has occurred, contaimnent and
mitigation protocols must be accessible, and remediation
goals and strategies must be assessed and evaluated.
The overall role of the risk assessment is to determine
how much of the contaminant would lead to an adverse
health effect. This information guides clearance
decisions, identifies how sensitive our analytical
detection capabilities must be to determine the presence
or absence of harmful concentrations of a contaminant,
and defines "how clean is clean" to determine if the
decontamination effort has been successful. EPA has a
history of developing chemical target concentrations and
using associated information to support decision making.
Target concentrations that are currently used by EPA's
regulatory programs include preliminary remediation
goals in the Superfund program, health advisories and
maximum contaminant levels in the Office of Water, and
reference concentrations and inhalation unit risk in the
Office of Air Quality Planning and Standards.
Risk-based goals are a function of the target risk, intake,
and pathogenicity. The risk-based goal can be used
to derive target concentrations that can be compared
with sampling results. To develop target concentrations
for pathogens, the approach must be determined and
the minimum data requirements should be identified
(Figure 1). To date, the minimum data set required for
setting clean-up goals for biological agents has not been
determined. Existing guidelines, such as the "Animal
Rule" (21 CFR 601) and Minimum Data Requirements
for Registering a Chemical Pesticide (40 CFR 158) may
provide insights. Considerations for determining the
minimum data set include: extrapolation of data obtained
from surrogates to pathogens, defining differences
between exposure routes, use of high dose data when
low dose data would be more relevant, applicability
of animal models, and determination of associated
correlates of disease.
SERA
Target Concentration
Animal Exposures
in vitro Studies
•	High Dose
•	Low Dose
•	Chronic Exposure
•	Challenge Dose
\ ~
MetaData
Analysis
~ \
•	Multiple animal species
•	Multiple microbial strains
Evaluation of Biological
Parameters of Exposure
•	Telemetry - clinical symptoms
•	Bactremia
•	Toxemia
•	Inflammatory cytokines
•	Antibodies
•	Histology
Physiological Modeling
•	Portal of Entry
•	Deposition
•	Replication
•	Host Immune Response
•	Clearance
•	Translocation
National Homeland Security Research Center	Minimum Data Requirements ?
Figure 1. The determination of risk-based target
concentration is complicated by the number of
potential approaches and the assessment of
minimum data requirements.
Because there is no consensus on the minimum data
required for deriving a cleanup goal for biological
contaminants. Dr. Nichols provided the following
questions to stimulate discussions on identifying the
research needed to derive a cleanup goal:
1.	How do we approach a no observable adverse
effect level (NOAEL) / lowest observable
adverse effect level (LOAEL)/ lowest
observable tolerable effect level (LOTEL) for
exposure to microorganisms?
2.	What dose-response models do we use and why?
3.	How do we design in vivo and in vitro studies to
better inform physiological modeling?
4.	How do we extrapolate animal study data to
humans?
5.	How do we account for uncertainty and
variability?

-------
Infection Transmission, Infection Control
Michael Bell, M.D.
CDC
From an infection control perspective, disease
management conducted in health care facilities focuses
on stopping transmission from person to person or
from person to enviromnent to person. With regard
to emerging diseases, there seems to be a consistent
pattern: 1) disturbance of or intrusion into ecosystems, 2)
primary entrance into human host, 3) secondary spread
among humans, and 4) potential amplification in health
care facilities.
Disease transmission requires a number of steps to
occur (Figure 2). Infection control to minimize the
transmission of pathogens can essentially lead to
zero exposure. Survival in the enviromnent during
transit can be affected by a number of enviromnental
factors (temperature and humidity), droplet size, and
composition. Transmission-based precautions for
droplet and airborne transmissions should address
infectivity relative to the time/distance of travel and the
predominant transmission mode. For example, the strict
5|iIVI cutoff value for aerosol inhalability originated
from studies specifically related to tuberculosis and does
not represent the upper size limit for inhalability for
other pathogens. It also should not be assumed that all
inhalation pathogens must reach the terminal alveolar
region to initiate infection.
Disease Transmission
To cause an infection, a pathogenic organism must:
| Leave original host|
C ISurvive in transit I
^ | Be delivered to a susceptible host|
^ | Reach a susceptible part of the hos 11
C | Escape hostdefenses |
Multiply and cause tissue damage
Figure 2. Necessary steps for disease transmission.
Systematic assessments of infectivity should consider
questions about assessing pathogens separately by their
features (e.g., viral envelopes), using representative
organism versus specific organism, and using time as a
surrogate for distance. CDC's current research agenda
includes aerobiology and improvement of protective
equipment for health care. Aerobiology considers
organism-specific measurements, enviromnental
variables, and substrate variables.
Mission Needs for Dose-Response at FDA-
CFSAN's [Center for Food Safety and Applied
Nutrition] Microbial Risk Assessment Program
David Oryang, M.S.
FDA-CFSAN
FDA has a long history (since 1906) of managing
risks, conducting safety assessments and performing
risk assessments for food additives, chemicals, and
microorganisms. CFSAN's mission is to promote and
protect public health by ensuring that the U.S. food
supply is safe, sanitary, wholesome, and honestly
labeled, and that cosmetic products are safe and properly
labeled.
U.S. food safety is challenged by constant changes in
the food system including: 1) significant increases in
the volume, variety, and complexity of imported foods,
2) shifting demographics, 3) more convenience foods
being eaten year round, and 4) new foodborne pathogens
with relatively little available data. Each day, industry
and government agencies must make decisions about
the safety of foods and food products. The public health
and economic consequences of "bad" decisions can be
substantial; not deciding is not an option. There has been
tremendous effort in the food safety community to make
consistent and transparent decisions that are informed by
science and risk.
Growing responsibilities and new challenges require
federal regulatory agencies to develop new tools and
approaches. CFSAN is moving toward a more risk
analysis based approach developing and using efficient
means to collect, organize, review and share information
used in regulatory decisions, and prioritizing activities
in view of limited resources. Risk assessment is one of
three components of the risk analysis triad: assessment,
management, and communication. It is a process to
describe what we know and how certain we are of what
we know, and to answer four key questions: a) What can
go wrong?, b) How likely is it to occur?, c) What are the
consequences?, and d) What factors can influence it?
Risk assessment is a tool, used by CFSAN to:
•	Support food safety decision-making -
particularly when decisions must be made under
uncertainty and all of the desired data are not
available.
•	Support decision making for import policies,
control strategies, inspection programs, and
safety tolerance levels.
•	Assess the effectiveness of interventions
by evaluating control measures, proposed
standards, and the contribution of compliance to
risk management.

-------
•	Inform communication and outreach strategies
by identifying subpopulations that are at high
risk, assessing uncertainty and variability, and
presenting a comparison of alternatives.
•	Assist food safety management with decisions
by providing advice on where to look for
hazards, by setting priorities and allocating
resources, and identifying risk drivers along the
"farm-to-fork" continuum.
One of CFSAN's typical targets for risk management is
microbial contaminants in food and food additives.
Looking forward, a significant microbial dose-response
need for FDA is the identification and characterization
of susceptible populations. A number of other FDA
microbial dose-response related needs (presented in
Figure 3) are identified. One recognized important need
is the evaluation of growth models that would provide
greater effectiveness in estimating exposure levels. A
further need is to move the focus from acute (where it
currently is) to transient and chronic effects (where there
is currently little emphasis).
There is a need to increase accessibility to data, models
and information, and to develop dose-response relations
for new food-borne pathogens, by extrapolating data
acquired in animal models to humans.
Key data needs are: a) descriptions of the variability
in susceptibility; and b) variation in the infection to
hospitalization ratio; within and between age groups and
susceptible populations.
Looking Forward:
Growth models: More effective estimates of exposure levels.
CFSAN focus on acute, as well as transient/chronic effects.	
Susceptible populations - IRAC workincigraupWgjpclBiBPSW^^
¦	Variation in susceptibility within ageSroups	1
¦	Variation in susceptibility between age groups	1
¦	Variation in fatality to hospitalization ratio	1
Increase accessibility to data, models and information.
Dose-response relations for new foodborne pathqgens.
Extrapolate data acquired in animal models to humans.
Web based tools for risk ranking across products and hazards
(iRISK) « ¦*
Development of risk prioritization framework to allocate
resources across programs on the basis of risk and othjer
factors.
Figure 3. Microbial dose-response research needs and
future directions of work as identified by the FDA.
(IRAC is the Interagency Risk Assessment Consortium)
CFSAN is developing a Web based tool (iRISK) to rank
risks across products and hazards.
FDA's ultimate goal is the development of a risk
prioritization framework to allocate resources across
programs on the basis of public health risk and other
factors.
To address microbial risk assessment issues, participants
were challenged to learn from past experiences, develop
new ways to address complex food safety issues, foster
involvement of multi-disciplinary expertise, and actively
participate in international activities.
Mission Needs at USDA
Janell Kause, MPH, MPP
USDA
USDA research is focused primarily on risk ranking
and other parameters of relative risk. Dose-response
analysis previously received minimal consideration as
other risk assessment elements were further developed
in the USDA; in particular, exposure analysis received
significant attention. From a USDA perspective,
exposure analysis evaluates foodborne exposure from the
plant to table (i.e., "farm-to-fork"). Currently, attention is
re-focusing on dose-response.
Data for dose-response are obtained from both animal
and human studies. Challenges with animal data
include limitations of scaling from animal to humans,
conversion from mortality to morbidity, and the
associated significant overall uncertainty. With regard to
human studies, these have been conducted with healthy
human populations, which may not be informative for
susceptible populations. More often, there is a reliance
on epidemiological data obtained during outbreaks
of foodborne illness. Limitations from outbreak data
include: 1) insufficient information on exposure level
or the amount of pathogen consumed, 2) unknown
pathogen sub-type or its associated virulence, 3) limited
ability to recall food or food vehicles, 4) unspecific
endpoints, and 5) the subjective determination of which
outbreaks should be included in the epidemiological
data. When comparisons between risk models and
outbreaks are conducted, it can be shown that outbreak
data and the modeled response data usually do not line
up. There appears to be an underestimate of the dose
necessary to induce health effects for some low dose
exposures. The key issue is the lack of understanding of
who is susceptible to what microbial hazard.
The inclusion of more specificity in the dose-response
assessment will allow for a more refined focus on the
hazard, better recall of contaminated food vehicles, and
enhanced risk communication with those who are truly
at risk. Abetter understanding of susceptible populations
is needed; currently, the FDA uses age as the proxy for
susceptible populations. An additional element that is
necessary to advance the dose-response assessment is
the ability to make decisions with increased certainty,
especially for low-dose exposures.

-------
Session 2: Dose-Response
Extrapolations
Presenters were asked to address the following stimulus
questions:
1.	How is uncertainty and variability addressed
in extrapolating dose-response data (e.g.
extrapolating across host species, exposure
levels, routes of exposure, durations of
exposures, pathogen strains or species,
endpoints, and/or sensitive populations)?
2.	Is it appropriate to group studies, animal models
or host species, and/or pathogen strains or
species in dose-response modeling of multiple
data sets?
Accounting for Uncertainty and Variability with
Mechanistic Knowledge in Dose-Response
Assessment
Margaret (Peg) Coleman, MS
Formerly affiliated with Syracuse Research Center
The short answer to both stimulus questions on
extrapolation is:
•	Be skeptical.
•	Examine the body of evidence available
on mode/mechanism of action and dose-
dependencies for disease resistance and
susceptibility.
•	Pool only with scientific justification.
•	Extrapolate using knowledge of the disease
triangle and mode/mechanism of pathogenesis/
virulence.
Examples presented to support the above answer
featured an integrated dose-response methodology that
linked empirical and mechanistic knowledge in mice
and humans for anthrax, salmonellosis, and tularemia
(See Appendix C for details of specific examples). Such
methodology provides a more robust assessment of
dose-response relationships by incorporating variability
in all aspects of the disease triangle (host resistance
and susceptibility, pathogen infectivity and virulence,
and environmental influences on exposure and disease
progression). Prototype physiologically-based biokinetic
(PBBK) models for anthrax and tularemia, akin to
physiologically-based pharmacokinetic (PBPK) models
for chemicals, support scaling of external and internal
doses to target tissues that are stronger predictors of
outcome (resistance or susceptibility to disease) and
disease severity. More integrated knowledge will better
inform decisions about extrapolation and pooling.
For many infectious diseases, the tissue tropisms
and pathology differ by route and host, so to reduce
uncertainty in extrapolation, critical species-common
effects must be identified.
PBBK modeling illuminates the black box of dose-
response assessment and expands our limited ability
to predict resistance and susceptibility to pathogens,
and the likelihood and severity' of human disease under
conditions of susceptibility. Two groups independently
developed mechanistic models for inhalation anthrax
adapting existing methodology from chemical risk
reported in the EPA Integrated Risk Information System
(IRIS) to interrelate empirical and mechanistic models
for respiratory system pathogens. A team of scientists at
Syracuse Research Corporation prepared anthrax PBBK
models for guinea pigs and primates, and subsequently
extended that work to develop a prototype PBBK model
for tularemia in primates and humans (see Lumpkin
presentation).
Empirical models for non-human primate datasets
are available for multiple endpoints and Francisella
tularensis strains that can expand when linked
with PBBK models using integrated dose-response
assessment methodology. The empirical model for
non-human primates, informed by the PBBK model,
predicts an internal dose-response function. This
predicted internal dose-response function then informs
the human PBBK model to predict human internal
and external dose-response models based upon the
additional endpoints and strains observ ed in the non-
human primates. In this case, existing human curves for
infectivity from volunteer studies can be used to exercise
the methodology in reverse i.e., predict non-human
primate curves as an additional check on the validity of
our approach.
The scientific basis of the current practices that focus
on empirical modeling can be improved by accounting
for mode/mechanism of action, particularly variability
in aspects of the disease triangle. Some datasets,
including salmonellosis and tularemia, offer both
human and animals dose-response datasets that could
prove useful for testing hypotheses and validating
dose-response assessment methodology. Advancing
prototype mechanistic models of disease in respiratory,
gastrointestinal, and dermal systems would facilitate
model-directed research that could further refine
our models and expand our knowledge of low-dose
behaviors for significant pathogens. Such knowledge
is key to selection of 'safe' levels unlikely to cause
disease (as per the statistical threshold demonstrated for
Salmonella pullorum for humans) or severe disease (as
per tularemia endpoints for primates).

-------
Let the Data Speak
Chuck Haas, Ph.D.
Drexel University
Intrinsic maximum likelihood fitting techniques can
be used to account for experimental variability; other
mathematical techniques can be used to evaluate
parametric uncertainty. Statistical tests can be conducted
to assess the appropriateness of pooling. Tests have been
conducted and decisions made to pool data between
strains, species, hosts, and sensitive subpopulations
for various published data sets. Pooling of data can be
conducted if the data justify it based on statistical and
biological rationale.
The dose metric should be an ingested or inhaled
number but other dose metrics may also be appropriate.
One current Drexel project is evaluating in vivo
pathogen dynamics to assess whether body burden
is an appropriate metric. It is also possible that body
burden, area under the curve, or other measures may be
appropriate.
The mechanistically derived dose-response models
(exponential. beta-Poisson) have been found to be
consistent with all data sets examined to date, including
human data on: 'Legionella pneumophila. Salmonella
typhimurivm, Giardia lamblia, E. coli 0157:H7,
Cryptosporidium parvum, B. anthracis (Sverdlovsk),
and severe acute respiratory syndrome (SARS). In many
cases, it has been possible to validate use of such models
against outbreak data.
NOAEL, LOAEL, and Dose-Response Curves:
Lessons from Anthrax
Thomas Whaien, Ph.D.
Georgia State University
Extrapolation models (e.g., linear, logit, probit, log
probit) have all been used for dose-response modeling
of anthrax data. Many publications involve extrapolation
from high dose animal studies (especially Jemski's
unpublished data from Glassman, 1966) to assess
potential adverse effects on humans arising from
extremely low doses (e.g., nine spores infecting 2% in
Meselson's analysis). However, close examination of the
published historical accounts of actual human exposures
do not support anthrax disease after exposure to such low-
dose levels. For example, Holty's review of diagnosed
anthrax cases over 107 years only found 32 documented
cases. Furthermore, the evidence indicates that the
1957 Manchester outbreak originated from egregiously
contaminated goat hair; however, air monitoring after the
outbreak had ended found hundreds of spores in the air
at a time when no new cases were identified.
Brachman exposed macaques for 47 days to B.
awfArac/'.v-contaminated air from a working South
Carolina mill. Measured environmental concentrations
were found to be highly variable (ranging from tens to a
few hundred spore-containing particles inhaled by each
macaque in a day). Under these varying conditions, a
number of monkeys developed inhalational anthrax.
Dr. Whaien and colleagues used Brachman's published
data (1966) to estimate the number of spore-containing
particles inhaled by humans working in the mill, based
on human respiratory rates and an air cleaning system
with approximately 90% efficiency. The result was
an estimate that workers in this mill setting would
likely have inhaled over 600 spore containing particles
per day for 36% of the 47 days of Brachman's study.
Extending this to the estimated number of worker-days
in the 60 years from the beginning of the twentieth
century mill ventilation in the United States to
widespread vaccination of mill workers around 1960
yields approximately 15,000 unvaccinated worker days
associated with doses greater than 600 spore containing
particles inhaled per day in the mill - with fewer than ten
cases of inhalational anthrax documented. Dr. Whaien
proposed 600 spore containing particles per day or
fewer as a potential NOAEL. Likewise, a LOAEL can
be developed based on the assumption of approximately
18 million worker days across all mills with over 600
particles (during the period of 1900 to 1960), with at
most nine cases of reported inhalational anthrax disease.
It was noted that these exposures are likely to include
spikes in anthrax spores in the air significantly greater
than 600 spores inhaled per worker per day.
Furthermore, observational data seem to contradict
the results obtained from low-dose extrapolation of
experimental data as conducted by Meselson and many
others. For example, millworkers' wives and children
were likely receiving more than nine secondhand spores
yet they were not experiencing anthrax-related illness.
Likewise, workers themselves were likely experiencing
workplace conditions with non-zero exposures to anthrax
spores - yet were not exhibiting anthrax illness in the
numbers anticipated based on the worker population pre-
1965 prior to introduction of the vaccine.
Dr. Whaien asked the following questions: Which is
better - the use of imperfect data based on thousands
or even millions of human exposures at low doses or
experimental laboratory data based on dozens of animals
at high doses? The challenge today is how to integrate
the two disparate results. Considerations of who will
make the decision and how it will be used are important
factors that may guide the integration process.

-------
Impact of Animal Models
Mary Alice Smith, Ph.D.
University of Georgia
One of the known issues in the low dose region of the
dose-response curve is the fact that the fitted models with
relatively similar estimates in the middle dose regions
may provide very different response estimates in the
lower end of the curve (Figure 4). The current challenge
is the determination of which curves may be correct.
These modeled example data represent doses outside the
range of the test that generated the original data and are
outside the realm of typical animal studies given the low
probabilities of occurrence.
Session 3: Physiological-Based
Modeling
Presenters were asked to address the following stimulus
questions:
1.	What overall assumptions are necessary for
valuable physiological models to predict human
consequences?
2.	What is the minimum data set required (i.e.,
what level of detail needs to be modeled for
acceptable human predictions (e.g., whole
species models, organ-specific models, and/or
cellular or toxin activity models)?)
Figure 4. Comparison of multiple fitted models to
show different extrapolation results in the low dose
region of the curve.
One concept meriting consideration is the International
Life Sciences Institute "Thresholds in the Dose-
Response for Bacterial Pathogens Key Events" approach.
Key events are those events along the pathway between
intake and ultimate effect. This approach has been
successfully used by EPA in the evaluation of thresholds
in the dose-response analysis of chemical hazards and
may have applicability for determination of thresholds in
microbial hazards. Overall, dose-response relationships
will reflect a summation of steps, and the process of
understanding and modeling each step will lead to a
superior modeling approach. It will allow for prediction
of the variation in response due to human variability
or microbial strain and may allow for evaluation
of the interaction between quorum sensing and the
immune system. There is value in the use of an iterative
approach: 1) develop animal models with the available
data, 2) match human incidence data on dose-response to
identify closest animal model dose-response relationship,
then 3) consider mechanisms involved to further refine
animal and model selections.
Physiologically-Based Modeling
Sarah Taft, Ph.D.
EPA
Assessing the human dose-response relationships for
microbial agents is often challenging as actual human
data is very limited and, in most cases, non-existent.
Therefore, to estimate these human dose-response
relationships, especially with regard to biological threat
agents, data must be extrapolated from experimental
animal models of infection and disease. To date,
however, there is no standard approach or consensus-
based methodology for animal-to-human microbial data
extrapolations.
For chemical risk assessments, the NOAEL observed in
the animal model, is divided by some magnitude of an
uncertainty number to account for the species differences
and interspecies extrapolations from animal-to-human.
EPA has a long history in conducting chemical risk
assessments, so Dr. Taft raised the question, can an
approach that is similar to that applied for chemical risk
assessments be applied to microbial hazards where an
uncertainty factor of some magnitude is used to account
for the uncertainty in the animal-to-human microbial
data extrapolation?
Dr. Taft went on further to ask the following questions:
What can we learn from these approaches? How can we
reduce the uncertainty in animal to human extrapolation?
And, are these chemical dose-response analysis
approaches appropriate for microbiological hazards?
Consideration of interspecies differences, as is done in
chemical risk assessments, could reduce uncertainty in
animal-to-human extrapolations for microbial agents.
Interspecies differences could be broken down into
kinetic and dynamic elements for physiologically-based
modeling (Figure 5). Kinetic elements are physiological
factors affecting the ability of the microorganism
to reach the target tissues. These typically include

-------
dosimetric concepts relating to a deep dose in the lung
and other targets; examples of kinetic differences with
regards to B. anthracis infections include bacteria
clearance from alveoli, bacterial germination rate,
lymph node bacterial dose, bacterial dose in circulation,
bacterial toxin production, and bacterial replication.
Dynamic elements are differences in the effects at the
target tissue (i.e., the response); these include bacterial
toxin activity and host inflammatory responses.
S Pra	Dose-Response Assessment for the
|Pi>mn	Development of Cleanup Goals
Agency"1611
How can we decrease uncertainty in animal-to-human
extrapolations for biothreat agents?
Physiological factors affecting the	Effects at the target tissues
ability of the pathogen to reach
the target tissues
Figure 5. Considerations of interspecies differences
that may reduce uncertainty in animal-to-human
extrapolations in microbial dose-response assessment
for the development of cleanup goals.
Physiological-Based Modeling in Microbial Risk
Assessment
Jeff Gearh art, Ph .1).,
Wright Patterson Air Force Base
Dr. Gearhart prefaced his presentation with three key
statements regarding physiologically-based biological
modeling in MRA:
Currently, physiologically-based biological
models in MRA are primarily research tools -
quantitative methods for hypothesis testing with
experimental data.
Physiologically-based biological models are
NOT intended to replace other modeling
approaches but are hopefully an adjunct to other
modeling approaches.
Physiologically-based biological models may
not be required or necessary for all MRA
applications.
The biggest assumption in physiologically-based
biological modeling is that there is an understanding
of the actual mechanism(s) of pathogenesis for the
microorganism of interest. The host-pathogen interaction
must be understood to derive quantitative measures of
this interaction. It is also imperative to understand the
immunological processes i.e., similarities and differences
between hosts and experimental animals, for both the
animal host and the human, to conduct animal-to-human
extrapolations. This understanding has been a challenge
with the re-analysis of historical experimental animal
data. Importantly, the assumptions should also depend on
the questions being asked (e.g., the focus on death as an
endpoint in past assessments limits the knowledge that
can be gained from these data).
The main motive for the development of different
physiologically-based biological models for various
agents is to understand the actual mechanisms of
pathogenesis. Physiologically-based biological model
development starts with an overall mechanistic
schematic of the infection process - route of microbial
entrance into the host, initial microbe response, and
subsequent host response. Quantitative laboratory
measurements can be used for endpoints input into the
model, and these endpoints can in turn also be used to
evaluate model predictions. The quantitative measures
of the host-pathogen interaction are the most critical
physiologically-based biological model elements.
The largest drawback of the modeling approach is how
"data hungry" the physiologically-based biological
models are. Determining the minimum data required
depends on how global the physiologically-based
biological modeling approach is. There is considerable
animal data available in the literature and from the lab,
but the question becomes, can the data be utilized in
and can it advance the physiologically-based biological
model (e.g., incorporating in vitro data in a whole animal
model)? Furthermore, the ultimate goal is for these
models to predict the potential human consequences;
therefore, the corresponding data and parameters first
modeled for experimental animals must subsequently be
coded for humans in the overall model.
Most of the existing studies in the historical literature
for modeling anthrax use high dose exposure data.
Reliance on high dose data confounds the identification
of relevant mechanisms, particularly with respect to
the relationship of time to dose. Physiologically-based
biological model development for anthrax requires
quantitative information on the following data elements:
spore deposition in the alveoli, ingestion by the alveolar
macrophage, germination and replication of vegetative
bacteria, transport to the lymph nodes, defeat of the
macrophages, and active B. anthracis replication
producing bacteremia and toxemia.

-------
Developing Mechanistic Models for Risk
Assessment of Biothreat Agents
Michael Lumpkin, Ph.D., DABT
Syracuse Research Corporation
To address the first question of what assumptions are
required for physiologically-based models, the ultimate
application of the model must first be defined. There are
three main uses of physiologically-based mechanistic
models for dose-response analysis: 1) retrospective
applications - these include extrapolation of an observed
dose-response relationship in animals-to-human
receptors and are commonly conducted for chemicals, 2)
prognostic applications - these allow for the prediction
of health outcomes after a biothreat incident, and 3)
prospective applications - these allow for exploration
of measurable forensic biomarkers and can also be used
to back-extrapolate from outcomes to exposures or to
inform identification of health outcomes from a given
exposure.
The output of the physiologically-based mechanistic
model is typically a computer simulation of events
from exposure to disease. To produce defined dose-
response empirical relationships, these models derive
outputs from the knowledge of biology along with the
understanding of the pathogen kinetics and dynamics. As
modeling increases in biological detail, it is anticipated
that uncertainty is reduced (Figure 6). However, the
simple addition of more biology doesn't necessarily give
you more certainty. The overall desire is that models are
useful outside very narrow applications.
Figure 6. Uncertainty in physiologically-based
mechanistic modeling may be reduced with the
addition of increasing biological detail.
Minimal data requirements increase as modeling moves
from empirical to the more physiologically-based
mechanistic approach. As additional data elements are
added with more mechanistic information, it is believed
that the predictive capacity increases and movement
is allowed along the continuum of retrospective,
prospective, and prognostic applications. This movement
appears to be a generally linear process to initially
gather and supplement the available data. However, Dr.
Lumpkin raised the question what happens when the
data gap is at the initial step of the identified empirical
relationship stage? To address this challenge, the best
overall approach may be to develop models for multiple
agents in concert and to share data parameters across
agents. With such a cross-cutting approach identifying
one data element for one tract that is missing in another
could be very beneficial and could greatly advance the
development of microbial physiologically-based models.
Syracuse Research Center is developing physiologically-
based models by using the general approach of
moving from empirical to mechanistic. The following
assumptions are utilized in their particle inhalation
model: 1) generalizations about deposited doses capture
relevant details of exposures, 2) in vivo pathogen
growth rates change as a consequence of host-pathogen
interactions, 3) the in vivo pathogen and toxin rates are
biologically justified, 4) a critical species-common effect
has been identified; and 5) an internal dose metric has
been identified that is sensitive to the critical effect.
Session 4: Dose-Response Method
Comparisons: Classical, Bayesian,
Epidemiology, and Benchmark Dose
Modeling
Presenters were asked to address the following stimulus
questions:
1.	Is the dose-response statistical method utilized
empirical or mechanistic?
2.	Is the method applicable for low-dose
extrapolations?
3.	Can the method accommodate data pooling and/
or the use of correction factors?
4.	How is the calculated dose-response
relationship verified and validated?
5.	How is model uncertainty adjusted for and
communicated to risk managers?
Dose-Response Method Comparisons:
Classical Studies: Quality assurance process
and techniques for leveraging new and old data
Tim Bartrand, Ph.D.
Clancy Environmental Consultants
Mechanistic models are derived based on assumptions
regarding the probability distribution of the dose and the
probability of a single pathogen initiating an infection. If
it is assumed that all pathogens have equal probability of
initiating infection, the model is exponential (i.e., if the

-------
probability is constant). If the probabilities of pathogens
initiating an infection are beta-distributed, the model is
a beta-binomial; the assumed host response may also be
bcta-binomially distributed.
Classical dose-response models are mechanistic as they
are based on biologically plausible processes (e.g.,
dose or dose distribution is known and infection can be
modeled as a single event or as a sequence of events).
Low-dose linearity represents a finite probability that
a single organism can initiate infection. However,
models can be increasingly complex to incorporate
considerations regarding competitive processes, survival
or time to infection modeling, and fractional dose
models.
The following quality assurance process has been used
by Dr. Haas' lab at Drexel University in the development
of classical dose-response models. First, the data are
screened to evaluate the suitability of the data for use
in dose-response modeling. Considerations for the
data evaluation include knowledge of the pathogen
strain, origin and features, exposure route definition,
dose or environmental concentration, defined end point,
animal description, and observation of intermediate
response measured (not necessary when using lethality
end point data). A test for trend (Cochran- Armitage) is
also conducted; it should be noted that some data sets
(e.g., Lassa virus aerosols) have a veiy flat curve where
the trend is not visually obvious. The data determined
suitable for dose-response modeling are fit with the
classical models using a maximum likelihood method.
Goodness of fit can be examined by a clii square test
with fit assumed when the deviance is less than the
clii square value based on the number of dose groups
minus the number of parameters in the model. The
best fit model has the lowest deviance. A bootstrap
process can be used to generate confidence intervals
for the communication of variability and uncertainty.
The outputs of this analysis include 1.) distributions
associated with parameter estimates and 2.) confidence
intervals on the percentile response value.
With regard to pooling, a statistical test in combination
with other biological considerations is used to
determine the acceptability of pooling data. Biological
considerations include the support of pathological
findings for pooling and determination that no systematic
differences exist between populations (e.g.. inbred versus
wild, prior exposure/immunity, age, diet.). An example
was presented regarding determining the appropriateness
of pooling two human and animal dose-response data
sets for tularemia inhalation exposure. Curves with
statistically valid fits were developed using the beta-
Poisson model for the human data and exponential
model for the monkey data. While the individual data
sets could be fit to available models, the pooled data
could not. Pooling analysis can be used to determine
whether responses of animal hosts come from the
same distribution, or to compare results of multiple
experiments.
Models and associated codes should be verified; model
components have to be benchmarked, the models
themselves have to be robust, and multiple users have
to produce the same results when using them. Models
should also be validated; multiple data sets, outbreak
data, time-to-response data, or comparison of models for
multiple routes can all be used to validate models. These
validation data sets oftentimes have limitations which
should be taken into account (e.g., limited dose data in
outbreaks).
Dose-Response Comparisons: Bayesian
Statistics
Jade Mitchell-Blackwood, M.S.
Drexel University->
Bayesian analysis can reflect a broad array of analyses.
Its essential concept is the application of Bayes'
Theorem to learn from available previous observations.
Parameters are defined as random variables, rather than
the discrete values used in Classical dose-response
models. Bayesian analysis can be applied to both
empirical and mechanistic dose-response models. The
use of exponential and Beta-poisson models for earlier
work was described by Haas et al. as mechanistic. The
rationale for terming these models as mechanistic is that
they have been based on biological plausibility relating
to random doses with a Poisson distribution in the dosing
medium. It is assumed that there is a probability that
one or more organisms may be ingested by the host and
there is a probability of survival for a single organism
once it is ingested, rather than quantitative assessment of
individual organism survival rates. These models have
also been termed "mechanistically-based empirical"
because the parameters are determined empirically from
curve-fitting response data.
Bayesian methods can be applicable for low-dose
extrapolations if the model being fit can be used for low
dose extrapolation. For example, the exponential and
Beta-Poisson models are based on assumptions that, if
true, would allow for low-dose extrapolations. However,
obtaining adequate low dose data to validate these
assumptions is a challenge.
Bayesian methods can allow for methods to
accommodate data pooling and the use of correction
factors. Hierarchical models can be used to perform
meta-analysis without the need for the pooling
assumptions required by the classical approach. It should
be noted that Bayesian analysis is unique in its ability
to handle hierarchical modeling relative to oilier dose-
response statistical methods. Using data from Bart rand

-------
et al. (2008), an inter-species variation example can
show the potential for Bayesian analysis to use available
data for a number of different species. In this example,
it should be noted that Bayesian analysis allows for the
use of the white rabbit data (with 100% lethality at all
doses tested) which would not normally be useful for
determination of dose-response analysis using other
methods. Bayesian methods can use this data along with
the data for the other host species to generate unique
parameter distributions for each species individually
and a generalized parameter distribution, based on the
hierarchical model, for all species observed (for which
observ ations are initially available) and unobserved (for
which predictive distributions are required).
When using Bayesian analysis, the calculated dose-
response relationship can be verified and validated
through a number of different complementary
approaches. As with other techniques, graphical plots
of model and data are a good first step to evaluate fit.
The Bayesian Information Criterion and Deviance
Information Criterion can be used to score fit. The
Bayesian Information Criterion is similar to the Akaike's
information criterion (AIC) in that it is calculated
as a log-likelihood with a penalty for the number of
parameters. Models can be cross-validated if there are
sufficient data that were not used to generate the model.
Verification of the lack of impact of the assumed prior
distribution on the resulting posterior distribution can be
assessed by using both an informed and an uninformed
prior distribution and comparing the results. To show
that the results are unbiased, the resulting posterior
distributions should be similar. Model uncertainty' can
also be evaluated and reported, using Bayesian analysis,
by calculating credible intervals from the posterior
parameter distributions.
Modes of Action in Low-Dose Extrapolation
Laurie Waisel, Ph.D.,
Concurrent Technologies Corporation
Modes of action are important to consider when
conducting low dose extrapolations. A mode of action,
as distinguished from a mechanism of action, reflects
a mathematical approximation that does not require
an understanding of the molecular level. It basically
requires the mode of action to be biologically plausible
and that it fit mathematically.
A number of important concepts for the development
of dose-response curves can be shown through a
comparison of the assumptions implicit in a horizontal,
vertical, and diagonal dose-response function. In the
horizontal function example, susceptibility is 0 or 100%.
any dose is lethal if susceptible and a host is either 0%
or 100% susceptible. In the vertical function example, all
have the same resistance and the lethal dose is the same
for all. However, not all doses are lethal. For the more
typically presented diagonal dose-response line, the risk
is proportional to the dose (e.g., carcinogenic radiation).
The probit curve is an example where a population's
hazard resistance is normally distributed and the dose-
response relationship follows a cumulative normal
distribution (Figure 7). Here the individuals who will
exhibit a response at a given dose follow the probit
model (cumulative normal distribution). However, it
should be noted that multiple parameters determine
resistance.
<
Dose
Figure 7. When a population's resistance is
distributed normally, the resulting dose-response
curve is a cumulative normal distribution
(i.e., probit curve).
Variability is another important concept in dose-response
modeling. Variability in dose-response is the result
of individual differences in resistance. With perfect
information, all dose-response queries can be answered
with certainty and models are essentially deterministic
(e.g., diagonal line model example). Uncertainty in dose-
response is a result of the element of chance. Chance
can be described as stochastic, or probabilistic. In these
scenarios, even perfect information will never allow
for certainty in the answer because of the influence of
chance (e.g., radioactive decay). The communication of
uncertainty when describing results should distinguish
between uncertainty and variability, and provide an
analogy that will help explain the model results.
Decision science can be used to inform development of
models. The starting point should be an identification of
the real-world decision(s) to be made using the model
and the drivers for the decision. To make decisions in
a well informed manner, the theoretical and empirical
considerations as well as qualitative and quantitative
information should be considered. To determine
appropriate models for use, biological plausibility and
validation with empirical data should be used.

-------
Microbial Dose-Response Methods
Comparisons - Benchmark Dose Approach
Jeff Gift, Ph.D.
EPA
The benchmark dose (BMD) modeling approach
involves the application of empirical modeling
(mathematical curve fitting) methods to available data
(e.g.. dose-response data for a given toxicological
endpoint). One advantage of empirical modeling is that
it is intuitive and it relies on all of the dose-response data
to derive a risk assessment point of departure (POD).
A disadvantage is that such empirical approaches can
provide very different curs e fits in the low-dose region
of the dose-response depending on the selected model.
There can be high uncertainty in these estimates with
respect to both the accuracy and biologically plausibility
of the results. As a result, in the application of BMD
methods, risk assessors need clear guidance that takes
these uncertainties into account. There are published
recommendations on determination of a model fit for
available data (i.e., Haas, Rose, and Gerba, 1999). There
is considerable overlap between the above referenced
book and the guidance contained in EPA's BMD
Technical Guidance (U.S. EPA 2000) and Benchmark
Dose Software (BMDS) (U.S. EPA, 2009).
Considerations in the process include evaluation of
the best parameter estimates for a given dose-response
model, determination of an adequate model fit,
approaches to determine among a set of plausible models
which model best fit the data, evaluation of uncertainty
in parameter estimates and the benchmark response
(BMR) level from which to derive a benchmark dose
POD. The U.S. EPA's BMDS (U.S. EPA, 2009)' and
BMD technical guidance documentation (U.S. EPA
2000) provide a set of tools and procedures for making
these determinations.
BMDS is an open source platform that facilitates the
application of BMD methods by fitting the mathematical
curves to dose-response equations. The BMDS can
run a suite of models and the results can be compared
in tabular or graphic form. The evaluation includes an
AIC value (Akaike's information criterion)2, goodness
of fit measure (p-value), calculated benchmark dose
and benchmark dose level. Chi-square residuals are
also available for dose groups, including those of lower
doses (the area in the dose-response that is generally of
greatest concern) in the data set. EPA's BMDS website
(www.epa.gov/bmds) contains training materials and
/ At this time, BMDS offers over 30 different models that are appro-
priate for the analysis of dichotomous, continuous, nested dichoto-
mous and time-dependent toxicological data.
2 Akaike's Information Criterion (AIC) (Akaike, 1973) is used for
model selection and is defined as -21, 2P where L is the log-likeli-
hood at the maximum likelihood estimates for the parameters and P
is the number of model degrees of freedom.
flow charts which walk users through the process of
doing a BMD analysis including determining the most
appropriate BMR and the best fitting model. The BMD
approach as it is applied by EPA has an additional
advantage; it accounts, in part, for the quality of the
study (e.g., study size) by estimating a BMDL, the 95%
lower bound confidence limit on the BMD. The BMDL
is closer to the BMD (higher) for larger studies and
further away from the BMD (lower) for small studies.
Thus, the BMDL accounts, in part, for a study's power,
dose spacing, and the steepness of the dose-response
curve.
The BMDS can accommodate current data gaps. For
example, low dose extrapolation for cancer dose-
response assessment and microbial risk assessment
earn some of the same challenges. In this case, policy
determinations have been made to assume linearity or
nonlinearity, and to use this as extrapolation mechanism.
EPA has also developed a set of categorical regression
models, CatReg 2009 R version (CatReg) that may
provide assistance in addressing some of the data
extrapolation gaps in question. The software has been
built to run on an R platform and the software is open
source (http://www.epa.gov/ncea/catreg). Categorical
regression allows for the use of categorical responses to
be modeled, with time and intercept parameters, which
could allow the data to be pooled and the probability
of getting x- responses at a specified severity to be
calculated. CatReg can also be used to evaluate response
over different time durations. CatReg also allows for the
stratification of dose-response data (e.g., by species, sex
or strain) so that the contribution of each stratification to
the overall model fit can be estimated.

-------
Session 5: Dose-Response Applications
for Vaccines and Therapeutics
Presenters were asked to address the following stimulus
questions:
1.	How are biomarkers utilized in dose-response
modeling of infection and/or disease?
2.	Can dose-response thresholds be estimated for
vaccines and/or therapeutics?
Dose-Response Applications for Vaccines &
Therapeutics
Conrad P. (Juinn, Ph.D.
CDC
Biomarkers of infection and disease are valuable tools
for formulating an earlier diagnosis, informing patient
management, and monitoring therapeutic intervention
and disease progression. Dr. Qiiimi presented Ms
current research on measuring exposure to B. anthracis
and understanding the potential clinically detectable
biomarkers of exposure, infection, and disease.
Exposure to environmental B. anthracis may be
innocuous because of the protection afforded by
host intact immune barriers or may elicit an innate
host response; exposure does not necessarily result
in infection and subsequent anthrax disease. Disease
progression from infection is dependent upon spore
uptake and germination; this is a secondary key step to
the breach in the host intact immune barriers. Potential
host biomarkers for environmental B. anthracis
exposure and infection include host responses associated
with, lethal factor, protective antigen (PA), and
capsular y-linked poly-D-glutamic acid. Anti-anthrose
trisaccharidc is antigenic, exposed on the surface of
spores, and contains a B. anthracis-specific epitope.
Lethal factor toxemia is specific to B. anthracis, is
quantifiable in serum/plasma and, in a rhesus macaque
model of inhalation anthrax, becomes detectable
approximately 12-18 hours after an initial spore aerosol
exposure. Seroconversion to anti-PA immunoglobulin
G (IgG) is also a host biomarker for anthrax. During the
anthrax letter attacks of 2001, anti-PA serology was a
contributing test in the confirmatory diagnosis of 12/22
cases, critical to the diagnosis in 6/22 cases and the
single confirmatory test in 3/11 cutaneous anthrax cases.
General trends that are consistent with survival observed
in experimental animals as well as one human case
include decreases in lethal factor with a concomitant
increase in anti-PA IgG.
Dr. Quinn proposed several issues for further discussion.
What should be the dose-response threshold for
biomarkers of exposure, infection, and seroconversion
responses? With current levels of knowledge, the
dose-response modeling may not be sufficient for
curve determination; it may be appropriate to consider
the development of a toolbox to begin to address this
data gap. Regarding the dose-response thresholds for
therapeutics - what is the timeframe post-exposure
within which the drag is effective? Is there a point of
no return? Could certain treatments actually exacerbate
disease?
Dose-response predictions can be estimated for vaccines
and therapeutics. For vaccines, field efficacy and
immunogenicity studies are part of vaccine development.
Combined measures of experimental models can assist
in evaluating vaccines to ensure effectiveness. There
are a number of measures for therapeutics which are
routinely conducted as part of product testing including
pharmacokinetics, therapeutic indexes, and therapeutic
window determinations.
How are Biomarkers Utilized in Dose Li
Response Modeling of Infection and/or
Disease?
Louise Pitt, M.I).
US Army Medical Research Institute of Infectious
Diseases (USAMRIID)
One of USAMRIID's current overarching research
goals is the development of well characterized animal
models for aerosol exposures of biological threat
agents. By developing these animal models of disease,
appropriate biomarkers can be identified that will
inform the proper timing of effective therapeutics and/
or vaccines. Biomarkers can be critically important in
aiding early diagnosis and useful in the identification
of therapeutic targets for treatment, and they can be
detected at various molecular or clinical investigative
levels. Molecular measurements can include microarray
analysis, proteomics, and metabolomics, while clinical
measures can include bacteremia, viremia, hematology,
chemistries, cytokine levels, temperature, immune
response, and toxemia. For the valuable use of a
particular biomarker, it is preferred that there is a rapid
assay available to detect the biomarker and that there
is good correlation between the biomarker and disease.
The biomarker should be evident in the relevant animal
model; the disease process in the animal has to mimic
human disease and the pathogenesis should be well
understood. The biomarker must not be pathogen strain-
specific, must be identifiable, and must have similar
expression regardless of strain.
To allow for estimation of dose-response thresholds
for vaccines or therapeutics, "humanized" vaccine or
therapeutics doses and schedules must be developed and
used in animal challenge studies. This allows for the
potential extrapolation of animal efficacy data to human
efficacy. Thresholds can be determined by increasing the

-------
challenge dose until breakthrough illness is observed;
however, increasing the challenge dose is technically
challenging for many biological threat agents. An
additional means to test for vaccine thresholds is to
reduce the vaccine dose and/or schedule while still
maintaining the same challenge level. It is anticipated
that the test animal would mount an incomplete
response; this information is then used to develop
correlates of immunity. A third method is to assess host
species susceptibility so that agent virulence can be
evaluated across species and to assess the durability of
immunity. Knowledge of the species-specific immune
responses can assist in making determinations regarding
the durability of immunity. For example, rabbits and
nonhuman primates present a similar disease course,
but they exhibit different rates of disease progression
and differences in the duration of immunity. IgG
predominates in the immune response of the rabbit, and
therefore, immunity does not persist as long in rabbits as
nonhuman primates.
Dose-Response: Economics and Public Policy
(or, the value of risk)
Martin Meltzer, Ph.D.
CDC
From the perspective of the policy maker, dose-response
analysis is all about the risk for the endpoint of concern.
The role of modeling for policy development is to
inform about potential trade-offs; for example, what are
the side-effects of spending resources? Is it possible to
maintain zero risk or the "perfect" vaccine or drug? If
so, what are the costs and side effects? Policy decisions
such as these are faced all the time in the public health
field. For example, a public health policy decision was
made as to whether to recommend vaccines to those
who were exposed to the anthrax letters. Available data
on spore survival showed that spores could survive in
vivo perhaps up to 60 days, though in potentially small
numbers. The policy decision was to not recommend
the vaccine. It was assumed that the risk of disease was
dependent upon the duration of antibiotics, and a 60-
day course of antibiotics was recommended instead. In
retrospect however, it was found that overall adherence
to antibiotics was poor; only 44% of those prescribed
antibiotics took them for the fully recommended
duration. This poor adherence was mainly due to the
gastrointestinal side effects of the prescribed antibiotics.
One way to look at this is that the risk-versus-tradeoffs
valuation changes over time, with newly available
information or personal experience. Even though in the
beginning those exposed initially wanted the antibiotics,
this original desire was modified by the experience of the
gastrointestinal side effects.
Pre-exposure smallpox vaccination is another area
where risk-benefit tradeoffs pose difficult challenges for
public health policy makers. Results from a survey of
the general public show that approximately 61% of the
public would desire and accept smallpox vaccination
if it were offered. The risk of smallpox is a function of
the number initially infected, the probability of release,
probability of contact, probability of transmission, and
vaccine effectiveness. The serious vaccine adverse
effects are a function of the probability of side effects.
However, it is only when the risk of smallpox is greater
than zero that pre-exposure vaccinations should be
considered (Figure 9). It was found that with a 1:10 risk
chance of 1,000 smallpox cases in a potentially exposed
population of 280,000,000 people, an individual would
have a greater risk of vaccine related adverse effects than
risk of contracting smallpox.
General populace
Risks: Smallpox vs. side-effects
(risk of side effects: 1:100,000
1,000 infected before detection)
Figure 8. Risk of smallpox release relative to
risk of smallpox infection for two different
exposed population sizes.
Dr. Meltzer concluded that the utility of using models
is not always to produce "a number"; the concepts
contained in the model are as important as the data/
numbers. It is more important to understand what
is being built and why. Models can also be useful
because they highlight data deficiencies, "how thin is
the ice" on which decisions are being made. Model
development should balance between simplification
and being simplistic. Utility is improved when one
model can be used for a limited set of questions. Utility
is also maximized when changes in results can be
demonstrated with changes in assumptions. Confidence
intervals are essential in describing outcomes and
associated probabilities. Model utility is improved when
probabilities for outputs can be described.
Workshop participant and presenter discussions were
held following each session throughout the two and a
half days. The following section summarizes the main
points according to overarching discussion topics; it is
not a complete record of all detailed discussions, nor
does it embellish, interpret, or enlarge upon matters that
were incomplete or unclear.
15

-------

-------
Group
Discussion Summaries
Microbial Dose-Response Modeling
Several general dose-response modeling concepts
were presented by workshop attendees. The best dose-
response models all have flaws - as tliev are, at best, a
simplification. The key to the dose-response assessment-
is to determine what can be done to make the models
useful and to recognize that the models do not need to be
perfect. Multiple models will most likely be necessary
to meet the challenges and decisions. The critical focus
for model selection is who is making what decision and
why; the model should be built and utilized to inform the
decision.
There are many different variations of dose-response
models, and assessors will have to select the appropriate
model for the required decision. The assumptions for
the model must be realistic and clearly understood,
and model results must be consistent even when the
assumptions are relaxed. As the model increases in
complexity, a more detailed estimate can be gained:
however, there will also be a significant increase in the
model uncertainty-. The model needs to have the "'right"
complexity; very often there are too many parameters
in the models. There must be a deliberate effort to
choose the appropriate number of parameters in the
dose-response model while avoiding confronting the
issue of forcing the data to fit. There was a continued
discussion regarding whether to let the data determine
the dose-response model selection or to modify the data
to fit the model. Overall, it was agreed that the models
should be, if possible, biologically plausible for the
specific pathogen exposure and the data should have
a statistically significant fit in the model. This point of
discussion as well as others throughout the workshop
was highlighted along with the notion that there are
opportunities to develop microbial dose-response
modeling approaches by learning and utilizing what has
been done in chemical dose-response assessments.
Mechanistic-Enough
Several discussions focused on the differences between
mechanistic and empirical dose-response models. For
termed mechanistic models, it was questioned whether
these models are "mechanistic enough." The beta-
Poisson dose-response model is termed a mechanistic
model as it is thought to be based on biological
plausibility; however, there was general consensus
that to better inform risk assessments, models need to
include mechanisms behind dynamic processes and not
just distributional processes as is described with the
beta-Poisson dose-response model. Risk assessment
requires evaluations that indicate the adequacy of a
model for the particular assessment being performed and
this, in turn, requires clarification of assumptions made
regarding processes involved in generating the data and
determining outcomes. A model where these assumptions
are obscure is insufficiently mechanistic. Furthermore,
too much mechanistic detail could burden assessments
with excessive degrees of parameter uncertainty.
Consequently, microbial dose-response models should be
just "mechanistic enough."
Thresholds
The issue of thresholds in microbial dose-response
modeling was addressed by several participants with
varying opinions. For example, some were of the opinion
that there can be no threshold when only one organism
has a probability of infection. It was questioned that
if a threshold were present in the data, would it be
discernable using the currently available dose-response
models? These single hit dose-response models may not
be acceptable for all pathogens and all endpoints. Most
agreed that the existence of a threshold would have to
be investigated on a pathogen-specific case-by-case
basis. For example, the human immunodeficiency virus
requires exposure to high viral numbers that reach the
mucosal surface; infection, however, is initiated by one
cell. In contrast, there are other microorganisms that
must act in concert to initiate infection (e.g. quorum
sensing).
Several individuals felt that the key processes or steps
of the infection cycle (invasion, infection, illness)
each have the potential for a dose threshold. Within
each of these steps, there are potential barriers that can
influence the threshold dose required to reach the next
step. The problem is that it is difficult to discriminate
between these key processes and to identify and isolate
the appropriate endpoints of invasion, infection, and/or
illness caused by a single pathogen. To address this need
for potential endpoints, generation and verification of
more mechanistic data are required.
Biomarkers
Biomarkers can be used in dose-response and potential
threshold modeling. However, it was noted that they
will most likely be disease-specific and therefore their
intended use will also be specific. The biomarkcr
of interest will be dependent on the endpoint being
modeled and what response the biomarker is designed

-------
to predict. For example, the biomarker for infection may
be different than the biomarker for early diagnosis of
disease aimed at successful treatment intervention.
Considerations of Exposure
Assumptions for Microbial Dose-
Response Modeling
Throughout the workshop, participants noted that it is
difficult to completely separate the exposure assumptions
from dose-response modeling. In addition, a thorough
understanding of the role of the environment in exposure
is necessary to better define the contexts in which a
microbe exhibits pathogenicity. For example, naturally
occurring Bacillus anthracis in the pasture is not
typically considered to be highly hazardous compared
to intentionally -released B. anthracis in a building.
Furthermore, for some microbes, the definition of
pathogenicity is highly dependent upon the host being
invaded. There are a number of microbes that may be
pathogens to individuals with compromised immune
systems, but may not be pathogenic to those with
competent immune systems.
Assessors also need to be attentive to the actual
reported measured doses. Microbial dose-response data
are challenged by a rather limited ability to measure
extremely low doses combined with an inability to
tell whether the challenge presented was viable or not.
For example, one referenced study delivered dose was
10 oocysts - which turned out to actually be 10 +/- 4
oocysts. Another issue with the historical data sets,
specifically for inhalational exposures, is that the particle
size was typically unknown. Particle size greatly impacts
the total internal doses and thus will impact the dose-
response estimates.
Variability in concentration is another important
consideration in both sampling and dose-response
modeling. Outbreaks may occur from outliers in doses
for a given medium. A concentration that is acceptable
if homogeneous, may pose a hazard when present in
hotspots of higher concentration and areas of lower
concentration. The average concentration is the same,
but the individual exposure doses can be considerably
higher.
Furthermore, historical assessments have typically
assumed microbes are Poisson-distributed in the
environment. However, it was noted that this assumption
has not always held up in environmental and laboratory
samples. Microbial environmental samples typically
present as a skewed distribution, and it is only when the
sample is a well-mixed sample from a laboratory that
a Poisson distribution may be able to be appropriately
assumed. However, it was argued that there is published
Cryptosporidium, data obtained from sampling of a
pristine water body that conclusively demonstrated that
environmental samples are Poisson distributed.
Pooling Microbial Dose-Response Data
The workshop attendees also considered the
appropriateness of pooling microbial dose-response data.
The first question that was asked during the discussion
was "why" pool the data; some participants felt that data
pooling can be very "tricky" and may not be appropriate
as it will raise red flags for most decision makers. The
example of pooling various experimental animal species
dose-response data was considered. One of the biggest
obstacles stated with this example is how to reflect and
account for the potential differences between the species
in the various parameters. It may be appropriate to pool
data between species, but how is this type of analysis
communicated clearly to the decision makers with
accompanying assumptions. Ultimately, pooling data
could allow more information to be gathered together to
make stronger inferences if differences in data sets were
accurately reflected and adjustments were made.
Microbial Dose-Response Modeling
Extrapolations
Throughout the workshop, there were many discussions
as to the utility of the dose-response models and the
need to extrapolate data. Almost always, the decisions
made from dose-response modeling efforts involve
extrapolated data. Three types of extrapolations were
discussed: animal-to-human, high-to-low doses, and
healthy-to-sensitive subpopulations.
Animal-to-Human Extrapolations
There was a clear understanding and agreement between
the attendees that one of the greatest challenges with
microbial dose-response modeling is the very limited
availability of human dose-response data. As a result,
the majority of dose-response estimates rise from
experimental animal studies. In the best case scenario,
the animal models used for human exposures should
meet the following assumptions:
1)	The disease is caused by the same mechanism
from the same agent w ith a comparable
progression and time course.
2)	There are similar immunological and
physiological responses, signs, and symptoms
in the animal model and human.
3)	The animal model provides the ability to
quantify information on levels of infection,
morbidity, and mortality.
To account for the uncertainty in the animal-to-human
extrapolations, the question of the appropriateness
of scaling and/or uncertainty factors was raised.

-------
Uncertainty factors are widely used in chemical dose-
response modeling and could be potentially valuable
for extrapolating microbial data to estimate human
exposures. It was noted that magnitude and degree of
application of these uncertainty and/or scaling factors
should depend on the decision to be made, the specific
situation, and the endpoint of concern.
Physiological-Based Modeling
To more accurately decrease the uncertainty from the
animal-to-human extrapolations of dose-response data,
assessors can utilize species-specific physiological-based
models. Physiological-based models are a relatively
new approach for microbial risk assessment; however,
these detailed models have been widely used for many
chemical dose-response assessments. It was noted that
physiological-based models can be very data hungry and
thus very expensive to advance. These complex models
need data for many different parameters as there are
many physiological interactions between the host and
pathogen that could be modeled.
For the important pathogens of concern, there is
significant interest and potential utility in developing
pathogen-specific physiological-based models. However,
for the majority of microbial agents, there may be
value in identifying commonalities among the complex
systems that are part of these models. For example,
intracellular microbial pathogens could be grouped
and considered by one type of model with parameter
modifications for agent- or host-specific characteristics.
Bacillus anthracis and Francisella tularensis have
similar characteristics that may allow for generally
similar models to be used with some re-parameterization
to fit them.
Several participants questioned how these physiological-
based models can be verified. It was noted that the model
predictions can only really be fully tested in animals;
however, some of the various parameters of the models
could be verified with human in vitro studies. Another
approach considers if these advanced models can be
predictive for - and then tested in - other experimental
animal species; if the model can accurately extrapolate
responses from animal-to-animal, then the model should
be able to then predict more accurately and extrapolate
from animal-to-human.
High-to-Low-Dos e Extrapolations
Another challenge with modeling microbial data from
dose-response studies is that most historical studies
administered extremely high doses to achieve effects,
and therefore, the data require extrapolation from high-
to-low doses to predict potential human responses at
low doses. There can be large orders of magnitude
differences in dose-response curve estimates in the
resulting low dose extrapolations depending on the type
and amount of data being modeled. It was noted that the
interest is not necessarily with the low-doses; the focus
is really about low responses and probabilities. However,
studies are not typically conducted in the very low
probability area of the dose-response curves, therefore,
most studies, at best, focus on the "middle" portion of
the data such as the lethal dose that caused death in 50%
of the test population (LD<0).
Participants also noted that while many organizations
have indicated an interest in primarily the low dose
region of the dose-response curve, the Department of
Defense also has an interest in mid to high level doses
as well as the low dose responses. Risk is characterized
on a sliding scale from negligible to catastrophic in
recognition of acceptable losses and of the potential for
mission importance to overcome adversity to risk. On the
other hand, EPA has the issue of determining low dose-
response relationships that will be applied to chronic
low dose exposures (i.e., multiple doses) for remediation
goals applicable to re-occupancy scenarios.
Healthy-to-Sensitive Subpopulations
Extrapolations
There was a great deal of discussion regarding sensitive
subpopulations among the workshop participants. It was
questioned whether the focus of dose-response modeling
should estimate sensitive subpopulations or is it adequate
to assess the risk for the majority of the population. Most
models assume a homogenous human population and
generally do not account for disease impact on sensitive
subpopulations. For example, Listeria outbreaks in
Europe have demonstrated that current dose-response
models are underestimating the hazard to those 60 years
and older. There was a general consensus that it may be
appropriate to evaluate different dose-response models
based on sensitive subpopulations as well as the majority
'"healthy" population.
The discussion next focused on how best to extrapolate
the dose-response data from "healthy" individuals and/
or experimental animal species and then apply the data
to the larger population to account for the presence
of potentially sensitive subpopulations. In chemical
risk assessment, the approach lias been to model the
"healthy" populations first and then to use uncertainty
factors to account for the large population including the
sensitive subpopulations. However, it was recognized
that one study will most likely not be adequate; it will be
knowledge gained from multiple dose-response studies
and outbreak data as was mentioned with the example of
the Listeria outbreaks in Europe. Outbreak data can be
particularly helpful for comparisons of the response in
the healthy populations with the response in potentially
sensitive subpopulation (e.g., children, elderly).

-------
Communicating Microbial Dose-
Response Modeling Results
Discussions also centered on effective communication
and interpretation of dose-response modeling results
to improve microbial risk assessment and mitigation
practices. The participants acknowledged the importance
of planning for communicating modeling results and
the uncertainties associated with these results. It was
noted that the relative effort that was typically spent in
building models did not include enough planning for
communication of model characteristics and results.
Attendees suggested that it was best to combine the risk
communication process with the risk assessment/risk
management process early in the planning and to do so
often during the entire process.
Communication of modeling results and risk should
be done in a manner that supports policymakers in
their decision making approach. It is also important
for the decision makers to try and bridge the gap by-
understanding the underlying science. Dose-response
modelers need to communicate the assumptions,
strengths and weaknesses of their models up front so
that risk managers can interpret and apply the models
correctly. Additionally, it is important to express
uncertainties in the numbers generated and include
confidence limits to explain confidence and probability
of illness. It should also be noted that for more
meaningful results, it is important to look at the total
distribution. The confidence in the data at any given
point depends on the overall meaning and the confidence
associated with that particular point in the distribution.
Risk communication approaches need to be tailored
to the respective audiences and accompanied by well
designed translation functions for specific audiences.
For example, when communicating to the large public,
there is not a likely difference in public perception of
a 45% versus 55% chance of getting ill. In fact, there
is no single "threshold" of public concern that should
be considered to be present. The percent of concern
changes every time and is highly dependent upon the
specific circumstances and potential consequences.
The percentage is not the key to public concern; it
is the public's understanding and acceptance of risk
greater than 0%, and the required learning curve for
the understanding of novel public health threats by the
public.
Microbial Risk Assessment Standard
Terminology
Several participants noted that developing standardized
microbial risk assessment terminology would be very
valuable and would facilitate successful collaborations
across disciplines. Communicating methods and results
between the disciplines has been difficult at times
as the various disciplines sometimes apply different
terms to define the same approach or the same term
to define different approaches. For example, the term
"mechanistic" was used to describe both empirical dose-
response models and physiological-based models even
though these methods are two very different approaches
to modeling. Others disagreed with developing
standardized terminology and felt that it was sufficient
to emphasize clear communication and associated
definitions while presenting and discussing work efforts.
A new term for microbial dose-response modeling,
the lowest observable tolerable environment level
(LOTEL), which is conceptually similar to a NOAJEL
or LOAEL type measure, was also discussed. The
basis for considering this new term is to describe the
lowest "tolerable" dose that can be identified for the
endpoint and receptor of concern. From a toxicological
perspective, NOAELs and LOAJELs have been
successfully used in chemical risk assessment. However,
given the current microbial dose-response models in use,
it may not be possible to identify how "tolerable" might
be defined. Furthermore, it is critical not to intertwine
the actual science data with the science policy and
perception. Participants agreed that care should be taken
prior to usage and acceptance of a new term.
Application of Microbial Dose-Response
Modeling
One goal of the workshop was to address how to use
dose-response data to support the derivation of risk-
based remediation goals following the release of a
biological agent.
There were discussions on developing risk-based
goals that provide direction on the selection of steps to
minimize risk for post-event and re-occupancy decisions,
on the use of antibiotics post-exposure, and oil providing
an alternative to a zero or no-growth cleanup goal. The
discussion about remediation and re-occupancy goals
was limited to the cleanup of aerosolized B. cmthracis
spores.
The following questions were used to guide the
discussion:
•	What endpoint is sufficient for remediation and
re-occupancy?
•	Can a cleanup goal of zero viability actually
be achieved with current decontamination
technologies?
•	Should cleanup be to the background level of
the agent in the environment?
•	Is there a dose of B. anthracis spores that an
immuno-competent individual can tolerate
without advancing to disease?

-------
•	How does the detection limit of the analytical
methods capabilities affect achievement of the
cleanup goal?
•	Should the extent of cleanup maximally
achievable with employment of engineering
controls with the best available technology be
considered in the derivation of a cleanup goal
or should only the health risk of exposure be the
driver'
Definitive answers to most of these questions require
more research, but some answers in the interim will be
based on the scenario at hand. Attendees gave insights on
how response and remedial actions have been employed
with other agents and other environmental media. In
general, the control of pathogens in drinking water has
been through the use of treatment or technologically
based standards. Health-care facilities utilize infection
control practices by focusing 011 blocking transmission
with engineering controls to minimize risk of exposure to
pathogens. Similarly, the food industry has implemented
the Hazard Analysis and Critical Control Point (HACCP)
system as an effective and rational means of assuring
food safety from harvest to consumption; preventing
problems from occurring is the paramount goal
underlying any HACCP system.

-------

-------
Conclusions
and
Future
Steps
This report was prepared as a summary of the
presentations and discussions held at the EPAI
CDC State-of-the-Science for the Determination
and Application of Dose-Response Relationships in
Microbial Risk Assessment Workshop, April 21-23.
2009. Participants of the conference shared knowledge,
explored differing opinions, and expanded overall
understanding in MR A dose-response relationships.
Because of the diversity of attendees' disciplines,
the different inputs and decision making required to
support each organization's mission, and the limited
timeframe, the primary goal of the conference was
to share knowledge, explore differing opinions, and
expand overall understanding in MR A dose-response
relationships. The report captures the main points and
highlights of the meeting, but does not embellish,
interpret, or enlarge upon matters that were incomplete
or unclear.
As a follow on to these discussions, a third MRA
conference/workshop is being planned for 2011. This
conference will be held to focus on the exposure
assessments of MRA.

-------

-------
Bibliography
Akaike H. 1973. Information theory and an extension of
the maximum likelihood principle. In: Petrov
BN, Csaki E, cds. Proceedings of the 2nd
International symposium on information theory;
September 1971; Tsahkadsor. Armenia, USSR.
Budapest (Hungary): Akademiai Kiado. pp.
267-281.
Bartrand T, Weir M, Haas C. 2008. Dose-response
models for inhalation of Bacillus anthracis
spores: Interspecies comparisons. Risk
Analysis, 28(4): 1115-1124.
Bracliman PS, Kaufman AF, Dalldorf FG. 1966.
Industrial inhalation anthrax. Bacleriol Rev.
30(3):646-59.
Catreg Software for Categorical Regression Analysis
CatReg 2009 R version, www.cpa.gov/ncea/
catreg. Accessed 1/14/2010.
Chappell CL, Okhuvsen PC, Langer-Curry R, Widmer
G, Akiyoshi DE, Tanriverdi S, T/ipori S.
2006. Cryptosporidium hominis: experimental
challenge of healthy adults. Am J Trop Med
Hyg75(5):851-857.
GlassmanHN. 1966. Discussion. BateriolRev.
30(3):657-659.
Glomski IJ, Piris-Gimenez A, Huerre M, Mock M,
Goossens PL. 2007.
Primary involvement of pharynx and peyer's patch
in inhalational and intestinal anthrax. PLoS
Patliog. 3(6): e76.
Haas CN, Rose JB, Gerba CP. 1999. Quantitative
Microbial Risk Assessment. New York
(NY) John Wiley & Sons, Inc.
Holty JE, Bravata DM, Liu H, Olshen RA, McDonald
KM, Owens DK. 2006. Systematic
review: a century of inhalational anthrax cases from
1900 to 2005. Ann Intern Med. 144(4):270-80.
Meltzer M. 2003. Risks and Benefits of Preexposure and
Postexposure Smallpox Vaccination. Emerging
Infect Dis 9(11): 11363-1370.
Mesclson M, Guillcmin J, Hugh-Jones M, Langmuir A,
Popova I, Shelokov A, Yampolskaya O. 1994.
The Sverdlovsk Anthrax Outbreak of 1979.
Science. 266(5188): 1202-1208.
Tumbull PC. 2002. Introduction: anthrax history, disease
and ecology. Curr
Top Microbiol Immunol. 271: 1-19.
U.S. EPA. 2000. Benchmark dose technical
guidance document [external review draft],
EPA/630/R-00/001. U.S. EPA, Risk Assessment
Forum. Washington, DC. http://oaspub.epa.gov/
eims/eimscomm.getfile?p_downloadJd=4727.
Accessed 02/25/2010.
U.S. EPA. 2009. Benchmark dose software (BMDS)
version2.1.1 [build: 11/06/2009], www.epa.
gov/ncea/bmds Retrieved 12/19/2009.

-------

-------
Appendix A: Workshop Agenda
Day 1: Tuesday, April 21, 2009
8:00 - 8:30 a.m. Registration
8:30 - 9:00 a.m. Welcome - Cynthia Sonich-Mullin, U.S. Environmental Protection Agency (EPA)
9:00 - 9:30 a.m. Keynote Address: "MRADose-Response Challenges"
Cynthia Chappell, University of Texas School of Public Health\
9:30 - 9:45 a.m. Participant Feedback and Discussion
9:45 - 10:05 a.m. Break
'10:05 - 11:30 a.m. Federal Mission Needs for Dose-Response
Tonya Nichols, EPA
Michael Bell, Centers for Disease Control and Prevention (CDC)
David Oryang, Food and Drug Administration
Jan ell Kcmse, U.S. Department of Agriculture
11:30 - 12:00 p.m. Participant Feedback and Discussion
12:00 - 1:00 p.m. Lunch
1:00 - 1:30 p.m. Dose-Response Extrapolations
1.	How is uncertainty and variability addressed in extrapolating dose-response data (e.g.,
extrapolating across host species, exposure levels, routes of exposure, durations of exposures,
pathogen strains or species, endpoints, and/or sensitive populations)?
2.	Is it appropriate to group studies, animal models or host species, and/or pathogen strains or
species in dose-response modeling of multiple data sets?
.Margaret Coleman, Syracuse Research Corporation
Charles Haas, Drexel University
Thomas Whalen, Georgia State University
Mary Alice Smith, University of Georgia
1:30 - 2:30 p.m. Participant Feedback and Discussion
2:30 - 3:00 p.m. Break
3:00 -3:30 p.m. Physiological-Based Modeling
1.	What overall assumptions arc necessary for valuable physiological models to predict human
consequences?
2.	What is the minimum data set required (i.e., what level of detail needs to be modeled for
acceptable human predictions (e.g., whole species models, organ-specific models, and/or cellular
or toxin activity models)?
Sarah Taft, EPA
Jeff Gearharl, The Henry M. Jackson Foundation for the Advancement of Military Medicine
Michael Lumpkin, Syracuse Research Corporation
3:30 - 4:30 p.m. Participant Feedback and Discussion
4:30 - 4:45 p.m. Preview of Day 2

-------
Day 2: Wednesday, April 22, 2009
8:30 - 8:45 a.m. Day 2 Opening Remarks
8:45 - 9:45 a.m. Dose-Response Method Comparisons:
Classical, Bayesian, Epidemiology, and Benchmark Dose Modeling
1.	Is the dose-response statistical method utilized empirical or mechanistic?
2.	Is the method applicable for low-dose extrapolations?
3.	Can the method accommodate data pooling and/or the use of correction factors?
4.	How is the calculated dose-response relationship verified and validated?
5.	How is model uncertainty adjusted for and communicated to risk managers?
Classical	Tim Bartrand, Clancy Environmental Consultants
Bayesian Jade Mitch el, Blackwood, Drexel University
Epidemiology Modeling	Laurie Waisel, Concurrent Technologies Corporation
Benchmark Dose Modeling	Jeff Gift, EPA
9:45 - 10:05 a.m. Break
10:05 - 11:45 a.m. Participant Feedback and Discussion
11:45 - 12:00 p.m. Closing Comments from Stimulus Presenters
12:00 - 1:00 p.m. Lunch
1:00 - 1:30 p.m. Stimulation Activity: Going Beyond the Dose-Response Curves!
1:30 - 3:45 p.m. Breakout Activity (4 teams)
3:45 - 4:45 p.m. Teams Report Back
4:45 - 5:00 p.m. Discussion on Playback Reports
Day 3; Thursday, April 23, 2009
8:30 - 8:45 a.m. Day 3 Opening Remarks
8:45 -9:15 a.m. Dose-Response Applications for Vaccines and Therapeutics
1.	How are biomarkers utilized in dose-response modeling of infection and/or disease?
2.	Can dose-response thresholds be estimated for vaccines and/or therapeutics?
Conrad Quinn, CDC
Louise Pitt, U.S. Army Medical Research Institute of Infectious Diseases
Martin Meltzer, CDC
9:15 - 10:30 a.m. Participant Feedback and Discussion
10:30 - 11:00 a.m. Next Steps
11:00 a.m.	Adjourn

-------
Appendix B: List of Participants
George Andrews
U.S. Navy
Naval Surface Warfare Center Dahlgren Division
Ken Andrews
High Impact Facilitation
Matthew Ardulno
Centers for Disease Control and Prevention
National Center for Preparedness, Detection, and Control
of Infectious Diseases
Division of Healthcare Quality Promotion
Prasith Baccanr
U.S. Department of Health and Human Services
Office of the Assistant Secretaire for Preparedness and
Response
Biomedical Advanced Research and Development
Authority
Mansoor Baloch
Centers for Disease Control and Prevention
National Center for Environmental Health
Environmental Health Sendees Branch
Tim Bart rand
Clancy Environmental Consultants
Irwin Baumet
U.S. Environmental Protection Agency
Office of Research and Development
National Center for Environmental Research
National Homeland Security Research Center
Monte Bawden
U.S. Food and Drug Administration
Division of Anti-Infective and Ophthalmology Products
Michael Bell
Centers for Disease Control and Prevention
Coordinating Center for Infectious Diseases
National Center for Preparedness, Detection, and Control
of Infectious Diseases
Marie-Claude Besner
U.S. Environmental Protection Agency
Office of Water
Office of Ground Water and Drinking Water
William Burrows
U.S. Army
Center for Health Promotion and Preventive Medicine
Cynthia Chappell
The University of Texas School of Public Health
Center for Infectious Diseases
Margaret Coleman
Formerly associated with Syracuse Research Center
John Decker
Centers for Disease Control and Prevention
National Institute for Occupational Safety and Health
Office of Emergency Preparedness and Response
Lisa Delaney
Centers for Disease Control and Prevention
National Institute for Occupational Safety and Health
Pamela Diaz
Centers for Disease Control and Prevention
Division of Bioterrorism Preparedness and Response
Jennifer Elin Cole
Frontline Healthcare Workers Safety Foundation, Ltd.
James Englehardt
University of Miami
Department of Civil, Architectural, and Environmental
Engineering
Laura Fra/ier
Centers for Disease Control and Prevention
Agency for Toxic Substances and Disease Registry
National Center for Environmental Health
Jeff Gearhart
The Henry M. Jackson Foundation for the Advancement
of Military Medicine
Jeff Gift
U.S. Environmental Protection Agency
Office of Research and Development
National Center for Environmental Assessment-
Bradford Gutting
U.S. Navy
Naval Surface Warfare Center Dahlgren Division

-------
Charles Haas
Drexel University
Civil, Architectural, & Environmental Engineering
Stephanie Hines
Battel le Memorial Institute
Alex Hoffmastcr
Centers for Disease Control and Prevention
Bacterial Zoonoses Branch
Division of Foodborne, Bacterial and Mycotic Diseases
National Center for Zoonotic, Vector-Borne, and Enteric
Diseases
Janell Kause
U.S. Department of Agriculture
Food Safety and Inspection Service
Risk Assessment Division
James Koopnian
University of Michigan
School of Public Health
Department of Epidemiololgy
Michael Kuhlman
National Biodefense Analysis and Counternieasures
Center
National Biological Threat Characterization Center
Heejeong Latimer
U.S. Department of Agriculture
Food Safety and Inspection Sendee
Office of Public Health and Science
Robyn Lee
U.S. Army
Center for Health Promotion and Preventive Medicine
Michael Lumpkin
Syracuse Resource Corporation
Environmental Science Center
Dennis Lye
U.S. Environmental Protection Agency
Office of Research and Development
National Exposure Research Laboratory
Debbie Massa
Frontline Foundation
Deborah McKean
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
Richard McNally
SAIC contractor
U.S. Department of Health and Human Sen. ices
Office of the Assistant Secretary for Preparedness and
Response
Biomedical Advanced Research and Development
Authority
Martin Meltzer
Centers for Disease Control and Prevention
Division of Emerging Infections and Surveillance
Sen ices
Jade Mitchell-Blackwood
Drexel University
Civil, Architectural, & Environmental Engineering
Christine Moe
Emory University
Rollins School of Public Health
Hubert Department of Global Health
Stephen Morse
Centers for Disease Control and Prevention
Division of Bioterrorism Preparedness and Response
Alison Myska
U.S. Department of Defense
Defense Threat Reduction Agency-
Tonya Nichols
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
David Oryang
U.S. Food and Drug Administration
Center for Food Safety7 and Applied Nutrition
Duanc Pierson
National Aeronautics and Space Administration
Johnson Space Center
Habitability and Environmental Factors Divison
Louise Pitt
U.S. Army Medical Research Institute of Infectious
Diseases
Center for Aerobiological Sciences
Regis Pouillot
U.S. Food and Drug Administration
Center for Food Safety and Nutrition

-------
Conrad Quinn
Centers for Disease Control and Prevention
MPIR Laboratory
Joan Rose
Michigan State University
Department of Fisheries and Wildlife
William Ross
Health Canada
Sean Shadomy
Centers for Disease Control and Prevention
Coordinating Center for Infectious Diseases
National Center for Zoonotic, Vector-Borne, and Enteric
Diseases
Bacterial Zoonoses Branch
Brandolyn Thran
U.S. Army Center for Health Promotion and Preventive
Medicine
Environmental Health Risk Assessment Program
Lesley Vazquez-Coriano
U.S. Environmental Protection Agency
Office of Science and Technology
Office of Water
Health and Ecological Criteria Division
Laurie Waisel
Concurrent Technologies Corporation
Thomas Wfaalen
Georgia State University
Sanjiv Shah
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
Erin Silvestri
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
Mary Alice Smith
University of Georgia
College of Public Health
Environmental Health Science
Marylynn Yates
University of California, Riverside
Department of Environmental Sciences
Max Zarate-Bermudez
Centers for Disease Control and Prevention National
Center for Environmental Health
Contractor Sunnort
Maria Smith
The Scientific Consulting Group, Inc.
Theresa Smith
Centers for Disease Control and Prevention
Coordinating Center for Infectious Diseases
Curtis Snook
U.S. Environmental Protection Agency
National Decontaminaton Team
Cynthia Sonich-Mullin
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center
Sarah Taft
U.S. Environmental Protection Agency
Office of Research and Development
National Homeland Security Research Center

-------

-------
Appendix
C: Slide Presentations
Coleman:
To supplement the Coleman summary in Session 2, the
following is a more detailed description of the specific
pathogen examples presented:
Anthrax: Aerosol challenge of rodents with Bacillus
anthracis may be misleading for primates with
different patterns of deposition due to anatomical and
physiological differences in respiratory systems. In vivo
images were generated in mice challenged by different
routes with high doses of a bioluminescent nontoxigenic
capsulated strain of B. anthracis (aerosol 108; intranasal
105; intratracheal 105; intravenous: 106 to 107; images
from Glomski et al, 2007). Spores depositing in the
turbinates of rodents infect nasal cav ity and throat
tissues before the lung, whereas spore deposition in
human respiratory tract system is deep in the lung due
to differences in anatomy and physiology-. Rodents also
swallow inhaled particles, and resultant gastrointestinal
pathologyr in mice may be poor predictor for human
effects by inhalation route. Therefore, rodents may not
be reliable predictive models for inhalation anthrax and
oilier human respiratory diseases. If rodent models are
to be useful for predicting human effects, deposition
and clearance models are needed for scaling doses and
translating system level knowledge.
Similarly, the relevance of mice to humans for oral
and dermal challenges with B. anthracis merits further
investigation and analysis. For gastrointestinal anthrax,
knowledge is so sparse that this demonstration of tropism
to Peycrs' patches in mice is relevant (intragastric
catheters or feeding needles at 108), as are conflicting
results from other animals resistant to high dose
challenges (guinea pig, rabbits, rhesus at 108 spores;
dog, guinea pig, sheep at -105). Future mechanistic
models may explain these inconsistencies and provide
more robust decision support for preparedness planning.
For dermal anthrax, systemic involvement in rodents
from sub cutaneous challenge (injection into dermis of
ear (500 or 10,000 spores)) is atypical of human cases
of cutaneous anthrax, largely localized infections via
damaged skin.
Historically, human cases of gastrointestinal and
cutaneous anthrax are associated with animal outbreaks
in hyper-endemic regions of the world. Epidemiologic
investigations report human cases occur in proportion to
animal cases. Approximately one human gastrointestinal
case per 30-60 animal cases and approximately one
human cutaneous case per ten animal cases were
associated with consuming, preparing, or butchering
meat from contaminated carcasses during epidemics of
anthrax (Turnbull, 2002). The paucity of human cases in
the US. despite outbreaks in livestock and wildlife, may
be due to more effective interventions (e.g., vaccination,
protective equipment) and inspection procedures to keep
diseased animals out of the US food supply.
Sound dose-response assessments must incorporate
knowledge of the mode/mechanism of anthrax in animals
and humans for robust extrapolations.
Salmonellosis: Outbreaks in peanut butter provides a
great example for discussion of extrapolation because
available animal and human data alone do not directly
address susceptibility of children, 3-13-fold more
susceptible than older age groups as reported in a 2003
FoodNet study.
Consider a family of dose-response curves from murine
studies conducted in the 1950s and 1960s by Bohnhoff
and colleagues. Fifty percent of normal healthy animals
are infected at approximately a million Salmonella
enteritidis cells. The dose-response curve is left-
shifted five orders of magnitude to an ID50 less than
ten cells when animals are rendered more susceptible
by treatment with antibiotics that disrupt the normal
protective effect of the indigenous microbiota of the
gastrointestinal tract. Susceptibility in treated mice
returns to normal as the interval between antibiotic
treatment and pathogen challenge increases to five days
and the protective effect of the indigenous microbiota is
restored.
Direct use of the murine curves for humans is not
recommended due to the systemic disease pattern in
mice, atypical of human disease. However, the existence
of rich human dose-response datasets for salmonellosis
and a species-common mechanism, antibiotic disruption
of the protective effect of the indigenous gut microbiota,
permits scaling of mouse and human dose-response
relationships to reflect variability in susceptibility within
and between hosts.
Much of the work with the human salmonellosis
datasets was published with my collaborator Ham
Marks at USDA. Nearly 400 human volunteers were
challenged with 13 strains of Salmonella (McCullough
& Eisclc, 1951). Accounting for strain differences
by ANOVA, nine strains can be pooled, but not with
the four Salmonella pullorum strains with a different

-------
mode of action consistent with a statistically significant
threshold dose-response relationship (Coleman and
Marks, 2000). ANOVA models provided significant fit
for multiple empirical models, with different low-dose
behaviors. Both models fit the data well in the observed
region, but differed by >75 order of magnitude when
extrapolated to the dose of a single Salmonella cell.
Model uncertainty and strain variability arc obviously
significant for salmonellosis. The Weibull model for
human salmonellosis was scaled to the murine family
of curves to generate a family of curves that represent
human populations of increasing susceptibility based
on the protective effects of the indigenous microbiota
common to mice and humans. Strain variability can
also be described for the most susceptible host, with
inflection points ranging from 1 or 1000 salmonella
cells that could cause illness in susceptible human hosts,
based on the available murine and human data.
To select models from these families of empirical
dose-response curves that are representative of infants,
children and adults, mechanistic knowledge and
models, as well as target in vitro or in vivo research, are
necessary to further illuminate the key events in host-
pathogen interactions for appropriate scaling.
***Note to Reviewers: The following section contains
presentation slides that were approved by the presenters
for inclusion in this document.

-------
Nichols
Microbial Risk Assessment
for the Development of Cleanup Goals
Tonya Nichols, Ph.D., Acting Associate Director
Threat and Consequence Assessment Division
National Homeland Security Research Center
U.S. Environmental Protection Agency
Cincinnati, Ohio, U.S.A
EPA's Mission:
To protect human health and to safeguard
the natural environment - air, water, and
land - upon which all life depends
5-EPA


The Problems We Face

• Protecting against environmental contamination

• Determining that a release has occurred

• Containing contamination

• Mitigating impacts

• Assessing and communicating risk

• Decontaminating impacted areas

• Disposing of contaminated materials

...EPA conducts risk assessment

to inform risk management decisions
—

v>ERA
Risk Assessment -> Risk Management

— U.S. EPA Cleanup Goals

Superfund program

• Preliminary Remediation Goals (Soil and Water)

Office of Water

• Health Advisories

- Maximum Contaminant Levels

Office of Air Quality Planning and Standards

• Reference concentrations

¦ Inhalation unit risk

e'SSr^eSTTcente,
5>EPA
Risk Assessment
Risk Management
Decision
vvEPA	Risk Assessment -> Risk Management
'	U.S. EPA Cleanup Goals
Exposure x Toxicity = Chemical Risk
Exposure x Pathogencity = Biological Risk
Target	Target
Cone, x Intake x Pathogenicity = Risk
Target Risk
Risk-based Goal = Target Cone. -
Intake x Pathogenicity
35

-------
^|=PA	Risk Assessment -> Risk Management
jfflSSifww	U.S. EPA Cleanup Goals
Target Concentration^
J
Estimated Inhaled Dose
J
Environmental Air Concentration
J
Environmental Surface Concentration
J
Surface Wipe Concentration
v»EPA
Target Concentration
How do we dervie the target concentrations for biologicals?
What are the minimum data requirements ?
•	pathogens vs surrogates
•	exposure route
•	dosing regimen
•	animal model
•	correlate of disease
Examples :
Minimum Data Requirements for Registering a Chemical Pesticide, CFR 158
New Drug and Biological Drug Products; Evidence Needed to Demonstrate
Effectiveness of New Drugs When Human Efficacy Studies Are Not Ethical
or Feasible, 21 CFR 601
SERA
Target Concentration
Animal Exposures
•	High Dose
•	Low Dose
•	Chronic Exposure
•	Challenge Dose
Evaluation of Biological
Parameters of Exposure
in vitro Studies
\ /
V /
Met aD at a
Analysis
•	Telemetry - clinical symptoms
•	Bactremia
•	Toxemia
•	Inflammatory cytokines
•	Antibodies
•	Histology
•	Multiple animal species
•	Multiple microbial strains
Physiological Modeling
•	Portal of Entry
•	Deposition
•	Replication
•	Host Immune Response
•	Clearance
•Translocation
Minimum Data Requirements ?
vvERA
Risk Assessment
Clean-up Decision Making Chaffenges for Biological Agents
•	Limited data available on which to base necessary immediate decisions
Unique agents
Unique exposure durations
Unique exposure situations/sites
•	No consensus-based microbial risk assessment methodology
Little infectivity/dose response data for agents of interest
Few transmission models
•	Communication and Transfer
Clear and understandable guidance
AEPA
Target Concentration
Questions:
How do we approach a NOAEL / LOAEL for microbganisms? LOTEL?
What dose - response models do we use and why?
How do we design in vivo and in vitro studies to better inform physiological modeling?
How do we extrapolate animal study data to humans?
How do we account for uncertainty and variability ?
A EPA


Risk Management

Considerations Impacting Cleanup

• Risk reduction

• R e gu I atory m an dat es

• Long-term effectiveness

• Reduction of hazard through treatment

• Short-term effectiveness

• Implementability

• Cost

• State acceptance

• Community acceptance


36

-------
Haas
Two Guiding Principles
"Let the Data Speak"
Use of D-R Models that can be
derived from plausible
mechanistic concepts
Chuck Haas, Ph.D.
How is uncertainty and variability
addressed in extrapolating dose-
response data (e.g. extrapolating
across host species, exposure levels,
routes of exposure, durations of
exposures, pathogen strains or
species, endpoints, and/or sensitive
populations)?
•	Intrinsic maximum likelihood fitting
accounts for experimental variability and
ascertains if other sources of variability
are present
•	Parametric uncertainty can be
determined
•	Tests for pooling between strains,
species, hosts, sensitive subpops have
all been made (we have examples of all
of these in our work)
Frequent Fallacious Understanding
of Dose Response Curves





w
o.»
f.
—J
M
//7

1=
M
BJ
J
—	J
J
—	3
| ...
• 0601
OJKao:
f
—	i
j
—	i
140
liXXflJ lOM-OO ;M!-M ttGt-Ot
Data
l.ot w i«-oi iDt«oc use.®* ;«-&•
Is it appropriate to group
studies, animal models or host
species, and/or pathogen
strains or species in dose-
response modeling of multiple
data sets?
37

-------
•	Yes if the data justifies it
•	Dose metric generally should be
ingested/inhaled #
•	In progress work - looking at in vivo
pathogen dynamics to assess body
burden as a metric
- Future - body burden, AUC, etc. as metrics
Dose Response Models Are
Consistent with Human Outbreak
Data (some examples)
•	Legionella pneumophila
•	Salmonella typhimurium
•	Giardia lamblia
•	E. coli 0157: H 7
•	Cryptosporidium parvum
•	Bacillus anthracis (Sverdlovsk)
•	SARS
38

-------
Taft
Physiologically-based Modeling
Sarah Taji,. PhD
Threat Consequence Assessment Division
National Homeland Security Research Center
U.S. Environmental Protection Agency
Cincinnati, OH
SEF	Disease is caused by the same mechanism of pathogenicity
>	Similar physiological and immunological responses to the agent
>	Comparable quantitative relationships between infectivity, morbidity,
and mortality
39

-------
Microbial Risk Assessment for the Development
of Cleanup Goals
¦BUB
i	i
Dose-Response H Exposure
Assessment	Assessment
~
Risk Management Decision:
Cleanup Goal
\>EFA
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Clean?
• No consensus based method for animal-to-human extrapolations for the
development of microbial cleanup goals!
• Can an approach similar to that applied for chemical risk assessments be used?
• Uncertainty Factors (UF)
NOAEL
UF
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Clean?
How can we decrease uncertainty in animal-to-human
extrapolations for biothreat agents?
Physiological factors affecting the
ability of the pathogen to reach
the target tissues
Effects at the target tissues
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Clean?
Dose
•	Lung deposition dose
•	Clearance from alveoli
•	Germination
•	Lymph node dose
•	Dose in circulation
•Toxin production
•	Replication
Response
•Toxin activity
• Increase inflammatory
responses
. ?
AERA
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Clean?
How can we decrease uncertainty in animal-to-human
extrapolations for biothreat agents?
Interspecies Differences |
I
•	Deposition Modeling
•	Physiologically-based Biokinetic
Modeling (PB/BK)
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Cban?
How can we decrease uncertainty in animal-to-human
extrapolations for biothreat agents?
Interspecies Differences
•Toxin Activity Modeling
40

-------
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: How Clean is Clean?
How can we decrease uncertainty in animal-to-human
extrapolations for biothreat agents?
Physiological factors affecting the	Effects at the target tissues
ability of the pathogen to reach
the target tissues
Dose-Response Assessment for the Development of
Cleanup Goals:
Bacillus anthracis: Physiologically-Based Biokinetic Modeling
41

-------
Oryang
DHHS/FDA/CFSAN
Background (1)
¦ FDA has a long history of managing risks, conducting
safety assessments and risk assessments for food
additives, chemicals, and microorganisms
Mission Needs for Dose Response
at FDA-CFSAN's Microbial Risk
Assessment Program
Background (2)
¦ CFSAN is moving toward a more risk analysis
based approach:
¦	Develop and use efficient means to collect,
organize, review and share information used in
regulatory decisions
¦	Prioritize activities because of limited resources
Microbial Contamination: FDA Centers
Microbial contamination is a source of concern
to several FDA Centers
¦	CFSAN: foods
¦	CVM: meat, eggs, seafood
¦¦ CDRH: medical devices (sutures)
¦	CBER: blood products and vaccines
DHHS/FDA/CFSAN	¦	DHHS/FDA/CFSAN
Center for Food Safety and Applied
Nutrition (CFSAN)
Mission:
Promoting and protecting
public health by ensuring
that our food supply is
safe, sanitary, wholesome,
and honestly labeled, and
that cosmetic products are
safe and properly labeled

Risk-based Decisions
Growing responsibilities and new challenges require
new tools and approaches
Risk assessment is a tool used
by regulatory agencies to
support decision making
for: import policies, control
strategies, inspection
programs, tolerance levels,
etc.
m
m
42

-------
Decisions...
Informed Decisions...
¦ Each day industry and government
agencies must make decisions about the
safety of foods and food products
During the past 15 years there has been a
tremendous effort both in the United States and
throughout the international food safety
community to make decisions that are:
•science-based
"risk-based
«: The public health and economic well-being
consequences of "bad" decisions can be
substantial
¦ Not deciding is not an option
¦transparent
consistent
DHHS/FDA/CFSAN
DHHS/FDA/CFSAN
Risk Assessment	I Uses for Risk Assessment
Risk Assessment is one of three components of
the risk analysis triad: Assessment,
Management, and Communication.
¦	A process to describe what we know and how
certain we are of what we know.
¦	Answers 4 key questions:
•	Whatcango wrong?	A
•	How likely is it to occur?
•	What tiro tht' consequences?	0E*
m What factors can influence it ?
43
Determination of the Dose-Response is part
of the Risk Assessment
¦ Answers 4 key questions:
I What can go wrong? -
Consumption = illness ? (Exposure -> Infection -> Illness)
¦	How likely is it to occur?
Likelihood/Frequency of adverse effect. f(dose, susc, path., etc)
¦	What are the consequences?
Severity of adverse effect: illness, death, etc)
¦	What factors can influence it?
(Pathogen, Host, and Environment factors)
Mathematically Modelled as:
1.	Exponential
2.	Beta-Poisson, Log-Normal, Log-Logistic, Extreme-Value
3.	Weibull-Gamma, Exponential-Gamma
DHHS/FDA/CFSAN
DHHS/FDA/CFSAN
Know where to look
¦	Set priorities/ allocate resources
¦	Identify steps along "farm to fork" continuum
that are "major contributors" to risk
Evaluate effectiveness of interventions
¦	Potential or equivalent control measures
¦	Proposed standards and criteria
¦	Contribution of compliance to risk
management
Inform communication/outreach
messages
¦	Determine subpopulations "at increased risk"
¦	Assess uncertainty and variability
¦	Present objective comparison of alternatives

-------
fiDA's Role irWrood Safety
¦	Every day across the country, people ec
buy groceries, and cook meals for their
Americans expect that all their food will
and FDA plays a critical role in making sure this is
true.
a FDA is responsible for the safety of the vast
range of food Americans eat; about 80 percent of
all food sold in the United States.
¦	This includes everything except for meat, poultry,
and processed egg products, which are regulated
by the U.S. Department of Agriculture (USDA).
DHHS/FDA/CFSAN
t out,
families,
be safe.
Defining the Challenge
Several factors are imposing increasing demands on
FDA's resources.. These are:
¦	Increases in the volume, variety and complexity of
imported foods.
¦	Shifting demographics.
¦	Americans are consuming more convenience foods.
¦	A greater variety of foods are eaten year round.
Also, foods that are consumed raw or with minimal
processing are often associated with foodborne
illness.
¦	The emergence of new foodborne pathogens.
DHHS/FDA/CFSAN
Global Food Supply
¦	The United States trades with over 150 countries/ territories
with products coming into over 300 U.S. ports
¦	It is increasingly important to understand changing
consumption patterns by susceptible population.
	DHHS/FDA/CFSAN	
Pathogens Newly Associated with Foodborne Illness Since the Mid-1970's
• Campylobacter jejuni
• Campylobacter jetus
• Cryptosporidium parvum
• Cyclospora cayetanesis
• Shiga toxin-producing E. coli
• Listeria monocytogenes
• No ro viruses
• Salmonella Enteritidis
• Salmonella Typhimurium DT104
• Vibrio vulnificus
• Vibrio cholerae 0139
• Yersinia enterocolitica
• Vibrio parahaemolyticus
• Enterobacter sakazakii
¦ The emergence of new foodborne pathogens requires
updated technologies that can detect the presence of
new agents in a variety of foods.
¦ Addressing these emerging hazards requires
cooperation among industry, academia, and
government to share information, establish testing
protocols, and develop dose-response data ana
relationships.
Looking Forward:
¦	Growth models: More effective estimates of exposure levels.
¦	CFSAN focus on acute, as well as transient/chronic effects.
¦	Susceptible populations - IRAC working group, Food Forum
symposium.
•	Variation in susceptibility within age groups
•	Variation in susceptibility between age groups
¦ 'Variation in fatality to hospitalization ratio
¦	Increase accessibility to data, models and information.
¦	Dose-response relations for new foodborne pathogens.
Extrapolate data acquired in animal models to humans.
¦	Web based tools for risk ranking across products and hazards
(iRISK)
¦	Development of risk prioritization framework to allocate
resources across programs on the basis of risk and other
factors.
FoodRisk.org
The online resource for food safety risk analysis
| Welcome to FoodRisk.org
># to Mjut orofeMkoruk tnvi
FoodRiik

Research
Exclusives
» fR N®. 229 (19911 |Pf«p«I«d KLEA)
onenw,	Mat. chk-ti.»->3 eiOliogr«oniC4i refer.
EcoSure 2007 Cold Temperature Database
cf. titvt »eojrr«4 in »I4 telreefttw* «;<»»« cuo» lati both «% •lUMi'.nv.nts
DHHS/FDA/CFSAN

-------
iRISK A web-based Comparative
Risk Tool
Conclusion
¦	Developed by FDA/IFT, and operationalized by RSI.
¦	Used to compare relative food safety risks across a
wide variety of chemical and microbial hazards,
foods and processes.
¦	Key feature: individual users have the ability to
securely develop risk models within the program
repository and can easily share data and models
with colleagues.
u Available thru www.foodrisk.org
¦	A key component of MRA is dose-response
modeling.
¦	Need to better define susceptible
populations for microbial hazards, and
address host susceptibility variation in the
dose-responses.
¦	Development of better process, survival,
and growth models -> better exposure
assessments,
Risk-based Decisions
Growing responsibilities and new
challenges require new tools and
approaches
Risk assessment is a tool used, by
regulatory agencies to support
decision making for: import policies,
control strategies, inspection
programs, tolerance levels, etc.
&
Each day industry and government agencies must make
decisions about the safety of foods and food products
¦	The public health and economic well-being consequences of "bad"
decisions can be substantial
¦	Not deciding is not an option
To Advance the Field of Food Safety
Risk Assessment We must continue to:
¦	Learn from our experiences
¦	Develop new ways to address
complex food safety issues
i ¦ feitSrilfivolvemeflt of multi-
discipljnary expertise
¦	Actively participate in
international activities
•Improve exposure assessment
•Improve dose-response modeling ,
•Define and characterize susceptible populations
45

-------
Waisel
Modes of Action in Low-Dose
Extrapolation
Laurie Waisel, PhD, Concurrent Technologies Corporation
Thomas Whalen, PhD, Georgia State University
Thomas Taylor, MS, PE, Centers for Disease Control
Murray Cohen, PhD, MPH, CIH, Frontline Healthcare Workers Safety Foundation
EPA-CDC Workshop on Dose-Response Relationships for Microbial Risk Assessment
Atlanta, Georgia
April 21-23, 2009



Thanks to Joey Kiernan for the artwork!
Technologies
W Corporation
Sn«
and Prevention)
Ions in this report are those of the authors and do 1
WBai -v l
Modes of Action
Not a mechanism of action
Dose-response extrapolation curve
•	Decision sciences (applied math)
Static vs. dynamic
•	Uncertainty vs. variability
Y Axis - Response
XAxis - Dose
Roadmap
Modes of Action
Low Dose-Response Extrapolation Workshop Paper
¦ Static Models
-	Horizontal Line
-	Vertical Line
-	Diagonal Line
-	Probit
Dynamic Models
* Variability and Uncertainty
-	Deterministic: Individual Differences {Variability}
-	Stochastic: Lucky Germs (Uncertainty)
-	Stochastic: Unlucky Hosts (Uncertainty)
Horizontal Line
Susceptible? Yes or No
-	Either you're susceptible or you're not.
-	If you're susceptible, then any dose is lethal
-	All susceptible individuals affected,
regardless of dose.
Street Clothes = Non-Swimm
Bathing Suit = Swimmer
Vertical Line
• LD is the same for everyone, so
LD01=LD50=LD100.
-	Dose is symbolized by depth of water
-	Resistance is symbolized by height.
-	Everyone has the same resistance.
Diagonal Line
Risk directly proportional to dose
-	Concept: Numberof targets hit is directly
proportional to the number of bullets fired
-	Example: Carcinogenic radiation
-	Number of snowballs symbolizes dose of
radiation

-------
Probit
•	Resistanceisnormallydistributed.
•	Dose-response relationship follows
cumulative normal distribution.
~	In this example, resistance is
symbolized by height.
~	In this example, the proportion of
people who will drown in a given
depth of water follows the probit
(cumulative normal) distribution
T3




/
<
/
/
/


Do,,
Multi-Parameter
• Multiple parameters determine resistance
Dynamic Models
•	Time matters.
•	Repair and reversibility.
•	Can the man bail out water faster
than it comes in?
•	How long can he keep bailing?
Deterministic: Individual Differences
•	Dose-response relationship
depends on quantifiable
individual characteristics of
human and microbe.
•	Is the virulence of the germ
bigger than your ability to
resist?
•	Whichever one is bigger will
win
Variability and Uncertainty
Variability
-	Dose-response relationship
depends on individual
differences
-	Deterministic
~ If you have all the
information, you know the
answer with certainty.
-	Example: diagonal line model
Uncertainty
-	Dose-response relationship has
an element of chance
-	Stochastic or probabilistic
~ Even if you have all the
information, you cannot
know the answer with
certainty because the
answer is partially
determined by chance.
-	Example: radioactive decay
The actual does entering the host may be uncertain depending on
the routes of exposure.
Stochastic: Lucky Germs
•	Each spore has a given
probability of germinating
(attack rate).
•	Lucky spores are the ones
that are randomly selected to
germinate.
•	Lucky germ hits the jackpot at
the one-armed bandit.
C7C2I Te.Jwriti/i/ct

47

-------
Stochastic: Unlucky Hosts
If exposed to disease, there is a
given probability that illness will
occur.
Unlucky hosts are the ones who are
exposed and get sick.
Lucky hosts are the ones who are
exposed and do not get sick.
Unlucky host loses all his money at
the one-armed bandit.
ThfoifVuiA*
Decision Science
•	Make sure who needs to make what actual real-world
decision, and why they need to make it.
•	Integrate theoretical and empirical, qualitative and
quantitative thinking to make that decision as well-
informed as possible.
•	Decide which model to use.
-	Passes within the error bars of the data?
-	Make biological sense?
•	Test selected model against empirical data.

-------
Meltzer
Dose-response:
Economics and public policy
(or, the value of risk)
Martin I. Meltzer, MS, Ph.D.
Senior Health Economist and Distinguished Consultant
DEISS/NCPDCID
qzm4@cdc.gov
Policy makers and dose-response
>Dose-response all about risk
>The "uh-oh" moment"
>Models inform about trade offs
>Zero risk, or "perfect" vaccine/ drug, BUT
>What about cost?
>Side effects?
> Practical - can it be achieved?
gig
Disclaimers
> The findings and conclusions in this
presentation are those of the authors and do
not necessarily represent the views of the
Centers for Disease Control and Prevention
0^3
Policy decisions: Example 1: The anthrax
letters: To recommend vaccine or not?
Figure 1: Effect of different duration of anthrax post-
exposure prophylaxis + spore survival data
Monkeys surviving
10 20 30 40 50 60 70 80 90
Days of compliance/ survival of spores
Sources; 1) Henderson et al. J Hyg (Lond). 1956;54:28-36: Fig 2 for Days 5,10, 20
2) Friedlanderet al. J Infect Dis. 1993; 167(5): 1239-43. Table 1 for day 30
gig
Policy decisions: Example 1: The anthrax
letters: To recommend vaccine or not?
Figure 2: (Assumed risk of disease by
duration of antibiotic compliance
70 80 90 100
Days of compliance
Sources; 1) Henderson et al. J Hyg (Lond). 1956;54:28-36: Fig 2 for Days 5,10,20
2) Friedlander et al. J Infect Dis. 1993;167(5): 1239-43. Table 1 for day 30
2J2
CDC
AfMIM?
Update: Investigation of B
rorlsm-Related In
6IOTERRORISM-REIATEO ANTHRAX
Antimicrobial Postexposure
Prophylaxis for Anthrax:
Adverse Events and Adherence
Colin W Sh«pa'd.' Montse Soriano-Gabarro.' Elizabeth R. Ml.' Jsmn
"Overall adherence during 60 days of
antimicrobial prophylaxis was poor (44%),. .
Cnwjmcj InfotbtHJ* DiwasiM • VW 8. No 10. Oclobm 300?
2j2
49

-------
Sensors and decision making:
Specificity* and PPV
2 per 2 per 10 2 perl 2 per 2per 2 per 2 per 2 per 10
100 million million 100,000 10,000 1,000 100
million
Prevalance
*Sensitivity fixed at 95%
252
Appreciating reality:
stem cell transplant cures
Transplant
Estimates
: cure %

type
Patient
MDs
Actual (CI)
Autologus
70
32
44 (30-58)
Allogeneic



Early
80
62
52 (40-64)
Intermediate
73
42
32 (21-44)
Advanced
80
31
10 (0-23)
Source: Lee et al. JAMA, 2001:285:1034-1038
Smallpox: Risks <& benefits of
pre-exposure vaccination
Hospital personnel
General populace
Investigation teams
Martin I. Meltzer, Ph.D.
DEISS/NCPDCID/CCID/CDC
(Emerg Infect Dis 2003:9:1363- 1370)
beh
First Example
>Response to smallpox as a bioterror
weapon
>Dec, 2002 survey: 61% accept smallpox
vaccination if ".... offered as a
precaution .. "
>Blendon et al. NEJM 2003:348-354. HQ!
The balance: Risk-benefits for
the INDIVIDUAL
Risk of smallpox
INPUTS:
# initially infected;
Prob. of:
Risk of Release;
Contact;
Transmission;
Vacc. Effectiveness;

Serious vaccine
side-effects
INPUTS:
Prob. of:
Side effects
General populace
Risks: Smallpox vs. side-effects
(risk of side effects: 1:100,000
1,000 infected before detection)
,£r -0.000004 -
-0.000008 -
3
If risk of smallpox > 0 = give pre - exposure |
Metro area of 9 million
Entire U.S. pop.: 280 million
1:100 1:1,000 1:10,000 1:100,000
Risk of release
Meltzer: Emerg Infect Dis 2003:9:1363- 1370
50

-------
Utility: First conclusions
Utility: Second conclusions
> Utility: not always = produce a
number
^Concepts as important as data
> Utility: Can we "capture" all the issues?
>	Use proxies for many items
>	Limit to how many issues can be model
> Balance: Simplify vs. Simplistic
> Highlight data deficiencies
>How "thin is the ice?"
> Utility: Improved when:
>One model for a limited set of questions
> Explain one model doesn't answer all
023
1333
Utility: Third conclusions
>	Maximize utility when:
>Show changes in results with changes in
assumptions
^Confidence intervals are essential!
>	Maximize utility when:
^Describe probabilities
^Formulas are no good for descriptions!

-------
Quinn
Dose Response Applications for
Vaccines & Therapeutics
SAFER* HEALTHIER - PEOPLE"
Anthrax Biomarkers
•	Exposure ^ Infection ~ Disease
•	Exposure
—	May be innocuous
•	Barrier protection
—	May result in a host response
•	Innate, non-specific
•	Specific
•	Infection
—	Usually results in a host response
•	Innate, non-specific
•	Specific
•	Disease
—	Spore uptake/germination
—	2° to integument breach
CDC SAFER • HEALTHIER* PEOPLE™
Anthrose trisaccharide (ATS)
is antigenic and exposed on
the surface of B, anthracis
Sterne spores
A B. anf/jrac/s-specific antigenic
Region is localized to a defined
terminal group of the oligosaccharide
SAFER•HEALTHIER* PEOPLE"
Host Biomarkers for Exposure
- Anti-ATSResponses -
anthrose
Daubenspeck et al., 2004
CDC SAFER • HEALTHIER* PEOPLE"
52
'SlXMiSnW,^-

-------
Host Biomarkers for Exposure
- Anti-ATS Responses -
RM101 (untreated)
RM062 (Cipro 48h)
1"
1

loflofiMum allusion
1LD50 ~ 55x103 cfu
SAFER•HEALTHIER* PEOPLE"
Biomarkers for Infection
• LF Toxemia
-	Specific for anthrax
-	Quantitative LF detection (serum/plasma)
-	Detectable ~18 hr post-exposure
-	T= (18-x) hr post-infection
CDC SAFER•HEALTHIER•PEOPLE™
Triphasic Kinetics of LF Toxemia in rhesus macaques


/ Phase 3

Phase 1




Phase 2


1	2	3 4	5
Days Post-Challenge
Other biomarkers;
PA detectable later in infection
PGA and bacteremia may be undetectable at end of phase 2
Lethal Factor and Neutrophil Frequency
for Death before Phase-3 vs Survival to Phase-3 or Later
0 12 18 24 30 36 48 72 96 120 144 168
Hours Post-Challenge
LF Levels may predict outcome trajectory

, GtAC Non-Surv
(SMC Survivors
Continued Decline
Host Biomarkers for Disease
- Seroconversion to Anti-PA IgG -
Presentation of PA to host immune system
Measure of host recognition & response
Contributing test in diagnosis of 12/22 cases
Critical contribution in confirmation of 6/22 cases
Single supporting test in 3/11 cutaneous cases
1 CA case did not seroconvert
No seroconverters other than 22 confirmed cases
Hours Post-Challenge
SAFER • H EALTHIER• PEOPLE
53

-------
Vaccines & Therapeutics
Vaccines
-	Field efficacy & immunogenicity studies
-	Non-inferiority vs. 'benchmark'
-	Defined correlates of protection
-	Combined measures of surrogacy
Therapeutics
-	Pharmacokinetics (PK)
•	AUC, Cmax, V*, CL
-	Therapeutic index
•	Ratio of TD50:ED50 (alt. TD1:ED99)
-	Therapeutic window
•	Estimate of effective drug doses within the safety range
SAFER•HEALTHIER*PEOPLE"
Discussion Points
Dose response modeling of
infection/disease
-	Spore biomarker threshold?
-	Infection threshold?
-	Seroconversion threshold?
Dose response thresholds for therapeutics
-	Time frame within which the drug is effective?
-	Is there a point of no return?
-	Could treatment exacerbate disease?
SAFER•HEALTHIER•PEOPLE"
54

-------
Mitchell-Blackwood
Dose-Response Method
Comparisons: Bayesian Statistics
EPA-CDC Microbial Risk Assessment (MRA) Workshop
Wednesday, 22 April, 2009
Jade Mitch el I-Blackwood
Ph. D. Candidate
Drexel University
Department of Civil, Architectural,
and Environmental Engineering
Philadelphia, PA
Dr. Patrick Gurian
Assistant Professor
Drexel University
Department of Civil, Architectural,
and Environmental Engineering
Philadelphia, PA
Bayesian Methods
Bayesian is a very broad term
"Subjectivist" vs. "Frequentist" view
Uses Bayes' Theorem to learn from observations
Parameters are random variables
Virtually any approach that uses probability
distributions to describe uncertainty in model
parameters can be considered Bayesian
In responding to these points we will try to first note
the many options available within a Bayesian
framework and then address the approach we have
been using as a specific example
1. Is the dose-response statistical method utilized
empirical or mechanistic?
Both types of models can be fit in a Bayesian
framework
We have generally fit exponential and Beta-Poisson
dose response models which can be described as
"mechanistic models" (Haas et al. 1999)
"Mechanistic" because these models are based on biological plausibility:
~	Dose is considered random and Poisson distributed in a medium.
~There is a probability of the host entering a disease state.
~ There is a probability that 1 or more organisms is ingested by the host.
~	There is a survival probability of the organism once it is ingested.
Sometimes called "mechanistically-based empirical" because the parameters are
determined empirically from curve fitting to host survival data rather than assessments
of individual organism survival rates.
2. Is the method applicable for low-dose
extrapolations?
In general it depends on the model being fit
in the Bayesian framework
Both Exponential and Beta-Poisson are based
on sets of assumptions which if true would
allow for low dose extrapolations
Getting adequate low dose data to validate
these assumptions is a challenge
3. Can the method accommodate data pooling
and/or the use of correction factors?
Bayesian methods allowfor Hierarchical Models
¦	Mean parameters generated from individual experiments are drawn
from a common distribution (hyperdistribution)
¦	Hyperdistribution has hyperparameters
¦	Parameters for each experiment are informed by the observable data
and the hyperdistribution	Distributions of
An Inter-species variation example
Exponential Dose Response Model
P(d) = 1 - e"rd
Where:
P(d) = Probability of death
r = pathogen-host survival probability
d = dose of organisms to host
55

-------
An Inter-Species Variation Example
Organism / Strain
Used
Host Species
(Reference)
Number of
Dose
Groups
Minimum
Dose
Maximum Dose
Bacillus Anthracis
Vollum Strain
Guinea Pig
(Altboum, 2002)
6
200
20000000
Bacillus Anthracis
ATCC 6605 Strain
Guinea Pig
(Altboum, 2002)
6
30
3000000
Ames Strain
White Rabbit
(Pitt, 2001)
3
9240000
1911"Duuu
Bacillus Anthracis
Vollum Strain
Rhesus
Monkeys
(Druett, H.A., et.
al., 1953)
y
70320
398400
An Inter-Species Variation Example
Human, r Guinea	Monkey, r Rabbit, r
Pig, r
Results of Hierarchical Approach
In r
4. How is the calculated dose-response relationship
verified and validated?
Graphical plots of model and data
Bayesian Information Criterion
(BIC)/Deviance Information Criterion (DIC)
Bayes Factors
Prob[data|Modeli]/Prob[data|Model2]
Cross-validation
If sufficient data are available
How is model uncertainty adjusted for and
communicated to risk managers?
Parameters are explicitly random variables
Posterior distribution reflects range of values
and likelihoods of different values given both
what was known initially and what was
learned from the data
A predictive distribution for unobserved
pathogenic agents or species can be
generated by integrating overthe posterior
distribution of the parameters and
hyperparameters with measurable
uncertainty
Prior vs. Posterior Distributions
of the Grand Mean for Unobserved Species
Inm„~n(-1L9,22)
-2TntJ2 ~n(-0.67, 0.842)
lnm„~n(-11.78,0.672)
-	Posterior
(Informed Prior)
-	Prior (Informed
Prior)
—Posterior(Uninfo
rmed Prior)
—Prior (Uniformed
Prior)
]nMto~n(-11.9,202)
-2TnU2 ~n(-0.67, 0.842)
56

-------
Strengths and weaknesses
Hierarchical approach
allows generalization
across experiments
and to real world
conditionsthat may
not be identical
Hierarchical approach
makes strong
assumptions about
appropriate
distributional forms
Large data requirements
to validate these
assumptions
References
•	Altboum, et. al 2002. Post exposure prophylaxis against anthrax:
Evaluation of various treatment regimens in intranasally infected
guinea pigs. Infection and immunity, 70:6231.
•	Bartrand, T., Weir, M. and Haas, C., 2008. Dose-Response Models for
Inhalation of Bacillus anthracis Spores: Interspecies Comparisons, Risk
Analysis, 28:4
•	Druett, HA., Henderson, D.W., Packman, L., and Peacock, S., 1953.
Studies on Respiratory Infection. I. The Influence of Particle Size on
Respiratory Infection with Anthrax Spores. Journal of Hygiene, 51:359.
•	Haas, C.N., Rose, J.B., and Gerba, C.P., 1999. Quantitative Microbial
Risk Assessment. John Wiley & Sons, Inc. New York, NY
•	Meselson, M., Guillemin, J., Hugh-Jones, M., Langmuir, A., Popova, I.,
Shelokov, A., Yampolskaya, 0.,i994. The Sverdlovsk Anthrax Outbreak
of 1979. Science, 266:5188
•	Pitt, M.L.M., Little, S.F., Ivins, B.E., Fellows, P., Barth, J., Hewetson,}.,
Gibbs, P., Dertzbaugh, M., Friedlander, A.M., 2001. In vitro correlate of
immunity in a rabbit model of inhalation anthrax. Vaccine, 19:4768.
57

-------
&EPA
United States
Environmental Protection
Agency
PRESORTED STANDARD
POSTAGE & FEES PAID
EPA
PERMIT NO. G-35
Office of Research and Development (8101R)
Washington, DC 20460
Official Business
Penalty for Private Use
$300

-------