&EPA
United States
Environmental Protection
Agency
EPA 600/R-10/138 | December 2010 | www.epa.gov/ord
Benchmark Dose Analysis for
Bacillus anthracis Inhalation
Exposures in the Nonhuman
Primate and Application to
Risk-Based Decision Making
Office of Research and Development
National Homeland Security Research Center
-------
-------
Benchmark Dose Analysis for
Bacillus anthracis Inhalation
Exposures in the Nonhuman
Primate and Application to
Risk-Based Decision Making
w
Office of Research and Development
National Homeland Security Research
-------
-------
Disclaimer
The U.S. EPA through its Office of Research and Development funded and managed the research
described herein under the Battelle CBRNIAC Contract No. SP0-700-00-D-3180, Delivery Order
Number 0396, Task 503 and CBRNIAC Contract No. SP0-700-00-D-3180, Delivery Order 0603,
Task 794.
It has been reviewed by the Agency but docs not necessarily reflect the Agency "s views. No official
endorsement should be inferred. EPA docs not endorse the purchase or sale of any commercial
products or services.
For questions on this report, please contact Dr. Sarah Taft of the U.S. Environmental Protection
Agency, National Homeland Security Research Center. 26 West Martin Luther King Dr.. Mail Stop
NG-16, Cincinnati. Ohio. 45268. Dr. Taft can also be reached by phone at (513) 569-7037 or email
at Taft.Sarah@epa.gov.
If you have difficulty accessing these PDF documents, please contact Nickel.Kathv@.epa. gov or
McCall. Amelia@epa.gov for assistance.
-------
iv
-------
Contents
List of Abbreviations/Acronyms ix
Foreword x
Executive Summary xi
Acknowledgements xiii
1.0 Introduction 1
2.0 Literature Review 3
2.1 Modeling Exposure and Lethality from Inhalation Exposures 3
2.2 Available Dose-Response Data 5
2.2.1 Nonhuman Primate Data 5
2.2.2 Human Data 5
3.0 Methods 9
3.1 Identification of Animal Model 9
3.2 Identification of Data Sources 9
3.3 Criteria for Use of Data Sets 9
3.4 Selected Data Sets 10
3.4.1 Department of Defense Anthrax Data Set 10
3.4.2 Defense Intelligence Agency Anthrax Data Set 10
3.4.3 Druett et al. (1953) Anthrax Data Set 10
3.5 Calculation of Inhaled Dose 11
3.5.1 Department of Defense Anthrax Data Set 11
3.5.2 Defense Intelligence Agency Anthrax Data Set 11
3.5.3 Druett et al. (1953) Anthrax Data Set 11
3.6 Bench Dose Analysis 11
4.0 Results 13
4.1 Statistical Description ofDosc-Respon.se Sets 13
4.1.1 Department of Defense Anthrax Data Set 13
4.1.2 Defense Intelligence Agency Anthrax Data Set 14
4.1.3 Druett et al. (1953) Anthrax Data Set 15
V
-------
4.2 Benchmark Dose Analysis Results 15
4.2.1 Department of Defense Anthrax Data Set 15
4.2.2 Defense Intelligence Agency Anthrax Data Set 17
4.2.3 Druett et al. (1953) Anthrax Data Set 19
5.0 Discussion 21
5.1 Variation in Dose-Response Lethality Estimates 21
5.1.1 Physical Characterization of Exposure Product 22
5.1.2 Receptor-specific Exposure Assumptions 22
5.1.3 Selection of Dose Metric 23
5.1.4 Statistical Assessment of Dose-Response Relationship 24
5.2 Using Available Anthrax Data to Derive a Remedial Target 25
6.0 Conclusion 27
7.0 References 29
Appendix A 33
-------
Tables
Tabic 1. Published Bacillus anthracis Data Sets and Corresponding Estimated Lethality Values
Identified from Literature Search for Nonhuman Primate Data 6
Table 2. Published Reanalvscs of Nonhuman Primate Original Data Sets Provided in Table 1 7
Table 3. Model Parameters, Standard Errors, 95% Confidence Limits, and AlC Values for the
Statistically Significant Mathematical Model Fits to the DoD Anthrax Data 16
Table 4. The BMD and BMDL at Identified BMRs for the DoD Anthrax Data 16
Table 5. Model Parameters, Standard Errors. 95% Confidence Limits, and AlC Values for the
Statistically Significant Mathematical Model Fits to the DIA Anthrax Data 17
Table 6. The BMD and BMDL at Identified BMRs for the DIA Anthrax Data 18
Table 7. Model Parameters, Standard Errors. 95% Confidence Limits, and AlC Values Associated
with a Statistically Significant Model Fit to the Druett Anthrax Data 19
Table 8. The BMD and BMDL at Identified BMRs for the Druett Anthrax Data 20
Table 9. Comparison of Median Lethality Estimate and Assumed Minute Volume 23
Table A-l. Unrestricted and Restricted BMDS Model Results for the BMD50 and BMDL50
Lethality Values for the DIA Anthrax Data 34
Table A-2. Unrestricted and Restricted BMDS Model Results for the BMD10 and BMDL10
Lethality Values for the DIA Anthrax Data 34
-------
Figures
Figure 1. Exposure assessment modeling of inhalation route of exposure to acrosoli/cd
B. anthracis
Figure 2. Histogram and cumulative curve showing the frequency and cumulative percentage
of the inhaled doses in the DoD Anthrax Data
Figure 3. Histogram and cumulative curve showing the frequency and cumulative percentage
of the inhaled doses in the DIA Anthrax Data
Figure 4. Histogram and cumulative curve showing the frequency and cumulative percentage
of the inhaled doses in the Dructt Anthrax Data
Figure 5. Visual fit of probit model to the DoD Anthrax Data
Figure 6. Loge Logistic model for the DIA Anthrax Data
Figure 7. Wcibull run as exponential for the Dructt Anthrax Data
Figure 8. Dose-response assessment steps in the development of dose-response relationships....
Figure 9. The probit model equation used by Dructt ct al. (1953) and the BMDS software
(U.S. EPA, 2009a) to fit dose-response data
Figure 10. Generalized approach to calculate a remedial target from animal dose-response data
for inhaled B. anthracis spores
-------
List of Abbreviations/Acronyms
AG I a glass impinger
AIC Akaikc Information Criterion
BAULADae biologically active units per liter of air as function of aerodynamic diameter
BMD benchmark dose
BMDL benchmark dose limit
BMDS benchmark dose software
BMR benchmark dose response
CBRN chemical, biological, radiological, and nuclear
CBRNIAC Chemical, Biological. Radiological, and Nuclear Defense Information Analysis
Center
CFU colony forming unit
DI A Defense Intelligence Agency
DoD Department of Defense
EPA United States Environmental Protection Agency
GSD geometric standard deviation
L liter
LD50 lethal dose for 50% of the test population
H in micrometer
MM AD mean median aerodynamic diameter
NA not applicable
NRC National Research Council
PI-Cat Pathogen Information Catalog
RDDR regionally deposited dose ratio
USAPHC United States Army Public Health Command
ix
-------
Foreword
Following the events of September 11, 2001, the U.S. Environmental Protection Agency's (EPA)
mission was expanded to address critical needs related to homeland security. Presidential directives
identify EPA as the primary federal agency responsible for the country's water supplies and for
decontamination following a chemical, biological, and/or radiological attack.
As part of this expanded mission, the National Homeland Security Research Center (NHSRC) was
established to conduct research and deliver products that improve the capability of the Agency in
carrying out its homeland security responsibilities. One specific focus area of our research is the
compilation, development and evaluation of information on the human health effects of pathogens
that might be used by terrorists. Such information is critical to understanding the risks associated
with biological contamination and to support the development of site specific cleanup goals.
This report demonstrated that the EPA's Benchmark Dose Software is a useful tool for evaluating
microbial dose-response data, including that from Bacillus anthracis inhalation exposures.
Furthermore, this study found that a number of disparities in the literature for B. anthracis lethality
estimates could be traced to differences in physical characterization of the spore product, receptor-
specific exposure assumptions, the calculated dose metric, and the statistical process employed to
assess the data.
NHSRC has made this publication available to assist the response community to prepare for and
recover from disasters involving microbial contamination. This information is intended to move
EPA one step closer to achieving its homeland security goals and its overall mission of protecting
human health and the environment while providing sustainable solutions to our environmental
problems.
Jon Herrmann, Director
National Homeland Security Research Center
-------
Executive Summary
The U.S. Environmental Protection Agency's (EPA)
National Homeland Security Research Center helps
to protect human health and the environment by the
development of risk assessment methodologies for
chemical, biological, and radiological threat agents.
There is no current technical or regulatory consensus
of an acceptable inhalation Bacillus anthracis dose-
response relationship that is reflective of a wide range of
doses and response levels. The lack of this relationship
is the main challenge in the development of an overall
risk-based approach for addressing B. anthracis releases.
This study reviewed available B. anthracis dose-
response modeling and literature for the nonhuman
primate, evaluated the use of the EPA's Benchmark Dose
Software (BMDS) (BMDS 2.1.1 Version 2.1.1.55) to fit
mathematical models to these data, and considered the
application of these dose-response data in risk-based
decision making.
The review of published dose-response data for B.
anthracis inhalation exposures identified significant
variability in study design and subsequent lethality
estimates. The reviewed studies varied with regard to
B. anthracis exposure products (e.g., strain, particle
si/c). nonhuman primates tested (e.g., rhesus versus
cynomolgus monkeys), and experimental designs
(e.g., animal number tested). A search was conducted
to identify available B. anthracis dose-response data
from inhalation exposures for BMDS evaluation. Three
data sets were selected: U.S. Department of Defense
historical data for Dugway Proving Ground outdoor
studies conducted during the 1950's (Janssen 1955a,
1955b, 1955c), a U.S. Defense Intelligence Agency study
conducted with multiple strains of B. anthracis spores in
2001 (Barnewall et al. 2001), and the classic exposure
study conducted by Druett et al. (1953).
The results of the benchmark dose modeling for
the Department of Defense Anthrax Data. Defense
Intelligence Agency Anthrax Data, and the Druett
Anthrax Data show no apparent consistency in the
calculated median benchmark response levels when
using the same model with different data sets and no
apparent consistency with the previously published
values or reanalyscs of original data. BMDS outputs
reported included the benchmark dose (BMD), and the
benchmark dose limit (BMDL) for identified response
levels, statistical measures of fit for models, and fitted
parameters and intercepts. As one indication of the
overall variation in results, the best fitting models
yielded BMDL50 (BMD50) values for the Department
of Defense. Defense Intelligence Agency, and Druett
Anthrax Data Sets of 660 (530), 10,000 (4,900) and
48,000 (37,000) inhaled spores, respectively. Even with
the use of criteria designed to increase the comparability
among the selected studies for this review, large
differences in derived values were still present. A
BMDL10 animal inhaled dose value of 550 inhaled spores
from the loge logistic mathematical model was identified
as a point of departure using BMDS guidance (U.S.
EPA 2008) and subsequently used in the interspecies
extrapolation and human equivalent dose development.
This value was derived from the Defense Intelligence
Agency Anthrax Data.
The development of human equivalent doses from
the animal dose-response value requires explicit
evaluation of the dosimetric differences between
the test animal and the human receptors to properly
conduct an interspecies extrapolation. Data elements
(i.e., physical characterization of the exposure product,
receptor-specific exposure assumptions, and particle
size-specific depositional data for both receptors
of interest) that arc critical to the development of
the dose-response relationship arc also important to
the extrapolation process. In lieu of reliance on the
calculated environmental air concentration as the human
equivalent dose, an environmental surface concentration
was developed due to its ease in sampling. However,
accompanying assumptions were necessary regarding the
expected deposition of airborne particles to the sampling
surface, the area! extent of the sampling surface, and the
efficiency at which these particles can be removed from
the surface and recovered from the wipe sample. The
surface concentration was then converted to an estimated
viable spore number recoverable from a wipe after
sampling. This hypothetical human equivalent dose was
presented to illustrate the use of animal dose-response
data in support of site-specific cleanup goal development
but did not include other considerations typically
considered as part of site-specific risk management
decisions.
The study demonstrated that the EPA's BMDS is a
useful tool for evaluating microbial dose-response
data, including that from B. anthracis inhalation
exposures. As with all statistical software applications,
users must identify the assumptions incorporated
within the software and the mathematical models it
supports. However, this concept is also important for
the evaluation of published dose-response data when
comparing results. This study found that a number of
disparities in the literature for B. anthracis lethality
estimates could be traced to differences in physical
-------
characterization of the spore product, receptor-specific
exposure assumptions, the calculated dose metric, and
the statistical process employed to assess the data. One
area that consistently has received less attention in study
design has been the determination of spore number
per particle. The reliance on data sets using single
spore particles may be an appropriate means to bypass
this concern. However, lack of these data or sufficient
confidence that exposure products used in exposure
products arc composed of single spore particles can
hinder confidence in historical and even more recent data
sets.
-------
Acknowledgements
The authors wish to acknowledge the support of all those who helped plan and conduct the
evaluation, analyze the data, and prepare the report. This effort built upon previously conducted
benchmark dose analysis for the guinea pig conducted as part of the Pathogen Information Catalog
project conducted jointly by the U.S. Environmental Protection Agency and the U.S. Army Public
Health Command (USAPHC; formerly U.S. Army Center for Health Promotion and Preventative
Medicine). This report benefited greatly from these earlier discussions. Members of this group
included: Dr. Brandolyn Thran (USAPHC). Ms. Robyn Lee (USAPHC). Dr. Patrick Gurian (Drexel
University), and Dr. Jade Mitchel 1-Blackwood (Drexel University).
We thank the Defense Intelligence Agency and the USAPHC for the use of their dose-response data.
We would also like to thank U.S. Environmental Protection Agency personnel for their reviews and
feedback: Dr. Tonya Nichols (National Homeland Security Research Center). Dr. Jeff Gift (National
Center for Environmental Assessment). Ms. Eletha Brady-Roberts (National Homeland Security
Research Center). Dr. Harlal Choudhury (National Center for Environmental Assessment). Dr.
Deborah McKean (EPA Region 8), and Dr. Femi Adeshina (National Homeland Security Research
Center). In particular, we wish to thank Dr. Gift for statistical assistance provided for model selection
and evaluation during the course of the project. We also wish to thank Dr. Michael Taylor from
Battelle Memorial Institute for reviews and technical assistance provided throughout the project.
For questions on this report, please contact Dr. Sarah Taft of the U .S. Environmental Protection
Agency, National Homeland Security Research Center. 26 West Martin Luther King Dr., Mail Stop
NG-16, Cincinnati, Ohio, 45268. Dr. Taft can also be reached by phone at (513) 569-7037 or email
at Taft.Sarah@epa.gov.
-------
xiv
-------
1.0
Introduction
The United States anthrax letter attacks of 2001
highlighted the need for Bacillus anthracis dosc-
rcspon.se data that arc suitable for risk-based decision
making (Gutting ct al. 2008). While a number of
publications identify 8,000 to 10,000 inhaled spores as a
median range of lethality estimates for human exposures,
Coleman ct al. (2008) noted that these values cannot be
attributed to an originating data set and arc seemingly
more reflective of best professional judgment. While
the lack of scientific evidence for this commonly cited
measure of lethality is problematic, it also is indicative
of much larger knowledge gap. There are a number of
published lethality estimates for identified data sets, but
the published values differ greatly (e.g., Dructt ct al.
1953, Classman 1966). Currently, there is no technical
or regulatory consensus on an acceptable inhalation B.
anthracis dosc-rcspon.se relationship that is reflective
of a wide range of doses and response levels. The lack
of an accepted dosc-rcspon.se relationship is the main
challenge in the development of a risk-based decision
making approach for B. anthracis releases.
Mathematical models (e.g., probit. exponential, bcta-
Poisson) used to describe chemical dosc-rcspon.se
relationships have also been used with microbial
hazards. As with chemical hazards, quantitative dose-
response relationships have been developed for microbial
hazards through the evaluation of mathematical models
with available data by using curve-fitting techniques.
Descriptions of microbial quantitative dosc-rcspon.se
relationships using mathematical models (e.g.,
Armstrong and Haas 2007, FAO and WHO 2003, Haas
ct al. 1999) have been available for a number of years as
well as comparisons of results obtained from deploying
different models with individual data sets (e.g., Holcomb
ct al. 1999). The resulting equations for the mathematical
models, percentiles of interest (e.g., Lethal Dose for 50%
of the test population [LD50]), and parameter values arc
typically reported. However, only two microbial dose-
response analyses have been identified that reported
results using benchmark dose concepts and terminology
(Moon ct al. 2004, 2005). Moon ct al. (2004) compared a
set of mathematical dosc-respon.se models when applied
to microbial dose-rcspon.se data sets and presented
results in the form of benchmark dose outputs. Building
on this approach, model averaging of benchmark results
has been conducted as a means to incorporate recognized
uncertainties in model choice (Moon ct al. 2005). In
contrast. Englchardt and Swartout (2006) have noted
concerns with the use of confidence limits associated
with classical statistics for microbial dosc-rcspon.se
data, which would also preclude the use of the classical
models associated with benchmark dose modeling for
these data sets. However, this area of concern is beyond
the scope of this paper.
Though originally developed for chemical hazards,
the U.S. Environmental Protection Agency's (EPA)
Benchmark Dose Software (BMDS) provides an
accessible tool to evaluate benchmark dose approaches
and facilitates the consideration of a number of
mathematical models for dosc-rcspon.se relationships.
While there have been publications describing the use of
benchmark dose analysis, there arc no published dose-
response results for microbial analyses that have been
generated utilizing the BMDS.
This report will review available B. anthracis dose-
response modeling and literature for the iionhunian
primate, evaluate the use of the EPA's BMDS to fit
mathematical models to these data, and consider the
application of dosc-rcspon.se data in risk-based decision
making.
-------
2
-------
2.0
Literature Review
2.1 Modeling Exposure and Lethality
from Inhalation Exposures
As with many biothrcat agents, the inhalation route of
exposure is the primary concern in a bioterrorist release
of B. anthracis. Exposure products arc the combination
of the biothrcat agent plus any additives or impurities
in the preparation that would be used in the bioterrorist
release or in the dosing of test animals. The exposure
product would be in the form of single or multiple spore-
containing particles. By design, spore products readily
form aerosols that maximize airborne time prior to
deposition on surfaces. Inhaled spores that deposit in the
alveolar region of the lung and then surv ive subsequent
phagocytosis by local macrophages have the potential to
germinate into vegetative bacteria (Hilmas et al. 2009).
Systemic and lethal illness may result when surviving
bacteria arc transported to the lymph nodes absent early
and effective treatment.
Two frameworks have been identified to provide a
comprehensive perspective for the discussion of dose-
response modeling of B. anthracis lethality. The first
framework, the National Research Council's (NRC
2008) Framework for Assessing the Health Hazard
Posed by Bioaerosols, describes an approach to
quantify the physical and biological factors driving
the health hazard posed by biothrcat agents during
inhalation exposures. The second framework, EPA's
exposure assessment process (U.S. EPA 1992), provides
terminology and concepts to consistently describe the
calculation and measurement of dose.
NRC (2008) identifies the physical characteristics of
the acrosoli/cd product mixture for determination of the
likelihood and number of inhaled spores that deposit
deep in the lung. Physical characteristics define the
concentration of the product in the air medium in units
of particles per unit volume, the median particle si/c
and standard deviation about that median measure, and
the spore number per particle. Biological factors include
the type of biothrcat agent, the viability of the biothrcat
agent, and the virulence of the biothrcat agent. Together,
tliesc factors comprehensively describe the exposure
product based on its potential to deposit in the lung and
the product's innate virulence.
Of the physical characteristics, particle si/c is a key
determinant of the potential for spore deposition in
the alveolar region of the lung. Knowledge of the
general relationship between particle si/c and B.
anthracis lethality was identified relatively early in
the B. anthracis dosc-rcspon.sc literature (e.g., Dructt
et al. 1953). However, technology at that time lacked
the capacity to identify particle measurements for
complex si/c distributions as is commonly performed
today. Early B. anthracis dosc-rcspon.sc studies targeted
environmental measurement of particles 5 jnii and less
as a proxy for rcspirablc particles (e.g., Classman. 1966,
Jansscn 1955a, 1955b, 1955c). Data were not typically
measured describing the particle distribution (e.g.,
median particle si/c and associated distribution of si/cs
about that median | geometric standard deviation]) or the
nontruncatcd elements of the particle si/c range (i.e.,
particles greater than 5 p,m).
Current publications (e.g., NRC 2008, Pitt and LcClairc
2005, U.S. EPA2004) describe the particle si/c range
of optimal deposition for humans to be 1 to 5 jnii
(measured as an aerodynamic diameter). However, it
should be noted that the 5 jnii value docs not represent
a strict cutoff for deposition as particles greater than 5
nm may still be deposited, albeit at lower rates, during
normal nasal breathing patterns. Breathing pattern
changes, including increases in tidal volume, breathing
frequency, and the use of oral versus nasal breathing,
may all increase the alveolar deposition fraction of
particles in the 1 to 5 jim range as well as facilitate
alveolar deposition of particles greater than 5 jim in si/c
(U.S. EPA 2004). As a point of reference, dcpositional
fractions in humans during nasal inhalation have been
identified to be approximately 20% for 1 jim particles
with a decline to 10% for 5 jim particles (U.S. EPA
2004). However, 5 jim particles may exhibit between 20
and 50% alveolar deposition during oral inhalation (U.S.
EPA 2004).
A unique contribution made by the NRC (2008)
framework is recognition of the importance in
quantifying the spore number per particle. Knowledge
of the spore number per particle, when combined with
the air concentration and the particle si/c and associated
dcpositional fraction, can be used to quantify the spore
number deposited in the alveolar region. With noted
exceptions (i.e., portions of the Dructt et al. 1953
published data), data sets arc not available that delineate
spore number per particle when single spore particles
were not identified as the exposure product.
In recognition of the multifactorial nature of particles
in bioacrosol exposures, the NRC (2008) proposed
3
-------
a new measure to describe particle characteristics
and concentration. This measure, the Biologically
Active Units per Liter of Air as a function of particle
aerodynamic diameter (BAULADae), incorporates the
measurement of airborne particle concentration, agent
number per particle, and particle distribution to capture
the dosinietricalh important elements of B. anthracis
aerosolized products. The use of the BAULADae
measurement process will facilitate the prediction of
the deposited dose from a measured environmental air
concentration.
In addition to the framework describing and quantifying
characteristics of the aerosolized spore product as
described by NRC (2008), traditional approaches (e.g.,
U.S. EPA 1992) for chemical exposure assessment
provide a useful complement to consistently define
exposure and the type of dose metric (Figure 1). The
distinction among the various exposure and dose metrics
is important in the evaluation of published dose-response
values to ensure that comparisons are made at similar
levels in the exposure assessment continuum.
The point of contact measurement is the environmental
air concentration of B. anthracis (Figure 1). The
environmental air concentration can then be combined
with exposure assumptions describing receptor contact
rates with the air medium to define the inhaled dose.
Absent accompanying particle size definition for the
product, this dose definition is not of sufficient rigor
to allow for extrapolation to humans or to evaluate the
applicability of these data to different biothreat agent
products or receptor exposure scenarios.
The deposited dose, also defined as an applied dose,
reflects the material available for absorption across a
body boundary, or in this case, movement across the
alveolar membrane. Particle deposition is driven by
particle size and density. For the monkey receptor,
off-the-shelf computer applications are not available to
derive the particle size distribution-specific depositional
fractions as are currently available for other receptors
(e.g., rabbits, guinea pigs, and humans using U.S. EPA's
Regionally Deposited Dose Ratio [RDDR] Model).
Additionally, spore per particle data are a necessary
accompaniment to particle size data to adequately
describe the number of spores that are deposited. These
data are not present in most published data sets, with the
exception of products identified as single spore particles.
As a result, the majority of historical data is of limited
utility for the development of a dose metric beyond the
potential dose, or inhaled dose, as identified in Figure
1. In the absence of sufficient data to derive an applied
dose, the potential dose is typically used as a proxy for
the applied dose (U.S. EPA 1992).
However, the inability to predict the relationship
between environmental air concentration and spore
deposition is a major uncertainty in the dose-response
modeling of B. anthracis exposures (Coleman et al.
2008). The determination of a deposited dose is critically
important to the extrapolation of dose-response data by
allowing explicit consideration of dosimetric differences
(e.g., inhalation minute volume, species-specific
deposition rates) between the test animal and the human
(Jarabek et al. 2005, Pitt and LeClaire 2004). This lack
of data also severely limits the application of dose-
response relationships for aerosolized products where the
particle size composition differs from the particle size of
the product for which the dose-response relationship was
originally developed.
Particle
Measurement Air
and Knowledge of
Particle Size
NRC Descriptor
(2008)
rr
Environmental
Air
Concentration
(Spores or
particles per Liter
air)
nha ed Dose
v
Particle Distribution Data
+ Spore per Particle
+ Particle Concentration
in Air
(BAULA r, )
Deposited Dose to
Alveolar Region of
Lung
Spores Accessible by
Macrophages
Point of
Contact
Potential
Dose
Applied
Dose
Internal
Dose
Figure I. Exposure assessment modeling of inhalation route of exposure to aerosolized B. anthracis.
-------
2.2 Available Dose-Response Data
Overall, published dose-response data for inhalation
exposures exhibited significant variability in lethality
estimates (Tables 1 and 2). The original studies included
differences in B. anthracis strain and product types,
particle si/c. nonhuman primate test animal, assumed
inhalation rate, and experimental design (Table 1). Table
2 describes available published reanalyzes of studies
identified in Table 1.
2.2.1 Nonhuman Primate Data
Published median lethality values for nonhuman primate
studies range from 4,130 (Classman 1966) to 61,800
inhaled spores (Vasconcelos et al. 2003) (Table 1). The
Classman (1966) study design included a high number
of subjects (i.e., 1,236 cynomolgus monkeys), and its
calculated median lethality value is considerably lower than
most other estimates. There is high interest in the data set
and it is often cited, but the raw data arc not available nor
arc statistical goodness of fit measures for the published
probit slope and median lethality value. Unfortunately, the
only available documentation is Classman's (1966) brief
discussion of the study and results.
Three B. anthracis dosing studies (i.e., A lb rink and
Goodlow 1959, Druett et al. 1953, Young et al. 1946,)
were conducted with nonhuman primates during the
1940's and 1950's that include published dose-response
data (Table 1). Median lethality estimates were provided
in metrics of environmental spore concentration and
inhaled spore dose. Young et al. (1946) identified a
median lethality estimate of 250,000 spores per liter
air. Albrink and Goodlow (1959) described individual
monkey lethal doses in units of inhaled spores. Druett et
al. (1953) provided a median lethality estimate of 45,000
spores per liter air and an inhaled dose of 53,000 spores.
All studies cited the use of single spore particles.
It should also be noted that Haas (2002) and Bartrand
et al. (2008) reanalyzed the Dreutt et al. (1953)
data to assess the fit of a number of commonly used
mathematical models (Table 2). Haas (2002) identified
a statistically significant fit to the exponential dose-
response relationship and reported an LD,;i value of
96,800 inhaled spores. Haas (2002) compared the
low dose extrapolation derived from Druett et al. 's
(1953) data and the exponential model with the low
dose extrapolation derived using Classman's (1966)
published probit slope value in the probit model. The
estimates in the low dose regions of these models varied
by almost three orders of magnitude, which is not
unexpected given the ty pical behavior of these curves in
the low dose regions (Haas 2002). Haas (2002) posited
that these differences may also indicate fundamental
differences between the two data sets, but that this
would not be possible to ascertain absent the original
data from Classman (1966). Bartrand et al. (2008) also
reanalyzed Druett et al. (1953) monkey data sets using
the exponential, probit. and bcta-Poisson models; the
best fitting model identified was the exponential model
with a calculated LD,;i of 92,000.
More recent studies using nonhuman primates have been
conducted to assess median measures of lethality. These
studies often have considerably lower animal numbers
(i.e., 14 or less subjects) than Classman's (1966) study
and have produced median lethality estimates that arc
considerably higher than Classman's published value
of 4,130 inhaled spores. Using a log10 probit model,
Vasconcelos et al. (2003) derived an LD50 of 61,800
inhaled spores for 14 cynomolgus monkeys exposed to
the Ames strain. In a different study using the log10 probit
model, Estep et al. (2003) derived an LD50 of 10,900
inhaled spores for the Ames strain and 10,300 spores for
the Vollum strain.
2.2.2 Human Data
There arc no human dosing studies conducted with B.
anthracis due to the known high lethality from inhalation
exposures. The limited human dose-response data
available are derived from dose reconstruction after
airborne release, and these values have questionable
levels of rigor relative to typically conducted animal
dosing studies.
Human anthrax incidence from the 1979 Sverdlovsk
B. anthracis release has been modeled in conjunction
with doses calculated using assumptions for the amount
of source material released, atmospheric dispersion
modeling, and human exposure locations (Meselson
1995, Wilkening 2006). These studies have shown
agreement in modeled human anthrax cases when using
log10 probit and exponential dose-response models.
In an attempt to identify a range in the amount of
released source material. Meselson (1995) demonstrated
potential agreement with human anthrax incidence
when using Druett et al."s (1953) LD50 value of 45,000
inhaled spores for rhesus monkeys as the input into an
exponential model and also Classman's (1966) log10
probit slope value of 0.7 as an input into a probit slope
model. Using more recently available geospatial and
weather data specific for Sverdlovsk in 1979, Wilkening
(2006) further demonstrated that these data could also
be shown to be consistent with the level of response
estimated using Classman's (1966) log10 probit slope
value of 0.7 as well as that of an exponential model
incorporating conipcting-risk components to replace the
deterministic potency value (i.e., ty pically identified as k
in exponential model equation).
5
-------
However, the actual volume of the released source
during the Sverdlovsk event remains unknown.
Therefore, these studies may provide confirmation of
general mathematical forms appropriate for the dose-
response relationship, but the derivation of potency
values is dependent on knowledge of the source amount.
As a result, there is high uncertainty associated with the
estimated dose used to confirm these relationships and
they will, at best, serve a complimentary rather than a
definitive role in any subsequent evaluations.
Table 1. Published Bacillus anthracis Data Sets and Corresponding Estimated Lethality Values Identif ed from
Literature Search for Nonhuman Primate Data.
Animal
Author
Reanalysis or
Strain
Particle Size
(Total Number
Exposed)
Model and Calculated
LDg0 (95% Conf dence
Interval)
Parameters
Original Data?
Assumed Inhalation
Rate, If Dose Metric
of Inhaled Spores
Young et al.,
(1946)
Unspecified Monkey
(16)
Bliss (1935)
Original
Detrick 25
Single Spore
NA
200,000 Spores
Environmental Air
Concentration
NA
Rhesus Monkey (72)
Logl0 Probit
45,000 Spores
Environmental Air
Druett et al.,
(1953)
Logl0 Probit
Slope = 3.19
with intercept
of 2.91
(Based on
Exposure
Concentration
x 10'4 as Dose)
Original
Vollum M36
Single Spore
1.2 L/minute
Concentration
(95% 30,000-52,000)
53,000 Spores Inhaled
Dose
Chimpanzee Monkey
(4)
Dose-response data
published, no analysis
provided
Albrink and
Goodlow,
(1959)
Original
Vollum rB
Single Spore
Developed Method
to Measure Minute
Inhaled Dose -
Response
NA
Volume, Values Not
Provided in Article
32,800 - Survival
34,350 - Survival
39,700 - Death
66,500 - Death
Glassman,
(1966)
Original Data
from Personal
Communication
by Jemski
Original publication
does not identify
strain, Haas (2002)
Assumed
to be less
Cynomolgus Monkey
(1,236)
Log10 Probit
4,130 Spores
Inhaled Dose
(95% 1,980-8,630)
Log10 Probit
Slope = 0.669
references Meselson
(2001) as the source of
the strain identification
of Vollum
than 5 jim
through use of
preimpinger
Unknown
No Intercept
Reported
1 and 2 p.m
(Mass Median
Aerodynamic
Diameter)
Cynomolgus Monkey
(14)
Log10 Probit
61,800 Spores
Inhaled Dose
(Fiellers* 95%
Conf dence Interval
34,800-110,000)
Log10 Probit
Vasconcelos
et al., (2003)
Original
Ames
Plethysmography
during Challenge,
Values Not Provided
in Article
Slope = 4.21
No Intercept
Reported
6
-------
Author
Reanalysis or
Original Data?
Strain
Particle Size
Animal
(Total Number
Exposed)
Model and Calculated
LDg0 (95% Conf dence
Interval)
Parameters
Assumed Inhalation
Rate, If Dose Metric
of Inhaled Spores
Estep et al.,
(2003)
Original
Ames
1.31 mM [sic]
(Cumulative
Mass Median
Aerodynamic
Diameter)
Rhesus Monkey
Log10 Probit
10,900 Spores
Inhaled Dose
(Fiellers 95%
Conf dence Interval
1,320-241,000)
No Slope
Reported
No Intercept
Reported
Plethysmography
during Challenge,
Values Not Provided
in Article
Vollum
1.31 mM [sic]
(Cumulative
Mass Median
Aerodynamic
Diameter)
Rhesus Monkey
Logl0 Probit
6,750 Spores
Inhaled Dose
(Fiellers* 95%
Conf dence Interval 21
-116,000
No Slope
Reported
No Intercept
Reported
Plethysmography
during Challenge,
Values Not Provided
in Article
*Fieller's confidence interval, as calculated using SAS™, is an inverse confidence limit describing the limit about the level of the independent
variable that results in the specified result (i.e., confidence limit about the dose related to an identified response level).
Table 2. Published Reanalyses of Nonhuman Primate Original Data Sets Provided in Table 1.
Author
Reanalysis or
Original Data?
Strain
Particle Size
Animal
(Total Number
Exposed) and
Assumed
Model and
Calculated LDg0
Parameters
and/or
Coeff cient
Inhalation Rate (If
Inhaled Spores Dose
Metric)
Haas, (2002)
Reanalysis of
Druett et al.
(1953)
Vollum M36
Single Spore
Rhesus Monkey (72)
Exponential
96,800 Spores
Inhaled Dose
(95% 70,700 -136,000)
k=7.16 x 10'6
2.4 L/minute
Bartrand et
al. (2008)
Reanalysis of
Druett et al.
(1953)
Vollum M36
Single Spore
Rhesus Monkey
(72)
Exponential
92,000 Spores
Inhaled Dose
(95% 29,440-70.932)
[sic]
k=7.16 x 10'6
2.4 L/minute
-------
8
-------
3.0
Methods
3.1 Identification of Animal Model
Nonhuman primates, by virtue of their close
phylogenctic relationship to humans, have been
identified as an appropriate animal model for human
inhalation exposure to B. anthracis spores. In particular,
rhesus and cynomolgus monkeys exhibit an overall
disease course and pathology of anthrax illness similar to
that identified in humans (Frit/ et al. 1995, Zaucha et al.
1998). A comprehensive literature review for rhesus and
cynomolgus monkey dose-response data was conducted
to identify dose-response analyses and potential data
sets for reanalysis with benchmark dose modeling
techniques.
Based on the literature review, the rhesus monkey
(Macaca mulatto) was selected as the animal model to
further evaluate benchmark dose approaches because of
the availability of both published analyses and suitable
dose-response data for reanalysis.
3.2 Identification of Data Sources
Two unclassified data sources were consulted to identify
suitable data for further dose-response analysis. The
first source was the Pathogen Information Catalog (PI
Cat); this data compilation was the product of a joint
effort between the EPA's NHSRC and U.S. Army Public
Health Command (formerly the U.S. Army Center for
Health Promotion and Preventative Medicine (U.S.
EPA 2010). The PI Cat was developed to identify and
collect available open source and limited distribution,
but unclassified, B. anthracis dose-response data. Data
collection procedures for the PI Cat are described in U.S.
APHC (2010). The second source was the Chemical.
Biological. Radiological, and Nuclear Defense
Information Analysis Center, also known as CBRNIAC.
The CBRNIAC is a Department of Defense Information
Analysis Center that is a comprehensive repository for
chemical, biological, radiological, and nuclear (CBRN)
technical information.
3.3 Criteria for Use of Data Sets
The primary criterion for selection of data sets was their
suitability for dose-response analysis using the dose
metric of inhaled spores and the measured response of
lethality. In lieu of the published dose metric of inhaled
spores, a dose metric of environmental air concentration
was acceptable if there were body weight or measured
minute volume data for the animal subjects. Suitability
for dose-response analysis was evaluated through the
consideration of three characteristics: 1) a description
of the B. anthracis exposure product allowing for
characterization of the physical factors described in
NRC (2008), 2) sufficient animal numbers in the dose-
response data set to allow for use of readily available
dose-response methods (i.e., preferably 24 or more
total animals), and 3) the inclusion of dose groups in
the lower range of responses (e.g., data including dose
groups with lethality at levels greater than 0% but less
than 20%).
The identification of the NRC (2008) comprehensive
set of physical factors was not available prior to the
generation of the evaluated data sets and has not been
routinely collected during the generation of B. anthracis
dose-response data. At a minimum, selected studies
must have defined the administered dose in a manner
(e.g., particle si/c. quantification of spores) that allows
for a calculation of inhaled dose. Biological factors were
not incorporated in the identification of data sets since
only limited data were available in the literature.
Three data sets were selected for benchmark dose
analysis based on their meeting all elements from the
identified three criteria, with the noted slight relaxation
of the NRC physical factors to allow acceptance of
particle si/c and spore quantification as sufficient
exposure product information. The first data set selected
was U.S. Department of Defense (DoD) historical data
from Dugway Proving Ground outdoor studies (Janssen
1955a, 1955b, 1955c). Strengths of this study included
an experimental design that incorporated relatively
low dose exposures and the largest total number of
exposed monkeys for which raw dose-response data
were available. The second data set selected was a
U.S. Defense Intelligence Agency study conducted
with multiple strains of B. anthracis spores in 2001
(Barncwall et al. 2001) Strengths of this study included
the direct measurement of respiratory parameters data
obtained during the exposure challenge and provided a
detailed particle size characterization of the exposure
product. One historical study, Druett et al. (1953), was
also identified for further evaluation using benchmark
dose techniques. Strengths of this study included the use
of a single spore dosing product, the availability of the
average monkey weight for the overall study, and the
large number of total exposed monkeys.
9
-------
3.4 Selected Data Sets
3.4.1 Department of Defense Anthrax Data Set
The Department of Defense Anthrax Data (hereafter,
DoD Anthrax Data) were developed from three outdoor
studies conducted at Dugway Proving Ground. Utah
(Janssen 1955a, 1955b, 1955c). These studies measured
the lethality of B. anthracis inhalation exposures of
monkeys, identified by U.S. APHC (2010) as the rhesus
monkey (Macaca mulatto). Monkey weights ranged
from 2.4 to 6.2 kilograms (Janssen 1955a, 1955b,
1955c). The B. anthracis strain was not identified in
the available DoD trial reports, but the Vollum strain
has been described in common use at the time by DoD
researchers (U.S. APHC 2010). Particle si/c data were
not available from Janssen (1955a, 1955b, 1955c).
Bacillus anthracis spores were deployed in an outdoor
environment from exploding E61R4 bomblets, and the
released spores traveled by natural air currents to the
exposure location of the monkeys. Air samples were
obtained from approximately 35 outdoor sampling
locations where a group of five monkeys was co-located
with three preimpinger/impinger air sampling devices.
The distance and height of the impingers relative to the
placement of the exposed monkeys is unknown. The
exposure duration of the monkeys is also unknown.
However, it was assumed that the air samplers were
maintained for the same time duration as the potential
exposure of the monkeys.
Air measurement devices consisted of a preimpinger and
impinger. Preimpingers were used to filter out particles
greater than 5 |im as this value was thought to be the
upper limit on rcspirablc particle si/c. Impinger fluid
was then plated and colony forming units (CFU) were
counted as the measurement for B. anthracis organisms
in the air per liter.
The work was conducted during the mid-1950s, and
study designs were limited by available knowledge.
However, there were some shortcomings in the study
design that warrant noting. The trial reports did not
identify the post-exposure observation period for
measurement of lethality after exposure. It is possible
that a 10-day observation period was used as that length
is consistent with B. anthracis studies of guinea pigs
conducted at Fort Oct rick at that time (Jemski and
Phillips 1964, U.S. APHC 2010). However, Jemski
and Phillips (1964) also note that an additional 3 to
5% mortality may be identified if monkeys are held for
up to six months. It is now known that the incubation
period for inhalation anthrax may extend up to 100
days (Inglesby et al. 2002), which would likely be most
relevant for low dose exposures. The reports did not
describe any steps to decontaminate the monkeys (i.e.,
cleaning or decontaminating of fur) after removal from
the exposure environment. Particle si/c data were not
described nor was information provided on the spore
number per particle.
3.4.2 Defense Intelligence Agency Anthrax Data
Set
The Defense Intelligence Agency Anthrax Data
(hereafter, DIA Anthrax Data) were derived from one
B. anthracis inhalation study conducted during 2001
(Barnewall et al. 2001). Thirty-four rhesus monkeys
{Macaca mulatto) were exposed to aerosolized strains of
B. anthracis spores in a head-only chamber enclosed in a
Class III biological safety cabinet. The strain information
is classified. A specially designed nebulizer delivered
B. anthracis spores in aerosol droplets; the delivered
particle si/c and their distribution were described as a
mean median aerodynamic diameter (MMAD) of 1.31
p,M and a geometric standard deviation (GSD) of 1.8.
During testing, monkeys were physically restrained
and dosed with 3 to 6 mg/kg Telazol® (a combination
anesthetic and tranquilizer). Monkeys were exposed for
10 minutes to the acrosoli/cd B. anthracis air mixture
and then were maintained in the same chamber for an
additional five minutes while clean air was flushed
through the system.
The air concentration was sampled through the use of
an all glass impinger (AGI)-6 impinger, pressure gauge,
and a vacuum pump to pull the sample. The impinger
collected air samples in sterile water, and plating was
conducted after serial dilution to determine the B.
anthracis aerosol concentration in CFU per milliliter of
liquid. This value was then used to calculate the value
for CFU per liter of air after incorporation of the air
sampling parameters. Plethysmography was conducted
during testing to measure individual-specific minute
volumes. After exposure to the acrosoli/cd B. anthracis,
the head of the each monkey was decontaminated prior
to removal from the biological safety cabinet. Monkeys
were then observed for B. anthracis-related death for 120
days post-exposure.
The DIA Anthrax Data were developed using state-of-
the-art practices for inhalation studies of acrosoli/cd
biological agents, and the exposure data (i.e., minute
volume and aerosol concentration in air medium) arc
presumed to be accurate. However, spore number per
particle was not identified in Barnewall et al. (2001).
3.4.3 Druett et al. (1953) Anthrax Data Set
The Druett et al. (1953) Anthrax Data (hereafter. Druett
Anthrax Data) were derived from published dose-
response data from B. anthracis inhalation studies. The
study measured environmental air concentration (in
units of single spores per liter) and associated mortality
-------
in 7 to 14 pound rhesus monkeys. The monkey species
was identified in original publication as Macacus rhesus
(historical name forMacaca mulatto). Single spore cloud
exposures were conducted in a Henderson apparatus.
Monkeys were exposed for one minute and then were
observed for three weeks post-exposure for mortality.
Since no measurements of minute volume were taken
prior to or during the aerosol challenge, allometric
equations relating monkey weight to minute volume
were used in the original study and in this subsequent
analysis. The original work docs not indicate if the
monkeys were tranquili/cd and/or sedated during the
exposure challenge and also docs not identify if animals
were acclimatized to the testing apparatus prior to the
exposure challenge. The use of tranquilizers or sedatives,
or conversely monkeys experiencing high levels of
stress, will a fleet the challenge animal's minute volume
beyond that predicted by the allometric equation. An
allometric equation relating weight to minute volume
is based only on the measured correlation and the
physiological state at which the measurements were
made.
3.5 Calculation of Inhaled Dose
3.5.1 Department of Defense Anthrax Data Set
Inhaled doses were determined using the calculated
minute volume inhalation rate and the environmental air
concentration derived from CFU counts of germinated
B. anthracis spores plated from the impinger fluid. The
exposure duration was unknown but it is assumed that
the impingers collected air samples during the same time
period as the monkey exposure duration.
The environmental air concentrations provided in the
original study reports (Janssen 1955a, 1955b, 1955c)
were used directly and not recalculated for this study.
These values were derived using Equation 1 and
incorporated the impinger sampling rate and the count of
B. anthracis colonics plated from the impinger fluid.
Equation 1.
Environmental Air Concentration {CFU per L /minute) =
f Impinger Count {CFU) ^
^Impinger Rate {LIMinute) J
The environmental air concentration (Equation 1) was
used with the estimated minute volume (Equation 2) to
derive the inhaled dose (Equation 3) that was used as the
dose metric in this reanalysis. Janssen (1955a, 1955b,
1955c) identified the arithmetic averaged group-specific
weight for each dose group. The minute volume was
calculated using body weight values in the allometric
equation described by EPA (1988) (Equation 2 following
a unit conversion to liters per minute). Equation 2 was
developed using regression analysis, with 0.81 and
0.4862 representing parameters that were fit to the data
used to derive the allometric equation (U.S. EPA 1988).
Equation 2.
Daily Inhalation Volume {m / Day) =
. 0.4862
0.81 x (Body Weightkg)
Equation 3.
Inhaled Dose {CFU) =
Minute Volume {LIminute) x
Environmental Air Concentration {CFU per LI minute)
3.5.2 Defense Intelligence Agency Anthrax Data
Set
The DIA Anthrax Data study design directly measured
the monkey respiratory parameters and used an active
sampling approach to derive B. anthracis environmental
air concentrations. Inhaled dose was calculated
by Barnewall et al. (2001) using the AG I sample
concentration, sampling parameters, and exposure
duration (Equation 4). The originally published inhaled
doses were used as the doses in this reanalysis.
Equation 4.
Inhaled Dose {CFU) =
{AGI Concentration {CFU / ml) x AGI Sampler Volume {ml) x
Minute Volume {L /Minute)
AGI Sampling Rate {L /Minute)
3.5.3 Druett et al. (1953) Anthrax Data Set
Druett et al. (1953) gathered environmental air
concentration data using impingers described by
Henderson (1952). Druett et al. (1953) described the
range of body weights (i.e., 7 to 14 pounds) for all
monkeys used in the overall study, and the midpoint of
this range (i.e., 10.5 pounds) was used after conversion
as the input for Equation 2. The calculated minute
volume of 1.2 L was used for all dose groups. The
inhaled dose was then derived by multiplying the
minute volume and the environmental air concentration
(Equation 3).
3.6 Benchmark Dose Analysis
The overall goal of benchmark dose analysis is to fit
a mathematical function that best describes the dose-
response relationship in the observable low dose region
of the data to enable extrapolation to doses lower than
those tested. Benchmark dose analysis estimates the
dose, termed a benchmark dose (BMD), for a specified
11
-------
level of benchmark dose response (BMR) observed. The
BMR is defined as the level of change in the response
rate. For example, a BMR of 10% would be equivalent
to a 10% response rate of the endpoint of interest. The
BMDS allows for the change in response rate to be
calculated as one of added or extra risk. Extra risk is
the increase in risk relative to the available risk; the
use of the extra risk calculation is recommended when
conducting benchmark dose analysis (U.S. EPA 2008).
However, the output of calculations for extra or added
risk is identical when conducting microbial dose-
response analysis when the background risk is assumed
to be zero.
EPA (2008) recommends a BMR value of 0.10 for
use with dichotomous data sets when deriving a point
of departure value, although users may make data-
specific determinations to select other values. For
this assessment, BMRs of 0.50, 0.10, and 0.01 were
selected for comparison of different model estimates at
various points in the dose-response relationship. These
values correspond to estimates of 50% lethality (i.e.,
LD50), 10% lethality, and 1% lethality, and the resulting
BMDs would be written BMD50, BMD10, and BMD01,
respectively. The lower levels of lethality (i.e., 1% and
10% levels) arc more appropriate for the selection of a
human equivalent dose than higher levels of response,
such as that exhibited by a BMD50. The primary BMDS
outputs of interest arc the BMD and the benchmark dose
limit (BMDL). The BMD is the dose that produces a
response at the level of the BMR. The BMDL is the 95%
lower statistical confidence limit of the BMD when the
95% lower confidence limit is applied to the estimated
slope parameter value.
For the benchmark dose evaluation, the current version
of EPA's BMDS (BMDS 2.1.1 Version 2.1.1.55)
(U.S. EPA 2009a) was used to fit models to the dose-
response data. Models from the BMDS dichotomous
and dichotomous-altcrnativc model suites were used
for analysis: the Weibull model, the Weibull model
run as exponential (with the power coefficient fixed as
one), probit. loge probit, logistic. loge logistic. Gamma
model, dichotomous Hill, probit-background response.
loge probit-background response, logistic-background
response, and loge logistic-background response. The
background parameter was directly specified as zero for
those models allowing this selection (i.e., loge logistic.
loge probit, Weibull, and Weibull run as exponential)
and the g parameter was specified as zero for the
dichotomous Hill model to ensure model fits did not
incorporate a background incidence of lethality.
The BMDS software places a number of default
restrictions on the slope and power values for some
models. These restrictions operate in the slope parameter
for the loge probit and loge logistic models, where the
value of the slope parameter is restricted to be equal
or greater than one. and in the power term for the
gamma, Weibull, loge logistic, and loge probit models,
where the value of the power term is restricted to be
greater than or equal to one. These restrictions prevent
the modeling of supra-linear response in the low dose
region. All default slope and parameter restrictions were
maintained in this analysis. Restrictions were maintained
based on recognition that historically used microbial
dose-response models (i.e., exponential. bcta-Poisson)
arc typically linear in the low dose region and arc
mathematically precluded from displaying supra-linear
behavior in this region. To test the potential impact
of unrestricted BMDS slope and model parameters
on model fits and output lethality values. Appendix
A compares results when varying the use of default
restrictions for the DI A Anthrax Data.
Statistically valid model fits and BMD values for a given
data set were identified using EPA guidance (U.S. EPA
2008). For each model, two BMDS outputs describing
the fit of an individual model to the data were evaluated:
the global goodness of fit as measured by the model-
calculated Chi-square p-value and the scaled residuals
calculated for each dose group. The p-value reflects the
overall goodness of fit. and a p-value of greater than 0.1
was used to identify a statistically valid fit. The scaled
residual is the difference between the model estimate
of response for an individual dose group relative to its
measured value. Scaled residuals closest to the BMD
arc of most concern for benchmark dose analysis as they
indicate the fit of the model to the data in the dose region
of greatest interest.
When comparing the fit of different models with valid
statistical fits and equivalent restrictions, the lowest
BMDL was selected when the calculated BMDLs were
not within a three-fold range (U.S. EPA 2008). However,
if the BMDLs were within a three-fold range, the
model with the lowest calculated value of the Akaike
Information Criterion (AIC) was selected. Since the
Chi-square p-values cannot be used to compare the
fits among different families of models or model with
differing numbers of parameters, the AIC value is more
appropriately used to compare fits across models. The
AIC value is calculated using the log-likelihood at the
maximum likelihood estimates for the model parameters
and the number of model degrees of freedom.
-------
4.0
Results
4.1 Statistical Description of Dose-
Response Sets
4.1.1 Department of Defense Anthrax Data Set
For each of the three studies comprising the DoD
Anthrax Data, there were 35 stations (with the exception
of one test using 36 stations), and a total of 285
monkeys had the potential for exposure to released
B. cmthracis spores. However, only those monkeys
where the co-located measurement devices captured B.
cmthracis spores were included in the dose-response
analysis. Additionally, available copies of the original
reports obtained through CBRNIAC included some
data points that were no longer legible. These data
points were removed from the data set if they could
not unequivocally be identified. Additional data were
removed at the higher doses (i.e., greater than 1,000
inhaled spores) to limit the inhaled doses to within a
2-log range to facilitate model fits to the data.
The selected data set consisted of 24 dose groups; 23
dose groups of five monkeys and one dose group of four
monkeys. To provide an indication of the distribution of
doses used for the benchmark dose analysis (i.e., 21 to
941 spores), inhaled doses were binned and displayed
in histogram form (Figure 2). The dose groups were not
evenly distributed across the range of doses, and the
higher doses in the range were underrepresented relative
to lower doses. Approximately 58% of the dose groups
in this data set were less than 400 spores. It was assumed
that the presence of these lower dose groups facilitated
more reliable dose-response fits in the lower response
regions of the curve.
Frequency
Cumulative %
Inhaled Spore Dose
Figure 2. Histogram and cumulative curve showing the frequency and cumulative percentage of the inhaled
doses in the DoD Anthrax Data.
13
-------
4.1.2 Defense Intelligence Agency Anthrax Data
Set
Individual dose-response data were provided for each
of the 34 monkeys in the study. The range of doses
was from 337 to 878,000 inhaled spores. To provide an
indication of the distribution of doses in the benchmark
dose analysis, the individual monkey inhaled doses were
binned and displayed in histogram form (Figure 3). In
sharp contrast to the DoD Anthrax Data, only 2.9% of
the individual doses were less than 400 spores.
Frequency
Cumulative %
Inhaled Spore Dose
Figure 3. Histogram and cumulative curve showing the frequency and cumulative percentage of the inhaled
doses in the DIA Anthrax Data.
14
-------
4.1.3 Druett et al. (1953) Anthrax Data Set
The data set consisted of a total of nine dose groups of
eight monkeys. The environmental air concentrations
tested ranged from 29,300 to 166,000 single spores per
liter air. To provide an indication of the distribution
of doses used for the benchmark dose analysis (i.e.,
35,000 - 198,000 spores), inhaled doses were binned
and displayed in histogram form (Figure 4). The Druett
Anthrax Data included inhaled doses considerably higher
than the other data sets. As a result, there were no doses
less than 400 inhaled spores and 22% of the doses were
less than 49,999 inhaled spores.
a.
3
o
i-
ID
O
a
aj
-Q
E
3
3
2
1
0
O)' O) O) O) O)' O)' O)' O)
V
-------
Table 3. Model Parameters, Standard Errors, 95% Conf dence Limits, and AIC Values for the Statistically
Signif cant Mathematical Model Fits to the DoD Anthrax Data.
Model
Slope
(Standard Error)
95% Conf dence
Limit
Intercept
(Standard Error)
95% Conf dence
Limit
Power
(Standard Error)
95% Conf dence
Limit
v and g
Parameters
(Standard Error)
95% Conf dence
Limit
AIC
Values
Value of Scaled
Residual Closest
to BMD1(|
Dichotomous Hill
(p=0.11)
2.85 (1.38)
0.144 to 5.55
-16.5 (7.48)
-31.1 to-1.82
Parameter Not in
Model
v: 0.576 (0.126)
0.329 to 0.822
g: Specified as 0
129.178
-0.7457
Loge Logistic
(p=0.13)
1.44 (*)
*
-9.44(*)
*
Parameter Not in
Model
Parameters Not in
Model
128.651
0.603
Gamma
(p=0.11)
0.00130
(0.000749)
-0.000167 to
0.00277
Parameter Not in
Model
1.24 (0.448)
0.368 to 2.12
Parameters Not in
Model
129.175
0.557
Weibull
(p=0.12)
0.000385
(0.000688)
-0.000963 to
0.00173
Parameter Not in
Model
1.14(0.284)
(0.582 to 1.70)
Parameters Not in
Model
129.274
0.516
LogeProbit
(p=0.11)
1 (NAT)
(NA)
-6.49(0.131)
-6.74 to -6.23
Parameter Not in
Model
Parameters Not in
Model
126.775
-0.748
Weibull as
Exponential
(p=0.18)
0.000913
(0.000154)
0.000610 to
0.00122
Parameter Not in
Model
Parameter
Specified as 1
Parameters Not in
Model
127.518
-0.595
* Standard Error not calculated by BMDS due to recognized error in its calculation
"I" NA signifies that the Standard Error and associated Confidence Limit were not calculated as parameter has hit a boundary condition.
Table 4. The BMD and BMDL at Identif ed BMRs for the DoD Anthrax Data.
BMR = 0.50
BMR =0.10
BMR = 0.01
Dichotomous Hill
BMD,0 = 630
BMDL,0 = 400
BMD10 = 190
BMDL10 = 98
BMD01 = 78
BMDL01 = 10
Loge Logistic
BMD,0 = 700
BMDL,0 = 540
BMD10 = 150
BMDL10 = 76
BMD01 = 28
BMDL01 = 6
Gamma
BMD,0= 720
BMDL,0 = 560
BMD10 = 140
BMDL10 = 89
BMD01 = 21
BMDL01 = 8
Weibull
gg
J2*0
II
II
S|
o °
BMD10 = 140
BMDL10 = 89
BMD01 = 17
BMDL01 = 8
Loge Probit
BMD,0 = 660
BMDL,0 = 530
BMD10 = 180
BMDL10 = 150
BMD01 = 64
BMDL01 = 51
Weibull as Exponential
BMD,0 = 760
BMDL,0 = 580
BMDl0 = 120
BMDLu = 88
BMDm = 11
BMDLm = 8
-------
4.2.2 Defense Intelligence Agency Anthrax Data
Set
The loge logistic and the dichotomous Hill BMDS
models exhibited acceptable fits as measured by p-values
and scaled residuals at BMDLs of interest (Table 5).
The calculated BMDL values did not vary by more than
three-fold at either the BMDL10 or the BMDL50 measures
for these models, and the log logistic model exhibited
the lowest AIC value (Table 5). The loge logistic
model calculated a BMDL50 of 4,900 inhaled spores
and a BMDL10 of 550 inhaled spores (Table 6). Model
parameters, confidence limits, and measures of fit are
provided in Table 5. Calculated BMDs and BMDLs for
identified BMRs are provided in Table 6. Figure 6 shows
the visual fit of the loge logistic model to the data.
Log Pro bit Model with 0.95 Confidence Level
LogProbit
50000 100000
dose
11:04 06/17 2010
Figure 5. Visual fit of probit model to the DoD Anthrax Data.
Table 5. Model Parameters, Standard Errors, 95% Conf dence Limits, and AIC Values for the Statistically
Signif cant Mathematical Model Fits to the DIA Anthrax Data.
Model
Slope (Standard
Error)
95% Conf dence
Limit
Intercept (Standard
Error)
95% Conf dence
Limit
v and g Parameters
(Standard Error)
95% Conf dence
Limit
AIC
Values
Value of Scaled
Residual Closest to
BMD„
Loge Logistic
(p=0.34)
!(*)
*
-9.23(*)
*
Parameters Not in
Model
36.814
-0.316
Dichotomous Hill
(p=0.48)
1 (NA)t
(NA)
-9.00 (0.706)
-10.4 to-7.62
v: 0.944(0.135)
0.679 to 1.21
g: Parameter
Specified as 0
38.636
-0.330
* Standard Error not calculated by BMDS due to recognized error in its calculation
"I" NA signifies that the Standard Error and associated Confidence Limit not calculated as parameter has hit a boundary condition
-------
Table 6. The BMD and BMDL at Identif ed BMRs for the DIA Anthrax Data.
BMR = 0.50
BMR =0.10
BMR = 0.01
Loge Logistic
BMD,0 = 10.000
BMDL,0 = 4.900
BMD10 = 1.100
BMDL10 = 550
BMD01 = 100
BMDL01 = 49
Dichotomous Hill
BMD,0 = 9.200
BMDL,0 = 3.500
BMD10 = 960
BMDL10 = 300
BMD01 = 87
BMDL01 = 26
Log-Logistic Model with 0.95 Confidence Level
Log-Logistic
1
0.8
0.6
0.4
0.2
0
BMDL BMD
0
100000
200000
300000
400000 500000
600000
700000
800000
900000
dose
10:12 05/05 2010
Figure 6. Log,, Logistic model for the DIA Anthrax Data.
18
-------
4.2.3 Druett et al. (1953) Anthrax Data Set
The tested BMDS models that exhibited acceptable fits
as measured by p-values and similar AIC values are
shown in Table 7. All tested BMDS models exhibited
acceptable fits to the data, with the exception of the
logistic-background response model. The calculated
BMDL50 and BMDL10 values did not vary by more
than three-fold across models with acceptable p-values
(p>0.1); therefore the model with the lowest AIC was
selected. This approach identified the Weibull model run
as exponential as the best fitting model. The Weibull run
as exponential calculated a BMDL50 of 37,000 inhaled
spores and a BMDL10 of 5,600 inhaled spores. Model
parameters, confidence limits, and measures of fit are
provided in Table 7. Calculated BMDs and BMDLs for
identified BMRs are provided in Table 8. Figure 7 shows
the visual fit of the log-logistic model to the data.
Table 7. Model Parameters, Standard Errors, 95% Conf dence Limits, and AIC Values Associated with a
Statistically Signif cant Model Fit to the Druett Anthrax Data.
Model
Slope
(Standard Error)
95% Conf dence
Limit
Intercept
(Standard Error)
95% Conf dence
Limit
Power
(Standard Error)
95% Conf dence
Limit
v and g
Parameters
(Standard Error)
95% Conf dence
Limit
AIC
Values
Value of Scaled
Residual Closest
to BMD1(|
Dichotomous Hill
(p=0.21)
2.30 (0.630)
1.06 to 3.53
-25.0(6.98)
-38.7 to-11.3
Parameter Not in
Model
v: 1 (NA|)
NA
g: Parameter
Specified as 0
78.4946
-0.9428
Loge Logistic
(p=0.21)
2.30 (*)
*
-25.0 (*)
*
Parameter Not in
Model
Parameters Not in
Model
78.4946
-0.943
Gamma
(p=0.25)
2.77E-05
(1.47E-05)
-1.05E-06 to
5.64E-05
Parameter Not in
Model
1.84(0.914)
0.0510 to 3.63
Parameters Not in
Model
77.9141
-1.07
Weibull
(p=0.25)
1.64E-07
(6.64E-07)
-1.14E-06 to
1.47E-06
Parameter Not in
Model
1.40(0.359)
0.693 to 2.10
Parameters Not in
Model
77.855
-1.106
LogeProbit
(p=0.23)
1.39 (0.358)
0.687 to 2.09
-15.1 (3.99)
-22.9 to -7.32
Parameter Not in
Model
Parameters Not in
Model
78.1857
-0.953
Weibull Run as
Exponential
(p=0.32)
1.44E-05
(2.35E-06)
9.81E-06 to
1.90E-05
Parameter Not in
Model
Parameter
Specified as 1
Parameters Not in
Model
77.1788
-1.568
Probit —
Background
Response
(p=0.26)
1.76E-05
(5.03E-06)
7.70E-06 to
2.74E-05
-1.0049(0.396)
-1.78 to -0.228
Parameter Not in
Model
Parameters Not in
Model
77.9739
-1.325
* Standard Error not calculated by BMDS due to recognized error in its calculation
"I" NA signifies that the Standard Error and associated Confidence Limit not calculated as parameter has hit a boundary condition
19
-------
Table 8. The BMD and BMDL at Identif ed BMRs for the Druett Anthrax Data.
BMR = 0.50
BMR =0.10
BMR = 0.01
Dichotomous-Hill
BMD,0 = 54.000
BMDL,0 = 40.000
BMD10 = 21.000
BMDL10 = 8.500
BMD01 = 7.200
BMDL01 = 1.400
Loge Logistic
BMD,0 = 54.000
BMDL,0 = 40.000
BMD10 = 21.000
BMDL10 = 8.500
BMD01 = 7.200
BMDL01= 1.400
Gamma
BMD,0 = 55.000
BMDL,0 = 40.000
BMD10 = 16.000
BMDL10 = 6.000
BMD01 = 4.200
BMDL01 = 580
Weibull
BMD,0 = 56.000
BMDL,0 = 40.000
BMD10 = 14.000
BMDL10 = 6.100
BMD01 = 2.700
BMDL01 = 580
Loge Probit
BMD,0 = 54.000
BMDL,0 = 40.000
BMD10 = 21.000
BMDL10= 11.000
BMD01 = 10.000
BMDL01 = 3.900
Weibull as Exponential
BMD,0 = 48.000
BMDL,0 = 37.000
BMD10 = 7.300
BMDL10 = 5.600
BMD01 = 700
BMDL01 = 540
Probit Background-Response
BMD,0 = 68.500
BMDL,0 = 57.000
BMD10 = 17.000
BMDL10 = 13.000
BMD01 = 2.000
BMDL01 = 1.400
O
'¦8
03
11:04 06/17 2010
Figure 7. Weibull run as exponential for the Druett Anthrax Data.
Weibull Model with 0.95 Confidence Level
Weibull
1
0.4
0.2
0
BMIpL
BMD
0
50000
100000
150000
200000
-------
5.0
Discussions
5.1 Variation in Dose-Response
Lethality Estimates
The results of the benchmark dose modeling for the
DoD Anthrax Data, the DIA Anthrax Data, and the
Draett Anthrax Data show no apparent consistency in
the calculated median benchmark response levels when
using the same model with different data sets and show
no apparent consistency with the previously published
values (Table 1) or reanalyses of original data (Table 2).
While obvious similarities in outputs of the probit and
logistic models can be shown in the results, these two
models often fit similarly to the same dichotomous data
set (U.S. EPA 2008). As one indication of the overall
variation in results, the best fitting models yielded
BMD,0 (BMDLJ0) values for the DoD, DIA. and Druett
Anthrax data sets of 660 (530), 10,000 (4,900) and
48,000 (37,000) inhaled spores, respectively. Even with
the use of criteria designed to increase the comparability
among the selected studies for this review, large
differences in derived values are still present.
To further understand reasons for these differences in
study results, a systematic approach to further evaluate
each element in the dose-response assessment process
is proposed (Figure 8). Elements of the dose-response
assessment process are defined to include: 1) physical
characterization of the spore product, 2) determination
of receptor-specific exposure assumptions, 3) selection
of dose metric and calculation of dose, 4) statistical
assessment of dose-response data, and 5) selection of
the dose-response relationship and identification of
point of departure values. The physical characterization
of the spore product and the determination of receptor-
specific exposure assumptions are conducted to inform
selection of the dose metric and the calculation of the
dose. After statistically testing the fit of mathematical
models to the dose-response data, the best fitting model
can be identified and the dose associated with the point
of departure identified.
Physical
Characterization of
Spore Product
Selection of
Dose Metric
and
Calculation
of Dose
Statistical
Assessment
of Dose-
Response
Data
Receptor-specific
Exposure
Assumptions
Figure 8, Dose-response assessment steps in the development of dose-response relationships.
21
-------
5.1.1 Physical Characterization of Exposure
Product
To evaluate identified B. anthracis dose-response
relationships across studies that tested different exposure
products, the products must be sufficiently described
to be comparable. The information must allow the
determination that the products arc either sufficiently
similar relative to the dosimetry of the receptor! s) being
evaluated, or the products must be characterized to allow
for the quantification of inhaled doses and deposited
doses.
From the perspective of test product characterization
for those data sets on which benchmark dose modeling
was conducted, the DoD Anthrax Data appears to be
the most suspect. Of the reviewed historical studies, the
Classman (1966) data likely also sutler from the same
measurement deficiencies as a similar process was used
in their study design. The only available particle size
data is the measurement of spore-containing particles
that were 5 p,M or smaller in size. However, neither
data set provides information to ascertain the median
particle size and associated distribution, the aerodynamic
diameter (i.e., no density information) or the spore per
particle measures.
Interestingly, the Classman (1966) data and the BMDS
reanalysis of the DOD Anthrax Data provide the lowest
estimates for LD50 and BMD50 values shown in Tables 1,
4, 6 and 8, respectively. One possible explanation for these
lower values is that particles greater than 5 pM in size are
contributing to exposure, but they arc not being counted
by the filtered inipingcr measurement process. Another
potential explanation is that there is an undcrcounting
of colonics from plates due to colony masking (Chang
et al. 1994), but a similar process was used for the DoD
Anthrax Data, Druett Anthrax Data, and Druett et al.
(1953) environmental air concentration measurements.
Therefore, colony masking is not likely to be causing the
differences in relative values observed for these data. For
both of the DoD Anthrax Data and Classman (1966) data,
there is little confidence in the generated values absent
more definitive particle size data.
The DIA Anthrax Data is an example of a current
research design that goes the furthest among
the reviewed studies in describing the physical
characteristics identified in the NRC framework (2008).
Specifically, the particle size of the aerosolized product
was measured in units of aerodynamic particle diameter
with values provided for a mass median aerodynamic
particle size and an associated geometric standard
deviation. Given the reported particle diameter, it is
highly likely that the exposure product utilized in the
DI A Anthrax Data consisted of primarily single spores.
The physical characterization and single spore nature
of the exposure product used to dose the animals is
also likely replicated by Vasconcelos et al. (2003) as
the studies were conducted within the same facility and
similar protocols were used to produce the material.
The Druett Anthrax Data, originally published in Druett
et al. (1953), were also obtained using a single spore
preparation based on information provided by the
original authors. When there is strong certainty that the
products were single spore with minimal aggregation of
spores during dosing, the use of single spore particles
obviates the need for particle measurement comparisons
among data sets.
When tested in the same animal species under
similar exposure scenarios, there should be relatively
small differences in the dosimetric potential, and
correspondingly, dose of the single spore products.
If the tested Bacillus strains have relatively similar
potencies, it should be expected that the calculated
doses associated with a given response level should be
relatively consistent. However, the Druett Anthrax Data
BMD50 value was approximately 4.7 times higher than
that of the DI A Anthrax Data. The calculated lethality
values in the DI A and Druett Anthrax Data should
be representative of the single spore dose associated
with lethality, if assertions regarding the single spore
nature of the particles arc correct. The next section will
consider the possibility that receptor-specific exposure
assumptions may also be contributing to differences in
the identified lethality values.
5.1.2 Receptor-specific Exposure Assumptions
The exposure assumptions for the minute volume used to
calculate inlialed dose varied considerably among those
studies reporting inhaled dose metrics. The assumed
minute volume varied from 1.2 liters per minute (Druett
et al. 1953) to 2.4 liters per minute (Bartrand et al. 2008,
Haas 2002) for studies described in Table 1 or Table 2. The
Druett Anthrax Data BMDS analysis assumed a minute
volume of 1.2 liters per minute (Equation 2). In contrast,
recent publislied studies (i.e., DI A Anthrax Data. Estep et
al. 2003, Vasconcelos et al. 2003) used plcthyniosgraphic
measurement techniques during testing that accurately
measured minute volume for each animal.
In three reanalyses of the original Druett et al. (1953)
data, estimates of the median lethality value varied
greatly. When comparing these estimates, the source of
the minute volume value should be carefully considered
(Table 9). For example, the minute volume of 2.4 liters
per minute assumed by Haas (2002) and Bart rand et al.
(2008) in their reanalysis was two times greater than
that the 1.2 liters per minute assumed for the BMDS
Druett Anthrax Data. The median lethality value when
-------
fitting the exponential model was 48,000 inhaled spores
(BMD50) using BMDS and ranged from 96,800 to 92,000
inhaled spores as published by Haas (2002) and Bartrand
et al (2008), respectively. The Haas (2002) and Bartrand
et al. (2008) median lethality values are 101% and 91%
greater than that of the Druett Anthrax Data BMDS
estimate for the same model, respectively. Differing
exposure assumptions can be shown to account for a
significant portion of the differences between the Druett
Anthrax Data BMDS results, Haas (2002), and Bartrand
et al. (2008) estimates.
Table 9. Comparison of Median Lethality Estimate and Assumed Minute Volume.
LD,8 or BMD„ Value Using Exponential Model
(Inhaled Spores)
Minute Volume Assumption (L/minute)
BMDS Druett Anthrax Data
48.000
1.2
Druett et al. (1953)
53.000
1.2
Bartrand et al. (2008)
92.000
2.4
Haas (2002)
96.800
2.4
In addition, it should be noted that allometric
relationships used to estimate minute volume do not
incorporate potential impacts on minute volume due to
the physiological state of the test animal (e.g., intense
stress, chemical restraint with tranquilizers). Since
the 1950's, most studies conducted using nonhuman
primates utilized some element of restraint (e.g.,
chemical, physical, or a combination of the two). An
early study (Berendt 1968) on the effect of straitjacket
physical restraints measured increases in the minute
volume of up to 200% beyond that derived by
application of Guyton's (1947) allometric equation.1
This increase was also corroborated by Jemski and
Phillips (1964) who described measured inhalation
rates of chimpanzees that were almost six times that of
the values estimated through use of Guyton's (1947)
equation. These inhalation rates were measured in
animals that were not sedated and were "securely
restrained in holding boxes specifically fabricated for
the size and species of animal involved." Conversely,
the use of chemical restraints (i.e., specifically, Telazol®)
is known to decrease the measured minute volume
from that calculated using Guyton's (1947) equation
by as much as 50% (Besch et al. 1996). Druett et al.
(1953), and by extension the Druett Anthrax Data
BMDS analysis, does not report the use or nonuse of any
restraints or acclimatization of the study animals. The
use of Guyton's, or any other allometric equation that is
not specific to the physiological state of the test animal,
may substantially underestimate the minute volume
of animals that are restrained and not sedated. The
underestimation of the minute volume is a significant
concern due to the resulting overestimate of the
calculated inhaled spore number associated with a given
level of response.
5.1.3 Selection of Dose Metric
Two dose metrics have been primarily used to describe
lethality, enviromnental spore concentration and inhaled
spore dose. Earlier studies (i.e., Dreutt et al. 1953,
Young et al. 1946) provided most, if not all, of their
lethality measures in dose units of enviromnental air
concentration. A number of studies, often conducted
more recently, used inhaled spores as the reported
dose metric for at least some of the reported data sets
(Druett et al. 1953, Estep et al. 2003, Glassman 1966,
Vasconcelos et al. 2003) (Table 1). Additionally, there
are reanalyses of older data sets that calculated inhaled
dose from available data (Table 2) (Bartrand et al.
2008, Haas 2002). The presence of both dose metrics
in the literature has sometimes led to inappropriate
comparisons of different dose metrics. For example.
Table 1 of Coleman et al. (2008) included lethality
values with both the dose metric of enviromnental air
concentration (Druett et al. 1953, Young et al. 1946) and
the dose metric of inhaled spores (Albrink and Goodlow
1959, Estep et al. 2003) identified as an inhaled dose.
However, Young et al. (1946) described the dose metric
as the "number of spores per unit of cloud when exposed
for 5 minutes" and the original Druett et al. (1953)
publication identified the dose metric as the "organisms-
minutes per liter of air." These two dose metrics as
described are enviromnental air concentrations. Overall,
the magnitude of difference between enviromnental air
concentration and inhaled dose lethality estimates will
vary based on the minute volume used to derive the
inhaled dose.
1 The allometric equation (Equation 2) used to derive minute volume is
based on Guyton's (1947) original data plus the addition of other data
sets developed after Guyton.
23
-------
5.1.4 Statistical Assessment of Dose-Response
Relationship
The published values for lethality estimates and
associated dose-response relationships (Tables 1 and 2),
and the BMDS results (Tables 3 through 8) were derived
using numerous approaches. Points of difference in
the approaches included the selection of mathematical
models (e.g., probit versus exponential) and the use of
the same mathematical model with different statistical
approaches (e.g., Finney's [1947] original probit
equation) or software (e.g., BMDS versus SAS®) to fit
model parameters with a given mathematical model. It is
expected that different mathematical models will derive
different parameter values and outputs for evaluated
dose-response relationships with an individual data set
or across different data sets. However, care should also
be exercised even when evaluating estimates developed
using the same mathematical model. For example, probit
slope models arc supported in the BMDS software and
were also utilized by early researchers, including Druett
et al. (1953). However, the equations and approaches to
derive slope parameters and coefficients arc considerably
different between the BMDS software and the original
Finney equations (Figure 9). Druett et al. (1953), using
the approach described in Finney (1947), fit log10
transformed dose-rcspon.se data to a probit model. The
output of the probit equation was in probit units and the
estimated parameter values were fit to the units of the
dependent variable. In contrast, the BMDS software
model incorporates a standard normal density function in
the equation used to fit the parameters and the response
variable units arc in probability, or percentage. The use
of the standard normal density function docs incorporate
an element of stochasticity in the BMDS modeling
process; values arc selected from the density distribution
as the model iterates until the data set is fit to the model.
While the distinction between the deterministic approach
of the original Finney (1947) approach and the stochastic
approach of BMDS may result in some relatively
minor differences in model results, it is likely that the
fundamental differences in the mathematical equation
structure arc driving the identified differences in fitted
parameters and intercepts.
When evaluating the use of fitted parameters for
application outside the original data set and statistical
approach, the dose data must be handled in a similar
manner as the original statistical analysis. For example,
the fitted slope parameter reflects the dose transformation
of the data originally used to derive it. A log10 dose
transformation will yield a different fitted parameter
for the same data as a loge or no dose transformation.
While these different dose transformations will yield
equivalent estimates once the dose is back-transformed
to its original units, the value of the slope factor cannot
be used to derive response estimates unless the slope
factor value is used in a similarly transformed equation.
These two previous considerations highlight the potential
hazard of using fitted parameter values generated using
one statistical approach as inputs into another statistical
approach, even when the same mathematical model is to
be used.
Equation used by Druett et al. (1953):
Response (Measured in Probits) = m(dose) + b
where in = slope and b = intercept
with use of companion table to transform probits to response percentage (Finney, 1947)
Equation used by BMDS software (U.S. EPA, 2009):
Response (Measuredas Probability) = ^ (a + p (dose))
where a = intercept, p = Slope, and = standard normal density function
Figure 9. The probit model equation used by Druett et al. (1953) and the BMDS software (U.S. EPA, 2009a) to
fit dose-response data.
Additionally, it should be noted that individual statistical
platforms (i.e., BMDS versus SAS®) may incorporate
differing default assumptions when fitting the same
mathematical model and that assumptions may not be
easily discerned when results arc reported. For example,
BMDS model fits for the probit model may assume a
background incidence of response or may incorporate a
default restriction that the value of the slope parameter
is greater than or equal to one (e.g., BMDS results for
DIA Anthrax Data, BMDS results for Druett Anthrax
Data). In contrast, the SAS probit model allows the user
to explicitly select for any model formulation whether a
background incidence is modeled and docs not default
to any restriction of fitted parameter values. As a result,
published parameter and lethality estimates should
be carefully evaluated in concert with the statistical
application and assumptions used to generate the
estimates.
-------
5.2 Using Available Nonhuman Primate
Data to Derive a Human Equivalent
Dose
Providing an adequate experimental animal dose-
response relationship is available, human equivalent
doses can then be developed. Human equivalent
doses typically take the form of a medium-specific
concentration (e.g., environmental air concentration)
or a surrogate measurement that serves as a proxy for
the medium of interest (e.g., surface concentration that
can be related to an air concentration). Depending on
the cleanup authority used, human equivalent doses
will support the development of site-specific cleanup
goals that will also reflect both technical and other
considerations (e.g., CERCLA's nine criteria for actions
conducted under the National Contingency Plan). The
following example will describe one potential technical
approach to derive a human equivalent dose but will
not consider other criteria that are also integral to the
development of site-specific or situation-specific cleanup
goals and risk management decisions. Additionally,
the use of uncertainty or variability factors, as is
typically used by EPA for reference dose or reference
concentration determination from experimental animal
data, has not been explicitly incorporated in this example
calculation. As such, the resulting calculation should not
be construed as standard setting or as a determination of
the appropriate level of B. anthracis risk management.
There are two technical elements of a human equivalent
dose that will be assessed in the development of a
cleanup goal subsequent to an aerosolized release of B.
anthracis spores. The first element is an interspecies
extrapolation process to account for dosimetric
differences between the test animal and human receptor.
The output of this process is the medium-specific
concentration (i.e., environmental air concentration)
that is assumed to produce an equivalent response in
humans as that exhibited by the test animal. The second
element is the conversion of the derived enviromnental
air concentration to a measurement more amenable
for use in sampling. It lias been assumed that a surface
concentration, as measured by viable recoverable spores
from a surface wipe, will be used for B. anthracis
spore surface sampling. Figure 10 traces the general
approach to combine the interspecies extrapolation
with the subsequent derivation of the sampling wipe
measurement.
The interspecies extrapolation process combines a
dosimetric evaluation to derive the applied dose from
the enviromnental air concentration for the animal
receptor with an accompanying dosimetric evaluation
to derive the enviromnental air concentration for the
human receptor from the applied dose. The dosimetric
adjustment process for inhalation exposures (Figure 10)
exhibits considerable overlap with elements of the EPA
(1992) exposure assessment process (Figure 1). While
there is considerable uncertainty in the interspecies
extrapolation process for microbial hazards, dosimetric
adjustment begins to address some known elements
of uncertainty in interspecies extrapolation. However,
the adjustment does not explicitly consider differential
susceptibility among species or differences in the
sequence of disease events at the cellular or molecular
levels (i.e., the microbial equivalent of the toxic dynamic
elements of uncertainty for chemical hazards). This
area is ripe for further evaluation, but was identified as
beyond the scope of this current study.
Interspecies Extrapolation
Point of Exposure
Animal
Environmental
Receptor
Concentration
Potential Dose
Animal
Inhaled Dose
BMDL10 = 550
Inhaled Spores
Deposited Dose
to Alveolar
Region of Animal
Lung
66 Deposited
Spores
Remedial Target for Surface
Wipe Measurement
Remedial Target Surrogate Measurement
Wipe
Measurement
2,500
Spores/Wipe
Environmental|
Surface
Concentration 1
860
Spores/100 cm2
Point of Exposure
Environmental
Air
Concentration
19,000
Spores/m3
Potential Dose
Human
Inhaled Dose
330 Inhaled
Spores
Deposited Dose
to Alveolar
Region of
Human Lung
66 Deposited
Spores
Human
Receptor
Figure 10. Generalized approach to calculate a sampling wipe measurement from animal dose-response data for
inhaled B. anthracis spores.
-------
The DIA Anthrax Data was selected for use in the
calculation of a human equivalent dose and wipe
measurement. The loge logistic model was identified as
the best fitting model for these data, and the BMDL10
value of 550 inhaled spores (Table 6) was selected as the
point of departure. This data set was chosen because it
was conducted with a superior study design that included
a 120 day observation period subsequent to exposure
and state-of-the-art measurement technologies for both
inhalation rate and particle si/c. The identified point of
departure value was then used for the animal inhaled dose
of the interspecies extrapolation. Although the benchmark
dose analysis did not utilize the dose metric of deposited
dose, the deposited dose was included in the following
discussion since its consideration is an important
component of the interspecies extrapolation process.
For the monkey, a particle depositional value of 12% for
1 to 2 p,M particles was assumed (Cheng et al. 2008).
This percent deposited value allowed for the calculation
of the deposited dose of 66 spores from the animal
inhaled dose of 550 spores. It was assumed that the
same deposited dose in the monkey and human would
result in equivalent levels of response. The deposited
dose of 66 spores for the human receptor was then
used to derive the human inhaled dose, using a human-
specific depositional rate of 20%. The depositional
value of 20% was based on the higher end of the range
of human depositional values for 1 to 2 \iM particles
(Figure 6-6 in U.S. EPA2004). Depositional rates may
differ based on the animal receptor used in the exposure
studies (e.g., guinea pig versus monkey), particle si/c
of the acrosoli/cd product, inhalation rate, and mode of
breathing (e.g., oral versus nasal) (Jarabck et al. 2005,
U.S. EPA 2004).
The conversion of the human inhaled dose to an
environmental air concentration requires the assumption
of an exposure duration and minute volume. For
simplicity, the exposure duration assumed was 1 minute.
The human receptor minute volume assumed in the
human equivalent dose calculation was 17 liters per
minute and was reflective of human adults undertaking
light activity (Table 5-23 in U.S. EPA 1997). Given
known relationships between age. si/c, activity level,
and minute volume; the selected human minute volume
should be representative of the exposure conditions in
which the human equivalent dose will be applied. In
contrast to the 17 liters per minute value assumed for
human receptors, the measured minute volume value for
sedated rhesus monkeys in the DI A Anthrax Data Set
was 0.5 to 1.0 liters per minute (Barnewall et al. 2001).
The calculated environmental air concentration resulting
in the same level of response would also be expected
to vary considerably between these receptors given the
difference in minute volume rates between humans and
rhesus monkeys.
With the assumption of a human minute volume of
17 liters per minute, the inlialcd human dose of 330
inhaled spores was then converted to an environmental
air concentration specific for the human receptor.
The resulting environmental air concentration was
19,000 spores per cubic meter. In lieu of reliance on
the calculated environmental air concentration as the
human equivalent dose, an environmental surface
concentration as measured by a wipe was selected.
The relative ease in sampling the surface concentration
when compared to the air medium drives the choice
of the environmental surface concentration value as a
human equivalent dose. To derive the environmental
surface concentration, assumptions arc necessary
regarding the expected deposition of airborne particles
to the sampling surface, the arcal extent of the sampling
surface, and the efficiency at which these particles can
be removed from the surface and recovered from the
wipe sample. There arc a number of different mode ling
approaclies that can be used to estimate surface
concentration from environmental air concentration
(e.g., Price 2009). However, it is beyond the scope of
this assessment to evaluate the available deposition and
resuspension models. A conservative proxy method to
derive the surface concentration from the known air
concentration has been described as one means to derive
a surface concentration. Using this approach. 90% of
the airborne spores, as measured by the environmental
air concentration, arc assumed to drop to a horizontal
surface and be available for surface wipe sampling. The
remaining 10% of the spores arc assumed to deposit on
vertical surfaces or remain suspended in the air. With
the assumption of an air volume of 75 cubic meters, the
calculated surface concentration using this approach
would be 860 spores per 100 square centimeters.
The surface concentration can then be converted to an
estimated spore number that can be recovered from
a wipe after sampling a 930 cm2 (i.e., 12 inch by 12
inch) surface area of a specified material type. Estill et
al. (2009) identified a 31% recovery efficiency for B.
anthracis deposited spores on a stainless steel surface
when using a moistened wipe. The 31% recovery
efficiency is used here to calculate a recovered spore
count from a surface wipe of 2,500 spores based on
the surface concentration of 860 spores per 100 square
centimeters. The resulting surface wipe measurement
is indicated by a recovered viable spore count from the
surface wipe of 2,500 spores.
-------
6.0
Conclusion
In summary, it has been shown that the EPA's BMDS
is an important tool for evaluating microbial dose-
response data, including that of B. anthracis inhalation
exposures. As with all statistical software applications,
users must identify the assumptions incorporated within
the software and the mathematical models it supports.
However, this concept is equally important for the
evaluation of published dose-response data and the
application of these data to the development of human
equivalent doses to support site-specific cleanup goals.
This study found that a number of disparities in the
literature for B. anthracis lethality estimates could be
traced to differences in physical characterization of the
spore product, receptor-specific exposure assumptions,
the calculated dose metric, and the statistical process
employed to assess the data. One area that consistently
has received less attention in study design has been
the determination of spore number per particle. The
reliance on data sets using single spore particles may be
an appropriate means to by pass this concern. However,
lack of these data or sufficient confidence that exposure
products arc indeed single spore particles can hinder
confidence in historical and even more recent data sets.
Knowledge of these contributors to the variability in
published estimates may facilitate common agreement
on a dosc-rcspon.se relationship based on a data set that
best characterizes these elements. As has been noted
previously, the NRC (2008) framework would provide
an excellent guide for proper characterization of products
used in the development of ideal dose-response data.
As a companion to the product characterization, the
receptor-specific exposure assumptions and dosimetric
evaluation should also receive equivalent consideration
in the study design of the ideal dose-response data set.
With an accepted B. anthracis inhalation animal dose-
response relationship, human equivalent doses can
then be developed. The development of these human
equivalent doses from test animal dose-response data
requires explicit evaluation of the dosimetric differences
between the test animal and the human receptors to
properly conduct an interspecies extrapolation. Again,
those data elements (i.e., phy sical characterization of the
spore product, receptor-specific exposure assumptions,
and particle-size specific depositional data for both
receptors of interest) that arc critical in the development
of the dose-response relationship arc also important to
the extrapolation process.
27
-------
28
-------
7.0
References
Albrink. William and Robert Goodlow. 1959.
Experimental Inhalational Anthrax in the Chimpanzee.
Am J Path. 35(5): 1055-1065.
Armstrong, T.VV. and C.N. Haas. 2007. A Quantitative
Microbial Risk Assessment Model for Legionnaires'
Disease: Animal model selection and dose-response
modeling. Risk Anal. 27(6): 1581-1596.
Barnewall. Roy. James Estep, and Robert DeBell. 2001.
Inhalation Median Lethal Dose (LD50) Determinations
in Rhesus Monkeys Exposed to Bacillus anthracis.
Final Report. Study Number CG463810D. Prepared
for Defense Intelligence Agency by Battelle Memorial
Institute. For Official Use Only.
Bart rand, Timothy. Mark H. Weir, and Charles N. Haas.
2008. Dose-Response Models for I nhalation of Bacillus
anthracis Spores: Interspecies Comparisons. Risk Anal.
28(4): 1115-1124.
Berendt. Richard F. 1968. The Effect of Physical and
Chemical Restraint on Selected Respiratory Parameters
ofMacaca Mulatto. Lab Anim Care. 18(8): 391-394.
Besch, Terry K., David Ruble. Paul Gibbs. M. Louise
Pitt. 1996. Steady-State Minute Volume Determination
by Body-Only Plethysmography in Juvenile Rhesus
Monkeys. Lab Anim Sci. October 1996. 46(5):539-544.
Bliss, C.I. 1935.The Calculation of the Dosage-Mortality
Curve. Ann. Appl. Biol. 22:134-167.
Chang, Ching-Wen. Yaw-Hui Hwang, Sergey
A.Grinshpun. Janet M. Macher, and Klaus Willeke.
et al. 1994. Evaluation of Colony Counting Due to
Colony Masking in Bioaerosol Sampling. Appl Environ
Microbiol. 60(10): 3732-3738.
Cheng. Y.S., H. Irshad. P. Kuehl. T.D. Homes, R.
Sherwood. C.H. Hobbs. 2008. Lung Deposition of
Droplet Aerosols in Monkeys. Inhal Toxicol. 20:1029-
1036.
Coleman. Margaret E., Brandolyn Thran, Stephen S.
Morse. Martin Hugh-Jones. Stacey Massulik. 2008.
Inhalation Anthrax: Dose response and risk analysis.
Biosecur Bioterror. 6(2): 147-159.
Druett. H.A., D.W. Henderson. L. Packman, and S.
Peacock. 1953. Studies on Respiratory Infection. I. The
influence of particle si/c on respiratory infection with
anthrax spores. J of Hygiene. 51:359-371.
Englehardt. James and Jeff Swartout. 2006. Predictive
Baycsian Microbial Dose-Response Assessment based
on Suggested Sclf-Organi/ation in Primary Illness
Response: Cryptosporidium parvum. Risk Anal. 26(2):
543-554.
Estep, J.E., R. Barnewall R., R. DeBell R., and N.
Niemuth, N. 2003. Inhalation median lethal doses of
Bacillus anthracis, Ames and Vollum strains, in the
rhesus monkey. Toxicologist. 62:161.
Estill, Cheryl. Paul Fairfield.. Paul. A. Baron,. Jeremy
K. Beard.. Misty J. Hein. Lloyd D. Larsen. Laura Rose.
Frank W. Schaefer III. Judith Noble-Wang, Lisa Hodges.
H.D. Alan Lindquist. Gregory Deye, and Matthew
J. Arduino. 2009. Recovery Efficiency and Limit of
Detection of Acrosoli/ed Bacillus anthracis Sterne
from Environmental Surface Samples. Applied Environ
Microbiol. July 2009. 75(13):4297-4306.
FAO and WHO. 2003. Hazard Characterization
for Pathogens in Food and Water Guidelines.
Microbiological Risk Assessment Series, No. 3. Food
and Agriculture Organization of the U nited Nations.
World Health Organization. Accessed from http://www.
fao.org/docrep/OQ6/y4666e/y4666eOO.Mm on August 28,
2008.
Finney, D.J. 1947. ProbitAnalysis. A statistical treatment
of the sigmoid response curve. Cambridge University
Press. Cambridge. UK.
Fritz, David L„ Nancy Jaax. Wade Lawrence, Kelly
J. Davis, Margaret Pitt. John Ezzell, and Arthur
Friedlandcr. 1995. Pathology of Experimental Inhalation
Anthrax in the Rhesus Monkey. Laboratory Invest.
73(5): 691-702.
Classman, Harold N. 1966. Industrial Inhalation Anthrax
- Discussion. BacteriolRev. 30(3): 657-659.
Gutting, Bradford. W. Stephen R. Channel. Alan E.
Berger. Jeffrey M. Gearhart. George A. Andrews.
Robert L. Sherwood, and Tonya L. Nichols. 2008.
Mathematically modeling inhalational anthrax. Microbe.
3(2): 78-85.
Guv ton A. 1947. Measurement of the Respiratory
Volumes of Laboratory Animals. Am. J. Phy siol. 150:70-
77.
Haas. Charles. 2002. On the Risk of Mortality to
Primates Exposed to Anthrax Spores. Risk Anal. 22(2):
189-193.
29
-------
Henderson. D.VV. 1952. An Apparatus for the Study of
Airborne Infection. J. Hyg. (London). 1952;50(1): 53-68.
Haas, C.N., IB. Rose IB., C.P. Gerba., C.P. 1999.
Quantitative Microbial Risk Assessment. New York. John
Wiley and Sons.
Hilnias. Corey.. Alexandre M. Katos. Patrick T.
Williams, and Jaime Anderson. 2009. "Anthrax" in
Handbook of Toxicology of Chemical Warfare Agents.
pp. 433-462. Academic Press.
Holcomb. David L., Mary A. Smith. Glenn O. Ware,
Yen-Con Hung, Robert E. Brackett and Michael P.
Doylcct al. 1999. Comparison of Six Dose-Response
Models for Use with Food-Borne Pathogens. Risk Anal.
19(6): 1091 --1100.
Inglcsby. Thomas V.,; Tara OToole. Donald A.
Henderson, John G. Bartlctt. Michael S. A seller, Edward
Eitzcn, Arthur Fricdlandcr. Julie Gerberding. Jerome
Haucr, James Hughes. Joseph McDade, Michael T.
Osterholm. Gerald Parker. Trish M. Perl. Philip K.
Russell, and Kevin Tonat; for the Working Group on
Civilian Biodcfcn.se. 2002. Anthrax as a Biological
Weapon, 2002: Updated Recommendations for
Management. May 1, 2002. JAMA. 287(17): 2236-2252.
Jansscn, R E. Jr. 1955a. Trial Report. BWAL. 6A-3-54.
Operation "Jungle Boy. " AD596073. CBRNIAC No.
CB022712. Distribution Limited to U.S. Govt. Agencies
Only.
Jansscn. R E. Jr. 1955b. Trial Report. BWAL. 6A-4-54.
Operation "Jungle Boy. " AD596081. CBRNIAC No.
CB022713. Distribution Limited to U.S. Govt. Agencies
Only.
Jansscn. RE. Jr. 1955c. Trial Report. BWAL. 6A-5-54.
Operation "Jungle Boy." AD596083. CBRNIAC No.
CB022714. Distribution Limited to U.S. Govt. Agencies
Only.
Jarabck, Annie M„ Bahnian Asgharian. and Frederick J.
Miller. 2005. Dosimetric Adjustments for Interspecies
Extrapolation of Inhaled Poorly Soluble Particles (PSP).
Lnhalation Toxicol. 17: 317-334.
Jcmski. Joseph and G. Briggs Phillips. 1964. Aerosol
Challenge of Animals. Technical Manuscript 144. U.S. Army
Biological Laboratories. Fort Detrick. Frederick. MD.
Mcsclson, Matthew. 1995. A Note Regarding Source
Strength. ASA Newsletter. June 8, 1995. 95-3(48).
Moon, Hojin, Hyun-Joo Kim, James JL. Chen, and
Ralph Kodcll. 2005. Model Averaging Using the
Kullback Information Criterion in Estimating Effective
Doses for Microbial Infection and Illness. Risk Anal.
25(5): 1147-1159.
Moon. Hojin. James Chen, David Gay lor. and Ralph
Kodcll. 2004. A comparison of microbial dosc-rcspon.se
models fitted to human data. Regul Toxicol Pharmacol.
40(2): 177-184.
National Research Council. 2008. A Framework for
Assessing the Health Hazard Posed by Bioaerosols.
Committee on Determining a Standard Unit of Measure
for Biological Aerosols. The National Academics Press.
Washington. D C.
National Research Council. 2006. Overcoming
Challenges to Develop Countermeasures Against
Aerosolized Bioterrorism Agents: Appropriate Use
of Animal Models. Committee on Animal Models for
Testing Interventions against Acrosoli/cd Bioterrorism
Agents. The National Academics Press. Washington,
D C. Accessed from http://www.nap.edu/cataloeJ11640.
html on July 7, 2009.
Pitt, M. Louise and Ross D. LcClairc. 20047.
"Pathogenesis by Aerosol" in Biological Weapons
Defense: Lnfectious Disease and Counterbioterrorism.
Ed. by Lindlcr, Luther E„ Frank J. Lcbeda, and George
W. Korcli. pp. 65-78. Humana Press. Totovva, NJ.
Price. Philip. K. Haniachi. J. Mc Williams, J., M. Sohn.,
M. 2009. Anthrax Sampling and Decontamination:
Technology Trade-Offs. (March 2, 2009). Environmental
Energy Technologies Division. Lawrence Berkeley
National Laboratory. Paper LBNL-1519E. Accessed
from http://repositories.cdIib.ora/ibnl/LBNL-1519E on
September 24, 2009.
U.S. APHC. 2010. Technical Guide 316 Supplement C2:
Dose-Response for Inhalation Anthrax in Guinea Pigs
using Historical Army Data and Classical and B aye si an
Statistical Methods. U.S. Army Public Health Command.
Unclassified/For Official Use Only.
U.S. EPA. 2010. The Pathogen Information Catalog (PL
Cat) Tool to Support Dose-Response Assessments. Dosc-
Rcspon.se Knowledge Base - Pathogen Information
Catalog. Technical Brief. U.S. Environmental Protection
Agency. EPA/600/S-08/029A.
U.S. EPA. 2009. Benchmark Dose Software. BMDS 2
.1.1 Version 2.1.1.55. U.S. Environmental Protection
Agency, National Center for Environmental Assessment.
Release Date: November 9, 2009.
U.S. EPA. 2008. Benchmark Dose Software (BMDS) On-
bmds/bmds_training/methodology/intro.htm#Decision
on August 28, 2008.
-------
U.S. EPA. 2004. Air Quality Criteria for Particulate
Matter. Volume 11 of II. EPA/600/P-99/002bF. U nitcd
States Environmental Protection Agency. Washington
DCD.C.. October 2004. EPA/600/P-99/002bF.
U.S. EPA. 1997. Exposure Factors Handbook. (Final
Report). 1997. EPA/600/P-95/002F a-c. Unitcd States
Environmental Protection Agency. Washington, DCD.C..
EPA/600/P-95/002F a-c.
U.S. EPA. 1992. Guidelines for Exposure Assessment.
EPA/600/Z-92/001. May 1992. Risk Assessment
Forum. Unitcd States Environmental Protection Agency ,
Washington, DCD.C. 1992. EPA/600/Z-92/001.
U.S. EPA. 1988. Recommendations for and
Documentation of Biological Values for Use in Risk
Assessment. Unitcd States Environmental Protection
Agency . EPA 600/6-87/008. Available at http://cfpub.epa.
gov/ncea/cfm/recordisplav.cfm?deid=34855. Accessed
on May 16, 2008.
Vasconcelos, Daphne. Roy Barncwall. Michael
Babin. Robert Hunt. James Estcp. Carl Nielsen,
Robert Carncs. and John Carney. 2003. Pathology of
Inhalation Cynoinolgus Monkey s (Macacafascicularis).
Laboratory Invest. 83(8): 1201-1209.
Wilkcning, Dean A. 2006. Sverdlovsk revised: Modeling
human inhalation anthrax. PNAS. 103(20):7589-7594.
Young George A., MR. Zclle M R., and Ralph Lincoln.
Ralph. 1946. Respiratory Pathogenicity of Bacillus
anthracis Spores I. Methods of study and observation of
pathogenesis. J. Infect Dis. 79(3): 233-246.
Zaucha. Gary M., Louise Pitt. James Estcp. Bruce
Ivins. and Arthur Fricdlandcr. 1998. The Pathology of
Experimental Anthrax in Rabbits Exposed by Inhalation
and Subcutaneous Inoculation. Arch Path Lab Med.
November 1998. 122: 982 - 992.
-------
32
-------
Appendix A
Evaluation of Benchmark Dose
Software Results with Unrestricted
versus Restricted Slope Values and
Zero versus Nonzero Background
Values
Two BMDS modeling assumptions were tested with the
DIA Anthrax Data using EPA's BMDS (BMDS Beta
Version 2.1.0.4) (U.S. EPA 2009). The following models
were used in the evaluation: probit, logistic, loge probit,
loge logistic, Weibull, Dichotomous-Hill, and Weibull
run as exponential. The first modeling assumption tested
was the impact of allowing for a nonzero background
value in the incidence of lethality for those models
that allowed it (i.e., probit and logistic). Statistically
significant model fits to the data were identified and the
modeled BMDL50 and BMDL10 values were generally
higher than estimates that did not allow for a background
incidence (Tables D-l and D-2). However, a direct
comparison of models that were able to be fit both with
and without an assumed background incidence is not
available. The probit and logistic models that could be
fit with a nonzero background were not able to fit to the
data when the background was set at zero.
The second modeling assumption tested was a
comparison of the lethality estimates obtained when
restricting the value of the slope parameter to 1.0 or
less. For the DIA Anthrax Data, model fits were able to
be identified for models with and without the restricted
slope value (Tables D-l and D-2). Depending on the
individual model compared (i.e., loge logistic versus
Dichotomous-Hill), differences between these estimates
varied by a factor of 2 or less.
33
-------
Weibull
Run as
Exponential
Zero
T» 1 1
3
I
51
¦3
Q
NF
Dichotomous
Hill
Zero
Background
Restricted
Slope
9,160
3,550
Dichotomous
Hill
Zero
Background
Unrestricted
Slope
9,820
3,270
Weibull
Zero
Background
NF
NF
Log Logistic
Zero
Background
Unrestricted
Slope
9,820
3,270
-a
si '
§? >N Jf
-1 1 =
it
s ^
=
10,200
4,940
Log
Pmbit
Zero
Background
i a;
s
J
10,100
3,340
Log
Pmbit
Zero
Background
T1 / * ,U- 1
* £-
NF
NF
LX)glSUC
Nonzero
Background
¦
j
55,900
30,200
Probit
Nonzero
Background
j
58,800
34,300
Data Set
l
j'
DIA Anthrax
Data Set
Weibull
Run as
Exponential
Zero
Background
NF
Dichotomous
Hill
Zero
Background
Restricted
Slope
963
302
Dichotomous
Hill
Zero
Background
Unrestricted
Slope
069
x>
Weibull
Zero
Background
NF
NF
Log Logistic
Zero
Background
Unrestricted
Slope
069
Log,
Logistic
Zero
Background
Restricted
Slope
1,140
549
-a
«a|
m s
O p
05 P
13 S
-------
-------
SEPA
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
------- |