U.S EPA, Health Canada, California
EPA, USDA, registrants of pesticides, food
companies, university researchers, and the
scientists at Exponent, Inc. are currently
using DEEM and Calendex.
Despite rumors the contrary,
EPA still makes extensive use of the DEEM
and Calendex programs to conduct exposure
assessments. EPA has told us that that it
is delighted that it now has several different
exposure programs available for its use
and that each of these programs brings a
somewhat different perspective to aggregate
and cumulative exposure analysis. The main
criteria that EPA uses in deciding which
programs are acceptable are its public availability,
SAP review, and openness of the computational
algorithms and databases used by the programs.
DEEM and Calendex qualify for use on all
three of these counts. (EPA has spent thousands
of hours of Quality Assurance (QA) time
in its review and use of DEEM and Calendex.)
No. The computational algorithms
used by DEEM and Calendex to generate exposure
amounts for individuals and report their
mean and percentile distributions were clearly
and publicly exposed at the time that these
programs were submitted for SAP approval
in 2000. Much of this information is still
available through EPA’s web pages.
[Link] Now all of the DEEM and Calendex
computational algorithm can be accessed
from this Durango Software website. Older
version of DEEM and Calendex (before the
current FCID versions) used proprietary
Novigen food translation factors to convert
foods as eaten in the CSFII to RACs (raw
agricultural commodities) and foodforms
as required for a dietary exposure analysis.
Even though the Novigen conversion factors
were proprietary, the actual intake calculations
for each individual were included in the
DEEM CEC (critical exposure calculation)
report. However, with the public release
of the USDA/EPA FCID database (where foods-as-eaten
have already been converted to RACs and
foodforms), and the incorporation of this
database into the FCID versions of DEEM
and Calendex, these FCID versions of these
programs are now completely transparent.
Of the three aggregate and cumulative exposure
programs now available to the risk assessor,
only DEEM/Calendex continues to base its
analyses directly on the publicly available
CSFII dietary intake surveys, which EPA
has been using for over 20 years in performing
dietary assessments and setting tolerances
for pesticide use.
The most significant limitation
of DEEM and Calendex at present and in the
foreseeable future is that there is no suitable
database of long-term dietary intake data
that follows a statistically significant
number of individuals with a broad range
of demographic characteristics.
Since the early 1980’s
EPA has relied on several USDA CSFII surveys
that follow individuals for only two or
three (non-consecutive) days. Moreover,
it is unlikely that the CSFII surveys will
be continued in the future; instead they
will be replace by the NHANES survey which
only contains one day of intake for most
individuals.
Research is critically needed
to develop alternative databases for estimating
long-term intake amounts for individuals
of all ages. We also need a reliable, statistically
significant, and publicly available database
with information related to the many potential
uses of pesticides in the residential environment,
including application rates, frequency and
timing, and the co-occurrences of these
applications in geographically distinct
regions of the country.
And just as importantly, we
need quantitative information related to
range of human contact with these chemical
residues in the residential environment,
especially for infants and children, which
can be used with calendar models. Calendex
cannot conduct chronic exposure analyses
for periods of time longer than one calendar
year.
However, when calculating
exposure for each individual at any time
in the current year, Calendex assumes that
an identical treatment schedule was used
in the prior year which could result in
active residues still present in the current
year. (Calendex combines any exposure from
those earlier applications with the exposure
from treatments in the current year.)
Finally, one must recognize
that, by its nature, conducting cumulative
and aggregate exposure analyses is a complex
and data intensive process that must be
conducted with utmost care and professional
expertise to avoid “garbage in-garbage
out” results.
What
is the best way to start learning to use
DEEM and Calendex?
DEEM is basically an easy
program to learn to use. When first learning
to use this program you should use single
estimates of residue levels for RACs and
foodforms. Once you have mastered this level,
then you can begin to include residue distribution
files and perform cumulative analysis (multiple
chemical residues found on the same foods).
There are no calendar considerations in
DEEM (unlike Calendex), other than the possibility
of restricting exposure analyses to specific
seasons of the year based on the seasonality
of the dietary intake data (for example,
performing an acute exposure analysis for
the summer months using only CSFII summer
participants, and residue data representative
of foods available in the market during
the summer months). The most difficult aspect
of a dietary analysis conducted with DEEM
is determining which foods and foodforms
should be included in the analysis and matching
appropriate residue data (either as deterministic
values or in a distribution) with those
foods and foodforms, before beginning to
enter data into the DEEM residue file editor
itself.
Once a dietary residue file
is set up, performing a dietary exposure
analysis with DEEM is relatively fast and
easy. When running your first analyses with
any residue file, start with a constrained
population (e.g., children 1-2, males only,
in one region only) and a small number of
Monte Carlo iterations, to see if your initial
results make sense. In general, it does
not make sense to run large numbers of iterations
(>100) until you are satisfied that you
have set up the analysis correctly and are
finalizing your analysis (e.g., reporting
for public policy making or submitting an
application for a tolerance).
Calendex is a much more demanding
program than DEEM, partly because of it
calendar-based data requirements but mostly
due to the fact that you are evaluating
a number of different pesticide uses in
the residential environment in addition
to dietary exposure, all in a probabilistic
sense. This means that you must be able
to describe the application and contact
frequencies and amounts as distributions
rather than as deterministic amounts so
that the results of the analysis can be
expressed as a percentile distribution of
exposure amounts in a given subpopulation
rather than as average or worst-case point
estimates.
The dietary exposure analysis
is the easiest part of a Calendex analysis
if you have already used DEEM to compute
dietary exposures, because Calendex simply
imports the DEEM dietary residue file with
the list of target foods and foodforms and
their associated residues (most appropriately
expressed as distributions). However, we
recommend that you begin your initial Calendex
analyses with one simple “AGX”
file (a file which describes (probabilistically)
application frequency and date(s), residue
amounts, and contact amounts for a single
application type, e.g., a lawn care product).
Set the residue amount to 1, the residue
persistence to one week, and the contact
amount to 1 to begin with, so that the exposure
amount will be forced to 1 (before division
by body weight). Run some preliminary analyses
for different times of the year and verify
that the reported exposures occur only in
the period of application, with exposures
equal to 1.0 for those individuals who are
exposed. Next, extend the period of residue
persistence to several months and then rerun
your analyses to see that the timing and
amount of exposures follow accordingly.
Change one parameter at a time in order
to verify that the calculations follow the
changes. After you gain confidence in the
program and your ability to use it to model
calendar-based applications and contacts,
you can begin to incorporate distributions
of residue and contact amounts in your AGX
file. Build and test one AGX file at a time
(one for each residential chemical application
type to be aggregated) before you begin
to run multiple AGX files in your analysis.
Finally, after you are satisfied with your
residential exposure analysis, include the
DEEM dietary residue file that you wish
to aggregate with your residential applications.
If you proceed systematically with your
analysis and begin by verifying its individual
components separately, you will be far more
likely to generate meaningful results than
by starting out by building and running
all of your input files together. Consult
the Calendex users-guide for more information
on setting up and running Calendex analyses.
Why
can’t exposure analysis be made any
easier for the analyst?
Exposure analysis is not rocket
science, but it is still very complex. Moreover,
you rarely will have enough data to be completely
confident with the validity of your results.
It is difficult enough to estimate deterministic,
worst case exposure amounts for either dietary
or residential chemicals. It is far more
difficult to set up meaningful parametric
distributions that can adequately characterize
the range of exposure amounts in the subpopulations
of interest to the risk manager. Other programs
that perform exposure analysis may have
fewer data requirements that Calendex but
they are more limited in their modeling
capabilities.
Why
do DEEM and Calendex report both “per
user” and “per-capita”
exposure calculations? Why do “per-user”
exposures sometimes decrease when multiple
sources of exposure are included in the
analysis?
DEEM and Calendex report the
mean and percentile distribution of exposure
in a given population both on a “per
user” basis and a “per capita”
basis. Both of these measures can provide
useful information to the risk assessor.
However, the per-capita measure is generally
more useful because (1) it estimates exposure
risk in the subpopulation of interest as
a whole, not just for users so that this
risk can be compared against other sources
of risk to the same subpopulation and (2)
because it is a more stable measure of risk
when multiple sources of exposure are being
assessed. To explain this latter assertion,
you must understand the definition of a
“user” in DEEM and Calendex.
In DEEM, a user is a survey
participant who consumes a target food (one
of the foods/foodforms listed in a DEEM
residue file with a non-zero residue) during
the period of analysis (a single day, since
each of the two days of CSFII food intake
are treated separately). (However, in a
Monte Carlo assessment this does not mean
that a “user” always has a non-zero
exposure, because a zero residue amount
can be drawn from a residue distribution
file (RDF) for that intake incident, resulting
in zero exposure for that individual.) When
performing an aggregate exposure assessment
with Calendex over more than one day, a
dietary “user” is any individual
who consumes a target food on any day during
the assessment period (one week up to one
year). When residential pesticide uses are
included in a Calendex analysis the definition
of a user expands to include individuals
who, on any day in the analysis period,
have contact with active residues from any
prior or present pesticide application,
whether as an applicator or as an occupant
of the building and surrounds, even if no
actual uptake occurs.
As multiple uses of pesticides
are evaluated in DEEM or Calendex, whether
in multiple foods or from one or more residential
uses included with dietary exposure, “per
user” exposure means can decrease
while per capita exposure means increase.
This is because adding a second exposure
source with relatively low residues can
increase the total number of “users”,
while decreasing the average per-user exposure
amount over this larger pool of users. The
per-user statistics are easy to calculate
(in fact, both DEEM and Calendex compute
the user statistics first, and then convert
these to per capita statistics) and may
be of some interest to the risk assessor
and for this reason they are reported, but
in most cases when multiple foods and\or
aggregate and cumulative exposures are assessed
the per capita statistics are more stable
and more meaningful to the risk assessor.
Why
do we need to run multiple iterations of
DEEM and Calendex?
DEEM and Calendex calculate
means and percentile distributions of exposure
amounts for user-specified populations within
the United States, e.g., all 1-2 year olds.
The number of individuals actually included
in the DEEM or Calendex analysis for any
given subpopulation is determined by the
number of these individuals in the CSFII.
Obviously the larger the survey size, the
more dependable will be the resulting mean
and percentile distribution. Since the size
of the subpopulation is fixed by the CSFII
database, we can only increase the number
of exposure calculations used to estimate
the mean and percentile distribution by
performing multiple iterations with each
individual in the same database. But performing
multiple iterations with each individual
only improves our estimate of the mean and
percentile distribution when probabilistic
residue data are being used in the analysis.
Otherwise we are just recalculating the
same exposures over and over again for the
same individual, which will not improve
the estimate of the mean or the distribution.
When dietary residue distribution
files (RDFs) are used to provide a range
of residues that might be found on treated
foods of interest to the risk assessor,
then for each iteration for the same individual
on the same day a new residue value is drawn
for each target RAC from the associated
residue distributions and an entirely new
total exposure amount is calculated for
that individual. The more distributions
used and the wider the distributions used,
the more important it is to run multiple
iterations to make sure that there are an
adequate number of high-end exposures computed
to adequately estimate the high-end percentile
distributions (especially above the 99th
percentile) for the population. In addition,
for foods that are eaten by a relatively
small percentage of the population, it is
impossible to adequately estimate the exposure
distribution at the high end without running
multiple iterations for each member of the
subpopulation of interest.
In Calendex, which calculates both dietary
and residential exposures using a wide range
of distributions to model usage, application
rates, available residue amounts, and contact
rates, it is all the more important to run
multiple iterations to adequately model
the underlying exposure distribution for
a given subpopulation.
Each of the 20,607 CSFII participants
has a statistical weighting factor (SWF)
which is used to make inferences about their
contribution to the total US population
and designated subpopulations. The average
SWF is approximately 12,700, so that each
participant represents, on average, about
that many individuals. The sum of all participants
times their own SWF is approximately 262
million people, the total US population
at the time that the 1994-96, 1998 CSFII
was completed. Thus using each individual
more than one time in an exposure assessment
of the US population or subset of that population,
with different draws from the appropriate
residue distributions or other probabilistic
parameters, improves the estimate of the
mean and distribution for the entire population
of interest as long as each participant
is weighted properly.
It is important to recognize
that the term Monte Carlo “iteration”
is used differently in DEEM and Calendex
than in some other programs which use Monte
Carlo analysis. For example, in Crystal
Ball ™ each individual would count
as one iteration, whereas in DEEM and Calendex
one run through the entire subpopulation
of interest is considered to be one iteration.
Multiple iterations in DEEM and Calendex
refer to computing multiple exposures for
each individual in the subpopulation, each
time using a different series of random
numbers to select the residues, application
schedules and contact amounts. All of the
exposures are used together to determine
the mean and percentile distribution for
the subpopulation.
How
many iterations are needed in a DEEM or
Calendex analysis?
In general, you need to run
enough exposure iterations for each individual
in the CSFII to reach a stable exposure
assessment for the actual subpopulation
of interest, especially at the high end
of the percentile distribution (e.g., 99.9%).
Stability is shown by running the same exposure
analysis with a different random number
“seed” and getting the same
practical result (i.e. within a few percent
of the same answer, depending on the requirements
of the assessment). There is no rule of
thumb for determining the actual number
of iterations needed to achieve stability;
stability depends on how many individuals
are in the subpopulation that you are evaluating,
the number and frequency of consumption
of target foods in the dietary residue file,
and the nature of the residue distributions
being processed. In complex DEEM analyses
with many residue distribution files (RDF’s)
and in Calendex analyses with multiple application
types (AGX files) 100 or more iterations
are usually needed to reach a stable result.
If there are only a few high residue numbers
and lots of low numbers in an RDF, then
the probability of matching a high residue
with a high consumption (or contact) amount
is low and it takes more iterations to make
sure that enough of these highly exposed
individuals end up being used to properly
shape the distribution at the high end.
However, achieving stability
is generally only important when finalizing
your assessment. When learning to use DEEM
and Calendex and when first experimenting
with new AGX files in Calendex it is smart
to use just a few iterations to make sure
that the results look reasonable.
Note that if you are using
an exposure program that does not allow
multiple iterations and a user-determined
random number seed you cannot tell if and
when you have reached a stable exposure
assessment.
What
is the “Calendar Model” for
computing human exposures to pesticides
and other chemicals?
Before the concept of aggregate
and cumulative analysis was introduced in
the mid 1990’s, most dietary exposure
analyses were based on intake data for two
or three days per individual, since the
best sources of these data were the two-
and three-day CSFII surveys conducted by
the USDA. Single-day dietary exposures for
acute analysis were computed for each individual
on each survey day separately and the results
(for as many as 40,000 person-days) were
used to generate a mean and a percentile
distribution for different populations of
interest (e.g., infants, children ages 1-2
years, etc.). Chronic (long term) exposures
could not be computed for individuals with
the available 2- or 3-day dietary intake
data. Instead chronic exposure values for
a given population were typically based
on the per capita mean intake amount for
each food consumed by the population of
interest, multiplied by an average residue
amount for each of those foods. Residential
exposure analyses for the same chemicals
were conducted using SOP methods [Link or
footnote] for the day of application, and
were typically based on worst-case scenarios;
for example, if two different pesticides
were used in the home, they would be assumed
to be applied and contacted on the same
day.
With the passage of the FQPA
in 1996, which required that cumulative
and aggregate exposure analysis would be
required for human exposure to chemicals
with a common action mechanism in foods
and in the residential environment, it was
clear that the single-day analysis models
used in the past would not provide a satisfactory
estimate of aggregate and cumulative human
exposures to chemicals in foods and the
residential environment. Instead a calendar
model would be required, one in which the
intake or uptake of chemicals in the diet
and in the residential environment would
be modeled with a time dimension included.
Only by adding a time dimension can you
make reasonable assumptions about the coincidence
of exposures to chemicals from different
sources at different times of the year.
In the mid 1990’s scientists
at Novigen Sciences, Inc (now the Foods
and Chemicals practice of Exponent, Inc)
developed a calendar-like matrix method
for evaluating exposure to two residential
pesticides uses in the same home that were
applied at specified time intervals but
not necessarily at the same time or at the
same interval. Using this matrix approach
they were able to assign probabilities to
the application dates of the two chemicals
and then make inferences about the distribution
of post-application exposure amounts that
would result for occupants over the period
of analysis. The name “calendar analysis”
was given to this new analytic technique.
Steve Petersen of Durango
Software greatly expanded this prototype
analysis technique to include both dietary
and residential exposure analysis, to use
it with multiple residential pesticide uses
applied at both regular and irregular intervals
during a calendar year, and to use it to
estimate both applicator and post-application
exposures. This expanded calendar analysis
concept was the eponymous basis for the
name “Calendex”.
The calendar model for exposure
analysis is simple in concept but difficult
to use to its full potential because of
the detailed amount of information that
is needed to execute it; not only does each
pesticide use need an application schedule
(including both number of times per year
and dates of application), initial residue
amount, and degradation rate, but the application
schedule parameters and initial residue
amount must be expressed probabilistically.
Moreover, a “contact schedule”
is needed to model individual behavior that
results in uptake of the available chemical
residues. This contact schedule will likely
have distributions of contact activities
as well, which can include both the duration
of the contact and the intensity of the
contact (for example, the number of times
that a child pets the family dog or rolls
on the lawn on a given day, as well as the
coincident clothing levels, when pet and
lawn treatments are being assessed). Different
contact schedules are needed for different
age groups and at different times of the
year and may even be appropriate for different
days of the week (at least differentiating
between week days and week ends).
To understand the concept
at its simplest, visualize a 365-day time
line from January 1 to December 31 on a
horizontal axis. (See figure 1.) Start by
including four applications per year of
pesticide use “A” occurring
at three month (91 day) intervals, say Feb
14 (day 45), May 16 (day 136), August 15
(day 227), and November 14 (day 318) in
a particular household. Assume that after
the chemical is applied it is active for
28 days, and that potency declines on a
straight line basis to 0 on the 28th day.
Moreover, assume that an individual’s
contact level with the treated surface (or
breathing rate if inhalation exposure is
being assessed) is the same each day, so
that the exposure amount is directly proportional
to the available chemical on the treated
surface. (In fact, the contact level may
vary from day to day and the amount of pesticide
applied may vary from application to application
in a stochastic manner as well; here we
use a simplifying assumption.) Figure 1
shows the declining pesticide availability
amounts for the four 28-day periods following
the day of application. Under these assumptions
the daily exposure amounts would be directly
proportional to the pesticide availability
on each day that there are active residues.
Now assume that pesticide use “B”
is used in the same household and occurs
twice a year during the summer months only
(say July 1 (day 182) and August 1 (day
213), and that its persistence is only 14
days. Again we use straight line degradation,
a constant application rate, and a constant
contact rate to make this an easier problem
to visualize. The availability of pesticide
use B during the calendar year at this location
is also shown in Figure 1. Based on the
application schedules described here, there
is no coincidence of exposures possible
and no aggregation of exposures is needed.
However, if stochastic methods
are used to determine when the applications
are made, then in some households that use
both treatment types there will be some
overlap between the applications. For example,
in figure 2 the quarterly applications of
use “A” are now shifted to March
14 (day 73), June 13 (day 164), Sept 12
(day 255), and Dec 12 (day 346), while the
applications of “B” are the
same as in Figure 1. Now the pesticide availability
from both uses are coincident during the
11 days from July 1 (day 182) and July 11
(day 192); these daily availability amounts
must be added together to determine the
total (aggregate) exposure amount during
each of these 11 days.
Figure 3, which is similar
to Figure 2 but includes daily dietary residue
estimates for the same individual to the
same chemical, shows how the dietary exposure
can be aggregated with the residential exposures
when they co-occur during the year in order
to determine total acute (one-day) exposure
amounts. Note that the dietary residues
are assumed to have 100% contact so that
the ingested residue amounts are the same
as exposure amounts. (The available residue
amounts from treatments A and B must be
converted to daily exposure amounts by using
daily contact factors before the resulting
exposure amounts can be combined with the
dietary exposures.)
It is easy to see how additional
residential pesticide uses can be included
with this model. However, it is also easy
to see that modeling the many different
exposure outcomes from one or more applications
of each pesticide use, as well as the dietary
exposures, can quickly overwhelm the risk
assessor if the modeling is not performed
with a robust computer program. Calendex
takes over the chores of determining which
pesticide uses are applied in a given household,
how many and when the applications are made
in a given year, what the application rate
is, how the chemical degrades from day to
day after application, how much contact
is made by the occupant, and what the resulting
daily exposure is for each day of interest
during the year. All of these calculations
are probabilistic, based on distributions
for each of these parameters provided by
the analyst. This process is repeated for
each pesticide use of interest to the risk
assessor, and combined on a daily basis
with the estimate of dietary exposure for
that same day. Calendex aggregates all individual
exposures on the same day and reports both
the total exposure and the components of
that exposure for later use in determining
mean exposure amounts, distributions of
exposures, and the contribution of the various
treatment types to the total exposure. Calendex
can report these exposure measures for any
given day of the year, or average the daily
exposures over any given time period during
the year. Calendex can also report the maximum
individual exposures for each individual
over the year (or other time period of interest
less than one year), regardless of when
these occurred, and then provide a “temporal
mapping” of those maximum exposures
so that the analyst can see which times
of the year are the most problematic with
regard to peak exposures.
Why
is Durango Software LLC, located in Bethesda,
Maryland, named after a small town in Colorado?
Because Durango Sotware founder
Steve Petersen
and his wife and family spent many summer
vacations in the family cabin just outside
of Durango and he still has dreams of working
from there someday.
