Slide
1
of
168
A
Model
Comparison:

Dietary
(
Food
and
Water)
Exposure
in
DEEM/
Calendex,
CARES,
and
LifeLine
F.
Bart
Suhre,
David
Miller,
Steve
Nako
U.
S.
EPA
Health
Effects
Division
Office
of
Pesticide
Programs
A
Model
Comparison:

Dietary
(
Food
and
Water)
Exposure
in
DEEM/
Calendex,
CARES,
and
LifeLine
F.
Bart
Suhre,
David
Miller,
Steve
Nako
U.
S.
EPA
Health
Effects
Division
Office
of
Pesticide
Programs
FIFRA
SAP
Meeting
April
29­
30,
2004
FIFRA
SAP
Meeting
April
29­
30,
2004
Slide
2
of
168
Model
Names
Model
Names

CARES

Cumulative
Aggregate
Risk
Evaluation
System

DEEM

Dietary
Exposure
Evaluation
Model

SHEDS

Stochastic
Human
Exposure
and
Dose
Simulator

LifeLine
and
Calendex

Not
acronyms
Slide
3
of
168
SHEDS
SHEDS

SHEDS­
Wood
is
a
probabilistic
model
which
aggregates
oral
(
incidental
hand­
to­
mouth),
dermal
and
inhalation
exposure
routes

SHEDS­
Wood
was
developed
by
EPA/
ORD
and
used
by
OPP
to
assess
Chromium
and
Arsenic
exposure
to
children
from
CCA
treated
wooden
play
sets
and
decks

FIFRA
SAPs:
Aug
2002,
and
Dec.
2003
Slide
4
of
168
Why
wasn't
SHEDS
considered
in
this
Model
Comparison
?

Why
wasn't
SHEDS
considered
in
this
Model
Comparison
?


Development
of
SHEDS­
Pesticide,
an
aggregate
model
for
oral
(
dietary
and
incidental
hand­
to­
mouth),
dermal
and
inhalation
exposure
routes
was
delayed
as
a
result
of
ORD
resources
being
channeled
to
the
CCA
risk
assessment

SHEDS­
Pesticides
was
recently
received
by
OPP
and
we
plan
to
include
it
in
future
model
comparisons
Slide
5
of
168
Acknowledgements
Acknowledgements

David
Miller,
EPA/
OPP

Steve
Nako,
EPA/
OPP

Sheila
Piper,
EPA/
OPP

Alan
Dixon,
EPA/
OPP

William
Smith,
EPA/
OPP

OPP/
HED,
DeSAC
members

Dietary
Exposure
Science
Advisory
Council

Muhilan
Pandian,
Infoscientific.
com
Slide
6
of
168
Materials
Provided
to
SAP
Materials
Provided
to
SAP

Document
entitled:


A
Model
Comparison:
Dietary
(
Food
and
Water)
Exposure
in
DEEM/
Calendex,
CARES,
and
LifeLine.
March
31,
2004.
Slide
7
of
168
Materials
Provided
to
SAP
Materials
Provided
to
SAP

Reference
material
from
past
SAP
Sessions
on
DEEM,
Calendex,

CARES,
and
LifeLine

Users
Manuals
and
Technical
manuals
for
DEEM,
Calendex,

CARES,
and
LifeLine
Slide
8
of
168
Materials
Provided
to
SAP
Materials
Provided
to
SAP

SAS
Programs
to
approximate
predictive
models
(
DEEM/
Calendex,
CARES,
and
LifeLine)
estimates

SAS
approximations
of
predictive
models
(
DEEM/
Calendex,

CARES,
and
LifeLine)
estimates
Slide
9
of
168
Materials
Provided
to
SAP
Materials
Provided
to
SAP

Questions
on:


The
impact
model
design
has
on
estimating
dietary
exposure

Methodology
for
testing
model
design
features
on
exposure
estimates
Slide
10
of
168
Presentation
Outline
Presentation
Outline

Background

F.
Bart
Suhre

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

F.
Bart
Suhre

Future
model
comparisons
in
OPP

F.
Bart
Suhre

Probabilistic
models
and
past
SAP
Reviews

David
Miller
Slide
11
of
168
Presentation
Outline
Presentation
Outline

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

David
Miller

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Steve
Nako

Case
Study
 
DEEM,
CARES,
and
LifeLine
Dietary
(
food
and
water)

exposure
estimates
using
a
common
dataset

David
Miller
Slide
12
of
168
Presentation
Outline
Presentation
Outline

Summary
(
very
brief)


F.
Bart
Suhre
Slide
13
of
168
Background
`
Legislative
Authority'

Background
`
Legislative
Authority'


FIFRA
and
FFDCA
authorize
EPA
to
regulate
pesticides

In
1996,
FQPA
amended
FIFRA
and
FFDCA
and
directed
EPA
to
consider
"
aggregate
exposure"
in
its
decision­
making
Slide
14
of
168
Background
`
Guidance
'
Background
`
Guidance
'


EPA
developed
Guidance
Documents:


Interim
Guidance
for
Aggregate
Assessments
(
1997)


Probabilistic
Guidance
(
1998)


Aggregate
Guidance
(
1999)


Cumulative
Guidance
(
2000)
Slide
15
of
168
Background
`
Models'
Background
`
Models'


EPA
encouraged
development
of
probabilistic
models
that
permit
time­
based
integration
of
residential
and
dietary
(
food
and
water)
exposures
to
pesticides

Models
were
developed
and
presented
to
FIFRA
SAP
by
EPA
and
various
model
development
teams
Slide
16
of
168
Background
`
Model
Implementation'

Background
`
Model
Implementation'


Three
Basic
Criteria:

1)
conform
to
EPA
guidance
2)
peer
review
3)
available
to
all
Slide
17
of
168
Goals
of
EPA
Current
Model
Comparisons
Goals
of
EPA
Current
Model
Comparisons

Better
understand
design
features
of
probabilistic
models
currently
available
for
estimating
pesticide
exposure

Develop
methodology
to
test
the
impact
model
design
features
have
on
dietary
exposure
estimates
Slide
18
of
168
Objectives
of
EPA
Current
Model
Comparisons
Objectives
of
EPA
Current
Model
Comparisons

Compare
dietary
(
food
and
water)

exposure
estimates
derived
from
DEEM/
Calendex,
CARES
and
LifeLine
using
common
data
sets

Investigate
specific
reasons
why
these
exposure
estimates
might
differ
Slide
19
of
168
Focus
of
Today's
SAP
Focus
of
Today's
SAP

Describe
design
features
of
DEEM/
Calendex,
CARES,
and
LifeLine
with
emphasis
on
the
dietary
exposure
pathway

Present
dietary
results
from
these
3
models

Demonstrate
why
dietary
exposure
estimates
may
differ
for
each
of
the
three
models
Slide
20
of
168
Future
Model
Comparisons
in
OPP
Future
Model
Comparisons
in
OPP

Using
the
dataset
discussed
today:


We
plan
to
investigate
model
design
features
for
SHEDS­
Pesticide
and
compare
its
1­
day
dietary
exposure
estimates
to
the
other
models
already
tested

We
plan
to
investigate
the
impact
model
design
has
on
estimating
exposure
for
periods
>
1
day
(
e.
g.
7,
14,
and
30
days)

for
all
available
pesticide
models
Slide
21
of
168
Future
Model
Comparisons
in
OPP
Future
Model
Comparisons
in
OPP

We
plan
to
more
fully
investigate
the
drinking
water
model
design
features
and
compare
exposure
estimates
for
each
model
using
a
common
dataset

We
plan
to
investigate
modeldesign
features
for
key
residential
pathways
and
compare
residential
exposure
estimates
for
each
model
using
a
common
dataset
Slide
22
of
168
Future
Model
Comparisons
in
OPP
Future
Model
Comparisons
in
OPP

We
plan
to
investigate
and
compare
aggregate
exposure
estimates
over
all
pathways
(
food,

water,
and
residential)
for
all
available
pesticide
models
using
a
common
dataset
Slide
23
of
168
Future
Model
Comparisons
in
OPP
Future
Model
Comparisons
in
OPP

We
will
be
performing
model
comparisons
on
a
routine
basis
as
a
result
of
stakeholders
using
different
models
to
support
pesticide
regulatory
activity
Slide
24
of
168
Future
Model
Comparisons
in
OPP
Future
Model
Comparisons
in
OPP

We
will
continue
to
work
closely
with
model
development
teams
to
improve
their
software

Future
development
priorities:


Standardized
formats
for
data
entry

More
sophisticated
output
reports
to
assess
model
sensitivity
and
exposure
contributors

Link
exposure
outputs
to
Physically
Based
Pharmacokenetic
(
PBPK)

models
Slide
25
of
168
Probabilistic
models
and
past
SAP
Reviews
Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline
Probabilistic
models
and
past
SAP
Reviews
Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline
David
J.
Miller
David
J.
Miller

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
26
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
27
of
168
Slide
28
of
168
Software
Models
Software
Models

A
number
of
software
models
are
now
available
which
allow
probabilistic
aggregation
of
exposures
through
multiple
pathways
and
routes

Lifeline

CARES

DEEM/
Calendex

SHEDS
Slide
29
of
168
Past
SAP
Reviews
Past
SAP
Reviews

DEEM/
Calendex

September,
2000:
DEEM

February
2000:
Calendex

CARES

April
2002

Lifeline

September
1999:
Design
Concepts
and
Theory

September
2000:
A
Case
Study.


March
2001:
Review
of
Lifeline
v.
1.0

SHEDS­
Wood

August,
2002

December,
2003:
CCA
Wood
Preservatives
Slide
30
of
168
Introduction
and
Purpose
Introduction
and
Purpose

This
presentation
will
provide
a
brief
introduction
and
review
of
three
models
being
considered
in
this
SAP

Purpose
is
to
review
general
background
information
on
each
of
the
models
and
how
they
operate
in
order
to
provide
overall
context
for
subsequent
presentations
involving
more
detailed
model
investigations
Slide
31
of
168
Presentation
Outline
Presentation
Outline
For
each
model:


Brief
Developmental
History

Overview/
Framework
of
Model

Base
Reference
Population

"
Binning"


Model
(
Sampling)
Weights

Body
Weights
Slide
32
of
168
Version
2.0
Version
2.0
Slide
33
of
168
Brief
Developmental
History
Brief
Developmental
History

Purpose:
to
create
publicly
accessible
risk
assessment
software
for
a
wide
range
of
anticipated
users

Developed
by
Hampshire
Research
during
1998
 
2000
under
Cooperative
Agreement
with
EPA
and
USDA

Transferred
to
The
Lifeline
Group
in
December,
2000

Made
publicly
available
on
Dec.
15,
2000

Currently
in
Version
2.0

Released
October,
2002

Website:


www.
thelifelinegroup.
org
Slide
34
of
168
The
LifeLine
 
Model
The
LifeLine
 
Model

Probabilistic
(
Monte­
Carlo)
model
of
aggregate
exposure
to
pesticides
that
occurs
to
each
member
of
a
simulated
population
of
individuals

Models
each
potentially
exposed
individual
within
a
population
as
an
individual

Attempts
to
define
each
simulation
of
an
individual
using
"
transition
rules"
Slide
35
of
168
The
LifeLine
 
Model
The
LifeLine
 
Model

Uses
data
from
well­
known
surveys
to
generate
and
evaluate
specific
daily
exposures
for
individuals

Uses
available
databases
to
address
each
component
of
simulation

Combined
datasets
allow
model
to
define
aggregate
exposure
for
each
day
of
an
individual's
life

Birth

Growth
and
aging

Movement
from
home
to
home
and
region
to
region

Extent
of
pesticide
use

Daily
activity
patterns
and
daily
dietary
patterns
Slide
36
of
168
Transition
Rules
Transition
Rules

Used
to
model
changes
in
values
of
exposure
factors
over
time

Ensure
that
selections
from
multiple
databases
are
consistent
and
scientifically
defensible

Ensure
consistency
with
prior
values
of
a
factor

Ensure
consistency
with
values
of
other
factors

Determine
when
a
value
should
change
and
when
should
remain
the
same

Accounting
for
Correlations

Record­
based
modeling

Contingent
modeling
Slide
37
of
168
Transition
Rules
Transition
Rules

Characteristics
of
individuals
generated
by
Lifeline
are:


Fixed
Characteristics

Include
sex,
race/
ethnicity,
birth
date,
body
type,

maternal
SES

Characteristics
that
very
slowly
over
time

Long­
term
trends/
progressions
(
weight,
height,

residence
location)


Episodic/
non­
periodic
State
changes
(
room
sizes,

pest
pressures,
etc)


Cyclic
(
periodic)
state
changes
(
weekday
vs.

weekend,
dietary
residues,
residential
pesticide
use
etc.


Characteristics
that
vary
day­
to­
day

Ephemeral
(
activity
patterns,
consumption,
residues)
Conception
and
Birth
Childhood
Retirement
Career
Home
ownership
Child
raising
Modeling
A
Person's
Life
Source:
HRI
SAP
Presentation
Slide
38
of
168
Conception
and
Birth
Childhood
Retirement
Career
Home
ownership
Child
raising
Product
A
Product
B
Product
C
Source:
Modified
from
HRI
SAP
Presentation
Modeling
A
Person's
Lifetime
Exposures
To
One
Active
Ingredient
Slide
39
of
168
Source:
HRI
SAP
Presentation
Slide
40
of
168
Repeating
the
Process
Yields
a
Model
of
the
Exposed
Population
Slide
41
of
168
Basic
Steps
in
LL
Analysis
Basic
Steps
in
LL
Analysis

STEP
1:
Creating
and
Modeling
Representative
Individuals

Assigning
fixed
and
initial
characteristics

Modeling
individual
growth

STEP
2:
Assigning
Physiological
and
Residential
Characteristics

STEP
3:
Assigning
Exposure­
Related
Behaviors

Residential

Food/
Water
Slide
42
of
168
STEP
1:
Creating
and
Modeling
Representative
Individuals
STEP
1:
Creating
and
Modeling
Representative
Individuals

Assigning
Fixed
and
Initial
Characteristics

Begin
with
a
national
survey
of
fixed
and
initial
characteristics

Natality
Database

Serves
as
Lifeline's
"
Reference
Population"


Determines
Lifeline's
Model
(
Sampling)
Weight

Two
option
for
Mortality

Based
on
standard
life
tables
(
NCHS
Vital
Statistics
of
the
U.
S.)


Mortality
fixed
after
at
85
years
Slide
43
of
168
STEP
1:
Creating
and
Modeling
Representative
Individuals
STEP
1:
Creating
and
Modeling
Representative
Individuals

Modeling
Individual
Growth

LifeLine
uses
a
anthropometric
model
of
the
patterns
of
change
in
physiology
based
on
NHANES
III
data

Each
year
the
physiology
of
the
individual
is
updated
Slide
44
of
168
STEP
2:
Assigning
Physiological
and
Residential
Characteristics
STEP
2:
Assigning
Physiological
and
Residential
Characteristics

Once
the
general
description
of
a
residence
is
determined:


AHS
database
of
specific
homes
is
used
to
define
the
residential
characteristics

American
Gardening
Survey
(
AGS)
and
National
Home
and
Garden
Pesticide
Use
Survey
(
NHGPUS)
used
to
evaluate
other
characteristics

US
census
tracks
mobility
in
a
series
of
multiple
and
partially­
overlapping
databases

Each
year
the
mobility
of
an
individual
is
modeled
in
binary
decision
Slide
45
of
168
STEP
3:
Assigning
Exposure
Related
Behaviors
STEP
3:
Assigning
Exposure
Related
Behaviors

Once
the
residence,
ethnicity,
age,
gender
and
race
are
defined
for
an
individual
at
any
given
point
in
time,
these
data
can
be
used
to
guide
the
selection
of
data
on

Activities/
locations
(
NHAPS)


Pesticide
use
(
NHGPUS)


Daily
diets
(
CSFII)


These
predictive
factors
are
usually
determined
by
selecting
records
from
individuals
that
match
the
modeled
individual

"
Binning"
Lifeline
Binning
Lifeline
Binning
°
Identify
key
parameters
related
to
exposure
and
determine
natural
groupins
based
observed
patterns
Slide
46
of
168
Slide
47
of
168
Lifeline
Dietary
Assessment
and
CSFII
Lifeline
Dietary
Assessment
and
CSFII

Lifeline
uses
USDA's
CSFII
(
Continuing
Survey
of
Food
Intake
by
Individuals)
as
a
base
data
set
for
the
dietary
component
of
its
assessment

Nationally
Representative/
Statistically­

Based

CSFII
reports
foods
"
as
eaten"
on
an
individual
basis.


200
g
apple
pie
­­
not
apples,
wheat,

soybean
oil,
etc.
Slide
48
of
168
CSFII
CSFII

Lifeline
uses
residue
data
on
individual
components
of
food
to
build
residue
distributions
on
foods
"
as
eaten"
using
recipe
file
(
aka
100
g
file)


Distribution
of
residues
on
apple
pie
"
built"
from
residue
distribution
in
apples,
wheat,
sugar,
etc.
Frequency
Chart
.000
.031
.062
.092
.123
0
30.75
61.5
92.25
123
0.00
0.33
0.65
0.97
1.30
1,000
Trials
1,000
Displayed
Forecast:
Apple
Residues
Frequency
Chart
.000
.021
.042
.063
.084
0
21
42
63
84
0.00
0.01
0.03
0.04
0.05
1,000
Trials
978
Displayed
Forecast:
wheat
residues
Frequency
Chart
.000
.007
.015
.022
.029
0
7.25
14.5
21.75
29
0.00
0.00
0.01
0.01
0.01
1,000
Trials
983
Displayed
Forecast:
sugar
residues
Forecast:
Residues
in
Apple
Pie
0
10
20
30
40
50
0.00
0.03
0.05
0.08
0.10
Frequency
Estimating
Residue
Distribution
in
Apple
Pie
Slide
49
of
168
Slide
50
of
168
Lifeline
Dietary
Consumption
Lifeline
Dietary
Consumption

Dietary
Bins
created
to
reflect
groupings
based
on
explicitly­
identified
eating
patterns/
similarities

age
x
season

Each
individual's
daily
food
consumption
in
LL
is
drawn
from
a
similar
individual
(
age
x
season)
from
the
CSFII

Daily
eating
patterns
"
built­
up"
for
each
day
in
an
individual's
lifetime

Uses
matched
individual's
intake
(
grams
food)
direction

No
"
normalization"
for
bodyweight
Slide
51
of
168
Exposure
through
Tap
Water
Exposure
through
Tap
Water

Water
Consumption
data
obtained
from
CSFII

Direct
and
indirect
water

Tap,
bottled,
Other,
Missing

Tapwater
concentrations
entered
by
user

Matched
to
each
residence
through
AHS
by

Census
region

Type
of
water
supply
(
public
or
private
system,

well,
or
other)


Setting
(
urban
vs.
rural)


Season
Slide
52
of
168
Exposure
through
Tap
Water
Exposure
through
Tap
Water

Census
Region
(
4)


Northeast,
South,
West,
Midwest

System
Type
(
3)


Public
or
Private
Water
system,
Well,

Other

Season
(
4)


Spring,
Summer,
Fall,
Winter

Setting
(
2)


Rural,
Urban
°
User
can
enter
up
to
96
different
water
concentration
distributions
°
Each
distribution
entered
as
cumulative
distribution
in
Lifeline
Slide
53
of
168
Exposure
through
Tap
Water
Exposure
through
Tap
Water

Each
house
is
then
randomly
assigned
a
percentile

Corresponding
water
concentration
value
for
that
region,
water
supply
type,
season
and
setting
is
selected

Upper
percentile
households
remain
in
upper
percentile
and
vice
versa

Same
4
season­
specific
residue
values
repeated
each
year
that
the
individual
resides
in
the
home
Slide
54
of
168
Characterizing
Exposure
Characterizing
Exposure

On
each
day
of
an
individual's
life
the
model
has
defined:


Physiology
(
weight,
skin
surface
area,
inhalation
rates)


Residence
(
region,
setting,
room
sizes,
air
exchange
rates,
type
of
floor
surface,
and
presence
of
a
lawn,
pets,
or
garden)


Daily
activities,
room
locations,
and
dietary
consumption

Pesticide
residue
levels
in
foods,
tapwater,
on
surfaces,
and
in
air

Using
this
data
the
model
calculates
route
specific
exposures
and
doses

These
routes
specific
doses
are
use
to
estimate
the
the
total
aggregate
(
systemic
dose)
Slide
55
of
168
Lifeline
Outputs
Lifeline
Outputs

Lifeline
determines
exposure
history
for
each
individual
on
each
day
of
a
person's
exposure
simulation
period

Lifeline
saves
only
a
portion
of
the
data
collected
for
each
individual.
For
each
season,
Lifeline
saves

Seasonal
average

Maximum
day

Random
Day

These
are
saved
for
each
routespecific
and
source
specific
exposure
Seasonal
Data
and
Report
Data
0
0.2
0.4
0.6
0.8
1
1.2
Days
in
Spring
Exposure
Value
1­
Day
Mean1
Max
1
Slide
56
of
168
Cumulative
and
Aggregate
Risk
Evaluation
System
Cumulative
and
Aggregate
Risk
Evaluation
System
Version
2.0
SRS
040401
CARES
Slide
57
of
168
Slide
58
of
168
Brief
Developmental
History
Brief
Developmental
History

Purpose:
CARES
is
a
software
tool
for
performing
Aggregate
and
Cumulative
Exposure/
Risk
Assessments

Developed
under
contract
with
member
companies
of
Crop
Life
America

First
made
publicly
available
on
April,
2002

Transferred
to
ILSI's
Risk
Science
Institute
in
January,
2004

Currently
in
Version
2.0

Released
January,
2004

QA
testing

Website:


http://
www.
infoscientific.
com/
software_
main.
htm#
cares
CARES
Reference
Population
CARES
Reference
Population

CARES
uses
a
Population
Generator
to
develop
a
Reference
Population
(
RP)


Selected
as
probability
sample
of
100,000
individuals
of
Census/
5%
PUMS
Stratified
so
as
to
support
analysis
of
upper
percentiles
of
exposure
for
subgroups
of
interest

Race/
Ethnicity
and
Age/
Gender
RP
appropriately
re­
weighted
to
be
similar
to
U.
S.
Census
Reference
Population:

5%
PUMS
Population:

Total
1990
Census:
241,000,000
12,000,000
100,000
(
0.04%
of
Census
)
Slide
59
of
168
Slide
60
of
168
CARES
Reference
Population
CARES
Reference
Population

U.
S.
Census/
PUMS
data
contains
many
descriptive
variables
of
interest
for
each
person
in
RP.
However,
PUMS
data
is
insufficient
to
completely
model
an
individual's
exposure 


No
data
on

Dietary
consumption

Human
activity
patterns

Pesticide
Use
patterns

Etc.


Necessary
to
match
individuals
in
Census
with
individuals
in
other
databases 

 
but
it
is
not
possible
to
make
an
"
exact"

match
Slide
61
of
168
CARES
Reference
Population
CARES
Reference
Population

Matching
between
individuals
in
RP
and
those
in
other
ancillary
databases
(
e.
g.,
CSFII,
NHAPS,

etc.)
done
based
on
distance
or
dissimilarity
measures
on
critical
parameters
(
e.
g.,
region,
age,

gender,
etc.)
Slide
62
of
168
CARES
Reference
Population
CARES
Reference
Population

General
Matching
strategy:


Determine
how
similar
each
sampled
Census
individual
person
is
to
each
person
in
the
the
auxilliary
database
of
interest,
and
match
based
on
this
similarity
measure

Similarity
measure
calculated
using
relevant
characteristics
that
each
database
has
in
common
(
Gower
Dissimilarity
Index)


Certain
parameters
receive
higher
priority
in
matching
than
others,
depending
on
characteristic
of
interest.
For
example,
for
food
consumption

Always
exactly
match
on
gender

Match
very
closely
on
age
for
infants,
children

Match
less
closely
on
age
for
youth

No
match
for
adults
(>
19
y.
o.)
Slide
63
of
168
Dietary
Matching
Dietary
Matching

For
initial
RP­
CSFII
match,
match
is
determined
based
on
characteristics
important
in
determining
food
consumption

Age/
gender

Race/
ethnicity

Region

CSFII
data
used
to
determine
body
weight,

nursing/
pregnancy
status,
and
one/
two
days
food
consumption

Combine
data
from
PUMS
and
CSFII
to
create
an
individual
that
is
both
realistic
and
more
"
complete"
Slide
64
of
168
Dietary
Profile
Dietary
Profile

CARES
next
creates
a
365
day
consumption
profile

Begins
with
1
or
2
days
present
in
CSFII

Uses
statistical
matching
(
Gower
Dissimilarity
Index)
to
match
similar
individuals
and
develop
consumption
diaries
as
"
fill­
in"
for
remaining
days
of
year

Priority
matching
on:
gender,
pregnancy
status,
age
in
months
for
infants,
age
in
years
for
youth,
etc.


To
preserve
seasonal
nature
of
consumption
data,
attempt
to
match
+/­
7
days
Slide
65
of
168
Slide
66
of
168
Residential
Exposures
Residential
Exposures

Residential
Factors
are
assigned
to
each
individual
based
on
U.
S.
Census,
AHS,
etc.


E.
g,
homeowner
status,
urban
vs.
rural,
housing
type,
presence
of
lawn
or
garden,
pets,
#
rooms,

AERs,
etc.


A
product
use
profile
is
constructed
to
assign
product
use
through
event
allocator

Considers
product
use
surveys,
label
information,

PPIS,
NHGPUS,
ORETF,
REJV
databases

Exposures
for
each
time
step
(
e.
g.
calendar
day)
and
activity/
pathway
are
calculated

Initial
media
concentrations,
degradation
rates,

transfer
coefficients,
activity
patterns/
duration,

absorption,
etc.
Slide
67
of
168
CARES
Outputs
CARES
Outputs

Broad
Range/
Selection
of
Available
Output
Options

Total
daily
exposures,
by
person­
day

Rank
order
from
high
to
low

Total
daily
exposures,
by
person­
day
and
food
Slide
68
of
168
Slide
69
of
168
CARES
Output
CARES
Output

Contribution
Analysis

Allows
the
assessor
to
evaluate
those
commodities
which
are
significant
contributors
to
the
highend
exposures

Useful
for
risk
management
purposes
for
determining
which
commodities
which,
when
removed
from
the
assessment,
would
result
in
the
largest
decreases
in
high­
end
exposures
CARES:
CEC
Report
CARES:
CEC
Report
Slide
70
of
168
Slide
71
of
168
Dietary
Exposure
Evaluation
Model
Dietary
Exposure
Evaluation
Model
Slide
72
of
168
Brief
Developmental
History
Brief
Developmental
History

Purpose:
DEEM
and
DEEM/
Calendex
are
software
tools
for
performing
Aggregate
and
Cumulative
Exposure/
Risk
Assessments

Developed
by
Durango
Software
and
originally
licensed
under
Novigen
Sciences,

Inc.
Licensing
now
being
done
by
Exponent,

Inc.


Currently
in
Version
8

Released
March,
2000

Website:


http://
www.
exponent.
com/
practices/
foodchemical/
Slide
73
of
168
DEEM
DEEM

DEEM's
base
population
is
derived
from
USDA's
CSFII

Limited
to
individuals
with
two
(
both)
days
of
consumption

DEEM
uses
MC
techniques
to
combine
crop
residue
concentrations
with
each
individual's
reported
daily
food
consumption

Result
is
a
distribution
of
1­
day
acute
dietary
exposures
across
a
population
Slide
74
of
168
DEEM
DEEM

Calendex
is
an
available
"
add­
in"

to
DEEM

Permits
aggregation
and
cumulation
of
water
and
residential
exposures
which
incorporate
temporal
sequence
of
events
Slide
75
of
168
DEEM
Outputs
DEEM
Outputs

Provides
Exposure
and
associated
risk
estimates
for
each
subpopulation
of
interest
at
each
percentile
of
interest

Provides
Commodity
Exposure
Contribution
Analysis
which
permits
evaluation
of
those
commodities
which
are
significant
contributors
to
the
highend
exposures
DEEM
Output
DEEM
Output
Slide
76
of
168
Slide
77
of
168
Side
by
side
comparisons
of
exposure
estimates
Side
by
side
comparisons
of
exposure
estimates
David
J.
Miller
David
J.
Miller

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
78
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
79
of
168
Slide
80
of
168
DEEM
vs.
Lifeline
Comparisons
DEEM
vs.
Lifeline
Comparisons

OPP
has
conducted
comparisons
among
models
to
serve
as
QA
checks

DEEM
vs.
Lifeline
Comparison

As
part
of
standard
procedures,
OPP
has
run
as
part
each
review
both
DEEM
and
Lifeline
in
parallel
to
compare
results

Represent
actual
assessments
performed
in
OPP

Standard
1000
ppm
single
commodity
file

1000
ppm
input
residue
distribution
for
single
commodity
will
produce
distribution
equivalent
to
consumption

Resulting
predicted
"
exposures"
correspond
exactly
to
consumption
values
Slide
81
of
168
DEEM
vs.
Lifeline
Comparisons
DEEM
vs.
Lifeline
Comparisons

DEEM
vs.
Lifeline
Comparison

Generally,
OPP
has
found
close
agreement
between
DEEM
and
Lifeline
results
at
percentiles
of
interest

Typically,
differences
have
been

Less
than
3%
at
95th
percentile

Less
than
10%
at
99th
percentile

Less
than
20%
at
99.9th
percentile
"
Exposure"
Comparison
 
1000
ppm
Fresh
Tomato
"
Exposure"
Comparison
 
1000
ppm
Fresh
Tomato
1.13
1.33
1.32
95
2.30
2.47
2.90
99
5.35
6.69
6.26
99.9
Adults
20­
49
1.17
1.13
1.11
95
3.58
3.57
4.02
99
9.10
10.82
10.45
99.9
Children
3­
5
DEEM
CARES
Lifeline
%­
ile
AGE
GROUP
Slide
82
of
168
Slide
83
of
168
Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations
Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations
Steve
Nako
Steve
Nako

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
84
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
85
of
168
Slide
86
of
168
Overview
Overview

Introduction

Framework
for
Model
Comparison

Simple
Example

SAS
Model
Approximations
Slide
87
of
168
Introduction
Introduction

Model
Comparisons
to
Date:


Same
inputs,
same
outputs
 
most
of
the
time

The
Olive
Example
Slide
88
of
168
Introduction
Introduction

Dietary
Exposure
Models
are
not
 
or
at
least
should
not
be
­
black
boxes

Basic
information
­
CSFII
Food
Diaries
and
FCID
Recipes
­
are
publicly
available
to
know
what
people
eat
Slide
89
of
168
Overview
Overview

Introduction

Framework
for
Model
Comparison

Model
Comparison
Framework

Outputs
 
Calculating
a
99.9th
percentile

Inputs:


Food
Diaries

Food
Recipes
(
Supplemental
Slides)


Simple
Example

SAS
Model
Approximations
Framework
for
Dietary
Exposure
Models
Framework
for
Dietary
Exposure
Models
Model
Weight
RefPop
Bodyweight
(
BSA)

Binning
Methodology
(
Consumption
Diaries)

Reference
Population
Model
Design
Components
°
The
Reference
Population
is
a
`
sample'
from
the
Modeled
Population
(
I.
e.,
U.
S.
Population)

°
The
Reference
Population
and
the
Binning
Methodology
determine
the
Model's
expected
frequency
of
using
each
of
the
CSFII
diaries
°
The
Model
Weights
project
the
simulated
exposure
days
to
the
modeled
population
°
The
bodyweights
for
the
Reference
Population
may
affect
exposure
if
these
bodyweights
are
used
to
determine
consumption
(
Lifeline)

Exposure
Distribution
for
Modeled
(
US)
Population
Slide
90
of
168
Model
Comparison
Framework
Model
Comparison
Framework
Equal
Weights
(
Random)

CARES
(
Stratified)

CSFII
Sampling
Weights
Model
Weight
US
Population
(
Natality,
Mortality)

US
Population
US
Population
Modeled
Population
CSFII
normalized
Gower
Dissimilarity
Index
Census
(
PUMS)
CARES
Lifeline
(
NHANES)

CSFII
normalized
RefPop
Bodyweight
Random
(
Age,
Season)

Random
(
2
Day
Diaries)

Binning
Methodology
Natality
(
NCHS)

CSFII
Survey
(
2
Day)

Reference
Population
Lifeline
Calendex
Model
Design
Components
Slide
91
of
168
Outputs:
Comparing
Upper
Percentile
of
Projected
Population
Outputs:
Comparing
Upper
Percentile
of
Projected
Population
Slide
92
of
168
Slide
93
of
168
Outputs:
Subpopulations
of
Interests
Outputs:
Subpopulations
of
Interests

Subpopulations
include:


Infants

Toddlers
(
1­
2,
3­
5
years
old)


Youths
(
6­
12)


Teenagers
(
13­
19)


Adults
(
20­
49,
50+)


Females
of
Childbearing
Age
(
13­
49)


Addressing
Acute
(
1
Day)
Exposure

Not
Addressing
Subchronic,
Chronic
and
Lifetime
Exposure
Inputs:
Food
Diaries
USDA
Consumption
Survey
Food
Intake
Individuals
(
CSFII,
1994­
98)

Inputs:
Food
Diaries
USDA
Consumption
Survey
Food
Intake
Individuals
(
CSFII,
1994­
98)

Further
information
is
available
at
the
web
site:

www.
barc.
usda.
gov/
bhnrc/
foodsurvey
Slide
94
of
168
Features
of
Aggregate
Exposure
Models
Features
of
Aggregate
Exposure
Models
No

No,
but
OK
to
use
DEEM
CEC

Acute
CEC
Critical
Exposure
Commodities
CSFII
(
soon
to
have
FCID)

FCID
(
pretty
recent)

FCID
(
recent
version)

FCID
(
recent
version)

Food
Recipe

Seasonal
Tables



%
aPAD
at
Upper
Percentile


Seasonal
CDF

Lifeline
No

w/
o
Temporal

DEEM


Calendar
Day

CARES

Residential

Calendar
Day
Water

Food
Calendex
Source
/

Output
Slide
95
of
168
Food
Diaries
&
Food
Recipes
Food
Diaries
&
Food
Recipes
RAC
Residues
Food
Diaries
Recipe
Slide
96
of
168
Food
Diaries
&
Food
Recipes
Food
Diaries
&
Food
Recipes
Calendex/
CARES
RAC
Residues
`
RAC'
Diaries
Food
Diaries
Recipe
DEEM­
Calendex
&
CARES
use
the
Food
Recipes
to
convert
Food
diaries
into
RAC
diaries.

These
models
perform
the
Dietary
Risk
Assessment
and
Critical
Exposure
Commodity
(
CEC)
Analyses
on
RACs
(
e.
g.,
Beef,
Lettuce,
Tomato­
fresh,
Tomato­
proc,
Wheat,
etc.).
Slide
97
of
168
Food
Diaries
&
Food
Recipes
Food
Diaries
&
Food
Recipes
LifeLine
Lifeline
uses
the
Food
Recipes
to
convert
RAC
residues
into
Food
residue
distributions.
The
Lifeline
model
performs
the
Dietary
Risk
Assessment
on
Foods
(
e.
g.,
Hamburgers).

`
Food'
Residues
RAC
Residues
Food
Diaries
Recipe
Slide
98
of
168
Food
Diaries
&
Food
Recipes
Food
Diaries
&
Food
Recipes
LifeLine
Calendex/
CARES
`
Food'
Residues
RAC
Residues
`
RAC'
Diaries
Food
Diaries
Recipe
Lifeline
use
the
food
recipes
differently
than
Calendex
&
CARES,
and
also
has
a
slightly
different
recipe.
Models
may
provide
different
predictions
due
to
current
use
of
different
recipes.
Slide
99
of
168
Slide
100
of
168
Overview
Overview

Introduction

Model
Comparison
Framework

Simple
"
10
Diary"
Example

Baseline
Model:
Equal
Contribution
from
Diaries
(
L2)


Different
Model
Weights
(
D1,
C1)


Stratified
Sampling
and/
or
Frequency
of
Use

Lifeline
Bodyweights
(
L1)


Lifeline
CSFII
Option
(
L3)


Estimating
Exposure
from
SingleRAC
Consumption
Approximations

Estimating
Exposure
with
Multiple
Commodities

SAS
Model
Approximations
Illustration:
Apple
Consumption
Illustration:
Apple
Consumption
This
`
baseline'
assumes
that
the
ten
food
diaries
all
have
the
same
expected
contribution
to
the
consumption
distribution
for
this
subpopulation.
Slide
101
of
168
Distribution
of
Apple
Consumption
(
L2)

(
10
hypothetical
3
to
5
yr
olds)

Distribution
of
Apple
Consumption
(
L2)

(
10
hypothetical
3
to
5
yr
olds)
Slide
102
of
168
Distribution
of
Apple
Consumption
(
D1,
C1)

Variable
Modeling
Weights
and/
or
use
Frequencies
Distribution
of
Apple
Consumption
(
D1,
C1)

Variable
Modeling
Weights
and/
or
use
Frequencies
Slide
103
of
168
Variable
Contributions
from
Diaries
(
D1­
C1)

Different
Model
Weights,
and/
or
use
Frequencies.

Variable
Contributions
from
Diaries
(
D1­
C1)

Different
Model
Weights,
and/
or
use
Frequencies.
Slide
104
of
168
Illustration:
Apple
Consumption
Illustration:
Apple
Consumption
Applying
Lifeline's
modeled
bodyweights
to
all
diaries
(
NHANES­
based
regression
for
12
groups
(
gender­
race­
ethnicities).
Slide
105
of
168
Accounting
for
Lifeline
Bodyweights
Accounting
for
Lifeline
Bodyweights
Shaded
cells
indicate
synthesized
diaries
Slide
106
of
168
Variability
due
to
Lifeline
Bodyweights
(
L1)

Variability
due
to
Lifeline
Bodyweights
(
L1)

Slide
107
of
168
Variability
with
Lifeline
Bodyweights
(
L1)

Variability
with
Lifeline
Bodyweights
(
L1)

Slide
108
of
168
Comparing
Model
Consumption
Approximations
(
D1­
C1,
L1)

Comparing
Model
Consumption
Approximations
(
D1­
C1,
L1)
Slide
109
of
168
Lifeline
CSFII­
Dietary
Bin
Option
Lifeline
CSFII­
Dietary
Bin
Option
Synthesized
diaries
Weights
adjusted
to
reflect
expected
frequency
of
drawing
diaries
using
CSFII
option
in
Lifeline
Slide
110
of
168
Comparing
Model
Consumption
Approximations
Comparing
Model
Consumption
Approximations
Using
CSFII
option
in
Lifeline
(
L3),
has
the
effect
of
weighting
each
diary
more
similar
to
Calendex
(
D1).
The
Lifeline
bodyweights
may
continue
to
cause
differences
in
model
predictions.
Due
to
these
two
effects,
this
option
may
not
provide
an
estimate
that
is
closer
to
Calendex.
Slide
111
of
168
Dietary
Exposure
=
Consumption
x
Residues
Dietary
Exposure
=
Consumption
x
Residues
Approximating
Consumption
Approximating
Residues
Approximating
Dietary
Exposure
from
Apples
In
this
simple
`
SingleRAC'
scenario
(
single
residue
=
1000
ppm,
PCT=
100%),
the
distribution
of
exposure
is
equal
­
in
value
­
to
the
consumption
distribution
for
that
subpopulation
Slide
112
of
168
Accuracy
of
Model
Approximations:

Fresh
Tomatoes
(
1000
ppm,
PCT=
100%)

Accuracy
of
Model
Approximations:

Fresh
Tomatoes
(
1000
ppm,
PCT=
100%)

3to5
Yr
Olds
Age
Group
1.02
1.10
1.09
1.13
1.17
1.11
95
10.84
9.14
11.48
10.82
9.10
10.45
99.9
CARES
(
C1)

DEEMCalendex
(
D1)

Lifeline
(
L1)

CARES
DEEMCalendex
Lifeline
SAS
Approximations
Actual
Model
Perce
ntile
Slide
113
of
168
Accuracy
of
Model
Approximations:

Fresh
Tomatoes
(
1000
ppm,
PCT=
100%)

Accuracy
of
Model
Approximations:

Fresh
Tomatoes
(
1000
ppm,
PCT=
100%)

20to4
9
Yr
Olds
Age
Group
1.07
1.31
0
1.33
1.13
1.32
95
4.77
6.19
10.65
6.69
5.35
6.26
99.9
CARES
(
C1)

DEEMCalendex
(
D1)

Lifeline
(
L1)

CARES
DEEMCalendex
Lifeline
SAS
Approximations
Actual
Model
%­
ile
Slide
114
of
168
Approximating
Dietary
Exposure
with
Residue
Data
Approximating
Dietary
Exposure
with
Residue
Data
Approximating
Consumption
Anticipated
Residues
Approximating
Dietary
Exposure
from
Apples
The
exposure
distributions
are
harder
to
anticipate
with
more
typical
residue
scenarios.
Slide
115
of
168
Approximating
Exposure
with
Complex
Residues
Approximating
Exposure
with
Complex
Residues
Slide
116
of
168
Estimating
Exposure
with
Multiple
Commodities
Estimating
Exposure
with
Multiple
Commodities
SAS
utility
uses
SingleRAC
consumption
approximations
to
estimate
exposure
at
some
upper
(
99.9th)
percentile
when
pesticide
is
used
on
multiple
commodities
Slide
117
of
168
Estimating
Exposure
Estimating
Exposure
We
estimate
exposure
at
the
99th%
to
be
0.04
mg
ai/
kg
bwt;
and
calculate
apples
contributing
51%,
bananas
32%,
and
olives
contributing
17%
of
the
Total
exposure
for
this
upper
1
percentile
Slide
118
of
168
Slide
119
of
168
Notes
on
Applying
SingleRAC
Approximations
Notes
on
Applying
SingleRAC
Approximations

Initial
intent
of
SAS
utility
was
to
provide
Lifeline
user
with
a
quick
(~
1
minute)
`
a
priori'
estimate
of
exposure
at
99.9th,
as
well
as
some
idea
of
the
Critical
Commodities
contributing
at
this
upper
percentile

Assume
that
simulated
exposure
days
at
the
upper
percentile
obtain
exposure
from
primarily
one
commodity

This
approach
underestimates
aggregate
dietary
exposure
since
it
does
NOT
account
for
co­
occurrent
exposure
from
multiple
commodities.
We
need
to
run
the
probabilistic
(
Monte
Carlo)
risk
assessment
model
to
properly
account
for
aggregate
dietary
exposure
Slide
120
of
168
Notes
on
Applying
SingleRAC
Approximations
Notes
on
Applying
SingleRAC
Approximations

If
a
Single
Commodity
provides
exposure
exceeding
the
aPAD
to
some
expected
percent
of
the
population
(>
0.1%),
then
the
probabilistic
models
will
also
exceed
the
aPAD
at
the
corresponding
upper
percentile
(
99.9th)


A
pesticide
with
100
uses
is
a
pesticide
with
100
stories
Slide
121
of
168
Overview
Overview

Introduction

Model
Comparison
Framework

Simple
Example

Constructing
SAS
Model
Approximations
for
Calendex,

CARES
and
Lifeline
Slide
122
of
168
Why
Approximate
Models?

Why
Approximate
Models?


We
highlighted
several
design
components
(
Reference
Population,

Binning
Method,
Model
Weights
and
Bodyweights)
that
explain
how
these
models
work

We
developed
the
SAS
model
approximations
to
help
us
identify
the
effects
of
each
design
component

Hypothetical
model
approximations
were
also
developed
(
e.
g.,
L2)
to
help
us
control
for
other
factors
while
focusing
on
the
effect
of
a
specific
component
Approximating
Calendex
Approximating
Calendex
CSFII
Sampling
Weights
Model
Weight
US
Population
Modeled
Population
CSFII
normalized
RefPop
Bodyweight
Random
(
2
Day
Diaries)

Binning
Methodology
CSFII
Survey
(
2
Day)

Reference
Population
Calendex
Model
Design
Components
Calendex
constructs
a
365
day
consumption
profile
by
randomly
drawing
from
the
2
daily
diaries
that
the
20,607
individuals
provided.
We
can
use
the
CSFII
sampling
weights
to
approximate
the
contribution
from
each
diary
since
all
diaries
have
the
same
expected
frequency
of
use.

Since
both
Calendex
and
DEEM
apply
the
CSFII
sampling
weights
to
project
simulated
exposure
days
to
the
US
Population,
the
CEC
utility
from
the
DEEM
can
be
used
for
Calendex.
Slide
123
of
168
Calendex:
Dietary
Bin
Calendex:
Dietary
Bin
CSFII
Diaries
20,607
2
1
RefPop
Person
2,603
31,887
74,723
CSFII
Weight
Calendar
Day
xx­
02­
1
xx­
02­
1
xx­
02­
2
xx­
02­
2
xx­
02­
1
xx­
02­
1
365
5
4
3
2
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
.
Since
all
diaries
have
the
same
expected
frequency
of
use,
the
Model
Weights
reduce
to
CSFII
sampling
weights
Slide
124
of
168
DEEM­
Calendex:
Model
(
CSFII)
Weights
DEEM­
Calendex:
Model
(
CSFII)
Weights
There
is
considerable
variation
in
the
CSFII
sampling
weights
within
age
groups.
Slide
125
of
168
Approximating
Calendex
(
D1)

Approximating
Calendex
(
D1)

Slide
126
of
168
Approximating
Calendex
(
D1,
D2)

Approximating
Calendex
(
D1,
D2)

Slide
127
of
168
Approximating
CARES
Approximating
CARES
CARES
(
Stratified)

Model
Weight
US
Population
Modeled
Population
CSFII
normalized
Gower
Dissimilarity
Index
Census
(
PUMS)
CARES
RefPop
Bodyweight
Binning
Methodology
Reference
Population
Model
Design
Components
CARES
constructs
a
365
day
consumption
profile
by
randomly
drawing
food
diaries
from
a
bin
constructed
by
use
of
a
Gower
Dissimilarity
Index.

Since
the
CARES
reference
population
is
a
stratified
sample
from
the
US
Census
 
PUMS,
the
CARES'

model
weights
are
needed
to
project
each
simulated
person­
day
to
the
modeled
population.

Since
the
Food
Match
Table
provides
the
actual
sequence
of
CSFII
diaries
used
in
each
CARES
run,
it
is
not
necessary
to
calculate
the
expected
number
of
times
that
each
diary
will
be
used
by
CARES.
Slide
128
of
168
CARES:
Dietary
Bin
CARES:
Dietary
Bin
CARES:
Dietary
Bin
CSFII
Diaries
100,000
2
1
RefPop
Person
2,603
31,887
74,723
CARES
Weight
Calendar
Day
xx­
02­
1
xx­
02­
1
xx­
02­
2
xx­
02­
2
xx­
02­
1
xx­
02­
1
365
5
4
3
2
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
CARES
(
bin)
use
a
Gower
Dissimilarity
index
to
create
bins
from
which
CSFII
diaries
were
selected
for
each
of
the
100,000
individuals
in
the
Reference
Population.

For
each
day,
the
CARES
model
weights
sum
to
241,343,436.
Slide
129
of
168
CARES
Variation
in
Frequencies
of
using
CSFII
diaries
CARES
Variation
in
Frequencies
of
using
CSFII
diaries
0
5,000
10,000
15,000
20,000
25,000
Count
Infants
1to2
3to5
6to12
13to19
20to49
50plus
Slide
130
of
168
Variation
in
CARES
Model
Weights
Variation
in
CARES
Model
Weights
Slide
131
of
168
Approximating
CARES
(
C1)

Approximating
CARES
(
C1)
Slide
132
of
168
Approximating
Lifeline
Approximating
Lifeline
Equal
Weights
(
Random)

Model
Weight
US
Population
(
Natality,
Mortality)

Modeled
Population
Lifeline
(
NHANES)

RefPop
Bodyweight
Random
(
Age,
Season)

Binning
Methodology
Natality
(
NCHS)

Reference
Population
Lifeline
Model
Design
Components
Lifeline
generates
its
reference
population
by
randomly
drawing
from
the
1996
NCHS
Natality
statistics.
All
individuals
are
equally
weighted.

Lifeline
constructs
lifetime
consumption
profiles
by
randomly
drawing
from
dietary
bins
based
on
age
and
season.

Since
consumption
profiles
are
generated
on
the
fly,
we
need
to
calculate
the
expected
number
of
times
that
each
diary
is
used
to
approximate
the
Lifeline
Model.

Slide
133
of
168
Lifeline:
Dietary
Bin
Lifeline:
Dietary
Bin
CSFII
Diaries
10,000
2
1
RefPop
Person­

Age
1
1
1
Lifeline
Weight
Day
of
Life
xx­
02­
1
xx­
02­
1
xx­
02­
2
xx­
02­
2
xx­
02­
1
xx­
02­
1
31,390
5
4
3
2
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
01­
2
xx­
01­
1
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
xx­
02­
1
xx­
02­
2
All
modeled
individuals
are
equally
weighted.
All
diaries
within
a
dietary
bin
have
the
same
expected
frequency
of
being
used
by
Lifeline.

Slide
134
of
168
Slide
135
of
168
Lifeline
Model
Approximation
(
L1)

Lifeline
Model
Approximation
(
L1)


For
each
subpopulation
(
3to5
yr
olds),
select
All
CSFII
diaries
in
DietBin(
s).


Match
each
CSFII
record
with
different
bodyweight
in
corresponding
bodyweight
bin
and
calculate
Consumption
(
grams
food/
kg
bwt).


Compute
Lifeline
Model
Weight
that
reflects
the
expected
number
of
uses
for
diary
(
Weight=
1
for
infants,
Weight=
4
for
8
yr
old
diary,
Weight=
1
for
a
12
yr
old
diary
in
the
6­
12
yr
old
subpopulation,
or
Weight=
2
for
a
12
yr
old
diary
in
the
13­
19
yr
old
subpopulation).


For
each
commodity
(
RAC­
FF),
calculate
the
Total
Number
of
Consumers
=
SUM
weights
for
all
diaries
(
w/
Consumption
>
0),
and
then
calculate
the
Total
Number
of
Non­

Consumers
=
Total
Pop.
 
Total
Consumers.


Rank
the
diaries
by
Consumption
(
grams
food/
kg
bwt),

assign
GroupID,
and
Calculate
the
Average
Consumption
and
corresponding
Weights
(
sum)
for
each
Group,
C_
1
to
C_
100
(
Supplemental
Slide).
Lifeline
Model
Weights
Lifeline
Model
Weights
Diary
Model
Weight
i
=
Expected
#
Uses
in
Lifeline
=
Expected
#
Draws
from
DietaryBin
x
Prob(
Selecting
Diary
|
DietaryBin).

While
each
diary
within
a
dietary
bin
has
the
same
chance
of
being
used,

we
need
to
make
adjustments
since
we
are
interested
in
calculating
percentiles
for
EPA
subpopulations
that
intersect
with
these
Lifeline
dietary
bins.

E.
g.,
if
we
are
calculating
an
upper
percentile
for
the
overall
US
population,
then
we
need
to
account
for
the
fact
that
we
have
a
disproportion
number
of
food
diaries
for
children.

[
About
54%
(
23,074
of
42,269)
diaries
were
from
respondents
<
20
yrs
old;
these
first
20
yrs
account
for
approx.
23%
(
21
of
86
yrs)
of
the
modeled
lifetime.]
Slide
136
of
168
Lifeline
Dietary
Bins
 
Number
of
CSFII
Diaries
Lifeline
Dietary
Bins
 
Number
of
CSFII
Diaries
Slide
137
of
168
Lifeline
Dietary
Bin
&
Model
Weights
Lifeline
Dietary
Bin
&
Model
Weights
Since
Lifeline
applies
its
own
bodyweight
model
to
calculate
consumption
&
dietary
exposure,
each
diary
needs
to
be
replicated
to
account
for
different
permutations
of
bodyweights
(
Lifeline
Age)
that
may
be
expected
in
Lifeline.

Using
Lifeline
data,
we
generated
bodyweight
bins
for
various
age
groups
to
minimize
the
number
of
replications
needed
to
account
for
consumption
variability
created
by
the
Lifeline.
Rather
than
creating
4
separate
permutations
of
a
8
year
old
food
diary
with
the
bodyweight
distributions
of
8,
9,
10
and
11
year
olds,
we
constructed
the
bodyweight
distribution
for
this
8­
11
year
old
group,
and
created
1
set
of
permutations
for
this
bodyweight
distribution.
Slide
138
of
168
Lifeline
Bodyweights
Lifeline
Bodyweights
For
the
children
subpopulations,
we
created
12
permutations
of
bodyweights
(
0th,
5th,
15th,
25th,
35th,
45th,
50th,
65th,
75th,
85th,
95th,
100th
percentiles).

For
adult
subpopulations
(
20to49
yrs
old,
50plus,
Females
13to49
yrs
old),
we
created
only
5
permutations
(
10th,
30th,
50th,
70th,
90th
percentiles)
due
to
increasing
size
of
data
base.

Slide
139
of
168
Approximating
Lifeline
(
L1)

Approximating
Lifeline
(
L1)
Slide
140
of
168
Slide
141
of
168
Olive
Consumption
(
D1,
C1,
L1)

Olive
Consumption
(
D1,
C1,
L1)
Slide
142
of
168
CASE
STUDY
CASE
STUDY
David
J.
Miller
David
J.
Miller

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
143
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
144
of
168
Slide
145
of
168
Case
Study
Case
Study

In
order
to
compare
exposure/
risk
estimates
from
the
three
models,
OPP
developed
hypothetical
case
study

Exposure
through
food,
water,
and
residential
turf
scenarios

Only
food
+
water
considered
here

Widely
used
insecticide

Agricultural
uses
include
fruits,

vegetables,
nuts,
and
grains

>
100
tolerances

Registered
on
over
300
crops
in
agriculture

High
usage
on
apples,
pecans
and
grapes
Slide
146
of
168
Case
Study
Case
Study

As
part
of
case
study,
OPP
assessed
exposure/
risk
to
two
groups

Children
3­
5
y.
o.


Adults
20­
49
y.
o.


Food

Primarily
PDP
data

Water

PRZM­
EXAMS
assessments
Slide
147
of
168
Case
Study
Case
Study

Given
different
assumptions
and
databases
used
by
each
of
the
three
models,
exposure
estimates
would
not
be
expected
to
be
identical,
but
should
be
reasonably
similar
to
each
other
Percent
aPAD
Occupied
Percent
aPAD
Occupied
CARES
DEEM
LL
CARES
DEEM
LL
68
62
52
4
5
3
Food
+

Water
60
52
37
2
2
1
Water
30*

29
32
1
3
1
Food
99.9th
Percentile
95th
Percentile
ADULTS
20­
49
y.
o.
Slide
148
of
168
Percent
aPAD
Occupied
Percent
aPAD
Occupied
CARES
DEEM
LL
CARES
DEEM
LL
129
116
115
9
14
7
Food
+

Water
89
80
78
3
3
3
Water
81*

81
77
5
12
4
Food
99.9th
Percentile
95th
Percentile
CHILDREN
3­
5
y.
o.
Slide
149
of
168
Slide
150
of
168
Exposure
Contributors
Exposure
Contributors

A
routine
requirement
for
OPP
is
to
identify
crops
which
contribute
significantly
to
exposure

Two
methods
available
for
this
comparison

Share
of
total
exposure

Number
of
High
exposure
events

DEEM
and
CARES
both
have
this
capability

Capability
currently
being
developed
for
Lifeline
Slide
151
of
168
Exposure
Contributors
Exposure
Contributors

Both
DEEM
and
CARES
identified
similar
"
top"
contributors
to
exposure
to
include

Strawberries

Olives

Blueberries

Almonds

Peaches
Exposure
Contributors,
Ranked
Exposure
Contributors,
Ranked
7
6
6
4
Peaches
5
5
5
5
Blueberry
4
3
4
3
Strawberry
juice
3
4
3
7
Almonds
2
2
2
2
Olives
1
1
1
1
Strawberries
CARES
DEEM
CARES
DEEM
No.
High
Exp.

Events
Share
Total
Exposure
Children
3­
5
y.
o.
Slide
152
of
168
Exposure
Contributors,
Ranked
Exposure
Contributors,
Ranked
3
12
6
13
Beet
Tops
5
7
5
6
Peaches
4
4
4
7
Blueberries
6
3
3
1
Olives
1
2
2
3
Almonds
2
1
1
2
Strawberries
CARES
DEEM
CARES
DEEM
No.
High
Exp.

Events
Share
Total
Exposure
Adults
20­
49
Slide
153
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
154
of
168

Background

Goals,
objectives
and
focus
of
current
model
comparisons
in
OPP

Future
model
comparisons
in
OPP

Probabilistic
models
and
past
SAP
Reviews

Side­
by­
Side
comparison
of
dietary
results
­
DEEM
and
Lifeline

Approximating
Consumption
Distributions
for
Probabilistic
Models
using
SAS
Simulations

Case
Study

Summary
Presentation
Roadmap
Slide
155
of
168
Slide
156
of
168
Summary
Summary

OPP
previously
sponsored
SAPmodel
review
sessions
 
for
the
LifeLine,
CARES,

DEEM/
Calendex
and
SHEDS
models

Each
of
these
models
have
been
used
­
and
continue
to
be
used
­

to
assess
pesticide
exposure
by
various
stakeholders
Slide
157
of
168
Summary
Summary
Today
we:


Discussed
the
results
of
our
investigation
of
model­
design
features
for
DEEM/
Calendex,

CARES,
and
LifeLine

Compared
1­
day
dietary
exposures
estimates
(
food
+

water)
for
these
models
Slide
158
of
168
Summary
Summary

We
applied
our
SAS
programs
to
single
commodity
and
multicommodity
case
studies
to
demonstrate
how
model
design
impacts
1­
day
dietary
exposure
estimates

We
found
the
1­
day
dietary
exposure
estimates
(
food
and
water)
for
the
three
models
to
be
very
close
Slide
159
of
168
Summary
Summary

We
found
it
to
be
tedious
and
labor
intensive
to
format
large
data
sets
for
multiple
models,

each
designed
to
accept
data
in
different
formats

We
found
that
apparently
large
differences
in
dietary
exposure
estimates
were
due
to
data
entry
errors
Slide
160
of
168
Questions
to
SAP
Questions
to
SAP
Slide
161
of
168
Background
for
Questions
1.1
(
paraphrased)

Background
for
Questions
1.1
(
paraphrased)


EPA
determined
that
the
3
models
investigated
differ
in
how
they:


Establish
a
reference
population

Bin
food
diaries
to
generate
longitudinal
profiles

Apply
model
weights

Estimate
body
weights
for
individuals
Slide
162
of
168
Question
1.1
Question
1.1

The
SAP
is
asked
to
please
comment
on
whether
the
above
cited
model
design
features
reflect
those
most
likely
to
result
in
differences
in
dietary
[
food
and
water]
exposure
estimates
based
on
identical
data
sets.
If
not,
what
other
model
design
features
are
likely
to
cause
different
dietary
exposure
estimates?
Slide
163
of
168
Question
1.2
Question
1.2

The
SAP
is
asked
to
please
comment
on
the
approach
taken
by
the
Agency
to
develop
and
use
SAS
approximation
models
(
see
Section
IV
of
the
background
document)
to
attribute
differences
in
model
predictions
from
differences
in
model
designs.
Please
suggest
possible
improvements
or
refinements
to
these
SAS
approximation
models
and
to
alternative
methods
for
comparing
model
predictions.
Slide
164
of
168
Question
2.1
Question
2.1

The
SAP
is
asked
to
please
comment
on
the
different
approaches
used
by
the
three
models
in
developing
their
Reference
Populations
and
model
weights.
Slide
165
of
168
Question
3.1
Question
3.1

The
SAP
is
asked
to
please
comment
on
the
frequency
that
CSFII
diaries
are
used
by
the
various
models.
Are
there
any
potential
biases
that
may
arise
in
the
respective
dietary
exposure
estimates
for
these
models
as
a
result
of
how
they
used
CSFII
records?

Considering
Lifeline's
current
dietary
bin
design
(
age,
season),
please
comment
with
respect
to
the
use
of
the
CSFII
survey
weight
option.
Is
either
Lifeline
option
(

CSFIIweighted
or
not)
generally
more
appropriate
than
the
other
or
are
there
circumstances
in
which
one
might
be
preferable
to
the
other?
Slide
166
of
168
Background
for
Question
4.1
Background
for
Question
4.1

Two
methods
for
estimating
dietary
contributors
at
the
high
end
of
the
exposure
distribution
were
described:

1.
The
Critical
Exposure
Contribution
(
CEC)
method
quantifies
the
contribution
of
specific
commodities
(
RAC­
FF)
to
the
total
exposure
at
the
upper
percentiles
of
the
exposure
distribution.
Slide
167
of
168
Background
for
Question
4.1,

Cont.

Background
for
Question
4.1,

Cont.

2.
The
`
frequency­
exceeded'
method
tabulates
the
frequency
that
consumption
of
a
particular
commodity
(
RAC­
FF)
causes
exposure
to
exceed
some
level
of
concern.
Slide
168
of
168
Question
4.1
Question
4.1

The
SAP
is
asked
to
please
comment
on
the
relative
merits
of
the
two
approaches
described
above
(
CEC
and
frequency­
exceeded)
for
identifying
significant
contributors
(
RAC­
FF)
to
exposure
at
the
upper
percentiles
of
exposure.
Are
there
other
methods
or
techniques
which
the
Panel
might
recommend
for
accomplishing
this
important
part
of
the
dietary
exposure
assessment?
