Attachment H (May 2003)

Methods Activity Group
Status Report for the CCL Work Group

Plenary Meeting
May 12, 2003
The Methods Activity Group has had three

conference calls since the March plenary mtg.
Participants of Work Group (not all participants on all calls) SARA CHECK
Laura Anderko
Douglas Crawford-Brown
Mike Dourson
Alan Elzerman
Brian Ramaley
Colin Stine
Craig Stow
Lynn Thorp
Dan Wartenberg
Tom Carpenter and other EPA staff
Jo Anne Shatkin and other Cadmus staff
Steve Via, AWWA
Jeff Rosen and Dave Drain, Perot Systems Govt Services
Abby Arnold and Sara Litke
Progress toward Deliverables
At the March 27meeting the CCL Work Group

reviewed the gate approach to screen from
the CCL universe of contaminants to the
PCCL.
The activity group has refined the gate
approach and today seeks review and
comment by CCL Work Group.

Gate Approach to screen from CCL Universe to
PCCL
Methods Activity Group proposes a gate
approach
Gates are paths only from the universe to
the PCCL for both microbes and chemicals
Knowledge base for microbes is forthcoming
Overview of Gate Approach

Assumptions:
CCL universe defined by data group, includes all
contaminants for which we are able to gather data
or information.
CCL universe may change over time
Not every contaminant in the CCL universe will be
characterized by having data directly reporting its
known or potential adverse health affects and its
known or potential occurrence.
Some contaminants may be characterized by
surrogate data, such as production data, or by other
indirect measures.
5
Overview of Gate Approach, (cont.)
Four gates describe the criteria for passing through any of the
gates ( correlate with terms demonstrated and potential as
used by NAS)
However, activity group goes beyond words demonstrated and
potential
Definition of terms:
Demonstrated
real measured data on which knowledge rests are available. For

health effects, this includes toxicology study endpoints from human
or animal studies. For occurrence, data means water concentration
data.
Potential
information on the contaminant or on a surrogate contaminant that is

suggestive of, or generally correlates well with, a specific effect or
measure of occurrence.
Overview of Gate Approach, (cont.)
Appropriate gate for a contaminant will be a

function of the nature of the knowledge about
that contaminant.
Contaminant that does not pass through a

gate remains on CCL universe, those that go
through a gate, are listed on the PCCL.
7
Overview of Gate Approach, (cont)

Four Gates:
I.
Quantitative data or measures of adverse

health effects and quantitative data on
concentrations in water.
II.
Information that there may be adverse

health effects and quantitative data on
concentrations in water

III. Quantitative data or measures of adverse
health effects and information that suggests
there may be significant presence or
potential to occur in water.
IV. Information that suggests there may be
adverse health effects and information that
suggests there may be significant presence
or potential to occur in water.
9

V. Nomination process
A nomination gate that allows a contaminant to move
to the PCCL because experts think it ought to
VI. Possible other gates?
10
The Universe of Potential Drinking Water

Contaminants
Contaminants that are
demonstrated to
occur in drinking water
II
Contaminants that have
the potential to cause
adverse health effects
Contaminants that are

demonstrated to cause
adverse health effects
III
IV
Contaminants that have

the potential to occur
in drinking water
Gate I - Quantitative data or measures of adverse health effects and quantitative data
on concentrations in water
Gate II - Information that suggests that there may be adverse health effects and
quantitative data on concentrations in water
Gate III - Quantitative data or measures of adverse health effects and information that
suggests there may be significant presence or concentrations in water
Gate IV - Information that suggests that there may be adverse health effects and
information that suggests there may be significant presence or concentrations in water.
Two Methods for Gates
Semi Quantitative
Bin effects into several

categories reflecting
different potencies (e.g.,
high, medium, low)
Bin exposures into several
categories reflecting
different concentrations
(e.g., NAS 1 to 10
magnitude choices)
Combine bins to establish a
score
Determine cut off score for
PCCL
Quantitative
Estimate reasonably
maximum potency by
Table 1
Estimate reasonably
maximum exposure by
Table 2
Compare exposure to
potency
If ratio is greater than a
certain value (e.g., 1),
then contaminant is on
the PCCL
12
Table 1. Uncertainty factors used in

development of potencies*
Type of toxicity data
Composite
factor
Confidence in
Estimate
Sensitive human NOAEL
High
Average human NOAEL
10
High
Experimental animal only
100
Medium to high
Less than lifetime study
1000
Medium
Lack of a NOAEL
3000
Medium to low
Insufficient studies to determine critical effect
10,000
Low
Lethal dose data only
100,000+
Not applicable
95% UCL of NOAEL based on QSAR
3000
Not applicable
* Choice of uncertainty factor, composite uncertainty factor, and confidence are as defined
by the EPA, except for lethal dose and structure data. Values for these latter two
categories can be found in the literature.
Table 2. Uncertainty factors suggested for

concentration data for the Universe to PCCL*
Composite
factor
Type of data
Confidence in
estimate
99 percentile of tap water measurements
High
Mean of tap water measurements
10
Medium to High
99 percentile of source water
Medium to High
Mean of source water
30
Medium
99 percentile of ambient water
10
Medium to low
Mean of ambient water
100
Low
Detection limit
Not applicable
Maximum concentration based on structure

(information)
Medium
* Choice of uncertainty factor, composite uncertainty factor, and confidence are

wild guesses for discussion purposes only. They should be based on a
comparison of available data for these measurements.
Gate Approach
Next steps:
Detailed review and discussion by the

CCL Work Group and ultimately,
approval of draft gates approach
Test the proposed approach
15
Progress toward Deliverables:

Model for classifying from PCCL to CCL
Deliverable: Recommended decision method

and associated prototype approach(es) for
classifying from the PCCL to the CCL
The group has reviewed a paper Model Fit to Example

Data to understand the model classes and identify
whether a preference could be determined at this point.
An attribute scoring approach to be used on an example
data set to test various approaches.
Reviewed example data set and test various
approaches with raw and scored data.
Next Steps: Identify process to select approach.
16
Progress toward Deliverables (cont.)

Classification Models (PCCL-CCL)
Purpose of model review was to understand how

the models work and what inferences can be drawn
about the models.
The models were evaluated based on how well they
replicated previous decisions. (list/no list)
Through a cross validation analysis, a mean
classification rate was calculated for each of the
models, using raw and scored data.
Multivariate Adaptive Regression Splines (MARS)
had the lowest misclassification rate for both raw
and scored data.
17
Progress toward Deliverables (cont.)

Participants noted however, that misclassification

could be caused by anomalies in the data set and/or
original decisions made in building the first CCL, rather
than due to the models themselves.
The group will continue to explore all four classes of
models using a larger and more robust training data
set.
The Models reviewed:
Artificial neural networks (ANN)

Classification and regression trees (CART)
Logistic regression (a specific form of a generalized
linear model (GLM) and
Multivariate adaptive regression splines (MARS)
18

Next Steps:
EPA is building a data set to fit the four classes of

models
19
Attributes
N. Kim provided overview of NRC rationale

and thinking behind attributes.
20

Attachment H (May 2003)

Uploaded by

Copyright:

Available Formats

Attachment H (May 2003)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Attachment H (May 2003)

Uploaded by

Copyright:

Available Formats

Methods Activity Group

Status Report for the CCL Work Group

The Methods Activity Group has had three

Tom Carpenter and other EPA staff

Jo Anne Shatkin and other Cadmus staff

Steve Via, AWWA

Jeff Rosen and Dave Drain, Perot Systems Govt Services

Abby Arnold and Sara Litke

Progress toward Deliverables

At the March 27meeting the CCL Work Group

Progress toward Deliverables

Knowledge base for microbes is forthcoming

Overview of Gate Approach

Overview of Gate Approach, (cont.)

real measured data on which knowledge rests are available. For

information on the contaminant or on a surrogate contaminant that is

Overview of Gate Approach, (cont.)

Appropriate gate for a contaminant will be a

Contaminant that does not pass through a

Overview of Gate Approach, (cont)

Quantitative data or measures of adverse

Information that there may be adverse

Overview of Gate Approach

Overview of Gate Approach

VI. Possible other gates?

The Universe of Potential Drinking Water

Contaminants that are

Contaminants that have

Two Methods for Gates

Bin effects into several

Table 1. Uncertainty factors used in

Sensitive human NOAEL

Average human NOAEL

Experimental animal only

Less than lifetime study

Insufficient studies to determine critical effect

Lethal dose data only

95% UCL of NOAEL based on QSAR

Table 2. Uncertainty factors suggested for

99 percentile of tap water measurements

Mean of tap water measurements

99 percentile of source water

Mean of source water

99 percentile of ambient water

Mean of ambient water

Maximum concentration based on structure

* Choice of uncertainty factor, composite uncertainty factor, and confidence are

Detailed review and discussion by the

Progress toward Deliverables:

Deliverable: Recommended decision method

The group has reviewed a paper Model Fit to Example

Progress toward Deliverables (cont.)

Purpose of model review was to understand how

Progress toward Deliverables (cont.)

Participants noted however, that misclassification

The Models reviewed:

Artificial neural networks (ANN)

Progress toward Deliverables

EPA is building a data set to fit the four classes of