Model Averaging Webinar Workshop Announcement

EPA’s National Center for Environmental Assessment (NCEA) hosted a public workshop on model averaging methods for dose-response analysis. The workshop was held by webinar in two half-day increments (12:00 PM–4:00 PM EST) on December 10-11, 2015. Most of the original text of this page has been preserved to document conditions and requirements set forth for the workshop.

04/15/2016: EPA has reviewed and approved for release the Model Averaging Workshop Report. The report includes the pre-meeting written responses from discussants, discussants' comments during the workshop to the EPA discussion questions, and their responses to questions asked by workshop audience participants.

On this page:

Workshop Details
Workshop Registration
Supporting Materials for the Workshop

Workshop Overview

The purpose of this workshop was to obtain expert peer consultation on model averaging methods for dose-response analysis. To facilitate the workshop, EPA provided invited workshop consultants with background and support materials for consideration, including a document describing methods that are consistent with published recommendations and software that illustrates how those existing methods could be implemented.

Workshop consultants examined EPA’s background and support material and provided written comments prior to the workshop. Discussions and conclusions reached during the workshop formed the basis of a workshop report that EPA considered for the identification of the model averaging method(s) that offer(s) the greatest potential for the development of chemical health assessments.

While the workshop support material and software describe candidate methods for use in evaluating continuous response measures, participants were also asked to comment on the potential for extension of any preferred method(s) identified to dichotomous response measures.

Conference Link and Participation Guide

Online conferencing was provided by Adobe Connect.

You can download a 2-page Virtual Participation Guide from the Workshop Support Materials page. It contains basic information on how to use Adobe Connect for this conference.

Location/Date

This workshop was held only by webinar. The workshop was open to attendance by interested public attendees and/or speakers on a first-come, first-served basis up to the limits of available webinar room.

This workshop was held December 10-11, 2015 from 12:00 pm - 4:00 pm EST.

Public Participation

Instructions for workshop registration were described on the Workshop Registration tab. Interested parties with expertise in this field of study were invited to provide their perspectives by registering for the workshop and preparing a brief presentation (= 5 minutes) for the workshop consultants. Instructions for requesting to make a presentation at the workshop were provided on the Workshop Registration tab.

Background

The benchmark dose (BMD) method is a key component of chemical risk assessments performed by EPA, other U.S Federal and State Agencies, and multiple international organizations. The BMD method is used to establish toxicity values such as EPA reference doses and cancer slope factors for chemicals expected to have human health effects. Existing BMD methods involve the selection of a single dose-response model from among a suite of models with little biological basis, based largely on a comparison of model fits to the data. As such, they have been criticized for not adequately accounting for model uncertainty and biological considerations. Model averaging is a well-recognized and published scientific approach that can assist in the characterization of model uncertainty, and has the potential for incorporating biological considerations. Bayesian methods can be applied to allow prior information, including biological considerations, to influence model averaging results. Given these potential benefits, model averaging and several model averaging approaches are being investigated and tested by EPA.

Discussants

Name	Affiliation
David Dunson	Duke University
Ruth Hummel	US EPA
Michael Messner	US EPA
Walter Piegorsch	University of Arizona
Woody Setzer	US EPA
Matthew Wheeler	National Institute for Occupational Safety and Health (NIOSH)

A PDF version of the discussants' response to the discussion questions is available for download from the Model Averaging Workshop Support Materials page.

Agenda

Thursday, December 10, 2015

Time	Activity
12:00 – 12:20	Welcome and Opening Remarks
12:20 – 1:00	Public Presentations
1:00 – 1:20	Discussion Question 1: Overall approach to model-averaging Are there other model averaging methods that EPA should consider?
1:20 – 2:00	Discussion Question 2, 5: Completeness of suite of models Are there other parametric models that should be included in model-averaging? If so, do you recommend testing performance of the new suite?
2:00 – 2:15	BREAK
2:15 – 3:00	Discussion Question 3: Implementation of methods Do you agree with the approaches used to implement the methods reviewed in the workshop support material?
3:00 – 3:30	Discussion Question 4: Testing approach Should additional testing be performed to identify a model averaging approach for dose-response analyses that offers the greatest advantage for the development of chemical health assessments?
3:30 – 4:00	Day 1 Audience Questions and Day 1 Wrap-Up
4:00	ADJOURN

Friday, December 11, 2015

Time	Activity
12:00 – 12:20	Welcome and Day 1 Recap
12:20 – 1:45	Discussion Question 6: Motives for using model averaging in health assessments a. How useful is model averaging for characterizing model uncertainty? b. How useful is model averaging for incorporating prior information? Discussion Question 7: Should alternatives or complements to model averaging be investigated? Examples: Isotonic regression; Non-parametric and semi-parametric (Bayesian and frequentist) modeling; Full Bayesian model averaging; Flexible parametric models
1:45 – 2:00	BREAK
2:00 – 2:40	Discussion Question 8: Dichotomous Data Describe any major concerns for the application of methods described in this report to dichotomous data. How do the results of the present background paper on models for continuous data compare to published work on model averaging for dichotomous models?
2:40 – 3:10	Discussion Question 9: Is Model Averaging Ready for Use in Chemical Health Assessment? Is model averaging as implemented in the workshop support material suitable for use in chemical health assessments, possibly with some reservations or precautions? Can you identify circumstances when model averaging may be helpful and informative or misleading?
3:10 – 3:30	Discussion Question 10: Conclusions Do you agree with the conclusions made in Section 4.1 of the workshop support material?
3:30 – 4:00	Audience Questions and Final Wrap-Up
4:00	ADJOURN

Workshop Registration

Registration for the workshop has closed.

To obtain clarification or provide comments on the workshop support materials, contact Jeff Gift, Ph.D., or by phone at 919-541-4828.

Supporting Materials for the Workshop

Model Averaging Webinar Workshop Support Materials. This page includes links to the following PDF documents:
- Report: Background and Support Materials for Peer Consultation Webinar Workshop on Model Averaging Methods for Dose-Response Analysis
- Discussion Questions
- Discussant Responses to the Discussion Questions
- Model Averaging Workshop Virtual Participation Guide
- Model Averaging Workshop Agenda
Model Averaging Application With Graphical User Interface (zip) (1.64 MB, 2015-11-12) - Requires Windows XP or better. A sample data file and instructions are included.
Software for performing model runs on multiple datasets in batch (zip) (5.03 MB)
Real and simulated data for Model Averaging test runs (zip) (3.49 MB)