Bayesian Models and Model Averaging Software (External Review Draft, 2018)
In 2018, the BMDS Bayesian models and Bayesian model averaging (BMA) software package for performing dose-response analyses of dichotomous data was externally reviewed by three experts in Bayesian risk assessment methods and dose-response modeling. Peer review comments are documented in a Bayesian Model Averaging for Dichotomous Data Peer Review Report.
The primary purpose of this review was to obtain external expert opinions on EPA’s proposed model averaging approach for the derivation of BMD and BMDL values from dichotomous dose-response datasets.
Reviewers were asked to read the review documents describing EPA's proposed BMA method and respond to several specific charge questions relating to the proposed method that are documented in the peer review report. They generally agreed that the BMA software package produced statistically valid results. They praised the innovative approaches with comments such as “substantive models going into model averaging are very well thought out,” “highly valuable,” and “…heavily encourage the Agency to adopt the kinds of methods proposed here into its next version of BMDS.”
Several recommendations were made regarding enhancements that could facilitate or expand its practical application, and how the documentation and outputs could be improved with respect to clarity.
EPA revised the BMA software package and documentation in response to these comments, incorporating most of the suggested revisions. The BMDS User Guide for BMDS Version 3 or higher contains details necessary to understand the program's assumptions and application.
The primary purpose of this review was to obtain external expert opinions on EPA’s proposed model averaging approach for the derivation of BMD and BMDL values from dichotomous dose-response datasets.
Reviewers were asked to read the review documents describing EPA's proposed BMA method and respond to several specific charge questions relating to the proposed method that are documented in the peer review report. They generally agreed that the BMA software package produced statistically valid results. They praised the innovative approaches with comments such as “substantive models going into model averaging are very well thought out,” “highly valuable,” and “…heavily encourage the Agency to adopt the kinds of methods proposed here into its next version of BMDS.”
Several recommendations were made regarding enhancements that could facilitate or expand its practical application, and how the documentation and outputs could be improved with respect to clarity.
EPA revised the BMA software package and documentation in response to these comments, incorporating most of the suggested revisions. The BMDS User Guide for BMDS Version 3 or higher contains details necessary to understand the program's assumptions and application.
Impact/Purpose
The primary purpose of this review was to obtain external expert opinions on EPA’s proposed model averaging approach for the derivation of BMD and BMDL values from dichotomous dose-response datasets.
Citation
U.S. EPA. Bayesian Models and Model Averaging Software (External Review Draft, 2018).
History/Chronology
| Date | Description |
|---|---|
| Mar 2018 | EPA hosted an external review to review and comment on the proposed Bayesian modeling methods. |
| Jun 2018 | An independent contractor submitted the comments from the review of the proposed Bayesian modeling methods. |
| Feb 2019 | EPA released the comments report on the draft documents. |