Best practices for exposure model peer review - A SciPinion advisory panel report

Abstract

The extent and rigor of peer review that a model undergoes during and after development influences the confidence of users and managers in model predictions. A process for determining the breadth and depth of peer review of exposure models was developed with input from a panel of exposure-modeling experts. This included consideration of the tiers and types of models (e.g., screening, deterministic, probabilistic, etc.). The experts recommended specific criteria be considered when evaluating the degree to which a model has been peer reviewed, including quality of documentation and the model peer review process (e.g., internal review with a regulatory agency by subject matter experts, expert review reports, formal Scientific Advisory Panels, and journal peer review). In addition, because the determination of the confidence level for an exposure model's predictions is related to the degree of evaluation the model has undergone, irrespective of peer review, the experts recommended the approach include judging the degree of model rigor using a set of specific criteria: (1) nature and quality of input data, (2) model verification, (3) model corroboration, and (4) model evaluation. Other key areas considered by the experts included recommendations for addressing model uncertainty and sensitivity, defining the model domain of applicability, and flags for when a model is used outside its domain of applicability. The findings of this expert engagement will help developers as well as users of exposure models have greater confidence in their application and yield greater transparency in the evaluation and peer review of exposure models.