Towards optimization of chemical testing under REACH: A Bayesian network approach to Integrated Testing Strategies

Integrated Testing Strategies (ITSs) are considered tools for guiding resource efficient decision-making on chemical hazard and risk management. Originating in the mid-nineties from research initiatives on minimizing animal use in toxicity testing, ITS development still lacks a methodologically consistent framework for incorporating all relevant information, for updating and reducing uncertainty across testing stages, and for handling conditionally dependent evidence. This paper presents a conceptual and methodological proposal for improving ITS development. We discuss methodological shortcomings of current ITS approaches, and we identify conceptual requirements for ITS development and optimization. First, ITS development should be based on probabilistic methods in order to quantify and update various uncertainties across testing stages. Second, reasoning should reflect a set of logic rules for consistently combining probabilities of related events. Third, inference should be hypothesis-driven and should reflect causal relationships in order to coherently guide decision-making across testing stages. To meet these requirements, we propose an information-theoretic approach to ITS development, the “ITS inference framework”, which can be made operational by using Bayesian networks. As an illustration, we examine a simple two-test battery for assessing rodent carcinogenicity. Finally, we demonstrate how running the Bayesian network reveals a quantitative measure of Weight-of-Evidenc