Economic benefits of using adaptive predictive models of reproductive toxicity in the context of a tiered testing program

Abstract

A predictive model of reproductive toxicity, as observed in rat multigeneration reproductive (MGR) studies, was previously developed using high throughput screening (HTS) data from 36 in vitro assays mapped to 8 genes or gene-sets from Phase I of USEPA ToxCast research program, the proof-of-concept phase in which 309 toxicologically well characterized chemicals were testing in over 500 HTS assays. The model predicted the effects on male and female reproductive function with a balanced accuracy of 80%%. In a theoretical examination of the potential impact of the model, two case studies were derived representing different tiered testing scenarios to: 1) screen-out chemicals with low predicted probability of effect; and 2) screen-in chemicals with a high probability of causing adverse reproductive effects. We define ‘testing cost efficiency’ as the total cost divided by the number of positive chemicals expected in the definitive guideline toxicity study. This would approach $$2.11 M under the current practice. Under case study 1, 22%% of the chemicals were screened-out due to low predicted probability of adverse reproductive effect and a misclassification rate of 12%%, yielding a test cost efficiency of $$1.87 M. Under case study 2, 13%% of chemicals were screened-in yielding a testing cost efficiency of $$1.13 M per test-positive chemical. Applying the model would also double the total number of positives identified. It should be noted that the intention of the case studies is not to provide a definitive mechanism for screening-in or screening-out chemicals or account for the indirect costs of misclassification. The case studies demonstrate the customizability of the model as a tool in chemical testing decision-making. The predictive model of reproductive toxicity will continue to evolve as new assays become available to fill recognized biological gaps and will be combined with other predictive models, particularly models of developmental toxicity, to form an initial tier to an overarching integrated testing strategy.