Using pre-existing control data to set expectations in preclinical studies

This work presents a conceptual approach to limit the use of live subjects for preclinical toxicological studies by leveraging pre-existing control data. Given a valid set of pre-existing control data, one can use probabilistic methods to set expectations for targeted study outcomes prior to undertaking a study. We do not address current efforts underway to generate, simulate, validate, or otherwise model or construct such data sets (e.g., mathematical models, virtual control groups). Rather, we assume the existence of an appropriately collected and curated data set of relevant metrics representative of control subjects and illustrate the use of probabilistic methods to set expectations a priori for experimental outcomes for commonly measured endpoints. We explore using the T-distribution to set expectations for small sample sizes when endpoints are continuous measures (e.g., organ weights) and the sample average of the data set is a good proxy for the true population mean. When endpoints are discrete measures (e.g., grades of specific pathologies) or the sample average of the data set does not serve as a good proxy, we employ bootstrapping to generate a distribution. We conclude that these probabilistic approaches can help investigators understand the behavior of endpoints in an untreated population and help set a priori expectations for "normalcy" when interrogating study results.