Abstract
BioCelerate is an industry consortium to increase efficiencies in early-stage R&D. Implementation of the SEND model represents an opportunity to apply large-scale data analytics to toxicology data. To realize this opportunity, differences in SEND implementation that make it difficult to conduct cross-study analysis must be addressed. In partnership with FDA, cross study analysis in SEND datasets that are significant drivers of variability that negatively impact cross-study analysis were identified. Herein the focus is on background-control data and approaches for improved data harmonization. The most relevant variables (eg MI, BW, LB, OM) of SEND datasets have been defined to function as robust and easy-to-use historical control databases. The FDA CDER repository of >1,800 SEND datasets was queried by FDA staff to gain insight into how SEND is being applied with respect to these parameters. Proposals for harmonization methods and test search scripts to transform data were developed to allow consistent extraction of these variables across SEND datasets. A framework for the development of proposed solutions from which a user could assemble the results for specific endpoints for all control animals in a user-defined subset of studies is provided. A BioCelerate/PHUSE collaboration is also engaged in further refinement of SEND database searching through the development of R search scripts. This R search script repository is publicly available on GitHub and can be used to provide greater context into the significance of toxicologic findings observed in toxicology studies.
