Virtual Control Groups, Actually: What SEND Makes Possible (and What It Doesn’t)

The CDISC Standard for Exchange of Nonclinical Data (SEND) has enabled the generation and exchange of structured, interoperable toxicology study datasets, providing a foundation for the development of innovative new applications of data science techniques in regulatory toxicology. One emerging application is the use of virtual control groups (VCGs), which leverage historical control data to reduce or replace the use of concurrent control animals in toxicology studies. Projects such as the IHI VICT3R initiative are using SEND-formatted toxicology study dataset repositories to develop robust matching criteria, e.g., study design, dosing regimen, and animal characteristics, to ensure appropriate VCG selection.

However, implementation of VCGs introduces significant challenges, including:

• the need for rigorous terminology reconciliation to integrate data across studies
• the risks of spurious selection of outlier animals
• variability in histopathology diagnostic thresholds – an issue that may require pathologists to re-read slides to ensure accurate interpretation

Fortunately, solutions are available, ranging from tools that automate terminology reconciliation to Bayesian methods that borrow information from historical control data to improve statistical power without direct one-to-one substitution.

By addressing these challenges, SEND-driven innovations can maximize data utility, reduce redundancy, and advance the 3Rs (Replacement, Reduction, Refinement). This presentation will highlight methodologies, obstacles, and future directions for integrating SEND-enabled data science into pragmantic regulatory toxicology strategies.