Statistical applications of virtual control groups to nonrodent animal toxicity studies: An initial evaluation

Utilization of data from historical control animals to form virtual control groups (VCGs) is an innovative approach to embody the 3Rs (reduce, refine, and replace use of control animals) principle in research. However, there is no available systematic comparison of statistical performance between concurrent control groups (CCGs) and VCGs in nonrodent safety assessment. The optimal selection criteria and combination of VCGs and CCGs also remain unclear. This study retrospectively evaluated VCGs’ statistical performance to detect test article effects on body weight and clinical pathology endpoints in dog and nonhuman primate (NHP) systemic toxicity studies. Body weight and six clinical pathology endpoints were analyzed against the reported study findings from a cohort of 22 previously reported nonrodent 1-month oral gavage toxicity using three different methods of generating VCGs. When the fold change from baseline was used, VCGs yielded a similar or higher statistical sensitivity to detect test article relatedness than CCGs. Compared to simple random sampling or using fixed criteria, the propensity score matching by BW, age, and year of study initiation yielded higher sensitivities. Our analysis supports the hypothesis that VCGs can be a viable instrument in nonrodent toxicity studies. •for VCGs yielded a similar or higher statistical sensitivity to detect test article relatedness than CCGs did.•Degree of improvement through VCG in sensitivity varied from endpoint to endpoint.•Propensity score matching for VCG generation yielded higher sensitivity.•Fold change from baseline overall yielded higher sensitivity than postdose measurements for majority of the endpoints.