1252 Incorporating virtual control groups in preclinical mouse studies: a historical data approach

<h3>Background</h3> The surge in anti-tumor drug development in recent years has led to an increased demand for preclinical animal testing. Vehicle controls in mouse studies have been well documented, with tumor growth data measured in tightly controlled laboratory settings. Accumulated historical control data, namely VCG (Virtual Control Groups), can potentially be used to reduce the number of mice required in current control groups and maintain the statistical power to detect a drug treatment efficacy. <h3>Methods</h3> Historical tumor growth data from syngeneic mouse models were curated from the Crown Bioscience databank. The data were categorized based on key factors affecting tumor growth, including mouse model type, treatment, and dosing method. To standardize historical growth data measured at different time points across studies, we introduced tumor growth rate by fitting a growth curve over the data. Several approaches were implemented to incorporate historical data into current studies, such as random sampling from matched tumors to serve as a VCG and meta-analysis-based methods. <h3>Results</h3> Compared to tumor growth inhibition (TGI), which is based on tumor volumes measured on a specific day, tumor growth rate, which integrates data from all time points, exhibited smaller variance and was thus used for downstream statistical analysis of drug efficacy. To understand how tumor growth rate relates to drug efficacy, the difference in tumor growth rates between the drug-treatment group and the vehicle group was also converted to TGI. By assuming a normal distribution of historical tumor growth rates, it can be inferred whether the current growth rate is a potential outlier due to hidden confounders. If the current growth rate is not an outlier, historical data will be considered for incorporation. In several mouse case studies a power analysis, where a simulation of a series of control groups under the framework of random sampling and meta-analysis-based methods was conducted, and an increase in the statistical power to detect a significant treatment effect by incorporating historical data was observed. However, the power gain is not always guaranteed. It depends on factors such as the consistency between the historical control group and the current control group. <h3>Conclusions</h3> Historical control data can be used to reduce animal usage in current control groups. Due to the heterogeneity of historical data and potential confounders in the current study, historical data should be handled with care. This study provides a general strategy of incorporation of historical tumor growth data into mouse studies.