Abstract 1615: Evaluation of efficacy metrics for in vivo experiments with immuno-oncology therapeutic agents

INTRODUCTION In vivo efficacy studies play an important role in the discovery of novel immuno-oncology (IO) therapeutics. Using syngeneic mice or humanized mice as mouse models, in vivo efficacy studies demonstrate the first handed evidence for anti-cancer activities of a drug. A variety of efficacy metrics, including tumor growth inhibition (TGI), growth rate, tumor response rate, time to tumor progression (TTP) and area under the curve (AUC), have been used to determine in vivo efficacy in literature. We set out to evaluate the efficacy metrics for assessing the in vivo activities of the IO therapeutic agents. METHODS A linear mixed effects model based on normalized tumor volumes is used to estimate the growth rate of each treatment arm and test if the growth rates between two treatment arms are significantly different. Tests of TGI at the last time point and the time point where TGI reached the maximum are compared with the growth rate tests. Tumor response, including a complete regression (CR), partial regression (PR), or stable response (SR) are defined based on the tumor volume changes compared to the baseline. Overall response (OR) is the summation of CR, PR and SR. Fisher's exact tests are used to obtain a p-value comparing the tumor responses between two treatment arms. The TTP with a reasonable tumor size limit is calculated and tested. Time to complete regression (TTCR) is calculated to reflect the time it takes for the tumor to achieve complete regression. The AUC based on the normalized tumor volume data is applied without an assumption of log linear distribution of the tumor volume data. With multiple in vivo experiment data sets focusing on different targets, the results from all metrics are compared against each other. Power and sample size are estimated for the key metrics as well. RESULTS To differentiate between two treatment arms, TGI at the last time points is mostly consistent with growth rate results. Growth rate tests appear to be more powerful than tumor response, TTP, and TTCR, while AUC tests are consistent with growth rate tests for the data we evaluated. AUC, as a nonparametric metric, appears to be more robust when the variability among individual tumor growth curves is relatively large and the log linear growth curve assumption for the growth rate model is severely violated. Based on growth rate or AUC tests together with last time point TGI, most of the in vivo experiments we evaluated are well powered with 8~10 mice per tre