Abstract 2791: Just how noisy is xenograft data? Using growth rate modeling and bootstrapping to optimize xenograft study design.

The evaluation of tumor growth inhibition in xenograft models, generated by subcutaneous implantation of human cancer cell lines, is an integral component of the drug discovery and development process. While xenograft data is ubiquitous, and opinions on its utility are just as common, systematic examinations of signal-to-noise ratio, overall precision of xenograft data and the best metric to report tumor growth inhibition are, remarkably, absent from the literature. To this end, we present a thorough retrospective analysis of xenograft studies from a large in-house database of 225 human xenograft efficacy studies performed from 2006 to present, using a wide range of anti-neoplastic drugs. Individual xenograft tumor growth curves were fit to an exponential model, and a novel measure for data analysis was developed (the model-fitted T/C ratio on Day 21). This novel measure possesses several advantages over a traditional (raw) T/C ratio, notably the use of all available data (not just Day 21), the absence of bias due to informative right-censoring (nonrandom removal of mice in the control group whose tumor volume reached a pre-defined humane endpoint), and the absence of bias due to the inherent slowing of growth kinetics in the control group. The model-fitted T/C ratio on Day 21 was then compared to the raw T/C ratio. For each pair of comparisons between treatment group and control, bootstrapping was used to determine the mean (μ) and standard deviation (σ) of the T/C ratio (either model-fitted or raw). A Z-score measure, (1-μ)/σ, was computed as a measure of the signal-to-noise ratio of the xenograft studies, for 1167 comparisons. For xenograft studies conducted with 10 mice per group over 21 days, the model-fitted T/C ratio outperformed the raw T/C ratio in terms of median Z-score (7.1 and 5.4 respectively). When the number of mice in each group was reduced to 6, and the study length was decreased from 21 to 14 days, a high Z-score (5.1) was still achieved for the model-fitted T/C ratio. The power of xenograft study comparisons for T/C in the range of 0.35 to 0.45 was examined and found to be 0.95 for studies with 6 animals per group using the model-fitted T/C measure. Finally, the misclassification frequency (fraction of comparisons where the treatment was misclassified using a binary cutoff of efficacious vs. non-efficacious) was calculated, and found to be 0.04 using the model-fitted T/C. Our calculations demonstrate that switching to a model-fitted T/C makes more efficient use of the data, yielding considerable cost savings while maintaining a high power and low misclassification rate. The exact degree of benefit from this novel measure and alternative design may vary for other animal facilities, since the noise levels could vary. However, the methods developed here to evaluate potential designs should still be widely applicable, and represent a general approach for the optimization of xenograft studies. Citation Format: Jill Donelan, Greg Hather, Ray Liu, Syamala Bandi, Wen Chyi Shyu, Mark Manfredi, Arijit Chakravarty. Just how noisy is xenograft data? Using growth rate modeling and bootstrapping to optimize xenograft study design. [abstract]. In: Proceedings of the 104th Annual Meeting of the American Association for Cancer Research; 2013 Apr 6-10; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2013;73(8 Suppl):Abstract nr 2791. doi:10.1158/1538-7445.AM2013-2791