Abstract LB-321: Re-evaluating sample sizes in preclinical testing of patient-derived xenografts

Abstract

Abstract The NCI Pediatric Preclinical Testing Consortium (PPTC) has established panels of patient-derived xenografts (PDXs) and cell lines of pediatric cancers for preclinical testing. We performed a systematic analysis of in vivo testing results from 2015-2018 to re-evaluate the number of mice needed for testing. Data were compiled from experiments performed on 30 anti-cancer agents; each agent was tested against 1 to 31 PDXs (median, 9) from a consortium-wide panel of 112 PDXs. We evaluated time-to-event across control mice, with “event” defined as quadrupling of tumor volume in solid tumors, mice becoming moribund or severely neurologically deficient in brain tumors, and hCD45+ > 25% or other leukemic event for leukemia. Observed median time-to-event and coefficient of variation (CV, standard deviation/mean) varied between PDXs. Overall, the median CV was 0.257 with an interquartile range of 0.172 to 0.363. We estimated the power to detect differential time-to-event across a range of parameters observed historically. We modeled time-to-event under two scenarios: a right-skewed, heavy-tailed log-normal distribution, and a more symmetric and light-tailed gamma distribution. We also varied CV and treatment effect size (T/C, ratio of median time-to-event between treated and controls). The PPTC defines a median T/C of 2 or greater as growth delay; progressive disease with growth delay is the lowest objective response measure beyond that of controls. Differential time-to-event was evaluated using the Gehan-Wilcoxon test. The PPTC has generally used sample sizes of 8 or 10 mice per group. Under both the log-normal and gamma models, a sample size of 5 mice per group yielded ≥87% power to detect a T/C of 2 assuming a CV of 0.3; over 60% of tested PDXs have a CV lower than this. Table 1 summarizes our results across a subset of parameters used in our calculations. Our results suggest that for most PPTC PDXs, power for detecting major growth delay effects can be maintained while employing fewer mice than used previously. Table 1Summary of Power Analysis for α = 0.05CoefficientTime-to-eventPower (α = 0.05)Distributionof VariationMedian T/Cn = 3n = 5n = 8n = 10Log-normal0.202.095.8%99.6%100.0%100.0%3.0100.0100.0100.0100.00.302.075.390.598.799.93.097.299.7100.0100.00.402.056.868.790.595.73.085.896.0100.0100.00.502.044.051.975.984.53.073.187.197.799.8Gamma0.202.094.499.1100.0100.03.0100.0100.0100.0100.00.302.072.587.098.0100.03.095.099.5100.0100.00.402.051.164.985.692.93.082.894.199.699.90.502.038.246.369.380.23.066.079.496.698.9 Citation Format: Eric Earley, Richard Gorlick, Peter J. Houghton, John M. Maris, Xiao-Nan Li, Richard B. Lock, Beverly Teicher, Malcolm A. Smith, Stephen W. Erickson. Re-evaluating sample sizes in preclinical testing of patient-derived xenografts [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr LB-321.