Alternative statistical strategies for biosimilar drug development

Abstract

Introduction: Many regulatory authorities have published requirements for the approval of biosimilar medicinal products. However, there is no guidance on which quantitative standards should be used to defi ne how similar a follow-on product must be to be considered biosimilar. Sample sizes for clinical biosimilar trials using traditional designs often exceed 500 patients. Several publications have referenced alternative methods to calculate biosimilar trial sample sizes. Few, however, provide actual case studies showing the order of magnitude change possible. We modelled alternate statistical approaches to practical case studies to test whether it is possible to reduce the sample size of clinical biosimilar trials. Methods: Clinical case studies of bevacizumab, adalimumab and rituximab biosimilars were used as models. A traditional frequentist model was compared to a repeated measures analysis, a batch-to-batch method, and a Bayesian method. Statistical modelling and sample size calculations were performed using PASS version 12 and SAS version 9.2. Repeated measures analysis was performed using a generalized estimating equation method. Sample size and power were estimated by simulations. Sample size estimation for the Bayesian method was based on beta-binomial distribution of the posterior distribution of response rate. Results: In all cases, the repeated measures analysis and Bayesian design resulted in reduced sample sizes compared to a traditional approach (3-11% and 26-29% reduction, respectively). The batch-to-batch method resulted in larger sample size estimates compared to the traditional approach.