A novel approach to rituximab biosimilar drug development.

Abstract

e13064 Background: Regulatory uncertainty exists about the amount of data needed to prove biosimilarity and to extrapolate across indications. Extrapolation is especially challenging for a reference compound licensed as both an anticancer and an immunomodulating agent and where the mechanism of action may play a different role in each. However, if a biosimilar is shown to be similar to the reference compound in one indication, it is potentially more likely to be similar in another indication. Using a Bayesian approach, this may be captured through hierarchical modeling (borrowing-of-strength) across indications. We model a Bayesian hierarchical modeling (BHM) approach and a frequentist approach to support extrapolation of rituximab efficacy between follicular lymphoma (FL) and rheumatoid arthritis (RA) indications. Methods: We compared sample sizes for separate frequentist equivalence tests to a BHM approach with identical type I error and power. Frequentist sample sizes were computed in PASS. For the BHM, we assumed the log-odds ratios between the reference and biosimilar, θi (thus θi = 0 is complete equivalence), are modeled as N(μ,τ) with hyperpriors on μ and τ. The BHM borrows information across indications according to empirical similarity of the results. Results: See table. The BHM approach results in a 35% reduction in total sample size. Discussion will be required with regulatory authorities regarding the acceptability of proposed equivalence margins and the degree of confidence required to support extrapolation of efficacy across all indications. Conclusions: With a 35% reduction in sample size, the BHM approach enables borrowing across oncology and autoimmune indications with equal power and confidence. Frequentist and BHM phase III assumptions and sample sizes. Assumptions Frequentist Bayesian Indication FL RA FL RA Drug regimen R+CVP R+MTX R+CVP R+MTX Power 0.9 0.9 0.9 0.9 Alpha 0.05 0.05 0.05 0.05 Equivalence margin (%) 10 15 10 15 Rituximab response rate (%) 81 51 81 51 Subjects per arm 353 254 225 170 Randomization (reference rituximab: rituximab biosimilar) 1:1 1:1 Total evaluable subjects 706 508 450 340 Total evaluable subjects overall 1,214 790 Abbreviations: FL, follicular lymphoma; R, rituximab; CVP, cyclophosphamide, vincristine, prednisone; MTX, methotrexate.