167 – Speed Session #2: Topics in Biopharmaceutical Research and Statistical Programming and Analysis, Part 1
Bayesian Hierarchical Bias Model for Demonstrating Biosimilarity
Joseph Wu
Boston University School of Public Health
Sandeep Menon
Pfizer
Gheorghe Doros
Boston University School of Public Health
Kerry Barker
Pfizer
Mark Chang
AMAG Pharmaceuticals
Traditional statistical methods used to test for average bioequivalence as in a generic drug development may not be the most efficient ways to apply to biosimilarity. We adopt a Bayesian approach to establish biosimilarity for a composite endpoint. Specifically, we propose a hierarchical bias model to capture the effect difference between the reference and follow-on products. Within a non-inferiority framework, we formulate a statistical test using the posterior distributions to demonstrate biosimilarity. We illustrate this proposed methodology using a recombinant polypeptide example used to treat rheumatoid arthritis and the composite endpoint of ACR20. Using simulation, we have shown that the Bayesian type 1 error is preserved even when reference product is performing worse in current trial than historical trial but not the frequentist type 1 error. Statistical power is better than the frequentist approach as sample size increases.