T-12: Design and Analysis of Biosimilarity Based on Interval Estimations
Poster Presenter
Chin-Fu Hsiao
Investigator
National Health Research Institutes Chinese Taipei
Objectives
The variability of biological products must be different between biological product and its follow-up drug. This leads to the so-called Behrens-Fisher problem — there does not exist an exact test or interval estimation.
Method
Approaches such as Satterthwaite’s, Cochran-Cox’s, and Howe’s approximations have been used to solve the problem.
Results
As suggested by the US Food and Drug Administration (FDA), the first assessment in a 3-tier approach of biosimilarity, or Tier 1, is to perform an equivalence test between an innovative biological product and its follow-up drug. Due to the nature of biological products, the variability must be different between two drugs. However, this leads to the so-called Behrens-Fisher problem — there does not exist an exact test or interval estimation. Approaches such as Satterthwaite’s, Cochran-Cox’s, and Howe’s approximations have been used to solve the problem. In this paper, we utilize two-sided confidence intervals for mean differences based on these approximations to assess the Tier 1 assessment of biosimilarity. On the other hand, literature for sample size determination of the assessment of biosimilarity is lacking.
Conclusion
Here, we show that the confidence lower and upper bounds follow a bivariate normal distribution asymptotically. Thus, the power function is built, and the sample size determination can be derived. Moreover, based on the power function, the relation between sample size and margin of the equivalence can be investigated in this study. A real example is represented for the illustration of our proposed approach.
