Menu Back to Poster-Presentations-Details

PP02-22: A Bayesian Approach to Design and Evaluation of Multiregional Clinical Trial

Poster Presenter

Lien-Cheng Chang

Section Chief, Department of Intellectual Property and Technology Transfer
Academia Sinica
Taiwan

Objectives

The objective of this study is to develop a Bayesian statistical approach for designs and evaluations of multi-regional clinical trials .

Method

In this study, we will use a Bayesian approach to design and evaluation of MRCTs under normality. We will assume that there is a difference in treatment effect due to regional difference. The posterior probability of the overall treatment effect is then calculated to evaluate efficacy of a treatment

Results

First of all, we derive the sample size required for the multi-regional clinical trial for different priors. Secondly, we present a randomized, double-blind, placebo-controlled MRCT on patients with schizophrenia to compare a new antipsychotic agent (a test drug) with dose ranging (400, 600, or 800 mg/day) and a placebo control as an example to illustrate our approach. The MRCT was conducted in 39 centers around the world, grouped three regions, South Africa, Europe, and Asia. The primary efficacy variable was the change in the total score of the Positive and Negative Syndrome Scale (PANSS) from baseline to week 6. For the purposes of this illustration, this investigation focuses on the efficacy of a 400 mg/day dose, and found that the overall placebo-adjusted change from baseline in the total score of the PANSS was -6.0, with a standard error of 2.80. For regions 1, 2, and 3, the treatment effects were -5.0, -5.0, and -7.6 with equal standard error 20.77, respectively, and the sample size per group was 22, 44, and 44, respectively. Based on original data, we try to apply different combinations of parameters of an inverse gamma distribution and a half-normal distribution to evaluate the marginal posterior probability of overall treatment effect. For an inverse gamma distribution with parameters a and b, if b increases with fixed a or a increases with fixed b, then the posterior probabilities decrease or increase, respectively. Additionally inverse gamma distribution is robust with increasing a with smaller fixed b or increasing b with greater fixed a. For a half-normal distribution with parameter , if increases, then the probabilities deceases. Furthermore, a half-normal distribution is robust in a smaller value of . If a probability value = 0.975 concludes a positive overall treatment effect, the appropriate parameters of the inverse gamma distribution and half-normal distribution are (2.5, 10) and 10, respectively.

Conclusion

We have established a Bayesian approach for the designs and evaluations of MRCTs. In the Bayesian approach, the overall treatment has a prior with a uniform distribution, and the between-region variability has a prior with a inverse gamma distribution or a half-normal distribution as priors. We calculate the sample size required for the MRCT with use different parameters of priors and different combinations of regional proportions and parameters of priors. We also illustrate our approach using an example.