Regional Consistency Assessment In Multi-Regional Clinical Trials

Objective: In recent years,multiregional clinical trial has become a widely applied and an efficient design in drug development worldwide.There are two main objectives in multiregional clinical trial,the first objective is to evaluate the global treatment effect,the second objective is to extrapolate the overall result to participant regions.However,the difference among regions make it difficult to achieve the second objective and global organizations such as FDA,EMA have not yet drafted the explicit methods for evaluating the consistency.Recent years,many experts and regional organizations have tried to propose methods or criteria for consistency evaluation.Most of these methods are correlated with superiority clinical trials and traditional frequentist methods neglect the variation of target parameters.To address these problems,we propose new methods to,(1)evaluate the consistency among regions in the condition of non-inferiority multiregional clinical trials;(2)propose a bayesian method to evaluate the consistency among regions in the condition of superiority multiregional clinical trials.Methods: First of all,four methods are proposed to evaluate consistency in non-inferiority multi-regional clinical trials.Monte-Carlo method is used to simulate non-inferiority data,and four simulation scenarios are set for continuous and binary endpoints and the performance of each method is evaluated.Secondly,in the bayesian framework,weighted Z statistic is used to evaluate consistency of results between target region and all the regions.The discounting factor is adopted to adjust the weight and a simulation method is proposed to construct the prior distribution of discounting factor.Also,the Bayesian Consistency Index is proposed to evaluate whether the results are consistent.In the proposed bayesian method,Monte-Carlo method is used to simulate data of superiority clinical trial.We set four simulation scenarios for continuous endpoint and compare the power and false rate in different scenarios,we also compare the results of Mixed prior model and the proposed bayesian method using Monte-Carlo method.Results:(1)The performance of four methods in the non-inferiority multiregional clinical trials: in method one,the consistent probability increases as the increase of significant level,sample size of target region and the treatment effect in target region.In method two and method three,the consistent probability increases as the increasing of population proportion,treatment effect in target region and the decrease of criterion.There is negligible difference between consistent probabilities of method two and method three in different scenarios.When the treatment effect in target region was smaller than 0,the consistent probability of method two is a little bigger than that of method three.When the treatment effect in China was bigger than 0,the consistent probability of method two is a little smaller than that of method three.In the results of method four,when f is small,consistent probability increases with a larger treatment effect in target region.When f is large,consistent probability increases and then decreases with a larger treatment effect in target region.And consistent probability decreases when there are more regions in MRCT.(2)Adopt weighted Z statistic in the bayesian method can highly improve the power.When the non-informative distribution is used as the prior of discounting factor,on average,50% information is borrowed from other regions no matter what values other parameters are.The bayesian method can adaptively borrow information from other regions with a fitted prior.When the difference of treatment effects between target region and non-target region is significant,the posterior mean of discounting factor approaches to 0 and the weight of Z statistic in non-target region becomes small,then the bayesian method borrows little information of other regions;when the difference is small,the posterior mean of discounting factor approaches to 1 and the weight becomes large,the bayesian method borrows most information of other regions.The bayesian method can also control false rate under 5% when there is no efficacy of test drug in target region using the fitted prior.Simulations show the results of Mixed prior model and proposed bayesian methods are the same.Conclusion: The three methods provides references at the design stage of MRCT,but there are both advantages and disadvantages in the methods.Proper method should be chosen according to the specific situation.In method one,it is recommended to choose a bigger significant level within the allowed range to achieve a higher power.In method two,the overall treatment effect is set as 0.However,in the real situation,the overall treatment effect is not known.So it is reasonable to replace 0 with the estimated overall treatment effect,which is method three.Method three has a better operability.And the simulation results show when the consistent probabilities are the same,method three needs smaller sample size compared to method two.The more rigid the criterion is,the more obvious the trend is.It is recommended to minimize the difference among sample size proportions in each region to achieve a larger power.In the proposed bayesian method,the fitted prior can adaptively borrow information from other regions according to the congruency of data between target region and non-target region.The fitted prior is applied well in the real situation.When the test drug is actually effective in all regions,the proposed method can adaptively borrow information and thus improve the power.When the test drug is not effective in all regions,the method can also control the false rate.So in the proposed method,it is recommended to use the fitting prior as the prior distribution of discounting factor.