The Study Of Sample Size Adjustment Of Adaptive Design And Its Realization On Web

The traditional randomized clinical trial(RCT)is often designed with parallel control in which the treatments are assigned to the patients equiprobably in different center. It has attracted many statisticians'interest for a long time because of its simplicity and efficiency. But it does not satisfy us for high drug development costs, long period to be pushed on market and escalating patient safety concerns. On one hand, we hope to shorten the development period as soon as possible. On the other, we hope to modify unreasonable aspects of the study coming from its initial design. One way to address these challenges that is receiving significant attention from pharmaceutical companies and regulatory agencies throughout the world is adaptive trial design.Adaptive design allows modifications to some unreasonable parameters of the trial by using accumulated data so far after its initiation without undermining the validity and integrity of the trial so as to find out the incorrect assumption in the planning stage; to estimate the parameters objectively in the next stage and correct the bias from the initial design to the most degree. The modifications to a trial of adaptive design include a host of aspects, such as sample size re-estimation, number of the stages and randomized assignment etc.From the 1990s till now the adaptive design has attracted more attention by the statistician and the clinical trial workers. The exploration to the principle of the design has never paused. However, there have not been a set of mature theories and believable operation regulations established, which results in the discussion of the design theory having become the hot topic of the clinical trial in full swing, but the practical application in clinical trials are very rare. In this study, we explored and discussed some theoretical and practical issues of two-stage adaptive design by using computer technology and Monte Carlo simulations. The main contents of our study include sample size adjustment (SSA), controlling the inflation of typeâ… error and maintaining the power etc. At the same time, we design and develop the adaptive design system based on Web, providing the adjustment process of adaptive design to users with active Web page and offering the interactive interface. The main works and results of the study are to be introduced as follows.1. There exist two operating ways in SSA by using IPS in adaptive design, which can be made in blind or unblind IPS respectively. But there is no verdict for which one is better. In order to study this issue, we make comparison for IPS sample size effecting on typeâ… error and power after SSA in blind and unblind IPS through Monte Carlo simulations. The results show us that the IPS sample size have great affection to typeâ… error and the power in both blind and unblind IPS if we analyze the final data by traditional statistical method. The results of two kinds of SSA have the following common charateristecs: When n1 is small, the inflation of typeâ… e rror increases; but the inflation of typeâ… error decreases and the power gradually increases with the growing of n1 . Typeâ… error stops inflating and the power reaches or exceeds the desired level when n1 is approaching 50 percent of the planning sample size. The typeâ… error is always below the nominal significance level and the power is higher than the desired level When n1 is over 50 percent of the planning sample size. Both results from different SSA methods show us the superiority of SSA and also remind us of paying more attention to the typeâ… error than power. Through the comparison of results of SSA in two conditions, we can see that the blinded SSA is easier to be implemented and more feasible in clinical trials, so we are inclined to recommend it.2. When IPS sample size is small, the inflation of typeâ… error is still obvious even if we make SSA with IPS blinded. Isn't the SSA with IPS suitable to clinical trials with small sample size? The arguments have never been stopped for this issue. The computer simulations tell us that the randomization test for the final data can not only prevent typeâ… error from inflating, but also maintain the desired power if we re-estimate the sample size under blinded IPS. We can say that SSA can be applicable to clinical trials with either big or small sample size if only we choose appropriate statistical method.3. When we make sample size recalculation by using the sample variance and the observed treatment difference, the IPS has to be unblinded. For the analysis of the data in this situation, there are two different methods currently, that is, traditional hypothesis test for final data and combining P values from different stages. There is no agreement for which method is preferred and how the IPS sample size affects on the results respectively when we use two different methods. The exploration to this issue presents that the IPS sample size should not be less than 1/3 of the original sample size, otherwise, the balance between controlling typeâ… error and maintaining power is hard to keep. From the comparison of the two methods being used to analyze the same clinical trial with two treatment groups, we can see that the traditional t method results in more inflating of typeâ… e rror and the relatively appropriate power; but combining P value method controls the inflation of typeâ… error with compromising the power. When the IPS sample size is more than 1/3 and less than 1/2 of the planned sample size, combining P value method should be preferred for it can control the typeâ… error with a little bit power lost. When the IPS sample size is bigger that 1/2 planning sample size, t test should be chosen to analyze the final data. The power is high enough and typeâ… error is not inflated in this condition.4. Considering covariances in clinical trials, should we use the sample variance and the observed treatment difference to recalculate the sample size? How the covariances affect on the SSA? Can it be ignored? The solutions for these issues have not been reported till now. We have made a lot of simulations for these issues and found that typeâ… error is obviously more inflated and the power is lowering if using variance analysis instead of using covariance analysis,with ignoring the covariance. It warns us that the effection of covariance should not be ignored, otherwise, typeâ… error will be hard to control and the power can not reach desired level.5. EDC is the hot topic in clinical trials and has attracted more attention currently. In order to push the combination of adaptive design with EDC, we designed and developed the adaptive design based on Web on the basis of above simulating exploration. It can provide reference protocols for the practical application of adaptive design in clinical trials. It can also help the users to determine the parameters of adaptive design and analyze the data input through Web pages.6. Based on the above Web system, we make SSA simulation for non-inferiority clinical trial with IPS blinded and comapare the results with the output of SAS program. The results from the Web system are just the same as that of SAS program.The achievements of this study consist in the following four points: Firstly, we make comparison for IPS sample size affection on typeâ… error and power after SSA in blind and unblind IPS through Monte Carlo simulations. Secondly, through simulations, we concluded that randomization test can both avoid the inflation typeâ… error and maintain the desired power when we make blinded SSA for clinical trials with small sample size. Thirdly, through simulations, we draw the conclusion that it is feasible to make SSA by the sample variance and the observed treatment difference when the covariances exist in the clinical trials. Fourthly, we designed and developed the adaptive design system based on Web, realizing the simulation of SSA online.This study mainly discussed several methods and its issues associated with adaptive design on the basis of SSA. Adaptive design based on Web is also one part of our study. The results of this study can provide reference for further researches of the program of National Natural Science Foundation of China (Adaptive design in clinical trial and its application, No: 30671823).