Group Sequential Test Based On Efficacy Stopping Probability

Objective:In the group sequential trial,the traditional group sequential test needs to correct the critical value of the current stage according to the amount of information obtained from the accumulated samples at the current stage.However,all current critical value correction methods cannot keep the efficacy stopping probability obtained in the design process unchanged,which leads to failure to meet the design requirements of researchers.Therefore,this paper proposed a new group sequential test algorithm based on efficacy stopping probability,which could maintain the designed efficacy stopping probability in the interim analysis.Methods:Based on the unilateral rejection type group sequential trial,the group sequential test based on effective stopping probability was constructed by numerical integration.Under the condition of different critical value calculation methods and different sample sizes,this paper compared the type Ⅰ error and power of traditional group sequential test and group sequential test based on efficacy stopping probability through statistical simulation,and evaluated the advantages and disadvantages of the two test methods.Results:When the sample size was small,the traditional group sequential test would have the problem of type Ⅰ error expansion.However the type Ⅰ error expansion of the group sequential test based on the efficacy stop probability was smaller than that of the traditional group sequential test.For example,when the design method was OBF method and the sample sizes in each stage of the test were 37,the type Ⅰ error of the traditional group sequential test was 0.056 in three stages trial,while the type Ⅰ error of group sequential test based on efficacy stopping probability was 0.053.The type Ⅰ error of latter method was closer to the 0.05.Both test methods could maintain the power at the set level.In the case of large sample size,the two test methods both could well control type Ⅰ error and power.Based on the same data,the conclusions of the two test methods were not completely consistent.In a group sequential trial,the change of the effective stop probability would lead to the change of the number of termination stages of the test,or even the change of the test conclusion.Therefore,it was necessary to maintain the efficacy stopping probability in the interim analysis.The larger the sample size,the higher the kappa coefficient and consistency rate of the two test methods.Except for the case that the efficacy stopping probability of a certain stage obtained from the design stage was small(for example,the efficacy stopping probability of the first stage of the three-stage O’Brien Fleming design is 0.106),the distribution of the updated efficacy stopping probability generated by the traditional group sequential test was almost normal and fluctuates up and down at the design level.Due to the need to recalculate the power according to the actual maximum information level,the efficacy stopping probability of the new algorithm in the last stage would also recalculate,but the change range was smaller than the traditional group sequential test.Conclusions:This research conducted an in-depth study on the important indicator of efficacy stopping probability in group sequential trials.The group sequential test based on the efficacy stopping probability was proposed for the first time to maintain the efficacy stopping probability at the design level in the interim analysis,so as to maintain the design characteristics.The simulation research proved that the algorithm had good statistical performance.This paper provided a new test process for group sequential test,and promoted the application of group sequential trial in clinical trials.