Estimation Of Treatment Effect And Related Statistical Inference On Randomized Controlled Clinical Trials With Treatment Switching

Parallel group clinical trials are widely used to compare a new treatment with an existing one to determine its treatment effect and safety.In the course of the trial,if some patients are insensitive to the original treatment and/or suffering from serious side effects,usual y they are allowed to switch to another treatment group based on ethical considerations,which is called treatment switching.Treatment switching puts a big chalenge to the existing statistical theory and methods,and also brings difficulties to the data analysis of clinical trials.This is because the primary goal of a clinical trial is to make statistical inferences on the efficacy of the test treatment on the basis that all patients complete the trial according to a pre-designed protocol.Even if some patients switch to other treatment groups during the trial,the most important issue that the trial managers want to solve through the designed clinical trials is still related to: what is the treatment effect of the testing therapy compared with the control in the absence of treatment switching.A variety of models and methods have been proposed to conduct statistical inference on survival data for clinical trials with treatment switching.However,the current research methods do not take switchers’ carry-over effect into account.From the perspective of drug metabolism,when a patient changes his or her treatment group,the therapeutic effect of the former treatment will not disappear immediately,but die down gradually;similarly,the latter treatment should also work slowly.Generally speaking,the carry-over effect is not of great interests to the managers(the main goal of conducting a clinical trials is to estimate the main effect of the trial therapy),but ignoring it will lead to bias to the estimation of the main effect.Taking switchers’ carry-over effect into consideration and describing it by statistical models,we conduct valid statistical inference on survival data for parallel group clinical trials with treatment switching in this paper.The main research issues include:(1)The traditional survival time model is expanded by including switchers’ carryover effect into the model.A full parameter model is employed to describe patient’s survival time distribution,and the maximum likelihood estimation of model parameters(including the parameter which measures the efficacy of the testing therapy)are obtained by theoretical derivation.(2)Some related statistical inference are also conducted based on the proposed new model,including the point and interval estimation of two groups’ median survival time,the hypothesis test on two medians,and the determination of required sample size for each hypothesis test,etc.(3)In order to evaluate the performance of the proposed new method,several numerical studies are conducted to compared the new method with several existing ones,including the rank preserving structural failure time model(RPSFTM)which ignores the carry-over effect and the intention-to-treat analysis(ITT)which ignores treatment switching.The results show that the existing methods yield biased estimation on the effect of the testing therapy since they ignores the carry-over effect of switchers.The higher the switching proportion and/or the longer of carry-over phase,the greater the bias.On the contrary,by describing switchers’ carry-over effect through statistical models,our proposed new method generates accurate estimation on the treatment effect in most of the cases.Therefore,the new model proposed in this paper provides a powerful tool for managers in dealing with survival data from clinical trials with treatment switching.