The Simulation Study And Application Of RPSFTM Method In Selective Crossover Design Trials

BackgroundRandomized controlled trials are widely used to assess the efficacy of a newtreatment compared to a control treatment. The placebo controlled treatment may bedesired from the scientific viewpoint but is usually unethical in oncology trials whenpatient in the placebo arm progressed. Selective crossover design is preferred that patientswill switch from the placebo arm to any active treatment with disease progressed. OverallSurvival is commonly used as primary endpoint in oncology clinical trials. As a traditionalmethod, ITT analysis might underestimate the treatment effect on overall survival inselective crossover design. The concern of this study is how to assess the true treatmentdifferences in selective crossover trials.Methods Rank Preserving Structural Failure Time Model (RPSFTM) was developed on thebasis of accelerated failure time model. This model retains all patients in the groups towhich they were randomized. GE (Grid Estimation）and IPE（Iterative ParameterEstimation）methods are usually considered for RPSFTM. The purpose of this studyincludes the review of RPSFTM analysis strategy, the coding of the SAS macro programand the comprehensive evaluation of ITT, GE and IPE methods.A Monte Carlo simulation study was conducted to assess GE, IPE and ITT methods.Different scenarios were considered which covers various switching rate, the switchingtime points and the true treatment effects.1000trials were repeated for each scenario. Inthe simulations, the bias, type I error, power, etc. were compared to assess the threemethods.ConclusionsFirstly, parameter estimations performed by IPE and GE methods were evaluated,which is the key point of RPSFTM analysis. Bias of the two methods was small, andbecame larger with the increased switching rate or earlier switching time points, or largertreatment effect. IPE method has the smallest bias among the three methods in mostscenarios. Secondly, power and type I error of ITT, GE and IPE methods were compared.It was found that, with switching rate increasing, the power of ITT analysis decreasedrapidly, so did GE and IPE methods. However, at any specific switching rate level, GE orIPE method had higher power compared with ITT analysis. Type I error of IPE methodwas controlled better within the acceptable range0.05and remained stable. Type I error ofGE method was controlled badly, which was above0.05and lacked of stability. Thirdly,bootstrap estimation of HR confidence interval was conducted for all three methods.Comprehensive comparisons showed that ITT method performed better at low switchingrate (below30%) while IPE method doing better at high switching rate. Lastly, thesimulation procedure proved that IPE algorithm was more systemic and efficient than GE.From the above, it is concluded that, PSFTM approach to adjust for switching should be considered in selective crossover design. And IPE method should be preferred whenconsidering RPSFTM analysis because of its more stability and efficiency compared withGE method. Traditional ITT analysis might not be appropriate to deal with treatmentswitching in final analysis separately due to low power but is still valuable in interimanalysis before switching or in final analysis when switch rate is not very high. Therefore,we suggest that both ITT analysis and RPSFTM analysis should be conducted to analyzesurvival data in selective crossover trials to increase the credibility and stability of theresults.