Abstract: AIM: Two methods, rank preserving structural failure time models (RPSFT) and the method of Branson and Whitehead (BW), are introduced in the applications of clinical trials with treatment switching. Comparisons of the two methods are made. METHODS: Based on simulated datasets of clinical trials with part of subjects in control group switching to experimental group, Monte?Carlo simulations were conducted to assess the two methods in estimating effects of the experimental drug under different levels of censoring and switching rates respectively. Besides, typeⅠerror rate and power of the two methods were compared with ITT. RESULTS: RPSFT and BW both showed high accuracies. Comparing with traditional methods, they introduced smaller biases. As censoring and switching rates gradually increased,biases of the two methods both increased with relatively smaller biases for RPSFT. And the estimated effects of the experimental drug were lower than the true efficacy. The mean squared errors (MSE) of the two methods were close. When the censoring rate was greater than 40%, the MSE of RPSFT was relatively smaller. With the switching rate increased, typeⅠerror rate of the two methods rised, generally higer than 0.05.Compared with ITT, the switching rate had less impact for the power of RPSFT and BW. CONCLUSION: Statistical analysis of clinical trials with treatment switching is mainly ITT, and RPSFT and BW are supplements. When censoring and switching rates are high, RPSFT has the priority.

Key words: treatment switching, rank preserving structural failure time model, accelerated failure time model, iterative parameter estimation algorithm