In crossover clinical trial studies, patients must receive two types of treatment in two periods, thereupon the possibility of not compliance of protocol in this design is more than parallel designs. Possibly, may occur non-compliance in one or both the periods. Also, the sample size in this type of design is usually smaller than the parallel design, and the elimination of patients who did not follow does a lot of bias in estimating the treatment effects. As a result, adjusting the effect of non-compliance in crossover designs is more important. In this study, a two-stage latent treat non-compliance model (LTGM) has been used to adjust the effect of non-compliance in a crossover study. In this model, first for each patient, the rate of compliance to the protocol as a latent variable is estimated using the latent treat analysis model, then placed in the grizzle model and the effect of non-compliance is adjusted. We compared the accuracy of three models: ordinary grizzle model, generalized grizzle model and LTGM model. Different scenarios were simulated to compare these techniques. The simulation study showed that the LTGM model has the lowest bias in all cases.

non-compliance, randomized clinical trials, simulation, grizzle model, latent treat analysis.