Background & Objectives: The standard methods for the comparison of two drugs in a randomized controlled clinical trial in the presence of non-compliance are intention-to-treat or per-protocol approaches. Both approaches have problems with estimation of drug effects, and researchers are not still certain to adopt which one. In this study, the bias of intention-to-treat and per-protocol approaches was calculated using Monte-Carlo simulation. We tried to choose the best approach (based on the AIC index) for comparing Risperidone plus Celecoxib and Risperidone plus Placebo.

Methods: This secondary study was conducted to compare the effect of Risperidone plus Celecoxib and Risperidone plus Placebo among 60 schizophrenic patients. To choose between the intention-to-treat and per-protocol approaches, Monte-Carlo simulation with Ackaike (AIC) and Baysian (BIC) indices was used.

Results: The results of Monte-Carlo simulation showed that when the sample size was small (n=30 or n=60) under fixed conditions of non-compliance equal to 5% and 10%, intention-to-treat had a better goodness of fit than per-protocol based on AIC and BIC. However, increasing the sample size in active and placebo groups (e.g., n=100) showed that per-protocol had a better goodness of fit than intention-to-treat.

Conclusion: When the sample size is large, the per-protocol approach may have a better goodness of fit than intention-to-treat to address the effects of non-compliance in randomized clinical trials.