Analysis of Randomized Clinical Trial in the Presence of Non-Compliance: Comparison of Causal Models

Year: 2021       Vol.: 70       No.: 1      

Authors: Ali Reza Soltanian, Hassan Ahmadinia, and Ghodratollah Roshanaei

Abstract:

Non-compliance is a common deviation from randomized clinical trials protocol. Standard approaches for comparing the effects of drugs in randomized clinical trials in the presence of non-compliance are intention-to-treat, as-treated and per-protocol analysis. Each of these approaches has disadvantages when evaluating the effect of medication in present of non-compliance. The current study compared the accuracy of instrumental variable (IV), intention-to-treat, as-treated and perprotocol technique. We assumed that non-compliance occurred for some patients in the new treatment group only, and independent of the patient outcomes. To compare these techniques, various scenarios were simulated. The MSE value for both PP and IV models changes only under the influence of the value of w (non-compliance ratio). That is, at all values of θ (treatment effect), the MSE of these two models increases with increasing non-compliance ratio, and changing the value of θ does not affect the MSE. The MSE value for the AT model if the non-compliance occurs only in the intervention group, this value changes only under the influence of the w value That is, in this case, in all values of θ, the MSE value increases with increasing non-compliance ratio, and changing the value of θ does not affect the MSE. But the MSE value of the ITT model is strongly influenced by the value of θ. At low θ values the MSE value of this model is lower than other methods and better estimates the therapeutic effect, and in this case with increasing the w, the MSE value increases very little. But as the θ increases, so does the MSE value, and in this case, as the w increases, the MSE value increases sharply.

Keywords: causal model, non-compliance, randomized clinical trials, simulation

Download this article: