Abstract
In randomized clinical trials, it is conventional to obtain an unbiased comparison of the effect of different treatment regimens by using an intention-to-treat-analysis. In many practical situations all patients do not receive the treatment regimen exactly as prescribed or it is well-established in advance that the treatment can take different forms over time depending on the patients' dynamic conditions. In drug trials, interest lies often not only in the pragmatic effect of a given treatment policy, but also in understanding which treatment pathway patients actually follow and what the biologic action is of the drugs that are actually administered. While an explanatory analysis based on treatment actually received acknowledges the heterogeneity represented by different compliance patterns, it may create biased comparisons because of possible associations between prognosis and compliance. In both cases, when ignoring noncompliance and when allowing for it, there is a risk of misjudging the true biologic action of a drug. The current practice of conducting and reporting clinical trials tends to give most weight to the intention-to-treat analysis. While not disagreeing with this viewpoint, more efforts should be made to define additional statistical analysis strategies that take account of noncompliance. The present paper presents two analyses which provide a deeper insight into the dynamics of noncompliance and disease progress without endangering scientific integrity. By emphasizing what is possible, more statistical efforts in this realm can be encouraged.
Get full access to this article
View all access options for this article.