Noncompliance in randomized controlled trials

Catherine Hewitt and colleagues1 have given an excellent brief account of the effect of noncompliance on the analysis of randomized controlled trials. We would like to add a few more points to make the discussion more complete.

Noncompliance can seriously decrease study power,2 resulting in widely varying estimates of the sample size required for a study.3 Noncompliance is thus a significant issue to be considered when planning trials involving long-term therapies. Moreover, the analysis of results for patients receiving treatment can be biased in situations in which participants decline treatment because they cannot afford to pay for their drugs, if they are not provided free of charge.4 Incorporation of patient preferences into the randomization process will also bias the results.5 With these practical issues in mind, it would be an interesting exercise to compare the results of well-designed observational studies with those of randomized controlled trials.