The hope and the hazards of using compliance data in randomized controlled trials.

This paper aims to elucidate both the advantages and limitations of using compliance data in the reporting of treatment differences in clinical trials, illustrating the issues with some recent examples. While analysis by intention-to-treat should remain the principal reporting approach for most major clinical trials, arguments are put forward as to why supplementary analyses taking account of compliance can be of value. However, continued recognition of the potential biases inherent in all such selective analysis is of key importance. Some of the possible analytical approaches are presented along with suggestions on interpretation. Particular emphasis is on one case study, a large European trial in obesity incorporating repeated measures of weight loss, drug plasma level and pill count data. In working on compliance data in clinical trials, the statistician's main responsibilities may be to undertake a cohesive analysis strategy not influenced by data dredging, to achieve clarity of exposition without undue complexity or oversimplification, and to provide appropriately cautious interpretations which take account of selection biases and data limitations.