A framework for the design, conduct and interpretation of randomised controlled trials in the presence of treatment changes.

BACKGROUND

When a randomised trial is subject to deviations from randomised treatment, analysis according to intention-to-treat does not estimate two important quantities: relative treatment efficacy and effectiveness in a setting different from that in the trial. Even in trials of a predominantly pragmatic nature, there may be numerous reasons to consider the extent, and impact on analysis, of such deviations from protocol. Simple methods such as per-protocol or as-treated analyses, which exclude or censor patients on the basis of their adherence, usually introduce selection and confounding biases. However, there exist appropriate causal estimation methods which seek to overcome these inherent biases, but these methods remain relatively unfamiliar and are rarely implemented in trials.

METHODS

This paper demonstrates when it may be of interest to look beyond intention-to-treat analysis for answers to alternative causal research questions through illustrative case studies. We seek to guide trialists on how to handle treatment changes in the design, conduct and planning the analysis of a trial; these changes may be planned or unplanned, and may or may not be permitted in the protocol. We highlight issues that must be considered at the trial planning stage relating to: the definition of nonadherence and the causal research question of interest, trial design, data collection, monitoring, statistical analysis and sample size.

RESULTS AND CONCLUSIONS

During trial planning, trialists should define their causal research questions of interest, anticipate the likely extent of treatment changes and use these to inform trial design, including the extent of data collection and data monitoring. A series of concise recommendations is presented to guide trialists when considering undertaking causal analyses.