Designing trials of complex interventions for efficacy and mechanisms evaluation

Objectives There have been recent developments in statistical methods for assessing the mechanisms through which a complex intervention acts, with particular applications in mental health trials. This literature has focused on causal mediation analysis, which differs from statistical mediation analysis by its explicit acknowledgement and exposition of the necessary assumptions, particularly that of no unmeasured confounding between the putative mediators or mechanisms and outcomes. Instrumental variable estimation procedures can produce valid causal estimates even in the presence of this unmeasured confounding, but the problem then becomes finding valid instruments to identify the statistical models. So far, however, little attention has been given to the issue of designing trials of complex interventions with this analysis in mind.