Identifiability of Components of Complex Interventions Using Factorial Designs

Abstract

Objective:

The aim of this contribution is to demonstrate how the component structure of a complex intervention (CI) can be efficiently exploited for study design and statistical analysis by using concepts of factorial designs. Many studies on CIs in complementary and alternative medicine exhibit the structure of factorial designs, where all possible combinations of the levels of two or more treatments occur together. In this contribution, the treatment arms of CI studies are explicitly viewed as factorial combinations of their components. Experimental design offers the general concept of identifiability of effects, that is, unique estimability of the components' effects from the observed data. For factorial designs, a simple cross table representation of the treatment arms can show the components or sums or interactions of components that are identifiable within a given study design. The question of identifiability arises particularly if some combinations of components are not observed (e.g., individualized homeopathic prescription without consultation). Study designs from published homeopathy studies are used for demonstration.

Conclusions:

CI studies should explicitly use an intervention's factorial component structure if it is inherent in the treatment arms being compared. In this way, investigators can avoid study designs from which the effects of interest cannot be uniquely estimated and improve the interpretation of estimated effects.