Keywords: multicomponent interventions, meta-analysis, hierarchical meta-regression
When synthesizing trial evidence about the effects of multicomponent interventions, standard meta-analysis methods reduce the data into simple pairwise comparisons (e.g., any component vs. no component). Such simplifications result in substantial loss of information and cannot isolate the effects of each intervention component. We explored the use of hierarchical meta-regression models to capture the complexity of multicomponent interventions. Our meta-analysis included 278 quality improvement trials that assessed at least one of 12 component strategies for diabetes management. A series of hierarchical models were implemented to assess the effects of these strategies. The models leverage different assumptions (e.g., grouping strategies into broader categories, assuming additivity of effects, ignoring co-interventions) to estimate the effects of the strategies. Models were extended to assess interactions between strategies and account for particular aspects of the available data (e.g., cluster trials). Compared to standard methods, our models were able to isolate the effects of individual components and resulted in different effectiveness rankings of the 12 strategies.