Network meta-analysis of complex interventions: accounting for component effects and control group risk

Freeman SC1, Scott NW2, Powell R3, Johnston M2, Sutton AJ1, Cooper NJ1
1University of Leicester; NIHR Complex Reviews Support Unit, 2University of Aberdeen, 3University of Manchester

Background: In many medical areas treatment interventions consist of multiple components. In 2009 Welton et al. (1) proposed 4 increasingly complex network meta-analysis models for assessing component effects of complex interventions. The additive-effects model allows a separate effect for each component. The two-way interaction model extends this to allow pairs of interventions to have bigger or smaller effects than would be expected from their individual components.

Objectives: To apply the additive-effects and two-way interaction models to an existing Cochrane review of psychological preparation interventions for adults undergoing surgery (2) and extend the models to account for control-group risk (CGR) for the continuous outcome length of stay (LOS). By doing this it is hoped that the most-effective components, and combinations of components, of the interventions can be identified.

Methods: We used a network of 36 trials comparing combinations of 6 components for psychological intervention before surgery. CGR was accounted for by allowing component effects to vary across different values of control-group LOS. Models were fitted within a Bayesian framework using WinBUGS and accounting for measurement error in CGR.

Results: Clinical and statistical heterogeneity were identified in the network. CGR was an important factor in determining the effectiveness of interventions. Specific component effects by comparison only explained a small fraction of the between-study heterogeneity.

Conclusions: We extended an existing Cochrane review to answer relevant clinical questions. This approach allowed component-specific effects to be estimated and to identify combinations of components responsible for a clinically significant improvement in LOS. This approach could be utilised when considering cost effectiveness by identifying the most-effective combinations of components for a specific type of surgery. More intensive interventions may be justified on cost-effectiveness grounds for certain types of surgery.

References

1. Welton NJ, et al. Am J Epidemiol 2009;169:1158

2. Powell R, et al. Cochrane Database of Systematic Reviews 2016, Issue 5 Art No:CD008646