Exploring indirect effects of comparative effectiveness in a Cochrane Review

Article type
Authors
Devine B1
1University of Washington, USA
Abstract
Background: Often the effects of alternative interventions designed to improve care are evaluated against a usual care or placebo control group and not directly against each other. In situations where controls are equivalent, one can compare the effects of alternative interventions indirectly.

Objectives: To explore the feasibility of examining indirect effects in a Cochrane meta-analysis where no direct comparisons are reported in the literature and only subgroup analyses have previously been reported. Indirect comparisons are used to rank order these interventions, estimate the probability of which intervention is best, and make pairwise comparisons.

Methods: Results from a published Cochrane review provided evidence from 12 trials (n = 1114) that asthma self-care education is effective in reducing emergency department (ED) visits (Random Effects SMD -0.27, 95% CI -0.45 to -0.09). An update of this review has re-confirmed this finding (28 trials; n = 3822$; SMD -0.26, 95% CI -0.33 to -0.13). Subgroup analyses were conducted to evaluate which of three interventions (individual, small group, or combined) resulted in fewer ED visits. Results from these subgroup analyses are now re-analyzed in a Bayesian framework using WinBUGS open source software to make these indirect comparisons.

Results: We combined 28 trials, employed a random effects linear model, and conducted the Bayesian, indirect treatment comparison. Results are consistent with the original results, suggesting that asthma self-care education is effective in reducing childrenás ED visits. Indirect comparisons using usual care as the common referent indicated the individual intervention was most effective (79% probability of being best), followed by the small group intervention (18%), followed by the combined intervention (3%). Pairwise comparisons revealed no significant differences between interventions. Results were robust to changes in the initial values of the priors.

Conclusions: In this example, the results of the indirect treatment comparisons are consistent with results of the traditional meta-analysis, and provide additional information to inform treatment decision-making.