P32 The Use of 'Placebo Adjusted Risk Difference Network Meta-Analysis' to Account for Confounding by Heterogeneity in Placebo Response Across Studies

Abstract

Variation in placebo response can ‘confound’ indirect comparisons as it effectively acts as a treatment effect modifier. At the placebo response increases it will tend to “crowd” out the response to active treatments reducing relative treatment effects. The potential confounding arising from heterogeneity in placebo response has been addressed by conducting meta-regression with placebo response as a predictor or by using the risk difference scale. However, meta-regression analyses often lack power, and some degree of confounding remains on the risk difference scale. For binary endpoints, based on an assumption regarding the correlation between treatment and placebo response, an explicit adjustment can made for the contribution of the placebo response to the observed response for active treatments. It is them possible to obtain an estimate of the response to active treatments in the absence of any placebo response. These estimates can be compared to estimate relative treatment effects on the risk difference scale that are not confounded by heterogeneity in treatment response.