How do Cochrane review authors interpret subgroup analyses?

Article type
Authors
Richardson M1, Garner P1, Donegan S1
1Cochrane Infectious Diseases Group, United Kingdom
Abstract
Background: Treatment by covariate interactions can be explored in reviews using subgroup analyses. Such analyses can identify how treatment effect varies by subgroup and are an important methodological approach for stratified medicine. A review of 52 recently published Cochrane Reviews found that 63% applied subgroup analyses [1]. However, interpretation of subgroup analyses in Cochrane Reviews is poor; the review found that only 3% of reviews reported whether or not there was an interaction (i.e. a difference between subgroup results) [1]. Reasons for the lack of interpretation have not been explored.
Objective: To identify how review authors interpret subgroup analyses and why analyses are not always interpreted.
Methods: We will develop and pre-pilot survey/interview questions based on subgroup analysis scenarios (i.e. no interaction; qualitative and/or statistically significant interaction). We will ask review authors: 1) to interpret each analysis; 2) whether there is a statistically significant interaction; 3) how they decided if the interaction was statistically significant. We will also ask why subgroup analyses were not interpreted in their review (if applicable).
We will survey correspondence authors for the 52 Cochrane reviews that were included in the recent review [1]. The review included the most recently published review for each Review Group published in the Cochrane Database of Systematic Reviews (searched 8 August 2013). Review authors based locally will be interviewed.
Results: We will summarise qualitative results (e.g. interpretations of analyses) and quantitative results (e.g. number of review authors who correctly identified a significant interaction). Review authors will be anonymised.
Conclusions: This research will identify how reviews authors interpret subgroup analyses, why subgroup analyses are often not interpreted, and improve how analyses are interpreted in future reviews.
Reference:
1. Donegan S, Williams L, Dias S, Tudur-Smith C, Welton N. Exploring treatment by covariate interactions using subgroup analysis and meta-regression in Cochrane Reviews: a review of recent practice. Submitted to PLoS one.