Laura Clark1, Caroline Fairhurst2, Elizabeth Cook2, David J Torgerson2. 1. York Trials Unit, Department of Health Sciences, University of York, Heslington, York, YO10 5DD, North Yorkshire, UK. Electronic address: laura.clark@york.ac.uk. 2. York Trials Unit, Department of Health Sciences, University of York, Heslington, York, YO10 5DD, North Yorkshire, UK.
Abstract
OBJECTIVES: An unknown number of randomized controlled trials (RCTs) have their treatment allocation subverted. If such trials are included in systematic reviews, biased results may be used to change policy. To assess whether a systematic review contains subverted trials, a meta-analysis of group differences regarding a baseline variable can be undertaken. In this article, the performance of age with another prognostic variable in detecting selection bias within systematic reviews is compared. STUDY DESIGN AND SETTING: Two Cochrane systematic reviews, one of low back pain and one of hip protectors for fracture prevention, were identified. The component RCT texts were obtained, and data were extracted on age, baseline back pain score (low back pain review), and baseline body mass (hip protector review). In this exemplar, we tested for baseline heterogeneity with a fixed-effects meta-analysis. RESULTS: Heterogeneity in age between the intervention and control groups was found. The observed heterogeneity increased with baseline back pain and body mass relative to age in each review. CONCLUSION: We found that covariates predictive of outcome demonstrate greater heterogeneity than age. However, there were fewer missing data relating to age. Reviewers should consider using age and another prognostic covariate in baseline meta-analyses to check the validity of their results.
OBJECTIVES: An unknown number of randomized controlled trials (RCTs) have their treatment allocation subverted. If such trials are included in systematic reviews, biased results may be used to change policy. To assess whether a systematic review contains subverted trials, a meta-analysis of group differences regarding a baseline variable can be undertaken. In this article, the performance of age with another prognostic variable in detecting selection bias within systematic reviews is compared. STUDY DESIGN AND SETTING: Two Cochrane systematic reviews, one of low back pain and one of hip protectors for fracture prevention, were identified. The component RCT texts were obtained, and data were extracted on age, baseline back pain score (low back pain review), and baseline body mass (hip protector review). In this exemplar, we tested for baseline heterogeneity with a fixed-effects meta-analysis. RESULTS: Heterogeneity in age between the intervention and control groups was found. The observed heterogeneity increased with baseline back pain and body mass relative to age in each review. CONCLUSION: We found that covariates predictive of outcome demonstrate greater heterogeneity than age. However, there were fewer missing data relating to age. Reviewers should consider using age and another prognostic covariate in baseline meta-analyses to check the validity of their results.
Authors: Tessa A Hulshof; Sytse U Zuidema; Peter J K van Meer; Christine C Gispen-de Wied; Hendrika J Luijendijk Journal: Int J Methods Psychiatr Res Date: 2018-12-04 Impact factor: 4.035