Subgroup analyses in systematic reviews: a survey of Cochrane and non-Cochrane reviews

Article type
Authors
Sun X1, Brignardello-Petersen R2, Carrasco-Labra A2, Neumann I3, Alonso-Coello P4, Akl E5, Johnston B6, Briel M7, Busse J8, Dahm P9, Granados C4, Glujovsky D10, Iorio A8, Irfan A11, Martinez L4, Reem Mustafa RM8, Ramirez-Morera A12, Sola I4, Sultan S13, Tikkinen K14, Vandvik P15, Zhang M4, Guyatt G4
1Centre for Health Research, Kaiser Permanente Northwest, Portland, Oregon, USA
2Evidence Based Dentistry Unit, Faculty of Dentistry, Universidad de Chile, Chile
3Department of Medicine, Pontificia Universidad Catolica de Chile, Santiago, Chile
4Iberoamerican Cochrane Centre, Institute of Biomedical Research (IIB Sant Pau), Barcelona, Spain
5Department of Medicine, State University of New York at Buffalo, NY, USA
6Research Institute, Child Health Evaluative Sciences, The Hospital for Sick Children, Toronto, Canada
7Basel Institute for Clinical Epidemiology and Biostatistics, University Hospital Basel, Switzerland
8Department of Clinical Epidemiology & Biostatistics, McMaster University, Canada
9Department of Urology, College of Medicine, University of Florida, Gainesville, Florida, USA
10IECS (Institute for Clinical Effectiveness and Health Policy), Buenos Aires, Argentina
11Internal Medicine Residency Program, University of Illinois at Urbana-Champaign, USA
12CCSS, Costa Rica
13Division of Gastroenterology, Hepatology, and Nutrition, Malcom Randall Veterans Affairs Medical Center, University of Florida College of Medicine
14Dept. of Urology, Helsinki University Central Hospital, Helsinki, Finland
15Norwegian Knowledge Centre for the Health Services, Oslo, Norway
Abstract
Background: Studies included in a meta-analysis often differ in methodological, patient, and intervention characteristics, and outcome measures, differences that may result in heterogeneity of treatment effects. Subgroup analyses may explain such heterogeneity. However, both inappropriate conduct and reporting of subgroup analyses may lead to misguided clinical and policy decisions, and the extent to which systematic review authors undertake subgroup analyses and the quality of their conduct and reporting remain uncertain.

Objective: To examine the extent to which subgroup analyses are conducted, and to compare the quality of conduct and reporting of subgroup analyses between Cochrane versus non-Cochrane reviews.

Methods: We searched the Cochrane Database of Systematic Reviews and the Medline for systematic reviews published in 2010 that included a meta-analysis of dichotomous outcome reported in randomized controlled trials. Teams of two reviewers screened studies for eligibility. We randomly selected 75 Cochrane and 75 non-Cochrane reviews from eligible studies. Teams of two reviewers are currently abstracting data using standardized, piloted forms guided with written instructions We will focus on subgroup analyses of aggregate data meta-analyses, and the unit of analysis is an individual systematic review. Our analysis will examine whether authors planned to explore heterogeneity, whether they were able to explore heterogeneity, reasons for the failure of exploring heterogeneity, and approaches to exploring heterogeneity. We will examine whether a subgroup hypothesis was pre-specified, the number of subgroup hypotheses tested, the number of outcomes and corresponding effect measures used for subgroup analyses, and whether a test of interaction was used. For each major issue, we will compare practice in Cochrane versus non-Cochrane reviews.

Results: The results of analyses will be available for presentation at the colloquium.

Discussion: The results of this study will provide an important insight about the quality of conduct and reporting of subgroup analyses in systematic reviews.