Risk difference estimation and reporting in systematic reviews

Article type
Authors
Alonso-Coello P1, Neumann I2, Carrasco-Labra A3, Brignardello-Petersen R4, Irfan A5, Sol\'{a} I1, Dahm P6, Glujovsky D7, Johnston B8, Martinez L1, Ramirez-Morera A9, Sun X10, Vandvik P11, Akl E12, Tikkinen K13, Iorio A8, Santesso N8, Brozek J8, Sch\"{unemann} H8, Guyatt G8
1Iberoamerican Cochrane Centre, Institute of Biomedical Research (IIB Sant Pau) Barcelona, Spain
2Department of Medicine, Pontificia Universidad Catolica de Chile, Santiago, Chile; Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada
3Evidence Based Dentistry Unit, Faculty of Dentistry, Universidad de Chile, Chile; Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada
4Evidence Based Dentistry Unit, Faculty of Dentistry, Universidad de Chile, Chile; Department of Health Policy, Management & Evaluation, University of Toronto, Canada
5Internal Medicine Residency Program, University of Illinois at Urbana-Champaign, USA
6Department of Urology, College of Medicine, University of Florida, Gainesville, Florida, USA
7Institute for Clinical Effectiveness and Health Policy, Buenos Aires, Argentina
8Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada
9IHCAI Foundation, Costa Rica
10Center for Health Research, Kaiser Permanente Northwest, Portland, USA
11Norwegian Knowledge Centre for the Health Services, Oslo, Norway
12Department of Medicine, State Univeristy of New York at Buffalo, NY, USA; Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada
13Department of Urology, Helsinki University Central Hospital, Helsinki, Finland; Department of Clinical Epidemiology & Biostatistics, McMaster University, Hamilton, Canada
Abstract
Background: Trading off desirable and undesirable consequences of alternative patient management strategies requires knowledge of estimates of both relative and absolute measures of effect (AME). It is unclear to what extent and how authors of systematic reviews (SRs) determine and report AME.

Objectives: To determine the proportion of SRs reporting AME and to evaluate how these are derived and interpreted. In particular, we will evaluate the proportion of SR that use relative measures for benefit outcomes and for harm. Additionally, we will test if AM reporting is associated with prespecified study characteristics.

Methods: We will include SRs of interventions that report dichotomous outcomes. We will search MEDLINE for a random stratified sample of Cochrane and non-Cochrane SRs. From a pilot study of 50 SRs, we will calculate the final sample size to test our hypothesis, which is that a considerable proportion of SRs do not report AME and that the methods used are often inadequate. We will determine the proportion of SRs reporting AME and how they are obtained (e.g. meta-analytically combined across studies or computed based on relative effect measures and assumed baseline risk) and reported (e.g. risk difference or number-needed-to-treat). We will also assess how baseline risk was obtained (e.g. as mean or median from included trials or from representative observational studies), and whether authors provided AME for more than one risk group. Finally, we will conduct multivariable logistic regression analyses to examine the relationship of reporting AME with a set of prespecified study characteristics (e.g. type of SR [Cochrane vs not Cochrane], statistical significance or source of funding).

Discussion: The project results will be presented at the Colloquium. Our results will inform the extent of limitations of determining and reporting AME in SRs and lay the foundation for further studies about the implications for decision-making.