Extent and handling of missing dichotomous outcome data in 100 Cochrane and non-Cochrane systematic reviews

Article type
Authors
Kahale LA1, Diab B1, Brignardello-Petersen R2, Mustafa R3, Busse JW2, Agarwal A2, Kwong J4, Li L4, Neumann I5, Lopes LC6, Olav Vandvik P7, Briel M8, Iorio A2, Guyatt GH2, Akl EA1
1American University of Beirut, Lebanon
2McMaster University, Canada
3University of Missouri-Kansas City, USA
4Sichuan University, China
5Pontificia Universidad Católica de Chile, Chile
6University of Sorocaba, Brazil
7Norwegian Knowledge Centre for the Health Services, Norway
8University Hospital of Basel, Switzerland
Abstract
Background: Reports of randomized clinical trials (RCTs) do not typically report whether certain categories of participants (e.g. non-compliers, those who withdraw consent) were followed-up for the outcome(s) of interest. It is not clear how systematic reviews (SRs) address this lack of information, i.e. assume they have missing participant data (MPD) or not.
Objectives: To describe how SRs report and handle categories of participants that could be potentially counted as having MPD.
Methods: We included SRs reporting a group-level meta-analysis of a patient-important dichotomous efficacy outcome, with a statistically significant effect estimate. Thirteen reviewers, working in pairs, independently extracted data from eligible SRs. We focused on 10 categories that could be counted as potentially having MPD: 'ineligible participants', 'did not receive any treatment', 'withdrew consent', 'outcome not assessable', 'explained and unexplained lost to follow-up', 'dead', 'experienced adverse events', 'non-compliant', 'discontinued prematurely', and 'cross-over'. We assessed whether the systematic reviewers: 1) planned in the methods section to collect information about those categories; 2) reported in the results section, the number of participants in those categories; and 3) reported on methods for handling MPD in their primary meta-analysis.
Results: We included a random sample of 100 Cochrane and non-Cochrane SRs published in 2012. Tables 1, 2, and 3 report our main findings. Twenty-five per cent of the SRs reported a plan to collect information about the categories of interest in their methods section, whereas 45% reported in their results section the number of participants in those categories. Only 13% reported a method to handle MPD in their primary meta-analyses, which were primarily complete case analysis.
Conclusions: Most SRs do not explicitly report sufficient information on categories of participants that could be potentially counted as having MPD, or handle MPD in their primary meta-analysis.