Online survey to identify methods used in meta-analysis to handle missing continuous outcome summaries in stroke rehabilitation systematic reviews




Poster session 5


Thursday 27 October 2016 - 10:30 to 11:00


All authors in correct order:

Na LH1, Brady MC2, Lewis SC1, Murray GD1, Langhorne P3, Weir CJ1
1 Centre for Population Health Sciences, University of Edinburgh, UK
2 Nursing, Midwifery and Allied Health Professions Research Unit, Glasgow Caledonian University, UK
3 University of Glasgow, UK
Presenting author and contact person

Presenting author:

Lumine Na

Contact person:

Abstract text
Background: Conventional methods for meta-analysis of continuous outcomes, based on the mean and standard deviation (SD), are often difficult to apply where; 1) outcomes have a skewed distribution, and 2) the mean and SD are not reported. Systematic review authors must then either seek the missing information from the trialists, omit the trial from the meta-analysis or use other statistical summaries if the trial is to be retained in the meta-analysis.

Objectives: We aimed to identify methods to assist systematic review authors in producing the best possible summaries of the evidence. The frequency with which missing mean or SD values occurred, and the methods used to address the issue were investigated in an online questionnaire of Cochrane Stroke Review (CSR) authors.

Methods: We approached authors of CSRs of rehabilitation interventions. Lead and second authors (plus the contact author, if different) of each review were asked to complete a questionnaire. Data were analysed descriptively to summarise the extent of unreported mean and SD data in CSRs and the methods used to handle this.

Results: The online survey was sent to 141 reviewers for 70 CSRs (Fig 1). Sixty-three responses (44% of 141) covered 53 CSRs (76% of 70). Most review authors (58 of 63) knew about analyses performed in the review; 56 had aimed to analyse continuous outcomes. Nearly all authors responded that mean, SD and sample size were to be extracted as part of the planned analysis; unreported mean and SD were encountered by most authors. Despite missing summary data, a meta-analysis was often performed. The majority of review authors contacted trialists to request missing items; however, most reported that trials were omitted from their analysis due to missing mean or SD.

Conclusions: Most reviewers omitted trials from meta-analyses due to missing summary data. In addition to the guidance on handling missing SD values available in the Cochrane Handbook for Systematic Reviews of Interventions, use of other available methods for imputing the mean and SD will help review authors maximise the information included in meta-analyses.