Handling trial participants with missing dichotomous outcome data when conducting a meta-analysis: a methodological survey of proposed approaches

Article type
Authors
Akl EA1, Kahale L1, Agoritsas T2, Brignardello Petersen R3, Busse JW2, Carrasco A3, Ebrahim S3, Johnston B3, Neumann I3, Sola I4, Sun X3, Vandvik P5, Alonso-Coello P4, Guyatt G3
1American University of Beirut, Lebanon
2 McMaster University, Canada
3McMaster University, Canada
4Iberoamerican Cochrane Centre, Spain
5The Norwegian Knowledge Centre for the Health Services, Norway
Abstract
Background:
Missing outcome data for trial participants represents a serious potential source of bias, particularly when associated with the likelihood of outcome events. Systematic review authors frequently deal with this problem when conducting meta-analyses.

Objectives:
To review the literature systematically to identify proposed approaches that systematic review authors should use to handle missing participant data (MPD) when conducting a meta-analysis.

Methods:
We searched MEDLINE and the Cochrane Methodology Group Specialised Trials Register. We included papers that devoted at least two paragraphs to discuss a relevant approach for missing dichotomous data. Ten pairs of reviewers selected relevant papers in duplicate and independently. One reviewer (LK) abstracted data in a tabular format from included papers and a second reviewer (EA) verified them. We then prepared narrative summaries of the results

Results:
Of 9138 assessed citations, six proved eligible: three provided general approaches and three addressed specific statistical issues when handling MPD for a dichotomous outcome when conducting a meta-analysis. The Table summarizes the three proposed general approaches. They recommend a complete case analysis as a primary analysis, and additional sensitivity analyses using different imputation methods, including: based on reasons for missingness, relative to risk among followed-up, best-case scenario, and worst-case scenario. Two approaches suggest taking uncertainty into account. Out of the three papers addressing specific statistical issues, one discussed correcting the bias resulting from missing data in a meta-analysis (Yuan 2009), and two discussed statistical methods for allowing for uncertainty due to missing data in meta-analysis (White 2008).

Conclusions:
The results of this study summarize current recommendations regarding when or how systematic reviews authors may handle MPD when conducting a meta-analysis. While these approaches require further testing, they will help in developing specific guidance for Cochrane review authors on how to address missing participant dichotomous data.