GRADE guidance for assessing risk of bias associated with missing participant outcome data in meta-analysis

Article type
Year
Authors
Guyatt G1, Ebrahim S1, Johnson B2, Alonso-Coelloe P3, Mathioudakis A3, Briel M4, Mustafa R5, Sun X6, Walter S1, Heels-Ansdell D1, Neumann I7, Akl E8
1McMaster University, Canada
2The Hospital for Sick Children, Toronto, Canada
3Iberoamerican Cochrane Centre, Barcelona, Spain
4Basel Institute for Clinical Epidemiology and Biostatistics, Switzerland
5Departments of Medicine and Biomedical and Health Informatics, University of Missouri-Kansas City, USA
6Center for Clinical Epidemiology and Evidence-Based Medicine, Xinqiao Hospital, Chongqing, China
7Department of Internal Medicine, Pontificia Universidad Catolica de Chile, Santiago, Chile
8Department of Internal Medicine, American University of Beirut, Lebanon
Abstract
Background: Detailed guidance for assessing the risk of bias associated with missing participant outcome data in meta-analyses has, until recently, been very limited. Available guidance has been available only at the individual study level and not at the body of evidence level.

Objective: To present recently approved GRADE (Grading of Recommendations, Assessment, Development and Evaluation) guidance for assessing the risk of bias associated with missing data at the meta-analysis level.

Methods: Systematic survey of existing methodological research, iterative discussions among the investigators, testing in systematic reviews, and feedback from the GRADE Working Group.

Results: Approaches begin with a primary meta-analysis using a complete case analysis (i.e. excluding those with missing data) followed by sensitivity meta-analyses imputing, in each study, data for those with missing data, and then pooling across studies. For binary outcomes we suggest use of 'plausible worst case' in which review authors assume that those with missing data in treatment arms have proportionally higher event rates than those followed successfully. For continuous outcomes, imputed mean values come from other studies within the systematic review, and the standard deviation from the median standard deviations of the control arms of all studies. For meta-analyses in which investigators have used different instruments to address the same construct, our approach involves choosing a reference measurement instrument and converting scores from different instruments to the units of the reference instrument. For all approaches, if the results of the primary meta-analysis are robust to the most extreme assumptions viewed as plausible, one does not rate down quality of evidence for risk of bias due to missing participant outcome data. If the results are not robust to plausible assumptions, one would rate down quality for risk of bias.

Conclusions: This GRADE guidance provides structured and transparent methods for establishing the extent to which missing participant outcome data impacts risk of bias in meta-analyses of randomized trials for both binary and continuous outcomes.