Addressing missing participant data for continuous outcomes assessed with different instruments: a guide for systematic review authors

Article type
Authors
Ebrahim S1, Johnston B2, Akl EA3, Mustafa RA4, Sun X5, Walter SD1, Heels-Ansdell D1, Alonso-Coello P6, Guyatt GH1
1Department of Clinical Epidemiology & Biostatistics, McMaster University, Canada
2Department of Anesthesia and Pain Medicine, The Hospital For Sick Children, Canada
3Department of Internal Medicine, American University of Beirut, Lebanon
4Department of Medicine, University of Missouri-Kansas City, USA
5Center for Clinical Epidemiology and Evidence-based Medicine, Xinqiao Hospital, China
6Iberoamerican Cochrane Centre, CIBERESP-IIB Sant Pau, Spain
Abstract
Background: We previously developed an approach to address the impact of missing participant data for continuous outcomes in meta-analyses of trials that used the same measurement instrument.

Objectives: To extend our approach to meta-analyses including trials that use different instruments to measure the same construct.

Methods: We reviewed the available literature, and conducted an iterative consultative process with nine methodologists. We applied our approach to an example systematic review of respiratory rehabilitation for chronic obstructive pulmonary disease.

Results: Our approach involves first choosing a reference instrument, typically the one that is most familiar to the target audience and/or has the best measurement properties. Second, we convert scores from different instruments to the units of the reference instrument. Third, we impute the means for participants with missing data using five sources of data that reflect observed outcomes from the trials in the systematic review. These range from the best mean score among the intervention arms of included trials to the worst mean score among the control arms of included trials. Fourth, we apply four increasingly stringent imputation strategies for addressing missing participant data (Table 1). To impute standard deviation (SD), we used the median SD from the control arms of all included trials. Finally, we calculate a pooled mean difference for the complete case analysis and each of the four imputation strategies. In the example review, pooled effect estimates diminished but lost significance only with the most stringent strategy (Strategy 4, Fig. 1). When judging the risk of bias as a result of missing participant data, one should consider both the plausibility of the more stringent strategies, and the importance of the apparent intervention effects.

Conclusions: Our extended approach provides guidance for addressing missing participant data in systematic reviews of trials using different instruments to measure the same construct.