Handling outcome multiplicity in systematic reviews on educational interventions: lessons learned from a review on first aid training for laypeople

Article type
Authors
Borra V1, D'aes T2, De Buck E2, Kendall I3, Laermans J2, McCaul M4
1Belgian Red Cross, Mechelen, Belgium
2Centre for Evidence-Based Practice, Belgian Red Cross, Mechelen, Belgium; Cochrane First Aid, Belgian Red Cross, Mechelen, Belgium
3Centre for Evidence-Based Practice, Belgian Red Cross, Mechelen, Belgium; Leuven Institute for Healthcare Policy, Department of Public Health and Primary Care, KU Leuven, Leuven, Belgium
4Division of Epidemiology and Biostatistics, Department of Global Health, Faculty of Medicine and Health Sciences, Stellenbosch University, Stellenbosch, South Africa
Abstract
BACKGROUND:
A priority exercise by Cochrane First Aid revealed that the effectiveness of first aid training for laypeople is a high-priority uncertainty among their stakeholders. While conducting a systematic review (SR) addressing this educational question, we were confronted with unanticipated outcome multiplicity for which no solutions were specified at the protocol stage.

OBJECTIVES:
To describe our approach to dealing with outcome multiplicity to enable synthesis using meta-analysis in our SR.

METHODS:
The 36 included (cluster-)randomized controlled trials reported on more than 190 outcome measures across 4 educational outcome domains, that is, first aid–related knowledge, skills, self-efficacy, and willingness to help. When examining the available data for synthesis, multiplicity arose in 2 ways: (1) multiple outcome measures were used for a single outcome domain within a single study and (2) single outcomes were measured at multiple time points after the intervention. In search of a practical solution for this issue of multiplicity, we reviewed Cochrane guidance; consulted the Cochrane Methods Support Unit, first aid experts, and the literature; and had interactive discussions among the core review team.

RESULTS:
We grouped individual outcome measures within the 4 educational outcome domains and categorized them as measured in the short (<1 month), intermediate (1-3 months), or long term (>3 months) after the intervention. In case multiple outcome measures within a study and outcome domain were available for inclusion in a synthesis, we applied the "median effect estimate" approach, ranking the effect estimates of the outcomes within the outcome domain and selecting the outcome with the median value as a representative effect to include in the meta-analysis. In case an outcome was measured multiple times within the same time frame (eg, short-term), the earliest time point was chosen. Applying this solution allowed us to select a single outcome measure per study and outcome domain to include in our meta-analyses.

CONCLUSIONS:
Outcome multiplicity is a common challenge in SRs with broad outcome domains. Our SR provides a workable approach that enables meaningful data synthesis to inform stakeholders. Although anticipating outcome multiplicity may be challenging, review teams may benefit from mapping outcomes through a scoping exercise and prespecifying solutions at the protocol stage.