Challenges and potential solutions for reporting findings from multicomponent meta-regression models in GRADE summary of findings tables
2Centre for Practice Changing Research, Ottawa Hospital Research Institute, Canada
3Institute for Health System Solutions and Virtual Care (WIHV), Women's College Hospital, University of Toronto, Canada
4Centre for Practice-Changing Research, Department of Medicine, University of Ottawa
Background: We recently completed a Cochrane review of diabetes quality improvement (QI) trials that aimed to identify promising QI strategies (or combinations of strategies) to deploy in practice or examine in future research. We coded interventions according to the presence of absence of 12 QI strategies and assessed the association between intervention components and post-intervention outcomes with Bayesian meta-regressions. This is atypical for a Cochrane review – our analysis and output do not match the usual Cochrane processes – and thus necessitated adjustments in our reporting of results to align with Cochrane procedures.
Methods: We will describe the misalignment between the goals and outputs of our analysis with the Grading of Recommendations, Assessment, Development, and Evaluations (GRADE) approach for producing Cochrane Summary of Findings (SoF) tables and our operationalizations and solutions.
Results: The standard approach for producing SoF table is to have a table for each intervention of interest. This was not feasible in our review, as studied interventions were diverse, rarely replicated, and did not capture the full range of possible interventions that could be comprised of combinations of the 12 QI strategies. Most importantly, the standard SoF table was inconsistent with the goal of our analyses which was learn across (diverse combinations of QI strategy) interventions to develop a theory of which QI strategies (or combinations) of strategies may be most promising for future practice or research. We therefore adjusted GRADE tables to report findings as they pertained to each QI strategy for each outcome. Additionally, we parsed data to inform GRADE assessments for each component (e.g., sample size, risk of bias, precision, consistency, and directness) to rate certainty of evidence. Discussion: GRADE is a transparent tool for communicating review findings. Although the goal of our analyses did not match the standard setup of GRADE SoF tables, we found it feasible to adapt SoF tables to better align with our synthesis objective. Further GRADE guidance may be required to help guide future and ongoing similar reviews aimed at teasing apart active ingredients of interventions, rather than interventions as a whole.