Meta-analysis is of critical importance for healthcare decision makers to assess the comparative efficacy and safety of interventions and is becoming an increasingly important component of health technology assessment. Across a wide range of therapeutic areas, many key efficacy measures are reported on a continuous scale; a key challenge for the meta-analysis of continuous outcomes is the failure of trials to report measures of uncertainty around estimates of treatment effect. The potential association of incomplete reporting of outcomes and non-statistical significance of trial results is particularly concerning for meta-analysis. The omission of studies from meta-analysis due to incomplete reporting may introduce bias and reduce statistical power. We have recently published the first review of methods to deal with missing variance data focusing on a single therapeutic area (1).
We systematically searched electronic databases for publications reporting meta-analyses of pharmaceutical interventions for type 2 diabetes mellitus (T2DM) populations. The review was limited to meta-analyses published after 2013 to ensure it was representative of the most current practices used by authors to address missing variance data in meta-analysis of continuous outcomes.
We found that a large proportion of recently reported meta-analysis publications in T2DM fail to report approaches to dealing with missing variance data for continuous outcomes. The trial-level imputation methods identified in this review can be considered as relatively basic. Where reported, details of approaches to imputation were insufficient and inconsistent across studies, with only a small minority reporting sensitivity analyses to assess the robustness of the results to the methods of imputation.
This research highlights a lack of standardisation of methods to address missing variance data in meta-analysis. We discourage authors from using a no-imputation approach unless it is inappropriate, such as in cases where a large proportion of the data has missing variances. A systematic approach to imputation is recommended which should include sensitivity analyses and authors are encouraged to improve the quality of reporting of their meta-analyses.