Conducting network meta-analysis with sparse data and many interventions: recent examples and the issues they raise

Article type
Authors
Norman G1, Westby M1, Dumville J1, Cullum N1
1Division of Nursing, Midwifery & Social Work, School of Health Sciences, Faculty of Biology, Medicine & Health,University of Manchester
Abstract
Background: Network meta-analysis (NMA) allows comparison and ranking of treatments across trials. This may be particularly important in conditions with large numbers of alternative treatment options such as complex wound care. This area is also characterised by largely under-powered studies at high risk of bias – since such studies largely constitute the current evidence base they are hard to ignore.
Objectives: To explore methodological issues in conducting and interpreting NMAs where there are many interventions and contributing trials are generally small and at high risk of bias.
Methods: We conducted two NMAs of dressings and topical treatments for complete wound healing, in pressure ulcers (PU) and venous leg ulcers (VLU). We used a frequentist approach and conducted GRADE assessment.
Results: These sparse networks included 39 studies of 21 interventions (2127 participants) (PU) and 59 studies of 25 interventions (5156 participants) (VLU). Figure 1 and Figure 2 show network diagrams with nodes, weighting and risks of bias.
Methodological issues arose in conducting and interpreting the NMAs. Many studies were excluded because they did not report complete healing, the outcome most important to people with complex wounds. Node-making decisions heavily influenced network composition and led to study exclusion, and ability to model time dependence was limited. Rank order was affected by both multiple interventions and possible outlier single studies, and most evidence for individual comparisons was considered to be low or very low certainty, due to risk of bias and imprecision. We ask ‘What is the value of NMAs when data are sparse?’ and whether better understanding of uncertainty in treatment and prevention of wounds and their complications can ultimately lead to patient benefit.
Conclusions: Our NMAs with sparse data and many interventions produced imprecise results and unstable treatment rankings. The outputs highlight uncertainty rather than returning clear data with implications for practice. However, understanding uncertainty is important to underpin honest conversations about research evidence in practice and can inform prioritisation of future research questions.
Patient or healthcare consumer involvement: We used open consultation on social media to inform inclusion criteria for one NMA. Consumers reviewed all protocols and reviews.