Node-making processes in network meta-analysis of non-drug treatments

Shi C1, Westby M1, Norman G1, Dumville J1, Cullum N2
1Division of Nursing, Midwifery & Social Work, School of Health Sciences, Faculty of Biology, Medicine & Health, University of Manchester, Manchester Academic Health Science Centre, Manchester, UK, M13 9PL, 2Division of Nursing, Midwifery & Social Work, School of Health Sciences, University of Manchester; Research and Innovation Division, Manchester University NHS Foundation Trust, Manchester Academic Health Science Centre, Manchester

Background: Network meta-analyses (NMAs) are common outputs of evidence synthesis. Treatment nodes are integral to network formation and represent the entities being evaluated. Because different node-making methods can alter the network and subsequent results drastically, robust conduct and reporting are needed. There is scope to expand current recommendations for NMAs of non-drug treatments so as to avoid arbitrary node-making.

Objectives: To enhance existing recommendations on the conduct and reporting of node-making process for NMAs of non-drug treatments.

Methods: We synthesised a node-making taxonomy based on a review of methodological papers. The taxonomy was developed and refined further in conjunction with methodological and clinical experts as we considered node-making processes in NMAs of treatments for pressure ulcers. We summarised our findings narratively and integrated them into existing recommendations.

Results: We summarise four node-making approaches. A broad lumping approach that groups similar treatments at a broad level, which is useful to estimate effects of treatment groups. A clinically-meaningful-element approach groups treatments with similar components, taking account of clinically important variables. A component lumping-and-dismantling approach is informed by meta-regression to investigate effects attributed to different treatment components. A class-effect model approach lumps similar treatments as a class, but assumes effect variations amongst these treatments, and uses modelling to estimate effects of specific treatments. Using this taxonomy and the practical implications it highlights, we suggest the following additional elements are needed in node-making guidance:

- explicitly considering the effect of lumping on the exclusion of studies from the NMA;

- use of additional sources of information to improve descriptions of non-drug treatments in included studies, which thus supports node-making judgments;

- reporting sources of information about treatments (e.g. manufacturers’ websites);

- following a reliable approach in order to avoid arbitrary node-making (e.g. undertaken independently by two people against defined criteria).

Conclusions: We propose the addition of four new elements to current recommendations, and suggest that node formation should follow a robust, preplanned process that is fully reported.