Node-making process in network meta-analysis of nonpharmacological treatment are poorly reported

Abstract

Objective

To identify methods to support the node-making process in network meta-analyses (NMAs) of nonpharmacological treatments.

Study Design and Setting

We proceeded in two stages. First, we conducted a literature review of guidelines and methodological articles about NMAs to identify methods proposed to lump interventions into nodes. Second, we conducted a systematic review of NMAs of nonpharmacological treatments to extract methods used by authors to support their node-making process. MEDLINE and Google Scholar were searched to identify articles assessing NMA guidelines or methodology intended for NMA authors. MEDLINE, CENTRAL, and EMBASE were searched to identify reports of NMAs including at least one nonpharmacological treatment. Both searches involved articles available from database inception to March 2016. From the methodological review, we identified and extracted methods proposed to lump interventions into nodes. From the systematic review, the reporting of the network was assessed as long as the method described supported the node-making process.

Results

Among the 116 articles retrieved in the literature review, 12 (10%) discussed the concept of lumping or splitting interventions in NMAs. No consensual method was identified during the methodological review, and expert consensus was the only method proposed to support the node-making process. Among 5187 references for the systematic review, we included 110 reports of NMAs published between 2007 and 2016. The nodes were described in the introduction section of 88 reports (80%), which suggested that the node content might have been a priori decided before the systematic review. Nine reports (8.1%) described a specific process or justification to build nodes for the network. Two methods were identified: (1) fit a previously published classification and (2) expert consensus.

Conclusion

Despite the importance of NMA in the delivery of evidence when several interventions are available for a single indication, recommendations on the reporting of the node-making process in NMAs are lacking, and reporting of the node-making process in NMAs seems insufficient.

Introduction

Clinicians should base their daily health-related decisions on the best evidence. The medical literature production is constantly growing; therefore methods to summarize evidence are strongly needed. Systematic review with meta-analysis has been developed to play this role [1], but this design can assess only two interventions in a pairwise comparison. However, several interventions are often available for a single indication, and clinicians need to assess the comparative effectiveness of each of them [2], [3]. Network meta-analysis (NMA) allows for combining both direct and indirect comparisons to estimate all possible pairwise comparisons between interventions for a single indication and ranking them according to their estimated treatment effect [4], [5]. This method has recently experienced incredibly fast use [6].

Nonpharmacological treatments include interventions such as surgery, technical procedures, devices, rehabilitation, psychotherapy, behavioral interventions, and alternative medicine [7]. They are widely used in clinical practice and represent 24% of assessed interventions in published randomized controlled trials [8], [9], [10], [11]. Nonpharmacological treatments are usually complex interventions, involving several components, each component possibly affecting the success of the intervention [12]. The complexity of these interventions raises important issues when performing NMAs. One major issue is to gather nonpharmacological treatments together in a homogenous group to allow for comparison. This process, called “lumping”, is known to feature limitations for further interpreting results [13], [14], [15], [16], [17], [18], [19], [20]. Indeed, for a single clinical indication, similar but not identical interventions can be lumped in different ways, which might impact the result of the NMA [18], [21]. For example, Del Giovane et al. proposed models to evaluate the assumption of similarity of dose effect [18] for situations of doubt as to whether the various doses of the same treatment can be lumped or not.

In 2015, the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) group developed a guideline dedicated to NMAs [22]. This guideline recommends specifying inclusion and exclusion criteria for each node and providing a justification when interventions are merged to form a single comparator (i.e., lumping of interventions). Indeed, different ways of lumping interventions could affect the results of the study [18], [21]. However, no method is available to support the node-making process, and to our knowledge, guidelines on this issue are lacking.

We aimed to identify methods for building nodes in NMAs assessing nonpharmacological interventions. To achieve this, we planned a literature review of methodological articles of NMAs, and a systematic review to describe methods currently in use and how they were developed.