Prospective use of network meta-analysis using formal statistical monitoring

Article type
Authors
Nikolakopoulou A1, Mavridis D2, Salanti G1
1Department of Hygiene and Epidemiology, University of Ioannina, School of Medicine, Ioannina, Greece
2Department of Primary Education, University of Ioannina, Ioannina, Greece
Abstract
Background: Pairwise and network meta-analyses (NMA) are traditionally used retrospectively to assess existing evidence. However, previous knowledge of trial results can introduce bias due to potential selective inclusion of the key components of the review question (PICO criteria). Prospective meta-analysis overcomes this limitation by requiring the identification of eligible trials before the disclosure of their results. It has been suggested that NMA can also be undertaken prospectively; this practice though has not been adopted yet.
Objectives: Designing a NMA prospectively creates a multiple testing scenario. Our objective is to develop a method for the prospective design of NMA by using formal monitoring.
Methods: We extend ideas of sequential monitoring of trials to random-effects NMA. We construct efficacy and futility stopping boundaries for NMA estimates and we present the sequential NMA procedure using repeated confidence intervals (RCI). When a RCI excludes the point of no effect the NMA effect is considered conclusive. Our method leads to recommendations of whether further research is required to inform NMA and for which comparisons. We illustrate the method using a network that evaluates the effectiveness of bare mental stent (BMS), coronary artery bypass (CAB) and drug eluting stent (DES) for all-cause mortality in diabetic patients. Studies are added in a chronological order to the NMA model.
Results: After the inclusion of eight studies none of the NMA summary effects was conclusive. The incorporation of the 9th and the 10th trial lead to a conclusive NMA effect estimate for the comparisons ‘BMS versus DES’ and ‘CAB vs BMS’ respectively showing the relative inferiority of BMS after accounting for multiple testing. If NMA had been planned prospectively, it would have stopped with the inclusion of the 13th study when NMA effects are conclusive for all comparisons.
Conclusions: Use of sequential methods in NMA can be adopted so that trials do not address already answered questions. Such a procedure would help to save resources and prevent the allocation of participants of RCTs to treatments proved to be inferior.