Understanding network meta-analysis

Abstract

Recently, in the World Journal of Clinical Cases, studied the different non-steroidal anti-inflammatory drugs (meloxicam, celecoxib, naproxen, and rofecoxib) for juvenile idiopathic arthritis with network meta-analysis (NMA). This manuscript aims to introduce to clinicians what NMA is. NMA represents a fundamental technique for simultaneously comparing three or more interventions within a single analysis, harnessing both direct and indirect evidence derived from a network of studies. It surpasses pair-wise meta-analysis which are confined to direct comparison of two items in clinical trials. This approach can estimate the relative effects between any pair of interventions within the network, often yielding more precise estimations than those generated from single direct or indirect analyses. NMA necessitates steps akin to those of conventional meta-analysis, involving a thorough literature search, assessment of potential trial biases, statistical analysis of reported pairwise comparisons for all relevant outcomes, and evaluation of overall certainty of evidence on an outcome-specific basis. However, NMA demands substantial resources, given its propensity to address broader inquiries, typically involving a larger number of studies at each phase of the systematic review, from screening to analysis, compared to traditional meta-analyses.

Key Words: Meta-analysis; Systematic review; Methodology; Research; Journal; Academic

Core Tip: Network meta-analysis stands as a potent instrument for comparative research of three or more. It surpasses pair-wise meta-analysis in complexity. Additionally, supplementary analyses, such as network meta-regression further elevate the intricacy of the analysis.

TO THE EDITOR

Meta-analysis is a statistical tool for pooling data and results across different studies hoping to arrive on a more precise conclusions to a clinical question. However, traditional meta-analyses are notably confined to direct comparison of two items (either intervention or treatment) in clinical trials. In reality, numerous trials contain more than one single active therapeutic arm comparing against placebos, usual practice, or the current standard of care as primary outcomes; whereas the comparison of results across different interventions may be the secondary outcomes. In response to these limitations, network meta-analysis (NMA) has emerged, enabling the computation of comparative effects among more than two interventions, even when lacking direct comparison within clinical trials.

NMA represents a fundamental technique for simultaneously comparing three or more interventions within a single analysis, harnessing both direct and indirect evidence derived from a network of studies. Zeng et al[1] studied the different non-steroidal anti-inflammatory drugs (meloxicam, celecoxib, naproxen, and rofecoxib) for juvenile idiopathic arthritis with NMA. This approach can estimate the relative effects between any pair of interventions within the network, often yielding more precise estimations than those generated from single direct or indirect analyses. Moreover, it facilitates the estimation of rankings and hierarchies of interventions[2].

HOW DOES NMA BEGIN

NMA necessitates steps akin to those of conventional meta-analysis, involving a thorough literature search, assessment of potential trial biases, statistical analysis of reported pairwise comparisons for all relevant outcomes, and evaluation of overall certainty of evidence on an outcome-specific basis. Zeng et al[1] thoroughly searched over different databases, and yielded 755 results. They also listed out the bias assessment of each study. NMA then identifies interventions linked by a common comparator. For instance, distinct active treatments may have been compared against placebos in separate trials. NMA enables the creation of a hypothetical trial comparing these active treatments based on their effects against a shared placebo, generating "indirect" evidence. These indirect comparisons serve to bridge knowledge gaps within existing evidence, yielding a more comprehensive understanding of treatment alternatives for clinicians. Once all treatments within a network have been compared, various methods exist for ranking treatments, conveying their relative net effectiveness[3].

The validity of NMA lies on the assumption that studies included in the analysis are similar in all major factors that would not induce a significant relative effect across studies. However, incoherence, also known as inconsistency, emerges when different input studies’ results were contradicting to each other’s. Grading confidence of evidence derived from a NMA commences with a meticulous evaluation of confidence in each direct comparison[4]. Domain-specific assessments were subsequently combined to evaluate the overall confidence in the evidence, encapsulating the multifaceted nature of this analytical approach while underscoring its potential impact on clinical decision-making and policy formation.

The utilization of a NMA encompasses the advantages of all accessible direct and indirect evidence. Research studies have indicated that this approach yields estimations of intervention effects with greater precision compared to individual direct or indirect estimates[5]. Moreover, it offers the capacity to furnish comparative data for interventions that have not been individually assessed within randomized trials[6]. This concurrent comparison of all pertinent interventions within a single analysis facilitates the estimation of their relative ranking concerning a specified outcome.

CONCLUSIONS

NMA stands as a potent instrument for comparative research of three or more. It surpasses pair-wise meta-analysis in complexity. Additionally, supplementary analyses, such as network meta-regression further elevate the intricacy of the analysis[7]. Notably, NMA demands substantial resources, given its propensity to address broader inquiries, typically involving a larger number of studies at each phase of the systematic review, from screening to analysis, compared to traditional meta-analyses.