The comparison of various inconsistency models in network meta-analysis
Date Issued
2014
Date
2014
Author(s)
Kuo, Yu-Chen
Abstract
Background：
Pairwise meta-analysis can only deal with comparisons between two treatments. Network meta-analysis is a new research synthesis method for comparing more than two different treatments. Inconsistency is an important issue in network meta-analysis and is defined as the difference between direct evidence and indirect evidence. There are two types of inconsistency in the literature, namely, “the loop inconsistency” and “the design inconsistency”.
Objective：
The main objective of this study is to compare different inconsistency models and different methods for evaluating inconsistency from both theoretical and practical perspectives with use of real examples for illustration.
Methods：
Five methods are compared for assessing inconsistency. Lu & Ades model compare estimating for differences in treatment effects between the uses of basic parameters and functional parameters; it is the first model that defines inconsistency. Unrelated Mean Effect model estimate all differences between pairs of treatments based on direct evidence. Unrelated Mean Effect model’s results can be viewed as results of multiple pairwise meta-analyses. Back-Calculation method evaluates the inconsistency in the whole network between direct and indirect evidence. Higgins & White model defines the inconsistency as the differences between study designs, i.e. the treatments compared in each study. The differences within each study design is considered heterogeneity Last, Krahn model combines pairwise meta-analysis with design inconsistency; it then uses the net heat plot to demonstrate where the design inconsistency is in network.
Results：
Pros and cons of the five methods and the potential problems arising from the analysis are discussed. Lu & Ades model is more robust than other models, but it cannot estimate design inconsistency. Unrelated Mean Effect model need neither the consistency assumption nor inconsistency parameters, but the model does not provide a holistic evaluation for the whole network. Back-Calculation method checks inconsistency in network but not design inconsistency. Higgins & White model is the first model to estimate design inconsistency parameters, but there were difficulties in setting and explaining the design inconsistency parameters. The net heat plot evaluates design inconsistency, but the Krahn model estimates differences between treatments without taking heterogeneity and inconsistency into account.
Conclusions：
Lu & Ades model is considered the best model for evaluating inconsistency in network meta-analysis than other models. Results from Lu & Ades model are easy to understand and interpret. Back-Calculation method can be conducted along with Lu & Ades’s model for identifying inconsistency in network meta-analysis.
Subjects
網絡統合分析
直接比較
間接比較
傳統不一致性
design不一致性
基礎參數
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-103-R01849033-1.pdf
Size
23.32 KB
Format
Adobe PDF
Checksum
(MD5):1fc79eef5810a938625a50fced71415f