Letter by Ye and Zhang Regarding Article, “Bayesian Methods Affirm the Use of Percutaneous Coronary Intervention to Improve Survival in Patients With Unprotected Left Main Coronary Artery Disease”
To the Editor:
Recently, Bittl1 et al performed a network meta-analysis of indirect evidence using Bayesian methods and proved that percutaneous coronary intervention (PCI) improves survival for patients with unprotected left main coronary artery disease compared with medical therapy. However, we are concerned about the validity of the network meta-analysis. Network meta-analysis is a statistical method for combining both direct and indirect evidence from multiple trials to obtain a single consistent quantitative synthesis. The key assumption underlying network meta-analysis is exchangeability of the trials or the consistency between the trials. If there is a violation of exchangeability (inconsistency exists between trials), the result of a network meta-analysis might be unreliable. Also, application of the consistency model blindly without inconsistency analysis will force the results to be consistent, even when the data are not, which could lead to incorrect conclusions. Generally, there are 2 major ways to assess whether the trials in the network are indeed consistent.
The first way is to assess the consistency statistically using inconsistency2 or node-splitting models.3 However, inconsistency can only be detected when both direct and indirect evidence are available for a comparison. In this network meta-analysis, there is no direct comparison between PCI and medical therapy groups, which makes it impossible to assess the inconsistency statistically.
The second way to ensure valid results is the careful selection of trials for inclusion in the analysis. Bittl et al included both randomized controlled trials and cohort studies in the meta-analysis. It is hard to believe there is no inconsistency between studies with different designs. Furthermore, as the authors mentioned, significant heterogeneity in baseline characteristics was identified among the trials in the network meta-analysis, including age, sex difference, and different medical treatment in medical therapy group. Although meta- regression was conducted to assess the confounding effect of theses covariates, it is definitely underpowered because only few studies were included and some of covariates could not be adjusted appropriately in a study-level meta-analysis.
As interventional cardiologists, we believe that PCI might be better than medical therapy in patients with unprotected left main coronary artery disease. The point we want to emphasize is the validity of network meta-analysis relies on the consistency between trials, and drawing conclusions from network meta-analysis without consistency is inappropriate and may be misleading. Result of network meta-analysis should be interpreted with caution, especially in clinical guideline development.
Yicong Ye, MD
Shuyang Zhang, MD
Department of Cardiology
Peking Union Medical College Hospital
Peking Union Medical College & Chinese Academy of
- © 2014 American Heart Association, Inc.
- Bittl JA,
- He Y,
- Jacobs AK,
- Yancy CW,
- Normand SL