Messori et al analyzed the raw data from our study using an alternative software titled trial sequential analysis (TSA; Copenhagen Trial Unit, Copenhagen, Denmark) and observed different results compared with our original meta-analysis in which the Comprehensive Meta-Analysis software, version 2 (Biostat, Englewood, New Jersey), was used. Based on their analysis, “the increased risk of myocardial infarction (MI) from oral direct thrombin inhibitors (DTIs) did not seem to be conclusively demonstrated. And the similar rates of MI observed between warfarin and nonwarfarin anticoagulants not only indicated no proof of difference but also demonstrated the proof of no difference.”
In response, we are relieved that the result of the second meta-analysis confirmed the lack of protective effect of warfarin against MI compared with nonwarfarin anticoagulants with the use of TSA software.
As for the primary meta-analysis, it is important to understand that we are dealing with a meta-analysis for the purpose of hypothesis generation using trials in which MI was not the primary end point. We therefore did not declare the data on the association of DTIs and MI conclusive as such a statement cannot only be based on a meta-analysis, and should be further examined in a randomized trial with MI as the primary end point. This study was inspired by the observations from several randomized trials showing increased rates of MI associated with the use of DTIs including the Randomized Evaluation of Long-Term Anticoagulation Therapy (RE-LY) with Dabigatran versus Warfarin in Patients with Atrial Fibrillation (RE-LY) and Thrombin Inhibitor in Venous Thromboembolism (THRIVE) trials. Besides the clinical observation from these randomized trials, there seems to be additional laboratory evidence that may support a hypothesis that certain mechanisms related to the direct thrombin inhibition may be the cause of these thromboembolic events, particularly at low concentrations of the direct thrombin inhibitor.
Messori et al raise an important point regarding the methodology of meta-analysis in general. The TSA methodology is based on the principle that when one is taking multiple peeks at the data, or rerunning a meta-analysis every time a new study comes out, there should be an adjustment of the parameters. Furthermore, a sample size is defined before proceeding with the meta-analysis to consider the study statistically valid.
Their critique, however, is not only directed toward our results but also toward most meta-analysis studies published in the contemporary medical literature. Very few of those studies would pass optimum sample size requirement set by TSA. TSA methodology provides a very conservative approach toward meta-analysis. For association of DTIs and MI, Messori et al calculated the optimal sample size of more than 185,000 subjects to achieve a power of only 80% compared with almost 40,000 subjects that were included in our original analysis. It is obvious that using a mammoth sample size would reduce the risk of false-positive results, but this comes at the price of potentially delaying a possible implementation of an intervention toward harmful effects of a drug for several years. A counter argument can be made when dealing with rare harmful cardiovascular events associated with commonly used drugs. In those circumstances, a more aggressive approach may be preferred such as application of fixed-effect model even in the presence of significant heterogeneity of the data may be necessary to prevent experiences such as the late withdrawal of rofecoxib from the market due to significantly increased risk of cardiovascular events.
Our meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines that is most commonly used by investigators for the performance and publication of meta-analysis. The conditions for application of the fixed-effect model in the primary meta-analysis were met. The conclusion reached regarding association of DTIs and coronary events is in concordance with other meta-analysis studies of association of dabigatran and risk of coronary events. Our study extends this observation beyond dabigatran and opens the question of possible problems with univalent direct thrombin inhibition. This information may be helpful for a more critical view of novel anticoagulants for clinicians who deal with daily patient care and investigators who plan to design future studies with this class of agents.