Serious methodologic issues limit the reliability and relevance of the results of the analysis by Singh and colleagues, in which they reported an “increased risk of serious adverse cardiovascular events” associated with varenicline based on a meta-analysis of 14 studies.1
First, the authors did not use a composite endpoint that is typically used in the medical literature to evaluate cardiovascular safety by, for example, including arrhythmias like atrial fibrillation. Such arrhythmias are not included in conventional definitions of major cardiovascular events because they generally are not associated with hemodynamic instability. The inclusion of atrial arrhythmias allowed Singh and colleagues to add seven events to the varenicline group but only one to the placebo group.
Second, the authors selected cardiovascular events in a manner that was inconsistent with their own methods. For example, they stated that they would include only patients with unstable angina, but they included patients who had any report of angina. Ischemic peripheral vascular events were not part of their defined endpoint, yet in their counts for the Rigotti study,2 Singh and colleagues included such events. They also planned to include only events reported during the double-blind period of each trial, but they included a number of events that occurred during the follow-up phase when the patients were no longer receiving drug treatment.
Third, Singh and colleagues failed to address potential statistical biases in their analysis. For example, the rate of patients lost to follow-up in most studies was greater in the placebo arm than in the varenicline arm. Furthermore, the authors excluded zero-event trials. Although this does not affect the relative risk, doing so distorts risk estimates, especially when the trials with zero events were designed to allocate more patients to varenicline than to placebo (e.g., 2:1 or 3:1 randomization).
In a July 2011 press release, the European Medicines Agency also noted “a number of limitations” in Singh and colleagues’ meta-analysis, including “the low number of events seen, the types of events counted, the higher drop-out rates in people receiving placebo, the lack of information on the timing of events, and the exclusion of studies in which no-one had an event.”3
Pfizer is working with the US Food and Drug Administration to conduct a meta-analysis that will address many of the methodologic deficiences in the meta-analysis by Singh and colleagues. This analysis will ensure that investigator-reported events are adjudicated by a committee of cardiovascular experts who are blinded to treatment to ensure correct diagnosis. We will publicly disclose the results when the analysis is completed.