We largely agree with Blankfield1 that the known elevation of blood pressure seen with varenicline could be one explanation for the increase in serious adverse cardiovascular events.2 Other possible mechanisms include the vasoconstrictive effects of varenicline because it is a nicotinic acetylcholine receptor agonist.
Takagi and Umemoto argue about the choice of the appropriate method for pooling uncommon events.3 However, they fail to realize that RevMan automatically adds a 0.5 continuity correction to zero event studies, and this continuity correction biased their reported Mantel–Haenszel estimates toward the null, which is bordering on statistical significance (odds ratio [OR] 1.56, 95% confidence interval 0.99–2.44). They interpret the lack of statistical significance as proof of the cardiovascular safety of varenicline. We chose the Peto–OR estimate because it is the recommended approach for uncommon events, particularly when there are trials with zero events.4 Sensitivity analyses using the fixed Mantel–Haenszel approach with appropriate continuity corrections showed similar results with the software package Stats Direct.
Woods and Caswell argue about the choice of reporting relative versus absolute risks.5 There is scientific and regulatory consensus that uncommon events should be modelled with the relative approach. We reported both relative and absolute risks. Because relative risks are transmitted equally across populations, we applied the pooled OR from our meta-analysis to the baseline event rate among smokers with stable cardiovascular disease to estimate the annual number needed to harm (NNH) among this group. Although our estimated NNH of 28 is applicable only to smokers with stable cardiovascular disease, the meta-analytic OR estimates could be applied to the lower baseline risk among smokers without cardiovascular disease to generate their NNH, which is likely higher.
Squire questions our choice of a population with stable cardiovascular disease when computing the NNH and recalculates the NNH as a reciprocal of the proportion of events in each group.6 Such an approach treats the summary estimate as a single trial and fails to leverage the benefits of randomization.
We partly agree with Alper that the method of interpreting the data can give discrepant results.7 However, all these results demonstrate the presence of a cardiovascular risk with varenicline. We also agree that boxed warnings about the serious neuropsychiatric effects of varenicline such as depression, suicidal behaviour, suicidal ideation, hostility and aggression should be factored into a risk–benefit assessment. Data from randomized controlled trials would be ideal, but the proposed 12-week post-marketing safety trial to assess these serious risks of varenicline among smokers with psychiatric problems will not be completed until 2016, about 10 years after regulatory approval.8