The meta-analysis by Singh and colleagues tries to quantify the risk of serious adverse cardiovascular events with varenicline.1 Many representations of this risk have been disseminated, including a large increase (number needed to harm [NNH] 28) extrapolated from applying the odds ratio from an overall meta-analysis to a high-risk group and a small increase (NNH 417) extrapolated from using the absolute numbers (pooled numerators and denominators) from the overall meta-analysis.
The method of interpreting the data can give widely discrepant results. Simple pooling of absolute numbers is inaccurate because it does not account for the variations in the distribution of patients between the varenicline and placebo groups, so results are more affected by how the populations of different trials were distributed rather than the varenicline–placebo difference. However, with the formal meta-analysis, it is not clear that a single trial with a high rate of cardiovascular adverse events (about 6% in a trial of patients with cardiovascular disease) and 13 trials with very low rates of cardiovascular events should be combined in a single meta-analysis.
To estimate the risk of serious adverse cardiovascular events in patients with stable cardiovascular disease, the best estimate may come from an analysis that is based on patients with cardiovascular disease at baseline from across these trials. Until that can be accomplished, the best estimate may be from a single trial in this population,2 in which the rates of cardiovascular adverse events were 7.04% with varenicline and 5.57% with placebo. The NNH would be 68, but it was not statistically significant.
A meta-analysis of the other 13 trials found a pooled cardiovascular event rate of 0.593% with varenicline and 0.237% with placebo. The Peto odds ratio was 2.54 with a 95% confidence interval (CI) of 1.26 to 5.12. This translates into an NNH range of 103 to 1627 across the 95% CI, using a control event rate of 0.237%.
The risk of serious neuropsychiatric symptoms may be of more concern, with hundreds of reported instances, including 272 completed suicides,3 but comparative evidence is limited, and no differences were found when compared with other smoking cessation medications.4
Of course, risks must be weighed against benefits, with number needed to treat (NNT) of nine for continuous smoking cessation at one year in the trial with the higher cardiovascular event rate2 and an NNT of six to nine at 24 weeks in a Cochrane review of 10 trials.5