Singh and colleagues found that varenicline was associated with a significantly increased risk of serious adverse cardiovascular events compared with placebo (1.06% [52/4908] in the group receiving varenicline v. 0.82% [27/3308] in that receiving placebo (odds ratio [OR] 1.72, 95% confidence interval [CI] 1.09–2.71).1 They used the Peto method to calculate ORs and 95% CI, stating that the Peto method provides the best CI coverage and is more powerful and relatively less biased than the random-effects analysis when dealing with low event rates.
The approximation used to calculate the log OR works well when the effects of intervention are small (i.e., ORs are close to 1.0), events are not particularly common and the studies have similar numbers in the experimental and control groups.2 As these criteria are not always fulfilled, the Peto method is not recommended as a default approach for analysis because it has been shown to give biased answers. On the other hand, when data are sparse (event rates are low or study size is small), Mantel–Haenszel methods have been shown to have better statistical properties.2
Sensitivity analyses by Singh and colleagues using the reciprocal of the treatment arm with a continuity correction (fixed Mantel–Haenszel OR 1.67, 95% CI 1.06–2.64) or without a continuity correction (fixed Mantel–Haenszel OR 1.77, 95% CI 1.09–2.88) showed results similar to those in the preliminary analysis using the Peto method. However, our recalculation of the same dataset as those in Figure 2 in Singh and colleagues’ article using RevMan showed a statistically insignificant increase in serious adverse cardiovascular events with varenicline compared with placebo (fixed Mantel–Haenszel OR 1.56, 95% CI 0.99–2.44).
This should be a proper result of the primary analysis in the meta-analysis. Therefore, there are no safety concerns about the potential for an increased risk of serious adverse cardiovascular events associated with the use of varenicline among tobacco users.