Varenicline: quantifying the risk

In their sophisticated review and meta-analysis, Singh and colleagues calculated the absolute risk for varenicline as 1.06% and that for placebo at 0.82%.¹ Hence, the correct absolute risk elevation (ARE) is the difference between the two percentages (0.24%, or 0.0024), which they also calculated correctly.

However, in their interpretation, Singh and colleagues chose to use pooled data in a way that was inconsistent with how NNH (number needed to harm) should have been calculated. They used a baseline cardiovascular risk of 5.57% — not the rate of cardiovascular events in the placebo groups of their meta-analysis — as the comparison group with varenicline. Doing so created problems with their subsequent analysis, interpretation and conclusion.

The problem was in part caused by their having combined their data with other data before coming to a conclusion. They used ARE as a summary statistic, then expressed it incorrectly. This highlights why NNH, as a summary statistic itself, is inferior to ARE, as Woods and Caswell have implied in the previous letter.²

On the basis of an article by Hutton,³ I believe that the interpretation of NNH would be better stated as the average number of comparable patients, which, if they received varenicline rather than placebo for the same period of time, would result in one patient being harmed who otherwise would not have been harmed. Since NNH is the inverse of the ARE, its correct value is 1/0.0024, or 417 patients, not 28 patients, as Singh and colleagues propose. By choosing to use the latter figure as the correct NNH, they have greatly overestimated the risk produced by varenicline. The failure to use 417 from their own data as the correct NNH contributes to a flawed analysis, interpretation and, probably, conclusion.

In the data Singh and colleagues cite from an FDA (US Food and Drug Administration) licensure submission,⁴ they report that the risk with varenicline was 2.32 per 100 patient years of exposure and with placebo, 1.63 per patient years of exposure. However, the authors stop at this point and do not calculate the ARE of these data (i.e., the difference between these two incidence rates, which equals 0.69 patients per 100 patient years of exposure to varenicline, which again, like the NNH of 417, is an exceedingly small number).