To the Editor:—
In the context of the current debate around the deleterious effects of the cyclooxygenase-2 inhibitors in patients with increased risks for cardiovascular diseases, we read with great interest the study published by Nussmeier et al. 1
The authors have studied, as intended, a population that was certainly at low risk for cardiovascular disease, and, indeed, the number of cardiovascular events was small and identical in both groups (five events per group [table 3 from the original article1]).
In the discussion section, the authors highlight the fact that the size of the studied groups is too small to show any statistically significant difference in the number of cardiovascular events, because the study was only powered to show a difference in a combined multisystem endpoint.
Nevertheless, we believe that readers should not be left with an impression of possibly false security and conclude that the administration of a cyclooxygenase-2 inhibitor in a low-risk population does not increase the occurrence of cardiovascular events. This has simply not been demonstrated in this study, partly because of two methodologic problems that should be addressed and have not been discussed in the article.
First, the analyses were performed on a modified intent-to-treat population, were the patients who did not receive the treatment were excluded from the analysis after randomization (four patients in the placebo group and eight patients in the parecoxib–valdecoxib group). The creation of a “modified intent-to-treat population” could lead to the destruction of the similarity of the groups after randomization. This problem is further increased by the fact that the reasons for not administering the treatment are not given in the article (as per the CONSORT statement2) and that twice as many patients were excluded from the parecoxib–valdecoxib group as from the placebo group.
In both groups, patients were lost to follow-up (five in the parecoxib–valdecoxib group and two in the placebo group). Although this is never clearly stated in the article, the usual definition of lost to follow-up implies that the primary endpoint (the combined incidence of postrandomization adverse events) is not known.
We computed the maximal possible bias to judge the robustness of the statistical results. In this method, all of the missing patients (eight excluded from analysis and five lost to follow-up) in the parecoxib–valdecoxib group are considered to have presented a cardiovascular event, in contrast to none of four excluded and two lost to follow-up in the placebo group. This is done with inclusion of all of the randomized patients. The computation of the risk ratio for presenting a cardiovascular event is given in table 1.
In contrast, when none of the 13 patients in the parecoxib–valdecoxib group are considered to have presented a cardiovascular event, but the 6 patients in the placebo group are considered to have presented a cardiovascular event, the results are given in table 2.
We concentrated this analysis on the cardiovascular adverse events because they are currently the most feared and highlighted adverse events of the cyclooxygenase-2 inhibitors.
One can see from these results that, with the data from the article, it is impossible to eliminate, from a statistical point of view, that the parecoxib–valdecoxib combination does increase the frequency of cardiovascular events in a low-risk population. But one can easily eliminate, with a high probability, that the placebo administration increased the frequency of cardiovascular events.
This should be clearly stated in the discussion section, where we have the impression that more emphasis is given to the demonstration of analgesic efficacy, which was a secondary endpoint, than to the lack of demonstration within a certain statistical probability of the primary endpoint, which was clearly stated to be the safety.
The second problem of this article is the marginal statistical power of the study. The authors state that “The sample size of 500 patients per treatment arm provided at least 80% power to detect a doubling of the 4% estimated background incidence of all predefined adverse events combined.” In fact, to provide this sort of power, you need at least 550 patients per treatment with the primary endpoint available. With the available data in the study, calculated on 525 patients per arm, you have, at best, a power of 74%, which leaves 26% of chance (or more exactly bad luck) to face a type II error of not being able to detect a difference of 4% that would really exist. This lack of power is further aggravated by the fact that the true incidence of the combined adverse event was even lower than expected, and demonstrating a doubling of the 3.2% event rate in the placebo group would necessitate 697 patients per treatment to maintain an 80% power. All of these power calculations are true if we suppose that the authors accepted a 5% risk of type I error, which in fact was never stated in the article.
The conclusion about the cardiovascular safety in a low-risk population, even after this study, is that we do not know with any certainty. And we should not be afraid to say so.
*Cliniques Universitaires de Bruxelles Hôpital Erasme, Brussels, Belgium. email@example.com