- © 2007 Canadian Medical Association or its licensors
[The authors respond:]
Falk Hoffman asserts that the conclusions in our meta-analysis were incorrect, as the random-effects confidence intervals for Figs. 1A and 2B appeared to include the value 1 when the analyses were repeated with Review Manager software. However, unlike Hoffman, we conducted our random-effects analyses using a software program that does not round off at 2 decimal places. Thus, in both random- and fixed-effects modelling, the confidence intervals approached but did not incorporate 1. The summary effect estimate for the comparison in Fig. 1A (β-blocker v. placebo in trials enrolling younger patients) with the random-effects model is 0.862 (95% CI 0.746–0.996; p = 0.044); This result (for which the 95% CI in our original publication was truncated, rather than rounded, to 0.74–0.99) clearly supports the interpretation stated in our paper that β-blockers are more efficacious than placebo in younger patients. Similarly, the summary effect estimate for the comparison in Fig. 2B (β-blocker v. other antihypertensives in trials enrolling older patients) with the random-effects model is 1.066 (95% CI 1.001–1.135, incorrectly reported in our original publication as 1.06 with 95% CI 1.01–1.10; p = 0.047); this result supports our interpretation that β-blockers were less efficacious than other antihypertensive agents in older patients. Unfortunately, this estimate and 95% CI were incorrectly reported in our paper; in rounded form, these values should have been reported as 1.07 (95% CI 1.01–1.10).1
Andrea Siebenhofer and colleagues note that our definition of the composite outcome was unclear. The initial version of our meta-analysis was substantially longer than the final published version and thus contained far more detail on trial entry criteria, baseline characteristics and outcome definitions. For publication purposes, we were asked to shorten the manuscript, and we regret that in doing so we inadvertently caused confusion for some readers. To clarify, our primary analyses were for the composite outcome of cardiovascular deaths (or all-cause deaths where cardiovascular deaths were not reported), MI or stroke. In light of the concerns of Siebenhofer and colleagues about the 12 trials included in Figs. 2A and 2B of our meta-analysis, it is important to point out that the composite outcomes incorporated cardiovascular mortality in 9 of those trials, total mortality in 2 and cardiac mortality in 1. Ten of the 12 trials incorporated fatal and nonfatal MI and stroke in their composite outcomes (one trial included only fatal and nonfatal MI and another included only fatal MI and stroke). We did not have access to the primary study data and therefore relied on the end-point definitions and methods of classification used by the primary trialists (recognizing that this was consistent within trials and thus would be identical across treatment groups within each trial).
In their table, Siebenhofer and colleagues report a different number of events in the composite outcome for the CAPPP trial than we did. Although we defined the composite event rates as the proportion of patients experiencing any of the composite events (and thus counted each participant only once), Sienbenhofer and colleagues appear to have pooled the number of events for the individual end points separately to arrive at their composite event rates. However, this individual counting of events assumes that they are independent and that no patients suffered more than one event; this is clearly not the case and we therefore favour our approach (which was that taken by the CAPPP authors in their primary publication2).
Siebenhofer and colleagues and Bo Carlberg questioned our inclusion of data from CAPPP and STOP2 (mixed β-blocker studies). We discussed the reasons for including these trials in our paper,1 where we also reported 2 sensitivity analyses that address these concerns.
Finally, Carlberg points out that the meta-analysis that he and his colleagues conducted differed from ours in that they examined the effect of β-blocker treatment on the incidence of MI, stroke or death separately and based their conclusions on the excess risk of stroke observed in patients treated with β-blockers. On the other hand, we focused on the composite end point of all 3 conditions together to account for competing risks and the potential for survivor bias. As discussed in our paper, an agent with a beneficial effect on one end point may appear to have a detrimental effect on another end point, given that treated patients survive without the first end point for long enough that the second end point occurs.