In a recent meta-analysis, Lindholm, Carlberg and Samuelsson claim that, in patients with hypertension, the β-adrenergic blocker atenolol is no different from other β-adrenergic blockers in terms of effects on hard endpoints 1. This assertion brings to mind the HAPPHY trial 2 in which patients on atenolol showed higher death rates and patients on metoprolol lower death rates than did patients taking diuretics. The difference between the two outcomes was not statistically significant. For other endpoints, only pooled results were released from that study.

We combined the information provided by the two sets of data, thereby obtaining stronger statistical power. Eight of the studies in the analysis compared atenolol and non-atenolol β-blockers, respectively, with other antihypertensive drugs and six compared atenolol and non-atenolol β-blockers, respectively, with placebo. We used a previously described Bayesian procedure 6 and first combined the results of studies in similar treatment groups, assuming that the relative effects of the treatment in the various studies were comparable, even though the absolute risks might differ. This is a natural assumption to make in this context as the basis of the Bayesian analysis in which results from different studies are combined. Generally, it is difficult to verify the validity of this assumption against data. However, from the data in the published meta-analysis 1 the assumption seems reasonable, except for the MRC Old comparison of atenolol with other antihypertensive drugs. Hence, calculations were also carried through without this comparison.

We then calculated the relative risk of non-atenolol β-blockers against atenolol by dividing the corresponding risk ratios. Finally, based on the 8 and 6 studies respectively, we combined the two estimates of the risk ratios for myocardial infarction in patients treated with non-atenolol β-blockers versus atenolol by simply updating one with the other by means of Bayes' formula.

Table 1 shows our findings, indicating that the frequency of myocardial infarctions is 13 % lower in patients on a non-atenolol β-blocker compared with patients on atenolol based on all 14 studies. This outcome was to be expected, since it agrees with previously published observations 2, which provide justification for relying on the 90 % credibility interval. This interval 0.750 – 0.992 has an upper border slightly less than unity. The 95 % credibility interval was 0.727 – 1.023. The probability of at least 10% lower risk (risk ratio ≤ 0.90), which could be considered to be of clinical interest, was 0.689. The results are slightly weaker when the MRC Old study is excluded (Table 1).

The present additional statistical analysis of the published data 1 thus shows a different outcome for atenolol in relation to the non-atenolol β-blockers, in disfavour of atenolol based on all 14 studies. Our finding also agrees with statements by Carlberg, Samuelsson and Lindholm in their earlier meta-analysis 7. One of the implications of this different outcome is that atenolol is not representative for the β-adrenergic blocker class of drugs as a whole. Observations relating to atenolol do not necessarily apply to other β-adrenergic blockers. A similar reservation has also been made recently by others 8. Therefore we do not agree with an editorial on the publication of the ASCOT trial and the meta-analysis, which advocates that the current guidelines for the treatment of hypertension should be changed in disfavour of all β-adrenergic blockers 9.

Claims have been made that β-adrenergic blockers are inferior drugs when used to treat patients with hypertension. We suggest that a recent meta-analysis shows that these claims are correct only in the case of atenolol. We have substantiated this suggestion by re-analysing the published data using a Bayesian technique. This re-analysis showed 13 % fewer myocardial infarctions among patients treated with non-atenolol β-adrenergic blockers than among patients treated with atenolol. We argue that non-atenolol β-adrenergic blockers should continue to be fundamental in the treatment of hypertension.

The authors declare that they have no competing interests except that JBO has received lecture fees from 'Orion' and 'Merck'.

IA and JBO, presented the problem and drafted the manuscript along with BN, who suggested the statistical solution based on earlier work by some of the present authors 6; IFT did the computations and took part in the planning along with JG. All authors read and approved the final manuscript.