Beta blockade during and after myocardial infarction: an overview of the randomized trials

Abstract

Long-term beta blockade for perhaps a year or so following discharge after an MI is now of proven value, and for many such patients mortality reductions of about 25% can be achieved. No important differences are clearly apparent among the benefits of different beta blockers, although some are more convenient than others (or have slightly fewer side effects), and it appears that those with appreciable intrinsic sympathomimetic activity may confer less benefit. If monitored, the side effects of long-term therapy are not a major problem, as when they occur they are easily reversible by changing the beta blocker or by discontinuation of treatment. By contrast, although very early IV short-term beta blockade can definitely limit infarct size, more reliable information about the effects of such treatment on mortality will not be available until a large trial (ISIS) reports later this year, with data on some thousands of patients entered within less than 4 hours of the onset of pain. Our aim has been not only to review the 65-odd randomized beta blocker trials but also to demonstrate that when many randomized trials have all applied one general approach to treatment, it is often not appropriate to base inference on individual trial results. Although there will usually be important differences from one trial to another (in eligibility, treatment, end-point assessment, and so on), physicians who wish to decide whether to adopt a particular treatment policy should try to make their decision in the light of an overview of all these related randomized trials and not just a few particular trial results. Although most trials are too small to be individually reliable, this defect of size may be rectified by an overview of many trials, as long as appropriate statistical methods are used. Fortunately, robust statistical methods exist--based on direct, unweighted summation of one O-E value from each trial--that are simple for physicians to use and understand yet provide full statistical sensitivity. These methods allow combination of information from different trials while avoiding the unjustified direct comparison of patients in one trial with patients in another. (Moreover, they can be extended of such data that there is no real need for the introduction of any more complex statistical methods that might be more difficult for physicians to trust.) Their robustness, sensitivity, and avoidance of unnecessary complexity make these particular methods an important tool in trial overviews.