Mega-trials and management of acute myocardial infarction

• Published in: The Lancet (British edition) Vol. 346; no. 8975; p. 611
• Main Author: Woods, K L
• Format: Journal Article
• Language: English
• Published: England 02-09-1995
• Subjects: Angiotensin-Converting Enzyme Inhibitors - therapeutic use; Bias; Humans; Magnesium - therapeutic use; Meta-Analysis as Topic; Multicenter Studies as Topic; Myocardial Infarction - drug therapy; Myocardial Infarction - mortality; Nitrates - therapeutic use; Randomized Controlled Trials as Topic - methods
• ISSN: 0140-6736
• DOI: 10.1016/S0140-6736(95)91440-4

Clinical management of acute myocardial infarction has been strongly influenced by large, simple trials (mega-trials) with unrestrictive protocols and limited data collection. The design has been adopted to increase statistical power to a maximum. Its validity rests on an effective randomisation procedure and intention-to-treat analysis of deaths. Experience has shown that mega-trials tend to generate effect-estimates nearer the null than those from conventional trials or meta-analyses. When a small or absent observed treatment effect (or subgroup effect) in a mega-trial contrasts with the results of conventionally designed trials, it is necessary to assess both null bias and failure to increase the true treatment effect to a maximum in the mega-trial. Null bias will arise when the contrast between treatment and no-treatment, or between subgroups, is blunted either by non-protocol therapy or by inaccuracy of data, including misclassification between subgroups. Each is more likely with an unrestrictive design. To increase the true treatment effect to a maximum, trial conditions must be specified with insight into mechanism, dose-dependence, and time-dependence. The mega-trial design is therefore unsuited to an exploratory role. These issues are illustrated by the examples of nitrates, angiotensin-converting-enzyme inhibitors, and magnesium in acute myocardial infarction but have general relevance to the validity and generalisability of simple trials.