Magnesium in acute myocardial infarctionBMJ 1995; 310 doi: https://doi.org/10.1136/bmj.310.6982.751 (Published 25 March 1995) Cite this as: BMJ 1995;310:751
- Salim Yusuf,
- Marcus Flather
- Director Senior research fellow Division of Cardiology, McMaster University, Hamilton, Ontario, Canada L8L 2X2
ISIS 4 provides no grounds for its routine use
Clinical investigators have long pursued the elusive dream of finding a cheap, safe treatment that reduces mortality and morbidity in a common condition. Many investigators had hoped that magnesium given intravenously to patients with acute myocardial infarction might be such a treatment. Were they justified?
Small trials reporting the use of magnesium in acute myocardial infarction have been reported intermittently for 20 years. The rationale for these studies came partly from observations of differences in heart attack rates associated with geographical variations in magnesium in the water supply and partly from laboratory studies showing that magnesium had cardioprotective effects during ischaemia and that myocardial magnesium concentrations were relatively low during acute ischaemia.1
Research on animals has shown that magnesium is a peripheral and coronary vasodilator, has antiarrhythmic effects, decreases reperfusion injury, and has antiplatelet effects in some species.1 The clinical importance of these findings is, however, still uncertain. Experimental studies often do not have the same rigorous design, conduct, and analysis that are now expected from clinical research. Few experimental studies are randomised or blinded. Furthermore, publication bias is likely, with positive results being more likely to be reported than negative ones—especially when the subject is a relatively new hypothesis.
An informal review of the results of the early clinical trials in acute myocardial infarction indicated a trend towards a lower mortality with intravenous magnesium, with this difference being statistically significant in only one trial.2 That trial also found a reduction in arrhythmias. These impressions were subsequently confirmed by a formal metaanalysis.3 The data came, however, from only 1300 patients with a total of 78 deaths; the analysis found a 55% reduction in the odds of death (P=0.001), with 95% confidence intervals ranging from about 30% to about 65%.
The authors recognised at least three limitations in their work. Firstly, the number of events on which the metaanalysis was based was so small that, even if one or two trials with unpromising results had been overlooked or not published, the results could have been substantially different. Secondly, few of the trials were in patients given treatment with aspirin or thrombolytic drugs, making the applicability of the results to the current era questionable. Thirdly, the large treatment effects apparent from the informal perusal of the data would have influenced the authors to invest the necessary effort to conduct the meta-analysis. This retrospective selectivity of meta-analysis is a potential source of error, as is the early termination of trials because of unexpectedly large treatment effects. The danger is that the observed benefits that trigger the analysis are quite likely to have been exaggerated by chance.
Investigators in Leicester had initially conducted a small trial (LIMIT: Leicester intravenous magnesium intervention trial) on around 200 patients; the results (included in the meta-analysis) indicated a trend towards fewer deaths.4 This study encouraged the design and organisation of the LIMIT-2 study, in which a total of 2316 patients were randomised to receive either an intravenous bolus of 8 mmol magnesium sulphate injected over five minutes followed by 65 mmol of constant infusion over 24 hours or a placebo infusion.5 There were 90 deaths in the magnesium group and 118 in the placebo group (a 24% relative risk reduction in mortality (95% confidence interval 1% to 43%), P=0.04) at 28 days' follow up, and these differences persisted in the follow up results (21% relative reduction in cardiovascular mortality (5% to 35%), P=0.01).6 Around one third of the patients in LIMIT-2 were given thrombolytic treatment and two thirds aspirin. Thrombolytic treatment and magnesium were generally started within one hour of each other.
In the earlier trials included in the meta-analysis there was a trend towards fewer arrhythmias and no apparent effect on outcomes related to heart failure; in LIMIT-2 little effect was seen on arrhythmic events, but the incidence of left ventricular failure was reduced in the treatment group. Those who conducted LIMIT-2 believed that it largely confirmed the results of the meta-analysis, and proponents of magnesium expected that a more definitive beneficial effect would be shown in the large ISIS 4 trial (fourth international study of infarct survival) with about 58000 patients.7 This optimism ignored the difference between the results of the metaanalysis and those of LIMIT-2 in that the main impact seemed to be on arrhythmias in the meta-analysis but on heart failure in LIMIT-2.
The results of ISIS 4 have just been published, and the findings related to magnesium are very disappointing.7 At 35 days of follow up there had been 2216 deaths among the 29011 patients who received magnesium compared with 2103 among the 29039 patients who received standard treatment (7.64% v 7.24%; odds ratio 1.06 (1.00 to 1.12), P=0.07). This lack of benefit was consistent across all major subgroups, including patients treated very early or those treated later and those receiving or not receiving concomitant thrombolytic treatment. The trend towards increased mortality was perhaps due to the excess incidence of cardiogenic shock and heart failure in the magnesium group, which is in direct contrast to both the results and the suggested mechanism of benefit observed in LIMIT-2. In contrast to the findings in LIMIT-2, but consistent with the results of older trials and experimental data, there was a significant reduction in early ventricular fibrillation with magnesium in ISIS 4.
When the unexpected results of ISIS 4 were reported at a meeting of the American Heart Association in November 1993 critics of meta-analysis pointed to the lack of consistency among the results of the meta-analysis, LIMIT-2, and ISIS-4. Supporters of magnesium questioned whether the large simple trial itself was unreliable: perhaps the patients enrolled were those unlikely to benefit from treatment with magnesium or perhaps the timing of the magnesium regimen was inappropriate. All the comments and analyses will need careful examination.8
The lack of consistency between ISIS 4 and the meta-analysis of the small trials does not necessarily invalidate the technique. Firstly, the meta-analysis was based on a relatively small amount of data (only 78 deaths), and despite the extreme P value (0.001) the result may not necessarily have been “robust.” The results of meta-analyses are subject to a number of biases that are not readily quantifiable. Traditionally, meta-analysis has been carried out on topics where the results are likely to be interesting. Indeed the decision to perform and publish a formal meta-analysis may itself be data derived. Although the LIMIT-2 trial seemed to confirm the results of the meta-analysis, its own results were only just statistically significant (P=0.04) with wide confidence intervals.
One explanation is that delays in giving magnesium after thrombolytic treatment could have confounded the results of ISIS 4. While it is true that the experimental treatments were started after proved treatments (aspirin, ß blockers, and thrombolytic drugs) had been given, further analysis of ISIS 4 indicates that there was no benefit even among the 10252 patients randomised within three hours of the onset of symptoms. In these patients recanalisation of the coronary artery and the infusion of magnesium would probably have been almost simultaneous. Furthermore, in LIMIT-2 the benefits of magnesium were similar in those receiving and those not receiving thrombolytic treatment. In ISIS 4 there was no evidence of benefit of magnesium in either of these groups. No coherent explanation has been offered for the differing results.
What are the lessons to be learnt? Firstly, a meta-analysis of small trials is not a replacement for large, carefully conducted trials. Secondly, since most treatments produce either no effect or at least only moderate effects on major outcomes such as mortality, investigators should be sceptical if the results obtained deviate substantially from this expectation (“too good to be true”). Thirdly, definitive trials should demand levels of evidence that are statistically more reliable, with the lower confidence limits of the risk reductions representing a clinically worthwhile difference.
Until further research evidence is presented there are no grounds for the routine use of magnesium for patients with acute myocardial infarction.