Intended for healthcare professionals


Article makes simple errors and could cause unnecessary deaths

BMJ 2002; 324 doi: (Published 19 January 2002) Cite this as: BMJ 2002;324:167
  1. Colin Baigent, MRC scientist (colin.baigent{at},
  2. Rory Collins, professor of medicine and epidemiology,
  3. Richard Peto, professor of medical statistics and epidemiology
  1. Radcliffe Infirmary, Oxford OX2 6HE

    EDITOR—The worldwide meta-analysis of antiplatelet trials shows that low dose aspirin (or some other effective antiplatelet regimen) reduces non-fatal myocardial infarction, non-fatal stroke, and vascular death in a wide range of patients who are at high risk of occlusive vascular disease.1 A paper disputing this was published concurrently in the For Debate section of the journal,2 but the arguments in it (some of which the author also published on the same date in an editorial in the Lancet)3 depend strongly on quite simple mistakes about the randomised evidence and could cause unnecessary deaths.

    Consider, for example, the ISIS-2 trial of short term antiplatelet therapy, in which 17 187 patients with suspected acute myocardial infarction were randomised, half to active aspirin and half to placebo.4 This trial showed a clear reduction in five week all cause mortality (811/8587 (9.4%) aspirin v 1030/8600 (12.0%) placebo deaths, 2P<0.00001).5 Bizarrely, in a section entitled “Trials do not show that aspirin saves lives,” the For Debate paper attempts to dismiss the ISIS-2 findings by suggesting that “all patients lost to follow up in the active group should be considered to have died and none of those in the control arm. Such an analysis would neutralise the benefit observed in one of the few seemingly convincingly positive studies of aspirin, the ISIS-2 trial.”2

    This is not even arithmetically correct, and such a statement should not be part of any serious debate in the BMJ. The five week follow up was 97% complete when this trial was first reported,4 and 99% complete when further follow up was reported in the BMJ.5 This slightly greater completeness yielded, in fact, only 13 extra deaths (6 in the aspirin group, 7 in the placebo), and the even slighter incompleteness that remains cannot, of course, be of any material relevance (especially since most of the few still untraced at five weeks are known to have been discharged alive from hospital: for only 0.2% of the patients given aspirin and 0.2% of those given placebo is there no follow up at all).

    Likewise, among about 20 000 patients in the 12 trials of long term (mean two years) antiplatelet therapy among patients with a history of previous myocardial infarction, the odds of having a non-fatal reinfarction were reduced by 30% (SE 6; 2P<0.00001), with no significant heterogeneity between the results in different studies.1 The For Debate paper purports to account for this 30% reduction by suggesting (without good evidence of any such effect) that the proportion of non-fatal infarctions that would be reported might be 70% with aspirin and 75% without. Again, however, this argument is arithmetically wrong, for 70 v 75 would represent a reduction of only 7%, not 30%.

    Furthermore, having suggested earlier that it is only analyses of all cause mortality that can be trusted, the paper then goes on to elaborate a curious theory that involves trusting the somewhat arbitrary distinction between mortality attributed to sudden death and to other cardiac causes. From this it eventually concludes that aspirin could be producing “an increased risk of sudden death among concealed, and therefore untreated, events.”2 But, there is no good evidence that this is true.

    More importantly, in the worldwide meta-analysis, vascular mortality—which is highly significantly reduced—already included both sudden death and death from unknown causes (as well as death from any type of stroke).1 In the 12 trials of long term antiplatelet therapy during the years after myocardial infarction the reduction in vascular mortality was 15% (SE 5; 2P=0.002) again with no significant heterogeneity between the effects in different antiplatelet trials (or 17%; 2P<0.0010), with even less heterogeneity, if the imbalance in prognostic features in the AMIS trial is appropriately allowed for).6 Moreover, all cause mortality was also reduced, as there was no significant excess of non-vascular deaths in this category of patient, or in any of the other four main categories of patients at high risk. Indeed, taking the 135 000 patients in all five categories together, non-vascular mortality was 1.1% with antiplatelet therapy and 1.2% without, which looks pretty safe. Thus, there is no good evidence from these trials that non-vascular mortality offsets the highly significant reduction in vascular death or in non-fatal myocardial infarction or stroke among high risk patients.

    A recent study of the costs of the secondary prevention of such vascular events by aspirin is cited in the For Debate paper as concluding that the cost per event prevented would be over £3000. If true, this could be money well spent, but it is included in a section misleadingly entitled “Neither safe nor cheap.”2 (No other cost estimates in that section are relevant to secondary prevention.) The author also suggests that “the greatest potential detriment of aspirin on health care, however, is that it diverts attention away from treatments that are of unequivocal benefit.” No good evidence for this assertion is provided and, moreover, there is no good reason why other effective treatments (such as angiotensin converting enzyme inhibitors, β blockers, and statins) should not be used in addition to aspirin, conferring additional benefit.7

    There are several other errors of judgment, partly from failure to understand the proper role of meta-analysis in the interpretation of randomised evidence. Given this, none of the substantive points in the For Debate article is of material relevance (except, perhaps, as a warning about the power of prejudice), and the chief ones have been dealt with adequately in the current or previous antiplatelet reports. In retrospect, it would perhaps have been better for the BMJ to have sought review of the paper from, among others, those whose work it criticises. This would have given the journal the opportunity to avoid publication of arguments and conclusions that are wrong for trivial reasons and potentially damaging to patients.


    1. 1.
    2. 2.
    3. 3.
    4. 4.
    5. 5.
    6. 6.
    7. 7.