Chronic aspirin is effective—if data are massaged sufficiently

Editor—Baigent et al argue passionately for the effectiveness of aspirin but also warn against the power of prejudice.¹ Perhaps there is more than one reasonable perspective on the aspirin meta-analyses, which they should have the courage to admit.

So to arithmetic—250 patients were reported lost to follow up on aspirin in the original ISIS-2 report, more than the 216 deaths' difference between aspirin and placebo.² This was subsequently revised.³ Which figure should we believe? However, my real concern is about the long term efficacy and risks of aspirin. Should aspirin be a short term intervention after a vascular event or a long term agent for cardiovascular prophylaxis?

Baigent et al have trouble in accepting the fact that the long term post-infarction studies failed to show a reduction in mortality, individually or overall.⁴ The reason the meta-analysis shows any reduction in long term mortality with anti-platelet agents after myocardial infarction is due to retrospective, unblinded, re-analyses of studies long completed.⁴ This exercise identified new deaths and also resurrections. We know neither how bias during this exercise was prevented nor why these newly acquired data were much more in favour of anti-platelet agents. Publication bias in favour of aspirin also exists.⁵ Considering the small treatment effect, bias could easily explain all the purported long term effects of aspirin. With enough post hoc revision many other failed medicines might also be shown to be effective.

All cause mortality tells us if treatment is effective; mode of death helps explain why a treatment is or is not effective. An increase in sudden death with aspirin was reported in the original post-infarction trials. It is curious that this fact “disappeared” in the meta-reanalyses. Baigent et al suggest that a modest concealment of non-fatal infarction could not account for the effects of aspirin. Perhaps such an effect is not subtle? Perhaps aspirin reduces the chance of a non-fatal myocardial infarction being reported from 75% to 50%? This could explain the results of the US Physicians' trial, in which a 39% reduction in non-fatal myocardial infarction was observed but no reduction in mortality or stroke and a trend to excess sudden deaths.⁶

The validity of recommendations based on meta-analysis alone is questionable because it is difficult to ensure freedom from bias and because it has frequently been misleading when supported only by individually inconclusive trials. At least one guideline group has decided that meta-analysis should be used as a method for deciding whether the mass of data is consistent with the outcome of a positive trial rather than taken as sufficient evidence by itself.⁷ Since the original reports of the best trials on aspirin failed to show a reduction in mortality, guideline recommendations may now have to be revised.

The authors conclude that it were better that my article had not been published. I am thankful that powerful people have a limited ability to censor the medical press. We need to be sure that long term aspirin is effective. The consequences of not doing so would be disastrous, for patients and future therapeutic developments.