Editor’s note: This post by Harlan Krumholz is republished with permission from CardioExchange, a new website for cardiovascular healthcare professionals from the New England Journal of Medicine. CardioBrief readers who are healthcare professionals are invited to join the site.

by Harlan Krumholz

I’m always scouting for papers to discuss in journal club with my students. Earlier this month, I found the perfect pair: two simultaneously published articles from the industry-funded CURRENT–OASIS 7 randomized trial, one in the New England Journal of Medicine and the other in the Lancet. Many of the authors of the two papers were the same.

First, a recap: In a 2×2 factorial design, investigators compared higher- with standard-dose aspirin (either 300–325 mg or 75–100 mg daily) and higher- with standard-dose clopidogrel (either 600-mg loading, then 150 mg/day for 6 days, and 75 mg/day thereafter or 300-mg loading and 75 mg/day thereafter) in patients with acute coronary syndromes. All 25,086 participants were scheduled for percutaneous coronary intervention no more than 72 hours after randomization; about two thirds eventually underwent PCI. The primary endpoint was cardiovascular death, myocardial infarction, or stroke at 30 days.

The NEJM article reports the main finding: a failure to demonstrate the superiority of the higher doses over the standard doses (4.2% vs. 4.4% incidence of the primary endpoint for both aspirin and clopidogrel). In subgroup analyses, a predetermined threshold for significance of P≤0.01 was used, appropriate (as the authors note) for the multiple comparisons performed. (Some might even argue that the threshold was generous, given that there were 13 subgroups.) Within those parameters, no evidence of a significant interaction was found for any of the prespecified subgroups. The authors conclude, without qualification, that outcomes did not differ significantly between the higher- and standard-dose groups.

In the simultaneously published Lancet article, the authors pivot from the NEJM article and focus on the subgroup that underwent PCI, which was one of the 13 subgroups reported in the NEJM. Although PCI was planned for everyone who was randomized, about a third of the patients did not undergo the procedure because they were not considered suitable based on angiography findings. In the group that did eventually have a PCI, there was no difference in the primary endpoint between higher- and standard-dose aspirin (4.1% vs. 4.2%) but a “significant” difference for higher- versus standard-dose clopidogrel (3.9% vs. 4.5%; P=0.039). The authors endorse double-dose clopidogrel for ACS patients who are treated with early PCI.

Teaching Point 1: Avoid simultaneously publishing articles that spin the same data in completely different ways. Anyone who believes everything he or she reads in these high-impact journals would have been quite confused this month.

Teaching Point 2: A negative result should be called what it is. For reasons they articulate themselves, the NEJM authors prespecified the threshold for significance in the subgroup analyses at P≤0.01. They then use the term “nominally significant” to describe the PCI subgroup findings yet acknowledge in their discussion that the result did not meet their significance threshold and, therefore, “could have been due to the play of chance.” I would have preferred that they state plainly that the result was not significant.

The Lancet authors, of course, tell a different story in their discussion, saying that the data “suggest a clear benefit of double-dose clopidogrel” in the PCI subgroup. In listing the study’s limitations, they make no mention that the interaction was negative by their prespecified standard. Read in isolation, the Lancet article appears to present persuasive evidence for giving the double dose to these patients; read in tandem with the NEJM article, it obscures the fact that the interaction is negative and, therefore, that the evidence for the use of double-dose clopidogrel is weak.

Teaching Point 3: It’s best to select subgroups using information available at randomization. Even if the interaction had been positive and the evidence had clearly favored the use of double-dose clopidogrel in patients undergoing PCI, the clinical implications still would have been murky because this subgroup could not have been identified at the time the decision was being made. (Angiography results determined who would and who would not undergo PCI.) That leaves the reader to speculate, as the Lancet authors do, about the implications for practice. The authors go ahead and suggest starting everyone with a double dose of clopidogrel and discontinuing that higher dose when it becomes clear that a patient will not undergo PCI. However, this trial simply did not test that strategy, so it must be understood as speculation.

These articles earn my top rating for use in a journal club. Rarely will you witness famous authors draw such different conclusions about identical data published simultaneously in two prestigious journals. It’s a perfect opportunity for students and others to learn about interaction testing, significance levels, and subgroup selection.

What lessons do you derive from these studies? Do you agree with the NEJM authors or the Lancet authors? You can’t agree with both.

2 Comments

  1. The controversy that involves the two large studies is so typical to:What is going on in medicine in these days?
    Today is the era of E.B.V.:Evidence Based Medicine.
    Today we talk numbers,and not mechanism of disease.We talk Treatment.We ignor Prevention.
    Let us not forget that:Vascular endothelium dysfunction
    is the MOTHER of all diseases in the world.This is my idea,and you try to disproove it.Let us remember the H.D.L. that plays the decisive role in vascular endothelium dysfunction.Let us stop and think if Aspirin and or Clopidogrel are the crucial medications
    for that given circumstances.
    The data in these studies are so many(and confusing)
    that:”From too many trees no one can see the forest”

  2. Perhaps the Lancet article authors misplaced the decimal point on the p value. If my calculations are correct, if one is using the Bonferroni correction for multiple testing with 13 subgroups, then a significant p value would be p=0.003846. If this is rounded up to p=0.0039, then a misplaced decimal point could almost explain the discrepancy. I think not, however.

Leave a Reply