Superiority of Surgery vs Stents for Long-term Mortality in Patients With Left Main Disease 'Unarguable'

John M. Mandrola, MD; John J.V. McMurray, MD, BSc, MB ChB (Hons)


December 06, 2019

This interview was recorded on November 18, 2019, at the American Heart Association Scientific Sessions. The transcript has been edited for clarity.

John M. Mandrola, MD: Hi, everyone. This is John Mandrola from | Medscape Cardiology, and I'm here in Philadelphia at the American Heart Association meeting. I'm delighted to be joined by Prof. John McMurray from Glasgow, Scotland. We are here to discuss the EXCEL trial.

This was a trial of left main percutaneous coronary intervention (PCI) versus coronary artery bypass surgery (CABG), and it was quite controversial. The 5-year results were recently published in the New England Journal of Medicine. I picked Prof. McMurray to talk with us because he's an experienced trialist, but he's neither an interventional cardiologist nor a surgeon, so I think he can give an independent view.

Let's start with the two controversies of EXCEL: first, that the overall results were concluded to be neutral or the same at 5 years, but there was divergence of the endpoints; and second, the discussion about the definition of periprocedural myocardial infarction (MI).

For this extremely important endpoint of all-cause mortality, there appears to be clear superiority of surgery over PCI.

John J.V. McMurray, MD, BSc, MB ChB (Hons): It's very interesting to come at it from an external clinical trialist's perspective because, of course, like many trials today, EXCEL used a composite endpoint. Composite outcomes are very important because they allow us to measure the complete burden of disease and they allow us to do affordable trials of a reasonable size.

That said, there is one weakness of looking at composite outcomes, and that's when the components of the composite don't show treatment effects in the same direction. It doesn't have to be significant for each component, but if there's a directionally different effect, then there's a problem, because then the whole concept of that composite is questioned. And, of course, that's exactly what we saw in EXCEL.

Overall, the primary composite outcome was essentially the same between the two treatment groups, but that was because there was a directionally different effect on two of the components of the composite. There was apparently less MI in the PCI group but, very important, there was a higher mortality rate, especially during longer-term follow-up.

A further complexity of looking at composites in surgical trials, of course, is that there is always an early price to pay for a surgical intervention. That has to be the case; there will be some people who die. Therefore, there's always the question of catch-up, duration of follow-up, and the statistical analysis of the events because of the potential loss of proportional hazards during follow-up. So this is a textbook case of a very complicated analysis of a clinical trial.

Mandrola: Let's start with the proportional hazards. What exactly do you mean and how does it apply to the EXCEL discussion.

Discrepant Effect: MI, Mortality

McMurray: With proportional hazards you want to see that the hazard function essentially remains constant over time, but that's clearly not the case in EXCEL with the two endpoints that we're talking about. With MI, there was an early excess and then the rate of MI changed over time and the two treatment curves crossed over. The same was true for mortality. The normal Cox proportional hazards analysis that we do may not be valid in that situation. The investigators did look at that, to some extent, by splitting the analysis into different time periods.

Mandrola: You mean the landmark analysis.

McMurray: Yes. That sort of helps address that issue, but I would say that's not the most important problem about the analysis of this trial. It really is the fact that there is such a discrepant effect on two of the key components of the composite outcome. And not just discrepant but directionally opposite, and that really creates a problem for clinicians interpreting the data. There's also the problem then of when we look at the prespecified hierarchy of events.

Mandrola: Let's talk about that because the issue that surgeon David Taggart, MD, PhD, made during one of the discussions at the EACTS meeting was that the overall death rate, which is an important endpoint—I mean, it's probably the most important—looked to be perhaps not statistically significant at the end of the 5-year follow-up, but it was going in this wrong direction. However, it's not the primary endpoint. When you have a composite but then you have a super-important component of the composite, how do you handle that as a trialist?

McMurray: Doctors and regulators usually take the view that overall mortality trumps everything. When you look at EXCEL, you must always put the data from one trial in the context of data from other trials and ideally perhaps from a meta-analysis. And the data from this trial and the data from the meta-analysis clearly show that for this extremely important endpoint of all-cause mortality, there appears to be clear superiority of surgery over PCI. As far as I can see, that is unarguable.

Clinical vs Biochemical MI

McMurray: Then we have the problem that the overall composite doesn't look different between the two arms. The reason it looks the same is because of this difference in the other direction for what is called periprocedural MI.

If you're comparing a biochemical change to being alive or dead, I think it's clear which endpoint trumps the other.

I'm a general cardiologist. I look after the coronary care unit and I look after patients on the wards; an MI to me is a clinical event. It's patients with symptoms who have changes in their ECG, and they have an elevation of their cardiac biomarkers. But we also know that there are degrees of severity of infarction. In a clinical trial like this, it's a biochemical change. And in the perioperative period is a biochemical change that is again inevitable as a consequence of operating on the heart. You cannot avoid a rise in biomarkers, and therefore it's even more difficult.

If it was spontaneous MI during follow-up that occurred more commonly in one group, and mortality was more common in the other group, first of all that would be very unusual because if you have more true MIs, ultimately that will typically lead to more death. That is almost inevitable. So it would be a very unusual discordance between the two treatment groups, but if it was there, that would be a different question and that would be a much harder case to resolve as a clinician.

But if you're comparing a biochemical change to being alive or dead, I think it's clear which endpoint trumps the other.

All-cause mortality was a component of the primary outcome, and my interpretation of the data is that it was nominally statistically significant at the end of the 5 years of follow-up. You cannot ignore all-cause mortality just because it may be at the bottom of your sequence of tested endpoints. It is too important to ignore. No trialist, no clinician, no regulator just looks at part of the data.

Of course, for very good reasons, we give primacy to the primary endpoint, and in some ways that's to protect against perhaps false claims being made about efficacy. Safety is a different issue altogether. All-cause mortality is an efficacy outcome but it is also a safety outcome, and those rules of hierarchical testing don't apply to safety considerations.

Mandrola: I know you'll give me a calm, reasoned response to this question about the conclusion in the manuscript, which says that there's no difference. How do we resolve this as clinicians when the manuscript says this or if it's incorporated in guidelines?

McMurray: Well, I don't think it will be incorporated in guidelines. I hope guideline writers are more astute than that. Nobody's saying anything wrong. For the primary endpoint, the truth is, it was not different, but it's more complicated than that. You have to read beyond the abstract; you have to read beyond the primary endpoint.

I hope the guideline writers will always take the view that all-cause mortality is something that has to be treated very carefully, and I can't imagine that it will be overlooked.

I may have been a bit more hesitant about what I would say because I'm looking at this from outside, and I've had to sit down and think about it all. But it's not just the fact that we see this in the EXCEL trial; it's the fact that we see it in a recent substantial meta-analysis of several trials like this. This isn't a chance finding. First of all, it's plausible, and second, it's externally validated by other data. I think it's real. And yes, for the composite outcome, there was noninferiority demonstrated—no question—for the primary composite outcome, but that's not the whole story and indeed that's maybe not the most important part of the story.

Mandrola: Thank you very much for being with us. I think that helps a lot.

McMurray: Thank you.

Follow | Medscape Cardiology on Twitter

Follow Medscape on Facebook, Twitter, Instagram, and YouTube


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.
Post as: