Thread by @drjohnm

Thread on my latest column @theheartorg on the faulty communication of the 5-year results of the EXCEL trial of Left Main PCI vs CABG

medscape.com/viewarticle/92…

A reader of the highly-influential NEJM manuscript would think the two treatments are similar

That's important because any human being given the choice b/w a PCI or open-heart surgery will choose PCI b/c it's far less invasive.

But are the two treatments similar at 5 years? I use the data in the paper to show that, no, CABG looks better in the long-term.

First the topline results of EXCEL, and some basics:

Left main coronary disease is a life-threatening condition and it's previously been shown (before PCI era) that CABG (coronary artery bypass grafting) was superior to medical Rx.
MI and death = endpoints that matter.

The problem with this composite endpoint is that EXCEL authors include peri-procedural MI -- as determined by an enzyme-dominant means--in the primary endpoint.

A Korean study shows how problematic this is: sciencedirect.com/science/articl…
In their series, the way you define MI DRAMATICALLY effects the incidence of MI after the two procedures.
The SCAI def of MI makes CABG look worse. This was the def used in EXCEL.

Dr David Taggert, at a session at the EACTS meeting, showed an analysis of Excel excluding periprocedural MI. This favored CABG -- by 44%

But wait, Mandrola, that is analysis after knowing the data. Garden of Forking Paths.
Ok... Maybe... But this exactly how NOBLE investigators analyzed their LM PCI vs CABG trial; thelancet.com/journals/lance…

MI rates in EXCEL are diverging like crazy. At the start of the trial, there are more in CABG arm (artifact of definition) and then later, there are many more in the PCI arm-- Likely real MIs
Early post CABG MIs have to be artifact b/c no signal of higher deaths.

Now to the death signal at 5 years: remember the mean age of patients in EXCEL was only 66. So death rates at 5 years is meaningful.

While overall mortality was a secondary endpoint --and the trial was not powered for detecting a diff; there were big differences:

Ah --maybe you would say: Mandrola, you are cherry-picking data. You and Taggert are biased.
Well, surely a statistician and trialist who does neither PCI or CABG has no turf to defend:
Here is Nick Freemantle --

In conclusion --
The definition of periprocedural MI biases the Primary endpoint in favor of PCI
Death signal clearly favoring CABG -- curves are widening over time
Thus ... the actual results in NEJM do NOT support the "no difference" conclusion.

If this paper stays as written, it will be cited to patients and written into guidelines.
Then, in the lab, which patient with left main would choose surgery if given the conclusions of EXCEL? A: None.
That is why I think @NEJM editors should require a rewrite of that paper:

Please take a look at my column medscape.com/viewarticle/92…
And consider leaving a comment thread there--as it will live with the article.
cc @RFRedberg @DavidLBrownMD @pomyers @dompagano