Latest EXCEL MI Analysis Settles Nothing; Flaws Remain

John Mandrola, MD


October 01, 2020

The EXCEL trial comparing revascularization of left main coronary artery disease with percutaneous coronary intervention (PCI) using drug-eluting stents vs coronary artery bypass surgery (CABG) should have been straightforward.

One group gets PCI, the other surgery, and you tally stroke, heart attacks, and death. Instead, the trial has been beset with controversy.

A new post hoc analysis of EXCEL focuses on just one of the controversies: how myocardial infarction (MI) was defined. It strives to end debate. It does not. Rather, the new paper obfuscates and distracts from the core flaws of EXCEL and its interpretation.

EXCEL Background

In 2016, the 3-year results of EXCEL showed that PCI was noninferior to CABG for the composite primary endpoint of death, stroke, and MI. The results were codified into European guidelines, and untold numbers of patients with left main disease had stents instead of surgery in the intervening years.

In 2019, EXCEL authors published the 5-year results, and again concluded that there was no difference between the two procedures in the primary outcome. Controversy soon followed and involved numerous issues; primarily rates of death and MI, as well as selective publishing.

First, the death rate at 5 years was 3.1% higher with PCI, yet the authors concluded that the two therapies were similar because the composite endpoint did not differ significantly. (The assertion “did not differ significantly” is a matter of debate as I outline below.)

Second, the EXCEL trial used a procedural-agnostic definition of periprocedural MI. This definition biases against CABG because surgery causes more release of cardiac enzymes than does PCI.

But the authors stated in the trial protocol that they would also measure MI, as a secondary outcome, according to the Third Universal Definition of Myocardial Infarction. This definition uses different levels of enzymes depending on the procedure and requires additional clinical confirmation, such as ECG changes. These MI results were not disclosed in either the 2016 or the 2019 articles.

Shortly after the New England Journal of Medicine (NEJM) published the 5-year results, the BBC Newsnight program reported an independent analysis of a leaked data set from EXCEL that showed an excess risk for MI with PCI according to the universal definition. The lead researchers told Newsnight that this was “fake information.” But then 9 months later, in response to numerous letters to the editor of NEJM , the EXCEL authors confirmed that the rate of MI according to the universal definition was indeed twofold higher in the PCI arm.

New Study: Periprocedural MI and Future Mortality

Four years too late, the EXCEL authors now reveal further details on the universal definition of the MI endpoint. The data are buried in a complicated analysis of the rates and clinical relevance of periprocedural MI.

The short version is that the results of EXCEL tip from neutral to positive for CABG depending on which MI definition is used. Periprocedural MI defined by using the protocol definition occurred in 3.6% patients in the PCI arm and in 6.1% of patients in the CABG arm. With use of the universal definition, the ratio was reversed: 4.0% for PCI and 2.2% for CABG. The authors write a sentence in the discussion that would have changed the conversation 4 years ago:

In an unplanned post hoc analysis, extrapolating these findings [rates of MI according to the universal definition] to the primary endpoint translated to fewer composite outcome events at 3 and 5 years after CABG compared with PCI.

Translation: EXCEL would have shown CABG as superior to PCI with that MI definition.

The focus of the latest paper is to correlate the two periprocedural MI definitions with future cardiovascular (CV) and overall death. This focus makes sense because periprocedural MI is important only if it predicts future adverse outcomes. Recall that in NOBLE, the other major trial comparing PCI to CABG in left main disease, periprocedural MI was excluded from the primary endpoint because those investigators believed that it was unreliable as a surrogate endpoint. 

According to the latest analysis from EXCEL, protocol-defined periprocedural MI was independently associated with CV death and overall death at 5 years with a similar hazard after either PCI or CABG. In contrast,  periprocedural MI per the universal definition was predictive of future mortality after CABG, but not after PCI. The difference was large: the hazard ratio for 5-year CV death after a post-CABG MI was nearly 12-fold higher than a post-PCI MI. The P value for interaction showed a significant difference (P = .004).

The EXCEL authors believe that these observations support their decision to use the protocol definition of MI and uphold the equivalence of PCI and CABG for patients with left main coronary artery disease. In his weekly podcast, the editor-in-chief of the Journal of the American College of Cardiology said he has "full satisfaction" and urged us to move on from controversy.

Editorial Rebuttal

Donald Cutlip, MD, from Beth Israel Hospital in Boston, pushed back on the EXCEL author’s conclusions in the accompanying editorial.

He delved into Table 7 of the appendix, which showed that 33 of the 56 protocol-defined MIs in the CABG group were based on substantial CK-MB elevations alone (>10x upper reference limit) without supporting clinical evidence. Cutlip points out that none of the 6 cardiac deaths associated with CABG-related periprocedural MI occurred in this group.

The authors wrote that discrepancies between the two periprocedural MI classifications were seen in approximately one quarter of PCI patients, and these differences were sufficient to abolish the prognostic utility of PCI-related MI classified by the universal definition. This is not false, but as Cutlip notes, the difference was 1 death. There were 4 deaths after PCI associated with protocol-defined periprocedural MIs, but only 3 deaths associated with the universal definition.

This one adjudication resulted in loss of significance in the correlation with future mortality.

My Comments

This post hoc observational analysis does not convince me that periprocedural MI is a useful endpoint.

The protocol-defined, periprocedural MI definition led to fewer procedural MIs in the PCI arm, but at 5 years, there was a higher rate of death with PCI than with CABG. This suggests that this endpoint was not a useful predictor of death. And it’s consistent with Cutlip’s point that none of the 6 cardiac deaths in the CABG arm occurred in patients with periprocedural MI defined by enzymes alone.

The EXCEL investigators did not report any data on the universal definition of MI in either the 3-year or the 5-year papers, despite it being a prespecified secondary endpoint. When Newsnight published the results according to the universal definition, the authors called it “fake information.” It wasn’t fake information. It was nearly exactly the same data published in their NEJM letter and now in their latest paper in the Journal of the American College of Cardiology.

I am searching for a word to describe having important prespecified data, choosing not to publish it (twice), calling it fake when someone else publishes it, and then later publishing essentially the same data dressed up in a complex analysis using multivariate regressions. Perhaps unseemly is as generous as I can be. And that is without mentioning that many of the authors have financial ties to stent manufacturers.

The EXCEL trial is not the first to grapple with definitions of procedural MI—that topic has been debated for decades. Had the universal definition of MI been used in the primary composite endpoint of EXCEL, the trial would have clearly favored CABG. In fact, the only reason EXCEL (barely) made noninferiority at 3 years was the higher rate of periprocedural MI in the CABG arm.

Brian Nosek, PhD, and colleagues showed in 2018 that how you choose to analyze a data set can have a bearing on the conclusions. Even with use of the protocol definition of MI exclusively, the 5-year results hinge on the analytic method.

EXCEL was designed as a noninferiority trial, and the 3-year results were reported with an agreed-upon margin of noninferiority of 4.2%. The 5-year results were presented as an analysis for superiority: the event rate was 22% for PCI vs 19.2% for CABG (95% CI, 0.9% - 6.5%; P = .13). Had the authors analyzed the 5-year results with a noninferiority design, the 6.5% upper bound of the confidence interval would have exceeded the noninferiority margin of 4.2%. The conclusion would have been that PCI was not noninferior to CABG.

This new paper does not change the fact that EXCEL—as analyzed using the protocol definition of MI—did not meet an arguably wide noninferiority margin at 5 years.

The latest JACC paper should not dampen the controversy surrounding this trial. I, along with 52 cardiologists, surgeons, and researchers, have signed a petition calling for a revision of the European revascularization guidelines and independent scrutiny of the EXCEL data.

Sorting this out transcends the specifics of the treatment of people with left main coronary artery disease. It is a matter of trust in clinical science.

John Mandrola practices cardiac electrophysiology in Louisville, Kentucky, and is a writer and podcaster for Medscape. He espouses a conservative approach to medical practice. He participates in clinical research and writes often about the state of medical evidence. 

Follow John Mandrola on Twitter

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: