Missing EXCEL MI Data Published, Prompting New Questions and Promised Answers

The EXCEL trial’s leadership has quietly published the “missing” rates of myocardial infarction using the Third Universal Definition (UDMI) with CABG surgery and PCI for the treatment of left main coronary artery disease that caused a tempest at the end of 2019. These new numbers, as critics alleged at the time, cast surgery in a much more favorable light.

The absence of the UDMI data—a prespecified secondary endpoint listed in the study protocol—was raised in a BBC Newsnight investigation that implied the numbers had been suppressed in the trial’s publication.

Now, as part of a series of letters in the New England Journal of Medicine, the EXCEL investigators provide the UDMI data. Lead investigator Gregg Stone, MD (Icahn School of Medicine at Mount Sinai, New York, NY), insists the higher rate of UDMI with PCI than with surgery doesn’t change the overall interpretation of the study results.

“The interpretation of the study is always based on the primary endpoint, and the primary endpoint used a definition of myocardial infarction which signified an extensive amount of myonecrosis after both PCI and CABG that was agreed upon by the surgical and interventional leadership of EXCEL,” he told TCTMD. “Every trial of course can have different outcomes if you change the definitions of different endpoints. And there was a very good reason that [our] leadership agreed on the primary definition.” 

As opposed to the protocol definition where there was no statistical difference in the risk of MI between the two revascularization approaches at 5 years, when the Third Universal Definition was used, there were 89 (9.6%) MIs in the PCI-treated patients and 43 (4.7%) MIs in the surgical arm.

In terms of the important procedural events, 3.3% of patients treated with PCI had an MI classified using the Universal Definition compared with 1.4% of surgical patients. In contrast, the rate of procedural MIs favored PCI (3.9%) over CABG surgery (6.0%) when EXCEL investigators used their protocol definition.

Missing Data Ignited a Firestorm

David Taggart, MD, PhD (University of Oxford, England), chairman of the EXCEL surgical committee during the design and recruitment phase of the trial, last year accused the investigators of overselling the study results and withdrew his name from the paper. At the European Association for Cardio-Thoracic Surgery (EACTS) annual meeting, Taggart alleged that noninferiority with respect to the primary composite endpoint of death, stroke, and MI was largely driven by a higher risk of periprocedural MIs in the CABG arm using the protocol definition and that investigators should have published the UDMI data as required by their study design.

In the brouhaha following Taggart’s presentation and the BBC Newsnight story, the EACTS formally withdrew support for the current treatment recommendations for left main coronary artery disease.

In one of several published letters to the editor this month, Taggart along with Mario Gaudino, MD (Weill Cornell Medicine, New York, NY), questioned why the UDMI data were missing from the 5-year EXCEL publication that came out in 2019, as well as other issues.  

In EXCEL, investigators used a modified version of the Society for Cardiovascular Angiography and Interventions (SCAI) definition for periprocedural MI and the Third Universal Definition for spontaneous events. They said they were unable to report the UDMI numbers, as the definition uses cardiac troponin (cTn) for biomarker assessment but the collection of cTn values was largely optional in EXCEL. As a result, only a limited number of patients had cTn concentrations measured.

To TCTMD, Stone said that when the clinical events committee (CEC) adjudicated UDMIs, it was unclear which biomarker threshold to use at the time of 5-year publication. With UDMI, CK-MB values can be used if cTn concentrations aren’t available—CK-MB values were routinely collected in EXCEL in support of the protocol definition of MI—but the definition doesn’t provide guidance on how to assess events in a situation where they have cTn concentrations for some patients and CK-MB values others.

“So frankly, the study leadership was uncertain how to proceed without a clear definition of the UDMI when there were mixed biomarkers available,” said Stone. “We have another manuscript that will be coming out soon that has looked at alternative definitions of myocardial infarction using numerous different combinations of biomarkers. I think that will be much more informative than what appears in our very brief letter [to the editor].”

Asked if the publication of these new data should quell critics, Stone said only that the EXCEL leadership has always tried to be as transparent as possible.

“The Universal Definition was one of numerous secondary endpoints we intended to look at and there are many other endpoints that we continue to look at in EXCEL,” he said. “We're getting to them as fast as we can. I hope all the data that comes out will continue to inform the interventional, surgical, and general cardiology community as to the best alternatives to treat patients with left main disease.”

Does the UDMI Change the Results?

For cardiac surgeon Marc Ruel, MD, PhD (University of Ottawa Heart Institute, Canada), however, the new data do change how the trial should be interpreted.

“The SCAI definition was an optimal definition—I'm not saying it was purposely so—in order for EXCEL to meet its noninferiority endpoint,” Ruel told TCTMD. “What I can say is that the SCAI definition had not been used before EXCEL in [any] other major venue, and it’s not been used since either. In fact, the fourth iteration of the Universal Definition of MI came out roughly 2 years ago now and it also didn’t espouse the SCAI definition.”

Rod Stables, DM (Liverpool Heart and Chest Hospital, England), an interventional cardiologist, said the headline study result from EXCEL is sensitive to the MI definition used in the primary outcome and while it showed PCI was noninferior to CABG surgery, there is more to a study than just the primary outcome.

“I have always seen EXCEL as a clear win for surgery, but then I read the paper and all the online additional data and studied all the figures,” he said.

Using the protocol definition of MI, the advantage for PCI appears related to the revascularization phase, with surgery having the edge for almost all outcomes afterwards, said Stables. “I suppose that for many readers, this will weaken the current conclusions of the study,” he said. The surgical advantage in EXCEL, he added, may be greater than seen in other studies because of a high proportion of participants with diabetes, approximately 30%, and the fact that the patients often had more complex coronary disease than the trial inclusion/exclusion criteria had demanded. 

Ruel stressed the inherent difficulties of comparing MIs using enzymatic release that occur from two distinct therapies. Instead, he said he believes investigators should exclude periprocedural MIs when comparing two unique procedures. The NOBLE trial, another study comparing PCI versus CABG surgery for left main coronary artery disease, did not count periprocedural MIs as part of their primary composite outcome of death, stroke, or MI, he noted.

“[Periprocedural MI] can be a secondary endpoint, but even then there will be important caveats because we’ll never agree on the definition,” said Ruel. “The two procedures just aren’t comparable. Sure, if you have ECG changes, if you have Q waves, or you have a new wall motion abnormality, you can't argue with that. But enzyme release is very, very difficult.”

Stone, along with Patrick Serruys, MD, PhD (National University of Ireland, Galway), and Joseph Sabik, MD (University Hospitals Cleveland Medical Center, OH), in the NEJM letter defend their primary endpoint and MI definition. They said the large discrepancy between the protocol and Universal Definition of MI was mostly due to a lack of confirmatory evidence of myocardial ischemia required as part of the UDMI. Additionally, the protocol definition of procedural MI was associated with a consistent hazard of cardiovascular death after both PCI and CABG surgery (P = 0.80 for interaction). In contrast, procedural events classified using the UDMI were strongly linked with cardiovascular mortality after CABG surgery but not after PCI (P = 0.04 for interaction).

“You choose your MI definition based on prior evidence, and the prior evidence before EXCEL suggested that the primary definition we chose, which did use a similar biomarker threshold and definition after PCI and CABG, was equivalently predictive after PCI and surgery, which is therefore fair to use as a common endpoint,” said Stone. “When we looked within the trial, the protocol definition was associated independently with all-cause and cardiovascular death and to a similar degree after PCI and CABG.”

Choosing an MI Definition

For Stables, who enrolled patients but wasn’t part of the leadership committee, EXCEL is an important study, one conducted to a very high standard in terms of attention to detail and research quality in the broadest sense. He pointed out that the primary endpoint—including how MIs were adjudicated—was prespecified.

“I agree that the reasons for the choice of this definition have logic and that this was a reasonable choice for a trial of this type,” he told TCTMD. “As an aside, the EXCEL leadership are presenting observational data to support the view that their 'type' of MI has a greater impact on subsequent mortality than would be observed with MI events declared using alternative approaches, but this would not have been known at the time of trial planning.”

He also noted that there can be some difficulty in adjudicating MI events using the Universal Definition among patients undergoing surgery, such as capturing patient symptoms or troponin release curves. “The decision not to provide immediate reporting of [UDMI] may have its basis on a concern for data integrity and reporting standards, rather than just a desire to 'bury' the findings,” he said.

Nonetheless, Stables stressed that the Universal Definition is one of the most important MI definitions available, one that was prespecified and should have been reported when the 3- and 5-year results were published. Given that it wasn’t, the investigators perhaps could have made an overt statement in the paper explaining why it wasn’t available at that time. Moreover, if data quality was an issue, real-time monitoring at an early stage of the trial could have identified problems with data collection, such as cTn values to adjudicate UDMI events, but Stables admitted it’s impossible to be 100% perfect in clinical trials of this scale.

Speaking with TCTMD, Richard Whitlock, MD (McMaster University, Hamilton, Canada), a cardiac surgeon, pointed out that the periprocedural MI definition used in EXCEL relied on similar thresholds for PCI and CABG rather than thresholds specific to the procedure, a point also raised by Pedro Lopez-Ayala, MD (Cardiovascular Research Institute, Basel, Switzerland), and colleagues in one of the letters to the editor.

“And when they're occurring at the rate of 4% to 6% using the protocol definition, that's way more frequent than anything else that was happening and everything gets drowned out by this outcome that is poorly understood, poorly validated.” Whitlock further stated that “this is a particular issue in a study that used a non-inferiority design; noise within the components of the composite primary outcome results in misleading results.”

The Mortality Question

In his letter, Taggart also raised the higher rate of all-cause mortality with PCI observed in EXCEL, arguing, as he did previously, that this is the most important endpoint for patients. To TCTMD, Whitlock said CABG surgery is the gold standard for the treatment of left main coronary artery disease, noting that a very early meta-analysis from 1994 showed CABG reduced the risk of death and had no effect on stroke or MI when compared with optimal medical therapy.

“If you have left main disease, the problem you face is that it’s going to kill you,” said Whitlock. “In EXCEL at 3 years the curves are diverging, but there was no statistical significance, but at 5 years there was a mortality benefit—very similar to the findings in the meta-analysis. If you're trying to say PCI is as good as CABG, the established standards for non-inferiority trials dictate that the comparison has to be on what CABG benefits: mortality, one of the most important outcomes to patients. Not surprisingly, every other outcome is really a wash in EXCEL.”

Ruel supports that interpretation of the trial, too.

“You can’t randomize 1,905 patients with robust distribution of random confounding factors and treat them for a lethal condition—they’re in their early 60s and this is the entity most likely to take their life—and then find that 5 years later you have about 35% more deaths in one group than the other and claim that it's not related to the intervention, that these are noncardiovascular deaths,” he stressed. “Really, in a randomized setting with a large number of patients with sufficient follow-up, one cannot argue with death. It is a very hard endpoint to bias in a randomized setting.”

Like they have done before, Stone, Serruys, and Sabik argue that mortality could be a false positive because EXCEL included 30 outcomes without adjustment for multiple testing. Moreover, cardiac mortality was similar at 5 years, and the overall difference in mortality was driven by noncardiovascular causes, such as late cancer or sepsis. They also point out that there have now been 5 trials involving 4,612 patients with left main disease treated with PCI or CABG surgery but no signal of higher mortality in a meta-analysis with a median follow-up of 67 months.

To TCTMD, Ruel said that diseases such as cancer and pneumonia are less likely to be survivable if the heart isn’t in great shape. Moreover, adjudication of the cause of death is a post hoc process and prone to bias. As to whether EXCEL should have placed more emphasis on the increased risk of all-cause mortality at 5 years when the results were published, Ruel said the answer is yes.

“I think the overall death signal, up 35% with PCI, is extremely relevant,” said Ruel. “That to me is the main signal from EXCEL, even beyond MI.”

Whitlock believes all-cause mortality should have been the primary endpoint in EXCEL but concedes there are roles for both procedures in the treatment of left main coronary artery disease.

“This trial provides information that you must apply to the patient in front of you,” he said. “In treating that patient, there's a lot more to think about than just where the blockage is. How old are they? Are they frail? Are there social circumstances that necessitate a fast recovery?. . .The one thing that EXCEL brings out is that PCI has come a long way. For certain patients, it for sure is in the armamentarium of how we can care for these individuals, but if a person has a prolonged life expectancy, I don’t think it should be the first-line therapy.”       

Note: Stone is a faculty member of the Cardiovascular Research Foundation (New York, NY), the publisher of TCTMD.