New EXCEL Analysis Reignites Debate Over PCI for Left Main CAD
James Brophy criticizes investigators’ view of the study, but the trial’s PI fires back by citing analytical errors and omissions.
(UPDATED) A statistical deep dive into the EXCEL trial, one that also includes evidence from other revascularization studies in the setting of left main coronary artery disease, shows that there is near-total probability that the risk of death, MI, or stroke, as well as overall mortality, is higher with PCI compared with surgery.
In a Bayesian analysis, there was an estimated 3.0% mean difference in the primary endpoint of death, MI, and stroke in favor of CABG, “with 95% probability of increased risk with PCI” and an “87% probability that this difference was greater than one extra event per 100 patients treated.” When it comes to the controversial all-cause mortality result, which was estimated to be 3.3% higher with PCI, the Bayesian analysis revealed there was “99% probability that the total mortality is increased with PCI compared with CABG” and a “94% probability that this difference was at least as great or greater than one extra death per 100 patients treated.”
For James Brophy, MD, PhD (McGill University Health Center, Montreal, Canada), sole author on the study published June 1, 2020, in JAMA Internal Medicine, the Bayesian approach reveals the possibility of a mortality signal with PCI compared with surgery. The scientific response should not be to ignore it, he says, but rather to systematically combine the finding with data from other trials, such as NOBLE, PRECOMBAT, and SYNTAX, “to provide an informed quantifiable estimate with its associated uncertainty.”
For example, when the Bayesian analysis included those randomized trials to inform the probability estimates of the primary endpoint, the mean difference between the two revascularization approaches was reduced to 2.6% in favor of surgery. “But the probability of more primary events with PCI remained high at 96%, with 86% probability of exceeding at least one event per 100 procedures,” according to Brophy.
When including other studies to inform the mortality endpoint, the difference in death was just 0.9% in favor of surgery. As a result, there was less certainty than when analyzing EXCEL alone, with Brophy reporting that the probability of more deaths with PCI was 85% and a 47% probability of there being at least one more death per 100 patients treated when the totality of evidence was considered.
While the different statistical approaches might appear a little “insider baseball” to the uninitiated, the new paper is important because it addresses one of the more controversial issues in cardiology right now, said Sanjay Kaul, MD (Cedars-Sinai Medical Center, Los Angeles, CA), who wrote an accompanying editorial. It also has real-world implications beyond a turf war between surgeons and interventionalists.
European guidelines have endorsed PCI as a class I indication (level of evidence A) for patients with low-to-intermediate complexity. The European Association for Cardio-Thoracic Surgery (EACTS) recently pulled their support from the section on left main disease following a BBC report showing a higher rate of MI adjudicated using the universal definition and allegations of ignored warnings from the Data Safety and Monitoring Board. In the United States, guidelines state PCI is class IIa indication for left main CAD in patients with a low SYNTAX score and a class IIb indication for those with an intermediate SYNTAX score.
“All of this becomes very important when it comes a truly reliable informed-consent process,” Kaul told TCTMD. “It depends on the transparent and trustworthy accounting of facts, which is a core objective of the guidelines. If you were the patient and I told you the outcomes between PCI and CABG surgery were similar, why would you choose the more invasive option that is associated with greater initial procedural morbidity and a longer, more painful recovery? On the other hand, if I told you bypass surgery was associated with improved survival and reduced risk of heart attack, you, like most, would be willing to play off short-term pain for long-term gain.”
Other Trials to Inform Probabilities
When the 5-year results of EXCEL were published and presented last year, the investigators reported there was no significant difference in the risk of death, stroke, or MI at 5 years among patients with left main CAD when treated with PCI or surgery. However, the risk of mortality was 38% higher with PCI, a difference that was statistically significant (13.0% with PCI vs 9.9% with surgery; OR 1.38; 95% CI 1.03-1.85).
That difference sparked a massive firestorm, leading David Taggart, MD, PhD (University of Oxford, England), who chaired the surgical committee during the design and recruitment phase of the trial, to pull his name from the paper and accuse the EXCEL researchers of downplaying the mortality risk. While Taggart also accused the EXCEL investigators of stacking the deck against surgery given the definition of MI used in the trial, he argued that being alive was the endpoint patients cared about most and this clearly favored surgery. The EXCEL researchers countered by stating the mortality risks were driven by sepsis and cancer and stressed there was no difference in cardiovascular mortality between PCI and surgery.
Kaul said the main advantage of Bayesian methodology is that it helps people understand the probabilities of a clinically important difference in outcomes. In contrast, the classical method of analysis, the frequentist approach that focuses on the frequency of a clinical event, utilizes P values and confidence intervals to determine statistical significance.
“If somebody is interested in a 1% absolute difference in mortality given all the other advantages of that intervention, the Bayesian analysis allows them to know the probability of that difference in mortality,” Kaul told TCTMD. “The frequentist analysis, strictly speaking, does not allow you to compute that.”
To TCTMD, EXCEL principal investigator Gregg Stone, MD (Icahn School of Medicine at Mount Sinai, New York, NY), said in an email that while he “wholeheartedly embraces” Bayesian methodology for analyzing clinical trials, Brophy made several nonstatistical errors and left many things out. For example, SYNTAX, NOBLE, PRECOMBAT, and EXCEL included different endpoints, but Brophy pooled the MACE and MACCE endpoints across the studies. NOBLE, he noted, didn’t include periprocedural MI, an endpoint that favors PCI over surgery.
“Thus, they combined apples with oranges,” said Stone. “[He] should instead have analyzed each component event separately. For example, procedural and nonprocedural MI, which would have shown that the former favored PCI and the latter favored CABG.”
With respect to the controversial all-cause mortality finding, Stone said the 0.9% excess with PCI over 5 years might be statistically correct, but that the posterior probability of more deaths with PCI was just 85%, a number that is far below the typical 95% probability usually used to be certain there is a high likelihood of difference.
“However, even if real, this is a difference in mortality of less than 0.2% per year, a clinically meaningless rate given the extra morbidity of CABG,” he said. “And [he] conveniently did not analyze cardiovascular mortality, which would have shown absolutely no difference between PCI and CABG. They ignored the fact that any small difference in all-cause death was driven by the increase in noncardiovascular death from EXCEL, due to late malignancies and sepsis, a finding without biologic plausibility and therefore likely due to chance. [He] inexplicably did not take into account the 10-year outcomes data from SYNTAX and PRECOMBAT, both of which showed nearly identical all-cause mortality between CABG and PCI. This would have further narrowed any differences between the two procedures in death.”
Additionally, Stone said the Bayesian analysis didn’t highlight the risk of stroke, which favors PCI, and while PCI is clearly associated with a higher risk of repeat intervention, revascularization after PCI and CABG surgery are not equivalent procedures.
“Patients are much more symptomatic when revascularization is performed after CABG than after PCI,” he said. “Thus, it is a biased endpoint. And in terms of importance, repeat revascularization is arguably less important that other findings such as atrial fibrillation, major bleeding, and acute renal insufficiency which have been shown to favor PCI and were not considered in the present report.”
Brophy countered by stating that the only separate endpoint prespecified as a primary or secondary endpoint by all the trialists was total mortality, which he did consider separately in his reanalysis. He also addressed the significance of the all-cause mortality result, telling TCTMD that the Bayesian perspective is quite insightful, noting that it “modestly” challenges the standard statistical interpretation of the mortality signal.
“Consider this simple analogy,” Brophy wrote in an email. “The null hypothesis is: ‘It will not rain today.’ The frequentist analyzes the available data and concludes that the null hypothesis can’t be rejected. What is the ‘actionable’ outcome from this analysis? Most would concur it would be to not take an umbrella. The Bayesian examines the same data and concludes there is an 85% (updated probability) of rain today and a 47% [chance] that it will be a deluge. What is now the actionable outcome? For many people, it would be to take an umbrella. In my opinion, the transparency of the Bayesian approach provides sharp contrasts with the frequentist approach that may, depending on personal values and risk assessment, indeed lead to different actionable decisions.”
Less theoretically, Brophy said, the Bayesian analysis suggests there is an almost 1% difference in all-cause mortality at 5 years. While Stone might feel that difference is not clinically meaningful, “I suspect that opinion is not universally shared by all physicians and more importantly by all patients,” he said.
Speaking with TCTMD, Kaul countered by stating that without doing an autopsy, it is difficult to adjudicate the causes of death. Even cardiovascular mortality, which tends to be adjudicated more accurately, was 1.3% higher in the PCI arm at 5 years, he observed, and while this difference wasn’t statistically significant, the difference is “clinically important,” especially when PCI doesn’t offer any other advantages in terms of MACE and MACCE. Stroke rates, he added, generally favor PCI over CABG surgery, although this was not observed in NOBLE. He noted that revascularization following PCI is not a trivial issue, saying that roughly 30% of patients are sent to surgery.
Kaul also highlighted an issue that has been raised before: that the prespecified analysis was noninferiority at 3 years, not superiority, but investigators chose to perform a superiority analysis at 5 years. The rationale for the change is unclear, but Kaul said that if a noninferiority analysis was conducted at 5 years it would not have met the criteria for statistical significance. Despite this, investigators claimed outcomes for PCI and CABG were interchangeable, even though there was a nonsignificant 3.1% absolute difference in the risk of the primary endpoint and a significant difference in mortality.
“This is what I believe is a classic case of interpretation bias, otherwise known as spin,” said Kaul. “You ignore the mortality difference and only focus on MACE. The MACE endpoint doesn’t meet the P value threshold but the initial construct of the study was never a superiority hypothesis.” Even at 3 years, he added, one excess event in the PCI arm and one less in the surgery group would have overturned the noninferiority endpoint, which exposes the “fragility” of the data.
On the whole, Kaul said the Bayesian analysis counters the EXCEL investigators’ claims that PCI and CABG surgery are comparable revascularization strategies in the setting of left main CAD. “That assertion is questionable at best, and at worst risks unnecessary and unavoidable deaths of patients with left main coronary disease,” said Kaul.
Brophy points out that 14 of the 34 EXCEL authors had a financial relationship with Abbott, the study sponsor, and that eight study authors list the Cardiovascular Research Foundation (CRF) as their affiliation. CRF, writes Brophy, received $937,000 from Abbott during the course of the trial.
To TCTMD, Stone said that “Abbott and many other companies make appropriate fair market payments to CRF to support numerous independent clinical trial activities and educational events,” and that “clinical trial interpretation by the authors was based solely on the data from the trial.”
Stone pointed to a recent meta-analysis led by Yousif Ahmad, MD (NewYork-Presbyterian/Columbia University Irving Medical Center, New York, NY), showing comparable rates of mortality, as well as similar rates of cardiac death, stroke, and MI, among patients with left main CAD treated with surgery compared with PCI. Rates of unplanned revascularization were higher in the PCI-treated patients. Those data include EXCEL, NOBLE, SYNTAX, PRECOMBAT, and a small German study, and they show that PCI and CABG surgery are two unique procedures—each with its own strengths and weakness—for left main CAD.
“Considering both the periprocedural and long-term outcomes, there are no major differences in cardiovascular or all-cause death and or total MI, and late quality of life is similar,” said Stone. “Thus, each heart team should assess each individual patient’s clinical characteristics and coronary anatomy and for those who are who are good candidates for both PCI and CABG, discuss the likely outcomes of both procedures with each patient, taking the patient’s preferences strongly into account.”
For his part, Brophy agreed with the importance of the heart team, and said his latest analysis was an attempt to supply a “transparent, systematic analysis of the totality of the data so as to best inform these discussions.”
Note: Stone is a faculty member of the Cardiovascular Research Foundation, the publisher of TCTMD.
Brophy JM. Bayesian interpretation of the EXCEL trial and other randomized clinical trials of left main coronary artery revascularization. JAMA Intern Med. 2020;Epub ahead of print.
Kaul SK. Should percutaneous coronary intervention be considered for left main coronary artery disease? Insights from a Bayesian reanalysis of the EXCEL trial. JAMA Intern Med. 2020;Epub ahead of print.
- Brophy and Kaul report no relevant conflicts of interest.
- Stone reports working with numerous device and drug companies but has no financial relationships with Abbott or any other stent manufacturer.