(Updated with a response from Bernard Gersh on August 18)
In a recent CVCTCardioBrief blog, Dr. Gersh questioned whether the recent controversy around the EXCEL trial in LMCA disease was overblown and lamented that the controversy had been a distraction. He offered his interpretation of the trial results dispelling the concerns around mortality and drawing the conclusion that there clearly is a role for PCI in selected patients with LMCA disease beyond trial data, driven by demographic, clinical, and anatomical information, operator proficiency and patient preference. I was invited by Larry Husten, a dear old friend, to write a brief rebuttal. So, here is the rest of the story.
The evidentiary support for guideline recommendations for low complexity LMCA disease (defined by a SYNTAX score of <22) is primarily derived from 4 randomized controlled trials: hypothesis-generating subgroup data from SYNTAX which showed noninferiority of PCI relative to CABG; NOBLE which demonstrated inferiority of PCI; and PRECOMBAT and EXCEL both of which established noninferiority of PCI relative to CABG. Noninferiority in SYNTAX and PRECOMBAT was met using absolute risk, but not relative risk difference margin because the observed primary outcome rates with CABG were lower than expected, thus biasing the results towards noninferiority. EXCEL barely met noninferiority at 3 years as the upper bound of 95% CI of 4.0% was just below an arguably wide margin of 4.2%. Moreover, noninferiority was primarily driven by periprocedural MI, 83% of which were defined by CK-MB elevation > 10xULN alone, a criterion which is biased against the more invasive CABG. Noninferiority was not established at 5 years as the upper bound of 6.5% exceeded 4.2% margin. Notably, a 2.1% advantage in periprocedural MIs for PCI did not translate into a mortality benefit, but instead was associated with a mortality disadvantage, thereby questioning the prognostic significance of these MIs. Despite a 3.1% absolute or 38% relative increase in mortality with PCI, the conclusion by EXCEL investigators that the treatment strategies are comparable is questionable at best, and at worst risks unnecessary and avoidable deaths in patients with LMCA disease. Furthermore, SYNTAX score has not been validated in prospective randomized trials as a predictor of outcome differences between PCI and CABG in LMCA disease, thereby questioning whether its use in guidelines is justified. Thus, it appears that the evidence is insufficient to justify a Class I Level of Evidence A recommendation in support of PCI for low complexity LMCA disease endorsed by the 2018 ESC/EACTS guidelines.
Dr. Gersh appears to be reassured by the null effect on mortality based on a trial-level meta-analysis. It is debatable whether the results from the 5 trials are poolable given the possible differences between the studies such as patient populations, complexity of LMCA disease, choice of drug-eluting stent, definition of MI, duration of follow-up, and methodologic differences. Lack of statistical heterogeneity is necessary but not sufficient to justify pooling. The inconsistency index, I2 of 43% indicates moderate heterogeneity driven by the results of EXCEL. The 2 trials that are most relevant to contemporary practice show substantial overlap in mortality outcome: EXCEL, HR 1.38 (95% CI 1.03, 1.85) vs NOBLE, HR 1.08 (95% CI 0.74, 1.59). Of note, the meta-analysis was published as a ‘fast track’ publication in European Heart Journal in March 2020 after the EXCEL controversy had erupted in late 2019, presumably to counter the adverse publicity surrounding the mortality result in EXCEL.
A previously reported individual patient-level meta-analysis of 4 trials (that included 3-year results of EXCEL and NOBLE) concluded no significant difference in mortality between PCI and CABG (HR 1·07, 95% CI 0·87 to 1·33).
This interpretation is misleading and Dr. Gersh perpetuates the error by ignoring the interaction term was not significant (unadjusted p=0.12). The correct interpretation is that the main effect in the overall cohort (harm of PCI over CABG: HR 1·20, 95% CI 1·06–1·37; p=0·004) is the most reliable estimate of treatment effect in subgroups, including patients with LMCA disease.
Dr. Gersh asserts that the difference in mortality in EXCEL is driven by an increase in non-cardiovascular mortality, primarily related to malignancy and infection occurring late during follow-up. I would argue that adjudicating the exact cause of death without autopsy is challenging and vulnerable to ascertainment error and misclassification. Even the 1.3% difference in CV mortality (definite plus undetermined cause) is arguably clinically important in light of no discernible long-term advantage with PCI in spontaneous MI, stroke or repeat revascularization.
Finally, the controversy surrounding the Universal Definition criterion for periprocedural MI cannot be resolved given the 6-fold greater lack of information for CABG compared with PCI regarding confirmatory evidence of myocardial ischemia that is required to support Universal Definition criterion for periprocedural MI. As a consequence, the differential prognostic impact of these periprocedural MIs cannot be estimated reliably. Thus, only data for nonprocedural MIs are reliable and indicate about 3% absolute risk increase with PCI using both the protocol definition and the Universal Definition.
In conclusion, based on the review of the totality of the evidence, a Class IA recommendation for low-complexity LMCA disease cannot be justified, thereby, warranting a downgrade in the ESC/EACTS guideline recommendation. While evidence is lacking to recommend PCI as a treatment of first choice, it might be a reasonable option to consider in patients in whom CABG poses a prohibitive risk. I am reminded of the quote from Nietzsche: “there are no facts, only interpretations”. Dr. Gersh and I have offered our interpretations of the LMCA disease trials. Clearly, our views are far from ‘close to consensus’. Feel free to recalibrate your interpretations, if you must.
Update (August 18):
Bernard Gersh sent the following response to Sanjay Kaul’s blog post:
“ Guess we have to agree to disagree.
From a clinical perspective, when faced with a decision in an individual patient, I believe the following are important issues:
a) The absolute magnitude of the differences in mortality is small but favor CABG.
b) Patients enrolled in these trials were by and large “good” surgical candidates and the PCI patients were “technically suitable”.
c) Irrespective of the statistical issues pointed out by Dr Kaul, the point I tried to make is that there many other clinical factors that need to be taken into account when recommending a therapeutic approach to an individual patient. In my experience patients are not that interested in “interaction” terms or other statistical issues such as the p value but they do want an educated opinion.
d) Whether the guidelines will change from a Class 1 recommendation to a 2A remains to be seen and this will not change clinical practice. Of course, there is a role for PCI in carefully selected patients with LMCA disease —as stated by Voltaire, “common sense is not that common ”.
e) Finally, I still believe that the controversy was overblown. Issues were raised that needed to be discussed further and that has indeed taken place. ”
Full Coverage of the PCI versus CABG for LMCA on CVCT Cardiobrief:
- Left Main Coronary Disease Trials: Close to Consensus by Bernard Gersh
- Left Main Coronary Disease Trials: Far from ‘Close to Consensus’ by Sanjay Kaul (with a response from Bernard Gersh)
- Don’t Confuse the Art and Science of Medicine: PCI vs CABG for Left Main Disease by Larry Husten