EXCEL: PCI inferior to CABG in Bayesian reanalysis


A Bayesian analysis of EXCEL and other randomised controlled trials comparing coronary artery bypass graft (CABG) and percutaneous coronary intervention (PCI) in left main coronary artery disease (LMCAD) suggests that PCI was associated with inferior long-term results for all events, including mortality.

James M Brophy (McGill University Health Center, Montreal, Canada) performed the analysis, the findings of which were published in JAMA Internal Medicine. He says: “This Bayesian interpretation of the EXCEL results, either considered in isolation or in the context of previous knowledge, suggested with a reasonably high probability that PCI was associated not only with statistical inferiority but also with clinical inferiority to CABG for treatment of patients with LMCAD, ceteris paribus. However, clinicians must also recognise the need for individual personalisation and consequently acknowledge that PCI may be an appropriate choice for selected patients, such as those with a reduced overall life expectancy of less than two to three years or those with very high surgical risk profiles.”

Brophy performed a systematic review using the PubMed database with the query string “(left main disease) and (PCI or CABG) and (five-year follow-up) and (clinical trial)” to identify all RCTs from January 1996 to January 2020 comparing CABG to PCI for treatment of patients with left main coronary artery disease and with five-year follow-up data. With the use of Bayesian methods, EXCEL (Evaluation of XIENCE versus coronary artery bypass surgery for effectiveness of left main revascularisation), the largest and most publicised RCT, was reanalysed as an isolated entity using noninformative priors, and in the context of previous knowledge using informative priors derived from similar trials. Published aggregate data were used with assignments from each trial following the original intention-to-treat principle. Combining EXCEL data with varying levels of prior information using Bayes theorem provided final (posterior) probability distributions for primary and secondary outcomes.

A composite endpoint of death, nonfatal myocardial infarction, and stroke was the primary EXCEL outcome and was therefore the primary outcome for the reanalysis. Secondary analyses were performed for isolated all-cause mortality and for the composite outcome, along with repeated revascularisation procedures.

Brophy noted that, when EXCEL data were analysed using the originally stated noninferiority design, the five-year primary outcome difference reported (2.8%; 95% CI −0.9%–6.5%) exceeded the predefined 4.2% noninferiority margin, and as a consequence the null hypothesis of PCI inferiority could not be rejected. However, the Bayesian analysis of the EXCEL primary outcome estimated 95% probability that the five-year primary outcome difference was increased with PCI compared with CABG and 87% probability that this difference was greater than one extra event for every 100 patients treated.

Bayesian analyses also suggested 99% probability that EXCEL total mortality was increased with PCI and 94% probability that this absolute difference exceeded one extra deaths per 100 treated. A systematic review identified three other RCTs that studied the same question. The incorporation of this prior knowledge reduced the estimated probability of any excess mortality with PCI to 85%. For the secondary composite endpoint, which also included repeated revascularisations, there were estimated probabilities of 98% for at least four extra events and of 90% for at least five extra events per 100 patients treated with PCI.

Brophy told Cardiovascular News: “The Bayesian perspective showing an 85% probability of increased mortality when all the trials are considered and a 47% probability of this mortality being greater than one extra death per 100 treated is actually quite insightful. The transparency of the Bayesian approach provides sharp contrasts with the frequentist approach that may, depending on personal values and risk assessment, lead to different ‘actionable’ decisions.”

And, in JAMA Internal Medicine, he writes: “Had the five-year EXCEL results been interpreted in alignment with their original primary noninferiority design and prespecified primary outcome margin of 4.2%, a different conclusion would have been reached, even within the framework of a standard frequentist analysis. This bayesian analysis was not restricted to the dichotomous world of null hypothesis statistical testing and provided additional quantitative insights into the mortality risk differences. By avoiding the dichotomisation of results into binary statistical significance bins at an arbitrary threshold, these Bayesian inferences arguably provide an enhanced appreciation of the results that may be helpful both to physicians and patients.”

In an accompanying editorial comment, Sanjay Kaul (Cedars-Sinai Medical Center, Los Angeles, USA) says the advantage of the Bayesian approach is that it “allows probabilities for different magnitudes of treatment effect, including clinically important effects, to be estimated. The results suggest high probability of a clinically important benefit with CABG over PCI especially in terms of MACCE, revascularisation, and even mortality. Thus, the suggestion by EXCEL investigators that the treatment strategies are comparable is questionable at best, and at worst risks unnecessary and avoidable deaths of patients with LMCAD.”

He adds: “The treatment of choice for LMCAD should be CABG regardless of anatomical complexity. Therefore, the guideline recommendation for PCI for low-complexity LMCAD should be downgraded in alignment with the new evidence.”

Brophy agreed, telling Cardiovascular News that “Kaul’s interpretation of the data and required actions are logical and consistent with this analysis”, although he pointed out that he was “less concerned about guidelines, and more concerned about supplying a transparent, unbiased, and systematic review of the literature so that meaningful conversations can occur between health care providers and their patients and optimised personalised decisions can be reached”.

Gregg Stone

In an extensive comment on the findings to Cardiovascular News, lead EXCEL investigator Gregg W Stone (Icahn School of Medicine at Mount Sinai, New York, USA) said: “While I wholeheartedly embrace Bayesian methodology, in the present report several non-statistical errors were made and there are glaring omissions. They pooled MACE and MACCE across the trials, even though NOBLE did not include peri-procedural MI (which favours PCI). Thus they combined apples with oranges. They should instead have analysed each component event separately, for example procedural and non-procedural MI, which would have shown that the former favoured PCI and the latter favoured CABG. As regards all-cause mortality, considering the totality of studies the finding of a 0.9% increase in mortality over five years may be statistically correct, although the posterior probability of more deaths with PCI was only 85% -that is, not meeting the level (95%) that is usually required to confidently be certain there is a true difference. However, even if real, this is a difference in mortality of <0.2%/year, a clinically meaningless rate given the extra morbidity of CABG. And they conveniently did not analyse 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 non-cardiovascular death from EXCEL, due to late malignancies and sepsis, a finding without biologic plausibility and therefore likely due to chance. They 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. They didn’t report stroke rates, which would have favored PCI. Finally, the difference in repeat revascularisation is real, and similar to what we and others have reported. However, revascularisation after PCI and CABG are not equivalent – patients are much more symptomatic when revascularisation is performed after CABG than after PCI. Thus it is a biased endpoint. And in terms of importance, repeat revascularisation 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.

“Thus, as more comprehensively shown in the recent meta-analysis by Ahmad et al in the European Heart Journal, the totality of data demonstrates that PCI with DES and CABG for patients with left main disease are two very different procedures: PCI offers advantages of being less invasive, with lower rates of stroke and procedural MI, fewer early episodes of major bleeding, acute kidney injury, atrial fibrillation requiring anticoagulation, early infections and repeat operations, and permits more rapid recovery and better early quality of life. CABG offers more durable protection from late MIs, and fewer patients after CABG undergo repeat revascularisation procedures than after PCI. 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. 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.”

In response to Stone’s comment that the analysis “combined apples and oranges”, Brophy stated that “the issue was extensively discussed in the article and is apparent upon a detailed reading”.


Please enter your comment!
Please enter your name here