Can New Meta-analysis of PCI vs CABG for Left Main CAD Settle the Debate?

John Mandrola, MD


November 19, 2021

After four randomized controlled trials, the best approach to treating patients with left main coronary artery disease ought to be clear.

The trials compared percutaneous coronary intervention (PCI) with drug-eluting stents against coronary artery bypass graft surgery (CABG) and measured important outcomes such as death, myocardial infarction (MI), stroke, and repeat procedures.

The largest of the four trials, EXCEL, has been mired in controversy regarding the signal of increased death in the PCI arm, the authors' definition of MI, and the decision not to publish a prespecified key secondary endpoint in the articles on the 3- and 5-year follow-up.

These controversies led a senior author to resign authorship from the 5-year-results paper and the European Association for Cardio-Thoracic Surgery to withdraw its support of European guidelines.

How to resolve these issues? Get an independent review of all the data from a skilled research team.

New Meta-Analysis of PCI vs CABG in Left Main CAD

At the American Heart Association (AHA) 2021 meeting, Marc Sabatine, MD, from the TIMI study group at Harvard Medical School, presented results of an independent meta-analysis of the four trials. The Lancet published the report along with an editorial.

One of the debates surrounding EXCEL was that at 5 years, death rates were significantly higher in the PCI arm. This was not seen in the NOBLE trial, the other large trial of PCI vs CABG in patients with left main disease. The EXCEL authors argued that the higher death rate in their trial was due to chance because cardiovascular death rates were not statistically different. But this requires adjudication, and everyone agrees that all-cause death remains the most unbiased endpoint.

The TIMI investigators chose all-cause death at 5 years as the primary endpoint of the meta-analysis. But they also presented data on rates of MI, stroke, and the need for repeat revascularization procedures.

Another decision the authors made was the choice of analytic method. As I have written previously, Brian Nosek and colleagues elegantly showed that the way you analyze a dataset can affect interpretation of the results.

Sabatine and colleagues chose to present the death outcome in two ways: first they used the standard frequentist statistical approach, with a risk difference and 95% CIs and P values.

But they also reported a Bayesian analysis, which tells you the probability of benefit given the data. It also tells you the probability of any benefit, or a 1% benefit, or a 5% benefit, and so on. You determine the degree of benefit that is clinically meaningful, and the Bayesian analysis tells you the probability of that occurring.

Mortality Findings

The main finding was that at 5 years, the rate of death was 11.2% in the PCI arm vs 10.2% in the CABG arm. This equated to a hazard ratio (HR) of 1.10 (95% CI, 0.91 - 1.32; P = .33).  The authors write that this resulted in a non–statistically significant absolute risk difference of 0.9% (95% CI, 0.9% - 2.8%).

In their Bayesian analysis, they found that the probability of any mortality benefit for CABG over PCI at 5 years was 86%, and the probability that the absolute mortality benefit for CABG was 1% or higher was 49%.

MI, Stroke, Repeat Procedures, and Composite Outcomes

In this meta-analysis, spontaneous MI was significantly more likely in the PCI group (HR, 2.35; 95% CI, 1.73 - 3.23; P < .0001). The absolute risk difference in spontaneous MI was 3.5%. Stroke rates were higher early after CABG, but at 5 years there was no significant difference.

The rate of repeat revascularization was higher after PCI than CABG (HR, 1.78; 95% CI, 1.51 - 2.10; P < .0001), with an absolute risk increase of 7.6%.

The broader composite outcome of death, MI, stroke, or revascularization was significantly higher for PCI vs CABG (HR, 1.31; 95% CI, 1.17 - 1.47; P < .0001), with an absolute risk increase of 6.6% (95% CI, 4.0% - 9.3%).


Consider the differing language surrounding the primary endpoint. The 10% higher death rate for PCI has 95% CIs that include both a 9% lower and a 32% higher death rate, and the P value is greater than .05. We say the 10% higher rate of death with PCI is not statistically significant. In other words, given that we test against the assumption of no difference, these data are not that surprising.

But then look at the Bayesian analysis: if you are a patient, one of the main things you want to know is which technique leads to a better chance of survival. These data, from four trials, tell you that there is an 85% chance that CABG is better and a 15% chance that PCI is better. And there is a near 50-50 chance that CABG is 1% better in absolute terms.

The Bayesian approach is how clinicians make decisions. We look at the rates of events in two arms of a trial, combine that with experiences, and then estimate the odds of one being better than the other in our patient. No matter the P value, medical decisions are never 100%; it is odds that eventually lead to a binary recommendation on treatment.

Note that the main controversy in Bayesian analyses stems from your prior assumptions. If you use too optimistic or too pessimistic of a prior, you can shift the interpretation of the data. The authors avoid this controversy by using a prior that is neither pessimistic nor optimistic. Statisticians call this a noninformative prior.

The Patient and the Heart Team

Of course, the treatment of patients with left main CAD involves more than just numbers.

The patient perspective turns on their view of the tradeoffs of the two procedures. Recovery from sternotomy and cardiopulmonary bypass looms large in the short run, but the need to stay on potent antiplatelet therapy and the higher probability of repeat procedures over the coming years factor in as well. Given the relatively young age of these patients (median, 66 years), long-term outcomes are important as well.

I am afraid this meta-analysis reveals no new answers. Death, MI, and repeat revascularizations favor CABG over PCI; stroke is similar at 5 years. Using the standard composite major adverse cardiac outcomes, the best average therapy on paper is CABG.

The question is how we present these data to our patients. Some patients would take the harder road of surgery to get the statistical advantages. Others might say the odds are not that different and going home the next day with a Band-Aid on the wrist sounds great.

The hardest part of these decisions is not letting our personal values affect framing. The framing problem is made even harder because both interventional cardiologists and surgeons have confidence in their skills—which is a positive.

An interventionalist might say, this is a favorable lesion, and with my experience and imaging tools, I can deliver an excellent result.

The surgeon might say, there are great targets, we have a great team, and the data support surgery.

There is a lot of talk about a heart team approach. In theory, it is a great idea, but in practice, getting a true multidisciplinary team together requires serious give and take among confident professionals—which is not always easily achieved.

But such a neutral approach seems worth the effort—for two reasons: one is that trials provide an average result but each person with left main coronary disease presents with specific nuances and goals.

The second reason neutral framing is needed is that I don't think another or even 10 more trials will yield different results. We are stuck with uncertainty.

For the record, I think the next trial in left main coronary disease ought to be revascularization vs optimal medical therapy. Recall that CABG beat medicines in the 1980s. And medical therapy has improved quite a lot since then.

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.