PCI vs CABG for left main disease: EXCEL Vs NOBLE

Pablo Lamelas –

In 2016 two large randomized trials testing the role of PCI in left main (LM) disease were published, which seem inconsistent. Are they? what the body of the evidence says? How do they impact my practice? Here we discuss both studies, address methodological issues and provide its interpretation in context.

EXCEL trial

This study, among people with left main disease, evaluated the efficacy of PCI vs CABG for death-stroke-myocardial infarction as primary outcome up to 3-years. The study showed superiority of PCI within 30-days, mainly driven by periprocedural myocardial infarction favouring PCI, while 3-year estimates remained similar (HR = 1.00) which met criteria for non-inferiority.


1 in 4 patients actually had high SYNTAX score with core lab: Although inclusion criteria was SYNTAX score under 32, this score has considerable inter-observer variability. Therefore, 1 in 4 patients randomized in the trial ended up having SYNTAX score over 32 by central core lab analysis. How this impact the results? This may impact the internal validity of the study since a considerable amount of patients were indeed “not eligible” from the start, favouring CABG based on prior data suggesting that SYNTAX score over 32 CABG seems superior than PCI. Although not reported, my guess is that those having a core-lab SYNTAX score over 32 had a mean score close to it, but still higher than 32.

3-year follow-up: Although not short, longer term is preferred. This data will come out soon.

Definite non-CV death higher in PCI group while CV death similar: This trial, as many other well-conducted trials in cardiovascular disease, defined non-CV death as a death clearly unrelated to a cardiovascular event. Deaths with unknown cause are not considered non-CV deaths. Still, this study observed more non-CV deaths in PCI. Given the small number of events (statistical fragility), lack of a clear biological explanation, and inconsistency with prior literature, this is likely a chance finding. To see another possible example of this please see the DAPT study.

How this impact the results? Since the primary outcome encompasses “all-cause mortality” (and not CV mortality) this may favour CABG as well assuming this is a chance finding. There is still no consensus in cardiovascular research whether to use all-cause mortality vs CV-mortality. Using all-cause mortality has the advantages of increased statistical power (larger absolute number of events) and avoiding the risk of miss-classification of a CV-death as non-CV-death. On the other hand, using strict definition of CV-death (unknown deaths are considered cardiovascular) then this outcome is more specific reducing the “noise” in this case caused by a likely spurious increase in non-CV deaths.

NOBLE trial

This study among people with left main disease evaluated the efficacy of PCI vs CABG for death-stroke-myocardial infarction plus repeat revascularization as primary outcome with maximum follow-up of 5-years. The study showed superiority of CABG, mainly driven by repeat revascularizations.


Repeat revascularization in primary outcome: This is a problematic outcome in PCI vs CABG trials for many reasons.

A) Is out of proportion vs the other components of the primary outcome (death, stroke and myocardial infarction).

B) The majority of repeat revascularizations (>80% in SYNTAX) are a repeat PCI, so two PCI is worst than one CABG?

C) Patients usually prefer multi-vessel PCI rather than one CABG even when told that CABG is associated with lower mortality at 1-year and up to threefold higher repeat revascularizations.

D) Atypical symptoms after PCI usually raise the concern of restenosis and patients undergo stress testing or even sent directly for invasive angiography, while post CABG patients atypical symptoms usually adjudicated to non-coronary causes like pericarditis, muscular, sternum incision, etc. Also, multimodality guidelines recommend systematic use of stress testing more frequent after PCI vs CABG. Altogether, post PCI patients more likely to undergo stress testing than post CABG counterparts.

G) Bypassed coronary territories progresses 3 to 10 times vs non-bypassed territories, so when graft failure occurs patients there is a high likelihood of hostile native targets for revascularization and treated medically.

All these limitations, among others, make repeat revascularization a likely biased outcome measure in PCI vs CABG trials, causing trouble specially when included in composite outcomes at the same level as death, stroke or myocardial infarction.

Peri-procedural MI (even large) not systematically measured: Is likely out of question that large myocardial infarctions (independent of mechanisms, whatever is secondary to plaque rupture, plaque erosion, spontaneous coronary dissection, coronary embolism, or peri-procedural) is prognostically important. In NOBLE there was no systematic screening of enzymes or ECG after index revascularization.

EXCEL observed that >70% of post CABG patients had enzyme elevation, and CK-MB ULR x 10 or Tnt x 70 ULR (large MIs) was 6% in CABG and 3% in PCI. Also, large MIs were associated with 2 to 3 fold increase in mortality during follow-up. In other words, a patient with CABG may have had CK-MB x 10 and missed, but a patient with initial PCI readmitted three months later with normal CK-MB but small elevation in troponins is considered a myocardial infarction. This aspect likely be a main driver of the difference MACE observed between NOBLE and EXCEL within 30 days and impacts long-term outcomes as well.

Frist generation DES used in some patients: Second and newer generation drug eluting stents have shown to be superior than newer generation stents.

Methodological comments

How much non-inferiority?

We do not know current benefit of CABG (the historical standard) over medical treatment, so difficult to put a line and say: we are sacrificing this X amount of uncertainty of effectiveness (this is the non-inferiority margin) of CABG to justify the benefit of lower morbidity with PCI. Therefore, how much uncertainty we need to put the limit and accept PCI as the standard for LM disease? this is unknown to me.

EXCEL longer-term data (5+yrs) lacking, but is not a comparison to see what is better, and even if non-inferiority is not maintained, inferiority (significant harm of PCI in death-stroke-MI) is very unlikely to happen given data so far (maybe wrong about this, time will say), so question of how much efficacy of long term CABG efficacy can be sacrificed may remain even after 5 years of follow-up.

Immortal bias in landmark analysis

To make this simple, if you have a randomized trial and you analyze any period that includes randomization time (from time zero to 1 month, or from time zero to two years) you have a randomized trial. If you do landmark analysis that excludes this period (from 3 months to end of follow-up, from 6 months to 12 months) randomization has been violated and confounding likely present.

Why? Because randomization created a prognostic balance at time zero (randomization time). If we exclude people from any analysis (in this case a landmark analysis from 1-month to 12 months) that cannot be included because they had the outcome, dropped the study or were lost to follow-up, then this initial prognostic balance created by randomization has been damaged.

In other words, in EXCEL there was a statistically significant reduction in deat-stroke-MI from time zero (randomization) to 30-days, that means the survivors of 30-days from PCI likely be composed of higher risk people compared to CABG. To be included in a landmark analysis that excludes time-zero (randomization time) you need to be alive without the outcome of interest, that why is called “immortal bias”: you need to be immortal to be included in such analysis.

Violation of proportional hazards assumption?

Survival analysis with Cox-regression assumes proportional hazards. In simple words, the hazard ratios between intervention and comparator should remain stable though the follow-up time. If one intervention is more effective than the other and then switches, then we may be violating this assumption. There are different ways to deal with this, all statistically complex, but you need to be aware of that when interpreting evidence. EXCEL showed better short-term outcomes with PCI, while probably better outcomes long-term with CABG. 5-year data vital to better understand this.

Research in context

What do the meta-analysis say?

No difference for death-stroke-MI with reasonably narrow confidence intervals, even including the outlier (NOBLE, the only one suggesting superiority of CABG over PCI) that did not measure periprocedural MI.

How should I manage my patient with left-main and SYNTAX score under 32?

Current 2018 ESC revascularization guidelines recommend PCI as IA for left main and SYNTAX score under 22, IIA for intermediate SYNTAX score and not recommended if SYNTAX score over 32.

Based on this sounds reasonable the following approach

1. Heart team (Minimum: clinical card, surgeon, and interventional cardiologist)
2. Shared decision making: involve the patient in the discussion and incorporate values and preferences
3. Be aware of your center results (CABG and PCI) for external validity of these trial results

One Reply to “PCI vs CABG for left main disease: EXCEL Vs NOBLE”

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s