Although all-cause mortality was a secondary endpoint, unadjusted for multiple comparisons, there was a significant gap of 3.1% (13.0% in PCI and 9.9% in CABG; OR, 1.38; 95% CI, 1.03-1.85) in all-cause mortality at five years in favor of CABG. This was considered by the EXCEL authors to be due to chance alone,10 based on the absence of a significant difference in cardiovascular mortality. However, it should be borne in mind that the adjudication process, i.e. determining the causes of death, in a non-blinded trial, can be subject to ascertainment and misclassification biases. The two most recent meta-analyses of all LMD PCI vs. CABG RCTs showed no statistically significant difference in mortality.11,12 However, if the results of EXCEL and NOBLE are pooled, there is a higher risk of death for PCI and therefore the question remains unresolved.13

The EXCEL protocol defining postprocedural MI used the same CK-MB thresholds for CABG and PCI (>10× upper reference limit [URL] in isolation or >5× URL with new Q waves or other confirmation of ischemia, as published by the Society for Cardiovascular Angiography and Interventions [SCAI]).14 It is well known that CABG is associated with the release of myocardial biomarkers due to aortic clamping and cardiac manipulation. This definition differs from those used in previous trials comparing PCI with CABG and has been shown to overestimate MI after CABG compared to other definitions. Cho et al. showed a substantial difference in the rates of periprocedural MI after PCI and CABG, depending on the MI definition used: 18.7% vs. 2.9% using the Second Universal Definition; 3.2% vs. 1.9% by the Third Universal Definition; and 5.5% vs. 18.3% by the SCAI definition.15 The use of the SCAI definition may have affected the primary endpoint and may explain why five-year MI rates in the CABG arm of the EXCEL trial7 (9.1%) are two to five times higher than in other LMD trials: 4.8% in SYNTAX,4 1.7% in PRECOMBAT and 3% in NOBLE.8 In a post-hoc analysis, Gaudino et al.16 showed that if the unpublished, but prespecified, universal definition had been used for procedural MI in EXCEL, the five-year composite endpoint would have demonstrated clear superiority of CABG (OR, 1.40; 95% CI, 1.09-1.81; p=0.009).

In the original trial protocol,17 and in successive versions (all including the Universal Definition as one definition of MI), the EXCEL investigators left the trial's specific MI definition open. The protocol defining MI was only published in 2016,5 using the SCAI definition,14 two months before the three-year results were published in the NEJM,18 which is most unusual. Finally, in the five-year results the EXCEL investigators failed to include the data on MI according to the prespecified secondary endpoint using the Second Universal Definition.

In a recent letter to the Editor of the NEJM,19 the EXCEL leaders reported that the large discrepancy between the two definitions in the number of CABG-related procedural myocardial infarctions (57 by the protocol definition vs. 13 by the universal definition) was mostly due to a lack of confirmatory evidence of myocardial ischemia that is required in the universal definition. However, reanalysis of the primary outcome using the protocol-prespecified Second Universal Definition is still unpublished.

The NOBLE trial8 appears to be less controversial, although critics have pointed out some flaws. An interim change of protocol, including primary outcomes reported and follow-up, due to lower-than-expected rates of MACCE, put the trial's internal validation at risk. The use of 11% first-generation along with second-generation (89%) DES may have influenced the PCI results, as the rate of stent thrombosis was significantly higher in NOBLE than in EXCEL (3% vs. 0.7%). Additionally, the inclusion of repeat revascularization in the composite endpoint may be debatable, since it is considered to be less clinically important compared to death, MI or stroke, although both SYNTAX1 and PRECOMBAT also included it in the composite endpoint.

The EXCEL investigators concluded that there was no statistically significant difference between PCI and CABG in the primary outcome at five years (difference, 2.8%; 95% CI, 0.9-6.5%). However, the 95% CI interval for the difference in MACCE rates between PCI and CABG is wide, ranging from 0.9% to 6.5%, mostly above 0%, favoring CABG. Importantly, there is an increasing gap over time between the two treatments after the second year in Kaplan-Meier curves of the primary endpoint, diverging in favor of CABG.

EXCEL was designed as a non-inferiority trial, and the three-year results were reported with an agreed margin of non-inferiority of 4.2%.5,18 If the protocol-defined, non-inferiority type of analysis, as reported at three years, had also been used for the five-year report, PCI was not non-inferior to CABG, as the 6.5% upper bound of the confidence interval would have exceeded the non-inferiority margin of 4.2%.

In all trials, results are only valid for populations with the same anatomical and clinical characteristics, the same techniques and expertise used in the selected centers. Extrapolation of trial results to wider, real-world populations could miss the target in a large proportion of patients, with potentially unpredictable outcomes. NOBLE and EXCEL7,8 included relatively young patients (around 66 years), with low comorbidity and favorable anatomical and clinical characteristics for both procedures. To put this in perspective, of the 2905 patients with >50% left main stenosis initially screened in EXCEL, only 1905 were randomized. The other 1000 patients were included in registries. Non-eligibility criteria included coronary disease that was considered too extensive (269/995; 27%) or too complex for PCI (317/995; 32%) and in 18% (183/999), specific coronary lesion characteristics or patient characteristics prevented randomization to CABG. This means that more than 30% (1000/2905) of patients with LMD were not represented in EXCEL. Worryingly, this issue of the external validity of RCTs, the backbone of translational and evidence-based medicine, is still, unforgivably, lacking in the guidelines.