NOBLE at 10 years: PCI versus CABG for unprotected left main disease

Summary

NOBLE randomized 1201 patients with unprotected left main disease, eligible for both treatments by heart-team consensus, between 2008 and 2015. 598 were assigned to PCI and 603 to CABG; 592 per arm entered the intention-to-treat analysis. Most received second-generation drug-eluting stents.

The original design tested non-inferiority on a MACCE composite at 5 years. At 10 years, the reported primary endpoint is all-cause mortality. Vital status was available in 98% of patients.

In the acute coronary syndrome subgroup, the HR was 0.57 (0.32 to 0.99), with an interaction p of 0.049. The Bayesian-Weibull analysis, with an informative prior HR of 1.10, gave a posterior median of 0.95 (95% credible interval 0.74 to 1.18).

The published conclusion is that PCI is “equally as safe as” CABG.

Strengths

• 98% follow-up at 10 years.
• The 0 to 5 and 5 to 10 year landmark analysis tested the hypothesis of late mortality catch-up with PCI raised after EXCEL. Both windows were neutral.
• The Bayesian sensitivity analysis used a conservative prior (HR 1.10, from SYNTAX and PRECOMBAT) and the result stayed neutral. Caveat: the prior comes from first-generation stent trials, applied to a comparison with mostly second-generation stents.

Methodological issues

• Analysis plan not prespecified. The authors state that the statistical plan for the 10-year endpoint was not in the protocol. They reused the 5-year plan. In practice, the 10-year primary endpoint is post hoc.
• Not powered for mortality. The original power calculation was for MACCE. With 281 deaths, the Schoenfeld formula gives about 80% power to detect an HR of 0.72. For an HR of 0.74, power is about 71%; about 346 events would be needed. The authors acknowledge that a type 2 error cannot be excluded.
• Equivalence claim without an equivalence test. The trial ran a superiority test. p = 0.56 means only that no difference was shown. “Equally as safe” requires a prespecified margin, a power calculation for it and a CI entirely inside the margin. With a hypothetical margin of 0.87 to 1.15, the upper bound of 1.18 would fall outside and equivalence would fail.
• Fragile ACS subgroup. The interaction p of 0.049 sits at the threshold. Nine subgroups were tested without correction. The chance of at least one false positive is $1 - 0.95^9 \approx 37\%$. The Bonferroni α would be about 0.0056, nine times lower than 0.049. CABG mortality in ACS was 31%, against about 22% in pooled SYNTAX and PRECOMBAT. Median delay to CABG in ACS was 6 days, against 1 day for PCI; the explanation is plausible, but it is post hoc reasoning on a borderline interaction.
• Proportional hazards tested only with log-log plots. No Schoenfeld residuals or formal test were reported. In the ACS subgroup, the curves separate late. For that subgroup, RMST would be more appropriate than the HR.
• Confounded IVUS substudy. Post hoc, the mortality HR against CABG was 0.79 (0.61 to 1.04) with final IVUS and 1.31 (0.94 to 1.83) without final IVUS. IVUS use varied by period, operator and center.
• No non-fatal events after year 5. MI, stroke and repeat revascularization were not collected from years 5 to 10. That window has only mortality, the least sensitive endpoint for comparing PCI and CABG.