SELUTION4ISR: sirolimus-eluting balloon versus DES or balloon angioplasty for in-stent restenosis

Summary

SELUTION4ISR randomized 418 of 1508 screened patients to a sirolimus drug-eluting balloon (DEB) or a “usual care” control designed to mirror NCDR 2009–2017. The control was contemporary DES (everolimus or zotarolimus) in 80% and balloon angioplasty in up to 20%. The operator chose DES or balloon before randomization. The balloon cap was reached after 176 patients; after that, randomization was DEB vs DES.

Lesions up to 26 mm, reference diameter 2.5–2.6 mm, 78% with a single prior stent layer and 88–92% with prior DES. Randomization took place after successful predilation, a requirement that failed in 15 of 1508 screened (1%). Intracoronary imaging in 98% of both arms.

P for interaction by selected control = 0.0026. The noninferiority margin was 10 percentage points, with a 97.5% threshold.

Strengths

• Per-protocol primary analysis, with the rationale stated: intention to treat is anticonservative in noninferiority.
• Declared noninformative priors, and tipping-point and competing-risk sensitivity analyses.
• Funder explicitly excluded from analysis, interpretation and writing.
• Event committee blinded to allocation and independent core lab.
• The 10-point margin was derived as 50% of an estimated DES-over-balloon effect of at least 20 points: an explicit M1/M2 construction.
• The authors report how many patients failed predilation (15 of 1508), which allows the enrichment concern to be assessed.
• The failed secondary, the interaction, the margin width and the screen failures are in the paper, several in the abstract and the conclusion.

Methodological issues

• Margin anchored on another contrast. M1 comes from the DES vs balloon contrast, but the comparator tested is 80% DES. Against the observed control rate (14.5%), the 10-point margin allows a risk ratio of 1.69. M1 was taken as a minimum estimate: a conservative M1 should shrink the margin, not widen it.
• Study resolution equal to the margin. Reconstructed from the published data, the minimum detectable effect at the design power (85%) is 10.9 points; at 80%, 10.2 points. The authors state that an upper limit below 8.9% “would have required a much larger trial.”
• Margin sensitivity and fragility. Against the 97.5% threshold, the trial fails at any margin of 8 points or less. At a 5-point margin the posterior probability is 81%. The fragility index is about 3 events in 390 patients. Realized power for noninferiority was 78% at a true difference of zero and 62% at the observed difference.
• Failed gated secondary. In single-layer ISR, DEB vs DES had an RR of 2.18 and an NNH of about 13. This was a formal test, gated on the primary succeeding. It is not a subgroup, so multiplicity arguments do not apply.
• Blended comparator. The two halves of the control gave opposite answers, with intervals excluding zero. The headline result is their arithmetic average. “Noninferior to usual care” does not describe a therapy a clinician can choose. The comparison by control is not randomized (chosen before allocation, not stratified) and the strata differ in complexity: it is exploratory, but it points the same way as the gated test.
• Difference concentrated in postprocedure MI. Postprocedure target vessel MI (13 vs 7; +2.9, CI −1.5 to 7.6) accounts for almost all of the composite difference. The interval crosses zero: it generates a hypothesis, it does not establish one.
• Different acute result and DAPT. The arms were identical after lesion preparation (MLD 1.9 vs 1.9 mm; DS 23.8% vs 22.8%) and diverged after device deployment (MLD 2.3 vs 2.5 mm; DS 15.6% vs 11.4%). The protocol prohibited post-DEB optimization. Recommended DAPT was 6 months after DES and 1 month after DEB or balloon, although DAPT or P2Y12 monotherapy remained high at 6 months in all groups.