onc brain

The longer read

The comparator is the first thing to check when an experimental arm wins by this margin, and here it holds up. The R+SRT arm was contemporary practice: multi-fraction cavity SRT in 92.2%, with a 12-month cumulative incidence of surgical bed recurrence of 15.4%. The authors set that against the 84-87% 12-month freedom-from-SBR reported in multi-institutional fractionated series, and it is ahead of the 72% Mahajan et al. reported with single-fraction cavity SRT. The HR of 0.06 is therefore not manufactured by an underperforming control.

What should temper the read is how few events carry it. One SBR in the tile arm against 12 with SRT gives a CI of 0.01-0.46, at a median follow-up of 12.9 months. The direction is not in doubt, the size is imprecise, and later or marginal failures after a brachytherapy source with a different dose delivery profile cannot be excluded at this follow-up. The SB-RFS co-primary is a different kind of evidence: with 30 vs 45 events against 1 vs 12 recurrences, most of what it counts is death, so it largely restates the survival difference instead of adding independent support for local control.

The OS result deserves the most skepticism. A median of 42.5 vs 17.6 months from swapping one focal cavity technique for another, with no difference in distant brain failure (HR 0.93) and none in LMD, is hard to attribute to local control alone. The mITT control arm includes 18 pts who never received SRT, 8 of whom died and 3 of whom entered hospice before it could be delivered. Those pts mark early systemic decline as much as a treatment failure. Restricted to pts who received assigned treatment, the OS HR moves to 0.73 with p=0.2156, while the SBR HR stays at 0.06. The authors argue the per-protocol cut introduces immortal time bias, which is fair, but the truth plausibly lies between the two analyses, and they themselves say the OS finding requires validation. The upper bound of the tile-arm median was not estimable.

Two design points cut in opposite directions. The SB-RFS definition was amended in May 2025 after the protocol was found to censor death, which is why there are two co-primaries; the authors state no outcome data had been reviewed by then, and time-to-SBR is the endpoint a radiation oncologist would have wanted anyway. Central review was by neuroradiologists not told the assignment, but seed artifact makes the tile arm identifiable on MRI, so blinding was nominal. RN was site-reported at 5.3% vs 5.7% at 12 months, and separating necrosis from recurrence in a seeded cavity is exactly where knowledge of the arm matters. The composite SBR-or-RN HR of 0.28 offers some protection against that misclassification.

The decision this moves is narrow: for a metastasis already going to resection, whether the cavity is treated in the operating room or at a median of 27.0 days afterward. A 17.8% non-completion rate for post-operative SRT, in line with the 20% Brennan et al. reported and the 73% completion Yeboa et al. reported, means part of the tile advantage is simply that every pt gets treated, a property of the pathway that should transfer to practice. It leaves unresected lesions untouched, and it does not say how tiles compare with pre-operative SRT, under test in two randomized phase 3 trials (NCT03741673, NCT05438212). The tile prescription dose is not in the source text, which limits any attempt to separate a dose effect from a timing effect.