onc brain

The longer read

Post-operative cavity SRT became the standard for resected brain metastases because it beat observation for surgical bed control (Mahajan et al.) without costing survival against WBRT. ROADS does not challenge that logic. It challenges when and how the dose is delivered. The comparator was not a straw man: over nine in ten control pts got fractionated SRT, and the 15.4% one-year surgical bed recurrence sits inside the range of contemporary multi-institutional series. Beating that arm with one SBR event against twelve says more than beating observation would.

Two mechanisms are in play and the trial cannot fully separate them. One is dosimetric: a permanent low-energy implant against the cavity wall gives a higher effective dose at the margin with steep fall-off. The other is adherence and timing: 18 pts assigned SRT never received it, a non-completion rate matching prior prospective data (Brennan et al. reported 20%). If the whole effect were adherence, the per-protocol analysis should have shrunk it. Time-to-SBR did not move there (HR 0.06), and SB-RFS weakened only modestly (HR 0.57). That pattern argues the implant controls the bed better even when SRT is actually delivered, while the access argument adds on top.

Confidence should be tempered in specific ways. The co-primary structure was rewritten in May 2025 after the original SB-RFS definition was found to censor death. The authors report no outcome data had been reviewed, and the correction moves the endpoint toward the conventional definition, but a mid-trial endpoint change in an open-label trial is the kind of thing a skeptic will cite. The hazard ratio of 0.06 rests on 13 events with a lower CI bound of 0.01; the direction is secure, the point estimate is not. Seed artifacts on MRI mean central reviewers could plausibly infer assignment. Median follow-up of 12.9 months is short for late cavity events, and radiation necrosis after permanent brachytherapy may not have fully declared itself, although the 12-month RN rates were close (5.3% vs 5.7%).

The OS result (42.5 vs 17.6 months, HR 0.59) should not be taken as a finding. A local therapy for one lesion producing that survival gap, confined to the single-metastasis subgroup and losing significance in the per-protocol set (HR 0.73, p=0.2156), is more likely imbalance, chance, or differences in systemic therapy timing than a direct effect. The trial did not collect enough systemic therapy data to test the authors' suggestion that avoiding post-op SRT gaps let drug therapy resume sooner. Treat it as hypothesis-generating.

The decision this moves is how to treat the cavity for a newly diagnosed, resectable 2-7 cm metastasis: implant at surgery versus a separate SRT course three weeks later. It requires neurosurgical and brachytherapy capability at the time of craniotomy, which will limit adoption more than the evidence does. It does not answer how TBRT compares with pre-operative SRT, which is being tested in two ongoing phase 3 trials (NCT03741673, NCT05438212). TBRT has a structural advantage there, since the tile goes in only after intra-operative pathology confirms metastasis, and ROADS itself enrolled pts who turned out to have glioblastoma. Longer follow-up for necrosis and a replicated survival analysis are what would convert this from a strong challenge to a new default.