ROADS
ForResected brain metastasis >2 cm, post-op cavity radiation candidates
NR vs 17 mo
GammaTile vs SRS; no HR, CI, or p reported in source
TL;DRSurgical bed recurrence 1% with GammaTile brachytherapy vs 12% post-op SRS in resected brain mets >2cm, N=230.
The RT read is the bed-control mechanism: GammaTile puts dose in the cavity at resection, closing the gap where post-op SRS fails in cavities >2 cm, with median time to bed recurrence not reached vs 17 mo. LMD was 10% GT vs 3% SRS, so the trade is bed control against meningeal seeding when picking the cavity strategy.
In a resected brain metastasis larger than 2 cm being planned for cavity radiation, this questions post-op SRS as the default bed strategy; it does not extend to intact metastases, cavities under 2 cm, or pts already carrying leptomeningeal disease.
Bed control is the read: median time to bed recurrence not reached vs 17 mo with post-op SRS, in cavities >2 cm where SRS control is weakest, with radiation necrosis flat at 8% GT vs 7% SRS. LMD 10% GT vs 3% SRS is the counterweight when choosing the cavity strategy.
The cavity radiation decision moves into the operation itself: GammaTile is implanted at resection, so a met >2 cm being taken out becomes a pre-op discussion about tile placement rather than a post-op SRS referral. The 10% vs 3% LMD signal is the intraoperative trade to weigh.
| Endpoint | GammaTile | SRS |
|---|---|---|
| Time to surg bed recur | NR | 17 mo |
| Surg bed recur FS | NR | 11 mo |
| 2 yr OS | 62% | 36% |
10 details
Randomized trial of GammaTile brachytherapy vs post-op SRS after resection of a brain metastasis, N=230, reported as final results at ASCO 2026 (Weinberg). Randomization ratio, number of sites, and follow-up duration not reported in source.
Resected brain metastasis >2 cm, the size band where post-op cavity SRS control is weakest. Histology mix, number of brain metastases allowed, systemic disease status, and performance status not reported in source.
Experimental arm is GammaTile, a Cs-131 collagen tile implanted in the cavity at the time of resection, so dose starts without the post-op delay. SRS dose, fractionation, cavity margin, and time from surgery to SRS are not reported in source, and those are exactly the parameters that decide whether the control arm reflects the reader's own practice.
Primary I: time to surgical bed recurrence. Primary II: surgical bed recurrence-free survival. 2 yr OS is reported alongside them; the source does not state whether OS was a prespecified secondary.
Both primaries favor GammaTile with medians not reached vs 17 mo and 11 mo. No HR, confidence interval, or p-value appears in the source.
Radiation necrosis 7% SRS vs 8% GT, essentially flat. Leptomeningeal disease 3% SRS vs 10% GT is the signal that cuts against the arm winning on bed control; timing and denominators not reported in source.
Post-op cavity SRS became standard on N107C/CEC.3 and the MDACC randomized trial, both of which traded whole-brain neurocognitive toxicity for weaker bed control, with failure concentrated in larger cavities. ROADS attacks that residual failure directly rather than re-litigating whole-brain RT.
The 2 yr OS separation, 62% vs 36%, is far larger than bed recurrence alone (12% vs 1%) would mechanistically support, which points at arm imbalance, differential salvage, or systemic therapy that the source does not report. The trial is also inherently unblinded, and the LMD excess in the GammaTile arm has no reported timing to judge whether it is procedure-related seeding.
If the bed-control numbers hold in the full report, the cavity strategy for a large resected met becomes a surgical-planning decision made before the operation rather than a radiation-planning decision made after it. The OS claim should wait for the manuscript.
Randomized, both primaries reported, final analysis. Verdict held below practice-changing: conference-slide source with no HR, CI, or p-value, and unexplained LMD excess.
- Is the 2yr OS separation confirmed with hazard ratios and cause of death?
- Does the leptomeningeal excess with GammaTile reflect seeding or longer survival?
- Does the benefit hold against fractionated post-op SRS rather than single fraction?
📚 Sources · 🐦 1 tweet
🚨🚨 ASCO 2026 Final Results Randomized trial resected brain met Brachytherapy vs Post-Op SRS🚨
— PDBrown (@PDBrownOnc) May 30, 2026
- Incredible Surg Bed Control with Brachy (↑↑OS as well)
- Surg bed recurrence 12% SRS vs 1% GammaTile pic.twitter.com/PCTsCluyUd
The longer read
The cleanest way to read ROADS is as an attack on the known weak point of the current standard rather than a challenge to it. Post-op cavity SRS displaced whole-brain radiotherapy on the strength of neurocognitive preservation, and both randomized trials that established it accepted worse intracranial control as the price. Bed failure in those trials was not evenly distributed: it concentrated in large cavities, which is precisely the population ROADS enrolled. A trial restricted to resected metastases over 2 cm is therefore sampling the part of the distribution where the standard is most vulnerable, and a bed-control win there is closer to confirmation of a known gap than to a surprise.
What makes the mechanism plausible is timing and geometry rather than dose alone. GammaTile is placed at resection, so the cavity receives dose immediately and conforms to the resection surface as it collapses, while post-op SRS waits for healing and then targets a cavity whose shape has already changed and whose margin expansion is a compromise between coverage and necrosis risk. Median time to bed recurrence not reached against 17 mo, and bed recurrence-free survival not reached against 11 mo, are consistent with removing that delay and that geometric mismatch. The radiation necrosis rates, 7% with SRS and 8% with GammaTile, argue that the gain did not come from simply pushing more dose into brain tissue.
The survival claim is where scepticism should concentrate. A 2 yr OS of 62% against 36% is a very large separation to attribute to a difference in bed control of 12% versus 1%, because most pts with resected brain metastases die of systemic disease or of distant intracranial failure rather than of recurrence at a single resection cavity. For the OS gap to be causal, salvage after bed failure would have to be far more damaging than it is generally observed to be, or bed recurrence would have to be acting as a marker for something else. The alternatives are more mundane: baseline imbalance in systemic disease control, differences in what pts received after progression, or a modest sample where survival is not the endpoint the trial was sized for. None of that is resolvable from a conference summary that reports no hazard ratios, no confidence intervals, and no p-values for any endpoint.
The leptomeningeal numbers deserve equal attention in the opposite direction. Leptomeningeal disease of 10% with GammaTile against 3% with SRS is the failure mode a reader would predict from placing radioactive material into an open cavity, and it is the one result that would constrain adoption if it holds. Without timing, denominators, or a competing-risk framing, it is impossible to know whether this reflects seeding, longer survival exposing pts to a later failure mode, or ascertainment differences. Those two possibilities point to opposite clinical conclusions, and the source does not separate them.
What would have to be true for this trial to be wrong about its central claim is fairly narrow: the bed-control difference is large, measured on a prespecified primary, and biologically coherent. What would have to be true for the OS claim to be wrong is much broader, and until the manuscript reports the hazard ratios, the arm characteristics, and the causes of death, the defensible position is to treat ROADS as strong evidence about the cavity and unproven evidence about survival.