⚡ IS GAMMA-TILE AFTER SURGICAL RESECTION OF BRAIN METS A NEW STANDARD OF CARE? A new brain mets study was published in @JCO_ASCO yesterday and discussed at #ASTRO2026. Hold your horses 🐎 The ROADS ph. 3 RCT of GammaTile (GT) vs postop stereotactic RT (SRT) after resection of brain metastases reported better surgical-bed control, better surgical bed recurrence-free survival, and even better OS with GT. Sounds practice-changing. “Sounds” is the key word here. Let’s look at what actually happened. 🧵👇 First, the trial: 👥 230 pts randomized pre-op: surgery + GT vs surgery + postop SRT; any additional intact brain mets still received SRT in both arms (~40% each). ⏱️ Median f/u: 12.9 months. Headline efficacy results: 🎯 Surgical bed recurrence (SBR): 1 vs 12 📉 12-mo SBR: 1.3% vs 15.4% ⏳ Time to SBR: not reached (NR) vs 17.4 mo HR 0.06 📊 Surgical bed recurrence-free survival (SB-RFS): NR vs 10.9 mo HR 0.48 LET’S UNDERSTAND SOME OF THE ISSUES WITH THIS STUDY: 1⃣ SB-RFS sounds like a LC endpoint. It is not. SB-RFS was defined as surgical bed recurrence OR death from any cause, whichever occurred first. So a patient can have a perfectly controlled surgical cavity, die from extracranial metastatic disease, and still have an SB-RFS event. So, “SB-RFS not reached vs 10.9 months” ≠ “LC NR vs 10.9 months.” That is an important distinction. 2⃣ The marked OS difference is the biggest red flag here. 🚩 Median OS 42.5 vs 17.6 months, HR 0.59, p=0.032 In metastatic disease, changing the local RT technique to a single resection cavity would not ordinarily be expected to more than double median OS without a very compelling mechanism. And this trial doesn’t provide an obvious one: no difference in LMD, distant brain failure, or radiation necrosis. 3⃣And here is where it gets even more interesting. 18% of patients assigned to postoperative SRT never actually received it. Credit to the authors for performing a prespecified per-protocol analysis restricted to patients who received assigned treatment. And voilà: OS: 42.5 vs 20.2 months, HR 0.73, p=0.216 ➡️ The OS difference is no longer statistically significant. 4⃣ Now back to the actual LC result. 1 vs 12 surgical-bed recurrences is impressive. There is clearly a LC signal here, and we shouldn’t dismiss it. But the headline HR of 0.06 is based on only 13 total SBR events, with a gigantic 95% CI of 0.01–0.46. So while the benefit may be real, the magnitude of benefit is much less certain than the headline HR suggests. 5⃣ The authors also argue that GT addresses the problem that some patients never make it to postoperative SRT. But of the 18 patients who didn’t receive SRT, nearly 70% died, entered hospice, or developed LMD. Those patients generally would not be candidates for postop focal RT regardless of technology. This is not really a GT vs SRT problem. 6⃣ And the “one and done” concept isn’t actually one-and-done for everyone: ~40% of patients in each arm had >1 brain metastasis and still required postop SRT. So for roughly 4 out of 10 GT patients, GT did not eliminate the need to come back for SRT anyway. 7⃣ Finally, there is a workflow question that remains unanswered, and obviously this was not the primary scope of ROADS. GT eliminates postop cavity simulation/planning/treatment. But it shifts the work to pre-op neurosurgery coordination, tile estimation/ordering, brachytherapy/physics, OR logistics, source handling, intraoperative involvement, radiation safety, and post-implant dosimetry. Does the magnitude of the oncologic benefit justify a potentially more resource-intensive treatment? 🧠 MY TAKE-HOME POINTS: 👏 Congratulations to the investigators. This is an important RCT, and the LC signal is real and deserves serious attention. But the story is much more complicated than the headline. ⚠️ Only 13 actual SBR events ⚠️ Very wide CI ⚠️ Death-inclusive SB-RFS endpoint 🚩 Unexplained OS signal that disappears in the per-protocol analysis ⚠️ SRT non-completion issue ⚠️ Unanswered workflow/value questions And this is where I disagree with the authors’ conclusion: “R+TBRT should be considered a standard-of-care option…” I don’t think that conclusion can be drawn from ROADS alone. The trial supports a meaningful local-control signal, but the limitations above make it premature to elevate GT to a standard-of-care option