onc brain

The longer read

The cleanest way to read this trial is that it separates two claims that reflex practice usually bundles together. Upfront cranial RT does what it is supposed to do to the brain: intracranial progression at two years was 21.7% versus 50%, sub-HR 0.35 (0.21-0.59), with death handled as a competing event rather than censored, which is the correct analysis and makes the estimate more believable rather than less. The second claim, that controlling the brain earlier translates into living longer, is not supported here, and the direction of the survival data (2y OS 48% upfront vs 60% delayed, HR 1.45) is the opposite of what an advocate for upfront treatment would predict.

That inversion is the whole story, and it deserves scepticism in both directions. OS was a secondary endpoint in a 208-patient trial, so the comparison is underpowered and the confidence interval around an HR of 1.45 is almost certainly wide enough to include no difference. A reader who wants to dismiss the survival signal has a fair statistical argument for doing so. But dismissing it does not rescue the case for upfront RT, because the fallback position then becomes that upfront RT buys intracranial control with no survival return, and the toxicity ledger is not empty: radiation necrosis was reported at roughly 6% upfront and absent in the delayed arm. A strategy that costs necrosis and returns a surrogate is a hard sell when the alternative keeps the same treatment available on demand.

What makes this trial different from the historic cranial RT literature is the competing intracranial therapy. In the WBRT era, deferring radiation meant deferring the only thing that acted on brain disease. Here the delayed arm was not untreated: it was on a TKI with real CNS activity, plus q3m MRI for the first year. The intracranial progression rate in that arm, 25.7% at one year, is the number that licenses the strategy, because it says most patients did not need cranial RT in the first year at all. The trial is therefore less a test of radiation than a test of whether modern systemic therapy plus disciplined imaging can safely hold the brain until it declares itself.

The biggest gap is that the source gives no RT prescription. Whether this was whole-brain or stereotactic, at what dose and fractionation, and to what target volume, determines almost everything about how the result transfers. A necrosis rate near 6% reads very differently as an SRS complication than as a WBRT one, and a whole-brain comparator would also raise the question of whether the survival direction reflects neurocognitive or performance-status decline rather than cancer biology. Neurocognition and patient-reported outcomes were prespecified secondaries and are not in the source; those endpoints would either explain the OS direction or undercut it, and until they are reported the mechanism behind the survival signal stays unresolved.

For the reader who has to act before the full publication, the defensible position is that a deferred strategy is now supported by randomised evidence in this specific population, provided the surveillance imaging is actually delivered on schedule. The trial does not license deferral where MRI follow-up is unreliable, and it does not extend to symptomatic disease, which was excluded by design.