onc brain

The longer read

Elective nodal irradiation has never been tested against omission in a randomised head and neck trial. It is inherited practice, built on surgical series showing occult nodal disease in undissected necks, and every subsequent de-escalation effort in this disease has attacked dose or systemic therapy rather than the elective volume itself. That makes the volume question the last large untouched source of toxicity in definitive chemoradiotherapy, because the elective levels are where most of the parotid, submandibular, constrictor and oral-cavity dose is deposited. A pooled long-term read of two prospective INRT protocols is therefore worth more attention than its sample size alone would justify.

The result that carries the argument is not overall survival. It is the absence of solitary elective nodal recurrence at 5 years, because that is the only failure pattern attributable to the omission itself. A regional recurrence adjacent to a treated node, or a recurrence occurring alongside local failure, would have happened under standard volumes too. The 3-year regional cumulative incidence of 4.3% sits inside the range conventionally-treated cohorts report, and the local recurrence figure of 9.5% is the larger number, which points the failure burden back at the gross primary rather than at the untreated neck. For a radiation oncologist that inverts the usual anxiety about this strategy.

Two features should temper confidence, and neither is sample size. The first is the AI-assisted nodal-selection step. The trials did not simply omit elective levels, they omitted them after an algorithm flagged suspicious nodes on staging PET/CT and neck CT, and the source describes no comparison between that step and unaided physician contouring. If the nodal selection is doing the work, the transferable intervention is the selection tool, not the omission, and a department adopting the volume reduction without it is not reproducing the trial. The second is that a patient-level pool of two protocols assumes their INRT definitions were interchangeable, which the source does not establish. Different margin conventions around an involved node produce materially different high-risk coverage while both being called involved-node treatment.

The toxicity claim also deserves a narrower reading than it will get. Mean composite MDADI of 84.9 at 12 months with no significant decline is a good number in absolute terms, but a single-arm swallowing score has no counterfactual attached. The entire rationale for omitting elective volumes is that it should produce better function than treating them, and this dataset cannot demonstrate that difference. It shows only that function was preserved, which is a necessary result rather than a sufficient one.

The presenters draw the right conclusion, which is that randomised evidence is needed before this moves off protocol. Their own framing matters here: this is a group with the imaging infrastructure, the algorithm and prospective follow-up, saying that even under those conditions the data are not yet a licence to shrink volumes in general practice. The value of the pooled analysis is that it establishes the safety signal is durable enough to justify the randomised trial, not that it substitutes for one. Selecting a patient for a protocol is a different act from changing a standard clinical target volume, and the distance between those two is the whole content of this abstract.