onc brain

The longer read

The question this cohort answers is not whether internal mammary irradiation works, which EBCTCG already established, but whether it still works once the systemic backbone got good. That framing matters because the argument for omission was never that IMNI is ineffective; it was that taxanes, trastuzumab, and aromatase inhibitors would sterilise the residual nodal disease that regional RT exists to treat, leaving only the cardiac and pulmonary cost. On that specific question the answer here is clean: the adjusted OS HR of 0.85 is indistinguishable from the EBCTCG rate ratio of 0.90 derived from trials that started before 2009, and the absolute 15-year gain of 4.2% sits essentially on top of IMN1's 4.7% from the pre-taxane era. If better systemic therapy were absorbing the RT benefit, the absolute gain should have shrunk. It did not.

How much confidence that deserves turns on the allocation. Assigning by laterality is not randomisation, and the authors lean on the Mendelian-randomisation analogy to argue that side of tumour is independent of prognosis. The baseline table supports them, with the only imbalance being nodes removed, but the design's real vulnerability is not confounding so much as dilution. Roughly 10% of left-sided pts received IMNI anyway and a quarter of right-sided pts got under 64.8% of the IMN volume to prescription dose because lung constraints bound first. Both errors push the estimate toward the null, which means the honest read is that the true effect of delivered IMNI is somewhat larger than 0.85, not smaller. That is the unusual position of a non-randomised study whose main bias runs against its own conclusion.

The subgroup that should actually move practice is the one with 1-3 positive nodes, where several major guidelines still withhold IMNI absent additional risk factors. At n=3100 the HR was 0.85 (0.73-0.97), the same as the overall estimate, and the authors could not find any measured factor identifying a group safe for omission. IMN1 had suggested lateral tumour location with 1-3 nodes was such a group; that did not replicate. The residual signal worth watching is the opposite one, the medial or central tumour with 4 or more nodes at HR 0.98 (0.79-1.21), which mirrors IMN1's 0.91 and is plausibly a disease-burden ceiling rather than a dosimetric artefact, though the interaction test was not significant.

Two caveats should constrain how far this travels. The ER-negative HER2-positive cell, with an HR of 1.49 crossing 1, is explorative, unadjusted for multiple testing, built on 270 pts, and contradicted by NSABP B-51; it is a hypothesis, not a contraindication. And the toxicity accounting is thin. Cardiac harm was captured only as death from ischaemic or valvular disease, which at 0.2% versus 0.7% at 15 years is reassuring at cohort level but says nothing about non-fatal events, and smoking status was unavailable. That matters because the dose-response between mean heart dose and ischaemic heart disease has no established threshold. The practical consequence is that the trial argues for IMNI as the default in node-positive disease while leaving the individual left-sided decision where it already sits, with the planner and the patient's cardiac risk profile.

What it does not address is the neoadjuvant population, excluded by design, where B-51 and RAPHCHEM are the live evidence and the omission question is genuinely open.