onc brain

The longer read

TORPEdO tested the premise behind most proton referrals in head and neck cancer: that lower dose to swallowing and salivary structures becomes function a patient can feel. The dosimetric half held, with IMPT delivering lower doses to most organs at risk, but the functional half did not. At 12 months the UW-QoL physical composite differed by 1.3 points, and the 97.5% CI of –3.7 to 6.2 leaves room for a small benefit while making a large one unlikely. The comparator is what gives the null its weight. IMRT here was planned to the same challenging OAR targets and planning priorities as IMPT, with central outline and plan review, so this is protons against well-optimised photons rather than against average community IMRT. That also bounds transferability: a centre whose IMRT falls short of that standard is not the setting tested.

The clinician-reported co-primary leaned the wrong way for the hypothesis. Events ran 18% vs 7%, almost entirely severe weight loss, and grade 3 weight loss was still present in 12 of 95 IMPT pts at 24 months with none on IMRT. The trial cannot say why. Several candidates are visible in the data: fewer prophylactic feeding tubes on IMPT (26% vs 38%), a higher baseline BMI in the IMPT arm (28.3 vs 27.2), a weight-loss excess concentrated in pts with baseline BMI of at least 25 kg/m², and a more disrupted delivery, with replanning in 61% vs 30% and cyclotron-driven interruptions. Each would push weight loss in the observed direction without saying anything about proton biology, and with p=0.079 and a wide interval, harm is not established. What the result does is remove any presumption that IMPT protects nutrition.

The pre-planned NTCP analysis matters for programmes that select pts by modelled toxicity reduction, as the Dutch thresholds do, rather than by randomisation. The enriched cohort scored 78.2 vs 75.8 with overlapping 95% CIs. That subgroup is small and was never powered, so it does not disprove model-based selection, but it offers no randomised support for it on the endpoint the models are meant to predict.

Confidence in the null should be moderately high for function and low for oncologic outcomes. An open-label design with a patient-reported co-primary would, if anything, bias toward reported benefit in pts who travelled to a proton centre, which makes a flat result more credible. Prespecified and post-hoc sensitivity analyses, including per-protocol and worst-outcome assumptions, were consistent with the primary analysis. The 2:1 allocation, though, left 69 IMRT pts, and loco-regional control and survival were exploratory: an LRR HR of 2.6 with a 99% CI of 0.3 to 20.3 neither reassures nor alarms on the split-target proton approach.

The decision it moves is referral of cisplatin-eligible, largely p16-positive oropharyngeal SCC for IMPT on functional grounds; the source concludes IMRT remains standard where IMPT is not used routinely. The read would be wrong if a real benefit sits below the trial's detection limit, or if late effects beyond 24 months, such as mandibular osteonecrosis or late dysphagia, separate the arms. Follow-up is ongoing.