onc brain

The longer read

The useful way to read RAD-IO is as a design decision rather than a result. It is a single-arm, multi-stage feasibility and safety study whose primary endpoint is a 12-month disease-free proportion read against a threshold the investigators set in advance, and the entire epistemic weight of the 80% figure rests on whether that threshold was calibrated honestly. Here the calibration is at least transparent: the reference is BC2001, the team's own randomised trial of radiotherapy with or without 5FU/MMC, and the GO bar sits at 75% with a contextual band of 60 to 75% below it. The result clears GO by 5 percentage points with a lower confidence bound of 0.67, which sits inside the contextual band. That is a pass, not a comfortable one, and with 2 pts still pending and 3 not evaluable out of 55 enrolled, the denominator itself is doing some work.

What should move a radiation oncologist's confidence in either direction is mostly about population rather than about durvalumab. Three quarters of these pts had prior neoadjuvant chemotherapy, and 80% were cT2. That is a favourable bladder-preservation cohort by the standards of most published chemoRT series, and a favourable cohort is exactly what inflates a 12-month disease-free proportion relative to the historic benchmark it is being compared against. The BC2001 experience being used as the yardstick was accrued in a different era of staging, of neoadjuvant chemotherapy uptake, and of imaging-based response assessment. When a single-arm trial beats a historic control by a margin, the first question is how much of the margin is stage migration and patient selection, and nothing in the source lets you answer it.

The radiotherapy itself is the least novel part of the trial and that is a genuine strength. 55Gy in 20 fractions to bladder, 46Gy in 20 fractions to nodes in the node-positive expansion, is standard practice in the UK, which means whatever this trial eventually shows about durvalumab transfers cleanly without any change to planning, dose or target volume. There is no confounding from an experimental RT schedule. The corollary is that this trial cannot inform any RT question: not elective nodal coverage, not fractionation, not the sequencing of RT relative to systemic therapy. A reader looking for an RT decision to change will not find one here.

The finding most likely to matter downstream is the delivery data. 33 of 54 pts completed the planned durvalumab course and 21 discontinued early. A 39% early discontinuation rate is a hard fact about a 12-month adjuvant tail bolted onto chemoRT, and it is reported in a trial whose stated takeaway is that the combination is feasible and tolerable. Both readings can be true if the discontinuations were driven by disease events or logistics rather than toxicity, but the source does not break that down, and the distinction determines whether the schedule taken forward to randomisation should be the one tested here. Checkpoint inhibition around bladder-preserving chemoRT is being pursued by several groups, and the question any randomised trial has to answer is not only whether durvalumab adds benefit but which component, the neoadjuvant, the synchronous or the adjuvant year, is carrying it. RAD-IO gives all three at once and so cannot separate them. For the result to be wrong in the direction that matters, the 80% would have to be a selection artefact that a randomised comparator erases, which is precisely what the next trial exists to test.