onc brain

The longer read

TREASURE is the first randomized test of a strategy that had been drifting into practice on the strength of single-arm data, and it lands hard in the opposite direction. The pre-IO rationale for consolidative thoracic RT in extensive-stage disease came from CREST, where thoracic RT after chemotherapy improved 2-year survival and intrathoracic control in a population that had not seen a checkpoint inhibitor. The unstated assumption when chemoimmunotherapy became first-line was that bolting the same modest thoracic dose onto maintenance atezolizumab would carry the same benign safety profile. Prospective single-arm series reported exactly that, and this trial shows why single-arm safety reporting in this setting was not sufficient: a 19.4% fatal AE rate is only interpretable against a randomized 3.0%.

The mechanism the authors advance is more persuasive than the OS number itself. PFS was superimposable at 2.4 versus 2.6 months, which means the intervention did not change tumor behavior in either direction, and an OS split without a PFS split points away from disease and toward treatment. The lymphocyte data give that argument a substrate: depletion was specific to arm A, specific to lymphocytes rather than leukocytes or neutrophils, persisted across maintenance cycles, and tracked with an accumulation of infectious and respiratory events. The AE time-to-event analysis separates the two exposures cleanly, with TRT at HR 2.47 (1.15-5.32) and longer atezolizumab exposure not associated. That is about as close to attribution as a 68-patient trial gets.

What should give a radiation oncologist most pause is that nothing about the plan was aggressive. Thirty Gray in ten fractions to postinduction volumes, with dosimetry inside or below usual ranges and under organ-at-risk thresholds, is a conservative prescription by any standard, and the authors could not find a dosimetric parameter associated with serious events within arm A. The usual reflex after a toxicity signal, drop the dose or shrink the field, has no evidence here that it would help. Nor does timing: concurrent versus sequential delivery relative to atezolizumab made no difference. If the effect is real, it is an interaction between thoracic irradiation of a lymphocyte-rich, highly perfused compartment and ongoing checkpoint blockade, and the exposures that would matter are ones the trial only gestures at, such as dose to circulating blood and to marrow-bearing volumes.

The skeptic's case is straightforward. Sixty-eight patients, an OS confidence interval running from 0.90 to 2.69, a P value of .34, and a stop triggered by the very events being interpreted. Curves that cross at 24 months with 2-year rates numerically favoring the irradiated arm are what an early hazard from toxic deaths looks like superimposed on a possible late local-control benefit, and that possibility cannot be excluded from these numbers. The attribution disagreement between investigators and the safety committee, four deaths called unrelated and three of them reclassified, shows how much interpretive weight rests on judgment calls in a small sample.

What holds the result up is external consistency. PACIFIC-2 and CheckMate-73L both found more fatal infectious events when thoracic radiotherapy ran alongside checkpoint blockade in non-small cell disease. Three randomized trials pointing the same way is a different evidentiary object from one underpowered phase 2. The reasonable position is that unselected use is off the table, and that the open question is selection, with baseline DLCO SB and lymphocyte kinetics the first candidates worth prospectively testing.