onc brain

The longer read

The case for PCI in SCLC was built before MRI could reliably show that the brain was clean at restaging or catch small relapses while they were still treatable. The abstract says as much: the reductions in brain metastases and the survival benefit come from pre-MRI era studies. In extensive stage, Takahashi's Japanese randomized trial had already undercut that case once MRI surveillance was the comparator, and many clinicians stopped offering PCI there. Limited stage is where the habit persisted, because the older pooled evidence was never retested. MAVERICK matters mostly because of who it enrolled: 68% LS-SCLC, with a stage interaction that is nowhere near positive on either the slide (p=0.79) or the abstract (p=0.92).

The primary endpoint deserves a careful read before the HR is taken at face value. CFFS counts a decline on any one of three tests, or death, as an event, so it is built to be sensitive. Both arms fail fast: 6 month CFFS is 37.8% with MRI alone and 16.5% with PCI, and by 12 months the figures are 17.3% and 5.9%. That is not a picture of preserved cognition without PCI. It is a population in which chemotherapy, disease and death drive most events, with PCI bringing them forward. The Fine-Gray analysis (subdistribution HR 0.62) argues that the difference sits in cognitive failure itself and is not carried by deaths. The 1-sided 0.05 test and 90% intervals are more permissive than the usual convention, though an upper bound of 0.78 leaves some distance from the null.

The most useful secondary result for a radiation oncologist is the hippocampal avoidance analysis. With 77% of PCI delivered as HA-PCI, and pts in the MRI arm who had been declared for HA-PCI before randomization still doing better (HR 0.63), the obvious rejoinder, that modern PCI is cognitively safe, loses its footing on this endpoint. NRG CC003 asked whether hippocampal avoidance improves on conventional PCI. MAVERICK asks the harder question of whether either beats not irradiating at all.

What the trial has not yet delivered is the survival answer that the standard-of-care claim depends on. The key secondary is non-inferiority with a margin of 1.25 on the upper 90% bound, and the interim estimate of 0.90 (0.67-1.20) comes at 128 of a planned 190 deaths. The direction is reassuring, but an interim read is not the final one. Brain metastases were less frequent with PCI and PFS was no different, which fits a model where MRI-detected relapses are salvaged with SRS or WBRT without a survival cost. That model depends on scan adherence the trial enforced and routine practice may not. The result would be wrong if late OS diverges in LS pts, where median follow-up of 22 months is short and an isolated brain relapse can cost a cure.

Two reporting details temper confidence at the margins. The slide and abstract give different p-values, which suggests different data cuts, and the CFFS table lists fewer pts than were randomized without explanation. The prespecified Day 180 patient-reported endpoints were also negative (p=0.07 and 0.69), so the objective tests show harm more clearly than pts report it. Finally, ADRIATIC changed LS-SCLC care during accrual, and only 41% received upfront immunotherapy. The IO interaction (p=0.25) is not a signal, but the durvalumab-consolidated LS pt is only partly represented here.