onc brain

The longer read

The cleanest way to read 0848 is as two results that point in opposite directions and are not equally reliable. The primary is unambiguous: after six cycles of gemcitabine-based chemotherapy in pts who had a gross total resection and no radiographic recurrence, adding 50.4 Gy of fluoropyrimidine-sensitized chemoradiation returned an OS HR of 0.96 with a 90% CI running from 0.79 to 1.18. That interval excludes the hypothesized HR of 0.76 the trial was designed to detect, so the underpowering (270 of a planned 316 events, 72% power) does not rescue the result; the trial did not miss a benefit of the size it went looking for. The as-treated sensitivity analysis, run because 14% of the CXRT arm declined radiotherapy, gave HR 0.98, which closes off the usual objection that noncompliance diluted a real effect.

Against that, the node-negative subgroup is the most interesting thing in the paper and the least secure. Five-year OS of 48.1% versus 28.6%, median OS 3.9 versus 3.0 years, and a treatment-by-nodal-status interaction at P=.022 in a multivariable model is a coherent biological story: pts with uninvolved nodes carry a lower competing risk of distant failure, and in a disease with a codominant pattern of failure that is precisely the population where locoregional therapy has room to change survival rather than merely change the site of recurrence. The problem is arithmetic. The subgroup is 91 pts, 42 and 49 per arm, and the interaction was one of nine tested in Figure 3. An interaction p-value near .02 in that setting is a hypothesis, and the authors say so.

What 0848 does settle, and this matters more than it first appears, is the safety question that ESPAC-1 left behind. Grade 4 or 5 events were 6% versus 5%. The grade 3 excess (38% versus 19%) is real but is diarrhea and lymphopenia, not the kind of toxicity that would make an adjuvant modality untenable. Since 2004 the argument against adjuvant chemoradiation in this disease has carried an implicit claim of harm; 0848 delivered radiation with credentialed centers, real-time plan review, and 92% per-protocol or acceptable-variation scoring among treated pts, and the harm did not appear. Whether ESPAC-1's detrimental signal was technique or chance, it is no longer a reason to avoid the question.

The part that limits how far any of this travels is the systemic backbone. Not a single pt received FOLFIRINOX; two-thirds got single-agent gemcitabine. If a more effective systemic regimen suppresses distant failure, the fraction of pts whose outcome is governed by locoregional control rises, which should make the node-negative signal larger rather than smaller. That is an argument for testing it, not for assuming it. The reverse case is at least as plausible: better systemic therapy may simply move the survival curves of both arms up and leave the RT contribution unmeasurable in a subgroup this size.

For a radiation oncologist, the practical read is narrow and honest. This does not restore adjuvant chemoradiation as a default after resection. It does make it defensible to discuss with a node-negative pt who has come through adjuvant chemotherapy without progression, and it gives a delivered, QA'd technique to discuss it with. The open question the field should be asking is not whether RT is safe here, which is now answered, but whether the node-negative hypothesis can be prospectively confirmed on a modern backbone before the neoadjuvant shift makes postoperative nodal status too unreliable to select on.