onc brain

The longer read

The question MASTERPLAN asks is an old one in pancreatic cancer: once systemic therapy controls micrometastatic disease better, does local failure become a problem worth radiating? Earlier randomized work mostly said no, or said it unclearly. LAP07 tested conventionally fractionated chemoradiation in locally advanced disease against a chemotherapy backbone that is now outdated, and PREOPANC tested preoperative chemoradiation in resectable and borderline disease. Neither used ablative-range stereotactic dosing, and neither was designed around local control as the primary question. MASTERPLAN changes both variables at once. It uses 40 Gy in five fractions with fiducials and real-time QA, and it puts locoregional recurrence at 12 months first, which is the endpoint SBRT can most directly move.

On that endpoint the result is clear, and the absolute gap is large enough that a trial of 74 patients could detect it. The difference in median OS from 14.0 to 22.0 months (P=0.006) is what will get attention, and also where caution is most needed. OS was secondary, the source reports no hazard ratio or confidence interval, and the trial was not powered for survival. Small randomized trials in pancreatic cancer have produced survival differences before that did not hold up at scale. The direction is coherent, because better local control, longer PFS (11.9 vs 6.3 months) and longer OS all point the same way. Still, coherence is not confirmation, and the reported OS gain is larger than one would expect from a local therapy alone when distant failure remains the dominant pattern.

Two design features should shape how far a reader generalizes. First, the population is mixed. The benefit is driven by the potentially operable stratum, while the inoperable stratum (n=26) shows a similar point estimate but does not reach significance. A radiation oncologist treating locally advanced disease should not take the pooled result as settling that setting. Second, only 80% of patients assigned to SBRT started it. That would normally bias an intention-to-treat comparison toward the null, so the true per-protocol effect on local control may be larger. It also means a fifth of the experimental arm did not get the intervention, and the source does not say why. If early progression or deconditioning was the reason, the issue is selection for SBRT fitness, not dilution.

The surgical data are honest non-results. Resection (68% vs 59%) and R0 rates (62% vs 40%) favor SBRT numerically but are underpowered. A surgeon should not read them as evidence that SBRT converts resectability. Toxicity is reassuring at the G3+ aggregate level, but that figure is dominated by chemotherapy, and the late duodenal and gastric effects that matter for stereotactic pancreatic RT are not described.

For the result to be wrong, the control arm's locoregional recurrence would need to be unusually high by chance, or recurrence assessment would need to be biased in an unblinded trial. The next trial should be larger, restricted by stratum, report RT-specific late toxicity, and test whether the OS signal survives.