onc brain

The longer read

The useful signal in this series is what did not predict recurrence. Across a cohort treated mostly with 50 Gy in 20 fractions or 40 Gy in 10, EQD2 carried no association with outcome (OR 1.007, p=0.4735), and whether a lesion was primary, recurrent, or treated postoperatively did not reach significance either. What remained after backwards selection were host and tumour features: age, size of 2 cm or greater, and immunosuppression, each with odds ratios in a range that matters clinically (2.090 for size, 2.939 for immunosuppression). For a radiation oncologist deciding on a schedule, the implication is that within the conventional hypofractionated range used here, prescription is not the lever; the lesion and the host define the risk, and the reader's effort belongs in target definition and surveillance.

The modality finding needs to be read the way the authors themselves read it. Orthovoltage and electrons outperformed photons (ORs 0.285 and 0.440), but photons were reserved by protocol for tumours thicker than 1 cm or larger than 3 cm. That is confounding by indication built into the treatment algorithm, and a three-level modality variable cannot be adjusted away by also including a dichotomized size term, because depth and true extent are not captured by the 2 cm cut. The photon OR is best treated as a marker of the deep, extensive lesion, not as evidence that megavoltage technique fails. If anything, it argues for scrutiny of margins and bolus on those cases, since inadequate field size has been raised elsewhere as an explanation for modality differences.

Several methodological choices should lower confidence in the absolute numbers. Recurrence was modeled with logistic regression rather than a time-to-event method, so a lesion followed for zero months counts as controlled alongside one followed for a decade. With a median follow-up of 18.4 months and discharge of BCC at 2 years, late recurrences, which are characteristic of BCC, are structurally undercounted, and the 84.2% control rate is both lower than many series and probably still optimistic for the longer horizon. The analysis is per tumour without correction for multiple lesions per patient, which matters most for the immunosuppressed group, where one transplant recipient can contribute several correlated events. Recurrence was a clinical judgment without a stated in-field versus marginal distinction, so pattern-of-failure inference is not possible.

Against the van Hezewijk comparison the authors raise, where higher-EQD2 schedules reported 3-year control around 96%, the temptation is to attribute the gap to dose. This dataset does not support that reading, both because its own dose variable was null and because the populations differ in recurrent and postoperative case mix. The age association is also worth holding loosely: the authors note most prior work finds no age effect, and a small per-year OR in an elderly cohort can reflect competing follow-up patterns as easily as biology.

What would make the main conclusions wrong is a clustering or censoring artefact concentrated in the high-risk groups. A competing-risk or Cox analysis with patient-level clustering and in-field failure coding would settle whether size and immunosuppression hold up, and a prospective dose comparison restricted to lesions of 2 cm or more would answer the question this cohort cannot.