onc brain

The longer read

The argument here is narrower and more defensible than the title suggests, and it turns on a distinction worth keeping straight. Nothing in the evidence Rowe reviews disputes that change in PSMA uptake after starting an ARSI carries prognostic information: the hazard ratios of 5.03 and 9.31 for overall survival by δ-percent and δ-absolute SUVmax point the same direction, and time to therapy change separates by roughly half in the unfavourable group. What is disputed is whether that association licenses reading a single patient's serial scans as response or progression, and those are different evidentiary bars. A prognostic marker needs only to stratify a cohort; a response criterion needs a stable, therapy-independent mapping from signal to disease burden, plus enough measurement reproducibility that a change exceeding some threshold means something other than noise.

The biology is what breaks the mapping. PSMA expression is regulated by androgen receptor signalling rather than fixed to tumor mass, so an agent that perturbs the AR axis perturbs the readout at the same time it perturbs the disease. Uptake can rise on effective therapy. This is not a hypothetical failure mode built from preclinical work alone, since the earliest post-ADT clinical imaging showed substantial expression change soon after initiation, and it is precisely why the timing of the follow-up scan relative to therapy start is not a scheduling detail. The counterpoint speaker will presumably argue that the direction and pattern of change are themselves informative, and the mixed-but-predominantly-increased pattern associating with worse outcomes is some support for that. But an imaging biomarker whose interpretation depends on knowing the drug, the mechanism, and the interval is a fragile one to standardize.

The reproducibility argument is the stronger half of the case and the one that should move a reader most. Repeatability work using Bland-Altman analysis, intraclass correlation, and within-subject coefficients of variation found that lesion identification and segmentation variability remain consequential even restricting to lesions above 1.5 cm³, which are not small lesions. If two readers or two acquisitions disagree about which lesions exist and where their boundaries lie, then SUVtotal and tumor volume, the parameters that would carry a burden-based response rule, inherit that instability. RECIP 1.0 is the field's serious attempt to impose structure on this, and it is a real advance, but it was derived for radioligand therapy where the tracer target and the therapeutic target coincide and where AR-driven expression modulation is not the dominant confounder. Extending it to AR-directed therapy is an assumption, not a validated step.

For a radiation oncologist the practical consequence sits upstream of any of this. Metastasis-directed therapy decisions increasingly get made off restaging PSMA PET, and if the scan follows ARSI initiation by a few months, an apparent increase in lesion avidity is not clean evidence that the systemic agent has failed or that new sites need treating. The honest position from this evidence base is that PSMA PET remains a strong staging and detection tool whose behaviour as a serial response metric under AR-axis manipulation is not yet characterized well enough to drive a per-patient decision. The work Rowe names as needed, kinetics of the AR effect on PSMA and better automated lesion identification, is the right list, and both are tractable.