POSEIDON
ForPost-prostatectomy biochemical recurrence, PET-negative, pre-PORT PSA the gating variable
HR 0·87
95% CI 0·76–1·01, p=0·06, not met
TL;DROS HR 0·87 (0·76–1·01), p=0·06 for adding hormone therapy to PORT; benefit only above pre-PORT PSA 0·5 ng/mL.
The gate is pre-PORT PSA, not risk features. Below 0·5 ng/mL point estimates favour PORT alone (HR 1·14 at ≤0·20); above it, HR 0·72 and 0·69 with NNT 22 and 12. Long-term HT only separates on spline at PSA ≥1·6 ng/mL, above what early salvage practice sees.
In PET-negative post-prostatectomy biochemical recurrence referred at PSA 0·2 to 0·5 ng/mL, this questions adding hormone therapy to prostate-bed radiotherapy; it does not address PSMA-PET-positive nodal recurrence or pts already above 0·5 ng/mL, where the survival signal sits.
For the pt referred at PSA 0·2 to 0·5 ng/mL, the point estimates favour PORT alone (HR 1·14 at ≤0·20, 0·94 at 0·21–0·50), so the routine 6-month ADT add-on has no survival footing in the range early salvage now targets. Above 0·5 ng/mL the HRs are 0·72 and 0·69 with NNT 22 and 12.
Duration is not the lever: prolonging 6 months to 24 within RADICALS gave OS HR 0·89 (0·68–1·16), and the long-term cohort's apparent advantage tracks its higher baseline PSA and worse pathology. RTOG 9601 used bicalutamide, not castration, so the one OS-positive trial does not validate GnRH intensification.
13 details
IPD meta-analysis of randomised phase 3 trials of PORT with or without hormone therapy, MARCAP consortium, PROSPERO CRD42019134376. One-stage framework, univariable Cox stratified by trial, intention-to-treat. Six randomised comparisons, 6057 pts, median follow-up 9·0 years.
Post-radical-prostatectomy pts receiving prostate-bed radiotherapy, adjuvant or salvage. Median pre-PORT PSA 0·3 ng/mL overall (0·5 in the long-term cohort). Positive margins in 57–60%, Gleason 7 in about two-thirds, seminal vesicle invasion 16–17%. Race and ethnicity data were not available.
Short-term arm was 4–6 months, predominantly GnRH agonist. Long-term was 24 months, GnRH agonist in 425 (86%) of 492 RADICALS three-way pts, with RTOG 9601 contributing bicalutamide monotherapy rather than castration.
PORT to the prostate bed with or without pelvic nodes, per each contributing trial. Dose, fractionation and nodal-volume detail are not reported in the source text, so the RT technique is not standardised across the six randomisations.
Primary: overall survival. Secondary MFS, event-free survival, biochemical recurrence (Fine–Gray). Prespecified interaction testing on pre-PORT PSA and hormone therapy duration, plus 10-year restricted mean survival time and NNT.
The primary endpoint was not met. The signal sits entirely in the PSA interaction and in MFS, both shown in the detail tables.
| Pre-PORT PSA (ng/mL) | HR (95% CI) | p |
|---|---|---|
| ≤0·20 | 1·14 (0·83–1·57) | 0·43 |
| 0·21–0·50 | 0·94 (0·74–1·19) | 0·70 |
| 0·51–1·00 | 0·72 (0·54–0·96) | 0·02 |
| >1·00 | 0·69 (0·48–0·98) | 0·03 |
| Comparison | OS HR (95% CI) | MFS HR (95% CI) |
|---|---|---|
| Short-term (4–6 mo) added to PORT | 0·93 (0·77–1·11) | 0·82 (0·71–0·95) |
| Long-term (24 mo) added to PORT | 0·79 (0·63–1·00) | 0·74 (0·60–0·91) |
| Prolong short to long | 0·89 (0·68–1·16) | 0·76 (0·61–0·95) |
RTOG 9601 remains the only contributing trial to show an OS benefit, and it enrolled late salvage at higher PSA with bicalutamide, exactly the high-PSA stratum where benefit persists here. The contrast with intact-prostate radiotherapy, where the same consortium's 2022 Lancet Oncology analysis found an OS benefit for added ADT, is the point: the postoperative setting does not inherit it.
The short-term and long-term comparisons draw on different trial populations, and the long-term cohort carried higher PSA and worse pathology, so the duration read is not a clean randomised contrast outside the 1523-pt RADICALS exploratory analysis. Spline thresholds of 1·6 and 2·0 ng/mL sit at the sparse tail of a cohort with median PSA 0·3.
MFS improved while OS did not, and the authors show the MFS effect never clears the ICECaP 0·81 threshold, so the divergence is not simply immature follow-up. What the analysis cannot settle is whether hormone therapy is inert at low PSA or whether its cancer-specific benefit is cancelled by other-cause mortality in pts with indolent disease.
IPD meta-analysis of six randomised trials, prespecified PSA interaction, 9-yr follow-up. Contradicts routine hormone therapy with PORT at low PSA. Duration contrast is exploratory.
- Biomarker to identify who benefits from hormone therapy with PORT
- Does hormone therapy benefit persist in PSMA-PET-staged salvage
- Why MFS gain does not translate to overall survival
📚 Sources · 📄 2 papers
Abstract
The longer read
The result a radiation oncologist has to sit with is not the p=0·06 headline but the shape of the PSA interaction underneath it. Across 6057 pts with nine years of follow-up, adding hormone therapy to postoperative radiotherapy produced a 0·7% absolute difference in 10-year overall survival, and the point estimate in the lowest PSA stratum (HR 1·14 at PSA ≤0·20 ng/mL) sits on the wrong side of unity. That is the stratum early salvage practice has spent a decade trying to move into. The field's own success at treating at lower PSA has, on this reading, moved most pts out of the range where the hormone question has a survival answer.
How much should this move confidence? The design is about as strong as the question permits: individual patient data rather than published summaries, stratification by trial so every comparison is against a concurrent control, and the RADICALS randomisations disaggregated so no pt is double-counted. The PSA interaction was prespecified rather than fished, and it survived the sensitivity analysis built to detect aggregation bias, strengthening on the two-stage within-trial model (continuous p=0·04, binary threshold p=0·03) where the one-stage model was weaker. That is the opposite of the pattern a fragile subgroup usually shows.
The duration comparison deserves more scepticism than the PSA one. The long-term signal (OS HR 0·79) comes from two trials whose pts had higher pre-PORT PSA and more adverse pathology than the short-term trials, so the apparent duration effect and the apparent PSA effect are partly the same observation viewed twice. The authors clearly know this, which is why the RADICALS-internal prolongation analysis matters: within a single randomised comparison, going from short to long bought no overall survival (HR 0·89, 0·68–1·16). The honest read is that duration is not the axis that separates responders, PSA is.
The MFS discordance is where this becomes interesting rather than merely negative. MFS improved convincingly (HR 0·79, 0·70–0·89) and did not translate, and the paper does the work of showing why that is not just short follow-up: the upper confidence bound never crosses the ICECaP 0·81 threshold at any pre-PORT PSA, so the effect is too small to have predicted a survival benefit even on the surrogacy model derived from this literature. Two explanations survive. Either delaying metastasis in a population whose competing risks are cardiovascular and metabolic does not extend life, or hormone therapy's own toxicity in low-risk pts consumes the cancer-specific gain. These have very different implications, and nothing in this dataset separates them.
What would have to be true for this to be wrong is a selection problem the trials cannot rule out: enrolment predates PSMA PET, so an unknown fraction of the low-PSA pts carried occult nodal or distant disease and were treated as though the prostate bed were the whole problem. Modern staging would sort those pts into a different pathway entirely. Koontz's restriction of the finding to PET-negative recurrence is the right one, and NRG GU006 testing PAM50-directed intensification is the shape of the answer this analysis argues for: not longer hormone therapy, but a way to tell which men need any.