COMPPARE
ForDe novo localized prostate cancer, excluding very high risk and metastatic
5.7% vs 6%
P=0.28, hypothesized 7% vs 15%
TL;DRProton vs IMRT: no difference in bowel urgency (5.7% vs 6%), ≥G2 GI toxicity (5.2% vs 5.6%), or 3yr disease control.
The spacer table is the actionable finding, not the modality comparison: 2yr GI G2+ fell to 4.4% (IMRT) and 4.7% (proton) with a spacer vs 7.2% and 8.7% without, P=0.009. Rectal separation, available at any IMRT center, delivered what particle therapy did not.
In de novo localized prostate cancer outside very high risk, this argues the rectal-sparing decision sits with spacer placement rather than referral to a proton center; it says nothing about late GU endpoints or very high risk disease.
The spacer stratum, not the modality arm, is where the toxicity separated: 2yr GI G2+ 4.4% (IMRT, spacer) and 4.7% (proton, spacer) vs 7.2% and 8.7% without, P=0.009. That moves the rectal-sparing decision toward spacer placement at your own center rather than proton referral.
| Outcome | Hypothesized IMRT | Hypothesized PT | Actual IMRT | Actual PT | P-value |
|---|---|---|---|---|---|
| Bowel urgency | 15% | 7% | 6% | 5.7% | 0.28 |
| Bowel frequency | 10% | 4% | 4% | 3.5% | 0.43 |
| GI toxicity CTCAEv5 ≥2 | 29% | 20% | 5.6% | 5.2% | 0.60 |
| Freedom from progression 3yr | 89% | 91% | 97.9% | 98.0% | 0.90 |
+2 more figures
| Group | 2yr cumulative CTCAE v5 GI G2+ | P |
|---|---|---|
| IMRT, no spacer | 7.2% (5.0%, 9.9%) | 0.009 |
| Proton, no spacer | 8.7% (5.0%, 14%) | |
| IMRT, spacer | 4.4% (2.8%, 6.4%) | |
| Proton, spacer | 4.7% (3.6%, 6.0%) |
8 details 4 trials watching
Prospective nonrandomised comparative-effectiveness cohort study funded by PCORI, comparing proton therapy and IMRT across 51 centers. Accrual 2524 pts from July 2018 to October 2022, allocated to a proton cohort (1500) and a photon cohort (1000).
All de novo prostate cancer except very high risk and metastatic. The exclusion is the boundary that matters: the pts in whom elective nodal coverage and integral dose arguments are strongest were never enrolled.
Primary: patient-reported bowel urgency and bowel frequency (EPIC) and CTCAE v5 ≥G2 GI toxicity, each powered at 90%. Freedom from disease progression at 3 years (PSA) was exploratory, not powered.
Every prespecified comparison was null. The more telling result is that observed rates undershot the design assumptions in both arms: ≥G2 GI toxicity 5.6% IMRT and 5.2% proton against 29% and 20% hypothesized.
Rectal spacer use separated the toxicity curves where modality did not. 2yr cumulative ≥G2 GI toxicity was 4.4% (2.8%, 6.4%) IMRT with spacer and 4.7% (3.6%, 6.0%) proton with spacer, vs 7.2% (5.0%, 9.9%) and 8.7% (5.0%, 14%) without, P=0.009 by Gray's test.
The ≥G2 GI rates here are far below the toxicity burden that motivated the proton hypothesis, and align with the modern IMRT plus spacer experience rather than the older photon series the 29% assumption was drawn from.
Cohort allocation, not randomisation, so the arms differ by referral pattern, geography, and insurance in ways baseline adjustment cannot fully absorb. The unequal cohort sizes (1500 vs 1000) reflect enrollment at proton-capable centers, not a design ratio.
A null comparative-effectiveness result in a low-event setting is weak evidence of equivalence and strong evidence that the toxicity target moved. The question the field now needs answered is late toxicity and second malignancy, which 3 years cannot address.
Nonrandomised prospective cohort comparison; residual confounding unaddressable. Null on every prespecified endpoint, but 3yr follow-up cannot capture the late toxicity protons are argued to prevent.
- Late GI and GU toxicity beyond 3 years n=303 · primary completion 2026-12 · proton vs IMRT hypofx registry, f/u to 2026n=400 · primary completion 2027-03 · randomised proton vs photon, late GI primary EPrecruiting Reduction of Gastrointestinal Toxicity in Prostate Cancer by Proton Spot Placement Phase NAn=500 · primary completion 2030-01 · proton LET vs rectal/bladder toxicity, n=500
- Second malignancy risk from integral dose
- Whether protons add anything once a spacer is placed n=50 · primary completion 2025-09 · 2-arm spacer trial in pts planned for proton
📚 Sources · 🐦 1 tweet
#COMPPARE early results: in localized #ProstateCancer, #proton therapy vs #IMRT showed no sig difference in pt-reported bowel urgency/frequency, ≥G2 GI toxicity, or 3-year biochemical control. Longer follow-up needed for late toxicity/long term outcomes #ASCO2026 pic.twitter.com/yli4l8nEOY
— QianJanieQin (@QianJanieQin) May 31, 2026
The longer read
The proton versus photon question in localized prostate cancer has been argued for two decades on a dosimetric premise: the exit dose IMRT deposits in the anterior rectal wall causes bowel toxicity that a Bragg peak avoids. COMPPARE tests that premise prospectively in 2524 pts across 51 centers and finds no separation on any of the three prespecified toxicity endpoints. The more consequential number is not the null P-values but the gap between what was assumed and what occurred. The design expected 29% ≥G2 GI toxicity on IMRT and 20% on protons; observed rates were 5.6% and 5.2%. When both arms come in far below the control assumption, the trial has not so much answered its question as documented that the question changed underneath it. There is very little room left between 5.6% and zero for a modality advantage to occupy, and a study powered for the gap between 29% and 20% is not the instrument to detect whatever difference remains.
What replaced modality as the driver of GI toxicity is visible in the spacer analysis, and it is the part of this presentation a radiation oncologist should carry forward. Two-year cumulative ≥G2 GI toxicity was 7.2% for IMRT without a spacer and 8.7% for protons without one; with a spacer, 4.4% and 4.7%, P=0.009. The ordering is worth sitting with: proton without spacer was numerically the worst of the four strata, and proton with spacer was not better than IMRT with spacer. Physical separation of the rectum from the high-dose region did what particle physics was supposed to do, at a fraction of the capital cost and at any center. This is a within-study comparison across nonrandomised strata, and spacer use is itself a choice correlated with practice patterns and anatomy, so it cannot carry the weight of a randomised finding. But the direction is consistent with the mechanism, and it names an intervention available to every reader.
The design constraints bound how far any of this travels. This is cohort allocation, not randomisation. Pts who receive protons self-select on insurance approval, geographic mobility, and often on comorbidity and performance status, and unequal cohorts of 1500 and 1000 reflect where enrollment was possible rather than an allocation ratio. Baseline adjustment cannot fully absorb that. The direction of such confounding usually favors the proton arm, which makes a null result somewhat harder to dismiss than a positive one would have been, but it does not make the comparison clean.
The follow-up is the deeper problem, and it is the one that should temper any reading of this as settling the question. Three years of PSA-based disease control (97.9% and 98.0%) is a floor effect in a population excluding very high risk and metastatic disease, and it establishes only that neither modality is failing. The toxicity endpoints protons are argued to improve are late ones: sustained rectal bleeding, GU strictures, and second malignancy attributable to low-dose integral bath. None of these mature by three years. A field that has spent a decade justifying proton capacity on late-effect and second-malignancy grounds should not treat an early GI null as a verdict, in either direction. What COMPPARE establishes now is narrower and still useful: for the pt in front of you with intermediate-risk disease and a spacer available, the acute and early GI argument for referral to a proton center is not supported by this data.