NRG-GI003 NCT03186898
ForUnresectable or locally recurrent HCC, CP A or B7, 1-3 lesions, PS 0-1
HR 1.30
90% CI 0.83-2.04, 1-sided p=0.83; not met, futility boundary crossed
TL;DRProtons did not improve OS vs photons in unresectable HCC: HR 1.30 (90% CI 0.83-2.04), futility boundary crossed at planned interim.
Local progression was no different (24-mo LP 19.8% vs 22.3%, HR 0.80, 95% CI 0.36-1.77) even though dose was individualized on liver-GTV mean dose, the parameter protons should improve. Delivered dose by arm is not reported, so whether protons bought escalation is unknown. The modality choice for HCC does not rest on OS or LC here.
In unresectable or locally recurrent HCC with CP A or B7 liver function and 1-3 lesions planned for a 5 or 15 fx course, this does not support selecting protons over photons for survival; it does not address pts beyond CP B7 or with more than 3 lesions.
Local progression was no different (24-mo LP 19.8% vs 22.3%, HR 0.80, 95% CI 0.36-1.77) even though dose was individualized on liver-GTV mean dose, the parameter protons should improve. Delivered dose by arm is not reported, so whether protons bought escalation is unknown. The modality choice for HCC does not rest on OS or LC here.
11 details
Phase III randomized trial, 1:1 protons vs photons, stratified by planned fractions (5 vs 15) and tumor vascular thrombosis. 117 screened June 2017 to January 2026, 116 randomized, 115 eligible/analyzed (protons 57, photons 58). Powered for HR 0.58 with 112 eligible pts and 88 OS events (80% power, 1-sided alpha 0.05), with 1 interim at 63 OS events.
Unresectable or locally recurrent HCC, pathologically or radiographically proven, Zubrod PS 0-1, Child-Pugh A or B7, 1-3 HCCs with size limits of 15, 10 and 6 cm for 1, 2 and 3 lesions; TVT allowed with a single lesion. Median age 72, 93% CP A, 75% no prior treatment, 57% single tumor, 23% TVT.
Protons or photons in 5 fx (30-50 Gy) or 15 fx (37.5-67.5 Gy), dose individualized on liver-GTV mean dose in both arms. Delivered dose by arm, proton delivery technique and motion management are not reported in source.
Primary: overall survival (Kaplan-Meier, log-rank). Secondary: PFS, local progression, adverse events (CTCAEv4).
Trial released at the planned interim after the futility boundary was crossed (HR>1), median follow-up 20 months. No endpoint favored protons with statistical support; arm-level figures are in the table.
| Endpoint | Protons | Photons | Effect |
|---|---|---|---|
| 24-mo OS | 56.2% (43.9-68.4) | 69.3% (57.3-81.2) | HR 1.30 (90% CI 0.83-2.04), 1-sided p=0.83 |
| 24-mo PFS | 31.6% (17.8-45.4) | 30.4% (15.9-44.8) | HR 1.02 (95% CI 0.65-1.62), p=0.92 |
| 24-mo LP | 19.8% (9.6-32.7) | 22.3% (11.3-35.5) | HR 0.80 (95% CI 0.36-1.77), p=0.54 |
| Treatment-related G3+ AEs | 11% | 24% | p=0.09 |
Treatment-related grade 3+ AEs were 11% with protons vs 24% with photons (p=0.09), numerically lower but not statistically different. AE type, hepatic decompensation and late effects are not broken out in source.
Retrospective proton vs photon comparisons in HCC had suggested a survival advantage attributed to liver sparing, and that hypothesis set the effect size this trial was powered for. Under randomization the OS estimate sits on the other side of unity. The source cites no comparator trial.
Cause of death is not reported, so an OS gap alongside PFS HR 1.02 and similar LP cannot be attributed to liver toxicity, competing mortality at median age 72, or chance. Subsequent systemic therapy by arm is unreported across an accrual period running June 2017 to January 2026. No subgroup data by Child-Pugh class, tumor size or fraction schedule in the abstract.
The trial rules against the large OS benefit it was designed to find; with a 90% CI of 0.83-2.04 it does not establish harm. The only signal in the hypothesized direction is the G3+ AE rate, which is underpowered. What stays open is whether protons help where liver reserve is the binding constraint, a group this cohort (93% CP A) barely samples.
Reported at planned interim after futility boundary crossed, abstract only, 115 pts with wide CIs. Negative for proton superiority, not evidence of harm; photons remain the default.
- Cause of the numerically worse OS with protons: hepatic decompensation, competing mortality, or chance?
- Does the lower G3+ AE rate with protons hold in the full report?
- Any proton benefit where liver reserve is limiting (CP B7, large tumors)?
📚 Sources · 📄 1 paper
Abstract
The longer read
The case for protons in HCC has always been dosimetric: less dose to uninvolved liver should mean less hepatic decompensation, and in a disease where pts die of liver failure as often as of tumor, that should show up in survival. Retrospective comparisons pointed that way, and this trial was sized on that optimism, powered for HR 0.58. What came back is an estimate on the wrong side of unity, HR 1.30, with the futility boundary crossed at the planned interim. The defensible reading is narrow. A 90% CI of 0.83-2.04 is compatible with a modest proton benefit and with a doubling of hazard, so this is not evidence that protons shorten survival. It is good evidence that the large advantage the design assumed is not there in this population.
The population matters for how far that travels. 93% had Child-Pugh A cirrhosis, 75% were previously untreated and 57% had a single tumor. That is a cohort with preserved reserve, in which a well-planned photon course can usually respect liver constraints, and so the cohort in which incremental liver sparing has the least room to convert into survival. The pts for whom the dosimetric argument is strongest, those with marginal function or little residual uninvolved liver, were a small minority or outside eligibility. The abstract gives no subgroup by Child-Pugh class, tumor size or fraction schedule, so the trial answers the question for the typical referral and leaves the constrained-liver case untested.
Several design features limit how confidently the mechanism can be read. Dose was individualized on liver-GTV mean dose in both arms, which in principle lets protons prescribe higher for the same liver cost. If that happened, local progression should favor protons, and an LP HR of 0.80 with a 95% CI of 0.36-1.77 is consistent with a small escalation effect or with none. Delivered dose by arm is not reported, nor is proton delivery technique or motion management, and range uncertainty in a moving liver target is a plausible reason a proton plan might underdeliver its promise. That is conjecture the abstract cannot test.
The survival and progression results also sit awkwardly together. PFS was superimposable (HR 1.02) and local progression similar, so whatever separates the OS curves is unlikely to be tumor control. Cause of death is not reported. Subsequent systemic therapy is not reported either, and accrual ran from June 2017 to January 2026, a period over which systemic options for HCC changed. In a 115-pt trial, chance imbalance in unmeasured liver function is a live explanation.
The one finding in the hypothesized direction is treatment-related grade 3+ AEs, 11% vs 24%, p=0.09. If it survives the full report it reframes protons as a toxicity question, not a survival one. Fewer severe AEs paired with numerically worse OS is an odd combination, though, and blinding is not stated. For now the result removes survival as a reason to prefer protons in Child-Pugh A HCC and leaves photon-based 5 or 15 fx treatment as the reference. The full manuscript, with events, causes of death and dose by arm, is what could change that read.