Why this matters
Hepatocellular carcinoma represents one of the clearest theoretical indications for proton therapy.
Radiation dose to uninvolved liver can constrain tumor dose and contribute to hepatic toxicity. Protons can substantially reduce low and intermediate dose outside the target, potentially allowing higher tumor doses while preserving functioning liver.
NRG-GI003 asked the clinically important question that dosimetry alone cannot answer: does that physical advantage translate into longer survival?
The randomized phase III result is negative.
Despite substantially lower dose to non-tumor liver, and despite higher delivered tumor doses in parts of the proton cohort, proton therapy did not improve overall survival, progression-free survival, or local progression compared with photons.
Study design
NRG-GI003 enrolled patients with pathologically or radiographically confirmed unresectable or locally recurrent HCC.
Key eligibility criteria included:
- Zubrod performance status 0-1
- Child-Pugh A or B7 liver function
- 1-3 tumors
- protocol-defined tumor size limits
- tumor vascular thrombosis permitted in selected patients
Patients were randomized 1:1 to proton or photon therapy and stratified by:
- planned fractionation, 5 vs 15 fractions
- presence or absence of tumor vascular thrombosis
Radiotherapy dose was individualized according to mean dose to liver minus GTV.
Allowed schedules were:
- 30-50 Gy in 5 fractions
- 37.5-67.5 Gy in 15 fractions
The primary endpoint was overall survival.
Secondary endpoints included:
- progression-free survival
- local progression
- treatment-related adverse events
A total of 116 patients were randomized and 115 were eligible for analysis:
- 57 to protons
- 58 to photons
Median age was 72 years, 93% had Child-Pugh A disease, and approximately one-quarter had tumor vascular thrombosis.
The trial was amended because accrual was slower than originally planned and was ultimately closed to accrual in January 2026.
Dosimetry
The expected physical advantage of proton therapy was clearly demonstrated.
For patients receiving 5 fractions, median mean dose to liver minus GTV was:
- 8.5 Gy with protons
- 12.8 Gy with photons
For patients receiving 15 fractions:
- 12.7 Gy with protons
- 20.3 Gy with photons
Protons therefore substantially reduced radiation exposure to uninvolved liver.
They also allowed greater delivered tumor dose in the 15-fraction cohort.
Median PTV dose was:
- 67.5 Gy with protons
- 58.05 Gy with photons
in patients treated with the 15-fraction schedule.
The trial therefore achieved what proton therapy was expected to achieve dosimetrically.
The critical question was whether patients benefited clinically.
Key results
At the planned interim analysis, the futility boundary was crossed and the Data Monitoring Committee recommended releasing the trial for reporting.
Proton therapy did not improve the primary endpoint of overall survival.
At 24 months:
- OS: 56.2% with protons vs 69.3% with photons
- HR 1.30
- 90% CI 0.83-2.04
- 1-sided p=0.83
This should not be interpreted as proof that photons are superior. The trial was designed to test whether protons improved survival, and that hypothesis was not supported.
Progression-free survival was nearly identical:
- 24-month PFS: 31.6% with protons vs 30.4% with photons
- HR 1.02
- 95% CI 0.65-1.62
- p=0.92
There was also no significant improvement in local progression:
- 24-month local progression: 19.8% with protons vs 22.3% with photons
- HR 0.80
- 95% CI 0.36-1.77
- p=0.54
Toxicity
Treatment-related grade 3+ toxicity was numerically lower with protons:
- 11% with protons
- 24% with photons
- p=0.093
The difference did not reach statistical significance.
A notable contributor to the numerical difference was laboratory toxicity, particularly lymphocyte decrease.
There were no treatment-related grade 5 events in either group.
The toxicity result therefore represents a potentially clinically meaningful signal, but not a statistically established proton advantage.
Exploratory tumor-size signal
The presentation included a post hoc multivariable analysis suggesting that the effect of proton therapy on OS may differ according to tumor size.
The treatment-by-tumor-size interaction was:
- p=0.046
For tumors smaller than 5 cm, the estimated proton-versus-photon HR was:
- 0.59
- 95% CI 0.24-1.45
For tumors 5 cm or larger:
- HR 2.13
- 95% CI 0.96-4.70
Neither subgroup estimate was individually statistically significant.
This finding is therefore exploratory and should not currently guide patient selection. It generates a hypothesis that tumor size may influence the balance between the dosimetric advantages and clinical effectiveness of proton therapy, but requires independent validation.
Interpretation
NRG-GI003 provides an unusually clean test of one of the central assumptions underlying proton therapy: that reducing normal-tissue dose should translate into better clinical outcomes.
The dosimetric hypothesis worked.
Protons substantially reduced mean dose to uninvolved liver and, in the 15-fraction cohort, allowed higher median tumor dose.
The clinical hypothesis did not.
There was no improvement in overall survival, progression-free survival, or local progression.
That disconnect is the most important result of the study.
A dosimetric advantage is not itself a patient benefit. Even when a modality clearly reduces normal-tissue exposure, the magnitude of that reduction may not be sufficient to change survival or disease control in an unselected population.
The toxicity signal deserves attention but does not rescue the primary result. Grade 3+ events decreased numerically from 24% to 11%, but the difference was not statistically significant.
Likewise, the tumor-size interaction is interesting but highly exploratory. It emerged from a post hoc analysis in a relatively small trial and cannot currently identify a subgroup in whom protons should or should not be used.
The negative primary endpoint is therefore clinically important precisely because the physical rationale for proton therapy was so strong.
Limitations
The trial did not achieve its original accrual target and underwent a statistical amendment before closing enrollment.
The current report is based on a planned interim analysis after crossing the futility boundary.
The total sample size was modest, limiting precision for subgroup analyses and toxicity comparisons.
Median follow-up was approximately 20 months, which remains relatively short for some late outcomes.
The tumor-size interaction was post hoc and should be considered hypothesis-generating.
The trial also tested individualized dose schedules across both 5- and 15-fraction regimens, introducing some treatment heterogeneity.