Proton therapy delivers less radiation to healthy liver tissue in patients with advanced hepatocellular carcinoma (HCC) compared with conventional photon therapy, but that doesn’t translate to longer survival, according to findings from the phase 3 NRG-GI003 trial.

The rationale for the trial was straightforward: Reducing radiation doses to normal liver tissue with proton therapy could decrease hepatic decompensation and ultimately improve survival compared with photon therapy.

However, at interim analysis, both radiotherapy approaches were providing similar local disease control and progression-free survival. Median overall survival was actually numerically higher with photon therapy, at roughly 55 months vs 27 months — though the difference was not statistically significant.

“There was no survival benefit with proton therapy,” lead investigator Theodore S. Hong, MD, of Dana-Farber Cancer Institute and Beth Israel Deaconess Medical Center in Boston, reported at the American Society for Radiation Oncology (ASTRO) 2026 Annual Meeting.

A Dosimetric Advantage

NRG-GI003 included 115 patients with unresectable or locally recurrent HCC who were randomized in a 1:1 ratio to receive proton therapy or photon therapy. In both groups, radiation could be delivered in five fractions (30-50 Gy) or 15 fractions (37.5-67.5 Gy), with the dose individualized according to the mean dose to the uninvolved liver.

Most patients had Child-Pugh Class A disease (93%), about one quarter had tumor vascular thrombus, and roughly 80% had a single tumor. Median total tumor size was just over 5 cm. Baseline characteristics and delivered tumor doses were generally similar between the groups.

As expected, proton therapy was associated with less radiation exposure to the healthy liver. Among patients receiving five fractions, the median mean dose to liver minus gross tumor volume was 8.5 Gy vs 12.8 Gy with protons photons. With 15-fraction treatment, the corresponding doses were 12.7 Gy and 20.3 Gy.

“The fundamental dosimetric premise of the trial was correct,” Hong said.

Patients in the proton arm had a lower rate of grade 3 or higher treatment-related adverse events, but the difference wasn’t significant (11% vs 24%; P = .093). Much of the numeric advantage was attributable to a lower rate of lab abnormalities, particularly decreases in lymphocyte counts. There were no treatment-related deaths.

Photon Survival Exceeds Expectations

Despite the reduction in radiation to healthy tissue, proton therapy did not improve the primary endpoint of overall survival. In fact, median overall survival was twice as long with photon therapy, at 54.9 months vs 26.6 months (hazard ratio [HR], 1.30; 90% CI, 0.83-2.04).

Hong noted that overall survival in the photon arm “dramatically exceeded” the expectation in the trial’s original statistical design: It assumed a median survival of only 14 months with photons and sought to detect an improvement to 24 months with protons.

Study discussant Erqi Pollom, MD, of Stanford School of Medicine in Stanford, California, offered a potential explanation for the much longer than anticipated survival in the photon arm: Improvements in immune checkpoint combinations alongside modern photon delivery have “dramatically improved patient outcomes since the trial conception.”

Meanwhile, other outcomes were comparable between the treatment arms, including progression-free survival, at a median of 16.9 months vs 13.7 months with protons vs photons (HR, 1.02). There was also no significant difference in local progression (HR, 0.80; 95% CI 0.36-1.77).

In both arms, regional intrahepatic progression outside the radiation field was more common than local failure, whereas distant failure was relatively uncommon.

Tumor Size Signal

A post hoc analysis raised the possibility that tumor size could influence the relative effect of proton and photon therapies, Hong noted. For tumors smaller than 5 cm, outcomes numerically favored protons, whereas for tumors 5 cm or larger, the direction reversed and numerically favored photons. In a multivariable analysis, the treatment-by-tumor-size interaction was statistically significant (P = .046).

Hong cautioned against overinterpreting this post hoc observation, calling it “hypothesis generating rather than definitive.” Still, he said the finding suggests that the value of protons may vary across patient groups.

Further work is needed, Hong said, to determine whether tumor burden, liver reserve, or other factors can identify patients most likely to benefit from proton therapy.

Pollom noted that the NRG trial set “quite a high bar for a local therapy” by targeting an HR of 0.58 for overall survival.

She said that future studies should examine the strategic use of protons in certain higher-risk patients, such as those with more impaired liver function, rather than trying to improve upon the already excellent outcomes of photon therapy for patients like those in the trial.

Ultimately, Pollom said, the NRG results show that modern liver radiation, “whether using photons or protons, is effective and safe.”

In a separate discussion, Marco Schwarz, PhD, medical physicist at the University of Washington in Seattle, noted that the trial may not have fully exploited the potential dosimetric advantage of protons.

Although proton therapy reduced radiation to uninvolved liver tissue, the tumor-dose distributions were similar between the groups. As a result, the expected separation between protons and photons in delivered tumor dose did not materialize. There’s still the question of whether greater dose escalation with protons could improve outcomes, according to Schwarz.

“The trial question remains open, remains sensible, and very much worth asking,” he said.

The study was supported by the National Cancer Institute. Hong reported consulting or advisory relationships with NeoGenomics, Merck, NextCure, and others, and research support from Taiho Pharmaceutical Co., Ltd.; AstraZeneca; Bristol Myers Squibb; and others. Pollom reported serving on the advisory boards of GT Medical Technologies, Inc. and Vysioneer, receiving research funding from ViewRay, and receiving an honorarium from Varian. Schwarz reported having no relevant disclosures.