Toxicity, Not Cure: Pooled Evidence Clarifies What Protons Offer — Particle Therapy in October 2026

October's most important papers converge on a single, clarifying conclusion: proton therapy's demonstrable benefit lies in what it spares, not in what it cures. Pooled evidence in oropharyngeal cancer and in paediatric medulloblastoma shows equivalent cancer outcomes with less toxicity, and a national selection model for lung cancer is built explicitly on predicted complications. Carbon ions, by contrast, continue to make the case for better tumour control in hard-to-treat cancers of the pancreas and lung. Meanwhile, liver and FLASH studies show how much of the field's progress now depends on adaptation, robustness and dose-rate engineering.

Oropharynx and medulloblastoma: the case is toxicity

A pooled analysis of two phase III randomised trials and one matched cohort study comparing IMPT with IMRT for oropharyngeal cancer is unusually clear in its conclusion. The hazard ratio for locoregional recurrence was 1.37 (95% CI 0.56–3.36), for progression-free survival 0.90 (0.61–1.33) and for overall survival 0.79 (0.32–1.98), all with confidence intervals spanning 1. Toxicity was another matter. Acute grade 3 dysphagia occurred in 24% of IMPT patients against 40.9% with IMRT (RR 0.65), and grade 3 xerostomia or oral and pharyngeal dryness in 19.9% against 32.5% (RR 0.72), both without heterogeneity. Exploratory nutritional outcomes, including 12-month gastrostomy dependence, showed no definitive benefit. The authors' summary is direct: IMPT's primary advantage is reducing acute toxicity, not improving efficacy.

The same pattern appears in paediatric medulloblastoma. The CURE meta-analysis compared proton and photon craniospinal irradiation across ten cohort studies — there are no randomised trials — totalling 1,034 children (537 proton, 497 photon). Overall survival did not differ (RR 0.984, I² = 0%). Protons significantly reduced growth hormone deficiency (RR 0.379, number needed to treat 2), hypothyroidism (RR 0.256, NNT 2) and neurocognitive decline measured by full-scale IQ (SMD 0.708, NNT 5), while grade 3 or higher ototoxicity did not differ. The authors support preferential use of proton craniospinal irradiation, while calling for prospective studies and an assessment of cost-effectiveness. Numbers needed to treat of two for endocrine outcomes are hard to ignore.

Selecting by predicted harm

Model-based selection for proton therapy depends on validated complication models, and October brought an important update. An NTCP model for grade 2 or higher acute oesophageal toxicity in lung cancer was validated and revised across 937 patients from four centres, of whom 444 (47%) developed the toxicity. The original model's performance varied between cohorts, so concurrent chemotherapy, sex and tumour stage were removed and overall treatment time added. The final model, based on mean oesophageal dose and overall treatment time, performed uniformly across centres (I² of 0% for all performance measures) with an optimism-corrected AUC of 0.75, and has been added to the national indication protocol for selecting lung cancer patients for protons.

For lung cancer patients unable to receive concurrent chemotherapy, the American Radium Society's appropriate use criteria, from a 15-member panel, judged both IMRT and proton therapy appropriate across presentations of locally advanced disease when using accelerated hypofractionation, with the aim of keeping lung V20Gy below 30% and heart V40Gy below 20%. Conventionally fractionated radiotherapy to 60 Gy in 30 fractions remains appropriate across variants in the absence of randomised evidence of superiority for hypofractionation.

Carbon ions: pancreas and early lung cancer

The final results of the prospective phase 2 PACK study (NCT04194268) add to the case for carbon ions in pancreatic cancer. Twenty-five patients with locally advanced or locally recurrent disease were enrolled and 23 analysed after treatment with 48 Gy (RBE) in 12 fractions using raster scanning, with weekly gemcitabine. One- and two-year overall survival were 78% and 59%, and median overall survival 26.5 months. Progression remained the problem: one- and two-year progression-free survival were 26% and 10%, with a median of 8.1 months. There were three treatment-related grade 3 toxicities — abdominal pain, upper gastrointestinal haemorrhage and a ureteric stricture — and no grade 4–5 events. The investigators call for a randomised phase 3 trial testing survival superiority against conventional chemoradiotherapy or chemotherapy.

In early lung cancer, the Japanese J-CROS-LUNG prospective registry analysed 231 patients with stage I non-small cell lung cancer treated with carbon ions at five institutions. Five-year overall survival was 69.6%, cause-specific survival 85.6% and progression-free survival 54.8%. Patients judged operable by a tumour board did markedly better than inoperable patients: 5-year overall survival 83.5% versus 49.1%, and 5-year cumulative local recurrence 3.8% versus 12.1%. The authors propose clinically assessed operability as a composite baseline marker for counselling patients, interpreting outcomes and designing future studies.

The liver: adaptation and the limits of escalation

Three liver studies show both the promise and the engineering demands of particle therapy in hepatocellular carcinoma. A multi-institutional phase II trial of hypofractionated ablative proton therapy (up to 67.5 GyE in 15 fractions) treated 92 patients with unresectable HCC or intrahepatic cholangiocarcinoma. Among evaluable patients, median overall survival was 29.2 months for HCC and 20.8 months for cholangiocarcinoma, with 5-year survival of 25.9% and 12.3%. In-field local failure occurred in only 2 HCC (4.5%) and 7 cholangiocarcinoma (17.9%) patients, and there were no treatment-related toxicities above grade 3. Failure was predominantly out-of-field, intrahepatic or distant, and exploratory plasma profiling linked progression to signatures of hypoxia and vascular remodelling — pointing towards combination strategies.

A planning study of 13 patients with large HCC adjacent to the gut asked how far dose could be escalated at equal toxicity. IMPT allowed more fractions than VMAT in 5, 8 and 8 of 13 patients under non-cirrhotic, Child-Pugh A and Child-Pugh B assumptions, with the advantage growing as liver function worsened. VMAT was mostly limited by mean liver dose, IMPT by gastrointestinal maximum dose — the constraint shifts from liver reserve to serial organs.

For carbon ions, daily online adaptive stereotactic treatment of 14 HCC lesions in challenging locations in 13 patients maintained PTV coverage at 95.7% against a baseline of 97%, whereas no adaptation or occasional offline adaptation reduced it to around 88%. Organ-at-risk constraint violations occurred in 38.5% of fractions without daily adaptation but 11.5% with it, and affected bowel, heart and oesophagus up to 6 cm from the target — a carbon-ion particularity the authors highlight.

BNCT, FLASH and the research frontier

An exploratory study of pretreatment biopsies from a phase I BNCT trial in cutaneous tumours compared three responders with three non-responders. Responders tended to have higher densities of T cells, CD8+ T cells and dendritic cells, while non-responders had higher proportions of PD-L1-positive cells. The findings are descriptive, but they support the growing interest in BNCT's immunological effects.

A critical review of proton boron capture therapy and neutron capture enhanced proton therapy concluded that, at clinically relevant boron concentrations, both generally produce negligible or only marginal dose enhancement, limited by the low proton–boron reaction cross-section and insufficient thermal neutron production, although some experimental biological effects remain unexplained.

For FLASH research, a redesigned synchrotron beamline now offers 42 dose and dose-rate combinations, from about 13 Gy/s to 2,200 Gy/s, with single-spill doses up to 53.8 Gy, without accelerator modification. A review of preclinical highlights from PTCOG 64 frames the broader direction: compact accelerators, prompt-gamma range verification, LET-aware quality assurance and uncertainty-aware NTCP models as near-term innovations, and digital twins, multi-ion optimisation and quantum technologies further out.

What October adds up to

October gives the field an unusually honest statement of what protons do: in oropharyngeal cancer and medulloblastoma they reduce toxicity without changing survival, and that is now the basis for patient selection. Carbon ions continue to show promise in pancreatic and lung cancer, but only randomised trials will settle their place. In the liver, the benefit of particles depends increasingly on adaptation and robust planning. The value of particle therapy is becoming clearer, more specific and more measurable.