Half of These Jaws Healed. The Oxygen Wasn't What Decided Which Half.

Clinical still life on a stainless steel surgical tray: a white 3D-printed model of a mandible with a missing segment, a curved titanium reconstruction plate laid beside it, and a printed panoramic radiograph underneath marked with an amber circle around a radiolucent defect

Of 318 people treated with radiotherapy or chemoradiotherapy for oropharyngeal cancer at Hokkaido University, thirty-six went on to develop osteoradionecrosis of the jaw. That is 11.3%, reported in the October issue of the Journal of Dental Sciences, and it is a number worth sitting with before the hyperbaric argument starts, because osteoradionecrosis is the condition that made hyperbaric oxygen a hospital service rather than a diving curiosity.

Sixteen of those thirty-six eventually needed an operation. Half of them healed. The variable that separated the healers from the rest was not whether they went into a chamber. It was what the surgeon did.

Two of eight, one of one, five of seven

The Sapporo group is candid about the arithmetic, which is the reason the paper is useful. Healing after the first operation ran at 25% for a sequestrectomy — two of eight — 100% for the single marginal resection, and 71.4% for segmental resection, five of seven. Across the whole series sequestrectomy was performed ten times, as a first or second procedure, and cured twice. Every patient whose segmental resection was the definitive operation ended up healed. Every patient who failed initial surgery had received at least 70 Gy.

Hyperbaric oxygen was part of the institutional routine: about 2.4 ATA, fifteen minutes of compression, sixty minutes at pressure, twenty minutes back down, with twenty preoperative and ten postoperative sessions offered to patients who consented. When the authors went looking for factors that distinguished healing from non-healing after sequestrectomy, oxygen was not one of them. In that subgroup six patients received hyperbaric oxygen and two healed; two did not receive it and neither healed. The authors write, plainly, that no clear background factor separated the two outcomes — and then, more usefully, that the numbers were too small for statistics to say anything at all.

None of this shows that hyperbaric oxygen is useless. It shows something subtler and more awkward: in a real cohort, the oxygen decision was invisible next to the surgical decision. Osteoradionecrosis is not one disease with one answer. It is a spectrum from a bit of exposed bone that will sequester on its own to a fractured, infected mandible that needs to be rebuilt with a leg bone. Whether a chamber helps depends entirely on where on that spectrum a patient sits, and almost no study is powered to look at one slice of it.

Three hundred and eleven jaws, and an argument in print

The largest recent attempt to isolate the oxygen question came from a United States tertiary centre. LaBanc and colleagues, writing in Facial and Plastic Surgery & Aesthetic Medicine, reviewed 311 patients who had segmental mandibulectomy and free tissue transfer for advanced mandibular osteoradionecrosis — 131 of whom received perioperative hyperbaric oxygen and 180 of whom did not. Complete flap failure occurred four times in the oxygen group and twice in the other. Nonunion, ten against fourteen. Exposed plate, seven against eight. Wound infections requiring a return to theatre, two against seven; infections managed with antibiotics alone, five against twelve. No comparison reached significance, and the authors concluded there was no difference in postoperative complications.

That is where the paper should have ended, and instead it produced a letter. In December 2025, two clinicians from Dartmouth Hitchcock’s hyperbaric medicine service published a one-page objection under a title that leaves nothing to interpretation: the data do not support the conclusions made. The surgical authors replied in the same issue. Both letters sit behind a paywall, so what can be reported here is that the exchange exists and that the disagreement is about inference from a retrospective cohort, not about the counts themselves.

The Dartmouth objection is not hard to reconstruct in principle. A retrospective series in which oxygen was given at clinician discretion is a study of who was chosen for oxygen, not of what oxygen did. The infection numbers — roughly five per cent of the treated group against eleven per cent of the untreated — are the kind of difference that looks like a signal and behaves like noise once a test is applied, and the authors were right to report no significant difference. But a null result in an observational design with a strong indication bias can be read two ways: as evidence of no effect, or as evidence that the comparison is too contaminated to read at all. The letter writers are entitled to the second reading. They are not entitled to declare the question settled in their favour, and neither are the surgeons.

Three trials, in the order they failed to settle it

The randomised record is short and its failures are instructive. In 2004, Annane and colleagues ran a double-blind trial in which patients with mandibular osteoradionecrosis received thirty preoperative and up to ten postoperative sessions at 2.4 ATA, with 100% oxygen in one arm and 9% oxygen in the other — a mix designed to reproduce the oxygen tension of room air at pressure. It was stopped at its second interim analysis, with 31% of the target enrolled, because recovery at one year looked worse in the oxygen arm: 19% against 32%, a relative risk of 0.60 with confidence intervals wide enough to contain almost anything. Not all participants underwent surgery, which by later standards is the treatment that counts.

HOPON, reported in 2019, asked whether oxygen prevents necrosis before it starts: 100 patients with a previously irradiated mandible who needed dental extractions or implants, randomised to twenty preoperative and ten postoperative sessions or to no chamber. At six months, 6% of each group had developed osteoradionecrosis. The trial stopped early at 65% of accrual because the event rate was lower than anyone had predicted, and the drop-out was driven by the logistics of attending thirty sessions.

Then came the pooled analysis of DAHANCA-21 and NWHHT2009-1, published in Radiotherapy and Oncology in 2022 — the same regimen, the same 2.4 ATA, patients selected because they had necrosis and an indication for surgery. Ninety-seven were enrolled, sixty-five were analysable, and healing at twelve months ran 70% with oxygen against 51% without. The odds ratio of 2.2 has a confidence interval from 0.7 to 7.0, and a p value of 0.13. The trial accrued 41% of its target and lost a third of its participants. The authors’ own sentence is the fairest summary available: the difference is not statistically significant because the study was underpowered and therefore prone to type II error.

Cochrane’s 2023 review bundled eighteen randomised trials and 1,071 participants and reached the conclusion that fits this mess: hyperbaric oxygen may result in complete resolution or substantial improvement of late radiation tissue injury, risk ratio 1.39, low-certainty evidence; it probably produces a modest reduction in osteoradionecrosis pain at twelve months, a mean difference of −10.72 points on a 0–100 scale, moderate certainty. A pooled analysis that folds in the two trials plus an earlier study of all necrosis grades put the healing risk ratio at 1.44 with no statistical heterogeneity at all. That is a real signal. It is also a signal assembled from small, unblinded studies with different endpoints, which is why it has not changed practice.

The harms the summaries skip

Cochrane also reports what the promotional material does not. Reduction in visual acuity, risk ratio 4.03, high-certainty evidence — usually transient, but real. Ear barotrauma on compression, risk ratio 9.08 when the control group was not pressurised, and no excess when a sham was used, which tells you most of that harm belongs to the pressure rather than to the oxygen. The FDA’s August 2025 letter to health care providers is blunter still, describing fires in hyperbaric devices that have caused serious injuries and deaths, and reminding operators that chambers are Class II devices cleared under product code CBF with specific instructions that are not always followed.

The guideline says one thing, the billing data another

In 2024, the ISOO-MASCC-ASCO guideline panel reviewed 80 studies and wrote the sentence that hyperbaric physicians have been arguing with ever since: “The use of hyperbaric oxygen in prevention and management of osteoradionecrosis remains largely unjustified, with limited evidence to support its practice.” This year’s clinically focused review in CA: A Cancer Journal for Clinicians softened the edges without reversing the verdict — prophylactic oxygen is not recommended, and for treatment of mandibular osteoradionecrosis the emphasis falls on patient selection, risk–benefit discussion and integration with surgery rather than on the chamber as the intervention.

Against that, two facts. First, Medicare pays. NCD 20.29 lists “osteoradionecrosis as an adjunct to conventional treatment” among covered indications, which means a clinician who orders it is practising inside American coverage policy even while an ASCO panel calls the practice unjustified. Second, the field’s own 2026 textbook chapter in Undersea and Hyperbaric Medicine notes that delayed radiation injuries remain the most frequent indication for hyperbaric treatment in the United States, accounting for roughly 40% of billings — and argues, at length, that the negative trials used inadequate protocols and flawed designs. That argument may even be correct. The difficulty is that it cannot be tested without the trial nobody has managed to finish.

What the choice actually contains

For a patient with exposed, non-healing jawbone after head and neck radiotherapy, the honest description of the evidence is this: the operation is the treatment, the oxygen is the adjunct, and the adjunct’s effect is somewhere between modest and unmeasured. The questions that follow are not rhetorical. Has a segmental resection with vascularised reconstruction been discussed, or only debridement? What protocol is being proposed — 2.4 ATA for thirty sessions with a defined relationship to surgery, or something improvised? Is the endpoint healing at twelve months, or the number of sessions completed? A 2025 systematic review from Amsterdam UMC found 17 studies and 640 head and neck patients treated with hyperbaric oxygen, positive results in 14 of them, and concluded that the evidence is mostly low quality with a high risk of bias. That is not a reason to refuse treatment. It is a reason to know which claim is being made.

The uncomfortable finding in the new cohorts is not that hyperbaric oxygen failed. It is that the therapy was never the variable that mattered most, and thirty years of research have not yet produced a study large enough to say so out loud.

Sources

← Back to blog