One Stroke Ward, Three Windows: A 429-Patient Cohort on When HBOT Starts

A wall clock past midnight above a stroke-bay gurney in a dim hospital corridor, with a steel hyperbaric chamber visible through the doorway behind it

Every day in a neurology department in Foshan, southern China, some of the patients admitted with an acute ischaemic stroke end up inside a hyperbaric chamber before their first day is out. Between January 2022 and April 2025 the hospital screened 1,627 stroke admissions. Of those, 486 received hyperbaric oxygen therapy within 72 hours of symptom onset — just under a third of the ward’s acute ischaemic stroke population. Four hundred and twenty-nine had records complete enough to analyse, and a paper published in Experimental and Therapeutic Medicine on 13 October sorts them by how quickly the chamber door closed.

Patients treated within 24 hours of symptom onset had 2.68 times the odds of a favourable outcome at 90 days — a modified Rankin Scale score of 0 to 2 — compared with patients treated between 48 and 72 hours (95% CI 1.55 to 4.63). The middle window, 24 to under 48 hours, came in at 1.72 (1.04 to 2.82). Ordinal analysis across the full Rankin distribution moved the same way: a common odds ratio of 2.44 for the earliest window, 1.68 for the second. Transcranial Doppler responses were more frequent in both earlier groups, while a higher rate of early neurological improvement and a lower rate of early deterioration appeared only in the under-24-hour group. Safety events — symptomatic intracranial haemorrhage, death, middle-ear barotrauma, oxygen toxicity — did not differ across the three windows.

A reader seeing only those numbers would conclude that the timing of hyperbaric oxygen after a stroke matters a great deal. The authors do not.

The schedule they were measuring

It is worth setting out what the treatment actually was, because “hyperbaric oxygen” covers everything from a pressurised bag in a wellness studio to a multiplace chamber in a trauma centre. Here it was 2.0 atmospheres absolute, 100% oxygen for 60 minutes at pressure, with 10 to 15 minutes of compression and decompression on either side — 80 to 90 minutes in the chamber per session, once daily, ten sessions planned over ten days. Patients completed a median of nine.

The cohort was also narrower than the word “stroke” suggests. Adults aged 18 to 85, anterior circulation only, baseline NIHSS between 4 and 20, no posterior-circulation events, no pre-stroke disability above a Rankin score of 2. Everyone received guideline-based acute care, including intravenous thrombolysis or endovascular thrombectomy when indicated — a detail that matters more than any other in this story, because the most dangerous version of hyperbaric enthusiasm is the one that postpones reperfusion.

The physiological marker was transcranial Doppler of the middle cerebral artery. A “haemodynamic response” was predefined as at least a 20% rise in mean flow velocity on the affected side without a rise of 0.10 or more in the pulsatility index. Combined with a neurological response, that became an exploratory “phenotype” carrying an odds ratio of 2.95 for a good outcome at 90 days.

Why the authors dismantle their own headline

The abstract contains a sentence most press releases would have deleted: “As the present observational study included no non-HBOT comparator, the present findings do not establish HBOT efficacy relative to standard care without HBOT or a causal benefit of earlier initiation.”

That is not boilerplate. It is the whole problem. Every patient in the cohort received oxygen under pressure, so nothing in the design can separate the drug from the patient who reached the chamber early — and the baseline table shows those patients were different. The earlier windows held the more favourable profile: higher ASPECTS, more thrombolysis, lower admission glucose, lower pulsatility index. Stroke severity was unevenly distributed too, and the subgroup analysis found the timing association varying by baseline NIHSS, an interaction the authors call exploratory and warn is unstable in a single-centre cohort with multiple comparisons.

Then there is the pathway. Onset-to-admission time was longer in the later windows, which is another way of saying that early treatment was available to people who arrived early, were recognised quickly, were stable enough to be moved, and were treated in a hospital with a chamber free. Multivariable adjustment, inverse probability weighting, a 72-hour landmark analysis, exclusion of everyone who had thrombectomy or thrombolysis, and an alternative Rankin cut-off of 0 to 1 all left the association standing. That is a thorough robustness programme. It still cannot touch confounding by clinical pathway.

The phenotype deserves its own warning. It was defined after treatment, by observing who responded, and an association between responding early and doing well at 90 days is close to tautological. To their credit, the authors treat it that way — they kept it out of the primary timing models and refuse to call it a validated surrogate. The paper is, in effect, an argument for a trial dressed as a cohort study.

Note the funding line as well: a Guangdong Provincial Medical Association clinical medicine fund with a dedicated hyperbaric oxygen stream. Speciality funding is not misconduct. It is simply the environment in which a positive timing signal is most likely to be generated, and most likely to be read as more than it is.

Eleven trials, seven hundred patients, no verdict

The randomised evidence has been stuck for a decade. The 2014 Cochrane review by Michael Bennett and colleagues pooled 11 trials and 705 participants and could combine data for only one outcome: death. Case fatality at six months was a wash, risk ratio 0.97 (95% CI 0.34 to 2.75). Four of fourteen disability and functional scales showed improvement — a mean 2.2-point reduction on the Trouillas disability scale, 27.9 points on the Orgogozo scale — and ten did not. The reviewers concluded there was “no good evidence” that hyperbaric oxygen improves clinical outcomes in acute ischaemic stroke, while conceding that a benefit “has not been excluded.”

A 2024 meta-analysis in BMC Neurology (Li and colleagues; eight trials, 493 patients) landed in the same place. NIHSS difference -1.41 points, confidence interval -7.41 to 4.58. Barthel index difference 8.85, confidence interval -5.84 to 23.54. Both straddle zero by a distance wider than any plausible benefit. The authors’ summary was that the findings “do not support the routine use” of the therapy, with the caveat that particular therapeutic windows and particular occlusion patterns remain unexplored.

The newest randomised data do not help either. A 30-patient pilot from Uttar Pradesh, published in Annals of African Medicine in April, enrolled patients three to six months after their stroke — a rehabilitation question, not an acute one — gave 24 sessions to fifteen of them, and reported improvements within both arms with no statistically significant difference between them.

Part of the reason is structural. Hyperbaric oxygen cannot be blinded; a sham requires altering pressure and oxygen fraction, and most protocols do not attempt it. The older trials predate modern reperfusion, so they tested oxygen in a population now treated very differently. And the group most likely to benefit — patients whose large vessel has been opened but whose microcirculation has not — barely existed as a trial population until a few years ago.

Oxygen is a drug, and this one has a dose ceiling

The largest randomised experiment on deliberately raising oxygen in acute stroke was not hyperbaric at all. The Stroke Oxygen Study, published in JAMA in 2017, randomised 8,003 adults at 136 UK centres within 24 hours of admission to continuous low-dose nasal oxygen for 72 hours, nocturnal oxygen only, or oxygen solely when clinically indicated. The primary outcome was the 90-day Rankin distribution. The odds ratio for a better outcome was 0.97 (0.89 to 1.05). No subgroup could be identified that benefited. No significant harm emerged either.

That is a negative trial at a dose — two to three litres per minute through nasal prongs — that a chamber exceeds by orders of magnitude. What the hyperbaric literature has lacked is any sense of where the upper edge of the curve sits. A Bayesian analysis in Neurocritical Care in September supplies one for acute brain injury. Using 11,264 ventilated patients with non-traumatic injury from an Australian and New Zealand registry, with priors drawn from two further datasets, the authors located the day-one arterial oxygen tension associated with the lowest ICU mortality. For the 2,099 patients with acute ischaemic stroke it was 100 to 160 mmHg. Extreme hyperoxia above 299 mmHg was rare, present in under a tenth of patients, and carried an 84 to 99% probability of an odds ratio above 1 for death.

Those are ventilated patients on a ward oxygen supply, not people at 2.0 ATA, where arterial oxygen tension runs far above the entire range that analysis could observe. The comparison does not condemn hyperbaric oxygen. It does puncture the assumption underneath it — that in an ischaemic brain, more dissolved oxygen must be better — and it leaves the burden of proof exactly where the trialists have left it.

There is a Doppler problem as well. Mean flow velocity is a velocity, not a flow; it shifts with vessel calibre, blood pressure, haematocrit and carbon dioxide. A 20% rise after a session is a real measured change, but it is not a measurement of oxygen reaching tissue, and the Foshan authors say so: transcranial Doppler, they write, “does not directly measure tissue oxygenation, infarct-core evolution, or penumbral salvage.”

In the rulebooks, stroke is not on the list

In the United States, hyperbaric chambers are Class II devices cleared through the 510(k) pathway under product code CBF, regulation 868.5470, with NFPA 99 and ASME PVHO-1 named as the recognised consensus standards for the pressure vessel and the room around it. Acute ischaemic stroke is not among the cleared indications, and the agency’s August 2025 letter to health care providers does not add it. That letter is about fire. It records reports of fires in hyperbaric devices that caused serious injuries and deaths, notes that the root cause is not known, and asks providers to follow the manufacturer’s instructions, ground equipment properly, train and maintain staff competence, monitor patients throughout, and keep potentially prohibited items — electrical or static devices among them — out of the chamber.

The Undersea and Hyperbaric Medical Society’s indications list, fifteenth edition, runs from air or gas embolism through carbon monoxide poisoning, clostridial myonecrosis, crush injury and compartment syndrome, decompression sickness, central retinal artery occlusion, selected problem wounds, severe anaemia, intracranial abscess, necrotising soft tissue infections, refractory osteomyelitis, delayed radiation injury, compromised grafts and flaps, thermal burns, idiopathic sudden sensorineural hearing loss, and avascular necrosis. Stroke is not on it. Nor is it on Medicare’s coverage list: NCD 20.29 limits payment to fifteen named conditions and, for wounds, to adjunctive use only.

So the practice documented in Foshan sits outside both the cleared indications and the reimbursement list of the country that regulates these devices most aggressively. It is not an American pattern. It is what happens in a health system with enormous stroke volume and a large installed base of chambers.

Two trials, 844 patients, open label

The question is at last being tested in the population where the penumbra argument actually bites. Two multicentre randomised trials are recruiting in China, both enrolling patients after endovascular treatment and both using the 90-day Rankin score as the primary outcome. One, sponsored by Beijing Tiantan Hospital, plans 420 patients with large vessel occlusion, opened in June 2026 and expects primary completion in September 2027. The other, from Beijing Chao Yang Hospital, plans 424 patients after endovascular therapy, began in July 2026 and expects completion in March 2028.

Both are open label. A patient knows whether they are in the chamber; an outcome assessor can be kept unaware, but expectation cannot be engineered away, and hyperbaric treatment is unusually good at generating expectation. That is not a reason to dismiss the results when they land in 2027 and 2028. It is a reason to read them slowly.

Until then, the fairest summary of hyperbaric oxygen in acute ischaemic stroke belongs to the Foshan authors themselves: earlier treatment was associated with better early physiology and better function at 90 days among people who received it, and nobody has yet shown that receiving it beats not receiving it. Anyone offered a chamber session in the days after a stroke should ask a simpler question first — whether the reperfusion therapy with randomised evidence behind it has already been given, and whether the chamber is being added to that therapy or substituted for it. Nothing in this literature supports a delay.

Sources

← Back to blog