Psychiatric HBOT's Big Number Meets Its Methodologists

Stylised forest plot on a deep navy field: a dashed vertical no-effect line, thirteen rows of teal square markers strung on confidence-interval whiskers that are wide and faint for the small imprecise studies and tight and bright for the precise ones, and a single large amber diamond marking the pooled estimate at the foot of the plot, with a dim amber-rimmed pressure gauge in the dark right-hand field

On 18 October, a journal called Psychiatry and Clinical Neurosciences published thirteen authors’ defence of a number. The number was Cohen’s d = −0.82: the pooled effect of hyperbaric oxygen on depressive symptoms in a meta-analysis their group had published six weeks earlier. The defence ran under a title that gives away the fight — a “Response to Commentary” — and its abstract is one sentence long. They were clarifying “the rationale for the meta-analytic models used” and adding “additional sensitivity analyses.”

The commentary they were answering had appeared in the same journal in July. Its authors were Yuki Matsuda, Yuta Suzuki, Aran Tajika and Taro Kishi, a group associated with the Kyoto evidence-synthesis school that has spent a decade auditing other people’s meta-analyses. Their title was a flat methodological assertion: “Model choice in meta-analysis should not be driven by Cochran’s Q test.”

One day after the Toronto reply landed, Frontiers in Medicine published a narrative review of hyperbaric oxygen for major depressive disorder whose closing assessment could hardly be blunter. There is, the authors wrote, “currently a dearth of direct evidence supporting the indication of HBO2 for MDD.”

Three documents, three weeks, one subject. This is what the psychiatric end of hyperbaric medicine looks like in late 2026: a strong pooled effect size, a live argument about the arithmetic behind it, and a clinical literature that its own reviewers describe as not yet there.

What the Toronto review actually pooled

The meta-analysis came out of the AI for Mental Health Program at St. Michael’s Hospital in Toronto, with a co-author list that includes Canada’s military and veteran psychiatry establishment. It searched three databases, set psychiatric symptom severity at end of treatment as the prespecified primary endpoint, and modelled in R version 4.4.0.

Seventeen studies covering 920 adults met inclusion criteria. Nine randomised trials fed the pooled analysis. For depressive symptoms the result was d = −0.82 with an I² of 82 per cent — a large effect carrying an enormous amount of between-study inconsistency. Anxiety, drawn from three studies, came in at d = −0.59 with I² = 42.8 per cent. Post-traumatic stress disorder produced the biggest figure of all, d = −1.96, from two sham-controlled trials with I² = 0 per cent.

The authors were not triumphalist about any of it. Their own abstract calls the evidence “preliminary,” the effects “variable,” and the estimates “heterogeneous,” and asks for larger, rigorously designed trials with standardised protocols. Adverse events were mild and transient, mostly ear discomfort and short-lived anxiety.

An effect that shrinks as the comparator gets better

Buried in the abstract is the finding that should govern how the −0.82 is read. A subgroup analysis by comparator produced significant heterogeneity (P = 0.038) and three quite different answers: d = −0.68 against sham, d = −1.33 against usual care or psychotherapy, and d = −0.60 against antidepressants that did not reach significance.

That gradient runs in the wrong direction for a specific treatment effect. The largest number comes from the weakest comparison. Against usual care — which in these trials often means little more than being seen regularly and given attention — hyperbaric oxygen looks transformative. Against an active drug it looks like a modest, statistically indistinct signal. Against a sham chamber, where the patient is pressurised, fitted with a mask and treated identically apart from the gas, the effect is real but roughly half the headline figure.

The single best-controlled trial in this space shows both why the sham comparison matters and how hard it is to run. Keren Doenyas-Barak and colleagues at the Sagol Center in Israel randomised 63 male veterans with combat-associated PTSD to sixty daily sessions of either 100 per cent oxygen at 2.0 ATA or 21 per cent oxygen at 1.02 ATA, with five-minute air breaks every twenty minutes. Fifty-six men completed. The treatment arm’s CAPS-5 score fell from 42.6 to 25.8 and held at 25.1 at follow-up. The sham arm’s rose, from 45.1 to 47.8 and then 49.2. Depression measures moved in the same direction, significantly on the BDI-II and the DASS-21 depression subscale; the stress and anxiety subscales did not shift.

The design is as clean as this literature gets, and the group difference is still an intersection of two trajectories — one falling, one drifting upward.

Why a statistics letter matters to a pressure chamber

Cochran’s Q is the standard test for whether the studies in a meta-analysis are estimating the same thing. When Q comes back non-significant, the temptation is to conclude the studies agree and switch to a fixed-effect model, which produces tighter confidence intervals and, often, a more impressive pooled estimate. The problem, well documented in meta-epidemiology, is that Q has low power: with the handful of small trials typical of hyperbaric research, a non-significant Q is weak evidence of homogeneity, and changing the model after seeing Q turns a pre-specified analysis into a data-dependent one. Matsuda and colleagues’ title states the rule that follows from this. Model choice should follow from the question the review is asking — whether one true effect is being estimated, or a distribution of effects across populations and protocols — not from a heterogeneity test.

What the commentary argues beyond that sentence, I cannot fairly report: the full text sits behind a paywall and PubMed indexes it as a one-line record. What is verifiable is that the parent review used random-effects models, that a critique of model choice exists in print, and that the authors answered it with sensitivity analyses rather than retraction. The exchange is not a scandal. It is the ordinary, slightly ugly machinery of evidence synthesis working in public, and it happens to be attached to the one paper that psychiatric hyperbaric marketing now cites.

For anyone weighing a course of treatment, the consequence is narrow but real: the pooled effect behind the pitch is not a fixed quantity. It moves when you change the model, and it moves a great deal more when you change the comparator.

Three reviews, three different answers on dose

Where the field disagrees most openly is not whether oxygen helps but how much pressure it takes. Here the Toronto review is unusually specific. It reports a clear dose-response: a robust benefit at 2.0 ATA (d = −0.93) and, in its words, none at 1.2 ATA.

That single sentence is the most consequential thing in the paper for the commercial hyperbaric sector, because 1.2 to 1.3 ATA is the range in which “mild” hyperbaric, soft-sided chambers and wellness lounges operate. On this analysis, the pressure that a great many paying customers actually receive produced no measurable antidepressant effect.

Andrews and Harch, reviewing PTSD rather than depression, drew a different boundary. Their 2024 dosage analysis covered eight studies and 393 subjects, seven of them randomised trials, and found statistically significant improvement alongside reliable or clinically significant change across a pressure range from 1.3 to 2.0 ATA, with a linear dose-response across cumulative oxygen exposures from 1,002 to 11,400 atmosphere-minutes. They also recorded something the promotional literature leaves out: at the highest oxygen doses, 30 to 39 per cent of subjects had a severe but reversible exacerbation of emotional symptoms. Their conclusion, that PTSD “can no longer be considered strictly a psychiatric disease,” is a claim about mechanism, not a dosing rule, and the two reviews are not measuring the same outcome in the same patients. Still, “none at 1.2 ATA” and “benefit from 1.3 ATA upward” cannot both be the last word.

A third data point is too small to settle anything and worth reporting anyway. Gur and colleagues at the Israel Naval Medical Institute randomised severe PTSD patients to sixty sessions at either 2.0 or 2.5 ATA and published preliminary results for the first nine completers: a mean CAPS-5 improvement of −14.0 at 2.5 ATA against −5.3 at 2.0, not statistically distinguishable (P = 0.8), with most of the gain arriving by weeks six to eight and then flattening despite four more weeks of treatment.

That plateau is a genuine puzzle. If the benefit is dose-dependent and cumulative, symptoms should keep falling. If the benefit is largely the structure, attention and hope of a ninety-minute daily ritual, a plateau around week eight is exactly what you would expect.

The trial that improved both arms, and the evaluation that inflated one

The most instructive randomised result in this whole area is not about PTSD at all. Lindell Weaver’s team at LDS Hospital in Salt Lake City ran a double-blind trial in 47 adults with persistent symptoms after non-stroke brain injury, giving forty chamber sessions or forty sham sessions over twelve weeks. The primary symptom inventory improved by 10.6 points with hyperbaric oxygen and 3.6 with sham, a mean difference of 7.0 (95 per cent CI 1.7 to 12.3, p = 0.01). The hyperbaric arm also did better on olfaction, anxiety, sleep and vestibular complaints.

But both arms improved on depression, headaches, PTSD symptoms, physical quality of life and the degree to which difficulties interfered with daily life. A chamber that is pressurised and a chamber that is not both make people feel attended to. Every uncontrolled programme evaluation in this literature inherits that problem, and the largest recent one shows the size of the inheritance: fifty Florida veterans who completed before-and-after surveys around a hyperbaric course saw their PCL-5 scores fall from 51.0 to 20.6, d = −1.91, sustained at six months. The authors were careful to call it a single-arm programme evaluation and to ask for controlled trials. Numbers of that magnitude, from that design, are the raw material of clinic websites.

America’s own psychiatric establishment has looked at this and held its ground. A 2024 state-of-the-evidence review in the American Journal of Psychiatry, written from the National Center for PTSD, grouped hyperbaric oxygen with stellate ganglion block and neurofeedback among treatments of rising public interest and judged the support for most novel somatic treatments “preliminary and highly variable.”

No cleared indication, and a safety letter that does not expire

None of this sits on a regulated indication. Psychiatric disorders do not appear among the fifteen indications in the Undersea and Hyperbaric Medical Society’s fifteenth edition, whose list runs from air or gas embolism through carbon monoxide poisoning, decompression sickness, refractory osteomyelitis and compromised grafts to avascular necrosis. They are absent from the FDA’s cleared uses for hyperbaric devices, which are regulated as Class II instruments under 510(k) Product Code CBF. A depressed or traumatised patient entering a chamber in the United States is, in regulatory terms, receiving an off-label treatment, and outside a trial usually paying for it.

The device side of that absence deserves more attention than it gets. In August 2025 the FDA wrote to health care providers about hyperbaric oxygen therapy devices in terms it does not use lightly: it was aware of “reports of fires that occurred with HBOT devices that resulted in serious injuries and deaths,” and the root cause was not known. The letter asks facilities to follow manufacturer instructions, to maintain fire prevention and safety measures, to train and monitor staff, to keep patients supervised for the duration of treatment, to observe cleaning and maintenance intervals, and to avoid prohibited electrical or static items and non-compatible clothing.

A sixty-session psychiatric course means sixty hours inside a pressure vessel breathing pure oxygen. Whatever one concludes about the effect size, the exposure is not trivial, and the safety expectations attached to it do not soften because the indication is unapproved.

What would actually settle it

The honest summary is that psychiatric hyperbaric medicine has produced a real, reproducible signal in small trials, an effect size that is sensitive to how it is modelled and what it is compared against, a dose question that three reviews answer three ways, and no adequately powered sham-controlled trial in major depressive disorder — the condition the marketing leans on hardest. The Toronto authors said as much themselves, and the Front Med review said it again on 19 October.

Settling it would take a trial designed the way the Weaver brain-injury study was designed, sized for depression rather than for feasibility, with a model chosen before the data arrive and a primary endpoint that is not a self-report questionnaire. Until then, patients are being asked to weigh a large pooled number whose authors call it preliminary, in a field whose methodologists are still arguing about the arithmetic, for a condition with no cleared indication and no shortage of clinics willing to sell the course.

That is not an argument against hyperbaric oxygen. Decompression sickness, carbon monoxide poisoning and a short list of wound, radiation and graft indications remain firmly evidenced, and none of them depends on this literature. It is an argument against letting one contested effect size do the work of a trial that has not been run. People considering hyperbaric oxygen for depression or PTSD should discuss it with the clinicians who already know their history, and should be wary of any programme that presents a pooled figure without the I² that sits next to it.

Sources

← Back to blog