Ketogenic therapy for anorexia: a 22-person signal that could matter more than its size suggests

TL;DR

• UC San Diego researchers put 22 women with a history of anorexia nervosa on a strict 70/20/10 ketogenic diet for 14 weeks. 18 completed it.
• 72% of completers finished the trial scoring in the recovered or normal range on the standard Eating Disorder Examination Questionnaire (EDE-Q). Restraint, depression, and concern with eating, shape and weight all improved.
• Body weight did not fall. No participant dropped below BMI 17.5. This matters — keto in this population was the obvious worry going in.
• The hypothesised mechanism is neurometabolic: brain cells in anorexia appear to process glucose poorly; ketones are an alternative fuel that bypass the defect.
• Published in Nature. Sample is small, mostly White, single-arm, no control. The result is a signal, not a protocol.

What happened

A team led by Guido Frank, MD, professor of psychiatry at UC San Diego School of Medicine, ran a 14-week single-arm pilot of ketogenic dietary therapy in 22 women aged 18–45 with a documented anorexia history and BMI above 17.5 — meaning they were weight-restored or only mildly underweight at entry. The protocol targeted 70% fat, 20% protein, 10% carbohydrate, monitored by weekly ketone testing, weight checks, EDE-Q questionnaires, and combined dietitian / psychiatrist / peer-support supervision.

Eighteen women completed the full 14 weeks. Of those, 72% finished in the EDE-Q recovered or normal range. Restraint scores, depression scores, and the eating/shape/weight concern subscales all improved. Three months after the intervention ended, those who voluntarily continued the diet had slightly better EDE-Q scores still. Weights held — the design's central safety question — and no completer dropped below the BMI 17.5 floor.

The paper was published in Nature. New Scientist and Fox News carried it on 3–4 June 2026.

Why a 22-person pilot is news

Anorexia nervosa kills more of its patients than any other psychiatric disorder — somewhere in the 5–10% lifetime mortality range, depending on the cohort. There has been no genuinely new mechanism-based treatment in twenty years. SSRIs do not work for the core illness. Family-based therapy works for adolescents and fails to translate cleanly to adults. The relapse rate after weight restoration sits around 40%.

So when a credible university lab in Nature reports any pilot in which 72% of completers crossed the recovered threshold, the bar for paying attention is met. The bar for prescribing it isn't. Those are different bars and the press is already blurring them.

The mechanism is the interesting part

Standard models treat anorexia as a psychiatric disorder with metabolic consequences. The neurometabolic hypothesis Frank's group is testing inverts that: it treats anorexia as a metabolic disorder of the brain with psychiatric consequences.

The argument runs:

1. Brain imaging in anorexia patients shows altered glucose metabolism in regions governing reward, interoception, and anxiety.
2. The cells of those regions appear to release energy poorly when fed glucose — the standard brain fuel.
3. Ketones (β-hydroxybutyrate especially) are an alternative fuel the same cells can use. They bypass the impaired glycolytic step.
4. If you reduce the anxiety load coming from a starved-feeling brain, the compulsion to restrict food may follow.

This isn't a wholly new idea — it has been kicking around metabolic-psychiatry labs (Chris Palmer, Iain Campbell, Shebani Sethi) for several years, mostly applied to bipolar disorder and schizophrenia. What's new is that someone ran the trial in an AN population without the patients losing weight. That was the central design risk and it cleared.

Where the numbers stop agreeing

Read past the headline.

• n = 22 enrolled, 18 completed. Four dropouts in a 22-person trial is an 18% attrition rate. Standard for psychiatric pilots, but it inflates the apparent response.
• Single-arm. There is no placebo group. The EDE-Q is self-report. Both inflate response in psychiatric pilots. A controlled trial with a similar-effort comparison diet is the next step, and not a small one.
• Recovered or normal on EDE-Q is not "cured." The questionnaire captures attitudes and behaviours over the prior 28 days. Long-tail relapse is the defining feature of AN. The three-month follow-up is encouraging but short.
• Demographics narrow. Predominantly White, all female, all weight-restored to BMI > 17.5 at entry. The trial cannot tell you what happens in lower-BMI, male, or non-White patients.
• The diet itself is hard. A 70/20/10 ketogenic plan is operationally demanding in a generally well population. In a population whose illness is centred on food restriction and control, the intervention design — a high-supervision, high-rigidity eating plan — collides with the disorder it's trying to treat. The trial provided weekly dietitian, psychiatrist and peer-support contact. That level of wraparound is not what an outpatient adult with AN typically gets.

None of these invalidate the result. All of them mean a 22-person pilot is the beginning of a research programme, not a treatment guideline.

What this isn't

This is not licence for anyone with an eating disorder to put themselves on a ketogenic diet. The trial wraparound — dietitian, psychiatrist, peer support, weekly ketone monitoring — is the active ingredient as much as the macros are. Unsupervised keto in an AN patient is a known way to relapse fast.

It is also not the moment for the "metabolic psychiatry" community to declare the case closed. They have one positive pilot in one disorder, in line with their working hypothesis. That is what the evidence supports and no more.

Cross-layer implications

• Treatment economics. If a replicated, controlled trial holds up, ketogenic dietary therapy could become the first metabolically-anchored treatment for AN. That reshapes how residential and outpatient programmes are staffed — dietitians and metabolic monitoring move from the periphery to the centre.
• Insurance. Dietary therapy is not a billable code in most US plans for psychiatric indications. A Nature-grade evidence chain forces that conversation.
• Research pipeline. Bipolar disorder, treatment-resistant depression and schizophrenia trials of ketogenic therapy are already running (Stanford, Edinburgh, McLean). An AN positive signal accelerates funding for that whole programme.
• Cultural risk. "Keto cures anorexia" headlines have already arrived. The disorder's relationship with restrictive eating frameworks is the exact place where careless coverage can do damage.

Recommendations

Addressed to the patients, families and clinicians for whom this story is operationally relevant. Not investment advice; not a treatment protocol.

If you or someone close to you has anorexia:

• Do not start a ketogenic diet on your own based on this study. The active ingredient in the trial includes the supervision package. Recreating just the macros recreates a relapse risk.
• It is reasonable to raise this study with a treating psychiatrist or dietitian. Ask whether a supervised metabolic-psychiatry programme exists near you. As of mid-2026, fewer than a dozen exist worldwide.

If you are a clinician treating eating disorders:

• The paper is Frank et al., Nature, 2026. Worth reading the methods in full before forming a view.
• The realistic near-term action is not implementation. It is referring suitable patients into the controlled trial that will follow this pilot.

If you are a payer, a programme director, or a policy lead:

• Watch for the controlled follow-up trial. That is the data point that should change procurement, not this one.
• Metabolic monitoring infrastructure (ketone strips, dietitian capacity, psychiatric integration) is the rate-limit if this scales. Building that capacity is a sensible no-regret move regardless of whether keto specifically wins.

Uncertainty ledger

• Will a controlled trial replicate the 72% figure? Most psychiatric pilots lose 30–50% of their effect under control. Plan for that.
• Does the effect hold in lower-BMI, male, and non-White patients? Unknown.
• Is it the ketones, the wraparound, the structure of having a tightly-defined "what to eat" rulebook, or some combination? The pilot cannot separate these.
• What is the 12-month relapse rate? Three-month follow-up is encouraging; AN relapse curves typically run longer.

Bottom Line

Anorexia nervosa is the deadliest psychiatric disorder, and the standard treatment toolkit has not meaningfully expanded since the 1990s. A 22-patient Nature pilot from UC San Diego is too small to change practice and too important to ignore: it is the first credible signal that the disorder may be treatable through the metabolism rather than around it. The right response is a controlled trial, not a TikTok protocol.

Sources

• Frank G et al., Nature, 2026 — primary publication. Tier 1.
• New Scientist, "Keto diet shows real promise for anorexia recovery," 3 June 2026. Tier 2.
• Fox News Health, "Keto diet may help reduce anorexia nervosa symptoms," 4 June 2026. Tier 3 (used for press-release framing, not load-bearing).
• UC San Diego Health press release, 3 June 2026. Tier 2.