The Ketogenic Diet for Treatment-Resistant Depression – and the Quieter Tool in the Control Arm

The ketogenic diet has been circulating in psychiatric conversation for several years on the strength of case reports, single-arm trials and a great deal of podcast enthusiasm. According to PubMed, we now have the first properly controlled randomised trial in treatment-resistant depression – 88 UK adults, an active comparator, 98% follow-up at six weeks, prepared meals and ketone strips supplied free (DOI). The result is technically positive and considerably more instructive than a yes-or-no verdict, provided you read past the abstract.

The number, and the arm nobody was watching

The primary outcome cleared significance by the width of a hair: a 2.18-point between-group PHQ-9 advantage for the ketogenic diet at six weeks, P=.05, with a confidence interval whose upper bound (-0.03) all but touches zero. The investigators had powered the trial to detect a 5-point difference and called that their minimal clinically important difference. They found less than half of it. By week 12 the gap was no longer significant. In per-protocol analysis – the participants who actually stuck to the diet – the difference lost significance, though the point estimate did not shrink: -2.81 in 61 people, with an interval that now crossed zero. In a post hoc sensitivity analysis using a different error structure, the six-week effect stopped being significant too. The authors, to their credit, say plainly that the clinical relevance is uncertain. They also push in the other direction, noting that antipsychotic augmentation in treatment-resistant depression buys roughly 3 placebo-adjusted PHQ-9 points, and arguing on that basis that 2.18 may not be clinically trivial.

Then there is the mechanism problem. If ketosis were doing the work, more ketones should mean more improvement. The opposite appeared: participants who spent the trial least ketotic improved by 12.9 PHQ-9 points, against 8.4 in those who sustained ketones at 1.5 mmol/L or above. This is exploratory and underpowered, but it is the wrong direction, and it sits badly with any account in which ketone bodies are the active ingredient. The trial's own gut-brain and microbiome analyses – the pathway most often invoked to explain why diet should move mood – have not yet been published. Until they are, the gut-brain story attached to this diet remains a hypothesis with a trial attached, not a finding.

Which leaves the control arm, and it is the most interesting thing in the trial – provided you do not misread it. These were people with a median 16-month current episode, mean baseline PHQ-9 of 19.5, 93% on antidepressant monotherapy that had already failed them twice. Asked to add one differently coloured vegetable or fruit each day, replace animal fats with plant oils, and take a 30-minute call each week, they fell 8.3 points. That figure cannot be read as the control diet working. There is no untreated arm in this trial, so there is nothing to measure 8.3 against – and the authors built the phyto diet deliberately as a credible placebo, describing it as having no known effect on depression and as unlikely on its own to influence the outcome. Regression to the mean, the natural course of a 16-month episode and the act of entering a trial all sit inside that number. What the two arms show together is narrower and more useful: most of the movement here was not attributable to what anyone ate. Meta-analytic placebo response in this population runs to a Hedges g of 1.05, which is precisely why the between-group comparison, and not the within-group drop, is the one carrying information.

What to do with this on Monday

Do not offer a ketogenic diet as a depression treatment on the strength of this trial. The effect is small, unstable across analyses, unrelated to ketosis, and it was produced under conditions no clinic reproduces: three prepared ketogenic meals a day supplied free, urine ketone strips, and – for participants who preferred to cook for themselves – £25 a week instead of the meals. The seven support sessions were not part of that scaffolding in any special sense; both arms received them. When the food stopped arriving, 48% of the ketogenic participants had stopped entirely within six weeks. A treatment that only works while someone is delivering your food is not yet a treatment; the authors say a more comprehensive intervention is needed before further clinical testing, and they are right.

There is one caveat worth holding. In prespecified subgroup analysis, the benefit sat entirely in those with severe depression, PHQ-9 20 to 27 (-4.73 at six weeks; 95% CI -8.16 to -1.30), and was absent – slightly negative, in fact – in the moderate group (0.16; 95% CI -2.30 to 2.63; P=.02 for interaction). Notably, the severe-group effect grew rather than faded by week 12 (-5.18; 95% CI -8.63 to -1.72). That is a hypothesis for the next trial, not a licence for this one. If a severely depressed patient with no metabolic contraindications arrives already committed to trying this, it is defensible to supervise rather than obstruct – with a dietitian, with the understanding that most people cannot sustain it, and with the honest framing that the evidence is one marginal trial.

What this trial does not hand you is a diet to prescribe. It is worth being exact about that, because the temptation is to swap one unsupported recommendation for another: if the ketogenic arm did not earn its claim, the phyto arm did not earn one either – it was built to be an inert comparator, and it was never tested against doing nothing. The honest summary for a patient who arrives having heard a podcast is that a well-run trial gave the ketogenic diet its best shot, with free meals and weekly support, and the diet-specific gain came in under the trial's own bar for mattering.

Two things do survive. The first is a research signal, not a prescription: the severe-depression subgroup separated (-4.73; 95% CI -8.16 to -1.30), which is exploratory and belongs in the next trial rather than in your clinic. The second is about what surrounds a diet rather than the diet itself – both arms received the same weekly contact, and both improved substantially. That is a hypothesis about structure and expectation, and this trial cannot test it, because it has no arm without them.