Published on 18 September 2026
Ketogenic diet for treatment-resistant depression: a result at the edge of significance, short of the trial’s own threshold
The essentials
Eighty-eight adults with treatment-resistant depression were randomized in the United Kingdom to a six-week ketogenic diet or a control diet designed to be inert. On the primary outcome, the PHQ-9 at six weeks, the between-arm difference is −2.18 points (95% CI: −4.33 to −0.03; p = 0.05). The trial had been powered to detect a 5-point difference, one the investigators themselves judged clinically relevant. Not only does the observed result fall short of that threshold, the confidence interval excludes it entirely. The effect is not recovered in the per-protocol analysis at six weeks (−2.81; 95% CI: −6.08 to 0.45; p = 0.09), nor at twelve weeks (−1.85; 95% CI: −4.04 to 0.33; p = 0.10), nor on remission, nor on five of the six secondary outcomes. By twelve weeks, 48% of participants in the ketogenic arm had stopped the diet entirely and 20% were still following it on more than half of days. Ketone concentrations were not associated with improvement, and those least in ketosis improved the most. The authors conclude that an antidepressant benefit is seen at six weeks whose “clinical relevance is uncertain.” PEB endorses that formulation and draws no dietary recommendation from it.
The context
Nutritional psychiatry occupies a place in public debate wildly out of proportion to its level of evidence. The ketogenic diet is the most visible case: it has circulated for years through case reports, open-label series, and testimonials, with claimed effects that no controlled design had yet put to the test in depression.
The scale of the gap is worth stating up front. In 2025, a study conducted in American college students with major depressive disorder reported a 69% reduction in PHQ-9 scores and a 71% reduction on the HRSD over ten to twelve weeks. That study was prospective, single-arm, with no control group, no randomization, and no blinding, with 24 participants enrolled and 16 analyzed. It was widely picked up by the press. It is against this backdrop that the work of Gao and colleagues has to be read: this is not one more study finding an effect, it is the first time the question has been asked in a design where the answer could come back negative.
The authors state their claim to priority narrowly, and it is worth keeping it exactly as worded: their controlled design provides, to their knowledge, “the first evidence of modest short-term antidepressant effects.” They do not claim the first randomized trial of ketogenic diets in psychiatry as a whole, a field in which other populations have been explored.
The study at a glance
| Population | 88 adults aged 18 to 65 with treatment-resistant depression, mean age 42.1 years, 61 women. At least two antidepressants at an adequate dose during the current episode per NICE guidance, current antidepressant for at least four weeks, baseline PHQ-9 of 15 or higher, mean 19.5 (SD 3.2). United Kingdom, trial conducted 22 February to 15 June 2024, intervention delivered remotely. |
|---|---|
| Intervention | Modified ketogenic diet for six weeks, less than 30 g of carbohydrate per day, 15% to 20% of energy intake from protein. Three prepared meals per day, free snacks and urinary ketone test strips, or £25 per week in food vouchers for those who preferred to cook. Morning urinary ketones measured at least twice weekly. Target of stable weight, with portion adjustment if weight changed by at least 0.5 kg on two consecutive days (n = 44). |
| Comparator | A so-called “phytochemical” diet: add a differently colored fruit or vegetable every day and replace saturated animal fats with unsaturated vegetable oils. £10 per week in food vouchers. The authors write that these minimal changes, to their knowledge, have no known effect on depression and served as a “credible placebo” (n = 44). |
| Primary outcome | Change in PHQ-9 score between baseline and week 6. |
| Secondary outcomes | Remission (PHQ-9 ≤4), anxiety (GAD-7), anhedonia (SHAPS), cognitive complaints (PDQ-5), quality of life (SF-12), occupational and social functioning (WSAS). The mechanistic analyses announced in the protocol, morning cortisol, gut microbiota, and reward sensitivity, will be reported separately according to the authors. |
| Design | Parallel-group randomized controlled trial, described as single-blind. The first 10 participants by simple randomization, the next 78 by minimization with a random element on BMI (below or above 30) and baseline PHQ-9 band (15 to 19 versus 20 to 27). Allocation concealed by the database until assignment. Seven 30-minute support sessions in both arms, six-week intervention followed by passive follow-up to week 12. Linear mixed-effects regression on time, group, and their interaction, intention-to-treat analysis on all 88 randomized, adjusted for the stratification factors. Sponsor: University of Oxford, NCT06091163. Individually randomized trial, CEBM 1b. |
Quality control
| Preregistration | Strong FINDING: ClinicalTrials.gov NCT06091163. Primary outcome, stratification factors, and subgroup analyses stated in the methods as prespecified. |
|---|---|
| Randomization and allocation | Strong FINDING: Minimization with a random element on both relevant prognostic factors, allocation concealed by the database until assignment. |
| Power | Strong FINDING: 32 participants per group were sufficient to detect a 5-point PHQ-9 difference with 90% power. 44 per group were randomized. Recruitment stopped at 88 instead of the planned 100 because retention was better than the 35% loss to follow-up that had been anticipated. On its primary outcome and for the difference it targeted, this trial was therefore not underpowered. |
| Retention | Strong FINDING: 86 of 88 at week 6 (98%), 82 of 88 at week 12 (93%). Two withdrawals in the ketogenic arm (5%), one in the control arm (2%). Rare, at this level, for a demanding dietary intervention. |
| Blinding | Major weakness FINDING: The trial is described as single-blind, but what is masked is the nature of the other arm’s diet, not the allocation itself. Each participant knows exactly what they are eating. The primary outcome is a self-report completed by a patient recruited for a ketogenic diet trial who knows whether they are following one. What is actually masked reduces to two things: the nature of the other group’s diet, and the fact that the data analyst did not deliver the intervention. |
| Material matching between arms | Major weakness FINDING: Human support is matched, seven sessions on either side. Nothing else is: three prepared meals delivered daily, free snacks and strips on one side, or £25 weekly for those who cooked for themselves, against £10 weekly on the other. The authors themselves acknowledge that providing prepared meals may have improved mood through convenience, independent of the diet’s composition. |
| Multiplicity | Caveat FINDING: Six secondary outcomes, several sensitivity analyses, two prespecified subgroup analyses, and post hoc analyses. The threshold used is two-sided p<0.05, with no correction for multiple comparisons mentioned. |
| Missing data | Strong FINDING: Intention-to-treat analysis, plus sensitivity scenarios: available cases, imputation with a PHQ-9 of 20, of 27, baseline value carried forward, last value carried forward, or remission with a score of 4. |
| Independence and transparency | Strong FINDING: Funded by the NIHR Oxford Health Biomedical Research Centre, non-industry. Open access under a CC BY license. One author, Dr Browning, reports personal fees from Alto Neuroscience, Engrail Neuroscience, Empyrean Neuroscience, and Janssen, outside the submitted work. The publication names no commercial supplier of the prepared meals and reports no product donations. |
| Observation horizon | Caveat FINDING: Six weeks of intervention, twelve weeks of total follow-up. The authors note that this duration is too short to assess the long-term adverse effects of ketogenic diets, citing hyperlipidemia, kidney stones, and vitamin deficiencies. |
The results
Mean difference between the two arms on the PHQ-9 at six weeks, in favor of the ketogenic diet (95% CI: −4.33 to −0.03; p = 0.05). The trial had been powered to detect 5 points. The confidence interval excludes that value: at the 95% threshold, the differences compatible with the data do not reach the magnitude the investigators had themselves defined as clinically relevant.
| PHQ-9 at week 6, intention-to-treat | Within-group change: −10.5 (SD 7.0) under the ketogenic diet, −8.3 (SD 5.1) under control. Difference −2.18 (95% CI: −4.33 to −0.03; p = 0.05). Cohen’s d −0.68 (95% CI: −1.35 to −0.01). PEB reading: Right at the conventional threshold: the interval’s upper bound is −0.03, and the printed value is p = 0.05. Cohen’s d carries the label of a moderate effect, but it corresponds to a raw difference of 2.18 points on a 27-point scale whose relevance threshold, set by the trial itself, was 5. The interval around d, from −1.35 to −0.01, is too wide to distinguish a substantial effect from no effect at all. |
|---|---|
| PHQ-9 at week 6, per-protocol (n = 61, of whom 34 on the ketogenic diet and 27 on control) | −2.81 (95% CI: −6.08 to 0.45; p = 0.09; standardized mean difference −0.88; 95% CI: −1.89 to 0.14). PEB reading: Not significant. The point estimate is somewhat larger, but the interval crosses zero. The robustness of the primary result is not established. |
| PHQ-9 at week 12, intention-to-treat | −1.85 (95% CI: −4.04 to 0.33; p = 0.10). Cohen’s d −0.58 (95% CI: −1.26 to 0.10). PEB reading: Not significant. Six weeks after dietary support stopped, the gap is no longer detectable. |
| PHQ-9 at week 12, per-protocol | −1.84 (95% CI: −5.17 to 1.48; p = 0.27). PEB reading: Not significant. |
| Remission (PHQ-9 ≤4) | Week 6: 11 of 44 (25%) versus 4 of 44 (9%). Week 12: 8 of 44 (18%) versus 4 of 44 (9%), after relapse in 3 patients. No significant difference between groups. PEB reading: A visible raw gap, not significant. The sample size cannot support a conclusion on a binary outcome. This is the figure most likely to be lifted out of context, and the least interpretable on its own. |
| Anxiety (GAD-7) at week 12 | −2.02 (95% CI: −3.95 to −0.10; p = 0.04; standardized mean difference −0.42; 95% CI: −0.82 to −0.02). PEB reading: The only secondary outcome to reach significance, at a time point where the primary outcome no longer does, with no correction for multiplicity. The authors themselves describe it as a transient effect on anxiety. Notably, the publication’s abstract states that no difference was observed on secondary outcomes, a statement the results section qualifies on this specific point. One isolated result within a series is not confirmation, it is a hypothesis. |
| Other secondary outcomes | Anhedonia (SHAPS), cognitive complaints (PDQ-5), quality of life (SF-12), functional impairment (WSAS): no significant difference between groups. PEB reading: Uniformly negative. If a diet were genuinely improving mood, one would expect a consistent signal on anhedonia and functioning as well. |
| Prespecified subgroup, severity | At week 6, severe depression (PHQ-9 20 to 27): −4.73 (95% CI: −8.16 to −1.30). Moderate depression (PHQ-9 15 to 19): 0.16 (95% CI: −2.30 to 2.63). p = 0.02 for the between-subgroup difference in effect. At week 12: −5.18 (95% CI: −8.63 to −1.72) versus 1.24 (95% CI: −1.27 to 3.78). PEB reading: The most interesting result, and the most fragile. The analysis is stated as prespecified and severity was a minimization factor, which counts in its favor. But the trial is not powered for subgroups, the interaction is one of several tested, and an effect concentrated in the most severe cases is also the expected signature of asymmetric regression to the mean. Treat it as hypothesis-generating, not as a criterion for selecting patients. |
| Adherence during the intervention | 28 of 44 participants (64%) maintained urinary ketones of 1.5 mmol/L or higher at least 60% of the time. At least 84% measured their ketones at least twice weekly as requested, at least 50% daily. PEB reading: Correct, under intensive support with seven coaching sessions, delivered meals, and weekly biological monitoring. This figure describes a research apparatus, not an office visit. |
| Adherence at week 12, without support | 21 of 44 (48%) report having stopped the ketogenic diet entirely. 9 of 44 (20%) follow it more than half of days. In the control arm, 2 of 44 (5%) stopped entirely and 21 of 44 (48%) still follow the instruction more than half of days. PEB reading: A collapse six weeks after support stopped: about one in five still holds to the ketogenic arm’s diet, against nearly one in two in the control arm. This is the single most transposable finding of the whole trial, and the least cited. |
| Ketones and improvement | Kendall’s τ-b correlations: no significant association between ketone concentration and PHQ-9. Mean urinary ketones below 1.5 mmol/L (n = 20): −12.9 (SD 6.3). At or above 1.5 (n = 23): −8.4 (SD 7.0). Exploratory post hoc analyses. PEB reading: This contradicts the presumed mechanism. Ketosis is supposed to be the active ingredient. Those who produced the least of it improved the most. An exploratory result, on two subgroups of about twenty people each and on an imprecise urinary measure, but one that does not run in the hypothesis’s favor. |
| Safety | No serious adverse event occurred, and none related to the research procedures. No significant weight change, weight stability being a protocol goal in both arms. PEB reading: Reassuring at six weeks, silent beyond it. The trial did not track metabolic parameters or changes to antidepressant treatment during the study, a limitation the authors themselves flag. |
A reading note on the control arm: the “phytochemical” diet was built to do nothing: add one colored fruit or vegetable a day, replace animal fats with vegetable oils. The authors state in so many words that it served as a credible placebo. In that arm, the PHQ-9 fell by 8.3 points over six weeks, a fall that cannot be attributed to the diet itself. In other words, close to four-fifths of the improvement seen in the ketogenic arm is also found in an arm receiving no intervention presumed to be active. That said, this should not be over-read in the opposite direction either: there was no true no-intervention arm, so the respective contributions of natural course, regression to the mean, expectancy, and weekly contact cannot be decomposed from this data.
Critical appraisal
| Execution quality | Strong FINDING: This is a well-conducted trial. Appropriate randomization, 93% retention at twelve weeks, intention-to-treat analysis, sensitivity tested under several scenarios, public funding, open access, a calibrated conclusion. The critique that follows does not concern the investigators’ rigor. |
|---|---|
| Blinding and a subjective outcome | Major weakness FINDING: The crux of the problem. The primary outcome is a self-administered scale, filled in by people who know which diet they are following, in a trial whose purpose was disclosed to them. Blinding is structurally impossible for a visible dietary intervention. This does not disqualify the trial, but it does mean that a 2.18-point gap on a self-report cannot be confidently attributed to the diet’s composition rather than to expectancy. A confirmation from an outcome independent of patient report, or from a blinded external rater, is missing. |
| Missing its own threshold | Major weakness FINDING: The investigators defined, before the trial, that a 5-point PHQ-9 difference would be clinically relevant, and powered the study to detect it with 90% power. They found 2.18, with an interval reaching up to −4.33. The 5-point difference is therefore excluded by the data. This point is rarely emphasized and yet decisive: the result is not merely modest, it is incompatible with the hypothesis the trial was built to test. |
| Statistical fragility | Caveat FINDING: p printed at 0.05, interval’s upper bound at −0.03, not significant per-protocol, not significant at twelve weeks, not significant on remission, not significant on five of six secondary outcomes. A result that survives none of the robustness checks is a result to confirm, not a result already established. |
| Comparator asymmetry | Caveat FINDING: Delivered prepared meals and free supplies against £10 weekly: the two arms did not receive the same amount of material attention. The authors identify this as a limitation. A design that had provided prepared meals in both arms would have isolated the diet’s composition, which this design does not do. |
| Mechanism | Caveat FINDING: If ketosis is the active ingredient, a dose-response relationship would be expected. None is found, and the observed direction is the reverse. The analysis is post hoc and the urinary measure is less precise than a plasma measure, as the authors themselves note. It nonetheless weakens the biological plausibility of the primary result. |
| External validity | Major weakness FINDING: Remote recruitment across the whole of the United Kingdom, a population selected through numerous exclusion criteria, a support apparatus that is not accessible in routine outpatient practice. The authors themselves note that remote recruitment raises questions about representativeness relative to routine practice. And the twelve-week adherence figure, one in five, shows what becomes of this diet the moment the infrastructure disappears. |
| Duration | Caveat FINDING: Twelve weeks in total. No data on long-term metabolic tolerability, none on the durability of the effect, none on weight regain after stopping. |
Three registers. What is demonstrated: in this population, with this comparator, and over six weeks, an adjunctive ketogenic diet produces a mean gap of 2.18 PHQ-9 points, below the prespecified clinical relevance threshold and not recovered in the robustness analyses; adherence collapses as soon as support stops; no serious adverse event tied to the procedures occurred at six weeks. What is suggested: that the effect, if it exists, might be concentrated in the most severe presentations, and that it does not appear to run through ketosis. Both of these propositions come from, respectively, a subgroup and a post hoc analysis, in a trial not powered for either. What belongs to opinion: how to interpret the 8.3-point fall in the control arm, and whether a properly blinded trial is even conceivable for a dietary intervention this visible.
Level of evidence
PEB’s assessment: low confidence in the existence of an antidepressant effect specific to the ketogenic diet, moderate confidence that this effect, if it exists, falls below the clinical relevance threshold the trial had set for itself, high confidence in the collapse of adherence outside a dedicated support apparatus. The trial is well conducted and its conclusion is honest. What limits confidence has nothing to do with how it was run, and everything to do with what a design of this kind can establish: a self-reported outcome in unblinded participants, a comparator unevenly resourced in material terms, a result sitting at the exact edge of the conventional threshold that disappears the moment the analysis changes. The authors themselves conclude that adherence required intensive support, that few patients chose to continue the diet once that support was withdrawn, and that a more complete intervention needs to be developed before any further clinical trial. The principal value of this work is not its result, it is that it sets a methodological standard where only open-label series existed before.
The colleague test
What an experienced colleague would say if shown this study in two minutes, between two consultations.
“They went looking for 5 points, they found 2, and the interval excludes 5. By twelve weeks there’s nothing left, and four patients out of five have stopped the diet. I’m not putting a depressed patient on 30 grams of carbohydrate a day for that. That said, I do notice the control arm also dropped 8 points, and the trial can’t tell us how much of that fall is time, expectancy, or being followed up weekly.”
Translation for practice: always ask what relevance threshold a trial set for itself before it started. A significant result that misses its own threshold is a negative result in disguise, and that habit of reading will serve you well far beyond nutrition.
What you can do with this
- Nothing changes in prescribing, and PEB issues no practicable dietary advice here. The conversation, however, does arrive in consultation, usually with a printed news article in hand.
- Answer with the exact numbers rather than a position. The study circulating most widely in the general public claimed 69% improvement: it involved 16 analyzed participants, with no control group and no randomization. The first controlled trial gives a 2.18-point PHQ-9 gap, against a target of 5, and nothing left at twelve weeks. It is about as clean a demonstration as one can offer of what a control group changes.
- Recognize the situations where the question does not even arise. The trial itself excluded, among others, any history of an eating disorder, a BMI below 18.5, pregnancy, plans for pregnancy and breastfeeding, epilepsy, bipolar disorder, schizophrenia and psychotic disorders, substance or alcohol dependence, suicidal ideation with intent or an attempt within the past two months, psychiatric hospitalization or electroconvulsive therapy within the past year, gallstones, renal tubular acidosis, urinary calculi, small-bowel malabsorption, a history of pancreatitis, severe food allergies, unstable medical conditions, a current vegetarian or vegan diet, and patients treated with insulin, sulfonylureas, GLP-1 agonists, or SGLT2 inhibitors. This list is not a recommendation, it is the safety perimeter the investigators set for themselves, and it is public in the registry.
- Know the documented contraindications. A narrative review published in Annals of Medicine in 2026 lists as absolute contraindications to the ketogenic diet: pyruvate carboxylase deficiency, primary carnitine deficiency, CPT1 and CPT2 deficiencies, carnitine-acylcarnitine translocase deficiency, medium-, long-, and very-long-chain fatty acid oxidation defects, and porphyria. As relative contraindications: acute pancreatitis, advanced hepatic or renal impairment, familial hypercholesterolemia, propofol use. As situations calling for caution: type 1 or type 2 diabetes on glucose-lowering treatment, treated hypertension, gallbladder disease, electrolyte disturbances, arrhythmias, pregnancy and breastfeeding, underweight status, intense physical activity, the postoperative period. The same source lists SGLT2 inhibitors among the clinically relevant drug interactions of the ketogenic diet, alongside insulin, sulfonylureas, metformin, GLP-1 agonists, antiepileptics, diuretics, and corticosteroids. Eating disorder is not listed among these interactions, which is not a clearance: the trial, for its part, excluded it.
- Document, if a patient has already started on their own. Many will have, without mentioning it. The useful conversation covers what can be measured: weight and its trajectory, blood pressure, a lipid panel, renal function, fluid intake, and above all a review of current treatment to flag a gliflozin, a sulfonylurea, or insulin. This list is not a validated monitoring protocol, none exists in this indication. Remind patients that an unsupervised restrictive diet, without dietetic oversight, has not been evaluated.
- State the adherence trajectory up front. It is the most honest figure to share. In a trial that supplied three meals a day and offered seven coaching sessions, 64% of participants maintained urinary ketones of at least 1.5 mmol/L on at least 60% of readings at six weeks, and 20% were still following the diet more than half of days six weeks after support stopped. Sharing this curve in advance protects the patient from experiencing a lapse as personal failure.
- Hold the control-arm observation as a lead, not as evidence. Seven structured appointments, a minimal dietary instruction, and one concrete daily target were accompanied by an 8.3-point PHQ-9 fall in treatment-resistant patients. The trial cannot attribute this fall to the support apparatus, for lack of a true no-intervention arm. It nonetheless recalls what a regular framework and an achievable daily task can represent, in a population whose follow-up often becomes sparse.
Can it be said that the ketogenic diet works in treatment-resistant depression?
No, and the authors do not say so. Their own wording is that an antidepressant benefit is observed at six weeks but that “the clinical relevance is uncertain,” the mean effect being modest and absent from the secondary analyses. The observed difference falls below the 5-point threshold the trial had itself set as clinically relevant, and the confidence interval excludes that threshold.
Why place so much weight on the absence of blinding for a dietary question?
Because the primary outcome is a scale the patient fills in themselves. When the person rating their own mood knows they belong to the experimental arm of a trial testing a diet they have heard about, the effect of the diet cannot be separated from the effect of expectancy. This is a structural limitation, not a flaw in conduct: what someone eats cannot be masked. It simply means a 2.18-point gap has to be treated as a signal to confirm, not as a settled finding.
Does the result in severe presentations justify targeting those patients?
Not yet. The subgroup analysis is stated as prespecified, and severity was one of the minimization factors, which makes it more credible than an improvised analysis. But the trial is not sized for subgroups, the interaction is one of several tested, and a larger improvement in the more severe cases is also what regression to the mean produces. It is a hypothesis for the next trial, not a criterion for selecting patients.
What does the absence of a link between ketones and improvement mean?
Ketosis is supposed to be the mechanism of action. The post hoc analyses find no significant association between ketone concentration and PHQ-9 change, and the participants least in ketosis were those who improved the most. These analyses are exploratory and rely on a urinary measure of ketones, less precise than a plasma assay. They do not refute the mechanistic hypothesis, but they do not support it either, which weakens the biological plausibility of the primary result.
Is the ketogenic diet dangerous in a depressed patient?
This trial reports no serious adverse event related to the research procedures over six weeks, with an explicit target of weight stability and weekly dietetic oversight. It says nothing beyond six weeks, and the authors themselves flag this, citing hyperlipidemia, kidney stones, and vitamin deficiencies as effects a duration this short cannot detect. The trial also excluded numerous profiles: a history of eating disorder, low BMI, pregnancy, epilepsy, bipolar disorder, psychosis, renal impairment, gallstones, pancreatitis, and several antidiabetic treatments. An intervention tested under this much protection does not transfer unchanged to an unselected patient.
What should be said to a patient who read that a ketogenic diet cuts depression by 70%?
That the figure is real, that it comes from a single-arm study of 16 analyzed participants with no comparison group, and that the question it poses is not the one it appears to answer. When the same intervention is tested against a comparator, the gap falls to 2.18 PHQ-9 points and disappears by twelve weeks. The difference between these two figures has nothing to do with the diet, it comes down to the study design.
Annotated bibliography
Gao M, Kirk M, Knight H, Lash E, Michalopoulou M, Guess N, Stevens R, Browning M, Weich S, Burnet PWJ, Jebb SA, Aveyard P (2026). A Ketogenic Diet for Treatment-Resistant Depression: A Randomized Clinical Trial. JAMA Psychiatry, 83(4), 331-340. DOI 10.1001/jamapsychiatry.2025.4431. PMID 41637092. PMC12874075. Source study analyzed here, full text consulted in open access under a CC BY license. Funded by the NIHR Oxford Health Biomedical Research Centre, non-industry. Dr Browning reports personal fees from Alto Neuroscience, Engrail Neuroscience, Empyrean Neuroscience, and Janssen; Drs Kirk and Stevens report NIHR funding; the statement “no other disclosures were reported” appears in the article. No commercial supplier of the prepared meals is identified in the publication.
ClinicalTrials.gov, NCT06091163. Ketogenic Diet for Treatment-Resistant Depression: Dietary Interventions for MEntal Health Study (DIME). clinicaltrials.gov. Sponsor: University of Oxford, 88 participants. Source of the detailed inclusion and exclusion criteria cited in this article. Registry entry consulted on 11 August 2026.
Decker DD, Patel R, Cheavens J, Hayes SM, Whitted W, Lee AJ, Buga A, Robinson BT, Crabtree CD, Kackley ML, Stoner JT, Sapper TN, Chebbi A, Volek JS (2025). A pilot study examining a ketogenic diet as an adjunct therapy in college students with major depressive disorder. Translational Psychiatry, 15, 322. DOI 10.1038/s41398-025-03544-8. PMID 40925905. A prospective, single-arm study with no randomization, no control group, and no blinding: 24 participants enrolled, 16 analyzed, ten to twelve weeks, a 69% reduction in PHQ-9 and a 71% reduction on the HRSD, with a 6.2% loss of body mass. Cited here as a methodological point of comparison illustrating the contrast between an uncontrolled design and a randomized trial, not as evidence of efficacy.
Dyńka D, Rodzeń Ł, Rodzeń M, Łojko D, Karakuła-Juchnowicz H, Ede G, Grzywacz Ż, Antosik K, Sethi S, Unwin D (2026). The ketogenic diet is not for everyone: contraindications, side effects, and drug interactions. Annals of Medicine, 58(1), 2603016. DOI 10.1080/07853890.2025.2603016. PMID 41486865. Source of the absolute and relative contraindications and the drug interactions cited in the practice box. A narrative review, not a graded systematic review. Several of its authors are engaged in the clinical promotion of metabolic therapies, which warrants caution regarding passages favorable to the diet, but this does not weaken the list of contraindications itself.
Correspondence. Two letters titled A Closer Look at Ketogenic Diet Effects in Depression were published in JAMA Psychiatry, 2026, 83(7), 773: one by Stubberud J (DOI 10.1001/jamapsychiatry.2026.0944), the other by Richardson K, Rocks T, and O’Neil A (DOI 10.1001/jamapsychiatry.2026.0947). References verified on Crossref; the content of these letters was not consulted and is therefore not reproduced here. They are flagged to let readers know that the interpretation of this trial has been the subject of published debate.
Editorial collections
Tags
Verified on August 11, 2026 against the full text of the publication and its supplementary material where available. This analysis underwent an independent double reading. The English version was checked for conformity on September 18, 2026, against the figures of the French version and against the source. How we verify what we publish
This analysis is intended for healthcare professionals. It does not constitute a prescribing recommendation and does not replace individual clinical judgment.
Analysis from Psychiatry Evidence Base, evidence-based psychiatry, explained with rigor.