Diets
ketogenic diet decreases seizure frequency in drug-resistant epilepsy
In plain terms: Does the ketogenic diet cut seizure frequency in drug-resistant epilepsy?
Strong support Diets 💰 Industry COI noted🔬 Includes disconfirming
Part of: 🥗 ketogenic diet
RefutedContestedStrong support
consensus score 0.79
Yes — this is its original, best-evidenced medical use: dietary therapy meaningfully reduces seizures, most robustly in children, with adult/RCT data thinner.
📅 Last reviewed: 2026-07-15 ⓘ
Evidence ladder
How far up the ladder this claim has climbed. A high consensus on a low rung means "consistent so far," not "proven in people."
Top evidence so far: All trials, pooled (Meta-analysis)
MechanismIn-vitroAnimalObservationalRCTMeta-analysis
How the studies fall
13 support 1 contradict 0 tested null 2 mixed · 16 sources, 14 independent groups
The evidence (20)
| Source | Grade | Stance | Quality | Finding |
|---|---|---|---|---|
| Kverneland 2018 · Epilepsia | RCT | contradicts | moderate | Adults >16y with drug-resistant focal epilepsy, 12wk adjunctive modified Atkins diet vs habitual diet; 37 assigned to diet, 38 to control. Primary endpoint null: among the 24 diet completers median seizure change -1.0 (IQR -13.7-8.8) vs +4.5 (IQR -4.8-33.5) in 32 controls, median difference -7.0 (95% CI -37.0 to 3.0; P=.21). ITT (37/38) RR for >50% seizure reduction 1.8 (95% CI 0.3 to 10.2; P=.65) and for >25% reduction 2.43 (95% CI 0.94 to 6.28; P=.06) - neither significant. The authors' claim of 'a significant reduction... but only for moderate benefit (>25% seizure reduction) among those who completed the intervention' rests on a completer-only analysis at the relaxed threshold, with attrition concentrated in the diet arm (9/37 excluded vs 4/38); no figure for it appears in the abstract and the paper is not open access. |
| Devi N, Madaan P, Kandoth N, Bansal D, Sahu JK 2023 · JAMA Pediatrics | meta-analysis | supports | moderate | Systematic review + frequentist network meta-analysis of RCTs of dietary therapies in CHILDHOOD drug-resistant epilepsy: 'Of 2158 citations, 12 randomized clinical trials (907 patients) qualified for inclusion.' The claim's subject gets its own arm-specific estimate within that 12-trial / 907-patient pool rather than being folded into a combined dietary exposure: 'In the short term, all dietary interventions were more efficacious than care as usual for 50% or higher seizure reduction (low glycemic index therapy: odds ratio [OR], 24.7 [95% CI, 5.3-115.4]; modified Atkins diet: OR, 11.3 [95% CI, 5.1-25.1]; ketogenic diet: OR, 8.6 [95% CI, 3.7-20.0]), while ketogenic diet (OR, 6.5 [95% CI, 2.3-18.0]) and modified Atkins diet (OR, 5.1 [95% CI, 2.2-12.0]) were better than care as usual for seizure reduction of 90% or higher.' ENDPOINT CORRECTION: the prior extract reported 'seizure-freedom odds'; this paper measured no such outcome. Its primaries were 'Short-term (<=3 months) 50% or higher and 90% or higher reduction in seizure frequency and treatment withdrawal due to adverse events' - a >=90% reduction is not seizure freedom, and 'classical KD and variants' concealed that KD, modified Atkins and LGIT are separately estimated arms of one network, not a single exposure. QUALITY high -> moderate on what the paper establishes about its own reliability: the efficacy window is short-term only ('<=3 months'), the network estimates are extremely imprecise (LGIT 'OR, 24.7 [95% CI, 5.3-115.4]'), the between-diet contrasts are indirect and null ('Indirectly, there was no significant difference between dietary therapies in efficacy and safety outcomes'), and the authors decline to call their own finding confirmed: 'Direct head-to-head comparison studies are needed to confirm these findings.' Heterogeneity, network inconsistency and publication-bias testing are not reported in the abstract and PMC9887534 is not in the open-access subset, so those are cannot-tell rather than absent. Safety, recorded here because it rides on the same claim's intervention: 'adverse event-related discontinuation rates were significantly higher for ketogenic diet (OR, 8.6 [95% CI, 1.8-40.6])'. Stance unchanged - the direction on the claim's object is large, prespecified and significant for the ketogenic diet specifically. |
| Hahn W et al 2026 · study_type: RCT | RCT | mixed | moderate | Crossover pilot, n=22 completers; 9% seizure-free on IF-MCT; >=50% responder rate numerically higher vs IF-alone but not significant (Breslow p=.233) |
| Manral, Dwivedi, Gulati, Kaur, Nehra, Pandey, Upadhyay, Sapra, Tripathi 2023 · Neurology | RCT | supports | moderate | Open-label RCT at a tertiary referral centre in New Delhi (CTRI/2015/07/006048), enrolling Aug 2015-Apr 2019: 'Persons with DRE aged 10-55 years attending outpatient epilepsy clinics between August 2015 and April 2019, who had more than 2 seizures per month despite using at least 3 appropriate antiseizure medications (ASMs) at their maximum tolerated doses' were randomised to standard drug therapy (SDT) plus modified Atkins diet or SDT alone; 'Intention-to-treat analysis was performed.' Primary endpoint, >50% seizure reduction at 6 months: 'At 6 months, >50% seizure reduction was seen in 26.2% in the intervention group vs 2.5% in the control group (95% CI 13.5-33.9; p < 0.001).' DENOMINATORS: '160 patients (80 adults and 80 adolescents) were randomized' - the 80/80 is the AGE split, not the arm split, and per-arm n is not stated in the abstract (26.2% is consistent with ~21/80 but that is inference, not text). The behaviour secondary outcome rests on a far smaller subset of the 160: 'behavior scores could be performed in 49 patients'. QOL: 'Improvement in QOL was 52.1 +/- 17.6 in the intervention group vs 42.5 +/- 16.4 in the control group (mean difference, 9.6; 95% CI 4.3 to 14.9, p < 0.001).' QUALITY high -> moderate on the authors' OWN reliability rating: 'CLASSIFICATION OF EVIDENCE: This study provides Class III evidence that the MAD increases the probability of seizure reduction in adolescents and adults with DRE' - an AAN Class III rating is the authors declining to call their own randomised trial Class I or II evidence, and the arm structure (SDT plus MAD vs SDT alone, no sham or attention control) precludes participant blinding on a patient-counted seizure endpoint; whether outcome assessors were masked is cannot-tell from the abstract. Every other RCT row on this claim already sits at moderate, so high was the outlier. Stance unchanged: the effect on the claim's object is prespecified, large and significant. |
| Vaccarezza MM et al 2026 · study_type: meta-analysis | meta-analysis | supports | moderate | SR/MA of 17 RCTs: peds 37% achieve >=50% seizure reduction (moderate-certainty); adults/adolescents +16 more per 100 vs usual care (moderate-certainty) |
| Zhu K et al 2025 · Epilepsy Behav | meta-analysis | supports | moderate | Systematic review + Bayesian random-effects NETWORK meta-analysis of RCTs of carbohydrate-restricted diets in drug-resistant epilepsy, PRISMA-NMA, literature search to 22 December 2023: 'A total of 17 RCTs involving 1468 DRE patients were included. The diets evaluated were the ketogenic diet (KD), modified Atkins diet (MAD), and low glycemic index treatment (LGIT).' The claim's subject is a SEPARATELY estimated network arm, not folded into a combined dietary exposure: 'For >50 % and >90 % seizure reduction from baseline, all three diets resulted in significant efficacy compared to the normal diet.' PRIOR EXTRACT CORRECTION: the previous text ('efficacy well-established in pediatric populations, adult data underexplored') was the paper's BACKGROUND rationale, not a result it measured - 'However, while the efficacy of these diets has been well-established in pediatric populations, their effectiveness on adult DRE patients remains underexplored' is the authors' reason for running the study, and the stored extract carried none of their findings, no denominator and no interval. DENOMINATORS: the 1468 spans all three diet arms plus normal-diet comparators across ALL age groups; no KD-specific trial count, arm n or effect estimate appears in the abstract, so no odds ratio here may be read as a figure for the ketogenic diet alone. The only per-diet estimates reported are MAD's, and both are OFF this claim's object: 'Notably, MAD was the only diet that demonstrated a statistically significant association with seizure freedom (OR 7.36, 95 % CrI 2.21-60.36), compared to the normal diet, while its lower compliance (OR 0.39, 95 % CrI 0.18-0.76) was likely influenced by the inclusion of adult subjects.' The between-diet contrast is a null on a DIFFERENT question and does not bear against the claim: 'This meta-analysis indicated that in RCTs, the included diet therapies were overall equivalent in efficacy and side effects, with the MAD showing a higher chance of seizure freedom.' QUALITY moderate stands on what the paper establishes about its own reliability: the credible intervals it does report are extremely imprecise (2.21-60.36), and heterogeneity, network inconsistency, risk-of-bias across the 17 included RCTs and publication-bias testing are cannot-tell from the abstract (Epilepsy Behav is Elsevier and the paper is absent from the PMC OA subset), not absent. Declared COI: 'The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.' INDEPENDENCE FLAG (not acted on): this is one of roughly ten syntheses counting on this claim over a randomised literature of about 17 trials, and s42391224 carries the same '17 RCTs' headline. STANCE, GRADE, QUALITY AND ROLE ALL UNCHANGED - the direction on the claim's object is significant for the ketogenic diet specifically; only the extract was wrong. |
| Farrar 2026 · Seizure | meta-analysis | supports | high | Systematic review/meta-analysis of 33 studies (1019 adolescents/adults with DRE) found 42.8% achieved over 50% seizure reduction on ketogenic diet therapy. |
| Yavuz M et al 2026 · study_type: observational | observational | supports | low | Retrospective n=36 infants <2yo: ITT responder rates 52.8% (1mo) declining to 36.1% at 24mo; one-third seizure-free at 24mo among continuers |
| Chavez-Castillo M et al 2026 · Epilepsy Behav Rep | observational | supports | low | Retrospective single-centre uncontrolled case series, Ketogenic Diet Clinic at Hospital Civil de Guadalajara (Mexico), pediatric drug-resistant epilepsy initiating KDT July 2022-July 2024; 21 identified, 1 excluded for missing baseline data, n=20 analysed, median 3 ASMs at baseline, median 72 weeks on diet, 85% with daily seizures. EXPOSURE IS NOT THE CLASSIC KETOGENIC DIET AND NO PER-REGIMEN OUTCOME IS REPORTED: 'Nine patients followed either a modified 1:1 ratio diet (45 %) or an LGIT (45 %), and only 2 patients initiated with a pre-KDT (10 %). None of the patients included in this study followed a classic ketogenic diet.' Every figure below is a pooled KD-variant estimate and is read on this claim the same way the pooled-exposure syntheses s38315271 and s40252525 are. WHOLE-COHORT RESULT: 'At the first follow-up visit where complete data were available for all 20 patients, 10/20 (50 %) achieved >=50 % seizure reduction, and 4/20 (20 %) achieved >=90 % reduction.' SUBSET DENOMINATOR, AND THE AUTHORS DISOWN IT AS AN EFFICACY ESTIMATE: 'Among the 14 patients who had reached Visit 3 by the study cutoff, 9/14 (64 %) achieved >=90 % seizure reduction' - the 14 are those recruited early enough to have accumulated a third visit, not survivors ('No patient discontinued dietary therapy during the study period') - and the paper instructs that 'this proportion reflects only those who had accumulated sufficient follow-up time and not the entire cohort' and that 'The higher responder rate observed at later visits should therefore be interpreted cautiously and within the context of program feasibility rather than as a definitive estimate of dietary efficacy', pointing readers to the 20-patient first-visit figure instead. THE ATTRIBUTION CAVEAT IS PART OF THE FINDING and the prior extract carried none of it: patients entered through a pediatric epileptologist who performed 'systematic etiologic reassessment, refinement of seizure classification, and active optimization of antiseizure medication regimens' at the same time the diet began, so 'Importantly, this was not a controlled trial, and enrollment in the ketogenic diet clinic involved comprehensive clinical optimization' and 'seizure improvement cannot be attributed exclusively to dietary therapy.' QUALITY LOW STANDS on what the paper establishes about itself: 'The relatively small sample size and the lack of a control group limit the generalizability of the findings', single centre, retrospective chart-based seizure counts with no blinded assessment and no pre-KDT metabolic screening ('universal testing was not feasible in our setting due to resource constraints'). Treatment burden, off this claim's object and filed as a candidate: median cumulative ASM dose change -25% (IQR -43.8% to 0), 12/20 (60%) with a dose reduction, 6/20 (30%) with a count reduction, p=0.018 Friedman across visits. COI: first author 'reports a relationship with Stendhal Pharma that includes: speaking and lecture fees', and 'Publication fees, if applicable, will be covered by Stendhal Pharma.' STANCE, GRADE, QUALITY AND ROLE ALL UNCHANGED - the direction on the claim's object is positive at both thresholds in the full 20; only the extract was thin. |
| Sharaf MA et al 2026 · Seizure | meta-analysis | supports | low | Pairwise meta-analysis of 8 studies comparing TWO ketogenic therapies head-to-head in CHILDREN with drug-resistant epilepsy - classical ketogenic diet (CKD) vs modified Atkins diet (MAD) - not keto vs no keto: 'Eight studies were included in this meta-analysis.' Every pooled estimate in the paper carries a ketogenic comparator: 'CKD demonstrated a statistically significant advantage over MAD for achieving >50% seizure reduction (OR = 0.55, 95% CI: 0.35-0.87; P = 0.01).' 'However, no statistically significant differences were observed between CKD and MAD for complete seizure freedom (100% reduction; OR = 0.72, 95% CI: 0.43-1.19) or >90% seizure reduction (OR = 0.87, 95% CI: 0.51-1.47).' OFF-SCOPE on the claim's exposure contrast: the claim asks whether the ketogenic diet decreases seizure frequency in drug-resistant epilepsy; this paper measures which ketogenic therapy decreases it MORE. The identical row would have to be recorded as CONTRADICTING a claim that the modified Atkins diet - itself a ketogenic diet therapy - decreases seizure frequency, on the same three numbers; a row that votes opposite ways depending on which ketogenic arm is called the subject is measuring relative, not absolute, efficacy. The single on-object sentence, 'Both MAD and CKD were effective in treating drug-resistant epilepsy', carries no estimate, no non-ketogenic comparator and no participant count; whether any single-arm response rate was pooled is cannot-tell from available text. SIBLING CONSISTENCY: s38908324 is also a two-ketogenic-arm study and is retained on this claim, but only because it reports an absolute within-subject result (71.1% mean seizure reduction, n=52); this paper reports no absolute efficacy figure at all, which is what separates the two rulings. DENOMINATORS: no total participant number, no per-outcome study counts, no I2 and no publication-bias test appear in the abstract, so none of the three ORs may be quoted with a study base; the prior extract's '8-study MA' was the only base available and the population (children) was missing from it. ORIENTATION FLAG: an OR of 0.55 read as a CKD 'advantage' implies the ratio is parameterised MAD-relative-to-CKD; which arm is the numerator is cannot-tell from the abstract, and the direction recorded here is the authors' stated one. On reread #686's note that Devi 2023's network found the MAD-vs-KD comparison null 'against sibling s41616756': this paper's contrast is a DIRECT pairwise pooling of head-to-head studies, not a network indirect estimate, and it is itself null at the two deeper thresholds - the disagreement is confined to the >=50% threshold. QUALITY moderate -> low on what the paper establishes about its own reliability: randomised and non-randomised designs are pooled into single odds ratios ('A meta-analysis included experimental and observational studies retrieved from relevant databases') and the measurement window is not fixed ('A 3-month follow-up was used to measure all outcomes, whenever available; otherwise, the nearest reported time point was used'). Abstract-only (Seizure 2026;135:77-87, Elsevier; no PMCID, not in the PMC OA subset; text via NCBI efetch). Declared COI: 'The authors declare that they have no competing interests.' |
| Meng X, Zhou K 2026 · Neurol Int | observational | supports | low | Narrative systematic review, no pooling ('A formal meta-analysis was not performed'); 8 primary studies, n=675, of NON-standard diets — standard KD, modified Atkins, carbohydrate-modified KD and isolated fish oil were excluded by design. KD-bearing evidence only, with its own denominators: Guzel 2019, single-arm olive-oil-based KD in 389 children, '83.1% responder rate at 12 months; 43.1% seizure-free', but 'the lack of a control group and a 25.7% dropout rate limit the practical conclusion' (the 12-month completer denominator is under 389 and is not reported here); Neal 2009 RCT (n=145) found 'MCT-based KD was not superior to classical vegetable oil-based KD, though both diets significantly reduced seizure in children with DRE' — a KD-vs-KD randomisation, so it yields no randomised KD-vs-no-KD comparison and is graded observational for this claim. The p=0.57 null is Sarlo 2023's low-glutamate diet (n=33), a different intervention, and does not bear on this claim's object. Authors on their own body: 'Evidence quality is generally low, with 75% of studies at high risk of bias', and they warn the placebo effect 'can account for up to a 20–30% reduction in seizure frequency'. Zero weight: review whose KD primaries are already counted inside the meta-analyses voting on this claim. |
| Mustafa MS, Shafique MA, Aheed B, Ashraf F, Ali SMS, Iqbal MF, Haseeb A 2024 · Ir J Med Sci | meta-analysis | supports | moderate | Systematic review + pairwise meta-analysis of RCTs of dietary therapy in PEDIATRIC drug-resistant epilepsy: 'Eleven RCTs with 788 participants were included in this study.' EXPOSURE IS POOLED, and the prior extract's 'ketogenic interventions' concealed it: every estimate below is for classical KD together with its variants, which the paper introduces as 'In addition to the traditional ketogenic diet, several variants have been introduced to enhance compliance and flexibility, such as the modified Atkins diet (MAD) and the low glycemic index diet (LGID).' No KD-specific estimate appears in the abstract, so none of these ORs may be read as a figure for the classical ketogenic diet alone. Results, all attributed to 'dietary interventions': 'The pooled effect estimates revealed a significant association between dietary interventions and seizure frequency reduction of > 50% (OR 6.68, 96% CI 3.52, 12.67) and > 90% (OR 4.37, 95% CI 2.04, 9.37).' and 'Dietary interventions also increased the odds of achieving seizure freedom (OR 4.13, 95% CI 1.61, 10.60).' The prior extract carried no intervals, dropped the >90% result, and wrote '788 children' where the paper says '788 participants'. The '96% CI' is the published text, not a transcription slip. QUALITY high -> moderate on what the paper establishes about its OWN reliability: it reports having done the work - 'Our approach involved predefined criteria for data extraction and the assessment of study quality' - but never states the result of that assessment, the intervals are very wide (seizure freedom 1.61 to 10.60), and the authors decline to call the finding settled: 'Further research is needed to address the limitations and investigate their long-term effects.' Heterogeneity, publication-bias testing, comparator definition and funding are cannot-tell from the abstract (Ir J Med Sci is not open access and the paper is absent from the PMC OA subset), not absent. High was also the outlier: the Cochrane review of the same overlapping RCT literature s32588435 rates its own evidence low-to-very-low certainty and sits at moderate. Safety, on the same intervention: 'The common adverse effects included constipation (39.07%) and vomiting (10%).' STANCE UNCHANGED - direction on the claim's object is large, significant and consistent across all three thresholds. |
| Dressler A, et al. 2025 · Nutrients | observational | supports | moderate | Prospective single-centre cohort, Medical University of Vienna (ethics 542/2007, enrolled June 2008-April 2017), reporting the prespecified observational follow-up of the authors' own parallel-cohort RCT of first-line KDT vs ACTH: 'Of 101 infants allocated to the initial treatment phase, N = 67 required further treatment.' ENDPOINT: the ONLY outcome measured was 'sustained seizure freedom (i.e., from all seizure types, including epileptic spasms) until last follow-up' - no >=50% or >=90% seizure-reduction responder rate was assessed, so this row bears on the claim's object only at its zero-frequency extreme. HEADLINE FIGURE WITH ITS OWN DENOMINATOR: 'KDT as the second treatment achieved sustained seizure freedom in 50% (12/24), compared to 19% (3/16) with ACTH and 9% (2/22) with ASMs' - the 24 is neither the 101 enrolled nor the 67 needing a second treatment, but the subset of the 35 initial-ACTH failures who were switched to KDT ('Of the 35 infants failing first treatment with ACTH, 24 were treated with a second KDT'), and the 16 ACTH-second infants are the mirror subset of the 32 initial-KDT failures. The prior extract ('showed clinically useful seizure control') was neither the paper's endpoint nor a number. CONFOUNDING BY INDICATION IS STRUCTURAL, not incidental: second KDT was given only to ACTH-failures and second ACTH only to KDT-failures, so first and second treatment are collinear by design and no adjustment can separate them; the KDT arm also had 'a longer duration of epilepsy before trial start and a more frequent use of concomitant VGB'. Adjustment covers gender, etiology, age and prior ASM count only. GRADE observational is the paper's own word: 'As the study was observational in the second treatment phase, no randomization was performed', and 'Randomization at this time point would not have been clinically and ethically appropriate, which is a limitation to the comparison of treatments.' QUALITY moderate stands on what the paper establishes about itself - prospective protocol-driven allocation with 'Two board-certified epileptologists independently assessed Video-EEGs and were blinded to treatment allocation and outcome', against a single centre, arms of 24 and 16, and a 'until last follow-up' bar of unequal length per infant ('96% of infants were still in follow-up after 6 months, 81% after 12 months, 65% after 24 months, and 39% after 48 months'). The VGB subgroup - 'sustained seizure freedom in response to second treatment was more frequent for KDT (56%) compared to ACTH (ACTH 0%; risk difference 53% (95%CI 4-73%), p = 0.047)' - lives in Supplemental Table S2, which this read could not obtain, so its denominators are cannot-tell. The result that does NOT favour keto is off this claim's object and is filed as a candidate rather than read as mixed: cumulative seizure freedom by initial allocation was near-identical, 'In all 53 infants who were initially allocated to KDT, the cumulative response after all treatment phases was 60% (32/53)' vs 63% (30/48) for ACTH - a keto-vs-ACTH question, not a keto-vs-worse-control one. FUNDING: 'This research received no external funding'; COI: 'AD has received travel reimbursement and speaker honoraria from Nutricia, UCB, Kanso, Eisai, Takeda, and Vitaflo', ketogenic-formula manufacturers among them, declared 'None were related to this article'. INDEPENDENCE FLAG (not acted on): these 101 infants are the cohort of the authors' parent PC-RCT (reference 29), which is not itself a vault source but may already vote inside the meta-analyses counting on this claim. STANCE, GRADE, QUALITY AND ROLE ALL UNCHANGED - only the extract was wrong. |
| Liu J, E R-H-M-Q-Q-G, Zhang J-Q, Wang D-P 2025 · Front Nutr | meta-analysis | supports | moderate | Systematic review + pairwise meta-analysis of randomised/quasi-randomised trials in CHILDREN (<=18y) with drug-resistant epilepsy, PROSPERO CRD420251122427, search to 16 March 2025: 'Ultimately, 9 studies met all stringent criteria and were included in the final meta-analysis'. ENDPOINT CORRECTION: the prior extract said the diet 'significantly reduced seizure frequency', but no continuous seizure-frequency outcome was pooled - all three endpoints are the ODDS OF CROSSING A RESPONDER THRESHOLD. >=50% reduction, 9 trials, random-effects, I2 = 60.5%: 'Pooled results revealed a significant association between KD intervention and a greater likelihood of reducing seizure frequency by 50% or more (OR = 7.69, 95% CI [3.42, 17.3])'. >90% reduction, 9 trials, fixed-effects, I2 = 0.0%: 'The combined results demonstrated a robust association between KD intervention and the likelihood of attaining a seizure reduction exceeding 90% (OR = 8.54, 95% CI [3.13, 23.31])'. THE EXPOSURE IS POOLED across KD variants, but unlike several siblings this paper gives the claim's own subject a separate estimate: 'Trials of the classical KD (n = 5) demonstrated an OR of 8.35 (95% CI [4.12, 17.0]; I2 = 36.1%), whereas those of MAD or LGIT (n = 4) yielded an OR of 6.12 (95% CI [2.78, 13.45]; I2 = 21.7%)'. SUBSET DENOMINATOR: the seizure-freedom OR 7.35 (95% CI [2.17, 24.88]) is not the 9 - 'Four studies that reported the proportion of patients achieving complete seizure freedom were included in this meta-analysis.' NO pooled participant total is stated anywhere in the paper, and Table 1's Total column contradicts its own arm counts in three of nine rows (de Kinderen 2016: 32 vs 26 + 26; Dressler: 101 vs 16 + 16; Sharma 2021: 91 vs 46 + 46), so no n may be read off it; whether those errors reached the pooled estimates is cannot-tell, since the event counts live in the forest-plot images. The authors' own conditions are part of the finding: 'while this relative effect size is substantial, the absolute rates of seizure freedom in the included trials were modest', the trials 'enrolled highly selected pediatric populations', and the publication-bias test was run but underpowered - 'Egger's test indicated no significant publication bias (p > 0.05), although the test had limited power due to the small number of included studies'. QUALITY moderate stands on what the paper establishes about itself: RoB 2.0, prespecified subgroups, leave-one-out sensitivity and a reported Egger test on one side; 9 small trials (largest 145 participants), I2 = 60.5% on the primary outcome, intervals as wide as 2.17-24.88, and the table errors above on the other. Safety, on the same intervention: constipation 37.5% (95% CI 21.6-56.6%), anorexia 19.5%, diarrhea 18.9%. FUNDING: 'The author(s) declare that no financial support was received for the research and/or publication of this article'; COI: none declared. INDEPENDENCE FLAG (not acted on): its 9 primaries - Neal 2008, Sharma 2013/2016/2021, El-Rashidy 2013, de Kinderen 2016, Lambrechts 2017, Dressler, Lakshminarayanan 2021 - are the same pediatric RCT pool already counted on this claim through the Cochrane review s32588435 and at least five further syntheses. STANCE, GRADE, QUALITY AND ROLE ALL UNCHANGED - only the extract was wrong. |
| Matairi AA et al 2025 · World J Pediatr | meta-analysis | mixed | low | Systematic review + meta-analysis of RCTs AND prospective cohort studies of dietary therapies in drug-resistant epilepsy (search to December 2024, updated May 2025), pooled in R with odds ratios. POOLED-EXPOSURE WARNING: the headline 'Dietary interventions significantly increased ≥ 50% seizure reduction [odds ratio (OR) = 3.46, 95% CI = 1.83-6.56] compared with standard care' combines KD, modified Atkins and low-glycemic-index treatment and may NOT be read as a figure for the ketogenic diet; the age subgroup 'stronger effects in pediatric patients (OR = 10.93 vs. 2.54 in adults, P = 0.007)' is that same pooled exposure, not KD. The paper does estimate the claim's subject separately, and the KD-specific results point BOTH WAYS on this claim's own object. Null at the primary responder threshold: 'MAD outperformed standard care (OR = 4.04), whereas KD did not (OR = 1.83).' Positive at the deeper threshold: 'For a ≥ 90% reduction, KD had greater efficacy (OR = 6.23) than MAD did (OR = 1.98).' Complete cessation is null but pooled across diets rather than KD-specific: 'No significant difference was found for complete seizure resolution (OR = 1.19).' The prior extract ('found significant reductions in seizure frequency across dietary therapies') carried only the pooled-exposure headline, and so reported as support for the ketogenic diet a result the paper's own KD estimate did not reach. STANCE supports -> mixed: the paper genuinely found both directions on this claim's subject and object, which is what mixed is for. DENOMINATORS: no confidence intervals, trial counts or participant numbers accompany ANY of the per-diet ORs in the abstract, and no total n appears in it at all - cannot-tell from available text, so none of these ORs may be quoted with a study base. QUALITY moderate -> low on what the paper establishes about its own reliability: 'Heterogeneity was substantial in most analyses (I2 > 75%)'; randomised and non-randomised designs are pooled into single odds ratios ('Eligible studies were randomized controlled trials and prospective cohort studies'); and bias was judged with a randomised-trials-only instrument ('The risk of bias was assessed via the Cochrane Risk of Bias 2 tool'). The authors decline their own strongest reading: 'However, claims of seizure freedom remain uncertain.' Abstract-only (World J Pediatr, no PMCID, not in the PMC OA subset). Declared COI: 'The authors have no conflict of interest to declare.' |
| Ruan Y, Chen L, She D, Chung Y, Ge L, Han L 2022 · Eur J Clin Nutr | meta-analysis | supports | moderate | Overview of reviews (umbrella review) that pools nothing of its own: 'We aim to evaluate the quality of the systematic reviews and meta-analyses (SRMAs) of KDT for epilepsy and summarize the evidence on their effects.' 'Two investigators independently performed study selection to include SRMAs, extracted data and assessed the quality of SRMAs with the AMSTAR-2 and PRISMA statement.' 'Twenty-four SRMAs were selected which encompassed a total of 255 original studies.' The seizure result is narrative, not pooled - no effect estimate, no interval: 'Overall, positive effects of KDT for epilepsy on seizure frequency reduction, as well as cognition and behavior were observed.' Two corrections to the prior extract. (1) The word 'consistent' is not the paper's; the paper says 'Overall', and reserves its contrast for other endpoints - 'In contrast, the effects of KDT on quality of life, growth and development were more controversial.' (2) The 24 is a borrowed denominator: the abstract breaks the 24 down by population (4 infant, 13 children/adolescent, 8 adults-or-all) and by other outcomes (17 adverse effects, 10 attrition, 4 death, 4 cognition/behavior, 3 quality of life, 2 growth) but never states how many reported seizure frequency - cannot-tell from the abstract. Zero weight as evidence_role: review (rubric axis 2, standing umbrella-review rule): with a search running to 3 September 2020, its 24 SRMAs / 255 original studies necessarily re-describe syntheses this claim already counts directly, including the Cochrane review s32588435, so counting it alongside them would double-count the same primary trials. Direction on the claim's object is unchanged and correct; only the weight is. |
| Ray S, Nathan J, Godhia M 2024 · Epilepsy Res | observational | supports | moderate | Open-label randomized trial comparing TWO ketogenic diets, NOT keto vs no keto: '52 children, aged 2-10 years, were randomized, 25 in the CKD group and 27 in the PUFAKD group.' Both arms were ketogenic - 'Fat: carbohydrate + protein ratio of 2.2:1-4:1 was maintained in both diets' - so the randomised variable was fat composition, not the claim's exposure; graded observational, not RCT, on the exposure-randomised test, because the seizure result is an UNCONTROLLED within-subject pre-post over 12 months with no non-keto comparator. Finding: 'A significant decrease (p=0.001) in seizures was observed (n=52), with no significant difference (p=0.537) between the two groups' and 'The mean seizure reduction was 71.1 %, with no significant difference (p=0.488) in both groups.' DENOMINATOR: 71.1% is a mean over all 52 randomised children POOLED across both keto arms; per-arm figures are not in the abstract. The prior extract's '50-90% reduction' was a PRE-SPECIFIED RESPONSE BAND, not a result - 'The percentage reduction in seizures at <50 %, 50-90 %, >90 %, and 100 % was also measured' - and no responder count in any band appears in the abstract. A compliance-response trend is reported (p = 0.042, Z=4.039, n=52) but is post hoc and non-randomised. Stance unchanged: the direction on the claim's object is real and large; the design cannot separate it from regression to the mean or natural fluctuation. Abstract-only (Elsevier paywall). |
| Han R et al 2026 · study_type: observational | observational | supports | moderate | Prospective cohort n=136 (73 KD/63 controls): KD arm showed higher proportion achieving >=50% seizure reduction at 6mo vs controls (no exact % given) |
| Feng Y et al 2025 · Frontiers in Nutrition | meta-analysis | supports | low | Umbrella meta-analysis whose UNITS are meta-analyses, not primary studies: 'S (Study Design): An umbrella meta-analysis of relevant meta–analyses.' 'Ultimately, 18 studies were included for quantitative synthesis'. Zero weight as evidence_role: review (rubric v1.1.3 axis 2 — umbrella status turns on the units pooled, not on whether pooling occurred), matching the structurally identical s35027683 already zero-weighted here. The double count is literal, not theoretical: at least four of the 18 pooled meta-analyses are themselves full-weight rows on this claim — s32588435 (its ref 52, 'Cochrane Database Syst Rev . ( 2020 ) 6 : Cd001903 . doi: 10.1002/14651858.CD001903.pub5 , 32588435'), s36716045, s38315271 and s40252525. Denominator corrections to the prior extract. (1) RR 1.95 (1.58–2.33) is not a dietary-therapy or LGIT figure: 'After analyzing 24 studies from 17 articles, the intervention significantly increased the likelihood of achieving a seizure reduction of ≥50% (relative risk RR: 1.95; 95% confidence interval CI: 1.58–2.33).' — 24 estimates from 17 articles, pooling antioxidants AND dietary therapies together (Figure 2A is captioned 'Impact of antioxidants and dietary therapy on seizure reduction'). LGIT's own subgroup figure is a different number: 'Notably, the low - glycemic index therapy (LGIT) (RR: 8.64; 95% CI: 4.23–13.05) showed excellent efficacy, while the use of antioxidants alone did not show significant benefits'. (2) RR 3.33 (1.51–5.14) for ≥90% reduction is dietary therapy pooled over 'analyzing 7 studies from 5 articles'; the ketogenic diet's own subgroup, the only KD-specific efficacy estimate in the paper, is 'Sub - group analysis indicated that the ketogenic diet (RR: 5.15; 95% CI: 1.87–8.43) was particularly effective for this outcome'. (3) The seizure-freedom null rests on 'analyzing 8 studies from 6 articles, dietary therapy was not effective in achieving the ultimate goal of seizure - free status (RR: 0.68; 95% CI: 0.00–1.36)'. Quality lowered to low on the paper's own reported statistics: every interval it computed is symmetric about its point estimate on the untransformed ratio scale to within rounding (1.95 ±0.37/0.38; 3.33 ±1.82/1.81; 5.15 ±3.28 exactly; 8.64 ±4.41 exactly; 1.56 ±0.37/0.36), so these are not valid RR intervals, and negative lower bounds are handled by fiat — 'When the lower limit of the confidence interval turns negative (due to small sample sizes or estimation variability), we truncate it to 0.00 to ensure conceptual correctness.' The paper also reads one of its own significant estimates backwards — 'Melatonin was ineffective in reducing seizure frequency (RR: 0.30; 95% CI: 0.00–0.63)' — and registers two different PROSPERO IDs ('Registration No: CRD420251003717' in Methods vs 'identifier: CRD420251166437' in the abstract). No overlap or corrected-covered-area assessment across the 18 meta-analyses appears anywhere in the full text. Population is 'P (Population): Patients with epilepsy.', not restricted to drug-resistant epilepsy as this claim's object is. Direction on the object is unchanged and correct; the weight and the reliability label are what move. |
| Martin-McGill KJ, Bresnahan R, Levy RG, Cooper PN 2020 · Cochrane Database Syst Rev | meta-analysis | supports | moderate | Cochrane systematic review with meta-analysis, RCTs/quasi-RCTs of ketogenic diets in drug-resistant epilepsy, ITT for all primary analyses: 'We identified 13 studies with 932 participants; 711 children (4 months to 18 years) and 221 adults (16 years and over).' The pooled estimates come from SUBSETS, not from the 932. Children, KD vs usual care, 4 studies / 385 participants: 'Seizure freedom (RR 3.16, 95% CI 1.20 to 8.35; P = 0.02; 4 studies, 385 participants; very low-certainty evidence) and seizure reduction (RR 5.80, 95% CI 3.48 to 9.65; P < 0.001; 4 studies, 385 participants; low-certainty evidence) favoured KDs'. Adults, 2 studies / 141 participants: 'In adults, no participants experienced seizure freedom. Seizure reduction favoured KDs (MAD only) over usual care but, again, we are not confident that the effect estimated is accurate (RR 5.03, 95% CI 0.26 to 97.68; P = 0.29)' - point estimate in the same direction, but so imprecise it settles nothing. The authors' own certainty rating is part of the finding, not a caveat to drop: 'We assessed all 13 studies to be at high risk of performance and detection bias, due to lack of blinding', 'We rated the evidence for all outcomes as low to very low certainty', 'We are not confident that these estimated effects are accurate.' Conclusion: 'KDs could demonstrate effectiveness in children with drug-resistant epilepsy, however, the evidence for the use of KDs in adults remains uncertain.' Quality moderate (not high) on that self-reported certainty; stance stays supports because the direction on the claim's object is positive wherever it was estimable. |
Disagree, or know a study we missed?
We grade by evidence, not opinions. The way to weigh in is to point us to a study we haven't cited (check the evidence table above first), or to flag a problem with one we have. Every submission is reviewed; if it holds up, the grade updates and shows in Science Changes Its Mind.
Opens a short form. You'll sign in with Google so submissions are tied to a real account — we don't display your identity, and we only accept a link we can verify (PubMed, DOI, ClinicalTrials.gov).
Educational only, not medical advice. Grades and scores reflect published evidence weighted by study design and quality; see the methodology.