Weaning from ketogenic diet therapy in children with epilepsy: Insights from a retrospective study

Abstract

Objective

This study was undertaken to describe weaning practices following ketogenic diet therapy (KDT) in children with epilepsy and to identify clinical factors associated with seizure exacerbation or antiseizure medication adjustments during or after weaning from KDT.

Methods

This retrospective observational study examined patients who initiated and discontinued KDT between 2016 and 2022 at a tertiary epilepsy center. Patients with GLUT1 deficiency, ongoing KDT, or less than 1 year of follow‐up postdiscontinuation were excluded. Clinical, electroencephalographic (EEG), magnetic resonance imaging, KDT response, weaning duration, and seizure outcomes were analyzed. Responders were defined by ≥50% seizure reduction. Seizure worsening and/or antiseizure medication (ASM) adjustments during weaning were the primary outcomes.

Results

Among 57 evaluable patients, 49% were responders and 25% became seizure‐free. During weaning, 62% experienced seizure exacerbation or required ASM modifications. Unfavorable outcomes were significantly associated with shorter KDT duration, shorter weaning periods, higher ASM burden, and abnormal preweaning EEG. Among responders, seizure worsening was not significantly linked to weaning speed alone. At 1 year, seizure freedom was more common in patients who did not experience any issue during the weaning (65% vs. 30%, p = .009).

Significance

Our findings underscore the importance of individualized KDT discontinuation plans based on efficacy, EEG activity, and ASM burden. Although prolonged weaning was more common in responders, weaning duration alone did not predict outcomes. These results highlight the need for prospective studies to define optimal weaning strategies for KDT.

Key points.

• Weaning from ketogenic diet is a critical period as with all antiseizure therapy.

• Weaning from ketogenic diet is associated with 62% having seizure worsening or needing ASM changes during discontinuation.

• Seizure worsening during weaning is associated with shorter use of ketogenic diet, high antiseizure medication burden, and abnormal EEG.

1. INTRODUCTION

Ketogenic diet therapy (KDT) has been used for more than a century as a treatment for drug‐resistant childhood epilepsy, mimicking the metabolic effects of fasting to reduce seizure frequency. ¹ Whereas the classical KDT uses a strict 4:1 lipid‐to‐protein‐plus‐carbohydrate ratio, alternative, less restrictive versions—such as the modified Atkins diet and low glycemic index therapy—have been developed to improve long‐term adherence. ² The efficacy of KDT is evidence‐based, with usually ≥50% seizure reduction reported in approximately half of patients. ³ , ⁴ , ⁵

When effective and well tolerated, KDT is typically continued for several months to years. However, the method used to discontinue the diet varies according to patient‐specific factors and clinician preference. ⁶ There is currently no consensus on the optimal approach to KDT weaning. The latest guidelines from the International Ketogenic Diet Study Group recommend a gradual taper over 1–3 months, although this recommendation is not based on comparative studies. ²

Early protocols, such as Talbot's 1930 recommendation, advocated for a very gradual reintroduction of carbohydrates over several months. By 2009, international consensus guidelines adopted a weaning approach similar to that used for antiseizure medications (ASMs), favoring a progressive taper. ⁷ Since then, clinical practices have continued to evolve. ⁶ A 2011 retrospective study involving 183 children found no significant association between weaning speed and seizure recurrence. However, children with incomplete seizure control (50%–99% reduction) and those on multiple ASMs were at greater risk of relapse. ⁸ The most recent guidelines, published in 2018, recommend a tapering period of 1–3 months for patients who responded well to KDT, with the option of more rapid discontinuation in nonresponders. ²

This study aims to describe real‐world weaning practices in children with epilepsy treated with KDT and to identify clinical factors associated with seizure worsening or treatment adjustment during weaning from KDT. We also investigated whether shorter weaning durations are linked to a greater risk of seizure recurrence during tapering or loss of seizure control after KDT complete discontinuation.

2. MATERIALS AND METHODS

This retrospective observational study was conducted at Robert‐Debré University Hospital (Assistance Publique–Hôpitaux de Paris, Paris, France) and received approval from the local ethics committee (CEERB Paris Nord). We reviewed the medical records of all children with epilepsy who initiated and subsequently discontinued KDT at this center between January 2016 and March 2022. Data collection included demographic information, electroencephalographic (EEG) and magnetic resonance imaging (MRI) findings, ASM history, KDT characteristics and efficacy, adverse events, weaning method and duration, and also seizure outcomes during weaning and up to 1 year post‐KDT discontinuation. When informations were missing from medical charts, caregivers were contacted by phone to supplement the data.

Epilepsy syndromes were classified according to the 2022 International League Against Epilepsy (ILAE) framework. Developmental and epileptic encephalopathies (DEE) that did not meet criteria for a specific ILAE‐defined syndrome were designated as “nonsyndromic DEE” with the underlying etiology specified. Etiologies were categorized according to ILAE definitions into structural (further classified as congenital or acquired), genetic, metabolic, immune, infectious, or unknown.

Patients were not included if they had a condition requiring prolonged use of KDT without any plan for weaning, such as GLUT1 deficiency syndrome or pyruvate dehydrogenase deficiency. We also excluded patients who were still on the diet at the time of data analysis, were lost to follow‐up, or had not completed at least 1 year off the diet following discontinuation.

All patients were treated with a modified ketogenic diet, initiated with a daily allowance of 10–15 g of carbohydrates, adequate protein intake, and high‐fat foods. We did not apply any fasting phase for KDT initiation. KDT was initiated during an inpatient department admission to allow monitoring for side effects and to provide structured therapeutic education by a trained dietitian. Patients received vitamin supplementation and, when appropriate, ketogenic formulas. Medical follow‐up visits were scheduled monthly during the first 3 months, and then every 3 months thereafter, including clinical assessments, laboratory tests, and imaging as needed. In cases of favorable response to the diet, it was typically continued for 2 years. After several months, carbohydrate intake was gradually increased to 15–25 g/day, and tapering of at least one ASM was encouraged when clinically appropriate.

The weaning process was conducted at home over 1–2 months, typically by replacing one ketogenic meal with a nonketogenic meal every 1–2 weeks. In cases of limited efficacy or poor tolerability, weaning was accelerated and completed over 7–10 days. The dietitians and physicians assessed adherence to the weaning protocol during the clinics throughout the weaning process.

KDT efficacy was assessed at 1 month, 3 months, and again within 1 month before weaning. Patients were defined as “responders” if seizure frequency was reduced by at least 50%, and “high‐responders” if seizure frequency was reduced at least 90% or if they became seizure‐free. Seizure worsening during weaning was defined as a clinically significant increase in seizure frequency (at least 50% increase compared to the previous assessment period), either reported spontaneously by caregivers or documented during routine follow‐up. Treatment modifications and their effects on seizure control were also recorded. During the weaning process, individual adjustments were possible by extending the duration of each step or modifying the ASM regimen. ASM adjustments during the weaning were done only in cases of increased seizure frequency based on the evaluation of the pediatric epileptologist (B.D.‐P., S.A.).

2.1. Statistical analysis

Statistical analyses were performed using Prism (version 10, GraphPad Software). Categorical variables were summarized as counts and percentages and compared using Fisher exact test. Continuous variables were summarized using median and interquartile range (IQR), with group comparisons made using the Mann–Whitney U‐test.

For categorical comparisons, odds ratios (ORs) with 95% confidence intervals (CIs) were calculated. For continuous variables, the Hodges–Lehmann estimator was used to estimate the median difference between groups, with 95% CIs when applicable. A two‐sided p‐value < .05 was considered statistically significant.

Patients with unevaluable KDT response (i.e., KDT duration < 1 month) were excluded. Variables with missing data (e.g., weaning duration) were analyzed using complete‐case analysis. The number of patients included in each comparison is noted in the relevant tables.

3. RESULTS

A total of 60 patients were included in our study according to our methods (Table 1). The median age at seizure onset was 7 months (range = 0–36), and the most frequent diagnosis was infantile epileptic spasms syndrome in 38% of cases. The first ASM was initiated at a median age of 9.5 months (range = 1–156), and KDT was initiated at a median age of 31 months (range = 4–168) and it was the third‐line treatment in 27% of patients. Nine children (15%) received KDT via enteral feeding, and two (3%) underwent KDT while maintaining the use of human milk by breastfeeding.

TABLE 1.

Patients characteristics at KDT onset [N = 60].

Characteristic	Value
Sex: male	32 (53%)
Age at first seizure occurrence, months	7 (0–36)
IESS	23 (38%)
DEE‐EtMAS	12 (20%)
Nonsyndromic–developmental and epileptic encephalopathy–not otherwise specified	9 (15%)
CAE‐IGE	8 (13%)
DEE‐LGS	3 (5%)
Focal epilepsy	3 (5%)
DEE‐EIMFS	1 (2%)
Rasmussen syndrome	1 (2%)
Brain MRI
Normal	25 (42%)
Abnormal	28 (47%)
Not available	7 (11%)
Unknown cause	25 (42%)
Genetic cause	17 (28%)
Congenital structural cause	9 (15%)
Acquired structural cause	7 (11%)
Metabolic cause	1 (2%)
Immune cause	1 (2%)
Age when first ASM was introduced, months	9.5 (1–156)
Number of ASMs received prior to KDT	3 (1–10)
Number of ASMs received when initiating KDT	2 (0–7)
Age at KDT initiation, months	31 (4–168)
Enteral administration of KDT	9 (15%)
Breastfeeding at KDT onset	2 (3%)

Note: Categorical variables are reported as n (%). Continuous variables are reported as median (range).

Abbreviations: ASM, antiseizure medication; CAE, childhood absence epilepsy; DEE, developmental and epileptic encephalopathy; EIMFS, epilepsy of infancy with migrating focal seizures; EtMAS, epilepsy with myoclonic–atonic seizures; IESS, infantile epileptic spasms syndrome; IGE, idiopathic generalized epilepsy; KDT, ketogenic diet therapy; LGS, Lennox–Gastaut syndrome; MRI, magnetic resonance imaging.

3.1. Efficacy and safety of KDT in our study

The median duration of KDT in this study was 11 months (range = 0–59). Twenty‐three percent of children remained on the diet for less than 3 months, whereas 48% continued beyond 1 year, including 20% who followed it for more than 2 years. Three children were excluded from the efficacy analysis because they discontinued KDT within the first month (Patient 1 stopped breastfeeding with both an increase in seizures and a behavior change; Patient 2: stopped due to gastrointestinal side effect–vomiting; Patient 3: stopped due to nonadherence to the diet).

Among the 57 children treated with KDT, 49% were responders based on the change in seizure frequency, including 32% classified as high‐responders, and 25% achieved complete seizure freedom. No significant association was found between KDT response and epilepsy diagnosis across the seven syndromic diagnoses (p = .79, chi‐squared test) nor with underlying etiologies (genetic, structural, or unknown; p = .39, chi‐squared test). Overall, no clinical factor in our study was found to be significantly associated with KDT efficacy. However, there was a nonsignificant trend toward better outcomes in children receiving KDT by enteral nutrition (25% vs. 7%, p = .08, Fisher exact test).

After KDT initiation, 22% of patients experienced gastrointestinal symptoms such as discomfort or constipation and 13% showed signs of growth deceleration. Laboratory monitoring revealed transitory hyperlipidemia in 15% of cases, and three patients presented with an asymptomatic decrease in carnitine levels. No vitamin or micronutrient deficiencies were detected and no cases of renal lithiasis were reported during the study period.

3.2. Weaning and discontinuation characteristics

The primary reasons for discontinuing KDT were lack of efficacy (55%), adverse effects (30%), and/or treatment duration exceeding 2 years (20%).

At the time of weaning, the median age was 4 years (range = 5 months–16 years), and children received a median of two antiseizure drugs (range = 0–5). The EEG before weaning was abnormal (epileptiform abnormalities or slow waves) in 81% of patients (49 of 59 with available data).

The median duration of the weaning process was 30 days (range = 1–150), with information available for all but five patients. Responders to KDT were significantly more likely to undergo a prolonged weaning process (>6 weeks) compared to nonresponders (74% vs. 17%, Fisher exact test, p < .001). Conversely, 41% of nonresponders experienced a short weaning period (<2 weeks) versus just 4% of responders. These findings likely reflect a more gradual tapering approach in patients for whom the diet had been beneficial.

3.2.1. Full cohort analysis by weaning outcome

Of the 60 patients included, 37 experienced seizure exacerbation and/or required ASM adjustments during weaning, whereas 23 remained stable without such events (Table 2). The two groups did not differ significantly in sex, age at seizure onset, diagnosis, or number of ASMs at KDT initiation. However, patients who remained stable were more likely to have received enteral nutrition (30% vs. 5%, p = .02) and responded favorably to KDT. High‐responders (≥90% seizure reduction) were notably more frequent in the stable group (57% vs. 15%, OR = 7.54, p = 0.001), and overall response (≥50% seizure reduction) showed a a similar trend, with a higher frequency in the stable group, but it is not significant (53% vs. 38%, OR = 3.03, p = .06).

TABLE 2.

Clinical and treatment characteristics of the full cohort (N = 60), comparing patients with and without seizure or treatment changes during KDT weaning.

Characteristics of all patients, N = 60	No seizure increase or ASM change, n = 23	Seizure increase and/or ASM change, n = 37	p and OR
Male gender	12 (52%)	20 (54%)	>.99
Age at first seizure occurrence, months	7 (0–96)	10 (0–96)	.57
First seizure at <2 years old	15 (65%)	23 (62%)	>.99
IESS	11 (48%)	12 (32%)	.28
DEE‐EtMAS	4 (17%)	8 (22%)	.75
DEE‐LGS	0	3 (8%)	.28
DEE‐EIMFS	1 (4%)	0	.38
NS‐DEE	4 (17%)	5 (14%)	.72
Rasmussen syndrome	0	1 (3%)	>.99
CAE‐IGE	2 (9%)	6 (16%)	.70
Focal epilepsy	1 (4%)	2 (5%)	>.99
Abnormal brain MRI a	13 (61%)	15 (47%)	.40
Unknown cause	7 (30%)	18 (49%)	.19
Genetic cause	8 (35%)	9 (24%)	.39
Congenital structural cause	2 (9%)	7 (19%)	.46
Acquired structural cause	5 (22%)	2 (5%)	.09
Metabolic cause	1 (4%)	0	.38
Immune cause	0	1 (3%)	>.99
Number of ASMs prior to KDT	3 (1–8)	3 (1–10)	.75
Age at KDT introduction, months	24 (4–168)	36 (3–180)	.91
Enteral nutrition	7 (30%)	2 (5%)	.02 b
Number of ASMs at KDT initiation	2 (1–4)	2 (0–7)	.61
KDT responders [≥50% reduction]	15 (53%)	13 (38%) c	OR = 3.029 p = .06
Superresponders [≥90% reduction]	13 (57%)	5 (15%) c	OR = 7.54 p = .001 b
KDT duration, months	21 (1–54)	5 (0–59)	.001 b
KDT weaning duration, days d	60 (7–150)	30 (1–120)	.001 b
Number of ASMs before weaning	1 (0–4)	2 (1–5)	.004 b
Abnormal EEG before weaning	14 (64%) e	34 (92%)	.013 b
Seizure freedom at 1‐year postdiscontinuation	15 (65%)	11 (30%)	OR = 4.432 p = .008 b
Reduction of 50%–99% in seizure frequency 1‐year postdiscontinuation	2 (9%)	20 (54%)	OR = .080 p = .001 b
Median media number of ASMs 1 year after discontinuation	1 (0–4)	3 (0–5)	.001 b

Note: Categorical variables are reported as n (%). Continuous variables are reported as median (range).

Abbreviations: ASM, antiseizure medication; CAE, childhood absence epilepsy; DEE, developmental and epileptic encephalopathies; EEG, electroencephalography; EIMFS, epilepsy of infancy with migrating focal seizures; EMAtS, epilepsy with myoclonic–atonic seizures; IESS, infantile epileptic spasms syndrome; IGE, idiopathic generalized epilepsy; KDT, ketogenic diet therapy; LGS, Lennox–Gastaut syndrome; MRI, magnetic resonance imaging; NS, nonsyndromic; OR, odds ratio.

^a MRI data were not available for two patients in the “no change” group (n = 21) and for five patients in the “change” group (n = 32).

^b Statistically significant.

^c KDT efficacy data were unavailable for three patients who were on KDT for less than 1 month.

^d Weaning duration data were not available for three patients in the “no event” group (n = 20) and for one patient in the “event” group (n = 36).

^e EEG data were unavailable for one patient (n = 22).

Both KDT duration and weaning duration were significantly longer in the group that did not experience any event during weaning (median = 21 vs. 5 months and 60 vs. 30 days, respectively, p = .0003 and p = .0018). Prior to weaning, these patients also had fewer ASMs (median = 1 vs. 2, p = .005) and less frequent EEG abnormalities (64% vs. 92%, p = .013) than those who experienced an ASM change and/or a seizure increase during KDT discontinuation.

At 1 year following KDT discontinuation, seizure freedom was more common in the group that did not experience any issue during weaning (65% vs. 30%, OR = 4.43, p = .009), along with lower rates of seizure reduction (9% vs. 54%, OR = .081, p = .0003) and reduced ASM burden (median = 1 vs. 3 ASMs, p = .0005).

Among the 57 patients who remained on KDT for at least 1 month and had evaluable outcome data, seizure worsening during weaning occurred in 32% of responders (9/28) and 38% of nonresponders (11/29), with no significant difference between groups (Fisher exact test, p = .78, OR = .77).

3.2.2. Weaning KDT in the responders to the diet

Among the 28 responders, 15 experienced no seizure worsening or change in ASM during the full weaning period, whereas 13 showed seizure recurrence and/or required ASM adjustments (Table 3). There were no significant differences between the two groups in terms of sex, age at seizure onset, syndrome, MRI findings, or number of ASMs at KDT initiation (Table 3).

TABLE 3.

Comparison of the occurrence of seizure or ASM modification during weaning in the responders to the diet.

Characteristics of KDT responders (n = 28) stratified by seizure or treatment evolution during weaning	No seizure increase or ASM change, n = 15	Seizure worsening and/or treatment modification, n = 13	p
Male gender	9 (60%)	7 (54%)	>.99
Age at first seizure occurrence, months	7 (0–36)	7 (0–96)	.69
IESS	9 (60%)	3 (23%)	.07
DEE‐EtMAS	3 (20%)	3 (23%)	>.99
DEE‐LGS	0	1 (8%)	.46
DEE‐EIMFS	1 (7%)	0	>.99
NS‐DEE	2 (13%)	3 (23%)	.64
Focal epilepsy	0	1 (8%)	.46
CAE‐IGE	0	2 (15%)	.21
Abnormal brain MRI a	8 (57%)	4 (36%)	.43
Unknown cause	5 (33%)	9 (69%)	.13
Genetic cause	5 (33%)	1 (8%)	.17
Congenital structural cause	0	1 (8%)	.46
Acquired structural cause	3 (20%)	2 (15%)	>.99
Metabolic cause	1 (7%)	0	>.99
Number of ASMs prior to KDT	2 (1–8)	3 (2–10)	.08
Age at KDT introduction, months	10 (4–168)	36 (5–180)	.14
Enteral nutrition	6 (40%)	1 (8%)	.08
Number of ASMs at KDT initiation	2 (1–4)	2 (0–2)	.44
Median KDT duration, months	24 (3–54)	18 (6–59)	.04 b
KDT weaning duration, days	120 (7–150) c	60 (14–120)	.07
Number of ASMs received before weaning	1 (0–4)	3 (1–5)	.001 b
Abnormal EEG before weaning	6 (42%) d	11 (85%)	.05 b
No seizure 1 year after discontinuation	13 (87%)	4 (31%)	.005 b
Number of ASMs 1 year after discontinuation	0 (0–4)	3 (1–5)	.001 b

Note: Categorical variables are reported as n (%). Continuous variables are reported as median (range).

Abbreviations: ASM, antiseizure medication; CAE, childhood absence epilepsy; DEE, developmental and epileptic encephalopathy; EEG, electroencephalography; EIMFS, epilepsy of infancy with migrating focal seizures; EMAtS, epilepsy with myoclonic–atonic seizures; IESS, infantile epileptic spasms syndrome; IGE, idiopathic generalized epilepsy; KDT, ketogenic diet therapy; LGS, Lennox–Gastaut syndrome; MRI, magnetic resonance imaging; NS, nonsyndromic.

^a MRI data were not available for one patient in the “no change” group and for two patients in the “change” group.

^b Statistically significant.

^c Weaning duration data were not available for two patients (n = 13).

^d EEG data were unavailable for one patient.

However, patients who experienced seizure or ASM adjustment had a significantly shorter duration of KDT (median = 18 vs. 24 months, p = .037), had been exposed to a higher number of ASMs before weaning (p = .001), and were more likely to have an abnormal EEG before tapering (p = .046). One year after KDT discontinuation, this group was also significantly less likely to be seizure‐free (31% vs. 87%, p = .006) and more likely to be treated with multiple ASMs (p = .001).

In patients with seizure or treatment modifications, a trend toward shorter weaning duration (p = .076) and lower use of enteral nutrition (p = .084) was also noted, although these differences did not reach statistical significance.

3.2.3. Comparison of KDT responders and nonresponders among patients with seizure exacerbation during weaning

Among the 20 patients who experienced seizure worsening during KDT weaning and had evaluable response data, 9ME were classified as responders and 11 as nonresponders. Clinical and treatment characteristics were compared between these two groups (Table 4). No significant differences were observed in sex, epilepsy etiology, diagnosis (including DEE–epilepsy with myoclonic–atonic seizures and other subgroups not shown), age at KDT initiation, number of ASMs used prior to KDT, or presence of abnormal MRI or EEG recording before weaning (all p > .05).

TABLE 4.

Clinical and treatment characteristics of patients with seizure increase during weaning, comparing KDT responders and nonresponders (n = 20).

Clinical variable	KDT responders, n = 9	Nonresponders, n = 11	p	OR/HL diff
Male gender	5 (55%)	7 (63%)	>.999	OR = .71
First seizure at <2 years of age	4 (44%)	5 (45%)	>.999	OR = .96
DEE‐IESS	1 (11%)	2 (18%)	>.999	OR = .56
DEE‐EtMAS	3 (33%)	3 (27%)	>.999	OR = 1.33
DEE‐LGS	1 (11%)	2 (18%)	>.999	OR = .56
CAE‐IGE	2 (22%)	1 (9%)	.566	OR = 2.86
NS‐DEE	1 (11%)	1 (9%)	>.999	OR = 1.25
Focal epilepsy	1 (11%)	0	.45	OR = ∞
RS	0	1 (9%)	>.999	OR = ∞
Abnormal brain MRI	1 (11%)	5 (45%)	.3043	OR = .16
Unknown cause	9 (100%)	6 (55%)	.038 a	OR = ∞
Genetic cause	0	2 (18%)	.479	OR = 0
Structural congenital cause	0	2 (18%)	.479	OR = 0
Immune cause	0	1 (9%)	>.999	OR = 0
Age at KDT initiation, months	36 (9–180)	43 (5–96)	.9256	HL diff = 0
Number of ASMs prior to KDT	4 (2–10)	3 (1–8)	.5238	HL diff = −1
KDT duration, months	17 (11–22)	3 (1–19)	.0006 a	HL diff = +12
Reason of discontinuation: side effects	3 (33%)	3 (27%)	>.999	OR = 1.33
Abnormal EEG before weaning	8 (88%)	11 (100%)	.4500	OR = 0
>3 ASMs before weaning	4 (44%)	0 (0%)	.0260 a	OR = ∞ (95% CI = 1.72 to ∞)
KDT weaning at >6 weeks b	5 (71%)	1 (9%)	.0128 a	OR = 25 (95% CI = 1.934 to 313)
KDT weaning at 2–6 weeks b	2 (29%)	5 (45%)	.6371	OR = .48
KDT weaning at <2 weeks b	0	5 (45%)	.1013	OR = 0

Note: Categorical variables are reported as n (%). Continuous variables are reported as median (range). One patient from the seizure increase group was not included in this analysis due to unevaluable KDT response (<1 month on diet).

Abbreviations: ASM, antiseizure medication; CAE, childhood absence epilepsy; CI, confidence interval; DEE, developmental and epileptic encephalopathies; EEG, electroencephalography; EMAtS, epilepsy with myoclonic–atonic seizures; HL diff, Hodges–Lehmann difference; IESS, infantile epileptic spasms syndrome; IGE, idiopathic generalized epilepsy; KDT, ketogenic diet therapy; LGS, Lennox–Gastaut syndrome; MRI, magnetic resonance imaging; NS, nonsyndromic; OR, odds ratio; RS, Rasmussen syndrome.

^a Statistically significant.

^b Weaning duration was unavailable for two KDT responders.

However, responders had significantly longer KDT duration compared to nonresponders (median = 17 months [IQR = 11–22] vs. 3 months [IQR = 1–19], p = .0006, Hodges–Lehmann estimated difference: +12 months). Additionally, receiving more than three ASMs before weaning was significantly associated with responder status (44% vs. 0%, p = .026, OR = ∞, 95% CI = 1.72–∞).

Weaning duration also differed significantly between the two groups. Responders were more likely to have undergone prolonged weaning (>6 weeks; 71% vs. 9%, p = .0128, OR = 25.0, 95% CI = 1.93–313), whereas nonresponders more frequently experienced shorter tapering periods, including weaning completed in less than 2 weeks (45% vs. 0%).

4. DISCUSSION

This study provides an analysis of the efficacy and discontinuation characteristics of KDT in a heterogeneous cohort of 60 pediatric patients with drug‐resistant epilepsy. Nearly half of the children were responders to KDT, with one in four achieving seizure freedom without any clinical predictor of KDT. The weaning process emerged as a critical phase. Among the 60 children included, more than half (62%) experienced seizure exacerbation and/or required adjustments to their ASM during or shortly after discontinuing ketogenic diet. An unfavorable outcome of the weaning process was significantly linked to a shorter duration of KDT, a more rapid weaning process, a higher number of ASMs prior to tapering, and the presence of EEG abnormalities before discontinuation. Abrupt discontinuation was associated with higher risk of seizure recurrence and adjustment of ASM, even among responders.

The weaning strategy used at our center consisted of progressively replacing one ketogenic meal with a nonketogenic meal every 1–2 weeks, rather than recalculating the lipid‐to‐carbohydrate ratio through precise nutritional adjustments. This pragmatic approach, which is less burdensome for families, may better reflect real‐world practices but differs from the structured, ratio‐based tapering described in some protocols. ²

Although a longer weaning duration is associated with better outcomes in our study, this trend did not reach statistical significance among KDT responders. In this subgroup, tapering duration was not significantly different between patients who maintained seizure control without ASM changes and those who experienced seizure worsening or treatment modifications. These findings suggest that, although clinicians may prefer a more gradual tapering approach in responders, seizure stability during weaning cannot be attributed solely to the weaning strategy itself. Other factors, such as baseline seizure frequency, preweaning EEG abnormalities, and the number of concomitant ASMs, may play a role in determining outcomes after KDT withdrawal. ⁸

Our findings are partly consistent with limited prior reports on KDT discontinuation. In the study by Worden et al., ⁸ no significant association was found between seizure relapse and the weaning strategy, categorized as immediate (<1 week), rapid (1–6 weeks), or slow (>6 weeks). They showed that relapse was more frequent in patients with only focal seizure reduction (50%–99%) and those receiving multiple ASMs. This is also in line with our data. ⁸ Although the 2018 International Ketogenic Diet Study Group consensus recommends tapering the diet over 1–3 months following a favorable response, these guidelines are not supported by prospective or comparative studies. ² A recent study proposed an alternative approach in which KDT is abruptly discontinued during hospitalization by introducing high‐carbohydrate meals under EEG monitoring; among 15 patients, 13 tolerated this protocol without seizure worsening. ⁹ This inpatient strategy allows close seizure monitoring and may alleviate caregiver anxiety during discontinuation. When considered alongside our results, these observations underscore the heterogeneity of current weaning practices.

In our cohort, 33% of patients experienced seizure worsening during the weaning phase. This seems more than in a previous study where it was reported a worsening in approximately 14% whenever the weaning was progressive or abrupt. ⁸ However, the follow‐up duration in our study may have enhanced the detection of delayed seizure relapse. Our findings suggest that seizure recurrence during KDT tapering may be more common than previously recognized, underscoring the need of careful monitoring throughout the discontinuation process.

This study presents several strengths, including a well‐defined pediatric epilepsy population, follow‐up extending up to 1 year after KDT discontinuation, and comprehensive documentation of both clinical characteristics and treatment trajectories. Conducted in a real‐world setting and encompassing both responders and nonresponders, the study offers valuable insights into the diversity of current weaning practices and their clinical implications.

However, this study has several limitations, including the single‐center setting and the retrospective design. The weaning protocol might present heterogeneity, including variability in the weaning process due to the absence of a standardized time frame and nonstandardized energy and carbohydrate content of replacement meals, all of which may introduce confounding factors or outcome bias. The small sample size limited the ability to perform multivariate analyses to identify independent predictors of seizure worsening. Furthermore, ketone levels were not consistently monitored, preventing an evaluation of the correlation between the degree of ketosis and seizure control or relapse. Consequently, we were unable to determine whether lower ketone levels prior to tapering could have served as predictors of adverse outcomes.

In conclusion, our findings highlight the clinical importance of the need for individualized weaning plans based not only on KDT efficacy, but also on EEG activity, ASM burden, and underlying etiology. Although a gradual taper was more commonly applied in responders, weaning speed alone did not predict outcomes in this subgroup. The pragmatic weaning approach used at our center, progressively replacing ketogenic meals without precise ratio recalculations, may enhance the generalizability of our results to settings where individualized, dietitian‐led tapering is not routinely available. These findings support cautious, tailored discontinuation strategies, particularly in partially responsive patients, and reinforce the need for prospective studies to determine the optimal weaning protocols that minimize the risk of seizure recurrence.

AUTHOR CONTRIBUTIONS

Noémie Donnard: Design of the study; data collection; analysis of the data; writing the manuscript. Blandine Dozières‐Puyravel: Design of the study; data collection; analysis of the data; editing the manuscript. Katia Geraldes: Data collection; editing the manuscript. Céline Perrot: Data collection; editing the manuscript. Virginie Quéméner: Data collection; editing the manuscript. Estelle Goujon: Data collection; editing the manuscript. Pierre Truchy: Data collection; editing the manuscript. Hala Nasser: Data collection; editing the manuscript. François‐Xavier Mauvais: Data collection; editing the manuscript. Marion Danse: Data collection; editing the manuscript. Stéphane Auvin: Design of the study; data collection; analysis of the data; writing the manuscript.

FUNDING INFORMATION

This work benefited from a government grant managed by the Agence Nationale de la Recherche under the France 2030 program, under the reference ANR‐23‐IAIIU‐0010.

CONFLICT OF INTEREST STATEMENT

S.A. is Deputy Editor for Epilepsia. He has received personal fees for lectures or advice from Angelini, Biocodex, Eisai, Encoded, GRIN Therapeutics, Jazz Pharmaceuticals, Longboard, Lundbeck, Neuraxpharm, Nutricia, Orion, Proveca, Servier, Stoke, UCB Pharma, and Xenon. He has been an investigator for Eisai, Lundbeck, Proveca, Servier, Takeda, and UCB Pharma. B.D.‐P. has received personal fees for lectures or advice from Biocodex, Jazz Pharmaceuticals, and UCB Pharma. None of the other authors has any conflict of interest to disclose. We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.