Discontinuing the Ketogenic Diet in Seizure-Free Children: Recurrence and Risk Factors


Address correspondence and reprint requests to Dr. Eric H. Kossoff at Suite 2158, The Johns Hopkins Hospital, 200 North Wolfe Street, Baltimore, MD 21287, U.S.A. E-mail: ekossoff@jhmi.edu


Summary:  About 10% of children become seizure-free after initiation of the ketogenic diet, and typically stop the diet after 2 yr. A retrospective chart review was performed of all children who became seizure-free on the ketogenic diet at our institution since 1993 and then discontinued the diet. Of 557 children started on the diet since 1993, 66 (12%) discontinued the diet after a median of 2.1 yr due to seizure freedom; 92% were also medication-free. Thirteen (20%) had recurrence of their seizures at a median of 2.4 yr (range: 0–5.5 yr) after the diet was stopped. Seven (58%) became seizure-free a second time, four with anticonvulsant therapy. Four patients (6%) continue to have daily seizures despite both medication and dietary therapies. The presence of recent EEG epileptiform activity, abnormal MRI, lower initial seizure frequency, and tuberous sclerosis complex all significantly increased the likelihood of recurrence.

The ketogenic diet, a treatment for intractable childhood epilepsy, was first developed by the Mayo Clinic in 1921 (Wilder, 1921). About 10% of individuals become completely seizure-free and often then discontinue use of anticonvulsant drugs while on the diet, despite often very intractable seizures at diet onset (Huffman and Kossoff, 2006). Traditionally, the diet is discontinued in children after 2 yr, especially in those who become seizure-free (Huffman and Kossoff, 2006). This is often an uncertain time for parents and children who had experienced dramatic success with the diet and are hesitant to return to their previous unlimited calorie-, carbohydrate-, and protein-containing diet. In a recent study of seizure recurrence after anticonvulsant discontinuation, 30% of children had recurrences, but only 1% were unable to resume seizure control again with anticonvulsants (Camfield and Camfield, 2005). Nonetheless, information about the risk of seizure recurrence, especially intractability, is not available for the ketogenic diet. The purpose of this study, therefore, was to determine the incidence of recurrence after discontinuing the diet because of seizure freedom, the risk of intractability in those whom seizures returned, and the presence of identifiable risk factors for recurrence.


Paper and electronic records were reviewed retrospectively for all 557 children who started the ketogenic diet at Johns Hopkins Hospital between November 1993 and April 2004, and who then discontinued the diet after at least 6 months without seizures. Sixty-six (11.2%) patients who met these criteria were identified. Diets were slowly discontinued in all children by lowering the ratio (fat: carbohydrate and protein) gradually over 2–3 months until urinary ketosis was no longer present, then increasing calories ad libitum (Huffman and Kossoff, 2006). No routine follow-up appointments were made for these children in epilepsy clinic; parents were asked to contact the physicians if seizures recurred. Information after the diet was discontinued for this study was obtained mostly from phone calls, emails, and parent appreciation letters. In June 2006, one investigator (EK) attempted to reach all families identified by phone to update information.

Categorical data were analyzed by Fisher's exact test while numerical data (medians) were analyzed by the Wilcoxon rank-sum test. The significance level for both tests was p = 0.05. The Johns Hopkins institutional review board approved this study.



Thirty-eight (58%) of the 66 patients were male. The median age at which children were started on the ketogenic diet was 3.1 yr (range, 0.2–9.7 yr) and patients remained on the diet for a median of 2.1 yr (range, 0.5–8.0 yr). These children had various seizure types, including absence (5%), atonic (2%), complex partial (9%), infantile spasms (21%), Lennox–Gastaut (6%), myoclonic (15%), and multiple seizure types (42%). Compared to the 557 children overall started on the diet, these seizure types are similar; 250 (45%) had multiple seizure types or Lennox–Gastaut, 70 (13%) had infantile spasms. Of the 13 (17%) with focal structural abnormalities on MRI, five children had dysplasias or focal atrophy, five had infarctions or hemorrhages, and three had tuberous sclerosis complex.

Within 1 week of starting the diet, 34 (52%) of the children were seizure-free. Of the remaining patients, 22 (33%) responded in the first 6 months, and 10 (15%) responded after that time. EEGs were obtained within 12 months of discontinuing the diet in 58 (88%) children.

The median duration of follow-up information was 3.3 yr (range, 0.1–10 yr). Forty-three (65%) patients were contacted or examined within the past 12 months. Fourteen (21%) children or their families could not be contacted after the diet was discontinued, despite several attempts to reach them by phone.


Thirteen children (20%) had recurrence of their seizures, at a median of 2.4 yr (range, 0–5.5 yr) after the diet was stopped (Table 1). In two children (patients 2 and 3), seizures recurred during the first month of weaning despite persistently high urinary ketosis. The diet was subsequently restarted at the previous (4:1) ratio, but patient 3 did not regain seizure control. EEG abnormalities after recurrence were similar to those seen during ketogenic diet use, except for patient 11, who had new onset of high voltage spike-wave discharges at 11 yr of age; 4.5 yr after the diet had been discontinued following a normal EEG.

Table 1. Demographic information and outcomes for children with seizure recurrence
PatientAge at diet initiation (yr)Seizure type(s)Prediet seizure frequency (per month)Time to initial diet responseDiet duration (yr)EEGaMRITime to recurrence (yr)bCurrent seizure frequency (per month)Diet restarted
  1. aWithin 12 months before discontinuation of the ketogenic diet.

  2. bZero implies recurrence took place during ketogenic diet wean.

 10.2Complex partial 6,1501st week1.2NormalNormal5.51No
 20.5Infantile spasms12,6001st month1.1EpileptiformNormal0.00Yes
 32.4Infantile spasms  1501st week2.5EpileptiformTuberous sclerosis0.030 Yes
 42.5Multiple  300>6 months2.8NormalLeft temporal atrophy3.50No
 53.1Multiple  3601st week2.0EpileptiformLeft parietal stroke0.50No
 63.3Multiple  7501st week3.0EpileptiformLeft temporal dysplasia5.00No
 73.9Multiple9,0001st week3.4EpileptiformTuberous sclerosis0.60No
 85Myoclonic  300>6 months5.0NormalNormal4.030 Yes
 95.1Infantile spasms  4501st week2.4Diffuse slowingTuberous sclerosis2.430 Yes
105.1Multiple  1501st month2.8NormalNormal1.30Yes
115.3Lennox–Gastaut  9001st week2.0NormalNormal4.51No
127.3Multiple  4201st week2.8EpileptiformNormal0.70Yes
137.8Complex partial  3001st month1.0Diffuse slowingNormal4.230 Yes

Children with and without recurrent seizures were compared to identify possible risk factors for seizure recurrence (Table 2). The three factors identified were lower initial median frequency of seizures (p = 0.03), presence of an abnormal EEG with epileptiform spikes within 12 months of diet discontinuation (p = 0.01), and the presence of a focal, structural abnormality on MRI (p = 0.02). The three children with tuberous sclerosis complex all had seizures recur, and only one child's symptoms were controlled a second time.

Table 2. Risk factors for seizure recurrence after discontinuation of the ketogenic diet, values expressed as medians (range)
Risk factorRecurrence (n = 13)Seizure-free (n = 53)p-value
  1. aWithin 12 months prior to discontinuation of the ketogenic diet.

  2. bMRI showed a focal, structural abnormality.

Age at first seizure (yr) 1.0 (0.0–5.9) 1.0 (0.0–6.3)0.63
Age at diet initiation (yr) 3.9 (0.2–7.8) 2.9 (0.5–9.7)0.34
Seizure-free duration (yr) 2.0 (1.0–3.3) 2.0 (0.5–7.8)0.83
Total diet duration (yr) 2.5 (1.0–5.0) 2.1 (0.5–8.0)0.42
Prediet seizure frequency (per month)420 (150–12,600)900 (2–10,800)0.03
Prediet AEDs tried4 (0–8)4 (0–9)0.59
Male gender8 (62%)30 (57%)0.50
Lennox–Gastaut or multiple seizure types7 (53%)27 (50%)0.55
Significant cognitive delay8 (62%)34 (64%)0.55
Infantile spasms3 (23%)11 (21%)0.56
Tuberous sclerosis complex3 (23%)0 (0%)0.01
Seizure-free after first week8 (62%) 26 (49%)0.31
Medication-free12 (92%) 49 (92%)0.68
Epileptiform EEGa6 (46%)5 (9%)0.01
Normal EEGa5 (38%)29 (54%)0.23
Abnormal MRIb6 (46%) 7 (13%)0.02

Various treatments were started in 12 of the 13 children with recurrence; one child (Patient 6) has had only two seizures in 6 yr and the family has elected not to treat. Seven (54%) children restarted the ketogenic or modified Atkins diet (Kossoff et al., 2006), and in two of the children (patients 2 and 12) seizures were completely controlled again. Patient 2 stopped the diet again after an additional year with no further seizures in the past 6 yr. Ten (77%) patients were placed on medications, half as an initial therapy and half after failing to respond to dietary therapy a second time. At this time, seven children (62%) are again seizure-free.


This is the first study, to our knowledge, to describe the risk of recurrent seizures after discontinuation of the ketogenic diet. Many prior studies of children who are seizure-free on anticonvulsant drugs have established that the rate of recurrence after discontinuation of anticonvulsant therapy ranges from 30 to 50% (Emerson et al.,1981; Gherpelli et al., 1992; Shinnar et al., 1994; Tennison et al., 1994; Peters et al., 1998; Camfield and Camfield, 2005; Sillanpaa and Schmidt, 2006). The recurrence rate assessed for the ketogenic diet appears to be slightly lower, with 20% having a recurrence. This is slightly higher than the 10% reported for children after resective surgery (Hoppe et al., 2006). Intractability in our study was 6%, midway between the distribution (1–20%) described for children after medication withdrawal (Gherpelli et al., 1992; Camfield and Camfield, 2005; Sillanpaa and Schmidt, 2006) and similar to the 3% after surgery (Hoppe et al., 2006).

Previously identified risk factors for seizure recurrence in children on anticonvulsants have been well-described in several studies, and include symptomatic epilepsy, age over 12 yr, and moderate to severe mental retardation (Shinnar et al., 1994; Tennison et al., 1994; Peters et al., 1998; Camfield and Camfield, 2005). Symptomatic epilepsy due to a focal cortical lesion(s) increased the risk for ketogenic diet patients, but older age and mental retardation did not. Evidence from abnormal EEGs suggests a correlation (Emerson et al., 1981; Shinnar et al., 1994; Tennison et al., 1994), with one series of 70 children describing an increased risk if EEG was abnormal in the year before the anticonvulsant taper, similar to our study (Gherpelli et al., 1992). Lastly, the median number of pre-diet seizures was surprisingly lower for patients with recurrent seizures as compared to the seizure-free group. The reason for this is unclear, as these children were neither younger at seizure onset nor more likely to have an epilepsy syndrome with frequent, but typically outgrown seizures (e.g., infantile spasms). Regardless of the explanation, a patient history of even the most severe intractability before success with the ketogenic diet should not dissuade an attempt to discontinue the diet.

There are a number of implications for patients who have had a complete response to the ketogenic diet. This study provides evidence that if patients have been seizure-free for at least 6 months; consideration should be given to discontinue the diet. In counseling families, it may be advisable to obtain a recent EEG and review previous MRI results prior to discontinuation. For patients with epileptiform spikes on EEG or focal abnormalities on MRI, the risk of recurrence is significantly higher and families may choose to keep their children on the diet for a longer duration. If seizures recur, however, the majority will be controlled again, often without the need for the ketogenic diet. Interestingly, three children with an abnormal MRI had tuberous sclerosis complex, all of whom had recurrence of seizures and two of whom continue to have seizures despite diet therapy. Previous studies of the ketogenic diet have reported excellent, even seizure-free, results in children with tuberous sclerosis complex (Kossoff et al., 2005). Perhaps the diet should be maintained for longer than 2 yr, or even indefinitely, in patients with tuberous sclerosis complex if seizure freedom is attained, based on the present results.

This study has several limitations. As this was not a prospective study, follow-up after diet discontinuation was not always considered to be necessary. Although a significant majority of patients were successfully contacted, it is possible that the others have had recurrent seizures but changed addresses, did not contact our group, and/or sought neurological care elsewhere. If these patients are excluded, 25% will have had recurrence. In addition, follow-up EEGs were not routinely obtained after diet discontinuation to confirm the absence of epileptiform activity. It is therefore possible that the diet may have normalized the EEG, as seen in five patients, similar to the effects of anticonvulsants in some generalized epilepsies (Duncan, 1987; Libenson and Caravale, 2001).

In conclusion, the risk of seizure recurrence after a seizure-free treatment course with the ketogenic diet appears slightly lower than after stopping use of anticonvulsant drugs, and similar to surgery. The risk of seizures becoming intractable in that situation fortunately was also rare. Children with abnormal MRIs or epileptiform EEGs may be more likely to have recurrences.


Acknowledgments:  Supported in part by the Pediatric Clinical Research Unit, NIH/National Center for Research Resources grant M01-RR00052. The authors gratefully acknowledge the editorial support of Drs. Pamela Talalay and Adam Hartman in preparing this manuscript.