Use of ketogenic diet therapy in infants with epilepsy: A systematic review and meta‐analysis

Ketogenic diet therapy (KDT) is a group of high‐fat, low‐carbohydrate diets used as an effective treatment option for children and adults with drug‐resistant epilepsy. There is limited research on the efficacy of KDT in infants, where there is the highest incidence of onset of the epilepsy. We aimed to systematically review studies that have reported on response to KDT in infants with epilepsy.


| INTRODUCTION
Epilepsy is a neurologic disorder characterized by an enduring predisposition to generate epileptic seizures, affecting 0.5%-1% of children. 1 The incidence of epilepsy is greatest in the first 2 years of life (56-88/100 000 children/y), 2 a population that remains most at risk for neurodevelopmental compromise in the longer term.
Approximately 20%-35% of children with epilepsy are drug-resistant, 3 having failed adequate trials of two tolerated and appropriately chosen antiepileptic drug (AED) schedules to achieve seizure control. 4 Early seizure control is associated with better developmental outcome, 5 but many of the epilepsies presenting in infancy have a poor prognosis for seizure control. 6 Ketogenic diet therapy (KDT) is a group of high-fat, low-carbohydrate diets used as a treatment option for drug-resistant epilepsy. Designed to mimic the effects of starvation on the body, fat is utilized as the principal energy source through production of ketones. KDT encompasses the classical ketogenic diet (KD), medium-chain triglyceride (MCT) KD, modified Atkin's diet (MAD), modified KD, and low glycemic index treatment (LGIT).
In the first randomized-controlled trial (RCT) of KDT for epilepsy, 38% children aged 2-16 years achieved ≥50% seizure reduction after 3 months, compared to 6% of controls (P < .0001). 7 Seven percent of children in the diet group had >90% seizure reduction, compared with no controls (P = .0582). There was no difference in effectiveness between patients who followed the classical diet or the MCT KD. 8 Further RCTs have corroborated the effectiveness of KDT, including the MAD, for epilepsy, with effects comparable to modern AEDs, in older children and adults. 9 This review aimed to systematically assess and summarize studies with seizure efficacy data in infants age <2 years following KDT as a treatment for epilepsy. The primary aim was to determine response rates and also to assess retention and report adverse side effects in this age group.

| METHODS
A systematic literature search was conducted in electronic databases (MEDLINE [PubMed], Embase [Ovid], the Cochrane Database of Systematic Reviews, Cochrane CENTRAL, and the National Institutes of Health clinical trial registry) with the following keywords: infant(s) OR child(ren), AND ketogenic OR medium chain triglyceride, AND epilepsy OR spasm(s) OR seizure(s). Reference lists of publications, including reviews, were manually searched. Publications including human participants only, written in English or Spanish were included; no date restrictions were set. The search was up to date as of October 16, 2019. This study was performed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. The protocol is registered on PROSPERO (https://www.crd.york.ac.uk/PROSP ERO/).

| Eligibility criteria
RCTs, case-control studies, and observational cohort studies fulfilling the following criteria were included: reviewed by LL and NS. Cases of disagreement were discussed until consensus was reached.

| Quality appraisal
The Effective Public Health Practice Project (EPHPP) Quality Assessment Tool for Quantitative Studies (https://merst.ca/ ephpp /, accessed October 11, 2019) was used to assess quality of the evidence. This tool includes questions on selection bias, study design, confounders, blinding, data collection methods, and withdrawals and dropouts. Each section is rated as "strong," "moderate," or "weak"; collectively, these are used to derive a global rating of "strong" (no weak ratings), "moderate" (one weak rating), or "weak" (two or more weak ratings). Global ratings are reported in this study. Quality assessment was performed for all included studies, rated independently by LL and NS, and disagreements were resolved by consensus.

| Outcomes
The primary outcome was efficacy of KDT for epilepsy in infants age <2 years, presented as the number or proportion of infants achieving ≥50% seizure reduction after ≥1 month of follow-up.
Secondary outcomes were: 1. Seizure freedom rates at ≥1 month of follow-up 2. Retention rates 3. Side effects

| Data extraction
The following data (where available), were extracted for each study:

| Data analysis
Descriptive analysis was conducted for the primary outcome and for seizure-freedom rates; data were summarized as aggregate rates (reported as intention-to-treat), ranges, median, and interquartile range (IQR), for numerical outcomes. Collective means were calculated from study means or individual patient data, where information was provided. Collective standard deviation could not be calculated, as these data were not available for all studies. Fisher exact or Mann-Whitney U tests were used to investigate associations of gender, age at seizure onset, age at diet onset, and number of failed AEDs with KDT response. Response rates were separated by epilepsy syndrome, etiology, or epilepsy/seizure type (including cohorts that included infants with only one epilepsy or seizure type) where such detail was provided. Narrative syntheses of retention rates, side effects, and diet-initiation protocols were compiled. Meta-analyses of proportions, including uncontrolled studies only (the RCTs had different treatment arms and outcome measures), were conducted using the statistical package meta 10 in R (version 3.6.2). 11 Inverse-variance meta-analyses were performed using the metaprop function, applying a logit transformation to the outcome, and the method of DerSimonian-Laird 12 used to estimate the heterogeneity variance. Response was defined, first, as ≥50% seizure reduction and, second, as seizure freedom, at 3 months or as close as possible to 3 months or at unspecified time points, where 3-month response was not reported. Residual estimates were calculated to identify outliers (Z-statistic >2). The degree of heterogeneity was evaluated through visual inspection of forest plots, using the Q-test and I 2 statistic. An I 2 of 25%, 50%, and 75% was assumed to indicate low, medium, and large heterogeneity, respectively; a Q-statistic with P-value < .1 was considered evidence of substantial interstudy heterogeneity. Subgroup analyses were conducted to investigate potential explanatory variables of heterogeneity with the following variables: study design (retrospective or prospective), response time point (3 months or "other"), and whether the study included just infants or infants as part of a wider cohort. The R 2 statistic was also evaluated in subgroup analyses to quantify the proportion of variance explained by covariates in each model.

| RESULTS
In total, 2206 publications were identified. After removing 685 duplicates, 1521 studies were screened and assessed for eligibility; 37 studies met our inclusion criteria. Four studies were excluded, [13][14][15][16] as these infants were reported in other included studies. 17,18 Thirty-three studies were included in the final analysis, of which five included solely infants and 28 included infants as part of a wider cohort ( Figure 1).
We also identified 32 case studies, including infants that met inclusion criteria but were not part of a cohort study (Table S1).

| Study characteristics
Eighteen studies were prospective, of which two were RCTs 19,20 (one of which also included a parallel cohort study); 15 were retrospective, single-arm cohort studies. Table 1 shows summary descriptive data for studies that included solely infants and Table 2 for studies including infants as part of a wider cohort.
A total of 534 infants who had followed a KDT for ≥1 month were included, of which 208 were from cohorts including solely infants. Eight studies only included infants with infantile spasms or West syndrome; most studies included patients with a range of epilepsy syndromes.

| Study quality
All included studies, classified by the EPHPP method, were rated as "low" quality.
No studies were given a "strong" rating for selection bias, although all prospective studies were classed as "moderate" ("not applicable" for retrospective studies). The two RCTs 19,20 were the only studies given a "strong" rating for study design. In no studies were assessors blinded to the intervention or exposure status of participants. Most studies were single-arm or did not mention any differences between groups, so ratings for confounders were "weak." Data collection tools consisted predominantly of medical record review and were not shown to be valid or reliable. Only one study was given a "strong" rating for withdrawals and drop-outs, 23 and one a "moderate" rating 24 ; the remainder were classed as "poor" ("not applicable" for retrospective studies).

| Aggregated efficacy
Data were available from 33 (controlled and uncontrolled) studies, including a total of 534 infants. The number of infants evaluated differs because efficacy rates were not reported at each time point in every study.   Single-arm cohort study (retrospective) Refractory status epilepticus 1 (0.5 y)     Presented as a percentage of individuals starting ketogenic dietary therapy (intent-to-treat).
b Inconsistent age data presented in article: Table 1 states Patient 3 was age 0.5 y at diet initiation, in text, stated 'age of the patients ranged from 2 to 9 y'. c Study includes comparative component but all cases exposed to KD treatment. d For patients who had already been treated with an enteral KD, age was calculated as 'age at initiation of parenteral KD minus 3 mo or, for patients who had followed an enteral KD for <3 mo, minus the number of months for which enteral KD had been followed. e When parenteral KD was used in children who had been already treated with an enteral KD, treatment response was defined as maintenance of the previously achieved ≥ 50% reduction compared to baseline, defined as the 3 mo before initiation of parenteral KD. In cases where parenteral KD was the initial treatment, response was defined as the absolute reduction in seizure frequency ≥ 50% after 3 mo of KD treatment.
f Results presented for individuals randomised to KDT. Age at epilepsy onset and Time from epilepsy onset to trial treatment (median, minimum and maximum) indicates that some individuals in the parallel cohort were aged > 2 y at diet start (no age-specific diet efficacy data given).
For studies giving efficacy rates at unspecified time points, 37 of 87 infants (43%) achieved ≥50% seizure reduction, 17 of which (20% of the whole group) were seizure-free.
The other RCT compared classical KD to standard adrenocorticotropic hormone (ACTH) treatment in infants with West syndrome. 19 Ten of 16 (62%) in the KD group and 11 of 16 (69%) in the ACTH group achieved the primary end point of electroclinical seizure remission at 28 days. Six of 16 (38%) in the KD group and 7 of 16 (44%) in the ACTH group remained seizure-free at last follow-up.

| Predictors of efficacy
Response (≥50% seizure reduction or seizure freedom) at 3 months (n = 33 from eight studies) and at unspecified time points (n = 49 from six studies) was not associated with gender, age at diet onset, or number of failed AEDs; age at seizure onset was higher in infants who achieved seizure freedom at 3 months (Z = −2.162, P = .036) (Tables S2-S5). Association with response at other time points was not evaluated as data were available for <7 infants.
Response rates separated by epilepsy syndrome are detailed in Supporting Information. Response rates were not separated into epilepsy or seizure type, as these were grouped inconsistently.

| Diet initiation protocols
Of the five studies that included solely infants, two of five o (40%) adopted a fasting protocol for KDT initiation, 17,33 and three of four (75%) admitted all, or almost all, infants. 23,25,33 Three studies 17,25,33 reported that they monitored blood glucose, either "periodically," with "any episode of emesis or reduced oral intake" or, for specific patients, regular checks every 6-12 hours. Two studies reported that full caloric requirements were prescribed. 25,33

| Retention rates
Minimum duration of dietary treatment was 1 month (due to our review criteria), and longest duration was 58 months. 38 Retention rates at 3, 6, 12, and 24 months were given in 12 studies. Aggregated rates were 197 of 235 (84%) at 3 months, 17 17,25,37 All individuals in the above studies were following a classical KD, with the exception of n = 1 on an MCT KD 42 (unknown diet duration), n = 3 on a MAD 24 (one on diet for 7 days; one still on diet at 6 months; and another still on diet F I G U R E 2 Proportion of responders to ketogenic diet therapy, defined as ≥50% seizure reduction at 3 mo follow-up (or at an unspecified time point where 3 mo data are not available). The vertical dotted line is placed at the point estimate of the summary proportion, with the horizontal tips of the diamond representing the 95% confidence interval of the summary proportion using a random-effects model. The squares represent the point estimate of each study, with the horizontal lines representing the 95% confidence interval around the point estimate. The size of each square is proportional to the weight of the study in the pooled estimate. Estimates are separated into prospective (prosp) and retrospective (retro) studies, and pooled proportions including all studies at 8 months at the time of publication), and n = 2 on LGIT 37 (one on diet for 16 months and one on diet for 24 months).
Dyslipidemia was reported in 17 of 104 infants (16%) in one study, 17 although only one child required dietary intervention to resolve this. Another study reported a mild elevation in serum triglycerides in one infant, who continued the diet, and "markedly elevated triglyceride level concerning for lipoprotein lipase deficiency" in another 25 ; one infant had type I hyperlipidemia after following KDT for 23 months. 33 Seventeen deaths were reported in infants either during or after KDT, of which 14 were believed to be due to the underlying disease or intercurrent illness, rather than the diet per se. 25,33,42 For the three deaths reported by Hong et al, 17 it was not stated whether they were thought to be related to KDT or not. Approximately 60% infants achieved ≥50% seizure reduction, with 33% becoming seizure-free. Results from RCTs were similar or higher than pooled results from uncontrolled studies. Caution must be exercised when interpreting overall  response rates due to interstudy heterogeneity and low-quality data. Our findings are consistent with a recent review of observational studies, in which 57.4% infants with infantile spasms achieved >50% seizure reduction and 33.63% became seizure-free within 6 months of KDT. 43 Response rates in infants seem comparable to rates in older children, if not more promising. A systematic review looking at the KDT efficacy in children younger than 18 years of age (one study also included adults), in uncontrolled studies only, estimated 56% participants to achieve ≥50% seizure reduction and 16% becoming seizure-free. 44 There have since been three RCTs comparing KDT to a control group in children 2 years of age or older 7,45,46 ; 38%-56% participants achieved ≥50% seizure reduction and 1%-15% became seizure-free.
Potential higher response rates in infants compared to older individuals may be due to increased compliance or biological factors. The suggestion that higher age at seizure onset is associated with increased likelihood of seizure freedom at 3 months is consistent with findings from individual studies in our review, but conflicts with others: Older age at spasm onset has been associated with increased likelihood of >90% improvement in spasm frequency at 6 months, 17 but higher seizure freedom rates have been reported in children age <1.5 years compared to >1.5 years, 22 and children <12 months of age have been reported to have a "tendency for a better response" compared to those aged >12 months. 42 Response rates in patients with Dravet syndrome, Ohtahara syndrome, tuberous sclerosis complex, and generalized encephalopathy may be particularly high, although numbers included in our review are small. Response rates for those with infantile spasms or West syndrome were in keeping with the wider cohort, although it should be noted that, in these patients, seizure freedom tends to be the desired clinical outcome.
Benefits of KDT in addition to seizure control were also reported, although inconsistently, in our included studies, such as cognitive and behavioral improvements 33 and developmental gains. 13,17,26,42 Authors of a systematic review commented that KDT seemed to have cognitive benefits in a lower proportion of infants compared to older children and adults, but evidence was available from only two infant studies. 47 Retention rates in infants were comparable to those reported in RCTs, including infants and older children: 74%-90% retention (presented as 10%-26% dropout) at 3-4 months, 66% by 6 months, and 58% at 16 months in individuals following a classical KD. 9 One may expect retention to be higher in infants younger than 2 years, as parents/ guardians are, in theory, able to exert a greater degree of control over the infant's diet. However, due to the clinical vulnerability of this age group, the trial period for dietary treatment may be shorter than in older children and adults, leading to more rapid diet discontinuation in the case of ineffectiveness.
Reported adverse side effects in infants were inconsistently reported and thus likely unreported. As in studies of older children, they were commonly gastrointestinal and rarely led to diet discontinuation. 7,48 Dyslipidemia was reported in 12% of infants in our review, similar to older children: 12.8% for hyperlipidemia, including prospective and retrospective studies. 48 Hypoglycemia and renal stones were also reported in a similar proportion of infants (2% and 3% in our review) compared to older children (1.8% and 1.4%, including prospective studies only). 48 The time points at which side effects occurred were inconsistently reported-it would be useful in future studies to guide health care professions and families regarding risk of specific side effects when KDT is initiated and in the longer term. As with all individuals initiating KDT, monitoring for hypoglycemia and excess ketosis is recommended for infants as well as gastrointestinal symptoms. 49 The report of an infant becoming comatose with hypoglycemia and acidosis was in a study where a fasting protocol was adopted, 33 although hypoglycemia also occurred in infants initiated without fasting 25 ; both occurred when admitted to start KDT, which is recommended for infants younger than 1 year of age. 49 There are several limitations of this review. All studies included were rated as "low" quality, including two RCTs, due to lack of blinding (understandably problematic with any "real food" dietary intervention) and the lack of overt validity or reliability of data-collection methods. No studies including age-specific diet efficacy data for infants compared a KDT to placebo or no change in treatment, although, in reality, clinicians are faced with the question of starting KDT or trying another AED, and high-quality evidence comparing these two treatment options may be more appropriate. Other treatment changes, including additional medications, dose reductions, and discontinuation, were not detailed consistently for infants, which may have affected diet-efficacy rates. More infants have been treated with KDT and are published in the literature, but without age-specific diet efficacy data, and so have not been included in this review. Alongside this data-collection bias, publication bias against negative results must also be considered.
Study design explained a significant amount but not all heterogeneity between ≥50% reduction outcomes; our subgroup variables could not explain heterogeneity for seizure-freedom outcomes. Clinical heterogeneity both within and between studies, for example, in terms of epilepsy syndrome, seizure type, concomitant AEDs, and type of KDT, may alter clinical outcome and likely contributed to heterogeneity. Higher quality, more homogenous data may allow