Motives and predictors of modified Atkins diet discontinuation as treatment of adults with drug‐resistant epilepsy

Abstract

Objective

This study was undertaken to explore the motives and predictors of modified Atkins diet (MAD) discontinuation as a treatment of adults with drug‐resistant epilepsy (DRE).

Methods

A single‐center observational longitudinal study was made of a cohort of patients treated with MAD during the first 6 years at a multidisciplinary outpatient clinic dedicated to adult patients. Time to MAD discontinuation analyses were used to explore sociodemographic, clinical, and biochemical predictors of MAD persistence.

Results

Eighty patients initiated MAD (54% females, median age = 29 years, median epilepsy duration = 21 years, median baseline seizure frequency = 30/month, median antiseizure medications = 4, 68% focal epilepsy, 64% structural etiology, and 26% genetic cause). The patients used the diet for a median time of 122 days. Fifteen percent of patients were very early dropouts (before 30 days), 20% were early dropouts (30–89 days), and 17.5% were late dropouts (>365 days). The motives for dropout (n = 64) were noncompliance (34.4%), inefficacy (23.4%), side effects (23.4%), and diet tiredness (18.8%). Female sex (adjusted hazard ratio [HR] = 1.75, p = .042), focal seizures with preserved consciousness (adjusted HR = 2.69, p = .017), higher glycemic level at baseline (adjusted HR = 1.03, p = .044), and lower level of serum total proteins at baseline (adjusted HR = .55, p = .052) were associated with shorter persistence on MAD. The HR for discontinuation was also higher for patients without daily seizures (HR = 1.72, p = .038). Among patients with at least 3 months in MAD (n = 51), reduced clinical response at month 3 was associated with earlier discontinuation of MAD, as measured by lower percent seizure reduction (HR = .98, p = .004) and <50% seizure reduction (HR = 2.44, p = .027).

Significance

The dropout rate of MAD in adults with DRE is high even at dedicated centers. Females, patients with less severe epilepsies, and patients previously on carbohydrate‐based diets may be more prone to earlier dropouts. These patients require closer monitoring in specialized ketogenic diet clinics to prevent MAD discontinuation before the minimum period necessary to evaluate response.

Key points.

• The dropout rate of MAD in adults with DRE was high, and most patients discontinued diet in the first 6 months, mainly for noncompliance.

• Female sex was a predictor of MAD discontinuation.

• Factors suggesting less severe epilepsies (not having daily seizures, having focal seizures with preserved consciousness, and shorter disease duration) predicted MAD discontinuation.

• Factors suggesting previous carbohydrate‐based diets (lower baseline proteinemia and higher baseline glycemia) were associated with earlier dropouts.

• Inefficacy at 3 months and epilepsy characteristics influencing response (focal epilepsies and nongenetic etiologies) correlated with earlier dropouts.

1. INTRODUCTION

High‐fat, low‐carbohydrate ketogenic diets (KDs) are treatment options for adults with drug‐resistant epilepsy (DRE), based on well‐established evidence from observational and randomized controlled studies. ¹ , ² Less restrictive variants like modified Atkins diet (MAD) have shown similar efficacy to the classic version and have expanded their use in the adult population. ³ Overall, the rate of response (defined as ≥50% seizure reduction) ranged from 17% to 60% depending on the study; other reported outcomes are decrease in the seizure severity and improvement in mood, energy, and overall quality of life. ⁴

However, availability of dedicated adult services and tools to measure and improve compliance are recognized as unmet needs. ⁵ Even for MAD in adults, the dropout rate is high, ranging from 35%–38% at 3–6 months of treatment to 61%–66% at 6–12‐month follow‐up. ⁴ This poses huge barriers in designing and developing studies on this treatment, especially clinical trials. ⁶ From a pharmacoeconomic perspective, according to a recent systematic review, the cost‐effectiveness of KD was not demonstrated, as the costs outweigh the seizure reduction and improvements in quality of life. However, these conclusions were based on limited data. ⁷ Even though the cost‐effectiveness of the treatment has not been clearly demonstrated, the availability of KD in health systems is currently recommended as a treatment option for children and adult patients with DRE, based on individual clinical need. ⁸ , ⁹

As KD is a potentially effective treatment, but complex and costly at the same time, it is important to better understand the large proportion of KD dropouts. This study aims to explore the motives and predictors of MAD discontinuation in the treatment of adults with DRE.

2. MATERIALS AND METHODS

2.1. Study design and population

We performed a single‐center observational longitudinal study of a cohort of patients treated with MAD in the context of a multidisciplinary outpatient clinic dedicated to dietetic treatment for adult patients with DRE since 2018. Our outpatient clinic is part of a national referral epilepsy center in a tertiary university hospital located in an urban area. However, the clinic also serves patients from rural areas. The referral area of the hospital is approximately 620 000 people (i.e., covering part of the northern region of Portugal). The clinic also receives patients from the country not covered by the regular referral area of the hospital.

The KD clinic consists of one neurologist–epileptologist (R.S.), two dietitians (A.C. and M.M.T.), and a nurse (C.T.). Patients with DRE, regardless of the underlying etiology, are referred by their assistant neurologist–epileptologist to the clinic. Then, the neurologist of the MAD team confirms that patients have DRE, according to the ILAE definition, ¹⁰ and no contraindications to MAD, according to current recommendations. ⁸ , ¹¹

2.2. Baseline and follow‐up procedures

Using a semistructured interview with the patient and/or the caregiver (Supporting Information S1) and the review of clinical records, patients' and caregivers' sociodemographic data were collected, as well as clinical data regarding epilepsy and comorbidities. The ILAE 2017 epilepsy classification and its 2025 update were used for epilepsy characterization. ¹² , ¹³

After starting MAD, patients were evaluated every 3 months. Patients (or their caregivers) were instructed to keep a seizure calendar for the duration of the treatment. Seizure frequency at baseline and at 3‐month follow‐up (i.e., 90 days) corresponded to the mean number of seizures per month in the previous 3 months. The percentage of seizure reduction was calculated at follow‐up. Additionally, the semistructured interview conducted by the neurologist–epileptologist at baseline and at follow‐up addressed the subjective impression of benefit in seizure severity (i.e., seizure and postictal period duration, falls, use of rescue medications, and frequency of bilateral tonic–clonic seizures) and subjective improvement of cognition/mood/behavior (Supporting Information S1). This qualitative assessment was performed independently of the quantitative evaluation, as the seizure calendars were only analyzed after the visit.

The biochemical monitoring was performed as recommended by international guidelines. ⁸ Among the biochemical parameters regularly monitored in clinical practice, the following were analyzed at baseline and trimonthly: glycemia, albumin, serum total proteins, ammonia, urea, total cholesterol, low‐density lipoprotein (LDL) cholesterol, triglycerides, dyslipidemia, uric acid, vitamin B12, folic acid, and vitamin D. All biochemical parameters were obtained from a single peripheral venous blood sample, collected at our clinic on the day of the clinical assessment (in the morning, but fasting was not mandatory). Dyslipidemia was considered when LDL > 116 mg/dL or triglycerides > 150 mg/dL. ¹⁴ In such cases, the dietician reinforced the use of unsaturated fats and recommended lifestyle changes.

2.3. Dietary treatment

MAD was prescribed by a dietitian of the multidisciplinary outpatient clinic, consisting of 10–20 g/day carbohydrate with liberal fat intake and adequate protein amount (ratio of 1–2:1 of fat to protein and carbohydrates combined). ¹⁵ Formula‐based 4:1 ratio KDs were added to encourage compliance. ¹⁶ Medium chain triglycerides (MCTs) were also added for some patients to improve ketosis and efficacy. ¹⁷ Decisions on whether to use MCTs and the velocity of daily carbohydrate amount reduction were made by the dietitian for each patient individually.

All patients received a daily multivitamin, calcium, and vitamin D supplement, and no changes in antiseizure medication (ASM) regimens were performed in the first 3 months of MAD.

The adherence to the diet was monitored through patient and/or caregiver report. Ketonemia (i.e., the level of beta‐hydroxybutyrate ketone body in blood) was home‐monitored on a regular basis for fine‐tuning the diet, using a finger prick and a test strip, and values of at least 2 mmol·L⁻¹ were targeted, when possible. The averages of measurements in the first month and before the third month follow‐up were calculated.

Time in MAD was calculated as the interval in days between starting the diet and the dropout date or December 3, 2024, if the patient was still on diet at this timepoint. In most cases, the decision to discontinue MAD was a shared decision between the clinician and the patient/caregiver. The motives for stopping MAD were recorded and classified by the clinician into one of the following categories: noncompliance, inefficacy, side effects, and diet tiredness. Noncompliance was considered when the patient did not adhere to the dietetic plan as based on patient or caregiver report. Dropout for inefficacy was considered when no quantitative or qualitative benefit was obtained in compliant patients. Dropouts classified as diet tiredness corresponded to patients who adhered to the diet for some time but became noncompliant due to continuous diet restrictiveness, regardless of the benefit.

2.4. Statistical analysis

Descriptive statistics were used for cohort characterization. Chi‐squared test (or Fisher exact test when appropriate), Mann–Whitney test, and Kruskal–Wallis test (with Dunn post hoc test) were used for group comparisons. The hazard ratios (HRs) for discontinuation associated with the baseline demographic, clinical, and biochemical variables were computed with univariate Cox regressions. Variables with a p‐value associated with the crude HR of <.1 were considered for multivariable analyses. A backward selection method (the threshold for variable removal was set at .1) was applied. Due to missing data (related to laboratory incidents and/or COVID19 pandemic restrictions), biochemical variables were analyzed separately. The first multivariable Cox regression analysis only considered demographic and clinical variables. The second analysis included biochemical parameters in addition to the final model of the first multivariable Cox regression analysis. The threshold for statistical significance was set at p < .1. Kaplan–Meier survival curves illustrate the significant impact of certain sociodemographic and clinical variables on time in MAD.

2.5. Ethics

This study was conducted according to the ethical standards of the 1975 Declaration of Helsinki, revised in 2013, and approved by the institutional ethical committee (Comissão de Ética CHUP/ICBAS; approval number 2019.031 [027‐DEFI/028‐CE]). Written informed consent was obtained from all the participating patients or their legal guardians.

3. RESULTS

3.1. Study population

Since May 2018, 80 patients with DRE initiated MAD, and as of December 3, 2024, 64 (80%) had already discontinued the diet. The median number of days on the diet for the whole cohort was 122 days (ranging between 3 and 2264 days; Q1 = 48 and Q3 = 540 days). Sociodemographic and clinical characteristics of the patients are presented in Table 1.

TABLE 1.

Baseline sociodemographic and clinical characteristics of adult patients with drug‐resistant epilepsy on modified Atkins diet: Whole cohort, dropouts (those who had already discontinued the treatment at time of data collection), on diet (those who were still on treatment at time of data collection), and according to sex.

	Whole cohort, N = 80	Dropouts, n = 64	On diet, n = 16	Female, n = 43	Male, n = 37	p
Sociodemographic
Sex, female	43 (54%)	39 (61%)	4 (25%)	–	–	–
Age, years	29 (21, 40)	28 (28, 35)	34 (23, 47)	27 (21, 36)	33 (22, 44)	.164
Intellectual disability	49 (61%)	40 (63%)	9 (56%)	25 (58%)	24 (65%)	.538
Residence, rural	46 (58%)	36 (56%)	10 (63%)	25 (58%)	21 (57%)	.901
Institutionalized	34 (43%)	26 (41%)	8 (50%)	19 (44%)	15 (42%)	.822
With caregiver a	61 (76%)	52 (81%)	9 (56%)	33 (77%)	28 (76%)	.911
Clinical [epilepsy]
Type, focal	54 (68%)	46 (72%)	8 (50%)	30 (70%)	24 (65%)	.641
Etiology	Structural, 51 (64%) Genetic, 21 (26%) Infectious, 5 (6%) Unknown, 3 (4%)	Structural, 42 (65%) Genetic, 14 (22%) Infectious, 5 (8%) Unknown, 3 (5%)	Structural, 9 (56%) Genetic, 7 (44%)	Structural, 31 (72%) Genetic, 6 (14%) Infectious, 5 (12%) Unknown, 1 (2%)	Structural, 20 (54%) Genetic, 15 (41%) Unknown, 2 (5%)	.015
Age at seizure onset, years	5 (1, 13)	5 (6, 11)	1 (1, 12)	5 (1, 14)	5 (1, 13)	.392
Disease duration, years	21 (17, 32)	20 (20, 26)	30 (20, 38)	20 (16, 27)	24 (18, 36)	.046
Baseline monthly seizures	30 (8, 74)	29 (37, 89)	40 (8, 74)	20 (8, 56)	30 (11, 75)	.390
Daily seizures	36 (45%)	27 (42%)	9 (56%)	16 (37%)	20 (54%)	.131
Bilateral tonic–clonic seizures	24 (30%)	17 (27%)	7 (44%)	9 (21%)	15 (41%)	.056
Focal seizures with preserved consciousness	10 (13%)	9 (14%)	1 (6%)	6 (14%)	4 (11%)	.745
Seizures causing falls	38 (48%)	30 (47%)	8 (50%)	22 (51%)	16 (43%)	.479
Concomitant ASM	4 (3, 5)	4 (3, 4)	4 (3, 5)	4 (3, 4)	4 (3, 5)	.028
VNS	24 (30%)	21 (33%)	3 (19%)	12 (28%)	12 (32%)	.660
Comorbidities	55 (69%)	46 (72%)	9 (56%)	28 (65%)	27 (73%)	.450
Caregivers, n = 59
Sex, female	52 (88%)	44 (86%)	8 (89%)	30 (94%)	22 (82%)	.229
Age, years	54 (50, 62)	54 (53, 58)	54 (45, 59)	54 (48, 60)	54 (50, 64)	.591
Education, years	9 (4, 12)	9 (7, 10)	8 (6, 12)	9 (4, 12)	8 (4, 12)	.896

Note: Data are presented as frequency (percentage) and median (Q1, Q3 percentiles). Chi‐squared test (or Fisher exact test when appropriate) and Mann–Whitney test were used to compare females and males.

Abbreviations: ASM, antiseizure medication; VNS, vagal nerve stimulation.

^a Two patients with in‐patient services.

The study cohort is heterogenous in terms of epilepsy classification, reflecting the broad indication for MAD in patients with DRE. Most of these patients have long‐lasting highly drug‐resistant and disabling epilepsies, as their median epilepsy duration was 21 years (Q1 = 17, Q3 = 32), baseline seizure frequency was 30 per month (Q1 = 8, Q3 = 74), and number of concomitant ASMs was 4 (Q1 = 4, Q3 = 5); 30% also had a vagus nerve stimulator implanted. Most patients had focal epilepsy (68%) and structural etiology (64%); characteristics included malformations of cortical development, n = 26; vascular, n = 9; nonlesional, n = 8; hippocampal sclerosis, n = 5; posttraumatic, n = 2; and postcranial radiotherapy, n = 1. All focal epilepsies had been previously considered for epilepsy surgery, but patients either were considered poor candidates at the time due to location(s) and/or extension of the lesions (n = 47), decided to postpone surgery (n = 3), or had already failed surgery (n = 4). Approximately one fourth had a genetic cause (26%): tuberous sclerosis complex (n = 5), Glut‐1 deficiency (n = 3), hypomelanosis of Ito (n = 1), mutations in ARHGEF9, STXBP1, and CHD2 genes (n = 1 for each), and genetic cause presumed by phenotype (n = 9).

Males had longer disease duration (median = 24 vs. 20 years, p = .046) and more bilateral tonic–clonic seizures (41% vs. 21%, p = .056) than females (Table 1). Epilepsy etiology was also different between females and males (p = .015), and the standardized residuals for males were as follows: −.6 for structural, 1.7 for genetic, −1.5 for infectious, and .1 for other. Subsequent analyses revealed that genetic etiology was more frequent among males (41% vs. 14%, p = .007) and that only females had infectious etiology (11.6% vs. 0%, p = .058).

Forty‐nine patients (61%) had intellectual disability, 24 (49%) had generalized epilepsy, 12 (25%) had Lennox–Gastaut syndrome, 24 (49%) had structural etiology (e.g., perinatal hypoxic–ischemic encephalopathy and large cortical malformations), 18 (37%) had genetic etiology, four (8%) had infectious etiology, and three (6%) had unknown etiology. Of those with intellectual disability, 34 (69%) attended an institution but only two with inpatient care. All patients were orally fed, except for two who had gastrostomy. Swallow studies were not systematically performed, but food consistency was individually adapted if dysphagia was suspected.

Fifty‐five patients (69%) had comorbidities, including the following: dyslipidemia (n = 25), psychiatric diseases (n = 22; eight depression/anxiety, 14 behavioral problems requiring sedatives to control behavior), hypothyroidism (n = 6), gastroesophageal diseases (n = 4), renal lithiasis (n = 3), hypertension (n = 3), skin diseases (n = 3; eczema, urticaria, and acne), osteoporosis (n = 1), rheumatoid arthritis (n = 1), hepatic adenomatous (n = 1), von Willebrand factor deficiency (n = 1), and biliary lithiasis and tachycardia (n = 1).

3.2. Timing and motives of MAD discontinuation

Fifteen percent of the 80 patients who started MAD were very early dropouts (before 30 days), 20% were early dropouts (30–89 days), and 17.5% were late dropouts (after 365 days). The motives for dropout among the 64 patients were noncompliance (n = 22, 34.4%), inefficacy (n = 15, 23.4%), side effects (n = 15, 23.4%), and diet tiredness (n = 12, 18.8%). Motives for diet discontinuation did not differ according to presence of intellectual disability (p = .406), any comorbidity (p = .350), or psychiatric comorbidity (p = .148). However, motives for diet discontinuation (at any time) were related to ketonemia levels at month 1 (p = .027). A post hoc analysis revealed that the noncompliance group had lower ketone levels in the blood than those with inefficacy (median = .82 vs. 1.80 mmol·L⁻¹, p = .014) or side effects (median = .82 vs. 2.03 mmol·L⁻¹, p = .014).

The number of days on the diet varied according to the motive for discontinuation (p < .001). The median time on the diet prior to dropout was 38 days for side effects (Q1 = 18, Q3 = 110), 60 days for noncompliance (Q1 = 14, Q3 = 95), 97 days for inefficacy (Q1 = 67, Q3 = 156), and 732 days for diet tiredness (Q1 = 480, Q3 = 918). Table 2 explores the relationship between timing and motive for dropout.

TABLE 2.

Characterization of patients according to modified Atkins diet status ^a and according to time on diet.

Characteristic	Very early, <30 days	Early, 30–89 days	90–179 days	180–364 days	Late, ≥365 days
Dropout	n = 12	n = 16	n = 16	n = 6	n = 14
Noncompliance	7	7	5	1	2
Inefficacy	0	5	7	3	0
Side effects	5	4	4	1	1
Diet tiredness	0	0	0	1	11
Female	6	13	9	3	8
Age, years	27 (18, 40)	32 (18, 68)	35 (18, 67)	30 (19, 40)	32 (18, 62)
Focal epilepsy	7	12	14	6	7
Genetic etiology	4	2	2	1	5
On diet	n = 1	n = 0	n = 1	n = 2	n = 12
Female	0	0	1	2	1
Age, years	48	–	28	39 (24, 53)	34 (20, 57)
Focal epilepsy	1	0	1	1	5
Genetic etiology	0	0	0	1	6

Note: Data are presented as frequency for all variables, except age. Age is presented as median (minimum, maximum).

Abbreviations: d, days; y, years.

^a At the time of the present review.

Among the 22 patients discontinuing MAD for noncompliance, nonadherence to the dietetic plan as confirmed by dietitian review translated into low ketonemia values during the first month (Q1 = .510, Q2 = .818, and Q3 = 1.638 mmol·L⁻¹). Patients and caregivers reported different reasons for nonadherence (e.g., difficulty in finding some ingredients, friends' and family's mistrust, uncontrollable appetite for carbohydrate‐based foods, and perception of a monotonous dietetic plan).

Regarding the 15 patients who stopped MAD for side effects, five had worsening of seizure frequency (four had focal seizures; infectious, posttraumatic, immune, nonlesional, and genetic etiologies), five had gastrointestinal symptoms (three had vomiting and two had diarrhea), two had prurigo pigmentosa, one had lethargy, one had worsening of behavioral problems, and there was one death (possible sudden unexpected death in epilepsy).

Among the 12 patients discontinuing MAD because of diet tiredness, their median ketonemia measurement at 3 months was 2.150 mmol·L⁻¹ (Q1 = 1.725, Q3 = 2.550). When they elected to stop MAD, one third of these patients had ≥50% seizure reduction (including one seizure‐free), seven (58%) reported qualitative benefits in seizures, and five (42%) reported benefits in cognition (Supporting Information S2).

Among the 64 patients who discontinued MAD (39 females and 25 males), the median time on the diet until discontinuation was 93 days (range = 1207, Q1 = 39, Q3 = 305). The 16 patients who remained on the diet at the time of the present study (four females and 12 males) had a median time on MAD of 638 days (range = 1553, Q1 = 341, Q3 = 1894).

3.3. Baseline predictors of MAD discontinuation

Time to MAD discontinuation analyses were used to explore sociodemographic, clinical, and biochemical predictors of MAD persistence (Table 3). The HR for dropping MAD was higher for female patients (HR = 2.04, p = .007), for patients without daily seizures (HR = 1.72, p = .038) or genetic etiology (HR = 1.72, p = .075), and for patients with focal epilepsy (HR = 1.63, p = .082) and focal seizures with preserved consciousness (HR = 2.38, p = .019). Kaplan–Meier curves are presented in Figure 1. Shorter disease duration (HR = .98, p = .090), lower level of serum total proteins at baseline (HR = .52, p = .027), and higher glycemic levels at baseline (HR = 1.02, p = .067) were also related to fewer days on MAD. Patients with intellectual disability (49/80) were not more prone to diet dropout (HR = .81, p = .411). When sex, disease duration, type of epilepsy, genetic etiology, daily seizures, and focal seizures with preserved consciousness were analyzed as covariates, only female sex (adjusted HR = 1.90, p = .018) and focal seizures with preserved consciousness (adjusted HR = 1.96, p = .073) continued to be associated with earlier dropout, after removal from the multivariable Cox regression model of independent variables with p > .1 (backward selection method). In a separate analysis, considering these two variables and the significant biochemical parameters, female sex (adjusted HR = 1.75, p = .042), focal seizures with preserved consciousness (adjusted HR = 2.69, p = .017), higher glycemic level at baseline (adjusted HR = 1.03, p = .044), and lower level of serum total proteins at baseline (adjusted HR = .55, p = .052) continued to be associated with shorter persistence on MAD.

TABLE 3.

Baseline predictors of modified Atkins diet discontinuation (N = 80 patients).

Predictor		HR	95% CI	p
Sociodemographic
Sex	Female	2.04	1.21, 3.41	.007
Age, years		.99	.97, 1.01	.483
Residence	Urban	1.16	.70, 1.90	.565
Intellectual disability	With	.81	.48, 1.35	.411
Caregiver	With	1.27	.68, 2.38	.462
Institutionalization		.73	.44, 1.21	.225
Epilepsy
Age at seizure onset, years		1.02	.99, 1.06	.130
Disease duration, years		.98	.96, 1.00	.090
Type of epilepsy	Focal	1.63	.94, 2.83	.082
Bilateral tonic–clonic seizures	With	.85	.49, 1.47	.552
Focal seizures with preserved consciousness	With	2.38	1.15, 4.92	.019
Seizures causing falls	With	.97	.60, 1.59	.911
Structural etiology	With	1.31	.74, 2.32	.351
Genetic etiology	With	.58	.32, 1.06	.075
Frequency of seizures		1.00	.997, 1.002	.489
Daily seizures	With	.58	.35, .97	.038
Number of ASMs		.81	.61, 1.06	.118
Vagal nerve stimulation	With	.96	.57, 1.63	.890
Comorbidities	With	1.09	.63, 1.89	.750
Biochemical
Glycemia		1.024	.998, 1.049	.067
Albumin		.64	.33, 1.24	.182
Total proteins		.52	.29, .93	.027
Ammonia		1.003	.996, 1.010	.430
Urea		.99	.96, 1.02	.627
Total cholesterol		1.005	.998, 1.011	.179
LDL cholesterol		1.003	.995, 1.011	.437
Triglycerides		1.003	.998, 1.007	.223
Dyslipidemia	With	1.31	.78, 2.20	.316
Uric acid		.84	.68, 1.04	.110
Vitamin B12		.999	.998, 1.000	.118
Folic acid		1.02	.96, 1.07	.561
Vitamin D		.998	.985, 1.011	.761
Caregiver, n = 59
Sex	Female	1.14	.51, 2.54	.743
Age, years		1.00	.97, 1.03	.945
Education, years		.99	.93, 1.06	.824

Note: For dichotomic variables, the HRs were calculated for the following: females in comparison to males; residents in urban areas in comparison to rural areas; patients with focal type of epilepsy in comparison to generalized type of epilepsy; and for patients with versus without a specific characteristic. Due to logistical issues, the following analytical data are missing: albumin (n = 1), ammonia (n = 2), cholesterol (n = 1), glycemia (n = 1), serum total proteins (n = 3), LDL (n = 1), triglycerides (n = 1), vitamin B12 (n = 3), vitamin D (n = 9), folic acid (n = 2), urea (n = 2), uric acid (n = 5). HRs are given for discontinuation.

Abbreviations: ASM, antiseizure medication; CI, confidence interval; HR, hazard ratio; LDL, low‐density lipoprotein.

FIGURE 1.

Kaplan–Meier curves of baseline predictors of modified Atkins diet (MAD) discontinuation (A, sex; B, type of epilepsy; C, etiology; D, presence of daily seizures; E, presence of focal seizures with preserved consciousness). Cum, cumulative.

3.4. MAD discontinuation after 3 months

Among patients with at least 3 months on MAD (n = 51), reduced clinical response at 3‐month follow‐up was associated with earlier discontinuation of MAD, as measured by lower percent seizure reduction (HR = .98, p = .004) and <50% seizure reduction (HR = 2.44, p = .027).

Qualitative benefits in seizures, such as reports of shorter seizure duration (HR = .20, p < .001), shorter postictal period (HR = .25, p < .001), and subjective improvement in cognition/mood and/or behavior (HR = .48, p = .032) at 3 months, were associated with longer persistence on MAD. Among those with falls at baseline and with 3‐month follow‐up (n = 31), a reduction in the number of falls was related with lower HR of dropout (HR = .39, p = .024).

Patients with lower levels of serum total proteins at baseline (HR = .29, p = .007), with dyslipidemia at baseline (HR = 2.25, p = .017) and at 3 months (HR = 2.07, p = .045), and with worsening of seizures at follow‐up (HR = 4.44, p = .003) had earlier dropout after 3 months. The median ketonemia measurement at 3 months was 1.905 (range = 3.2, Q1 = 1.60, Q3 = 2.50). Higher ketonemia measurements were related to shorter duration on MAD (HR = 1.65, p = .063; Table 4). No association was found between ketonemia and seizure reduction ≥ 50% (median = 1.94 vs. 1.91, p = .723) nor with subjective improvement in cognition/mood and/or behavior (median = 1.90 vs. 1.95, p = .520) at follow‐up. However, patients with any qualitative benefit in seizures had lower ketonemia measurements (median = 1.80 vs. 2.41, p = .091). Among patients with bilateral tonic–clonic seizures at baseline and with 3‐month follow‐up (n = 16), those with reduction in this type of seizure (4/16) had lower ketonemia (median = 1.29 vs. 1.93, p = .025). No other qualitative benefit in seizures was significantly related to ketone levels in the blood (p > .100).

TABLE 4.

Clinical and biochemical predictors at 3 months of modified Atkins diet discontinuation afterward (n = 51 patients).

Predictor		HR	95% CI	p
Quantitative response
Seizure frequency		.998	.995, 1.001	.288
Seizure reduction, %		.98	.97, .99	.004
Seizure reduction ≥ 50%	Yes	.41	.19, .90	.012
Qualitative response
Fewer rescue medications	Yes	.03	<.001, 735.42	.488
Shorter seizure duration	Yes	.20	.10, .40	<.001
Fewer bilateral tonic–clonic seizures	Yes	.03	<.001, 12.40	.244
Fewer seizures causing falls	Yes	.39	.17, .88	.024
Shorter postictal period	Yes	.25	.12, .50	<.001
Subjective improvement of cognition/mood/behavior	Yes	.48	.24, .94	.032
Side effects
Any	Yes	1.22	.62, 2.38	.565
Gastrointestinal	Yes	.73	.34, 1.57	.426
Lethargy	Yes	1.65	.68, 3.99	.268
Worsening of seizures	Yes	4.44	1.67, 11.78	.003
Glycemia	At 3 months	1.01	.98, 1.05	.521
	At baseline	1.01	.97, 1.04	.771
Total proteins	At 3 months	.73	.40, 1.32	.292
	At baseline	.29	.12, .72	.007
Dyslipidemia at 3 months	Yes	2.07	1.02, 4.23	.045
Dyslipidemia at baseline	Yes	2.25	1.16, 4.38	.017
Average ketonemia measurements		1.65	.97, 2.81	.063

Note: HRs are given for discontinuation.

Fewer patients with dyslipidemia at 3 months had seizure reduction ≥ 50% at follow‐up (12.5% vs. 52.2%, p = .004). The effect of baseline dyslipidemia on seizure reduction ≥ 50% at 3 months was not statistically significant (22.2% vs. 41.9%, p = .162).

4. DISCUSSION

4.1. Timing and motives of MAD discontinuation

In more than 6 years of experience at our dedicated center treating adult epilepsy patients with MAD, we observed that patients remained on average 122 days in treatment. Most of them discontinued diet in the first 6 months (55%), and the main reason was noncompliance.

The literature on MAD in adults with DRE is very heterogeneous in terms of study design, clinical setting, number of patients, and follow‐up times, factors that might influence the reported dropout rates. Even in controlled trials with short treatment periods of 2–6 months, ⁶ , ¹⁶ , ¹⁸ the dropout rate approaches 35%. In observational studies like ours, the dropout rate increases, with the follow‐up time reaching 66% at 12 months ¹⁹ and 71%–83% at 12–24 months. ²⁰ The Johns Hopkins center ¹⁵ reported a 56% dropout rate in the first 5 years of MAD treatment, which is lower than in our cohort. Differences in patient characteristics, organization, and available resources may partially explain this variability. Our cohort included a higher percentage of intellectually disabled patients, a lower age at epilepsy onset, and more seizures at baseline than the Johns Hopkins center cohort, suggesting more severe epilepsies.

Noncompliance was the main cause of MAD discontinuation in our cohort, and several reasons were given by patients/caregivers. In the literature, noncompliance is a broad category that encompasses different terms, such as “not supported by family,” ⁶ “did not adhere to MAD,” ¹⁶ “restrictiveness,” ¹⁵ , ¹⁹ , ²¹ “lack of motivation,” ²² “financial and logistical difficulties associated with adherence to the diet,” ²³ and “socio‐familial problems.” ²⁴

In our study, inefficacy was also a prevalent cause of dropout, and side effects were not a major reason to stop treatment. In accordance with the literature, ¹⁵ , ¹⁹ , ²² , ²⁵ the main side effect causing dropout was increased seizure frequency, which occurred early in the treatment. Possible reasons for worsening seizure frequency may include interactions with ASM, variations in compliance with ASM, or spontaneous fluctuations in underlying seizure frequency. Consistent with another study, ¹⁵ diet tiredness was a cause of MAD late dropout despite some patients experiencing ≥50% seizure reduction.

Motives for discontinuation were related to ketonemia levels at the first month but not to presence of intellectual disability or comorbidities (including psychiatric comorbidity) at baseline. As expected, noncompliant patients had lower ketonemia levels in the first month.

4.2. Predictors of MAD discontinuation

4.2.1. Sociodemographic factors

Regarding potential sociodemographic predictors of MAD discontinuation, only sex was a significant predictor. Female patients dropped out of MAD earlier than male patients. This association between sex and MAD discontinuation remained significant even when relevant clinical variables were taken into account. Another study that explored the effect of sex on the retention rate of modified KD reported an opposite trend but failed to reach statistical significance. ¹⁹ This apparent incongruency may be related to cohort differences. For instance, self‐referral was not a possibility in our cohort. It has been argued that self‐referred women may be more motivated to undergo KD with the secondary goal of weight loss, ¹⁵ even though KD may be less effective in weight loss in women than in men due to hormonal and metabolic differences. ²⁶ The association between sex and MAD discontinuation found in our DRE cohort warrants further investigation in future studies.

4.2.2. Clinical factors

Regarding baseline clinical characteristics, focal epilepsy and nongenetic etiologies were predictors of MAD discontinuation. This association between type of epilepsy and MAD discontinuation may be related to poorer response to the diet. In a previous study of our cohort, it was reported that focal epilepsies were less common and genetic causes were more frequent in responders than in nonresponders. ²⁷ Notably, type of epilepsy did not influence dropout in another cohort. ¹⁹

Patients without daily seizures, particularly those with focal seizures with preserved consciousness, and with shorter disease duration were more likely to discontinue MAD, suggesting that less severe patients with a narrower margin to improve are less motivated to engage with the diet for longer periods and therefore are more prone to discontinue earlier. It has been argued that chronic epilepsy is a main motivator to engage with KD. ²⁰ In our cohort, intellectual disability did not have a significant effect on MAD discontinuation. However, a recent study conducted in children found that patients with development delay had higher frequency of discontinuation of KD at 3 months. ²⁸

As expected, reduced clinical response at month 3 was associated with earlier discontinuation of MAD. Reversely, qualitative benefits in seizures and subjective improvement in cognition/mood/behavior at month 3 were associated with longer persistence with MAD, emphasizing the wide range of outcomes that have been increasingly recognized and valued by professionals, patients, and caregivers. ²⁹

4.2.3. Biochemical factors

Regarding biochemical predictors, patients with lower levels of serum total proteins and higher glycemic levels at baseline spent fewer days on MAD. The effect of serum total protein levels at baseline on MAD persistence was present even in the subgroup of patients with at least 3 months of treatment, suggesting that higher protein intake prior to enrollment may potentiate adherence to MAD treatment for DRE. However, no pre‐MAD food diaries were available to confirm the relationship between protein intake and serum total protein levels. Furthermore, serum total protein levels may be affected by factors other than protein dietary intake (e.g., liver and kidney function). However, we should point out that all participants had normal liver and kidney function at baseline. The effects of the glycemic levels at baseline on MAD duration were less consistent.

In recent years, high‐protein low‐carbohydrate diets have been studied mainly in the field of cardiovascular health, due to their positive effects on reducing body weight in obese patients and improving glycemic control in individuals with type 2 diabetes. An increase in the protein content of the diet induces satiety mainly due to oxidation of amino acids fed in excess, and specific amino acids may also serve as precursors for specific neurotransmitters involved in appetite. ³⁰ , ³¹ We think these mechanisms associated with previous dietetic habits may potentiate persistence on MAD for longer periods. Dietetic habits are a complex topic that involves social and environmental influences on food choice, psychological influences on eating behavior, and eating behavior profiling. ³² This complexity may explain the difficulty in sustainably changing previous eating habits, regardless of the studied population. ³³ , ³⁴ , ³⁵

In our cohort, higher ketonemia measurements in the first 3 months were related to shorter duration on MAD. It has been suggested that in less restrictive variants of KD (e.g., MAD), reduced levels of glucose, rather than elevated ketones per se, are more relevant for diet effectiveness. ³⁶ Our data suggest that lower ketonemia targets may improve DRE patients' persistence on MAD without compromising its efficacy, namely, in quantitative and qualitative indicators of benefit in seizures at 3‐month follow‐up.

No patient was excluded from enrollment due to dyslipidemia, and none abandoned treatment due to dyslipidemia. However, the presence of dyslipidemia (both at baseline and at follow‐up) was predictive of earlier dropout in the subgroup of patients with at least 90 days on MAD, but not in the overall cohort, suggesting a delayed effect of dyslipidemia. It is noteworthy that patients with dyslipidemia at 3 months had poorer treatment response (i.e., fewer patients with dyslipidemia had seizure reduction ≥ 50%). Both cholesterol and ketone body synthesis occur in hepatocytes from the excess acetyl‐CoA, which is an oxidation product of fatty acid components of the ingested lipids. ³⁷ , ³⁸ Therefore, dysregulation of fat metabolism in the liver may constitute an interplay between the complex pathways of dyslipidemia pathogenesis and KD mechanisms of action in seizure control.

4.3. Limitations

Our study has several limitations. This is a unicentric observational longitudinal study, translating the first 6 years of clinical practice at a single dedicated center, limiting the generalizability of results.

Although the treatment protocol was maintained throughout the study, the results may have been affected by the learning curve of the treating team. With increased experience, the clinical team learned to better anticipate difficulties in implementing diet, deal with adverse events, and adjust to individual needs. The clinical characteristics of the enrolled patients varied across the recruitment period, with more severe patients included at first. The study protocol required stability of ASM during the first 3 months on MAD. Any change in ASM after that timepoint was made according to the individual clinical need. Potential effects of ASM changes after 3 months on the studied outcomes were not properly explored in this study.

Treatment compliance was evaluated through patient or caregiver report in regular interviews (without a systematic food diary), which is susceptible to recall bias. Ketonemia levels provided an additional indicator of compliance.

Seizure frequency and characteristics of the seizures at baseline and follow‐up were based on patient/caregivers' calendars, which are vulnerable to inaccuracies; this is a common limitation of this type of study. Nonseizure benefits (i.e., in cognition, mood, and/or behavior) were based on subjective report and were not addressed individually. The use of objective measures would be ideal, although highly challenging in practice, given the wide variability in clinical severity (e.g., intellectual ability at baseline).

A broad list of variables potentially influencing the persistence on MAD was explored in a relatively small cohort, increasing the risk of type I and II statistical errors. Multiple subgroup analyses were performed to address relevant questions; however, the results ought to be interpreted with caution due to the small sample size. On the other hand, potentially relevant factors were insufficiently explored, such as economic income, social and family context, type of psychiatric comorbidity, specific motives for noncompliance (i.e., only the main motive for treatment dropout was recorded, and only broad categories were considered in this study), and previous dietetic habits.

5. CONCLUSIONS

The dropout rate of MAD in adults with DRE is high, even at dedicated centers. Most patients discontinue diet in the first 6 months, mainly due to noncompliance. In our cohort, female had earlier MAD discontinuation. Factors suggesting less severe epilepsies (i.e., not having daily seizures, having focal seizures with preserved consciousness, and shorter disease duration) and previous carbohydrate‐based diet (i.e., lower baseline proteinemia and higher baseline glycemia) were associated with earlier dropouts. Lower efficacy at 3 months and epilepsy characteristics possibly correlating with poorer response (i.e., focal epilepsies and nongenetic etiologies) were also correlated with shorter persistence on the diet. These findings suggest that patients with such characteristics should have closer monitoring in specialized ketogenic diet clinics to prevent MAD discontinuation before the minimum period necessary to evaluate response. In sum, the identification of motives and potential predictors of MAD discontinuation contributes to improvement of patient selection and promotion of treatment adherence. This largely overlooked topic in the literature ought to be addressed in larger studies.

AUTHOR CONTRIBUTIONS

Conceptualization: Raquel Samões, João Chaves, Sara Cavaco, and Bárbara Leal. Clinical methodology: Raquel Samões, Ana Cavalheiro, Catarina Teixeira, and Maria Manuel Tavares. Data curation: Raquel Samões and Sara Cavaco. Statistical analysis: Sara Cavaco. Writing—original draft preparation: Raquel Samões and Sara Cavaco. Writing—review: All authors. Funding acquisition: Raquel Samões, João Chaves, and Bárbara Leal. Supervision: Sara Cavaco, João Chaves, and Bárbara Leal. All authors read and approved the final version of the manuscript.

FUNDING INFORMATION

This study was supported by a Liga Portuguesa contra a Epilepsia 2020 scientific grant and Centro Hospitalar Universitário do Porto 2020 grant for investigation projects. The Unit for Multidisciplinary Research in Biomedicine is funded by FCT Portugal (grant numbers UIDB/00215/2020 and UIDP/00215/2020) and ITR ‐ the Laboratory for Integrative and Translational Research in Population Health by a BICE Tecnifar grant (LA/P/0064/2020).

CONFLICT OF INTEREST STATEMENT

R.S. has served as a paid consultant for Nutricia and Glutamine. R.S., A.C., and M.M.T. have received support from Nutricia for participating in educational courses about ketogenic diet. A.C. and M.M.T. have received support from Glutamine for participating in educational courses about ketogenic diet. The remaining authors have no conflicts of interest. 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.

Supporting information

Supporting Information S1 Semistructured clinical interview applied to patients and/or caregivers before initiation of the diet and at every 3 months after initiation of the diet.

Supporting Information S2 Characterization of the response to modified Atkins diet in the 12 patients who discontinued treatment because of diet tiredness, at the time of diet withdrawal.

Samões R, Cavalheiro A, Tavares MM, Teixeira C, Leal B, Chaves J, et al. Motives and predictors of modified Atkins diet discontinuation as treatment of adults with drug‐resistant epilepsy. Epilepsia. 2026;67:199–212. 10.1111/epi.18657

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.