Evaluating the therapeutic efficacy of a modified ketogenic diet in children with autism spectrum disorder: a randomized controlled trial

Abstract

Background

Autism spectrum disorder (ASD) is a heterogeneous neurodevelopmental disorder marked by persistent impairments in social communication and interaction, together with restricted and repetitive behavior or interests. Recent epidemiological data indicate that the prevalence of ASD has continued to increase worldwide. Current behavioral and pharmacological interventions remain limited in efficacy, particularly for associated comorbidities. In this context, metabolic interventions, especially the ketogenic diet (KD), have attracted growing interest as a potential therapeutic strategy.

Methods

A total of 62 young children with ASD admitted to Hefei Maternal and Child Health Hospital between January 2024 and January 2025 were prospectively enrolled and randomly assigned to a treatment group or a control group (31 per group). Both groups underwent hospital-based rehabilitation training on weekdays and home-based rehabilitation on weekends over a 2-month period. The treatment group additionally received a modified KD, comprising ketogenic nutritional powder and a structured dietary plan, with lunch provided at the hospital on training days while breakfast, dinner, weekend dietary implementation, and home monitoring were managed by caregivers. The control group followed a regular diet along with the same rehabilitation program. Changes in ASD-related symptoms were assessed using the Autism Behavior Checklist (ABC) and Childhood Autism Rating Scale (CARS).

Results

After two months of intervention, both groups showed significant reductions in ABC and CARS scores compared with baseline (P < 0.05). Between-group comparison of individual change scores showed larger reductions in the treatment group for both ABC (U = 324.000, P = 0.027) and CARS (U = 273.500, P = 0.003).

Conclusions

In this cohort of young children with ASD, the modified ketogenic diet combined with rehabilitation training was associated with greater short-term reductions in ABC and CARS scores than rehabilitation alone. The intervention demonstrated acceptable short-term feasibility.

Trial registration

The Project was registered on China Clinical Trial Registry (ChiCTR) with the identifier ChiCTR2300075057 ,Registered 24 August 2023, https://www.chictr.org.cn/.

Introduction

Autism Spectrum Disorder (ASD) is a behaviorally defined neurodevelopmental disorder characterized by persistent deficits in social communication and interaction, along with restricted and repetitive patterns of behavior, interests, or activities [1]. The global prevalence of ASD is estimated to range from 1.0% to 1.5% [2]. Over the past few decades, the reported prevalence of ASD has increased substantially, and current estimates suggest that approximately 1 in 100 children worldwide are affected [3]. This increase is thought to reflect greater clinical awareness, revisions to diagnostic criteria, and the combined influence of genetic and environmental factors [4].

Growing evidence suggests that ASD is accompanied, in at least a subset of children, by measurable alterations in metabolic and nutritional profiles, including abnormalities in amino acid metabolism, lipid handling, trace elements, and oxidative stress-related pathways [5]. These findings have been reported not only in Western populations but also in regional studies from Asia and the Middle East. For instance, a Chinese study reported altered whole-blood trace element profiles in children with ASD, particularly involving zinc-related changes [6], while Jordanian studies have described elemental abnormalities in blood and urine as well as altered heavy metal and trace element patterns in scalp hair samples from children with ASD [7]. Such observations do not establish a single biochemical signature for ASD, but they support the view that nutritional and metabolic dysregulation may contribute to clinical heterogeneity and provide a rationale for testing diet-based interventions. Dietary strategies that modify energy metabolism and nutrient composition, including ketogenic approaches, have attracted increasing interest as adjunctive interventions in ASD.

The ketogenic diet (KD) was first proposed by Dr. Russel Wilder in 1921. It is a high-fat, moderate-protein, and low-carbohydrate dietary regimen that mimics the metabolic state of fasting. Under this condition, the body shifts from carbohydrates to fat as its primary energy source, leading to the production of large amounts of ketone bodies. Through this metabolic shift, the KD may exert beneficial effects on energy metabolism and is regarded as a specialized nutritional therapy [8]. The ketogenic diet has been reported to be effective and safe, and can be used synergistically with other therapeutic options [9]. The KD has established clinical efficacy in drug-resistant epilepsy, and modified ketogenic approaches have more recently been explored as potential adjunctive interventions in ASD [10]. However, adherence to the KD remains challenging for many patients and their families because of its restrictive nature. To address this issue, modified ketogenic diets, such as medium-chain triglyceride (MCT)-enriched and gluten-free ketogenic diets, have been explored as potential alternatives for managing symptoms in individuals with ASD. Despite these promising findings, the evidence supporting these modified approaches remains limited and warrants further investigation [11].

ASD is a lifelong neurodevelopmental condition for which no definitive cure is currently available. Its management is further complicated by factors such as family socioeconomic status, parental psychological well-being, and the broader social environment [12]. Currently, risperidone and aripiprazole are the only medications approved by the U.S. Food and Drug Administration for the treatment of behavioral symptoms in ASD, although their use may be associated with adverse effects [11]. Non-pharmacological interventions for ASD are generally classified into three categories: (1) relationship-based interventions, such as Relationship Development Intervention (RDI) and Floor Time; (2) skill-based interventions, including the Picture Exchange Communication System (PECS), Discrete Trial Training (DTT), Social Stories, and Peer-Mediated Interventions; and (3) comprehensive intervention models, which target multiple developmental domains and individualized treatment goals. Although these approaches are widely applied, their efficacy is often limited, and early intervention remains critical for improving outcomes [12]. Given these challenges, further research into the etiology and treatment of ASD remains essential to address the growing global burden of the disorder.

In this prospective randomized controlled trial, we evaluated whether adding a modified ketogenic diet to rehabilitation training would result in greater short-term improvement in ABC and CARS scores in young children with ASD than rehabilitation training alone.

Methods

Participants

In this prospective randomized controlled study, 62 children with ASD admitted to the Children’s Neurological Rehabilitation Center of Hefei Maternal and Child Health Hospital between January 2024 and January 2025 were randomized in a 1:1 ratio to the treatment or control group (31 participants per group). The allocation sequence was generated by an investigator not involved in participant recruitment or outcome assessment using a random-number table. Group assignments were placed in sequentially numbered, opaque, sealed envelopes and revealed only after participant enrollment. Participants were enrolled by the attending clinicians and assigned to the intervention by a designated study coordinator according to the concealed sequence. Simple randomization was used without blocking or stratification. The sample size was calculated for a two-group comparison assuming a standardized effect size of 0.80, a two-sided significance level of 0.05, and 80% statistical power. A minimum of 26 participants per group was required. Allowing for an anticipated dropout rate of approximately 15%, the target sample size was increased to 31 participants per group, resulting in a total planned enrollment of 62 participants.

Time since ASD diagnosis was defined as the interval, in months, between the first documented clinical diagnosis of ASD according to DSM-5 criteria and study enrollment. The inclusion criteria were as follows: (1) diagnosis of ASD according to DSM-5 criteria by a qualified clinician at the Children’s Neurological Rehabilitation Center of Hefei Maternal and Child Health Hospital [13]; (2) normal findings on routine blood and urine tests, blood glucose and lipid assessments, liver and renal function tests, and ultrasonography of the liver, gallbladder, pancreas, and spleen; (3) age 2–5 years; (4) no previous ketogenic diet treatment; (5) written informed consent provided by the participant’s legal guardian; (6) willingness of the child and guardian to comply with the randomized study protocol; and (7) no use of psychotropic medications or nutritional or metabolic supplements that could influence metabolic status during the study period. The exclusion criteria were as follows: (1) severe gastrointestinal, cardiovascular, respiratory, hepatic, urinary, genetic metabolic, or congenital immunodeficiency disorders that contraindicated KD treatment; (2) severe vomiting; (3) fever or active infectious disease; (4) other established contraindications to KD treatment; and (5) organ transplantation, stem cell therapy, or transfusion of allogeneic blood products within the previous 6 months, or the presence of a hemolytic disorder. This trial did not include predefined stratification by epilepsy comorbidity, developmental quotient or intellectual level, or functional subtype, nor were these characteristics systematically incorporated into baseline comparisons or outcome analyses.

Procedure

The treatment group received a modified ketogenic diet consisting of Chi Ketone Nutritional Powder (hydrolyzed whey protein type), which was supplied free of charge by Shenzhen Zeneca Biotechnology Co., Ltd., together with a structured intervention diet. Per 100 g, the nutritional powder contained 50.7 g fat, 14.3 g protein, 8.9 g carbohydrate, and 17.9 g dietary fiber, corresponding to 78%, 10%, 6%, and 6% of energy, respectively. The ketogenic ratio [fat: (protein + carbohydrate)] was approximately 2:1. The MCT content was approximately 23.5 g per 100 g, accounting for 44.1% of total fat and 32.6% of energy. Each sachet weighed 17.5 g and provided approximately 100 kcal. Participants followed a gradual dose-escalation regimen, starting with half a sachet three times daily. The dose was progressively increased within one week according to individual tolerance to one sachet three times daily, providing approximately 10%-40% of the total daily energy intake. The remaining energy requirements were provided by the ketogenic diet. The intervention diet restricted the consumption of staple foods, high-starch items, and sugary foods during breakfast and dinner while ensuring a balanced intake of essential nutrients. For lunch, which was provided in the hospital, participants were encouraged to limit or avoid rice, noodles, steamed buns, and other carbohydrate-rich grain-based foods, and instead focus on vegetables, meat, and eggs. Throughout the observation period, no additional rehabilitation programs or extended training times were introduced. Urine ketone and glucose levels were regularly monitored and recorded. Caregivers were also instructed to record body weight and relevant clinical symptoms. Gastrointestinal symptoms and adverse events, including diarrhea, vomiting, gastroesophageal reflux, rash, hypoglycemia, excessive ketosis, and dehydration, were monitored throughout the intervention period. During the intervention period, participants followed a hybrid hospital-home management model. On weekdays, children attended hospital-based rehabilitation sessions and received lunch there, while breakfast, dinner, weekend dietary plans, rehabilitation exercises, and home monitoring were carried out under caregiver supervision.

The control group received a regular diet combined with the same hospital-to-home rehabilitation schedule. From Monday to Friday, participants attended hospital-based rehabilitation sessions, and on weekends, they performed home-based exercises under caregiver supervision. Outcome assessors administering the ABC and CARS were blinded to treatment allocation throughout the study. Participants and caregivers were instructed not to disclose dietary assignment during assessments, and treatment-allocation information was not available to the assessors. Blinding was maintained until completion of the end-of-treatment assessments. Owing to the nature of the dietary intervention and the need for diet-specific monitoring in the treatment group, blinding of participants, caregivers, dietitians, and treating physicians was not feasible. Both groups received the same standardized rehabilitation framework throughout the 8-week study period, including verbal-cognitive training, guided education, sensory integration training, and transcranial magnetic stimulation. Rehabilitation was delivered from Monday to Friday, with comparable session frequency and total training time between groups. Individual adjustments were permitted according to each child’s functional needs but did not alter the prescribed overall rehabilitation intensity or duration. No systematic between-group differences in rehabilitation exposure were introduced during the study. Training sessions were conducted from Monday to Friday, with home-based rehabilitation exercises on weekends, for a total intervention duration of two months.

In this study, the KD intervention was implemented over an 8-week period and consisted of three phases: a baseline period (1 week), a titration period (3 weeks), and an efficacy observation period (weeks 4–8). The diet was introduced gradually. The test sample, a ketogenic nutritional powder, was increased stepwise from 10% to 40% of total daily caloric intake over 5 days, while the remaining calories were provided by the participants’ usual diet. Daily caloric intake was determined according to the recommended dietary intake for Chinese residents. Both the treatment and the control groups received a similar rehabilitation training protocol, with the modified ketogenic diet being the only difference between the two groups. The macronutrient composition was carefully adjusted to ensure comparability between groups. The KD induces a metabolic state in which energy is derived primarily from fat metabolism, leading to the production of ketone bodies and potential “ketosis.”

During the intervention period, dietitians provided standardized instructions for home-based metabolic monitoring, including fingertip blood glucose testing with glucometers and urinary ketone assessment using reagent strips. The monitoring regimen followed a phased approach: daily evaluations during the baseline phase (7 days), weekly assessments during the 3-week titration phase, and regular monitoring throughout the efficacy observation period from weeks 4 to 8. Clinical staff (dietitians and physicians) conducted systematic reviews of parental documentation, cross-referencing ketone chart records with corresponding glycemic measurements to verify dietary adherence. Protocol-specified thresholds triggered clinical interventions, whereby undetectable ketonuria or aberrant glucose levels necessitated immediate dietary recalibration through practitioner-parent consultations. In this study, all blood glucose levels remained within normal ranges, and ketones were detected on day 4 of the baseline phase (Fig. 1).

Fig. 1.

Urinary ketones in participants (n = 31) eating a KD (Day 1–56, x-axis). Each bar (100%, y-axis) corresponds to all 31 participants

Adverse reactions, including hypoglycemia, excessive ketosis, and gastrointestinal disturbances, were closely monitored and recorded. During the initiation period, participants received daily follow-ups from a physician or dietitian, either by telephone or through in-person visits, to ensure safety and adherence.

To support growth and development during KD treatment, caloric and nutrient intake was carefully calculated. (1) Daily Caloric Intake: Total daily energy intake was determined according to the child’s age, height, and weight, corresponding to 75–100% of the recommended caloric intake for children of the same age with standard height and weight. For overweight children, caloric intake was adjusted according to the standard body weight for age. When catch-up growth was required (e.g., in infants and young children), caloric adjustments were made based on the child’s metabolic rate and activity level to ensure optimal growth. (2) Protein Intake: Protein requirements were calculated to meet the recommended daily intake for healthy children of the same age, ensuring adequate support for growth and development.

Study Implementation and Monitoring: A registered dietitian provided a 2-hour training session for family caregivers to ensure the proper implementation of the modified KD program. Throughout the study, the dietitian closely monitored each child’s dietary adherence. A multidisciplinary team, including a dietitian, specialist nurse, and two neurorehabilitation physicians, provided continuous oversight and support to all participants. The intention-to-treat (ITT) population included all 62 randomized participants (31 in each group). For participants without an end-of-treatment ABC or CARS assessment, the last available score was carried forward to the 2-month endpoint. Baseline ABC/CARS scores, time since ASD diagnosis, age, and sex were complete, and no covariate imputation was required. A sensitivity analysis was also performed among participants who completed the 2-month intervention and end-of-treatment assessment (treatment group, n = 24; control group, n = 27) (Fig. 2).

Fig. 2.

Flow chart

Instruments

Autism Behavior Checklist (ABC): The ABC is a screening and assessment tool used to aid in the identification of ASD in individuals aged 18 months to 35 years. It consists of 57 items across five domains: sensory (9 items), relational (12 items), stereotyped behaviors and object use (12 items), language (13 items), and self-help and social skills (11 items). Each item is scored on a scale of 1 to 4, reflecting the significance of the behavior in relation to ASD-related symptoms. In this study, enrolled children were assessed using the ABC scale once before and once after the intervention. A total score of 53 or higher was used as the diagnostic threshold for ASD [14].

Childhood Autism Rating Scale (CARS): The CARS is a quantitative and objective tool developed by Schopler et al. in 1988. It is used to differentiate ASD from other developmental disorders and to evaluate the severity of ASD symptoms. The CARS consists of 15 items, each scored from 1 to 4 based on the degree of symptom severity, yielding a total possible score of 60. The assessment incorporates both therapist observations and parental reports, with the evaluator determining the extent to which the child’s behavior deviates from that of a typically developing child of the same age. In this study, the children were assessed using the CARS scale once before and once after the intervention. A cutoff score of 30 was used to diagnose ASD, with scores between 30 and 36 indicating mild-to-moderate autism and scores above 37 indicating severe autism [15]. The ABC and CARS were selected as outcome measures because both are established instruments for assessing ASD-related behavioral characteristics and symptom severity and have demonstrated acceptable psychometric properties in clinical populations. The ABC was used to capture changes across multiple symptom domains, whereas the CARS was used to reflect overall autism severity.

Outcome assessment

Changes in ASD-related symptom scores were assessed using the ABC and CARS. Both scales were administered once before the intervention and once at the end of the two-month treatment period in both groups. Within-group changes from baseline to the end of treatment were assessed using the Wilcoxon signed-rank test, whereas between-group comparisons were performed on individual change scores, calculated as baseline minus post-treatment score, using the Mann-Whitney U test.

Statistical analysis

SPSS 25.0 software was used for statistical analysis. Continuous variables were summarized as mean ± SD or median (IQR), as appropriate, and categorical variables were summarized as counts. ABC and CARS scores at baseline and post-treatment, together with individual change scores, were summarized as median and interquartile range (IQR), with ranges reported for change scores. For between-group comparisons of change scores, rank-biserial correlation was reported as the effect estimate with 95% bootstrap confidence intervals based on 10,000 resamples. The primary efficacy analysis for ABC and CARS used baseline-adjusted linear regression/ANCOVA models, with the post-treatment score as the dependent variable, treatment group as the main independent variable, and the corresponding baseline score, time since ASD diagnosis, age, and sex as covariates. The primary ANCOVA analyses were performed in the ITT population (n = 62), whereas within-group pre-post comparisons and between-group comparisons of change scores were considered supportive analyses; analyses restricted to participants who completed the end-of-treatment assessment (n = 51) were performed as sensitivity analyses. To assess the robustness of the findings to the handling of missing outcomes, a multiple-imputation sensitivity analysis was also performed for missing end-of-treatment ABC and CARS scores using chained equations with 50 imputed datasets. The imputation model included treatment group, age, sex, time since ASD diagnosis, baseline ABC and CARS scores, and available outcome data, and estimates were combined according to Rubin’s rules. A two-sided P < 0.05 was considered statistically significant.

Results

Comparison of baseline characteristics between groups

A total of 62 children met the inclusion criteria and were initially enrolled in the study. Baseline age and sex distribution were comparable between the two groups (P > 0.05). However, time since ASD diagnosis was significantly longer in the control group than in the treatment group (P < 0.001) (Table 1).

Table 1.

Baseline characteristics of children in the two groups

Variable	Treatment group (n = 31)	Control group (n = 31)	Statistic	P
Age, year	3.35 ± 0.95; median 3 (IQR 3–4), range 2–5	3.68 ± 0.94; median 4 (IQR 3–4), range 2–5	U = 386.5	0.165
Time since ASD diagnosis, months	5.76 ± 6.97; median 1.5 (IQR 1.0–12.0), range 0.0–24.0	16.58 ± 8.20; median 20.0 (IQR 12.0–24.0), range 0.0–30.0	U = 188.5	< 0.001
Sex, male/female	18/13	16/15	χ² = 0.261	0.610

Comparison of ABC scores before and after intervention

In the ITT population (31 participants per group), ABC scores decreased from a median of 67 (IQR, 62.5–69.5) to 56 (IQR, 55.0-60.5) in the treatment group and from 66 (IQR, 62.0–69.0) to 59 (IQR, 57.5–64.0) in the control group; both within-group changes were significant (both P < 0.001). Individual ABC change scores (baseline minus post-treatment) were 9 (IQR, 4.5–12.0; range, 0–21) in the treatment group and 4 (IQR, 3.0–8.0; range, 0–14) in the control group. The Mann-Whitney U test comparing these change scores showed greater improvement in the treatment group (U = 324.000, P = 0.027; rank-biserial correlation = 0.326, 95% CI, 0.032–0.605) (Table 2). In the baseline-adjusted analysis including baseline ABC score, time since ASD diagnosis, age, and sex as covariates, the treatment group remained significantly associated with a lower post-treatment ABC score than the control group (β = -4.12, 95% CI: -6.65 to -1.60, P = 0.001). The multiple-imputation sensitivity analysis yielded a consistent treatment effect for ABC (β = −4.44, 95% CI: −8.37 to − 0.51, P = 0.027).

Table 2.

Comparison of ABC scores before and after intervention between the two groups

Group	n	Baseline ABC, median (IQR)	Post-treatment ABC, median (IQR)	Change score, median (IQR); range	Z	P
Treatment group	31	67 (62.5–69.5)	56 (55.0-60.5)	9 (4.5–12.0); 0–21	-4.294	< 0.001
Control group	31	66 (62.0–69.0)	59 (57.5–64.0)	4 (3.0–8.0); 0–14	-4.555	< 0.001

Between-group comparison of change scores: U = 324.000, P = 0.027; rank-biserial correlation = 0.326 (95% CI, 0.032–0.605)

Comparison of CARS scores before and after intervention

In the ITT population (31 participants per group), CARS scores decreased from a median of 36 (IQR, 34.5–37.5) to 32 (IQR, 31.0–34.0) in the treatment group and from 35 (IQR, 34.0-36.5) to 34 (IQR, 32.0–35.0) in the control group; both within-group changes were significant (both P < 0.001). Individual CARS change scores (baseline minus post-treatment) were 4 (IQR, 2.0-5.5; range, 0–7) in the treatment group and 2 (IQR, 1.0–2.0; range, 0–6) in the control group. The Mann-Whitney U test comparing these change scores showed greater improvement in the treatment group (U = 273.500, P = 0.003; rank-biserial correlation = 0.431, 95% CI, 0.145–0.693) (Table 3). In the baseline-adjusted analysis including baseline CARS score, time since ASD diagnosis, age, and sex as covariates, the treatment group remained significantly associated with a lower post-treatment CARS score than the control group (β = -1.33, 95% CI: -2.38 to -0.27, P = 0.014). The multiple-imputation sensitivity analysis similarly showed a consistent treatment effect for CARS (β = −1.83, 95% CI: −3.19 to − 0.47, P = 0.008).

Table 3.

Comparison of CARS scores before and after intervention between the two groups

Group	n	Baseline CARS, median (IQR)	Post-treatment CARS, median (IQR)	Change score, median (IQR); range	Z	P
Treatment group	31	36 (34.5–37.5)	32 (31.0–34.0)	4 (2.0-5.5); 0–7	-4.307	< 0.001
Control group	31	35 (34.0-36.5)	34 (32.0–35.0)	2 (1.0–2.0); 0–6	-4.527	< 0.001

Between-group comparison of change scores: U = 273.500, P = 0.003; rank-biserial correlation = 0.431 (95% CI, 0.145–0.693)

Follow-up and dropout analysis

Of the 62 children initially enrolled, 51 completed follow-up and assessments, yielding a completion rate of 82.26%. 11 children were lost to follow-up, corresponding to a dropout rate of 17.74%. Among those who discontinued the trial, three participants in the treatment group withdrew because of noncompliance with the dietary protocol, and two others in the treatment group terminated participation due to adverse reactions. The incidence of adverse reactions in the treatment group was 6.45% (Table 4). No hypoglycemia, hyperlipidemia, dehydration, vomiting, or gastroesophageal reflux was observed during the 8-week intervention. Caregiver monitoring did not identify clinically apparent weight loss during the treatment period.

Table 4.

Follow-up dropout analysis

group	reason	numbers	incidence rate
treatment group (n1 = 31)	dietary protocol noncompliance	3	9.68%
	Failure to undergo treatment as scheduled	1	3.23%
	Failure to attend follow-up as scheduled	1	3.23%
	Diarrhea (adverse reactions)	1	3.23%
	Rash (adverse reactions)	1	3.23%
control group (n2 = 31)	Failure to undergo treatment as scheduled	2	6.45%
	Failure to attend follow-up as scheduled	2	6.45%
total(N = 62)		11	17.74%

Discussion

This study showed that a modified ketogenic nutritional powder combined with an intervention diet produced greater short-term reductions in ABC and CARS scores than the control regimen. The observed differences were limited to short-term changes in ABC and CARS scores, while the intervention remained feasible for a considerable proportion of families during follow-up. Earlier ketogenic studies in ASD have often been constrained by small sample size, heterogeneous dietary protocols, and a strong dependence on caregiver execution [11–20]. Our data support that a simplified ketogenic strategy can be implemented in daily care without the full preparation burden of a classic food-based KD. The intervention used in this study was not a strict classic 3:1 or 4:1 protocol, which may have facilitated short-term tolerability and adherence, particularly because family routines, feeding rigidity, and rehabilitation schedules can strongly influence whether a diet can be maintained [17, 21–23].

The direction of change observed in our cohort aligns with earlier clinical reports on ketogenic interventions in ASD. Evangeliou et al. described behavioral improvements in a subset of children who adhered to the diet for a sufficient duration [16]. El-Rashidy et al. found that both a modified Atkins ketogenic regimen and a gluten-free casein-free diet improved ATEC and CARS scores, with greater gains in certain cognitive and sociability domains in the ketogenic group [20]. Lee et al. further demonstrated that a modified gluten-free ketogenic diet supplemented with medium-chain triglycerides enhanced core ASD-related behaviors, supporting the clinical promise of less restrictive ketogenic variants [11]. Meta-analyses have not established ketogenic therapy as a standard treatment for ASD, but pooled evidence suggests that selected dietary interventions can yield measurable benefits in at least some symptom domains and in carefully selected patients [18, 19]. Our findings suggest that measurable changes in symptom scores can be achieved with a formulation designed to reduce implementation burden. This distinction carries practical significance, as a diet with theoretical efficacy may fail in real-world settings if families cannot maintain it.

Although the treatment group showed larger reductions in ABC and CARS scores, the clinical relevance of these changes remains uncertain. Established minimal clinically important differences for short-term changes in ABC and CARS scores are lacking, particularly in young children with ASD, and observed score changes may also be influenced by measurement variability. As both instruments were originally developed to characterize ASD-related behaviors and symptom severity, future studies incorporating outcome measures specifically validated for treatment responsiveness would help determine whether the observed statistical differences translate into meaningful clinical benefit.

Diet tolerability often determines whether dietary studies in autism succeed or fail. The retention rate in our ketogenic arm fell within the range reported in earlier ASD ketogenic cohorts and feasibility work, where adherence has been highly individualized and strongly shaped by child-specific sensory and behavioral features [16, 17]. In our study, dietary intolerance and early non-compliance accounted for most withdrawals from the intervention arm [16, 17]. Feeding problems occur far more frequently in children with ASD than in typically developing peers, and meta-analytic data indicate higher rates of feeding difficulty alongside lower intake of some nutrients, including calcium and protein [22]. Food selectivity may persist over time, narrow dietary variety, and affect both family mealtimes and parental stress [21, 23]. These features make strict ketogenic therapy difficult to sustain, even among highly motivated families. Our data suggest that a powder-supported protocol may improve acceptability, yet this approach does not eliminate the need for close dietary counseling during the early adaptation period.

Children with ASD exhibit a substantially higher burden of gastrointestinal symptoms than controls, and these symptoms can interact with irritability, sleep, behavior, and caregiver-reported daily functioning [24]. A simplified diet that does not account for gastrointestinal tolerance is unlikely to succeed in this population. The formulation used in our study may have offered practical advantages because it combined a moderate ketogenic ratio with medium-chain triglycerides, hydrolyzed whey protein, fiber, and micronutrient fortification, which may be more acceptable to some children than a rigid classic ketogenic meal plan. Previous work has shown that modified ketogenic regimens containing medium-chain triglycerides can support behavioral improvements in ASD while allowing greater dietary flexibility [11]. Preclinical evidence also indicates that ketogenic treatment can reshape gut microbial composition in ASD-like models, suggesting that part of the clinical response may be mediated through gut-related pathways [25]. The low incidence of diarrhea and the absence of a broader gastrointestinal safety signal suggest that the present regimen achieved a workable balance between ketosis-oriented design and digestive tolerability.

Ketogenic therapy alters brain amino acid handling and can shift glutamate metabolism toward greater GABA availability, a mechanism first explored in epilepsy research and also relevant to ASD, where excitatory-inhibitory imbalance has long been proposed [26, 27]. Human electrophysiological studies have shown that short-term ketogenic exposure increases cortical inhibitory tone, providing a physiological basis for behavioral change beyond seizure control [28]. Mitochondrial dysfunction, including abnormalities in respiratory chain activity and markers of inefficient energy metabolism, has also been described in a subset of individuals with ASD [29, 30]. Neuroinflammatory signaling represents another candidate pathway, with elevated NF-κB expression reported in postmortem ASD brain tissue [31]. Ketone metabolism may partially counter this burden by improving cellular energy handling and dampening inflammatory signaling; PPARγ-related anti-inflammatory effects have been demonstrated in experimental ketogenic models [32]. Our study did not measure GABA, ketone concentrations, lactate, pyruvate, cytokines, or mitochondrial markers. The mechanisms underlying the observed score changes remain uncertain. Existing evidence raises the possibility that ketogenic interventions may influence ASD-related symptoms through coordinated effects on metabolic regulation, neurotransmitter balance, and inflammatory signaling.

Trials of vitamin and mineral supplementation in ASD have reported improved nutritional and metabolic status alongside benefits in selected clinical measures, suggesting that correcting biochemical imbalances can lead to symptomatic change in at least a subset of patients [33, 34]. Similarly, studies of levocarnitine and L-carnosine have shown improvement on certain behavioral scales, supporting the idea that targeted metabolic support may modify ASD manifestations when the treated pathway is relevant to the child [35, 36]. Biomarker studies provide a parallel line of evidence. In Jordanian children with ASD, blood and urine analysis revealed higher aluminum levels and lower calcium and manganese in blood, along with reduced urinary calcium [37]. Hair analysis in severe cases from the same population showed elevated aluminum, lead, and potassium, with decreased magnesium and zinc [7]. Amino acid studies have also described reduced essential or neutral amino acids and altered glutamate-related profiles in subsets of children with ASD, although the specific pattern varies across cohorts and specimen types [38–41]. These data support that ASD involves measurable metabolic heterogeneity. The key difference is that a ketogenic regimen does not function as a single-nutrient replacement strategy but instead alters substrate utilization, ketone availability, redox balance, and likely neurotransmitter flux simultaneously.

There are several limitations. A marked baseline imbalance in time since ASD diagnosis was observed despite random allocation. Although time since ASD diagnosis was included as a covariate in the adjusted analyses, residual confounding cannot be excluded, particularly if its association with treatment response is non-linear. The ketogenic group also received more intensive diet-specific education, monitoring, and clinical contact than the control group, and the observed between-group differences therefore cannot be attributed exclusively to the dietary intervention. First, the intervention period captured only short-term responses, so the durability of benefit after longer treatment or after diet withdrawal remains unknown. Second, attrition in both groups, particularly early non-compliance in the ketogenic arm reduces precision and may enrich the final analysis with families already more capable of sustaining dietary change. Third, because the study did not include serial measurements of ketone bodies, micronutrients, amino acids, inflammatory markers, mitochondrial markers, or gut microbiota, mechanistic inference is limited. Fourth, safety interpretation is confined to the observation window and cannot address later effects such as dyslipidemia, nephrolithiasis, slowed growth, or micronutrient depletion, which have been described with long-term ketogenic therapy. Despite these limitations, the present findings support further evaluation of modified ketogenic dietary interventions in young children with ASD. The findings included both changes in symptom scores and a short-term tolerability profile that appeared manageable for many families. Future studies should combine longer follow-up with biochemical phenotyping so that children most likely to respond can be identified before dietary treatment begins.

Conclusions

In this cohort of young children with ASD undergoing rehabilitation, the modified ketogenic diet was associated with greater short-term reductions in ABC and CARS scores than rehabilitation alone. The clinical significance and longer-term relevance of these score differences require further investigation. Because epilepsy comorbidity, developmental level, cognitive function, and functional subtype were not systematically characterized, the generalizability of these findings to high-functioning ASD or other ASD subgroups remains unclear.

Authors’ contributions

KYH, DW conceptualized the study design and supervised all research phases. LL coordinated participant recruitment, implemented therapeutic interventions, conducted longitudinal follow-up assessments, and performed statistical analyses. QSZ, JZ administered standardized evaluations using the Autism Behavior Checklist (ABC) and Child Autism Scale (CARS). ZL provided nutritional counseling for caregivers and monitored dietary compliance throughout the observation period. All authors critically reviewed the manuscript, approved the final version for publication, and assume collective responsibility for the integrity of this work.

Funding

This study was supported by the Hefei Health Commission Research Fund under the Health Applied Medicine Research Initiative (Award ID: Hwk2023zd025).

Data availability

The datasets used and analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in strict compliance with the ethical principles outlined in the Declaration of Helsinki. Written informed consent was obtained from all participating children’s legal guardians prior to their inclusion in the research. The study protocol received formal approval from the Institutional Review Board of Hefei Maternal and Child Health Hospital (Ethical Approval No.: YYLL2022-07-01-01) following rigorous ethical review procedures.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.