Ketogenic Diet and Neurological Health: A Narrative Review with Natural Language Processing-Assisted Thematic Mapping Using BERTopic

Abstract

Background

The ketogenic diet (KD) is an established therapy for drug-resistant epilepsy and is increasingly investigated across other neurological disorders.

Objective

To provide an updated narrative synthesis of recent mechanistic and clinical evidence on KD across neurological disorders, including its therapeutic applications, potential mechanisms, and current evidence gaps, and to use BERTopic as a supplementary natural language processing tool to characterize major research themes and temporal changes in research emphasis.

Methods

Literature published from 2021 to early 2026 was retrieved from Scopus, Web of Science, and PubMed. After deduplication and eligibility screening, 1017 records were included in BERTopic analysis. PubMedBERT, UMAP, HDBSCAN, and c-TF-IDF were used for thematic mapping; interpretation and evidence synthesis remained expert-guided. Temporal analyses were restricted to 2021–2025.

Results

Seven thematic clusters were retained, with increased representation of gut–brain-axis-related publications after 2023. Narrative synthesis highlighted energy metabolism, mitochondrial function, oxidative stress, neuroinflammation, neurotransmitter regulation, and the gut–brain axis as major mechanistic domains. Clinical evidence was strongest for drug-resistant epilepsy, while findings for Alzheimer’s disease and migraine were less mature and those for Parkinson’s disease remained preliminary.

Conclusion

KD is an established therapy for drug-resistant epilepsy and shows potential across other neurological disorders, particularly Alzheimer’s disease and migraine. Its effects may involve energy metabolism, mitochondrial function, oxidative stress, neuroinflammation, neurotransmitter regulation, and the gut–brain axis. BERTopic-assisted mapping complemented the narrative synthesis by summarizing major research themes and temporal trends.

Introduction

The ketogenic diet (KD), characterized by high fat, moderate protein, and very low carbohydrate intake, has evolved from a historical therapy for refractory epilepsy into a promising intervention for broader neurological health.¹ Originally introduced in the 1920s as an alternative to fasting, KD is now recognized for its ability to modulate critical neurophysiological pathways, including energy metabolism, inflammation, oxidative stress, and neurotransmitter balance.² Over the past decade, accumulating evidence has expanded its potential application to neurodegenerative diseases, neurodevelopmental disorders, mood disorders, and traumatic brain injury.³

Neurological disorders represent a significant global health burden and often lack effective disease-modifying treatments. The rising prevalence of these conditions, driven by aging populations and lifestyle changes, underscores the urgent need for non-pharmacological interventions that target underlying mechanisms rather than just symptoms.⁴ By shifting the body’s primary fuel source from glucose to ketone bodies, KD offers a unique metabolic strategy. Ketones serve as an efficient energy substrate for neurons, particularly in states of glucose hypometabolism common in neurological diseases.⁵ In addition, KD has been reported to influence pathways related to mTOR and Nrf2 signaling, mitochondrial function, neuroinflammation, neurotransmission, and gut–brain-axis communication.⁶ A substantial number of narrative reviews, systematic reviews, and meta-analyses have summarized the mechanisms, clinical efficacy, safety, and disease-specific applications of the KD across epilepsy, neurodegenerative disorders, migraine, and other neurological conditions. These studies provide an important evidence base, but they primarily focus on predefined mechanistic pathways or individual disease categories. An important gap therefore remains in understanding how the rapidly expanding KD literature is organized across mechanistic and disease-specific domains, how research emphasis has shifted over time, and where major clinical evidence gaps persist across neurological disorders. In particular, it remains unclear which mechanistic themes currently dominate the field, which neurological conditions have accumulated comparatively stronger or weaker clinical evidence, and which areas have received increasing research attention in recent years. Accordingly, this review addresses the question of how contemporary research themes, clinical applications, and evidence gaps in KD-related neurological research are distributed and evolving across the recent literature. Within the KD literature, BERTopic offers several complementary advantages to conventional narrative synthesis. KD research spans multiple neurological disorders and intersecting mechanistic domains, including energy metabolism, mitochondrial function, neuroinflammation, neurotransmission, and the gut–brain axis, making the recent literature difficult to organize within a single predefined framework.⁷ BERTopic enables large numbers of abstracts to be grouped according to semantic similarity, thereby providing a corpus-level overview of how disease- and mechanism-related themes are distributed across the literature. It also facilitates comparison of topic representation across publication years, which is useful for identifying shifts in research emphasis within this rapidly evolving field. Importantly, these advantages relate to literature organization and trend mapping rather than assessment of evidence quality; interpretation of mechanisms, clinical efficacy, and research gaps remained based on expert appraisal of the underlying studies. The present study was therefore designed as an updated narrative review supplemented by BERTopic-assisted thematic mapping, rather than as a replacement for existing systematic reviews or meta-analyses.⁸ More broadly, machine-learning-assisted literature synthesis should be viewed as complementary to, rather than a replacement for, conventional systematic reviews and expert-driven narrative synthesis, because algorithmic methods can facilitate large-scale thematic organization while critical appraisal and contextual interpretation remain dependent on domain expertise.⁹ BERTopic was applied as a supplementary, exploratory tool to characterize corpus-level thematic structure and temporal changes in research emphasis, while mechanistic and clinical interpretations remained based on expert appraisal of the underlying literature. Accordingly, this review had two primary objectives: (1) to provide an updated narrative synthesis of recent mechanistic and clinical evidence on KD across neurological disorders, including therapeutic applications and remaining evidence gaps; and (2) to use BERTopic-assisted thematic mapping to characterize major research themes and their temporal distribution.

Materials and Methods

Study Design

This study was conducted as a narrative review supplemented by BERTopic-assisted thematic mapping.⁸ Although a structured literature search and eligibility-screening procedure was used to construct the corpus for topic modeling, the review was not designed as a systematic review or meta-analysis, and no quantitative pooling of effect estimates was performed.BERTopic was applied as a supplementary, algorithm-assisted tool to support thematic organization and temporal mapping of the recent ketogenic diet literature.¹⁰ The resulting topic structure is influenced by methodological choices, including the embedding model, dimensionality-reduction parameters, clustering settings, corpus composition, topic-reduction decisions, and subsequent expert interpretation. Importantly, BERTopic did not determine which mechanisms, clinical findings, or evidence gaps were emphasized in the narrative synthesis; these decisions remained based on expert appraisal of the underlying literature. Accordingly, BERTopic was used as a descriptive organizational tool rather than as an unbiased, objective, or fully automated representation of the evidence.¹¹

Methodology Workflow

The methodological workflow consisted of literature retrieval, eligibility screening, document embedding, dimensionality reduction, topic clustering, topic representation, expert-guided topic refinement, and stability assessment, as summarized in Figure 1.

Figure 1.

Methodological workflow and record-screening process. (A) BERTopic-assisted text-analysis workflow, including literature retrieval from PubMed, Scopus, and Web of Science; PubMedBERT-based document embedding; UMAP dimensionality reduction; HDBSCAN clustering; and c-TF-IDF-based topic representation. (B) Flow diagram of record identification and screening. A total of 1156 records were retrieved from the three databases. After removal of 139 duplicates, 1017 unique documents were retained for the final BERTopic analysis.

Data Sources and Eligibility Criteria

To construct the corpus for BERTopic-assisted thematic mapping, we performed a structured literature search across three major bibliographic databases: PubMed, Scopus, and Web of Science. The search covered publications from 2021 to early 2026. Because data for 2026 were incomplete at the time of retrieval, temporal trend analyses were restricted to the complete calendar years 2021–2025. This recent time window was selected to characterize contemporary changes in KD research emphasis rather than to reconstruct the full historical development of ketogenic therapy. Earlier foundational studies were incorporated into the subsequent narrative synthesis when necessary to provide mechanistic and clinical context. The search strategy incorporated terms related to ketogenic dietary interventions and neurological disorders. Rather than relying solely on the broad term “neurological health”, we included specific neurological conditions to improve retrieval sensitivity. The target population comprised individuals with neurological disorders, including epilepsy, Alzheimer’s disease, Parkinson’s disease, migraine, neurodevelopmental disorders, and other relevant neurological conditions. Because this review also incorporated mechanistic and translational evidence, relevant animal and cellular models of neurological disease were also considered eligible. No restrictions were imposed on age or sex. The PubMed search strategy was: (“Ketogenic Diet”[Mesh] OR “Ketogenic Diets” OR “Low Carbohydrate Diet” OR “Ketosis” OR “Fasting”) AND (“Nervous System Diseases”[Mesh] OR “Neurology”[Mesh] OR “Neurological” OR “Neurodegenerative Diseases” OR “Alzheimer” OR “Parkinson” OR “Epilepsy” OR “Migraine” OR “Cognitive Dysfunction” OR “Neurodevelopmental”). Eligible records included peer-reviewed original clinical, preclinical, and translational studies, as well as review articles. Editorials, letters, case reports, and non-peer-reviewed publications were excluded. For transparency, the eligibility framework was further defined using an adapted PICO approach. The Population included individuals with neurological disorders and relevant preclinical disease models; the Intervention comprised KD and related ketogenic dietary strategies; Comparators included usual diet, standard care, alternative dietary interventions, placebo, or no intervention where applicable; and Outcomes included seizure control, cognitive and behavioral outcomes, neurological symptoms, mechanistic biomarkers, safety, tolerability, and treatment adherence. Because this was a narrative review incorporating preclinical, translational, and review literature, a formal comparator was not required for all eligible studies.

Deduplication and Screening

Records retrieved from the three databases were exported in RIS format and combined for screening. Deduplication was performed using BibExcel based on unique identifiers, including DOI, PMID, and EID, together with title-similarity matching. Potential duplicate records were subsequently reviewed manually by two independent reviewers to resolve ambiguities.¹² Following deduplication, records underwent structured eligibility screening according to the predefined inclusion and exclusion criteria. Of the 1156 initially retrieved records, 139 duplicates were removed, resulting in a final corpus of 1017 documents for BERTopic analysis. The record-identification and screening process is summarized in Figure 1B.

Document Embedding

Document embeddings were generated using PubMedBERT (Microsoft Research, 2021), a domain-specific BERT model pretrained on biomedical literature.¹³ PubMedBERT was selected because its biomedical pretraining is well suited to capturing semantic relationships among clinical, mechanistic, and disease-specific terminology within the KD literature.

Dimensionality Reduction

Because the document embeddings were high-dimensional, Uniform Manifold Approximation and Projection (UMAP) was applied before clustering. UMAP reduces dimensionality while preserving relevant local and broader structural relationships within the embedding space, thereby facilitating subsequent topic clustering.¹⁴

Topic Clustering

Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) was applied to the UMAP-reduced embeddings to identify groups of semantically related documents.¹⁵ HDBSCAN was selected because it can accommodate clusters of different sizes and densities, is robust to noise, and does not require the number of clusters to be specified in advance. Nevertheless, the resulting cluster structure remains dependent on parameter selection and corpus composition.

Topic Representation Using C-TF-IDF

Class-based Term Frequency-Inverse Document Frequency (c-TF-IDF) was used to characterize the semantic content of each cluster. This approach identifies terms that are comparatively important within each topic relative to the overall corpus, thereby supporting topic representation and subsequent expert interpretation.¹⁶

Topic Reduction, Expert Refinement, and Stability Assessment

Initial HDBSCAN clustering using a minimum cluster size of 15 produced 12 preliminary topics. To improve interpretability, BERTopic’s reduce_topics function was subsequently applied to merge semantically similar clusters, resulting in seven broader topic groups. Two neurologists independently reviewed the preliminary topic solution and refined the final topic labels and grouping for clinical interpretability. For example, closely related themes involving mitochondrial dysfunction and oxidative stress were consolidated within a broader neuroprotection domain, while Alzheimer’s disease-related mechanistic content was aligned with the broader neurodegeneration theme where appropriate. The reproducibility of the final topic structure was assessed using bootstrap resampling with 100 iterations and 80% subsampling. The final seven-topic solution showed a stability score greater than 0.85. This stability assessment was used to evaluate the reproducibility of the topic structure and was not interpreted as validation of the biological or clinical significance of individual topics.¹⁷

Data and Code Availability

To enhance methodological transparency and reproducibility, the Python code used for the BERTopic pipeline, including model configuration and analytical parameters, is publicly available through the project GitHub repository: https://github.com/BennenJin/KetogenicDiet-Bert.

Results

Bertopic-Assisted Thematic Mapping and Temporal Trends

Following BERTopic model construction and expert-guided topic refinement, seven thematic clusters were retained to describe the organization of recent KD research.¹⁸ These clusters were used as a descriptive framework for thematic mapping and temporal trend assessment rather than as independent evidence of novel mechanistic or clinical relationships. Temporal analysis suggested an increased representation of gut–brain-axis-related publications after 2023, indicating growing research attention to this area during the study period (Figure 2). However, changes in topic frequency may also reflect broader publication trends, database composition, and shifts in terminology and should therefore be interpreted cautiously. The present review focuses on literature published between 2021 and 2025 to characterize recent developments in the field, while recognizing that several foundational mechanisms of the ketogenic diet, including ketone body metabolism and GABA/glutamate regulation, were established in earlier work. Accordingly, BERTopic-assisted mapping was used to support the organization of the recent literature and to highlight changes in research emphasis, whereas the interpretation of mechanistic significance and clinical relevance remained based on expert appraisal of the underlying evidence.

Figure 2.

Annual distribution of all topics.

Thematic Organization of the Literature

BERTopic-assisted mapping provided a descriptive representation of recurring semantic patterns within the recent KD literature. Topic 1 was characterized primarily by terms related to mitochondrial function, inflammation, oxidative stress, therapeutic mechanisms, and neurotransmission, whereas Topic 2 contained terms more closely associated with Alzheimer’s disease, β-amyloid, cerebral function, neurocognition, and the hippocampus. The separation of these themes within the selected embedding and clustering framework may reflect differences in terminology and research emphasis between mechanism-oriented and disease-specific studies. However, this pattern should not be interpreted as evidence of a formal or persistent “terminological silo”, because topic boundaries are influenced by the embedding model, clustering parameters, corpus composition, and subsequent expert interpretation. Accordingly, these clusters were used primarily as an organizational aid for the narrative synthesis. The observed thematic separation informed the decision to first discuss broadly applicable metabolic and neuroprotective mechanisms in Discussion and then consider disorder-specific clinical applications in Emerging Research Directions, while the substantive interpretation of these domains remained based on expert appraisal of the underlying literature.

Temporal Distribution of Research Themes

Temporal mapping of the seven BERTopic-derived clusters suggested changes in the relative representation of research themes across the 2021–2025 period. In particular, Topic 3, characterized by terms related to the gut–brain axis, microbiome, barrier integrity, and lipopolysaccharides, showed an increased representation after 2023 (Figures 2 and 3). This pattern suggests growing research attention to gut–brain-axis-related mechanisms within the recent KD literature. However, changes in topic frequency should be interpreted cautiously because they may reflect broader publication trends, shifts in terminology, database composition, or changes in research emphasis rather than a discrete scientific inflection point. Accordingly, the temporal pattern was used to inform the emphasis placed on the gut–brain axis in Mechanistic Basis of KD in Neurological Disorders, while the mechanistic and clinical significance of this topic was determined through expert appraisal of the underlying literature.

Figure 3.

Seven thematic clusters derived from BERTopic-assisted mapping.

Cross-Cutting Thematic Patterns

Topic 6 contained a heterogeneous combination of terms related to brain injury, biomarkers, drug resistance, and triglycerides, and did not correspond clearly to a single neurological disorder or mechanistic category. This pattern may reflect growing interest in cross-cutting approaches that integrate clinical phenotype, treatment response, and metabolic or biomarker-related characteristics across different neurological conditions. However, the composition of this cluster should be interpreted cautiously. Because topic boundaries are influenced by corpus composition, model parameters, and subsequent expert interpretation, the cluster does not by itself establish the emergence of a distinct biomarker-driven therapeutic paradigm. Instead, it was treated as a hypothesis-generating thematic signal that informed the discussion of biomarker-guided and individualized approaches in the Future Directions section. The clinical relevance of these concepts was evaluated on the basis of the underlying literature rather than inferred directly from the topic model.

Discussion

BERTopic-Informed Framework for the Narrative Synthesis

The Discussion was organized with explicit reference to the seven BERTopic-derived thematic clusters shown in Figure 3. Rather than treating these clusters as independent evidence of mechanistic importance or clinical efficacy, we used them as an organizational scaffold to connect corpus-level semantic patterns with expert-guided interpretation of the underlying literature. Topic 0, characterized by epilepsy, cognition, ketone bodies, energy, and carbohydrate-related terms, informed the discussion of cerebral energy metabolism and epilepsy. Topic 1, dominated by mitochondrial function, inflammation, oxidative stress, therapeutic mechanisms, and neurotransmission, informed the synthesis of major neuroprotective mechanisms. Topic 2, characterized by Alzheimer’s disease, β-amyloid, cerebral function, neurocognition, and hippocampal terms, informed the Alzheimer’s disease-specific discussion. Topic 3, centered on the gut–brain axis, microbiome, barrier integrity, and lipopolysaccharides, informed the gut–brain-axis section. Topic 4 captured broader neurodegenerative, neurodevelopmental, headache, and synaptic themes and therefore informed the cross-disorder clinical synthesis. Topic 5 was strongly associated with migraine, inflammation, prevention, and nutritional intervention and informed the migraine-specific discussion. Topic 6 contained a more heterogeneous combination of brain injury, drug resistance, biomarkers, and triglyceride-related terms and was therefore used cautiously to inform the discussion of treatment resistance, metabolic phenotyping, biomarkers, and future individualized approaches. Importantly, these topic-to-section linkages were used to structure the narrative rather than to determine evidentiary strength. Mechanistic interpretation, clinical appraisal, and conclusions regarding therapeutic efficacy remained based on critical evaluation of the underlying experimental and clinical studies.

Ketogenic Diet Definition and Main Variants

Definition of the Ketogenic Diet

The ketogenic diet (KD) is a specialized, metabolically targeted dietary pattern characterized by high fat, moderate protein, and extremely low carbohydrate intake.¹⁹ The core physiological goal of KD is to induce nutritional ketosis, a metabolic adaptation in which the body shifts its primary energy source from glucose to endogenously produced ketone bodies (β-hydroxybutyrate, acetoacetate, and acetone) due to insufficient carbohydrate availability.²⁰ Typically, KD is formulated to achieve a fat-to-carbohydrate plus protein ratio ranging from 3:1 to 4:1 by weight. In this distribution, carbohydrates account for less than 5–10% of total daily caloric intake, protein contributes 20–30%, and fat comprises 70–80% of total calories. This macronutrient profile suppresses insulin secretion, reduces hepatic gluconeogenesis,²¹ and promotes lipolysis, thereby driving ketone body synthesis in the liver and their subsequent utilization by peripheral tissues and the central nervous system (CNS).²² Notably, nutritional ketosis differs from pathological ketoacidosis, a life-threatening condition characterized by excessive ketone accumulation and severe acidosis. Nutritional ketosis maintains systemic pH within a physiological range (7.35–7.45) and ketone concentrations between 0.5 and 3.0 mmol/L.²³ The neuroprotective and therapeutic effects of KD are closely linked to this controlled ketogenic state, which modulates CNS metabolism, reduces neuroinflammation, enhances antioxidant capacity, and regulates neurotransmitter homeostasis, as elaborated in subsequent sections.²⁴ Over the past century, KD has undergone iterative modifications to improve tolerability, compliance, and clinical applicability, leading to the development of several distinct variants. These variants retain the core principle of inducing nutritional ketosis but differ in macronutrient composition, implementation flexibility, and target populations. Key variants are summarized below (Table 1):

Table 1.

Summary of Ketogenic Diet and Main Variants

Diet Variant	Year Introduced	Fat-To-Non-Fat Ratio	Carbohydrate Content	Fat Source	Key Features	Ref
Classic Ketogenic Diet	1921	4:1 or 3:1	5% of energy	60–80% long-chain fats	Most stringent; difficult to maintain; highly restrictive.	[17–22]
Modified Atkins Diet (MAD)	2003	Variable	~10% of energy	Variable fat sources	Greater flexibility; improved palatability and adherence; outpatient-initiated; popular for adolescents/adults	[24–27]
Medium-Chain Triglyceride (MCT) Diet	1970s	Variable	~10% of energy	~45% medium-chain fats	More efficient ketone generation; better tolerability; allows more carbohydrate	[28–31]
Low Glycemic Index Treatment (LGIT)	Late 1990s-2000s	Variable	Low glycemic index carbohydrates	Variable fat sources	Focuses on glycemic control; easier implementation; fewer adverse effects; good acceptance	[32–34]

Classic Ketogenic Diet (CKD)

Ketogenic diet, first developed in the 1920s, is the most well-established variant and serves as the gold standard for refractory epilepsy, particularly in pediatric populations. CKD typically follows a strict fat-to-non-fat (carbohydrate plus protein) ratio of 4:1 or 3:1 by weight.² In the 4:1 regimen, carbohydrates are restricted to less than 10 g per day (or less than 5% of total calories), while the 3:1 ratio allows slightly more protein and carbohydrates to improve palatability and growth in some children. Protein intake is generally limited to 1–1.5 g/kg of ideal body weight per day to avoid gluconeogenesis from excess amino acids. Fats are primarily derived from saturated and monounsaturated sources, such as butter, coconut oil, avocado, and fatty meats, to maximize ketone production.²⁵

Recent randomized controlled trials have reinforced the superior efficacy of CKD in drug-resistant epilepsy. In a non-inferiority RCT (NCT02708030) comparing CKD (n=55), the modified Atkins diet (MAD, n=58), and low glycemic index treatment (LGIT, n=57) in children aged 1–15 years, CKD achieved the greatest median seizure reduction at 24 weeks (−66%; IQR −85% to −38%), with a ≥50% seizure reduction rate of 67.3% (35/52). Neither MAD nor LGIT met the predefined non-inferiority margin of −15% points relative to CKD, although treatment-related adverse events were similar between CKD and MAD (~56%) and significantly lower with LGIT (33.3%).²⁶ The KIWE trial (NCT02205931), a UK-based Phase 4 multicenter RCT in infants aged 1–24 months with drug-resistant epilepsy, compared CKD (n=78, ratios 2:1 to 4:1) with further antiseizure medication (n=58). Although no superiority was demonstrated (median seizures per day: CKD 5 vs medication 3; IRR 1.33, 95% CI 0.84–2.11), the responder rate of approximately 40% was consistent with prior ketogenic diet studies, and tolerability and safety profiles were comparable between groups, supporting CKD as a viable treatment option for infants after failure of two antiseizure medications.²⁵ A 24-month RCT by El-Shafie et al randomized 40 children with drug-resistant epilepsy to CKD or MAD, with 30 completing the study. CKD achieved seizure freedom in 60% of patients (vs 53.33% with MAD), with the remainder achieving ≥50% seizure reduction. Notably, lipid profiles remained within acceptable levels throughout the 24-month follow-up, adverse effects were mild and medically manageable, and CKD significantly reduced interictal epileptiform discharge frequency while improving EEG background rhythm.²⁷ While highly effective for seizure control, CKD is associated with challenges in long-term compliance due to its strict dietary restrictions, and individualized diet selection based on patient needs and caregiver capacity remains recommended.

Modified Atkins Diet (MAD)

The modified Atkins diet (MAD), first introduced as a clinical therapy in 2003, was developed to improve tolerability while maintaining ketosis.²⁸ Unlike CKD, MAD does not utilize a fixed fat-to-non-fat ratio. Instead, it focuses on restricting carbohydrates to 10–20 g per day, primarily from non-starchy vegetables, while allowing unlimited fat and protein intakee.²⁹ This flexibility eliminates the need for precise meal weighing and simplifies dietary adherence, making MAD suitable for adolescents and adults with neurological disorders.³⁰ MAD has been shown to induce ketosis and achieve therapeutic effects comparable to CKD in epilepsy, AD, and PD.³¹

Recent randomized controlled trials have provided robust evidence for MAD across age groups and neurological conditions. In a prospective, randomized, open-label, blinded-endpoint RCT (CTRI/2015/07/006048) conducted at the All India Institute of Medical Sciences, 160 adolescents and adults (aged 10–55 years) with drug-resistant epilepsy were randomized 1:1 to MAD plus standard drug therapy or standard drug therapy alone for 6 months. The MAD group achieved significantly higher ≥50% seizure reduction rates compared with controls (ITT: 26.2% vs 2.5%, p < 0.001; per-protocol: 45.7% vs 3.2%, p < 0.001), with a 10-fold greater relative risk of response (RR = 10; 95% CI 2.54–43.3). Seizure freedom was attained in 5.0% (ITT) and 8.7% (per-protocol) of MAD participants vs none in controls. Quality of life scores also improved significantly (mean difference 9.6; 95% CI 4.3–14.9, p < 0.001), and the diet was well tolerated, with only one case of weight loss and two cases of diarrhea reported.²⁸ In a separate open-label RCT (NCT04172311) comparing add-on MAD (n=51) with add-on levetiracetam (n=50) in children aged 2–12 years with non-surgical drug-resistant epilepsy, the MAD arm demonstrated significantly superior ≥50% seizure reduction at 12 weeks (52.9% vs 22%, p < 0.001), with a mean seizure frequency reduction of −47.33 ± 39.57% vs −31.15 ± 32.18% (p = 0.03). Constipation was the most common adverse effect with MAD (41.1%), while sedation/lethargy (18%) and anxiety/irritability (14%) predominated in the levetiracetam group. Lennox-Gastaut syndrome was the most common electroclinical syndrome (46%), and the majority had generalized seizures secondary to structural brain lesions, supporting MAD’s utility in severe pediatric epilepsy phenotypes.³² Beyond epilepsy, a 12-week randomized feasibility trial of MAD in older adults with mild cognitive impairment due to early Alzheimer’s disease (n=38 randomized; 20 MAD, 18 control) found a moderate effect size favoring MAD on the Memory Composite Score (Cohen’s D = 0.57; 95% CI −0.67 to 1.33), though the difference did not reach statistical significance due to high attrition and low adherence. Notably, lipidomic and metabolomic analyses revealed significant changes in 13 metabolites and 10 lipids (including triacylglycerols, sphingomyelins, acetoacetate, and fatty acylcarnitines) between MAD and control participants, suggesting biologically meaningful metabolic shifts even with suboptimal ketosis levels.³¹

Medium-Chain Triglyceride (MCT) Ketogenic Diet

The MCT ketogenic diet incorporates medium-chain triglycerides (MCTs), a type of saturated fat that is rapidly absorbed and metabolized into ketone bodies, to enhance ketosis while reducing total fat intake.³³ MCTs, found in sources such as coconut oil or MCT oil, are directly transported to the liver where they are converted to ketone bodies more efficiently than long-chain triglycerides.³⁴ Due to the higher ketogenic potential of MCTs, this variant allows for greater protein and carbohydrate intake compared to CKD.³⁵ Consequently, the effective fat-to-non-fat ratio is significantly lower than in CKD, making it suitable for patients with malabsorption or difficulty tolerating high levels of long-chain fats. MCT-KD has been increasingly studied for its potential in neurodegenerative disorders, as MCT-derived ketone bodies may provide enhanced neuroprotection.³⁶

In the context of epilepsy, a 16-week open-label, randomized, controlled, crossover pilot study conducted at Severance Children’s Hospital (Seoul, South Korea) between August 2022 and September 2023 directly compared a decanoic acid (C10)-enriched MCT ketogenic diet with the classic KD in 15 pediatric patients with refractory epilepsy (10 completers).³⁷ Each patient received both diets for 8 weeks in crossover fashion. Initial treatment with the C10-enriched MCT-KD resulted in seizure reduction in all five patients in that sequence, with two achieving seizure freedom, compared with two of five responders in the classic KD-first group. Upon crossover, patients initially on the classic KD showed additional seizure reduction when switched to the C10-enriched diet, while those initially on the C10-enriched diet maintained their improvements. Adverse effects including transient hypoglycemia, metabolic acidosis, hypercalciuria, and gastrointestinal symptomswere manageable in both groups. This study, the first to directly compare a C10-enriched MCT-KD with the classic KD, highlights the potential synergistic anticonvulsant effects of decanoic acid, which is known to directly inhibit AMPA receptors and enhance mitochondrial biogenesis via PPARγ activation.³⁸ Regarding neurodegenerative applications, a 6-month double-blind, placebo-controlled RCT (UMIN000031281) by Fortier et al (2021) evaluated a ketogenic MCT drink (15 g twice daily) vs placebo in 83 participants with mild cognitive impairment (MCI). The MCT group showed significant improvements in free and cued recall (Trial 1; p = 0.047), verbal fluency (categories; p = 0.024), Boston Naming Test (p = 0.033), and Trail-Making Test total errors (p = 0.017) compared with placebo, with some cognitive outcomes correlating positively with plasma ketone levels.³⁴ Additionally, a 12-month randomized, double-blind, placebo-controlled trial (ChiCTR2200059641) by Duan et al randomized 280 older adults with MCI to four groups: placebo, MCT alone (14 g/d octanoic acid + 10 g/d capric acid), DHA alone (800 mg/d), or MCT + DHA. After 12 months, all three intervention groups demonstrated statistically significant improvements in full-scale IQ, elevated serum total ketone body levels, reduced mitochondrial DNA deletions, and decreased plasma Aβ42 levels compared with placebo. The combined MCT + DHA group showed the greatest benefits, suggesting additive neuroprotective effects of MCT-derived ketones and omega-3 fatty acids.³⁹ These findings build on earlier PET imaging work by Croteau et al, which demonstrated that 30 g/d of MCT supplementation doubled brain ketone body consumption without affecting brain glucose utilization in patients with mild-to-moderate AD, confirming that MCT-derived ketones can directly compensate for the cerebral glucose deficit characteristic of AD.⁴⁰

Low Glycemic Index Treatment (LGIT)

The low glycemic index treatment (LGIT) is a less restrictive variant designed to further improve compliance, particularly for individuals who struggle with the extreme carbohydrate restriction of CKD and MAD.⁴¹ LGIT allows a higher carbohydrate intake of 40–60 g per day but restricts carbohydrates to those with a low glycemic index (GI < 50), such as non-starchy vegetables, legumes, and low-GI fruits.⁴² Unlike CKD, LGIT does not enforce a strict ratio; instead, the fat-to-non-fat ratio is variable, generally adjusted to maintain ketosis while accommodating the higher carbohydrate allowance. LGIT is well-tolerated and has demonstrated efficacy in seizure control, making it a preferred option for long-term management in some patients.⁴³

Recent randomized controlled trials have provided direct evidence supporting LGIT as an effective and well-tolerated dietary therapy. In an open-label RCT by Lakshminarayanan et al, 40 children aged 2–8 years with drug-resistant epilepsy were randomized to add-on LGIT (n=20) or antiseizure medication alone (n=20) for 3 months. 30% (6/20) of children in the LGIT group achieved >50% seizure reduction compared with none in the control group (p = 0.02), with one child achieving seizure freedom and one achieving >90% seizure reduction. The number needed to treat for >50% seizure reduction was 3. Compliance was high (88.5%), and adverse events were limited to transient lethargy (n=2) and vomiting (n=1), confirming LGIT as the least restrictive ketogenic diet variant with minimal side effects.⁴⁴ A 24-week RCT (NCT03464487) by Panda et al (2024) compared daily vs intermittent LGIT in 132 children aged 1–15 years with drug-resistant epilepsy (66 per group; 122 completers). In the intermittent LGIT arm, children received a liberalized diet allowing medium-GI foods and up to 20% carbohydrate calories on two days per week (Saturday and Sunday). Mean weekly seizure frequency reduction was comparable between intermittent and daily LGIT in both ITT (−50.95% ± 22.34% vs −47.16% ± 23.41%, p = 0.36) and per-protocol analyses (−53.88% ± 20.54% vs −49.20% ± 21.87%, p = 0.23). The proportion achieving ≥50% seizure reduction was also similar between groups. Importantly, caregiver difficulty was significantly lower with intermittent LGIT (p = 0.001), with trends favoring fewer adverse events and better compliance, supporting intermittent LGIT as a practical strategy to reduce caregiver burden without sacrificing efficacy.⁴⁴ Furthermore, the first head-to-head RCT comparing LGIT directly with MAD (Anand et al, 2025) randomized 91 children aged 1–15 years with drug-resistant epilepsy to MAD (n=45) or LGIT (n=46) for 24 weeks. Mean seizure reduction was 57% ± 39.4% with LGIT vs 60.7% ± 41.3% with MAD (p = 0.664), with no statistically significant difference. Adverse effects were lower in the LGIT arm (50% vs 66.6%), serious adverse effects were more frequent with MAD, and dyslipidemia was notably rarer with LGIT (1 vs 6 cases). These findings suggest that LGIT may be non-inferior to MAD with the advantage of increased acceptance and fewer adverse effects, reinforcing its role as a practical first-line dietary alternative for children with drug-resistant epilepsy.⁴⁵

Each KD variant possesses distinct advantages and limitations regarding their specific macronutrient ratios, MCT content, and dietary flexibility. Selection is guided by patient age, comorbidities, dietary preferences, and the specific neurological condition being treated.⁴¹ The subsequent sections will focus on the therapeutic applications of these variants.

Mechanistic Basis of KD in Neurological Disorders

Based on expert appraisal of the recent literature, the mechanistic evidence was organized into three broad domains: regulation of cerebral energy metabolism, neuroprotective mechanisms, and modulation of the gut–brain axis. BERTopic-assisted thematic mapping was used only to support the organization of these recurring themes and did not independently determine their biological importance or evidentiary strength. These domains are not mutually exclusive; rather, they interact through overlapping metabolic, inflammatory, redox, neurotransmitter, and microbiome-related pathways. The following sections therefore synthesize the underlying experimental and clinical evidence for these mechanisms in greater detail.

Regulation of Energy Metabolism

KD exerts its therapeutic effects by reshaping cerebral energy metabolism through two complementary pathways: (i) ketone body-mediated alternative energy supply, and (ii) mitochondrial function optimization (Figure 4). These mechanisms are particularly relevant in neurological disorders characterized by glucose hypometabolism, including Alzheimer’s disease, epilepsy, and migraine.

Figure 4.

Ketogenic diet (KD) reshapes cerebral energy metabolism through dual synergistic pathways. (left) Ketone body-mediated alternative energy supply: hepatic fatty acid oxidation generates β-hydroxybutyrate (β-OHB), acetoacetate, and acetone, which cross the blood-brain barrier via monocarboxylate transporters (MCT1/MCT2) to fuel neuronal mitochondrial metabolism. Compared to glucose, ketone oxidation yields greater ATP per oxygen unit while reducing reactive oxygen species (ROS) generation. (right) Mitochondrial function optimization: KD upregulates PGC-1α signaling, promoting mitochondrial biogenesis, enhancing electron transport chain efficiency, and reducing mitochondrial DNA (mtDNA) damage. Together, these pathways enhance neuronal energy resilience in states of glucose hypometabolism characteristic of neurological disorders. Created with BioRender.

Ketone Body-Mediated Alternative Energy Supply

As illustrated in the left panel of Figure 4, ketone bodies provide an efficient alternative fuel for neurons when glucose utilization is impaired. Under physiological conditions, the adult brain relies almost exclusively on glucose as its primary energy substrate. However, in neurological disorders, glucose hypometabolism is frequently observed, for instance, regional cerebral glucose metabolism is reduced by 20–40% in affected brain regions in AD, while in epilepsy, abnormal neuronal hyperexcitability leads to excessive glucose consumption and subsequent energy depletion.⁴⁶ KD-induced nutritional ketosis provides an alternative energy source via three ketone bodies: β-hydroxybutyrate (β-OHB), acetoacetate (AcAc), and acetone. These molecules are synthesized in hepatocytes through mitochondrial β-oxidation of fatty acids, which generates excess acetyl-CoA that is subsequently condensed into acetoacetyl-CoA and then converted to 3-hydroxy-3-methylglutaryl-CoA (HMG-CoA) by HMG-CoA synthase 2 (HMGCS2), the rate-limiting enzyme of ketogenesis. HMG-CoA lyase then cleaves HMG-CoA to produce AcAc, which is either reduced to β-OHB by β-hydroxybutyrate dehydrogenase (BDH1) in an NADH-dependent reaction or spontaneously decarboxylated to acetone.⁴⁷

Circulating ketone bodies cross the blood-brain barrier (BBB) via monocarboxylate transporters (MCTs), primarily MCT1 on endothelial cells and MCT2 on neurons. Under fasting or ketogenic conditions, plasma β-OHB concentrations rise from basal levels of <0.1 mmol/L to 1–5 mmol/L, and MCT expression at the BBB is upregulated, facilitating increased cerebral ketone uptake.⁴⁸ Notably, recent in vitro studies using iPSC-derived brain microvascular endothelial cells demonstrated that β-OHB diffusion across the BBB is saturable and partially MCT1-dependent, and that prior ketone body exposure primes endothelial cells for enhanced β-OHB transport.⁴⁸

Within neuronal mitochondria, β-OHB is oxidized back to AcAc by BDH1, generating NADH. AcAc then receives CoA from succinyl-CoA via succinyl-CoA:3-oxoacid CoA-transferase (SCOT/OXCT1), producing acetoacetyl-CoA, which is cleaved by thiolase into two molecules of acetyl-CoA that enter the tricarboxylic acid (TCA) cycle.⁴⁹ This pathway yields a higher ATP-to-oxygen ratio compared to glucose oxidation: β-OHB oxidation elevates the mitochondrial NAD⁺/NADH ratio and shifts the coenzyme Q couple (CoQ/CoQH₂) toward a more oxidized state, thereby increasing the thermodynamic driving force (ΔG’) across Complex I–II of the electron transport chain (ETC) and enhancing ATP synthesis efficiency.⁵⁰

In a preclinical AD model (3xTg/POLβ+/− mice), 12 weeks of dietary ketosis (blood ketones 1–2 mM) significantly restored the ΔG’ of Complex I–II in hippocampal mitochondria, with the CoQ/CoQH₂ couple becoming more oxidized and the NAD⁺/NADH couple more reduced, demonstrating direct bioenergetic rescue.⁵⁰ Beyond its role as a metabolic fuel, β-OHB functions as a potent signaling molecule through at least three distinct mechanisms. First, β-OHB is an endogenous and specific inhibitor of class I histone deacetylases (HDAC1, HDAC2, HDAC3), as demonstrated by Shimazu et al in a landmark study showing that β-OHB at physiological concentrations (1–5 mM) increases histone acetylation at the promoters of the oxidative stress resistance genes FOXO3A and metallothionein 2 (MT2), thereby upregulating their transcription and conferring substantial protection against oxidative stress in vivo.⁵¹ Second, β-OHB promotes brain-derived neurotrophic factor (BDNF) expression in hippocampal neurons through a cAMP/PKA-CREB(S133) phosphorylation cascade that increases H3K27 acetylation at Bdnf promoters I, II, IV, and VI, while simultaneously reducing the repressive H3K27me3 mark via upregulation of the demethylase JMJD3, the effects observed even at physiological β-OHB concentrations (0.02–2 mM) under adequate glucose supply.⁵² Third, β-OHB serves as a substrate for a novel post-translational modification, which directly modifies histones and non-histone proteins. ChIP-seq studies in mouse cerebral cortex revealed that fasting-induced ketosis significantly enriched H3K9-bhb marks on over 8000 DNA loci, with enrichment on enhancers and promoters correlating with active gene expression, particularly of circadian rhythm genes.⁵³ Collectively, these epigenetic mechanisms enable β-OHB to reprogram neuronal gene expression programs favoring survival, antioxidant defense, and synaptic plasticity. While BHB signaling is extensively characterized, it should not be conflated with the entirety of KD action. For instance, the MCT-KD variant exerts direct anticonvulsant effects via decanoic acid independent of ketosis, and LGIT’s efficacy correlates more strongly with glycemic control than with absolute ketone concentrations. Furthermore, metabolic adaptations such as enhanced fatty acid oxidation and gut microbiome remodeling contribute to neuroprotection through BHB-independent pathways.

Optimization of Mitochondrial Function

As shown in the right panel of Figure 4, KD directly targets mitochondrial dysfunction, a pathological hallmark of neurological disorders characterized by impaired oxidative phosphorylation, increased ROS production, reduced mitochondrial biogenesis, and disrupted mitochondrial dynamics through multiple convergent molecular pathways.⁵⁴

Ketone body oxidation in neuronal mitochondria reduces ROS production through two complementary mechanisms. First, β-OHB metabolism shifts the mitochondrial redox state by elevating the NAD⁺/NADH ratio and oxidizing the CoQ/CoQH₂ couple, which reduces electron leakage at Complex I and Complex III—the primary sites of superoxide (O₂⁻) generation in the etc⁴⁴ In isolated neocortical mitochondria, β-OHB and AcAc significantly decreased ROS levels and raised the threshold for calcium-induced mitochondrial permeability transition (mPT), a key event in oxidative neuronal death.⁵⁵ Second, KD upregulates mitochondrial uncoupling proteins (UCPs), particularly UCP2, UCP4, and UCP5, in hippocampal neurons. Which demonstrated that KD-fed juvenile mice exhibited significantly increased UCP2, UCP4, and UCP5 protein levels in the hippocampal dentate gyrus, with maximum mitochondrial respiration rates significantly higher than controls (p < 0.001), indicating enhanced UCP-mediated proton conductance. Critically, oligomycin-induced ROS production was significantly lower in KD-fed mice (p < 0.05), confirming that UCP activation dissipates the proton motive force (Δp) and reduces protonic backpressure on the ETC, thereby diminishing superoxide generation.⁵⁶

KD triggers a biphasic antioxidant response. An initial, transient increase in mitochondrial ROS (mtROS) during the metabolic transition to fatty acid oxidation activates the redox-sensitive transcription factor nuclear factor erythroid 2-related factor 2 (NRF2). Under basal conditions, NRF2 is sequestered in the cytoplasm by its repressor Keap1 and targeted for proteasomal degradation. mtROS and AMPK promote dissociation of NRF2 from Keap1, allowing NRF2 to translocate to the nucleus, where it binds the antioxidant response element (ARE) in DNA and drives transcription of antioxidant enzymes including heme oxygenase-1 (HO-1), glutathione-S-transferases, and glutamate-cysteine ligase (GCL), the rate-limiting enzyme in glutathione (GSH) biosynthesis.⁵⁷ Elevated GSH levels further inhibit mTOR signaling, contributing to anti-inflammatory and antiepileptogenic effects. In parallel, β-OHB directly activates NRF2 through metabolic intermediates: during β-OHB oxidation in the TCA cycle, fumarate accumulates and acts as an endogenous NRF2 activator by succinating Keap1 cysteine residues, as demonstrated in retinal neurons where β-OHB treatment upregulated NRF2 and preserved the antioxidant pentose phosphate pathway.⁵⁸ Concurrently, β-OHB-mediated HDAC inhibition upregulates FOXO3A, superoxide dismutase 2 (SOD2/MnSOD), and catalase, establishing a second layer of antioxidant defense that directly scavenges mitochondrial superoxide and hydrogen peroxide. In Aβ-injected rat hippocampus, β-OHB pretreatment activated NRF2, recovered SOD and catalase activities, decreased intracellular ROS and Ca²⁺ levels, and inhibited the apoptotic cascade (decreased p53, caspase-3/9/12, and cytochrome c release), providing direct evidence of neuroprotection against AD-related oxidative injury.⁵⁹

KD upregulates PGC-1α, the master regulator of mitochondrial biogenesis, through AMPK-mediated phosphorylation. AMPK, the principal cellular energy sensor, is activated by the transient reduction in ATP levels during the metabolic shift to fatty acid oxidation and by the modest mtROS spike characteristic of KD-induced mitohormesis.⁶⁰ Activated AMPK directly phosphorylates PGC-1α, which then co-activates nuclear respiratory factors (NRF-1, NRF-2/GABP) and mitochondrial transcription factor A (TFAM), driving transcription of both nuclear-encoded and mitochondrial-encoded oxidative phosphorylation subunits and promoting de novo mitochondrial synthesis.⁶¹ Previous study provided direct evidence for this pathway in a mouse model of progressive mitochondrial dysfunction (mutUNG1 mice with impaired mtDNA repair): KD treatment significantly upregulated PGC-1α, SIRT3 (a mitochondrial NAD⁺-dependent deacetylase), and UCP2 in hippocampal CA1 pyramidal neurons. In cultured hippocampal neurons and human fibroblasts exposed to oxidative stress, β-OHB treatment increased the oxygen consumption rate (OCR) and NAD⁺/NADH ratio, confirming enhanced mitochondrial bioenergetics. Notably, KD selectively upregulated the β-OHB receptor GPR109a (HCAR2) but not the lactate receptor GPR81 (HCAR1), suggesting a specific ketone body receptor-mediated signaling mechanism that initiates the PGC-1α–SIRT3–UCP2 biogenesis cascade.⁶² SIRT3, in turn, deacetylates and activates key mitochondrial enzymes including SOD2 and components of the ETC, further linking biogenesis to antioxidant defense. Mitochondrial integrity depends on a tightly regulated balance between fusion (mediated by mitofusin 1/2 [MFN1/2] and optic atrophy protein 1 [OPA1]) and fission (mediated by dynamin-related protein 1 [DRP1] and mitochondrial fission factor [MFF]). In neurodegenerative diseases, this balance is disrupted: AD is associated with excessive DRP1-mediated fission and reduced MFN2/OPA1-mediated fusion, leading to mitochondrial fragmentation, impaired cristae structure, and compromised oxidative phosphorylation capacity; PD-associated PINK1/Parkin mutations impair mitophagy and disrupt fusion-fission equilibrium.⁶³ Preclinical evidence indicates that KD can positively impact mitochondrial dynamics, though the precise molecular targets remain under active investigation. The KD-mediated upregulation of PGC-1α and AMPK is expected to restore fusion-fission balance, as PGC-1α promotes MFN2 expression and AMPK regulates mitochondrial fission through phosphorylation of MFF, while simultaneously activating mitophagy to clear damaged mitochondria.⁶⁴ These effects collectively enhance neuronal energy production, maintain mitochondrial network integrity, and improve cellular resilience against neurodegenerative insults.

Neuroprotective Mechanisms

Neurodegeneration, oxidative stress, neuroinflammation, and dysregulated neurotransmission are key pathological features of neurological disorders. KD exerts notable neuroprotective effects by targeting oxidative stress, neuroinflammation, and dysregulated neurotransmission, creating a favorable microenvironment for neuronal survival (Figure 5).

Figure 5.

Neuroprotective mechanisms of the ketogenic diet (KD). KD exerts notable neuroprotection through three interconnected pathways: (Top left) Anti-oxidative stress: β-hydroxybutyrate (β-OHB) activates the Nrf2 pathway, upregulating antioxidant enzymes and increasing reduced glutathione levels, thereby reducing reactive oxygen species accumulation and lipid peroxidation. (Top right) Anti-neuroinflammation: KD inhibits NLRP3 inflammasome activation via HDAC3 inhibition, reduces pro-inflammatory cytokine production and promotes microglial polarization from pro-inflammatory M1 to anti-inflammatory M2 phenotype. (bottom) Neurotransmitter homeostasis: KD enhances GABAergic signaling through glutamate decarboxylase activation and reduced GABA degradation, decreases glutamate excitotoxicity via glutaminase inhibition, and restores GABA/glutamate balance. These convergent mechanisms collectively improve mitochondrial biogenesis, enhance ATP production, and prevent neuronal death across neurological disorders. Created with BioRender.

Anti-Oxidative Stress

As depicted in the upper left panel of Figure 5, KD enhances the brain’s antioxidant defense system through multiple convergent pathways. Oxidative stress, defined as an imbalance between ROS production and antioxidant defense capacity, contributes to neuronal damage and death in virtually all neurological disorders.⁶⁵ KD enhances the brain’s antioxidant defense system through multiple pathways. β-OHB, acting as a signaling molecule, activates the nuclear factor erythroid 2-related factor 2 (Nrf2) pathway. This activation upregulates the expression of antioxidant enzymes and Phase II detoxification proteins. Consequently, activation of the Nrf2 pathway reduces ROS accumulation, mitigates lipid peroxidation, and protects neuronal membranes from oxidative damage.⁶⁶ Beyond enhancing antioxidant gene expression, KD also reduces the production of ROS at its source by optimizing mitochondrial function. By shifting the brain’s primary fuel source from glucose to ketone bodies, the diet decreases the metabolic flux through glycolysis and the electron transport chain associated with glucose metabolism, which is often a major source of ROS in metabolically stressed or dysfunctional neurons. This metabolic shift promotes a more efficient electron flow, reduces electron leakage, and thereby lowers the generation of superoxide radicals. Furthermore, preclinical studies have consistently demonstrated that KD significantly increases the levels of reduced glutathione (GSH), the brain’s most abundant and crucial endogenous antioxidant, in various animal models, including those of epilepsy, AD, and PD.⁶⁷ This elevation in GSH levels further supports the notable anti-oxidative stress effects of the KD, as summarized in Figure 5, highlighting its potential as a neuroprotective strategy by restoring redox balance in the central nervous system.

Anti-Inflammation

Neuroinflammation, characterized by the activation of microglia and astrocytes, the release of pro-inflammatory cytokines and the infiltration of peripheral immune cells, is a key driver of neurological disease progression.⁶⁸ KD exerts anti-inflammatory effects by modulating both central and peripheral immune responses. In the central nervous system, KD reduces microglial activation and polarizes microglia from a pro-inflammatory (M1) phenotype to an anti-inflammatory (M2) phenotype. This phenotypic shift decreases pro-inflammatory cytokine release (IL-6, TNF-α, IL-1β) and increases anti-inflammatory IL-10 production, as shown in Figure 5. Peripherally, KD modulates the gut microbiome, reducing the production of lipopolysaccharides, a potent inducer of systemic inflammation.⁶⁹ By decreasing LPS translocation across the gut barrier into the bloodstream, KD subsequently reduces neuroinflammation via the gut-brain axis. BERTopic analysis indicated a surge in research on KD and neuroinflammation after 2023, particularly in the context of AD and PD, reflecting growing recognition of this mechanism.

Neurotransmitter Regulation

The lower panel of Figure 5 illustrates KD’s modulation of neurotransmitter homeostasis through multiple pathways. Dysregulation of neurotransmitter systems, including γ-aminobutyric acid (GABA), glutamate, and dopamine, is a hallmark of neurological disorders. Glutamate excitotoxicity contributes to neuronal death in epilepsy and neurodegenerative diseases, while reduced GABAergic inhibition and dopamine dysregulation are associated with epilepsy and PD, respectively. KD modulates neurotransmitter homeostasis through multiple pathways, thereby restoring neuronal excitability balance.⁷⁰ KD increases GABA levels in the brain by enhancing the synthesis of GABA from glutamate via glutamate decarboxylas activation and by reducing GABA degradation. This increase in GABAergic inhibition directly counteracts neuronal hyperexcitability, which is critical for seizure control in epilepsy.⁷¹ Additionally, KD reduces glutamate levels by inhibiting glutaminase and by enhancing glutamate uptake into astrocytes, thereby mitigating glutamate excitotoxicity.⁷² In PD models, KD has been shown to increase dopamine levels in the substantia nigra by upregulating tyrosine hydroxylase and reducing dopamine degradation, providing a potential mechanism for its therapeutic effects in PD.

Modulation of the Gut-Brain Axis

The gut–brain axis represents a complex bidirectional communication network integrating the gut microbiome, the enteric nervous system (ENS), and the central nervous system (CNS), and is increasingly implicated in neurological health and disease. Temporal mapping suggested an increased representation of gut–brain-axis-related publications after 2023, indicating growing research attention to this area during the study period. This descriptive pattern should be interpreted cautiously and does not establish the biological importance of the gut–brain axis relative to other mechanisms. As illustrated in Figure 6, KD may influence this system through two major pathways: (i) remodeling of the gut microbiome and associated microbial metabolites, and (ii) reinforcement of gut barrier integrity with reduced LPS-mediated systemic inflammation.⁷³ Specifically, KD has been reported to remodel the gut microbiome, with enrichment of specific taxa such as Akkermansia muciniphila and Parabacteroides merdae in experimental models. In preclinical epilepsy studies, these KD-associated bacteria, together with Parabacteroides distasonis, have been linked to altered microbial amino-acid metabolism and increased seizure resistance. More broadly, KD-induced microbial changes may influence the production of bioactive metabolites, including short-chain fatty acids, and thereby contribute to gut–brain communication. However, the direction and magnitude of microbiome changes vary across dietary protocols, host populations, and experimental models.⁷⁴ SCFAs, including butyrate, acetate, and propionate, may influence the CNS through systemic circulation and neuroimmune, endocrine, and vagal signaling pathways. These metabolites have been associated with modulation of neuroinflammation, blood–brain barrier integrity, and neurotransmitter-related processes, although the magnitude and clinical relevance of these effects remain context dependent. In parallel, preclinical studies suggest that KD may strengthen gut barrier integrity by increasing the expression of tight-junction proteins such as occludin and zonula occludens-1, thereby reducing intestinal permeability and limiting the translocation of LPS and other pro-inflammatory molecules. These effects may attenuate systemic inflammatory signaling and indirectly influence neuroinflammatory processes.⁷⁵ Collectively, modulation of the gut–brain axis may represent an additional mechanism through which KD influences neurological function, complementing its more direct effects on energy metabolism, mitochondrial function, and inflammatory signaling.

Figure 6.

Modulation of the gut-brain axis by the ketogenic diet (KD). KD restructures gut microbiome composition and reinforces intestinal barrier integrity, collectively reducing neuroinflammation via the gut-brain axis. (left) Gut microbiome remodeling: KD induces a proliferation of ketone- and lipid-utilizing bacteria and a reduction in glucose-fermenting taxa. These compositional shifts enhance short-chain fatty acid production (SCFAs) while decreasing pro-inflammatory lipopolysaccharide (LPS) generation. (right) Gut barrier integrity regulation: KD upregulates tight junction protein expression, reducing intestinal permeability and preventing LPS translocation into systemic circulation. The resulting decrease in bacterial translocation mitigates both systemic and neuroinflammation, preserving neuronal function. SCFAs and ketone bodies are transported to the central nervous system via systemic circulation and the enteric nervous system, where they exert neuroprotective effects by attenuating neuroinflammation, fortifying blood-brain barrier integrity, and regulating neurotransmitter synthesis. Created with BioRender.

In summary, the therapeutic efficacy of the KD in promoting neurological health is orchestrated through a synergistic interplay of energy metabolism regulation, direct neuroprotection, and gut-brain axis modulation. Taken together, these mechanisms provide a useful framework for organizing the current mechanistic evidence on KD in neurological disorders. Their biological and clinical relevance was evaluated through expert appraisal of the underlying literature rather than inferred from the topic model.

Other Mechanisms

In addition to the core mechanisms identified by BERTopic analysis, several other potential therapeutic pathways of the ketogenic diet (KD) in neurological health have been reported in recent preclinical and clinical studies, although they were not prominently featured in the derived research themes. These complementary mechanisms include the reduction of β-amyloid (Aβ) deposition, improvement of synaptic plasticity, and enhancement of hippocampal function.⁷⁶ These processes are particularly relevant to neurodegenerative disorders such as Alzheimer’s disease and contribute to the multifaceted neuroprotective effects of KD. Briefly, KD reduces Aβ plaque accumulation by inhibiting Aβ synthesis and enhancing its clearance, improves synaptic plasticity via the upregulation of neurotrophic factors and synaptic proteins, and enhances hippocampal function through a stable energy supply, reduced neuroinflammation, and the promotion of hippocampal neurogenesis.

Clinical Applications and Current Evidence Across Neurological Disorders

Based on expert appraisal of the recent clinical literature, the following sections focus on epilepsy, Alzheimer’s disease, Parkinson’s disease, and migraine, which represent major areas of clinical investigation for KD. BERTopic-assisted mapping was used only to support thematic organization and did not determine the strength of clinical evidence for these conditions.

Epilepsy

Epilepsy represents the earliest and most well-established clinical indication for the KD, with therapeutic application dating back to the 1920s.⁷⁷ It remains a first-line adjuvant strategy for drug-refractory epilepsy, which is defined as the failure of response to at least two appropriately chosen and tolerated antiepileptic drug.⁷⁸ This is particularly significant in pediatric populations. The clinical efficacy of the KD in epilepsy is substantiated by extensive preclinical and clinical evidence. In pediatric refractory cases, the diet achieves a seizure reduction of at least 50% in 50% to 70% of patients, with 10% to 20% of patients attaining complete seizure freedom.⁷⁹ While the classic ketogenic diet is considered the gold standard for pediatric management, modified protocols such as the modified Atkins diet and low glycemic index treatment are frequently preferred for adolescents and adults due to superior tolerability and adherence. Recent clinical trials have confirmed that ketogenic diet interventions not only reduce seizure frequency but also ameliorate quality of life, cognitive function, and behavioral outcomes, particularly in patients with genetic epilepsies such as Dravet syndrome and Lennox-Gastaut syndrome.²⁷ Importantly, the efficacy of the ketogenic diet in epilepsy is not restricted to a specific seizure type, with beneficial effects observed in both partial and generalized seizures.

Alzheimer’s Disease

Alzheimer’s disease is the most prevalent neurodegenerative disorder and is characterized by progressive cognitive decline, amyloid-β accumulation, tau pathology, and impaired cerebral energy metabolism.⁸⁰ Studies from several independent groups have reported potential benefits of ketogenic interventions in patients with mild cognitive impairment or Alzheimer’s disease, particularly with modified Atkins or medium-chain triglyceride-based approaches. A randomized crossover trial in patients with Alzheimer’s disease reported improvements in selected clinical outcomes during a modified ketogenic diet intervention, whereas a separate randomized feasibility trial in older adults with mild cognitive impairment found only a moderate effect size favoring the modified Atkins diet on memory performance, with the between-group difference not reaching statistical significance because of limited adherence and attrition. MCT-based randomized trials have also reported improvements in selected cognitive measures and cerebral ketone utilization, although responses have varied across cognitive domains and study populations. Preclinical studies provide additional mechanistic support, including reports of improved memory, reduced neuroinflammatory or oxidative stress-related changes, and modulation of amyloid-associated pathology in animal models.⁸¹ Nevertheless, these findings have not yet established that ketogenic interventions modify the underlying course of Alzheimer’s disease in humans. Systematic and critical reviews similarly emphasize the small sample sizes, heterogeneous dietary protocols, short follow-up periods, and variability in cognitive and biomarker outcomes across studies.⁸² Accordingly, ketogenic interventions should currently be regarded as a promising but investigational approach in Alzheimer’s disease, with larger and longer randomized trials needed to clarify their effects on cognition, daily function, biomarkers, and disease progression.

Parkinson’s Disease

Parkinson’s disease is a neurodegenerative disorder characterized by progressive loss of dopaminergic neurons in the substantia nigra, resulting in cardinal motor symptoms including tremor, rigidity, and bradykinesia, together with non-motor manifestations such as cognitive impairment, depression, fatigue, and sleep disturbance. The rationale for ketogenic interventions in PD is based primarily on their potential effects on mitochondrial metabolism, neuroinflammation, and dopaminergic homeostasis.⁸³ Preclinical studies have reported that ketogenic interventions may preserve nigrostriatal function, attenuate neuroinflammatory changes, and improve motor behavior in experimental models. However, the available human evidence is substantially less consistent. A pilot randomized controlled trial comparing a ketogenic diet with a low-fat diet reported possible improvements in selected non-motor symptoms but did not establish a consistent benefit for core motor outcomes. More recently, an independent randomized feasibility study of an MCT-supplemented ketogenic diet supported the feasibility of ketogenic intervention in patients with PD, but did not provide definitive evidence of clinically meaningful motor improvement. Systematic reviews have similarly concluded that the clinical evidence remains limited by small sample sizes, short intervention periods, heterogeneous dietary protocols, and predominantly pilot or open-label study designs.⁸⁴ Taken together, ketogenic interventions in PD remain investigational. Current evidence suggests possible benefits in selected non-motor domains, whereas consistent improvement in motor symptoms has not been demonstrated. Larger, adequately powered randomized controlled trials are needed to clarify clinical efficacy, determine the optimal dietary formulation, and assess long-term safety, adherence, and interactions with dopaminergic therapies.

Migraine

Migraine is a common primary headache disorder characterized by recurrent, severe headaches often accompanied by nausea, photophobia, and phonophobia. The pathogenesis involves disturbances in energy metabolism, neuroinflammation, and the dysregulated release of neurotransmitters such as serotonin and glutamate—pathways that are directly modulated by the ketogenic diet.⁸⁵ Recent clinical studies identify the ketogenic diet as a potential strategy for migraine prevention, particularly in patients with medication-refractory symptoms or comorbid metabolic conditions such as obesity and insulin resistance.⁸⁶ The medium-chain triglyceride ketogenic diet and modified Atkins diet are the most commonly utilized variants due to their tolerability and ability to induce rapid ketosis.⁸⁷ Proposed mechanisms underlying its efficacy include a stable cerebral energy supply, reduced oxidative stress and neuroinflammation, and the regulation of neurotransmitter balance, although the precise clinical pathways require further elucidation. Notably, the ketogenic diet may also benefit patients by reducing triggers associated with carbohydrate intake and glycemic fluctuations.

In summary, the ketogenic diet exhibits distinct therapeutic profiles across various neurological disorders, with varying levels of clinical evidence supporting its application. While its efficacy is most notablely established in epilepsy, growing research highlights its potential in neurodegenerative disorders and migraine, offering a novel, non-pharmacological approach to complement conventional treatments.

Neurodevelopmental and Neurodegenerative Disorders

In addition to the four core conditions prioritized by BERTopic analysis (epilepsy, AD, PD, and migraine), emerging evidence suggests that the ketogenic diet may exert therapeutic potential across a broader spectrum of neurodevelopmental and neurodegenerative disorders (NDDs).⁸⁸ This section synthesizes preliminary clinical and preclinical findings in multiple sclerosis (MS), autism spectrum disorder (ASD), amyotrophic lateral sclerosis (ALS), Huntington’s disease (HD), and major depressive disorder (MDD), with explicit attention to disorder-specific pathophysiological mechanisms.⁷

Multiple Sclerosis (MS)

Multiple sclerosis (MS) is a chronic autoimmune demyelinating disease characterized by neuroinflammation, oligodendrocyte injury, axonal degeneration, and progressive neurological disability.⁶⁵ Preclinical evidence suggests that ketogenic interventions may influence MS-related pathology through several complementary mechanisms, including modulation of microglial activation, inflammasome signaling, mitochondrial function, and adaptive immune responses. In experimental autoimmune encephalomyelitis (EAE) models, KD has been reported to attenuate neuroinflammation and demyelination through pathways involving NLRP3 inflammasome inhibition, HDAC3-related signaling, mitophagy, and altered Th1/Th17 responses.⁶⁹ However, these mechanistic findings are largely derived from preclinical models, and their relative importance and reproducibility across independent experimental settings remain uncertain. Human evidence is substantially more limited. Current reviews indicate that the clinical effects of ketogenic interventions in MS have not yet been established, and adequately powered randomized controlled trials remain scarce.⁸⁹ Accordingly, KD should presently be regarded as an investigational adjunctive strategy in MS rather than an established therapy. Future studies should determine whether preclinical anti-inflammatory and metabolic effects translate into clinically meaningful improvements in fatigue, cognition, disability progression, or inflammatory disease activity, while also defining the optimal dietary formulation, treatment duration, and patient-selection criteria.⁹⁰

Autism Spectrum Disorder (ASD)

ASD is a neurodevelopmental disorder characterized by social interaction deficits, repetitive behaviors, and cognitive dysfunction. While BERTopic analysis did not identify ASD as a core cluster, emerging literature supports its inclusion within the NDD framework.⁹¹ KD may improve behavioral symptoms through modulation of neurotransmitter balance (GABA/glutamate ratio) and reduction of neuroinflammation. Preclinical studies in BTBR mice demonstrated that KD improves core autism-like behaviors, potentially via gut microbiome modulation and immune response alteration.^91,92 The diet’s ability to enhance mitochondrial function and reduce oxidative stress may address the bioenergetic dysfunction frequently observed in ASD.⁹³ Small-scale clinical studies report improvements in social interaction, repetitive behaviors, and cognitive function.⁹⁴ However, large-scale RCTs are needed to establish efficacy, optimal protocol, and safety in pediatric populations.

Amyotrophic Lateral Sclerosis (ALS) and Huntington’s Disease (HD)

ALS and HD are severe neurodegenerative disorders with substantial unmet therapeutic needs. Although ketogenic interventions may theoretically provide alternative energy substrates and reduce oxidative stress, the available evidence remains limited and heterogeneous. Preclinical studies have reported mixed or negative findings, and human evidence is largely restricted to isolated case reports or small uncontrolled studies. Individual reports have described improvements in functional or symptomatic outcomes during ketogenic or time-restricted ketogenic interventions; however, such observations cannot be generalized or interpreted as evidence of efficacy.⁹⁵ At present, there is insufficient clinical evidence to conclude that KD improves motor function or alters disease progression in ALS or HD. Larger, controlled studies are required before therapeutic conclusions can be drawn.

Major Depressive Disorder (MDD)

MDD is increasingly recognized as having neuroinflammatory and metabolic components. KD may alleviate depressive symptoms by regulating cerebral energy metabolism, enhancing brain-derived neurotrophic factor (BDNF) expression, and modulating the gut-brain axis.⁹⁶ These mechanisms converge on the neuroplasticity deficits and hippocampal atrophy characteristic of MDD. Large-scale clinical validation is still required. The anti-inflammatory effects of KD, particularly via β-hydroxybutyrate-mediated HDAC inhibition and NLRP3 inflammasome suppression, provide a plausible mechanistic rationale for future investigation.⁹⁷

Comparative Strength of Clinical Evidence

The clinical landscape for the ketogenic diet across neurological disorders is marked by substantial heterogeneity, ranging from well-established protocols in epilepsy to preliminary observations in Parkinson’s disease. To ensure methodological transparency, the evidence levels assigned in Table 2 were determined based on a framework adapted from established criteria in neurological dietary interventions. This framework evaluates the quality of literature according to study design and sample size, categorizing the evidence into three distinct tiers.

Table 2.

Summary of Clinical Evidence and Limitations

Neurological Disorder	Primary KD Variant(s) Used	Key Clinical Efficacy	Evidence Level	Key Limitations	Ref
Epilepsy	CKD (pediatrics); MAD, LGIT (adolescents/adults)	Consistent seizure reduction in pediatric drug-resistant epilepsy; seizure freedom achieved in a subset of patients	High (multiple RCTs, meta-analyses, and real-world studies)	Long-term adherence remains challenging; adult evidence is less extensive; optimal treatment duration and predictors of response remain uncertain	[79–81]
Alzheimer’s Disease (AD)	MAD, MCT-KD	Modest short-term improvements in selected cognitive outcomes have been reported, but effects are not consistently demonstrated across studies	Moderate (small-scale trials with heterogeneous protocols and outcomes)	Small sample sizes; heterogeneous KD protocols; limited long-term follow-up; uncertain effects on disease progression and advanced AD	[82–84]
Parkinson’s Disease (PD)	MAD, MCT-KD	No consistent motor benefit demonstrated in human studies; possible improvement in selected non-motor symptoms	Moderate-Low (small, mostly open-label or uncontrolled studies; inconsistent clinical findings)	No large-scale RCTs; small sample sizes; inconsistent motor outcomes; unclear long-term safety and interactions with dopaminergic therapy	[85,86]
Migraine	MCT-KD, MAD	Reduction in headache frequency and/or severity has been reported in some adult studies, but durability and comparative efficacy remain uncertain	Moderate (limited RCTs and short-term clinical studies)	Short follow-up; heterogeneous dietary protocols; limited evidence in refractory migraine; few direct comparisons with standard preventive therapies	[63,64,87,88]

Notes: Evidence Grading: High: Consistent findings from multiple randomized controlled trials (RCTs), meta-analyses, or large-scale real-world studies with notable methodological quality. Moderate: Findings supported by well-conducted but limited trials (eg, small sample sizes), inconsistent results among multiple RCTs, or single RCTs with minor limitations. Moderate-Low: Preliminary findings primarily derived from small, open-label trials, animal studies, or observational studies, where the evidence is suggestive but not yet conclusive.

Abbreviations: CKD, Classic Ketogenic Diet; MAD, Modified Atkins Diet; LGIT, Low Glycemic Index Treatment; MCT-KD, Medium-Chain Triglyceride Ketogenic Diet.

Epilepsy: A Cornerstone Application with Evolving Nuances

In the realm of epilepsy, the KD stands as a therapeutic mainstay, particularly for pediatric populations with drug-resistant conditions. Clinical data in this domain are notable, demonstrating that a significant proportion of patients achieve meaningful seizure reduction. The diet’s efficacy in infants has been shown to rival that of additional antiseizure medications, underscoring its potency.⁵⁷ However, the transition to adult populations reveals a different picture; long-term adherence is often poor, with a high rate of treatment discontinuation and a rapid decline in seizure control post-intervention.⁹⁸ Key challenges remain in defining the optimal duration of therapy and identifying reliable predictors of response, which are essential for tailoring treatment strategies to individual patients.⁹⁹

Alzheimer’s Disease: Biomarker-Driven Efficacy and Protocol Variability

For Alzheimer’s disease, the clinical evidence base is promising but less mature.¹⁰⁰ A critical insight from recent studies is the differential response based on APOE genotype, where ε4-negative individuals appear to derive greater cognitive benefit from ketone supplementation.¹⁰¹ While there is a clear correlation between blood ketone levels and neurocognitive performance, the field suffers from a lack of standardization.¹⁰² Variations in dietary protocols ranging from macronutrient ratios to the specific KD variant used, which hinder the ability to compare study outcomes directly. Furthermore, the absence of large-scale, long-term randomized controlled trials means the potential of KD to modify the trajectory of neurodegeneration remains an open question.

Parkinson’s Disease: A Focus on Non-Motor Symptoms

Parkinson’s disease currently has among the weakest clinical evidence for ketogenic interventions. Although preclinical studies suggest potential neuroprotective effects, human studies have not demonstrated consistent improvements in core motor symptoms.¹⁰³ Some small, predominantly open-label or uncontrolled studies have reported possible benefits in selected non-motor domains, including fatigue, pain, and cognitive symptoms;¹⁰⁴ however, these findings remain preliminary and are limited by small sample sizes, short follow-up periods, and substantial risk of bias. At present, there is insufficient evidence to conclude that ketogenic interventions provide clinically meaningful motor benefit in PD, and the effects on non-motor symptoms also require confirmation in adequately powered randomized controlled trials. Important uncertainties remain regarding the optimal dietary formulation, long-term safety and adherence, and potential interactions with dopaminergic therapies.

Migraine: Prevention with Persistence Challenges

In migraine management, the KD has emerged as a potent preventive strategy. Notably, very-low-calorie variants have shown sustained efficacy in reducing headache frequency, with benefits extending even after the reintroduction of carbohydrates.⁸⁷ The reduction in inflammatory markers suggests a direct link between the state of ketosis and symptom alleviation.¹⁰⁵ Despite this, the sustainability of the intervention is a major hurdle; benefits are often lost once the diet is discontinued.¹⁰⁵ The evidence base is currently limited by short follow-up periods and a lack of comparative data against first-line pharmacological preventatives, leaving clinicians without clear guidance on its positioning within standard care pathways.^86,106

Safety, Risks, and Clinical Managements

While the KD demonstrates therapeutic potential, its implementation is associated with adverse effects primarily driven by metabolic adaptation and macronutrient restriction. Although most effects are mild and reversible, severe complications may limit clinical applicability.¹⁰⁷ This section focuses specifically on risks relevant to neurological patient populations, emphasizing risks that directly impact therapeutic outcomes and differential diagnosis.

General Metabolic Adaptation

Adverse effects common to all KD implementations include nutritional deficiencies, gastrointestinal disturbances, and metabolic perturbations. These risks are well characterized in general literature and require standard monitoring. The following subsections emphasize disorder specific considerations that directly impact neurological therapeutic outcomes.

Neurological Disorder-Specific Risks and Clinical Considerations

The safety profile and clinical feasibility of the ketogenic diet vary substantially across neurological disorders because of differences in age, disease-specific symptoms, nutritional vulnerability, comorbidities, and concomitant medications. In addition to the general metabolic and gastrointestinal adverse effects of KD, clinicians should consider disorder-specific risks that may influence tolerability, treatment adherence, symptom interpretation, and medication management. Accordingly, the following sections summarize key clinical considerations for epilepsy, Alzheimer’s disease, Parkinson’s disease, and migraine.

Epilepsy-Specific Risks

In pediatric epilepsy, unique risks warrant particular attention. Growth deceleration may occur with overly strict protein restriction in classic KD protocols, necessitating careful anthropometric monitoring.¹⁰⁸ The neurocognitive effects of metabolic adaptation may transiently mimic or exacerbate post-ictal symptoms, complicating seizure diary interpretation.¹⁰⁹ Critically, KD may alter the pharmacokinetics of antiepileptic drugs, potentially necessitating dose adjustments. For adults with epilepsy, where adherence is already poor, these acute effects contribute to high discontinuation rates.¹¹⁰

Alzheimer’s Disease-Specific Risks

Alzheimer’s patients present distinct vulnerability profiles. The high saturated fat content of KD raises concern given the established cardiovascular dementia continuum.¹⁰¹ Furthermore, the reduced fiber intake inherent to KD may compromise gut microbiome diversity, potentially counteracting gut brain axis benefits.⁸² Polypharmacy interactions with cholinesterase inhibitors require monitoring, as gastrointestinal side effects of both KD and Alzheimer’s medications may synergistically worsen nutritional status.

Parkinson’s Disease-Specific Risks

Parkinson’s patients face unique challenges arising from the intersection of KD with standard dopaminergic therapy.¹¹¹ Protein intake timing critically affects levodopa absorption. This creates a therapeutic tension: insufficient protein may compromise nutritional status, while excessive protein may reduce medication efficacy.¹¹² Additionally, non motor symptoms including dysphagia and gastroparesis may render high fat KD meals poorly tolerable. Orthostatic hypotension, exacerbated by KD induced diuresis, may worsen fall risk in this vulnerable population.¹¹³

Migraine-Specific Risks

For migraine patients, several KD related effects require careful differentiation from disease symptoms. Electrolyte imbalances during KD initiation may themselves trigger headaches, potentially confounding efficacy assessment. The neurocognitive fog of metabolic adaptation overlaps temporally with prodromal migraine symptoms, complicating patient reported outcome interpretation.⁸⁵ Notably, the very low calorie KD variants showing greatest efficacy carry higher risks of gallstone formation and orthostatic symptoms.

Neurocognitive and Behavioral Effects

The adaptation phase neurocognitive effects, often termed “keto flu”, typically resolve within one to four weeks but carry particular significance in neurological populations. In epilepsy, these effects may be misattributed to seizure activity. In Alzheimer’s and Parkinson’s, they may be confused with disease progression.¹¹⁴ In migraine, they may trigger or mimic headache episodes. Clinicians should proactively counsel patients regarding the expected temporal course to prevent premature discontinuation of potentially effective therapy.

Patient-Specific Vulnerabilities Requiring Enhanced Monitoring

Certain neurological subpopulations warrant intensified surveillance. Pediatric patients with epilepsy face heightened risks of growth impairment due to high anabolic demands.¹¹⁵ Elderly patients with neurodegenerative diseases frequently present with polypharmacy and comorbid cardiovascular or renal disease, amplifying drug diet interactions and metabolic risks. Patients with diabetes mellitus require antidiabetic medication adjustment to prevent hypoglycemia during KD initiation. Finally, patients with preexisting eating disorders or poor baseline nutritional status may experience accelerated deterioration on restrictive protocols.

In summary, the therapeutic benefits of the ketogenic diet must be balanced against its potential risks and side effects, which are primarily driven by its extreme macronutrient composition and metabolic adaptation. Most side effects are mild, transient, and manageable with dietary modifications and medical monitoring. However, severe or persistent complications may limit the long-term clinical utility of the diet. Future research should focus on identifying patient-specific risk factors, optimizing ketogenic diet protocols to minimize adverse effects, and developing personalized monitoring strategies to enhance safety and adherence. These are critical steps for advancing the ketogenic diet as a standardized non-pharmacological intervention for neurological disorders.

Future Research Directions

Based on the current state of clinical evidence, identified knowledge gaps, and emerging trends from BERTopic analysis between 2021 and 2025, future research on the ketogenic diet (KD) in neurological health should focus on addressing existing limitations. Priorities include optimizing clinical applicability and expanding the evidence base for its therapeutic potential.

Personalized KD Protocols and Biomarker-Driven Patient Selection

One of the most critical unmet needs in the field is the lack of personalized KD protocols tailored to individual patient characteristics. These characteristics include age, neurological diagnosis, comorbidities, genetic background, and metabolic status. Future research should prioritize the identification of reliable biomarkers that predict treatment response and adverse effect risk. This approach enables precision medicine strategies for KD implementation. For example, in Alzheimer’s disease (AD), the APOE ε4 genotype has already been identified as a modifier of KD response. However, additional biomarkers such as cerebrospinal fluid Aβ levels, blood ketone kinetics, and gut microbiome composition need validation to refine patient selection. Similarly in epilepsy, genetic markers associated with mitochondrial dysfunction or glucose metabolism may help identify patients most likely to benefit from KD. This is particularly relevant for adult populations where efficacy remains uncertain. Furthermore, comparative studies evaluating the efficacy and tolerability of different KD variants are needed. These studies should compare MCT-KD, MAD, and LGIT across specific neurological subgroups. For instance, determining optimal KD variants for elderly patients with PD or for adolescents with refractory epilepsy could improve adherence and clinical outcomes. Additionally, research should explore flexible KD adaptations such as cyclic KD or targeted carbohydrate reintroduction. These strategies aim to balance therapeutic efficacy with long-term compliance, particularly for patients requiring lifelong intervention.

Long-Term Clinical Trials and Real-World Evidence

Current clinical evidence for KD in neurological disorders is largely limited to short-term trials of 12 months or less. There is a paucity of data on long-term safety and efficacy, particularly for neurodegenerative disorders like AD and PD where disease progression spans years. Future research should prioritize large-scale, multicenter, long-term randomized controlled trials (RCTs) lasting 24 to 60 months. These trials are necessary to evaluate whether KD can slow disease progression, reduce long-term disability, and improve mortality outcomes. These trials should also incorporate comprehensive safety monitoring. This includes assessing long-term metabolic, cardiovascular, hepatic, and renal outcomes to address concerns about the potential chronic risks of KD. Complementary to RCTs, real-world evidence (RWE) studies will be critical. Studies using electronic health records, patient registries, and longitudinal cohorts can assess KD effectiveness in diverse clinical settings. This includes patients with comorbidities such as diabetes or cardiovascular disease who are often excluded from RCTs. RWE studies can also provide insights into long-term adherence patterns, factors associated with treatment discontinuation, and real-world safety profiles. Such data are essential for guiding clinical practice.

Elucidation of Unclear Therapeutic Mechanisms

While core mechanisms of KD neuroprotective effects such as energy metabolism, neuroinflammation, and the gut-brain axis have been well characterized, several critical knowledge gaps remain. These gaps are particularly evident regarding disorder-specific mechanisms. For example, the exact pathways through which KD exerts preventive effects in migraine need clarification beyond general energy metabolism and neuroinflammation. Research should focus on neurotransmitter regulation involving serotonin and glutamate, as well as trigeminal nerve activity. In PD, research should investigate why KD improves non-motor symptoms more consistently than motor symptoms. Studies should determine whether this discrepancy is related to differential effects on dopaminergic vs non-dopaminergic pathways.

Additionally, emerging mechanisms require further preclinical and translational research to validate their clinical relevance. These mechanisms include the role of KD in synaptic plasticity, hippocampal neurogenesis, and epigenetic regulation such as HDAC inhibition by β-OHB. Advanced imaging techniques like functional MRI and positron emission tomography, along with multi-omics approaches including genomics, transcriptomics, metabolomics, and microbiomics, can provide novel insights. These tools can elucidate the molecular effects of KD in the human brain and enable the identification of new therapeutic targets.

Mitigation of Adverse Effects and Improvement of Adherence

Low long-term adherence and adverse effects are major barriers to KD clinical adoption. Adverse effects include nutritional deficiencies, gastrointestinal disturbances, and metabolic risks. Future research should focus on developing evidence-based strategies to minimize adverse effects and enhance adherence. For example, studies should explore targeted nutrient supplementation with B vitamins, fiber, and electrolytes to prevent nutritional deficiencies. Developing standardized guidelines for gradual KD initiation is also necessary to reduce acute “keto flu” symptoms. Additionally, patient-centered interventions may improve long-term adherence, particularly in adult and elderly populations. These interventions include dietary counseling, digital tools such as mobile apps for meal planning and ketone monitoring, and support groups. Research should also investigate the safety and efficacy of KD in high-risk populations. These populations include pediatric patients under 2 years of age, elderly patients with multiple comorbidities, and patients with preexisting hepatic or renal dysfunction. Developing tailored monitoring protocols for these vulnerable groups will help mitigate risks. This will expand KD applicability to a broader patient population.

Expansion of KD Applications and Combination Therapies

While BERTopic analysis identified epilepsy, AD, PD, and migraine as core research themes, preliminary evidence suggests KD may have potential in other neurological conditions. These conditions include autism spectrum disorder (ASD), depression, traumatic brain injury, and multiple sclerosis. Future research should conduct well-designed preclinical and clinical studies to evaluate KD efficacy in these understudied conditions. The focus should be on identifying disorder-specific mechanisms and optimal KD protocols. Additionally, combining KD with conventional therapies or emerging interventions may enhance therapeutic outcomes. Conventional therapies include antiepileptic drugs, cholinesterase inhibitors, and dopaminergic medications. Emerging interventions include probiotics, exercise, and cognitive training. For example, research should explore whether KD can improve the efficacy of conventional AD medications or reduce their side effects. Studies should also determine whether combining KD with probiotics can further modulate the gut-brain axis and enhance neuroprotection. These combination strategies have the potential to provide synergistic benefits and improve patient outcomes.

Methodological Considerations for Future Evidence Synthesis

Algorithm-assisted approaches such as topic modeling may complement conventional evidence synthesis by facilitating large-scale thematic organization and trend visualization. However, their outputs remain dependent on model selection, parameter settings, corpus composition, and expert interpretation. Future applications should therefore emphasize transparent reporting, reproducibility, and integration with established systematic and narrative review methodologies rather than treating algorithmic clustering as an independent source of scientific inference.

Emerging Research Directions

The recent literature suggests several areas that warrant further investigation. These directions are based on the combined appraisal of emerging publication patterns and the underlying mechanistic and clinical evidence, rather than on topic-model outputs alone. In particular, increasing attention has been directed toward the gut–brain axis, biomarker-guided patient stratification, and long-term metabolic and neuroprotective effects of KD. These areas should be regarded as evolving research priorities rather than validated future trajectories.

Growing Research Attention to the Gut–Brain Axis

Temporal mapping showed an increased representation of gut–brain-axis-related publications after 2023, suggesting growing research attention to this area within the recent KD literature. This pattern should not be interpreted as evidence that the gut–brain axis has become the dominant biological mechanism of KD, because changes in topic frequency may also reflect publication trends, terminology, and corpus composition.¹¹⁶ Nevertheless, the increasing literature on microbiome composition, intestinal barrier function, and microbial metabolites supports further investigation of the gut–brain axis as a potentially important component of KD-related neurological research.¹¹⁷ Future studies should clarify whether specific microbiome changes contribute causally to neurological outcomes and whether microbiome-targeted strategies can meaningfully enhance KD efficacy.¹¹⁷

Biomarker-Guided Precision Nutrition

The heterogeneous biomarker-related cluster identified by BERTopic included terms related to brain injury, drug resistance, metabolic characteristics, and biomarkers across several neurological conditions. This pattern may reflect increasing interest in individualized treatment approaches, but it does not establish the emergence of a distinct precision ketogenic therapy paradigm.¹¹⁸ Future studies should evaluate whether genetic, metabolic, neuroimaging, or microbiome-derived biomarkers can reproducibly predict treatment response, tolerability, or adherence and thereby support more individualized KD protocols.¹¹⁹

Long-Term Neuroprotective Effects in Aging Populations

The recent literature also showed increasing attention to KD in Alzheimer’s disease, Parkinson’s disease, and other age-related neurological disorders.¹²⁰ This publication pattern suggests that long-term metabolic and neuroprotective effects are becoming an important area of investigation, but it should not be interpreted as evidence of a definitive shift from seizure control toward disease modification. Future studies should determine whether KD can produce sustained effects on cognition, neurodegeneration, mitochondrial function, or inflammatory processes in aging populations and whether such effects translate into clinically meaningful disease modification.

Conclusion

This review highlights KD as an established non-pharmacological therapy for drug-resistant epilepsy and an increasingly investigated metabolic strategy across other neurological disorders. Its neurological effects appear to involve interconnected regulation of cerebral energy metabolism, mitochondrial function, oxidative stress, neuroinflammation, neurotransmission, and the gut–brain axis. Clinical evidence beyond epilepsy remains heterogeneous: findings in Alzheimer’s disease and migraine are encouraging but still evolving, whereas evidence for Parkinson’s disease and other neurological conditions remains preliminary. BERTopic-assisted thematic mapping complemented the expert-guided narrative synthesis by organizing the recent literature into seven thematic clusters and illustrating changes in research emphasis over time, without independently determining mechanistic or clinical significance. Overall, the available evidence supports the broader neurological relevance of KD while underscoring the need for standardized, adequately powered, and long-term clinical studies to clarify disease-specific efficacy, optimize patient selection, and establish its safety and therapeutic role beyond epilepsy.

Limitation

This review has several limitations that should be considered when interpreting its findings. First, BERTopic-assisted thematic mapping does not eliminate subjectivity or methodological bias; rather, some analytical decisions are shifted from manual classification to model specification and subsequent expert interpretation. The resulting topic structure depends on the embedding model, dimensionality-reduction settings, clustering parameters, corpus composition, topic-reduction strategy, and expert-guided labeling. Although PubMedBERT was selected for its biomedical domain specificity and bootstrap resampling demonstrated good stability of the final seven-topic solution (>0.85), such stability reflects reproducibility under resampling and does not establish semantic validity, biological significance, or the existence of discrete scientific domains. More fundamentally, AI-assisted literature synthesis is constrained by the quality, representativeness, and completeness of the retrieved corpus. BERTopic operates on textual representations of published studies and does not independently assess study design, risk of bias, statistical validity, reproducibility, or overall evidence quality. Consequently, publication bias, inconsistent terminology, uneven representation of disease areas, and lower-quality or redundant evidence may influence the resulting thematic structure and temporal patterns. Alternative model specifications or expert interpretations could also produce somewhat different topic boundaries. Accordingly, BERTopic outputs should be interpreted as descriptive maps of the published literature rather than as independent evidence of biological relationships, causal mechanisms, or translational priority. Second, there are inherent limitations regarding database selection and potential bias. Our literature retrieval was confined to Scopus, Web of Science, and PubMed. While these are authoritative sources, this selection may exclude grey literature, non-English publications, or region-specific journals, potentially introducing geographical or linguistic bias. Additionally, the field of ketogenic diet research is susceptible to publication bias. Positive results and studies on novel mechanisms, such as the gut-brain axis, are more likely to be published than null findings. This “file-drawer problem” may skew the thematic landscape identified by BERTopic, potentially overestimating the clinical efficacy of KD interventions. The observed surge in literature post-2023 may partly reflect this hype-driven publication trend rather than a linear translation of clinical benefit.

Beyond the methodological constraints of the review process, critical limitations persist regarding the primary clinical evidence. There is a significant heterogeneity in KD protocols across studies, which complicates the comparison of results and the establishment of standardized guidelines. Additionally, most mechanistic insights are extrapolated from classic KD models with specific fat-to-protein ratios. Caution is warranted when generalizing these findings to modified variants, as differences in macronutrient composition, MCT content, and glycemic load may engage divergent molecular pathways that remain underexplored. Furthermore, long-term safety and efficacy data remain scarce, particularly for neurodegenerative disorders where disease modification requires sustained intervention. Low long-term adherence is another major barrier to clinical adoption, as the strict dietary restrictions often lead to treatment discontinuation. Patient-specific response predictors, such as genetic markers and metabolic phenotypes, are not yet fully characterized, making personalized treatment challenging. Additionally, the KD is associated with potential risks including nutritional deficiencies, gastrointestinal disturbances, and metabolic or cardiovascular concerns, all of which require careful monitoring to ensure patient safety.

Funding Statement

This research was supported by the Qing Lan Project of Jiangsu (JS2023-27) and the General Project of Philosophy and Social Sciences Research in Jiangsu Higher Education Institutions (2025SJSZ0979) to Benneng Jin and Chen Zhe; the Yangzhou Children Friendly Project (2026YFE-135); the Research Fund of Yangzhou Polytechnic College for the Innovative Development of Modern Vocational Education and Natural Science Research (2025XJ17); the Young Elite Scientists Sponsorship Program of Jiangsu Province; and the Yangzhou Municipal Basic Research Program (YZ2025162) to Ruijin Xie.

Declaration of Generative AI and AI-Assisted Technologies in the Writing Process

During the preparation of this manuscript, the authors used ChatGPT (OpenAI, GPT-5) solely for language polishing, grammatical refinement, and improving the clarity and readability of English expression. The tool was not used to generate scientific data, perform data analysis, interpret results, formulate scientific conclusions, or create references. All AI-assisted text was carefully reviewed, verified, and revised by the authors. The authors confirm that all scientific content in the manuscript is original and accurate and take full responsibility for the integrity of the entire manuscript, including the accuracy and appropriateness of all references and citations. The authors also confirm that the terms of use of ChatGPT were reviewed and complied with and that its use is considered suitable for publication.

Data Sharing Statement

Additional data will be made available upon reasonable request to the corresponding author.

Disclosure

Illustrations were created with BioRender.com On behalf of all authors, the corresponding author states that there are no conflicts of interest to report.