Ketogenic diets and cardiovascular health balancing metabolic benefits against lipid-related risk

Abstract

Background

The cardiovascular impact of ketogenic diets (KD) remains debated. In this review, KD refers to adult dietary interventions intended to induce nutritional ketosis through marked carbohydrate restriction, while acknowledging that ketogenic diet therapy (KDT) is a broader umbrella term encompassing several protocol variants.

Methods

We systematically searched PubMed, Embase, Web of Science, and the Cochrane Library from inception to March 2026, and we additionally searched gray literature sources (ClinicalTrials.gov, the WHO International Clinical Trials Registry Platform, medRxiv, conference proceedings, and reference lists). Eligible studies included randomized controlled trials (RCTs), prospective cohorts, and observational studies of adult KD interventions (generally < 10% of energy from carbohydrate or < 50 g/day). Primary outcomes were lipid changes; secondary outcomes included glycemia, anthropometrics, and blood pressure; exploratory outcomes were mortality and atherosclerosis. Random-effects meta-analyses generated pooled estimates. The protocol was not prospectively registered.

Results

Twenty-four studies (20 RCTs, 2 prospective cohorts, 1 non-randomized clinical trial, and 1 pilot intervention; >47,000 participants) were included. Compared with control diets, KD increased low-density lipoprotein cholesterol (LDL-C) (mean difference [MD] + 12.2 mg/dL, 95% CI 7.5 to 16.9; I2 = 67.7%) and total cholesterol, while high-density lipoprotein cholesterol (HDL-C) rose modestly. Triglycerides decreased overall (MD -16.1 mg/dL, 95% CI -20.3 to -11.8; I2 = 17.1%). KD also improved fasting glucose, glycated hemoglobin, body weight, body mass index, and systolic blood pressure. Cohort evidence suggested lower all-cause mortality but no clear reduction in cardiovascular mortality. Imaging evidence for atherosclerosis was limited and mixed.

Conclusions

KD was associated with improvements in adiposity, glycemic control, and blood pressure, but also with higher LDL-C and total cholesterol. Evidence for clinical cardiovascular endpoints remains limited, mixed, and mostly low certainty, supporting individualized risk-benefit assessment rather than universal adoption of KD.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12872-026-06189-7.

Background

Cardiovascular disease (CVD) remains the leading cause of morbidity and mortality worldwide, accounting for an estimated 17.9 million deaths annually [1]. Diet is a central determinant of cardiometabolic health, and dietary interventions are widely recommended for the prevention and management of cardiovascular risk factors. In recent years, the ketogenic diet (KD)—characterized by a very low intake of carbohydrates (typically < 50 g/day or < 10% of total energy) with fat providing the majority of caloric intake—has gained increasing popularity not only for weight reduction but also for the management of metabolic disorders such as type 2 diabetes mellitus (T2DM) and nonalcoholic fatty liver disease [2–4].

Despite its rising use, the cardiovascular implications of KD remain controversial and incompletely understood. Importantly, terminology is inconsistent across the literature. Recent work has emphasized that ketogenic diet therapy (KDT) is a broader umbrella term that includes several protocols, whereas adult cardiometabolic “ketogenic” interventions are often heterogeneous in carbohydrate restriction, protein content, degree of energy restriction, and achieved ketosis [5]. In the present review, we use KD to denote adult dietary interventions intended to induce nutritional ketosis, typically by restricting carbohydrate intake to < 50 g/day or < 10% of total energy with fat as the predominant energy source. On the one hand, KD has been shown to promote weight loss, improve glycemic control, and reduce insulin resistance [6–9]. On the other hand, KD can increase lipids, particularly LDL-C and total cholesterol, raising concerns about long-term safety with respect to atherosclerosis and cardiovascular outcomes [7, 10].

Given the growing adoption of KD worldwide and the conflicting evidence regarding its cardiovascular effects, a comprehensive synthesis of the current literature is warranted. Recent analyses have highlighted marked heterogeneity in LDL-C responses, particularly according to baseline body mass index [11], whereas broader macronutrient epidemiology from the PURE study suggests that the cardiovascular effects of carbohydrate restriction cannot be inferred simply from low-fat versus high-fat contrasts alone [12]. The present study therefore aimed to systematically review and meta-analyze available evidence on the impact of ketogenic diets on cardiovascular risk factors and cardiovascular outcomes, including lipid profile, glycemic control, anthropometric measures, blood pressure, mortality, and atherosclerosis-related endpoints. By integrating evidence from RCTs, cohort studies, and observational investigations, this review seeks to clarify both the potential benefits and the potential harms of KD in relation to cardiovascular health.

Methods

Protocol and reporting standards

This systematic review and meta-analysis was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 guidelines [13]. The completed PRISMA checklist with page numbers is provided as Supplementary Appendix 1. The review protocol was developed before study selection began but was not prospectively registered in PROSPERO or an equivalent registry; this is acknowledged as a limitation because prospective registration improves transparency and reduces the risk of outcome-driven analytic decisions.

Eligibility criteria

We included studies that met predefined eligibility criteria. Eligible designs comprised randomized controlled trials (RCTs), prospective cohort studies, non-randomized clinical trials, and pilot intervention reports. Studies were restricted to adult populations aged 18 years or older, encompassing healthy individuals as well as those with obesity, type 2 diabetes mellitus (T2DM), or metabolic dysfunction-associated steatotic liver disease (MASLD). Because definitions of KDT vary across indications [5], we limited eligibility to adult dietary interventions intended to induce nutritional ketosis, generally defined as carbohydrate intake contributing < 10% of total daily energy or an absolute intake < 50 g/day, with fat providing the majority of energy. We excluded broader low-carbohydrate diets that did not clearly meet this threshold. Eligible comparators included low-fat diets, balanced diets, the Dietary Approaches to Stop Hypertension (DASH) diet [14], or other standard dietary advice. The primary outcomes of interest were changes in lipid parameters, specifically LDL-C, HDL-C, triglycerides (TG), and total cholesterol (TC). Secondary outcomes included glycemic indices such as fasting glucose, glycated hemoglobin (HbA1c), fasting insulin, and homeostatic model assessment of insulin resistance (HOMA-IR), as well as anthropometric measures including body weight, body mass index (BMI), and waist circumference, and blood pressure [15]. Exploratory outcomes were all-cause mortality, cardiovascular mortality, major adverse cardiovascular events (MACE), and imaging or surrogate markers of atherosclerosis. Only articles published in English were considered.

Literature search strategy

A comprehensive literature search was performed in PubMed, Embase, Web of Science, and the Cochrane Library from inception to June 2025 and was updated through March 2026. The search strategy combined controlled vocabulary (e.g., MeSH and Emtree terms) and free-text keywords related to ketogenic diets and cardiovascular outcomes. The main search terms included “ketogenic diet”, “very low carbohydrate diet”, “low-carbohydrate high-fat diet”, “cardiovascular”, “lipid”, “cholesterol”, “mortality”, and “atherosclerosis”. To improve completeness, we also searched gray literature sources, including ClinicalTrials.gov, the WHO International Clinical Trials Registry Platform, medRxiv, conference proceedings where available, and the reference lists of relevant reviews and eligible articles. No restrictions were applied to publication year, but only studies published in English were considered. The search and study selection process adhered to the guidance of the Cochrane Handbook for Systematic Reviews of Interventions [16] and the PRISMA 2020 statement [13].

Study selection

Two reviewers (Hanxiu Liu and Wei Wu) independently screened all records in a two-stage process. First, titles and abstracts were reviewed to exclude clearly irrelevant reports, such as animal studies, pediatric epilepsy trials, narrative reviews, and studies unrelated to cardiovascular or metabolic outcomes. Full texts of potentially eligible articles were then retrieved and assessed in detail against the predefined inclusion and exclusion criteria. Disagreements were resolved by discussion and, when necessary, consultation with a third reviewer (Zhongwei Liu). Because the screening workflow was maintained in a consensus library rather than as preserved reviewer-specific decision logs, a valid post hoc Cohen’s kappa statistic could not be reconstructed; this is acknowledged as a methodological limitation.

Data extraction

Data from the included studies were extracted independently by two investigators (Ni He and Xiqiang Wang) using a predesigned standardized form. Extracted information included study characteristics (first author, year of publication, country, and study design), participant characteristics (sample size, age, sex distribution, baseline body mass index, and underlying conditions), intervention details (definition of ketogenic diet, macronutrient composition, duration of intervention), comparator diet, and outcomes of interest. For randomized controlled trials, we collected baseline and post-intervention values or change-from-baseline means with corresponding standard deviations (SDs) for continuous outcomes. When effect sizes were not directly reported, values were estimated from medians, interquartile ranges, or graphical data using validated conversion methods [17]. In cohort studies, we extracted effect estimates such as hazard ratios (HRs), relative risks (RRs), and odds ratios (ORs), with their corresponding 95% confidence intervals (CIs). All extracted data were cross-checked, and discrepancies were resolved by consensus with a senior reviewer (Zhongwei Liu).

Risk of bias assessment

The methodological quality of the included studies was assessed using validated tools appropriate to study design. For randomized controlled trials, we applied the Cochrane Risk of Bias 2.0 (RoB 2) tool [18], which evaluates potential bias across five domains: (1) randomization process, (2) deviations from intended interventions, (3) missing outcome data, (4) measurement of the outcome, and (5) selection of the reported result. Each domain was rated as “low risk,” “some concerns,” or “high risk,” and an overall judgment was assigned accordingly. For prospective cohort studies, methodological quality was appraised using the Newcastle-Ottawa Scale (NOS) [19]. The NHANES cohort scored 8 of 9 stars (selection 4/4, comparability 2/2, outcome 2/3), whereas the KETO imaging cohort scored 6 of 9 stars (selection 3/4, comparability 1/2, outcome 2/3). We also used the GRADE approach to rate the certainty of evidence for the major outcomes as high, moderate, low, or very low according to risk of bias, inconsistency, indirectness, imprecision, and publication bias [20]. The non-randomized clinical trial and pilot intervention were narratively evaluated, with emphasis on confounding, lack of randomization, adherence assessment, and completeness of outcome reporting. Two reviewers (Shuo Pan and Ling Zhu) independently performed the quality assessments, with disagreements resolved by consensus with a third reviewer (Yuming Zhang). Visual summaries of RoB 2 judgments for RCTs were generated using the traffic-light plot and the domain-level summary plot.

Statistical analysis

All quantitative syntheses were conducted using a random-effects model (DerSimonian-Laird approach), given the expected heterogeneity across study populations and dietary interventions. For continuous outcomes, pooled effect sizes were expressed as mean differences (MDs) with 95% confidence intervals (CIs). When outcomes were reported using different units or scales, standardized mean differences (SMDs) were calculated. For time-to-event or dichotomous outcomes (e.g., all-cause mortality, cardiovascular mortality, or MACE), pooled estimates were reported as hazard ratios (HRs) or relative risks (RRs) with 95% CIs. Between-study heterogeneity was evaluated using the Cochrane Q test (with p < 0.10 indicating statistical significance) and quantified by the I2 statistic, with values of 25%, 50%, and 75% representing low, moderate, and high heterogeneity, respectively [21]. Potential publication bias was assessed by visual inspection of funnel plots and formally tested using Egger’s regression asymmetry test when at least ten studies were available for a given outcome. Forest plots for prespecified secondary outcomes were provided in the Supplementary Figures, and funnel plots with Egger test results were provided for LDL-C and triglycerides. Subgroup analyses were prespecified based on population characteristics and intervention duration. Sensitivity analyses included leave-one-out analyses and restriction to studies judged at low risk of bias. All analyses were performed using Review Manager (RevMan) version 5.4 (Cochrane Collaboration, London, UK) and R version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria) with the “meta” and “metafor” packages.

Results

Literature search and study selection

A total of 1,240 records were identified through electronic database searches (PubMed, Embase, Web of Science, and the Cochrane Library), and an additional 12 records were obtained through manual and gray-literature searches. After removing 310 duplicates, 942 unique records remained for screening by title and abstract. Of these, 860 were excluded as irrelevant (animal studies, pediatric epilepsy trials, narrative reviews, or studies unrelated to cardiovascular or metabolic outcomes). A total of 82 full-text articles were assessed for eligibility. After excluding 58 articles (non-ketogenic low-carbohydrate diets, lack of appropriate comparator, or insufficient outcome data), 24 studies met the inclusion criteria and were incorporated into the qualitative synthesis. These comprised 20 randomized controlled trials (RCTs), 2 prospective cohort studies, 1 non-randomized clinical trial, and 1 pilot intervention report. Twenty-two studies (20 RCTs and 2 cohorts) contributed to quantitative synthesis, while two studies were retained for narrative assessment [2, 7, 22–47]. The updated database and gray-literature searches through March 2026 did not identify additional eligible studies; therefore, the final study set remained unchanged. The process of study selection is illustrated in Fig. 1.

Fig. 1.

PRISMA flow diagram of the study selection process. A total of 1,240 records were identified through database searching and 12 from manual or gray-literature sources. After duplicate removal, 942 titles/abstracts were screened, 82 full texts were assessed, and 24 studies were included in the qualitative synthesis. Twenty-two studies contributed to quantitative synthesis, and two were narratively synthesized

Characteristics of included studies

A total of 24 studies were included in the qualitative synthesis, comprising 20 randomized controlled trials (RCTs), 2 prospective cohort studies, 1 non-randomized clinical trial, and 1 pilot intervention report. Twenty-two studies (20 RCTs and 2 cohorts) provided sufficient quantitative data for meta-analysis, while two studies were included for narrative assessment. The RCTs were conducted between 2003 and 2024 and enrolled between 12 and 307 participants, with intervention durations ranging from 4 weeks to 24 months. Populations included healthy normal-weight women, adults with obesity, individuals with T2DM, adults with metabolic syndrome, cancer patients, and trained athletes. Ketogenic diets were generally defined as carbohydrate intake < 10% of total energy or < 50 g/day, although dietary composition, energy restriction, and achieved ketosis varied across studies. Control groups received low-fat, low-calorie, balanced, high-carbohydrate, or habitual diets, or usual care. Primary outcomes included lipid profile (LDL-C, HDL-C, TG, TC), glycemic indices, anthropometric measures, blood pressure, mortality, and vascular imaging endpoints. Key characteristics of the included studies are summarized in Table 1.

Table 1.

Characteristics of included studies

Study (Year)	Design	Population (n)	Intervention (Ketogenic diet)	Comparator	Duration	Primary outcomes
Foster et al. (2003) [23]	RCT	Obese adults (n = 63)	Low-carbohydrate ketogenic diet (initial phase < = 20 g/day carbohydrate)	Low-fat diet ( < = 30% energy from fat)	12 months	Lipid profile, weight
Yancy et al. (2004) [40]	RCT	Overweight adults with hyperlipidemia (n = 120)	Ketogenic diet ( < = 20 g/day carbohydrate)	Low-fat calorie-restricted diet ( < = 30% energy from fat)	24 weeks	Lipid profile, weight, BMI
Volek et al. (2004) [22]	RCT	Overweight women (n = 28)	Very-low-carbohydrate ketogenic diet ( < = 20 g/day carbohydrate)	Low-fat diet ( < = 30% energy from fat)	12 weeks	Lipid profile, insulin resistance
Segal-Isaacson et al. (2004) [24]	RCT	Adults with overweight/obesity (n = 18)	Low-carbohydrate ketogenic diet ( < = 30 g/day carbohydrate)	Low-fat diet matched for energy and protein	12 weeks	Weight, lipid profile
Johnstone et al. (2008) [41]	RCT	Obese men (n = 20)	High-protein ketogenic diet (ad libitum, <=30 g/day carbohydrate)	Balanced diet (50–55% carbohydrate, 30% fat)	4 weeks	Appetite, weight, lipid profile
Foster et al. (2010) [25]	RCT	Obese adults (n = 307)	Low-carbohydrate ketogenic diet ( < = 20 g/day carbohydrate)	Low-fat diet ( < = 30% energy from fat)	24 months	Lipid profile, weight, glycemic indices
Bazzano et al. (2014) [26]	RCT	Obese adults (n = 148)	Low-carbohydrate ketogenic diet ( < = 40 g/day carbohydrate)	Low-fat diet ( < = 30% energy from fat)	12 months	Lipid profile, weight
Tay et al. (2014) [42]	RCT	Adults with T2DM (n = 115)	Very-low-carbohydrate ketogenic diet ( < = 50 g/day carbohydrate, high unsaturated fat)	High-carbohydrate diet ( > = 55% carbohydrate)	24 weeks	HbA1c, lipid profile, weight
Moreno et al. (2014) [27]	RCT	Obese adults (n = 89)	Very-low-calorie ketogenic diet (VLCKD, 800 kcal/day)	Low-calorie diet	12 weeks	Weight, lipid profile, glucose indices
Zajac et al. (2014) [32]	RCT	Athletes (n = 20)	Ketogenic diet (very-low-carbohydrate, <=50 g/day)	Habitual high-carbohydrate diet	4 weeks	Lipid profile, performance
Veum et al. (2017) [28]	RCT	Adults with metabolic syndrome (n = 46)	Ketogenic diet (very high fat, <=10% carbohydrate)	Low-fat diet ( < = 30% energy from fat)	12 weeks	Visceral adiposity, lipid profile
Saslow et al. (2017) [45]	RCT	Adults with T2DM or prediabetes (n = 34)	Very-low-carbohydrate ketogenic diet (< 30 g/day carbohydrate)	Plate method / moderate-carbohydrate diet	12 months	HbA1c, lipid profile, weight
Burén et al. (2021) [7]	RCT	Healthy young women (n = 17)	Ketogenic diet ( < = 10% energy from carbohydrate, < 30 g/day)	Balanced diet (55% carbohydrate, 30% fat)	4 weeks	Lipid profile
Sánchez et al. (2021) [31]	RCT	Adults with obesity (n = 79)	Very-low-calorie ketogenic diet (VLCKD, 800 kcal/day)	Low-calorie balanced diet	12 weeks	Vasa vasorum density, endothelial markers
Kysel et al. (2020) [37]	RCT	Healthy young males (n = 36)	Cyclical ketogenic diet (ketogenic phase < = 50 g/day carbohydrate)	Nutritionally balanced reduction diet	8 weeks	Body composition, strength, endurance
Perissiou et al. (2020) [43]	RCT	Obese adults (n = 16)	Ketogenic diet ( < = 50 g/day carbohydrate) combined with exercise	Exercise + balanced diet	8 weeks	Lipid profile, body composition
Basciani et al. (2020) [34]	RCT	Obese patients (n = 90)	Very-low-calorie ketogenic diet (VLCKD, 800 kcal/day; protein sources: whey, vegetable, animal)	Balanced low-calorie diet	12 weeks	Lipid profile, glucose indices, weight
Harvey et al. (2019) [30]	Pilot intervention report	Adults (n = 28)	Ketogenic diet (nutritional supplements to induce ketosis)	Balanced diet	6 weeks	Ketosis symptoms, lipid profile
Kackley et al. (2022) [44]	RCT	Overweight/obese adults (n = 30)	Well-formulated ketogenic diet ( < = 50 g/day carbohydrate +/- exogenous ketone salts)	Hypocaloric balanced diet	6 weeks	Cognition, mood, lipid profile
Graybeal et al. (2023) [38]	Randomized crossover trial	Endurance athletes (n = 12)	Ketogenic diet ( < = 50 g/day carbohydrate)	High-carbohydrate / habitual diet	4 weeks	Lipid profile, metabolism, performance
Lim et al. (2024) [47]	RCT	Adults with obesity (n = 80)	Healthy ketogenic diet ( < = 50 g/day carbohydrate, high unsaturated fat)	Energy-restricted balanced diet	6 months	Lipid profile, HbA1c, weight
Athinarayanan et al. (2019) [46]	Non-randomized clinical trial	Adults with T2DM (n = 2625)	Nutritional ketosis intervention (low-carbohydrate diet with remote care)	Usual care	24 months	HbA1c, lipid profile, weight
Budoff et al. (2024, KETO trial) [48]	Prospective cohort (CCTA imaging)	Adults with diet-induced LDL-C rise (n approx. 400)	Ketogenic diet (low-carbohydrate, high-fat)	Non-KD controls	2 years	Coronary plaque progression
Qu et al. (2024, NHANES cohort) [39]	Prospective cohort study	General US adults (n = 43,776)	Ketogenic-like diet pattern ( < = 25% energy from carbohydrate)	Higher-carbohydrate diet pattern	Median 10 years	All-cause and CVD mortality

Abbreviations: BMI body mass index, CCTA coronary computed tomography angiography, DASH Dietary Approaches to Stop Hypertension, HbA1c glycated hemoglobin, HDL-C high-density lipoprotein cholesterol, HOMA-IR homeostatic model assessment of insulin resistance, KD ketogenic diet, LDL-C low-density lipoprotein cholesterol, MASLD metabolic dysfunction-associated steatotic liver disease, NHANES National Health and Nutrition Examination Survey, RCT randomized controlled trial, TC total cholesterol, TG triglycerides, T2DM type 2 diabetes mellitus

Risk of bias assessment

The risk of bias for the 20 randomized controlled trials (including randomized crossover trials) was assessed using the Cochrane Risk of Bias 2.0 (RoB 2) tool. Overall, the methodological quality was judged as moderate, with most trials rated as having “some concerns” rather than high risk. Randomization and allocation concealment were adequately reported in the majority of trials, although several smaller studies provided insufficient detail on the random sequence generation process. Blinding of participants and personnel was generally not feasible due to the nature of dietary interventions, introducing a potential for performance bias. However, the primary outcomes were based on objective biochemical measurements (e.g., lipid profile, fasting glucose), which reduced the likelihood of detection bias. Attrition bias was low across trials, and selective reporting was not evident in most studies. The two prospective cohort studies were evaluated with the Newcastle-Ottawa Scale (NOS): the NHANES cohort scored 8/9 (selection 4/4, comparability 2/2, outcome 2/3), and the KETO imaging cohort scored 6/9 (selection 3/4, comparability 1/2, outcome 2/3). The non-randomized clinical trial and pilot intervention were evaluated narratively because formal pooling with RCTs was not appropriate. A visual summary of the domain-level assessments is presented in Fig. 2.

Fig. 2.

A Risk-of-bias traffic light plot for the 20 included randomized controlled trials. Each row is labeled by first author and year, and each column represents a RoB 2.0 domain. Green = low risk, orange = some concerns, red = high risk. B Domain-level summary of risk-of-bias judgments across the 20 randomized controlled trials

Certainty of evidence (GRADE)

Using the GRADE framework [20], certainty of evidence was judged as moderate for HDL-C, fasting glucose, HbA1c, and systolic blood pressure; low for LDL-C, total cholesterol, triglycerides, body weight, and BMI because of inconsistency and/or imprecision; and very low for mortality and atherosclerosis outcomes because these were derived largely from observational evidence and sparse event data. This certainty assessment was incorporated into the interpretation of the results and discussion.

Effects on hard outcomes

Two prospective cohort studies reported on the association between ketogenic dietary patterns and clinical cardiovascular outcomes. The NHANES analysis (n = 43,776 adults, median follow-up > 10 years) suggested that adherence to a ketogenic-like dietary pattern was associated with lower all-cause mortality (HR 0.76; 95% CI 0.63 to 0.90), but no statistically significant association was observed with cardiovascular mortality (HR 1.13; 95% CI 0.79 to 1.60). These findings suggest that while ketogenic-like dietary patterns may be linked to lower overall mortality, cardiovascular-specific outcomes remain uncertain. The KETO imaging study examined individuals with marked LDL-C elevations while following a ketogenic or very-low-carbohydrate diet. The published report did not demonstrate greater coronary plaque burden than in matched controls, nor a clear association between LDL-C concentration and plaque burden, although sample size and follow-up were limited [48]. No randomized controlled trial directly assessed cardiovascular events or mortality as a prespecified endpoint. Current evidence on hard outcomes is therefore limited to observational data and remains low certainty.

Effects on cardiovascular risk factors

Across the 20 randomized controlled trials included in the lipid meta-analyses, ketogenic diets were associated with a statistically significant increase in LDL-C. As shown in Fig. 3, the pooled random-effects estimate was + 12.2 mg/dL (95% CI 7.5 to 16.9; I2 = 67.7%), indicating substantial between-study heterogeneity. Triglycerides moved in the opposite direction. The pooled estimate showed an overall reduction of -16.1 mg/dL (95% CI -20.3 to -11.8; I2 = 17.1%) (Fig. 4). HDL-C increased modestly across most trials, with pooled analyses showing an average rise of approximately + 2 to + 3 mg/dL, while total cholesterol tended to increase in parallel with LDL-C. Forest plots for additional secondary outcomes, including HDL-C, total cholesterol, fasting glucose, HbA1c, insulin, HOMA-IR, body weight, BMI, waist circumference, and blood pressure, are provided in Supplementary Figure S1. Regarding glucose metabolism, ketogenic diets improved multiple glycemic indices. Fasting glucose decreased significantly (-0.27 mmol/L, 95% CI -0.40 to -0.14; 18 RCTs), as did HbA1c (-0.36%, 95% CI -0.55 to -0.17; 7 RCTs). Fasting insulin levels declined (-2.31 uU/mL, 95% CI -3.43 to -1.18; 15 RCTs), and HOMA-IR improved (-0.63, 95% CI -0.96 to -0.31; 9 RCTs). Anthropometric outcomes also favored ketogenic diets: pooled analysis of 18 RCTs showed an average weight reduction of -8.5 kg (95% CI -10.4 to -6.6), while BMI decreased by -3.0 kg/m2 (95% CI -4.1 to -1.9; 12 RCTs). Waist circumference declined significantly (-6.3 cm, 95% CI -9.4 to -3.2; 9 RCTs). Across 7 RCTs, ketogenic diets reduced systolic blood pressure by -4.2 mmHg (95% CI -7.1 to -1.3) and diastolic blood pressure by -2.3 mmHg (95% CI -4.4 to -0.3). Funnel plots for LDL-C and triglycerides are shown in Supplementary Figures S2 and S3. Egger testing did not indicate significant asymmetry for LDL-C (p = 0.403), whereas asymmetry was detected for triglycerides (p = 0.001), suggesting that small-study effects or selective reporting may have influenced the TG estimate.

Fig. 3.

Forest plot of LDL-C change comparing ketogenic diets with control diets across randomized controlled trials. Each horizontal line represents the mean difference (MD) and 95% confidence interval (CI) for an individual study, and the size of each square reflects the random-effects study weight. The vertical line at 0 indicates no between-group difference. The diamond represents the pooled random-effects estimate. Overall, ketogenic diets were associated with a significant increase in LDL-C levels (MD + 12.2 mg/dL, 95% CI 7.5 to 16.9). Between-study heterogeneity was substantial (I²=67.7%). Abbreviations: CI, confidence interval; LDL-C, low-density lipoprotein cholesterol; MD, mean difference

Fig. 4.

Forest plot of triglyceride changes comparing ketogenic diets with control diets across randomized controlled trials. Each horizontal line represents the mean difference (MD) and 95% confidence interval (CI) for an individual study, and the size of each square reflects the random-effects study weight. The vertical line at 0 indicates no between-group difference. The diamond represents the pooled random-effects estimate. Overall, ketogenic diets were associated with a significant reduction in triglyceride levels (MD − 16.1 mg/dL, 95% CI − 20.3 to − 11.8). Between-study heterogeneity was low (I²=17.1%). Abbreviations: CI, confidence interval; MD, mean difference

Figure 3. Forest plot of LDL-C (mg/dL) comparing ketogenic diets with control diets across randomized controlled trials. Each line represents an individual study with its mean difference and 95% confidence interval, while square size reflects random-effects study weight. The diamond indicates the pooled random-effects estimate. Heterogeneity for the pooled estimate was I2 = 67.7%.

Atherosclerosis and vascular imaging outcomes

Direct evidence on atherosclerosis progression was scarce and inconsistent. The KETO imaging cohort study evaluated coronary plaque burden using coronary computed tomography angiography. Despite markedly higher LDL-C concentrations in the ketogenic-diet group, no significant difference in coronary plaque burden was reported versus matched controls, and LDL-C concentration itself was not associated with plaque burden in that study [48]. None of the 20 randomized controlled trials included in this review reported structural measures of atherosclerosis such as carotid intima-media thickness, coronary calcium score, or arterial stiffness. Consequently, no pooled meta-analysis could be performed for these outcomes, and available evidence should be interpreted as hypothesis-generating.

Sensitivity and subgroup analyses

Sensitivity analyses supported the direction of the primary findings. Exclusion of any single trial in leave-one-out analyses did not materially alter the pooled estimates for LDL-C, triglycerides, HDL-C, or total cholesterol. When the analysis was restricted to studies judged at low risk of bias, the effect sizes remained directionally consistent, although confidence intervals were wider because fewer studies were included. Subgroup analyses revealed clinically relevant heterogeneity in LDL-C response. Trials with shorter intervention durations ( < = 12 weeks) tended to show more pronounced LDL-C increases than longer studies, and LDL-C elevations were larger in healthy normal-weight participants than in those with T2DM or MASLD. Comparator diet also appeared to matter, with larger LDL-C increases when KD was compared with DASH or balanced diets than with low-fat diets. These subgroup patterns are directionally consistent with recent BMI-focused work showing that LDL-C increases are greatest among adults with lower baseline BMI [11].

Discussion

In this systematic review and meta-analysis of 24 studies, including 20 randomized controlled trials, 2 prospective cohort studies, 1 non-randomized clinical trial, and 1 pilot intervention report, ketogenic diets showed a trade-off between metabolic benefit and lipid risk. On the benefit side, KD improved glycemic control, adiposity-related outcomes, and blood pressure. On the risk side, KD increased LDL-C and total cholesterol, with substantial between-study heterogeneity. The certainty of evidence was not uniform across outcomes: moderate for several metabolic markers, low for the principal lipid outcomes because of inconsistency and imprecision, and very low for mortality and atherosclerosis because these outcomes were supported mainly by observational evidence. Taken together, the current evidence supports KD as a potentially effective short-term cardiometabolic intervention, but not as a universally benign dietary strategy.

Our findings are broadly consistent with recent meta-analyses and reviews, which have likewise reported improvements in triglycerides, body weight, and glycemic parameters alongside higher LDL-C and total cholesterol concentrations [11, 49]. At the same time, the present review extends prior work in two important ways. First, by integrating cohort and imaging data, it highlights the gap between favorable changes in intermediate metabolic markers and the still-uncertain consequences for clinical cardiovascular endpoints. Second, it underscores that not all “ketogenic” interventions are equivalent: differences in baseline BMI, fatty-acid composition, comparator diet, energy restriction, and achieved ketosis likely explain a substantial part of the observed heterogeneity. This point aligns with recent calls to distinguish the broader concept of KDT from the more heterogeneous adult cardiometabolic KD literature [5]. Broader macronutrient epidemiology, including the PURE study, also suggests that the health effects of carbohydrate restriction depend strongly on the dietary context and food quality rather than on carbohydrate reduction alone [12].

From a mechanistic perspective, several factors may underlie the adverse LDL-C response. Ketogenic diets often increase saturated fat intake, which may enhance hepatic cholesterol production and reduce LDL receptor activity [50]. Carbohydrate restriction can also alter lipoprotein trafficking and may preferentially expose lean, insulin-sensitive individuals to larger increases in circulating LDL-C, a pattern supported by recent BMI-stratified analyses [11]. Conversely, the improvements in glycemia, insulin sensitivity, and body weight are biologically plausible mediators of lower cardiometabolic risk. The unresolved question is which axis dominates clinically over time. Importantly, currently available imaging evidence does not justify either complacency or alarmism: the published KETO trial did not show greater plaque burden in matched ketogenic-diet participants despite markedly elevated LDL-C [48], whereas small case reports and case series continue to raise concern in selected susceptible phenotypes. Thus, the atherosclerosis signal remains hypothesis-generating rather than definitive.

Several methodological features strengthen the present review. We performed a comprehensive search across multiple databases, updated the search through March 2026, included gray literature sources, separated randomized evidence from observational evidence, and incorporated formal risk-of-bias and GRADE assessments. We also used random-effects models, subgroup analyses, sensitivity analyses, secondary-outcome forest plots, and publication-bias assessments to address anticipated heterogeneity. However, several limitations warrant emphasis. First, heterogeneity remained substantial across lipid outcomes, reflecting differences in study design, dietary composition, intervention duration, comparator diets, and participant phenotype. Second, most RCTs were of relatively short duration, limiting inference about long-term cardiovascular risk. Third, hard outcomes such as cardiovascular events and mortality were available only from observational cohorts and were therefore at higher risk of confounding. Fourth, structural measures of atherosclerosis were rare. Fifth, the review protocol was not prospectively registered, and reviewer agreement statistics could not be reconstructed retrospectively because screening-specific decision logs were not preserved. These limitations reduce certainty and should temper causal interpretation.

In clinical practice, these findings argue for individualized implementation rather than routine or universal recommendation of KD. Ketogenic diets may be reasonable for carefully selected adults seeking short-term improvement in obesity, glycemic control, or blood pressure, but lipid monitoring is essential, particularly in lean or otherwise low-risk individuals who may develop disproportionate LDL-C elevations. Future studies should prioritize standardized KD definitions, prospective protocol registration, longer follow-up, and adjudicated cardiovascular endpoints. Until such evidence is available, KD should be viewed as a double-edged dietary strategy whose metabolic benefits must be balanced against potential lipid-related cardiovascular risk.

Conclusions

Ketogenic diets were associated with improvements in adiposity, glycemic control, and blood pressure, but also with higher LDL-C and total cholesterol. Evidence for cardiovascular events, mortality, and atherosclerosis remains limited and mainly observational. These findings support individualized use of KD with attention to lipid monitoring rather than routine universal recommendation.

Abbreviations

BMI

body mass index

CCTA

coronary computed tomography angiography

CI

confidence interval

CVD

cardiovascular disease

DASH

Dietary Approaches to Stop Hypertension

GRADE

Grading of Recommendations Assessment, Development and Evaluation

HbA1c

glycated hemoglobin

HDL-C

high-density lipoprotein cholesterol

HOMA-IR

homeostatic model assessment of insulin resistance

HR

hazard ratio

KD

ketogenic diet

KDT

ketogenic diet therapy

LDL-C

low-density lipoprotein cholesterol

MACE

major adverse cardiovascular events

MASLD

metabolic dysfunction-associated steatotic liver disease

MD

mean difference

NHANES

National Health and Nutrition Examination Survey

NOS

Newcastle-Ottawa Scale

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

RCT

randomized controlled trial

RoB 2

Cochrane Risk of Bias 2.0

RR

relative risk

SMD

standardized mean difference

T2DM

type 2 diabetes mellitus

TC

total cholesterol

TG

triglycerides

WHO

World Health Organization

Authors’ contributions

HXL and WW performed study selection. NH and XW extracted the data. SP and LZ performed the quality assessments. ZWL and YMZ resolved methodological disagreements and supervised the study. SM contributed to manuscript revision. All authors read and approved the final manuscript.

Funding

This work was supported by the Shaanxi Provincial Innovation Capacity Support Program-Science and Technology Innovation Team Project (Grant No. 2025RS-CXTD-053).

Data availability

All data generated or analyzed during this study are included in this article and its supplementary information files.

Declarations

Ethics approval and consent to participate

Not applicable. This study is a systematic review and meta-analysis of previously published studies and does not involve new data collection from human participants.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.