GRAPHICAL ABSTRACT

INTRODUCTION

Continuous glucose monitoring (CGM) has become a cornerstone of diabetes management since its introduction more than two decades ago, providing real-time data on glucose levels and fluctuations [1-3]. Unlike hemoglobin A1c (HbA1c), which reflects long-term glycemic control but fails to capture daily variability, CGM records glucose values every 5 minutes, offering granular insights into glycemic dynamics [4]. Key CGM-derived indices—including coefficient of variation (CV), time in range (TIR), time below range (TBR), and time above range (TAR)—allow a comprehensive evaluation of glucose control and guide optimization of therapeutic strategies [5].

Beyond monitoring, CGM data have been applied to classify individuals with diabetes or at risk of the disease by revealing characteristic patterns of metabolic dysfunction [6-8]. For instance, one study demonstrated a progressive shift in CGM indices, documenting increasing glycemic variability and deterioration in glucose homeostasis from healthy individuals to those with prediabetes and subsequently to patients with diabetes [9]. Such classifications aim to enhance early detection and risk stratification of individuals at high risk for diabetes progression, as well as to support personalized treatment approaches [10-12].

Despite these advances, limited research has integrated dietary patterns with CGM-based classification in diabetes. Given the pivotal role of nutrition in shaping glycemic responses, we categorized individuals with diabetes according to CGM-derived clusters and investigated associated nutritional factors. We further evaluated whether CGM-based clustering could predict changes in HbA1c after 3 months compared with baseline. This approach seeks to clarify how real-world dietary patterns affect glucose profiles and long-term glycemic outcomes.

METHODS

Study participants and study design

This was a prospective, single-center observational study conducted at Seoul National University Bundang Hospital. Eligible participants were adults aged ≥20 years with a diagnosis of diabetes and an HbA1c level of ≥7.0%. Exclusion criteria included acute or active skin lesions unsuitable for CGM placement, current chemotherapy or immunosuppressive therapy, inability to perform fingerstick glucose monitoring, and planned contrast-enhanced computed tomography within 1 week of study initiation.

All participants underwent CGM and were categorized into three clusters according to CGM-derived metrics. The primary aim was to determine how glucose patterns and variability differed across clusters. Glucose variability was evaluated using the CV (%) and standard deviation (SD). The secondary aim was to identify dietary characteristics that distinguished the clusters.

The study was approved by the Institutional Review Board of Seoul National University Bundang Hospital (No. B-1803-459-301), and all participants provided written informed consent. The study adhered to the principles of the Declaration of Helsinki and good clinical practice guidelines.

CGM and clustering

The iPro2 CGM system (Medtronic, Seoul, Korea) was applied to all participants for 7 days. Glucose values were recorded every 5 minutes, and the data were exported into Excel files (Microsoft, Redmond, WA, USA) for analysis. CGM metrics included mean glucose (mg/dL), CV (%), SD (mg/dL), TAR (>250 mg/dL and >180 mg/dL), TIR (70–180 mg/dL), and TBR (<70 mg/dL). K-means clustering was applied to categorize participants into three clusters based on these metrics.

Dietary assessment and day-to-day fluctuations in nutrient intake

During CGM monitoring (days 1–7), dietary intake was recorded using food diaries, with entries organized by meal type (breakfast, lunch, dinner, and snacks). Participants documented detailed descriptions and portion sizes, such as ‘one glass of juice,’ ‘one serving of cooked rice,’ ‘three chopsticks of seasoned bean sprout salad,’ or ‘half an apple.’ All reported food quantities were converted into grams for analysis. Because starting and ending times varied, dietary records from day 1 and day 7 were excluded, leaving days 2–6 for analysis.

Nutrient composition was calculated using the Food and Nutrient Ingredients Database of the Ministry of Food and Drug Safety of South Korea. Data were summarized by meal and macronutrient: carbohydrates (g), proteins (g), fats (g), saturated fats (g), and sodium (g). Analyses included daily energy intake (kcal), energy by meal, energy distribution across meals (% kcal), macronutrient intake (g), and macronutrient ratio (% kcal). All records were reviewed by a registered dietitian at the study hospital.

We also aimed to examine how day-to-day and meal-to-meal energy and nutrient intake fluctuations differed between the three clusters, and to understand potential dietary patterns. Day-to-day fluctuations in energy and macronutrient intake were defined as the CV (%) or SD (kcal) of daily intake from days 2–6. Meal-to-meal fluctuations were assessed as intra-individual CV (%) and SD (kcal) between meals. Intra-individual CV (%) and SD (kcal) were also calculated for CGM metrics. Spearman’s partial correlation was performed while controlling for insulin use.

Other outcomes

Baseline demographic, clinical, and biochemical characteristics were collected, including comorbidities and glucose-lowering therapies. HbA1c was reassessed 3 months after study initiation. Additional outcomes included the proportion of participants achieving HbA1c <7.0% at 3 months, within-cluster changes from baseline, and differences between clusters. These analyses were intended to evaluate whether baseline clustering predicted short-term glycemic improvement.

Statistical analysis

Descriptive data were presented as mean±SD for continuous variables or as frequencies (numbers and percentages) for categorical variables. Comparisons among the three clusters (cluster 1, cluster 2, and cluster 3) were conducted using one-way analysis of variance (ANOVA) for normally distributed continuous variables, the Kruskal-Wallis test for non-normally distributed continuous variables, and the chi-squared test for categorical variables, as appropriate. A P<0.05 was considered statistically significant. For variables with P<0.05, post hoc analyses were performed using the Tukey’s honest significant difference test for ANOVA or the Dunn’s test for non-parametric comparisons. Associations between continuous variables were examined using Spearman’s rank correlation and partial correlation analysis. All statistical analyses were performed with R version 4.1.2 (R Core Team, Vienna, Austria). Figures were created using Graph-Pad Prism version 10.4.0 (GraphPad Software Inc., San Diego, CA, USA).

RESULTS

Study population

A total of 34 participants with type 1 diabetes mellitus (T1DM, n=10) or type 2 diabetes mellitus (T2DM, n=22) and HbA1c levels >7.0% were enrolled between July 2018 and August 2020. Two individuals were excluded: one due to technical failure preventing retrieval of CGM data, and another due to poor adherence to food diary completion. Thus, 32 participants (14 male and 18 female) were included in the final analysis. One additional individual was excluded from the analysis of HbA1c changes at 3 months because a follow-up HbA1c measurement was unavailable.

As shown in Fig. 1, the 32 participants were categorized into three clusters (cluster 1, cluster 2, and cluster 3) using K-means clustering based on CGM data. Multiple clustering trials were conducted using different combinations of CGM metrics, including mean glucose (mg/dL), CV (%), % TAR >250 mg/dL, % TAR >180 mg/dL, % TIR 70–180 mg/dL, and % TBR <70 mg/dL. After evaluating the results, CV (%) and % TIR 70–180 mg/dL were selected as the independent variables that best distinguished glucose profiles across clusters.

Baseline characteristics

Table 1 presents the baseline characteristics of the three clusters: cluster 1 (well-controlled), cluster 2 (high variability), and cluster 3 (high mean glucose). The sex distribution, age, and proportion of participants with T1DM were similar across clusters. The mean duration of diabetes did not differ significantly: cluster 1 (11.8±10.2 years), cluster 2 (12.9±7.1 years), and cluster 3 (14.6±7.5 years). Although body mass index (BMI) and blood pressure were comparable across clusters, obesity prevalence (BMI ≥25 kg/m²) was highest in cluster 3 (87.5%), compared with cluster 1 (40.0%) and cluster 2 (35.7%) (P=0.060).

Baseline fasting plasma glucose, HbA1c, fasting C-peptide, and fasting insulin levels were similar between clusters. However, cluster 3 exhibited significantly higher 2-hour postprandial glucose (286.7±52.0 mg/dL) than cluster 1 (169.1±82.8 mg/dL) and cluster 2 (203.2±71.6 mg/dL) (P=0.023). Triglyceride levels were also significantly higher in cluster 3 (180.6±88.0 mg/dL) than in cluster 1 (110.2±41.8 mg/dL) and cluster 2 (99.4±48.5 mg/dL) (P=0.013).

Other laboratory results, including renal and liver function tests, were comparable across clusters. Comorbidities such as hypertension and dyslipidemia were similarly distributed. Insulin regimens (basal, premixed, or basal bolus) were also used in comparable proportions across the three clusters.

CGM profiles

CGM parameters for each cluster are presented in Table 2 and Fig. 2. Significant differences were observed among clusters for mean glucose (mg/dL), CV (%), SD (mg/dL), % TAR >180 mg/dL, and % TIR 70–180 mg/dL (all P<0.001). Cluster 1 (well-controlled) had the lowest mean glucose (138.1±16.8 mg/dL) and the highest % TIR 70–180 (94.5%±7.0%). Cluster 3 (high mean glucose) showed the highest mean glucose (203.4±40.5 mg/dL) and the lowest % TIR 70–180 (32.6%±18.1%). Cluster 2 (high variability) displayed intermediate mean glucose (168.3±22.6 mg/dL) and TIR (61.2%±14.2%). Post hoc analyses confirmed that both mean glucose and % TIR differed significantly among all three clusters (P<0.05). Cluster 2 exhibited the highest CV (27.3%±5.4%), which was significantly greater than in the other two clusters. Similarly, SD was highest in cluster 2 (46.1±11.5 mg/dL), followed by cluster 3 (35.3±7.4 mg/dL) and cluster 1 (21.0±7.4 mg/dL). Cluster 3 had the highest % TAR >180 mg/dL (51.6%±20.7%), followed by cluster 2 (30.7%±9.8%), while cluster 1 had the lowest (5.5%±7.0%). Post hoc analysis confirmed that the TAR >180 mg/dL in cluster 1 was significantly lower than in clusters 2 and 3.

Nutritional profiles

Table 3 summarizes the energy intake (kcal) of participants in clusters 1, 2, and 3. Total daily energy intake, energy intake by meal, and the distribution of energy (% kcal) across meals did not differ significantly among clusters. Similarly, macronutrient intake (g) and the macronutrient ratio (% kcal) were comparable across groups.

However, significant differences were observed in carbohydrate intake fluctuations. The day-to-day fluctuations of carbohydrate intake (5-day SD) were significantly higher in cluster 3 (high mean glucose) (502.1±363.4 kcal) compared with cluster 1 (well-controlled) (260.2±94.1 kcal) and cluster 2 (high variability) (250.2±126.1 kcal), with a P value of 0.024. In contrast, day-to-day variability in total daily energy intake, protein, and fat did not differ significantly among clusters.

Table 4 presents the Spearman’s partial correlations of intraindividual meal-to-meal fluctuations in total daily energy intake and macronutrient composition with the energy percentage of breakfast, lunch, and dinner, as well as mean energy intake per meal. Both intra-individual meal-to-meal CV (%) and SD (kcal) for total daily energy intake were positively correlated with the energy percentage of lunch meals (rho=0.49, P=0.005; and rho=0.52, P=0.002, respectively).

In addition, both CV (%) and SD (kcal) of total daily energy intake were negatively correlated with mean energy intake per meal. Specifically, carbohydrate CV (%) and SD (kcal) were negatively correlated with mean energy intake per meal (rho=–0.62, P<0.001; and rho=–0.43, P=0.012, respectively). Similarly, fat CV (%) and SD (kcal) also showed negative correlations.

Furthermore, both CV (%) and SD (kcal) of carbohydrate intake were positively correlated with CGM mean glucose levels, with the SD correlation reaching statistical significance (rho=0.88, P=0.023). A sensitivity analysis excluding one visually distant data point from cluster 3 confirmed the robustness of this finding, as the correlation between carbohydrate intake SD and mean glucose remained strong (rho=0.81, P=0.049).

3-Month follow-up

At the 3-month follow-up, HbA1c improved in cluster 1 (well-controlled), decreasing from 8.5%±1.2% at baseline to 7.7%±1.9%. In contrast, no significant change was observed in clusters 2 and 3. HbA1c shifted from 8.7%±0.8% to 8.6%±0.8% in cluster 2 (high variability) and from 8.2%±1.1% to 8.3%±0.8% in cluster 3 (high mean glucose). In cluster 1, the proportion of participants achieving HbA1c <7.0% increased from 0.0% at baseline to 33.3% at follow-up. By comparison, no participants in clusters 2 or 3 achieved or improved to HbA1c <7.0%.

DISCUSSION

In this study, K-means clustering was performed using standardized CGM metrics derived from ambulatory glucose profile reports—including % TIR, % TBR, mean glucose, SD, and CV—to ensure clinical relevance and comparability across participants [13]. To confirm consistency, clustering analyses were repeated using multiple combinations of validated metrics. % TIR was used to represent overall glycemic control, while CV reflected glycemic variability (%) [14]. Based on these metrics, participants were categorized into three clusters: cluster 1 (well-controlled), cluster 2 (highest CV), and cluster 3 (highest mean glucose).

Cluster 3 (high mean glucose) was characterized by the highest mean glucose level and intermediate glycemic variability, indicating a persistent hyperglycemic profile. This cluster also exhibited the greatest day-to-day fluctuations in carbohydrate intake, suggesting irregular or unbalanced dietary patterns. In contrast, cluster 1 (well-controlled) demonstrated the most favorable glycemic metrics, including both lower mean glucose levels and more consistent carbohydrate intake. Additional analyses showed that higher intra-individual meal-to-meal fluctuations in carbohydrate intake were associated with lower mean energy intake per meal and higher CGM mean glucose levels. Collectively, these findings suggest that in individuals with persistent hyperglycemia and relatively low glycemic variability, inconsistent carbohydrate intake may contribute to suboptimal glycemic control.

The American Diabetes Association’s clinical practice guidelines emphasize lifestyle and behavioral interventions as the foundation of diabetes prevention [15] and highlight the importance of healthy dietary patterns [16]. Dietary approaches such as the Mediterranean diet, intermittent fasting, and low-carbohydrate diets have been associated with improved glycemic control [17-19]. In particular, low-carbohydrate diets have been linked to reduced risk of T2DM [20], improvements in HbA1c, and decreased reliance on glucose-lowering medications [21-23]. These findings underscore that carbohydrate intake patterns play an essential role in diabetes management.

Previous studies have applied CGM-based clustering to identify individuals at risk of diabetes progression and to enhance early risk stratification [9-11]. However, limited work has integrated dietary patterns into CGM-based classification. To our knowledge, this is the first study to suggest that intra-individual fluctuations in energy and macronutrient intake may be linked to distinctive features of glucose profiles.

Our results further imply that dietary patterns characterized by large lunch meals may reflect uneven distribution of carbohydrate or fat intake, consumed in disproportionately high amounts at a single sitting. CGM mean glucose levels were positively correlated with fluctuations in energy and carbohydrate intake, suggesting that carbohydrate-loading dietary patterns may exacerbate hyperglycemia. Thus, dietary patterns involving large, carbohydrate-heavy meals may compromise glucose regulation.

These findings highlight the importance of balanced meal sizes and even nutritional distribution, particularly for individuals with diabetes and high mean glucose levels. Dietary interventions aimed at improving the regularity and balance of carbohydrate consumption may be especially beneficial for those with high mean glucose and low glycemic variability, as observed in cluster 3 (high mean glucose).

Using one week of CGM data, this simplified clustering approach identified individuals more likely to achieve optimal glycemic control in the short term. Cluster 1 (well-controlled), which had favorable baseline CGM profiles, demonstrated HbA1c improvement at 3 months, potentially reflecting greater responsiveness to general lifestyle guidance. In contrast, clusters 2 (high variability) and 3 (high mean glucose) showed no significant improvement, suggesting that these individuals may require more intensive pharmacologic interventions. Overall, the clustering approach demonstrated predictive value for glycemic outcomes over the subsequent 3 months. These findings suggest that early CGM-based clustering could help identify patients more likely to reach short-term glycemic targets and provide prognostic insights useful for clinical decision-making, including determining optimal timing for elective procedures.

However, this study has several limitations. The clustering analysis was exploratory, and the small sample size—with limited numbers of participants in each cluster—reduced the statistical power to detect between-group differences and limited the robustness of subgroup comparisons. Formal power calculations are inherently challenging in unsupervised clustering, as no predefined group structure or effect size exists prior to analysis. Therefore, the findings should be considered hypothesis-generating rather than confirmatory.

In studies with small samples, we acknowledge the potential risk of overfitting. To minimize this, we selected a minimal set of variables for clustering, applied careful parameter selection to avoid over-specification, and interpreted the cluster patterns cautiously. For the one variable that showed a statistically significant difference between clusters—carbohydrate intake variability—we conducted a post hoc power analysis. The observed effect size (Cohen’s f=0.55) corresponded to a statistical power of 76%, suggesting that the finding is reasonably reliable despite the small sample. Nevertheless, confirmation in larger and more homogeneous cohorts is required to validate and generalize these observations.

Our analysis demonstrated that greater variability in carbohydrate intake was associated with higher mean glucose levels. However, given the observational design of this study, the possibility of reverse causation must also be considered. Previous research has indicated that postprandial glucose dips may increase hunger and subsequent energy intake, while irregular meal timing can influence glucose fluctuations. Thus, a bidirectional relationship between glycemic status and dietary behavior is plausible, underscoring the need for further investigation through longitudinal or interventional studies.

CGM data were collected over a 7-day period, but the start time on day 1 and the end time on day 7 varied among participants. To ensure consistency and data quality, only days 2 through 6 were included in the final analysis. Although this monitoring window was relatively short, daily energy and macronutrient intake were similar across all clusters, suggesting that the restriction is unlikely to have introduced substantial dietary bias. Nonetheless, longer monitoring periods would be valuable to determine whether the dietary and glycemic patterns observed here remain consistent over time.

Raw CGM data could have provided additional insights, such as intra-day glucose trends or postprandial excursions. However, these features were excluded due to limited standardization and potential noise. Instead, K-means clustering was performed using standardized CGM metrics derived from ambulatory glucose profile reports, ensuring clinical relevance and comparability across participants.

This study included both T1DM and T2DM patients. The distribution of diabetes type was similar across clusters, and the proportion of participants with T1DM did not differ significantly. Given the small sample, further stratified analyses by diabetes type were not feasible.

All participants received standardized lifestyle education based on Korean Diabetes Association materials, covering balanced meals, regular physical activity, and weight management. However, individual adherence was not formally assessed, and longitudinal behavioral data were not collected, which may have introduced residual confounding.

Alongside carbohydrate intake variability, glycemic variability can be influenced by non-dietary factors such as insulin dosing, physical activity, and hypoglycemic events. Although participants were blinded to CGM data, self-monitored capillary glucose data in insulin users were not considered in this analysis. Moreover, as the study population had a low incidence of hypoglycemia, the findings may not be generalizable to individuals at higher risk for hypoglycemia.

Recognizing the limitation of not accounting for multiple non-dietary influences, we acknowledge that the relationship between carbohydrate intake variability and glycemic variability is likely multifactorial. Nonetheless, this study emphasizes that variability in carbohydrate intake was closely associated with higher mean glucose levels, suggesting that inconsistent dietary intake may contribute to sustained hyperglycemia.

Since this is an observational study, the dietary implications reported here should be regarded as exploratory and hypothesis-generating. This study suggests that CGM can categorize individuals with diabetes according to their glucose profiles, and that such clustering reflects underlying dietary behaviors. High variability in carbohydrate intake was associated with large lunch meals and impaired glycemic control. These findings suggest that, for individuals with a CGM profile of high mean glucose and relatively low CV, dietary strategies emphasizing balanced meal size and distribution of nutrients may be beneficial. Future prospective studies are needed to evaluate whether dietary interventions tailored to CGM-based cluster profiles can improve glycemic outcomes.

Article information

CONFLICTS OF INTEREST

No potential conflict of interest relevant to this article was reported.

ACKNOWLEDGMENTS

Tae Jung Oh was supported by an investigator grant from the Seoul National University Bundang Hospital Research Fund (18-2018-0015).

We would like to thank Ms. Nam Yeon Seo, a registered dietitian at the affiliated hospital, for her valuable advice and support during the early phase of this study.

We acknowledge Ji-hyung Kook’s initial efforts in planning the research.

AUTHOR CONTRIBUTIONS

Conception or design: T.J.O. Acquisition, analysis, or interpretation of data: H.A.K., K.H.K., Y.L., Y.J.S., J.H.M., S.H.C., T.J.O. Drafting the work or revising: H.A.K., T.J.O. Final approval of the manuscript: H.A.K., K.H.K., Y.L., Y.J.S., J.H.M., S.H.C., T.J.O.

Fig. 1.

Patient distribution by K-means clustering based on coefficient of variation (CV) and time in range 70–180 mg/dL (TIR).

Fig. 2.

Percentage of time above range (TAR; >180 or >250 mg/dL), time in range (TIR; 70–180 mg/dL), and time below range (TBR; <70 mg/dL) in clusters 1, 2, and 3.

References

• 1. Aleppo G, Webb K. Continuous glucose monitoring integration in clinical practice: a stepped guide to data review and interpretation. J Diabetes Sci Technol 2019;13:664–73.ArticlePubMedPMCPDF
• 2. Deshmukh H, Wilmot EG, Gregory R, Barnes D, Narendran P, Saunders S, et al. Effect of flash glucose monitoring on glycemic control, hypoglycemia, diabetes-related distress, and resource utilization in the Association of British Clinical Diabetologists (ABCD) nationwide audit. Diabetes Care 2020;43:2153–60.ArticlePubMedPMCPDF
• 3. American Diabetes Association Professional Practice Committee. 6. Glycemic goals and hypoglycemia: standards of care in diabetes-2024. Diabetes Care 2024;47:S111–25.ArticlePubMedPMCPDF
• 4. Parkin CG, Davidson JA. Value of self-monitoring blood glucose pattern analysis in improving diabetes outcomes. J Diabetes Sci Technol 2009;3:500–8.PubMedPMCPDF
• 5. Grunberger G, Sherr J, Allende M, Blevins T, Bode B, Handelsman Y, et al. American Association of Clinical Endocrinology clinical practice guideline: the use of advanced technology in the management of persons with diabetes mellitus. Endocr Pract 2021;27:505–37.ArticlePubMed
• 6. Zhang L, Tian Q, Guo K, Wu J, Ye J, Ding Z, et al. Analysis of detrended fluctuation function derived from continuous glucose monitoring may assist in distinguishing latent autoimmune diabetes in adults from T2DM. Front Endocrinol (Lausanne) 2022;13:948157.ArticlePubMedPMC
• 7. Inayama Y, Yamanoi K, Shitanaka S, Ogura J, Ohara T, Sakai M, et al. A novel classification of glucose profile in pregnancy based on continuous glucose monitoring data. J Obstet Gynaecol Res 2021;47:1281–91.ArticlePubMedPDF
• 8. Montaser E, Brown SA, DeBoer MD, Farhy LS. Predicting the risk of developing type 1 diabetes using a one-week continuous glucose monitoring home test with classification enhanced by machine learning: an exploratory study. J Diabetes Sci Technol 2024;18:257–65.ArticlePubMedPMCPDF
• 9. Cichosz SL, Kronborg T, Laugesen E, Hangaard S, Fleischer J, Hansen TK, et al. From stability to variability: classification of healthy individuals, prediabetes, and type 2 diabetes using glycemic variability indices from continuous glucose monitoring data. Diabetes Technol Ther 2025;27:34–44.ArticlePubMedPDF
• 10. Tao R, Yu X, Lu J, Shen Y, Lu W, Zhu W, et al. Multilevel clustering approach driven by continuous glucose monitoring data for further classification of type 2 diabetes. BMJ Open Diabetes Res Care 2021;9:e001869.ArticlePubMedPMC
• 11. Metwally AA, Perelman D, Park H, Wu Y, Jha A, Sharp S, et al. Prediction of metabolic subphenotypes of type 2 diabetes via continuous glucose monitoring and machine learning. Nat Biomed Eng 2025;9:1222–39.ArticlePubMedPMCPDF
• 12. Shilo S, Keshet A, Rossman H, Godneva A, Talmor-Barkan Y, Aviv Y, et al. Continuous glucose monitoring and intrapersonal variability in fasting glucose. Nat Med 2024;30:1424–31.ArticlePubMedPDF
• 13. Kim K, Choi SH, Jang HC, Park YS, Oh TJ. Glucose profiles assessed by intermittently scanned continuous glucose monitoring system during the perioperative period of metabolic surgery. Diabetes Metab J 2022;46:713–21.PubMedPMCPDF
• 14. Yoo JH, Kim JH. Time in range from continuous glucose monitoring: a novel metric for glycemic control. Diabetes Metab J 2020;44:828–39.ArticlePubMedPMCPDF
• 15. American Diabetes Association. 3. Prevention or delay of type 2 diabetes: standards of medical care in diabetes-2021. Diabetes Care 2021;44:S34–9.ArticlePubMedPDF
• 16. American Diabetes Association Professional Practice Committee. 5. Facilitating positive health behaviors and well-being to improve health outcomes: standards of care in diabetes-2025. Diabetes Care 2025;48:S86–127.PubMedPMC
• 17. Estruch R, Ros E, Salas-Salvado J, Covas MI, Corella D, Aros F, et al. Primary prevention of cardiovascular disease with a Mediterranean diet supplemented with extra-virgin olive oil or nuts. N Engl J Med 2018;378:e34.PubMed
• 18. Salas-Salvado J, Guasch-Ferre M, Lee CH, Estruch R, Clish CB, Ros E. Protective effects of the Mediterranean diet on type 2 diabetes and metabolic syndrome. J Nutr 2015;146:920S–7S.ArticlePubMedPMC
• 19. Stentz FB, Brewer A, Wan J, Garber C, Daniels B, Sands C, et al. Remission of pre-diabetes to normal glucose tolerance in obese adults with high protein versus high carbohydrate diet: randomized control trial. BMJ Open Diabetes Res Care 2016;4:e000258.ArticlePubMedPMC
• 20. Gow ML, Garnett SP, Baur LA, Lister NB. The effectiveness of different diet strategies to reduce type 2 diabetes risk in youth. Nutrients 2016;8:486.ArticlePubMedPMC
• 21. American Diabetes Association. 5. Facilitating behavior change and well-being to improve health outcomes: standards of medical care in diabetes-2021. Diabetes Care 2021;44:S53–72.ArticlePubMedPDF
• 22. Yuan X, Wang J, Yang S, Gao M, Cao L, Li X, et al. Effect of the ketogenic diet on glycemic control, insulin resistance, and lipid metabolism in patients with T2DM: a systematic review and meta-analysis. Nutr Diabetes 2020;10:38.ArticlePubMedPMCPDF
• 23. Saslow LR, Daubenmier JJ, Moskowitz JT, Kim S, Murphy EJ, Phinney SD, et al. Twelve-month outcomes of a randomized trial of a moderate-carbohydrate versus very low-carbohydrate diet in overweight adults with type 2 diabetes mellitus or prediabetes. Nutr Diabetes 2017;7:304.ArticlePubMedPMCPDF

Figure & Data

References

Citations

Citations to this article as recorded by