Association of diet composition and meal intervals with short-term glycemic control and glycemic variability among type 2 diabetes mellitus patients Authors: Mohammed Azhar Hussain, H Anil Kumar, Mahadevamma Lingaiah Published: 2026 ABSTRACT AbstractBackgroundGlycemic variability (GV) is an independent contributor to diabetes related complications beyond glycated hemoglobin. While dietary macronutrient composition influences glycemic responses, the importance of meal timing and nutrient distribution in real world settings remain unclear particularly in Indian diets.MethodsIn this observational study, 47 adults with type 2 diabetes mellitus who completed 10–14 days of continuous glucose monitoring (CGM) were included in the final analysis. The median CGM wear duration was 14(14–14) days and the median CGM analyzed days was 11(10–11) days. Associations between dietary variable and CGM derived metrics including Time in Range (TIR), Time above Range (TAR) and percentage coefficient of variation (%CV) were assessed using Spearman correlation and multivariable mixed-effects regression models adjusted for total energy intake, body mass index (BMI), HbA1c, and insulin/insulin-secretagogue use.ResultsHigher carbohydrate intake was associated with higher %CV (ρ = 0.354, p = 0.017), higher TAR (ρ = 0.343, p = 0.021), higher average glucose (ρ = 0.355, p = 0.017), and lower TIR (ρ = −0.378, p = 0.010) in subject-level analyses. Longer lunch–dinner and breakfast–dinner intervals were associated with higher TIR, whereas the breakfast–lunch interval showed no clear association with glycemic outcomes. In mixed-effects analyses accounting for repeated observations and participant-level covariates, habitual dietary carbohydrate composition remained associated with glycemic variability, whereas meal-timing associations were less consistent after adjustment.ConclusionsHabitual dietary carbohydrate composition remained associated with glycemic variability, whereas meal-timing associations were less consistent in adjusted analyses. Future studies are needed to clarify the role of dietary timing strategies in optimizing CGM-derived glycemic outcomes.Clinical trial numberNot applicable.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12902-026-02381-0.Keywords: Type 2 diabetes mellitus, Glycemic variability, Time in range, Diet composition Introduction Glycemic control plays a vital role in preventing the progression of diabetes-related complications in patients with type-2 diabetes mellitus (T2DM) [1]. In India, glycemic control has been achieved in only 45.75% of adults aged ≥ 45 years with previously diagnosed diabetes [2]. When both diagnosed and undiagnosed individuals with diabetes are considered, only 49.4% received glucose-lowering therapy and just 43% achieve adequate glycemic control [2]. Continuous glucose monitoring (CGM) enables detailed characterization of ambulatory glucose profiles and enables assessment of time in range (TIR), time above range (TAR), time below range (TBR) and glycemic variability [3]. Glycemic variability (GV), defined as fluctuations in blood glucose over time, has emerged as an important component of dysglycemia. Beyond traditional markers such as fasting plasma glucose and glycated hemoglobin (HbA1c), GV independently contributes to oxidative stress, endothelial dysfunction and the development of microvascular complications [4]. In individuals with T2DM, glycemic variability is associated with an increased risk of cardiovascular disease even in presence of well-controlled HbA1c [5]. Dietary approaches may represent as a key strategy for improving glycemic management in individuals with diabetes. Dietary intake is a modifiable factor of postprandial glucose responses and overall glycemic control. Lifestyle factors such as reduced refined grain consumption, increased physical activity and longer sleep duration have been associated with an improved glycemic control [6, 7]. Macronutrient composition, meal timing, glycemic index and overall dietary patterns influence postprandial glucose excursion [8–13]. Diets characterized by lower carbohydrate content, higher dietary fibre, increased protein and healthy fat intake improves glycemic control and reduce glycemic variability [14–17]. A recent CGM-based investigation found that macronutrient composition was not associated with glycemic variability, but may play a role in modulating hypoglycemic events [18]. However, evidence addressing inter-individual variability in glycemic responses to Indian foods remains limited. Therefore, this study aimed to examine the association between dietary pattern and macronutrients composition with CGM-derived glycemic variability among individuals with T2DM. Methods Study design and setting This was a single centre, observational study conducted at a tertiary care teaching hospital. The study followed a prospective design to evaluate dietary patterns and their association with glycemic outcomes. The study was conducted in the Department of Internal Medicine at CDSIMER hospital between March 2024 to February 2025. The study protocol was approved by the Institutional Ethic Committee and all the participants provided written informed consent prior to enrolment. The study was conducted in accordance with the Declaration of Helsinki. Study population Adult participants aged ≥ 18 years were recruited consecutively. Eligible participants included individuals diagnosed with type 2 diabetes mellitus receiving stable anti-diabetic treatment for at least 3 months. Exclusion criteria comprised pregnancy, lactation, chronic kidney disease, use of medications which significantly affect glucose metabolism (other than antidiabetic agents) and inability to provide informed consent. Data collection Baseline demographic and clinical data including age, sex, anthropometric measurements, medical history and medication use were obtained through patient interviews and medical record review. Body mass index (BMI)was calculated as weight (kg) divided by height squared (m2). Dietary data recording and verification Participants were provided with a dietary diary and instructed to prospectively record all foods and beverages consumed during the monitoring period, including approximate portion sizes and timing of intake. The diary was unstructured and allowed free text recording of dietary intake. Portion size estimation was performed using standardized serving sizes recommended in the Dietary Guidelines for Indians, published by the Indian Council of Medical Research–National Institute of Nutrition. To improve completeness and accuracy of dietary records and minimize recall bias, participants were contacted daily by trained study personnel throughout the monitoring period to review entries and verify food items and portion sizes. Dietary records were reviewed by trained investigators prior to nutrient analysis. Meal-based dietary pattern analysis Dietary intake was grouped by meal timing (breakfast, lunch, dinner and snacks). Meals were classified according to what participants reported meal type recorded in the dietary diary. The proportion of total daily energy and macronutrient intake consumed at each meal was calculated. Dietary patterns were described based on percentage of macronutrients distributed across meals rather than total daily intake. Nutrient analysis Dietary intake data were analysed using the Healthify: AI Diet & Fitness mobile application (https://play.google.com/store/apps/details?id=com.healthifyme.basic) which provides nutrient composition estimates based on Indian food composition databases. All food items were entered manually by trained investigators using household portion measures. Macronutrient quantities, carbohydrates, proteins, fats and dietary fibre were extracted for individual food item and combined to derive meal-wise and total daily intake. Macronutrient percentages were calculated by dividing the amount of each macronutrient with the total amount of carbohydrates, proteins and fats combined and multiplying by 100. The application was used solely as nutrient calculation tool. No automated dietary feedback, scoring systems or behavioural recommendations were utilized. Nutrient estimates derived from an app-based database may be subjected to measurement error related to portion-size estimation and recipe variability. Glycemic metrics All participants underwent a 14-day blinded continuous glucose monitoring (CGM) period. The FreeStyle Libre pro sensor was used, which measures interstitial glucose levels at 15-minute intervals, yielding 96 readings per day [19]. The sensor was activated on day 0 and remained fixed for 14 days or until earlier deactivation, continuously capturing glucose measurements. To reduce measurement error associated with sensor stabilization and potential degradation toward the end of the wear period, the first two and last two complete calendar days (00:00–23:59) of CGM data were excluded from analysis. CGM data adequacy (≥ 10 days of CGM) was assessed prior to analysis in accordance with the International Consensus on Time in Range [20]. A valid CGM day was defined as a 24-hour period comprising 96 glucose readings. No imputation was performed for missing glucose values and all included days demonstrated complete intraday capture without missing observation. Participants were required to have a minimum of three valid monitoring days after applying exclusion criteria. For each valid day, 24-hour mean glucose and intraday glycemic variability (%CV) were calculated using interstitial glucose measurements obtained at 15-minute intervals. The percentage coefficient of variation (%CV) was calculated as: As per the recommendations from the Advanced Technologies and Treatments for Diabetes (ATTD) Congress, standard CGM-derived metrics were utilized [20]. Glycemic variability was measured as percentage coefficient of variation (%CV). Glycemic control metrics were derived from continuous glucose monitoring (CGM) data, collected over a minimum of 10–14 consecutive days. Metrics included average glucose, coefficient of variation (%CV), time in range (TIR), time above range (TBR), number of hypoglycemic (< 54 mg/dl, 54-69 mg/dl) and hyperglycemic episode (181-250 mg/dl, > 250 mg/dl). Statistical analysis Continuous variables were summarized as mean ± standard deviation or median (interquartile range), depending on data distribution. Categorical variables were presented as frequencies and percentage. Participant-level averages of macronutrient composition and meal intervals were used to minimize pseudo replication in correlation analyses. Associations between dietary variables and subject-level CGM metrics were assessed using Spearman’s rank correlation and univariate linear regression. For daily repeated measures analyses, mixed effect linear regression models were constructed to evaluate associations between dietary composition, meal timing variables and glycemic variability (%CV). Participant specific random intercepts were included to account for within-subject correlation of repeated observations. A first order autoregressive [AR(1)] residual correlation structure was additionally specified for serial within-subject dependence across study days. Study day was included in all models to account for potential temporal trends during CGM monitoring. Time-varying dietary exposures and meal interval variables were decomposed into participant-specific means (between-person effects) and day-specific deviations from individual means (within-person effects) to distinguish associations attributable to differences between participants from those attributable to day-to-day variation within the same participant. Macronutrient variables were treated as compositional data; therefore, regression models were parameterized using two macronutrient components while interpreting the omitted macronutrient as the reference component. Multivariable mixed-effects models were adjusted for total energy intake (per 100 kcal/day), BMI, HbA1c, and insulin/insulin-secretagogue use (Sulfonyl Ureas or Meglitinides). Additional crude mixed-effects models containing only the exposure variables of interest were also performed to facilitate comparison between unadjusted and adjusted associations. Potential multicollinearity was assessed using variance inflation factors (VIFs). Model performance was evaluated using marginal and conditional R² values. Glycemic variability (%CV) was considered as the primary outcome for daily mixed-effects analyses because of its greater sensitivity to short-term glycemic fluctuations. Daily analyses of additional CGM metrics including TIR, TAR, TBR, and average glucose were performed as exploratory outcomes to characterize short-term day-to-day glycemic patterns alongside the primary %CV analysis. Subject-level summaries across the monitoring period were considered the primary basis for interpretation of these CGM metrics. Statistical analyses were performed using R version 4.5.2, and a two-sided p-value < 0.05 was considered statistically significant. Analytical code, model formulas, and the data dictionary used for mixed-effects regression analyses are provided in Supplementary File 2. Results A total of 69 patients were enrolled, of whom 47 completed the study and had usable CGM and diet data. A total of 664 CGM monitoring days were recorded. After excluding the first and last two days to minimize sensor-related variability, 476 days remained. The final dataset comprised 395 CGM days (46 Patients), restricted to participants with at least three valid monitoring days and a consistent traditional meal pattern of breakfast, lunch, and dinner. A detailed description of participants included and the CGM days analyzed are described in Fig. 1. The median CGM wear duration was 14 (14–14) days with a range of 10–14 and median CGM analyzed days was 11 (10–11) days with a range of 6–11. Baseline demographics, clinical, dietary and glycemic control metrics are described in Table 1. Flow diagram of participant inclusion and continuous glucose monitoring (CGM) data selection Table 1Baseline demographic, clinical, dietary, and CGM characteristics of the study population (n = 47)CharacterMean ± SD / Median (IQR)Sex (males), n (%)26 (55.32)Age (years)53.74 ± 12.81Diabetes duration (years)9.77 ± 6.15Weight (Kg)71.58 ± 12.94Height (cm)158.0 (154.0–172.0)Body mass index (kg/m2)27.25 (22.39–30.37)Systolic blood pressure (mmHg)130.0 (120.0–140.0)HbA1c (%)7.5 (6.9–8.6)FPG (mg/dL)145.0 (136.0–160.0)PPG (mg/dL)178.0 (158.0–220.0)Total Cholesterol (mg/dL)177.0 (166.0–218.0)HDL (mg/dL)51.0 (46.0–55.0)LDL (mg/dL)101.0 (86.0–141.0)VLDL (mg/dL)25.42 ± 7.65Triglycerides (mg/dL)127.1 ± 38.25Mean glucose (mg/dL)140.0 (113.0–187.0)Glycemic variability (%CV)22.25 (18.28–27.98)Time in Range (%)77.0 (45.25–95.0)Time below Range (%)0.0 (0.0–1.0)Time above Range (%)17.5 (0.0–51.75)Protein (% total protein intake/day)15.40 (13.39–17.79)Carbohydrate (% total carbohydrate intake/day)69.63 (64.41–75.42)Fat (% total fat intake/day)14.60 (10.12–18.08)Fiber (g/day)21.60 (16.30–27.60)Energy intake (kcal/day)1529 (1230–1899)Continuous glucose monitoring duration (days)14 (14–14) Antidiabetic therapy Oral hypoglycemic agentsOnly InsulinBoth (Oral hypoglycemic agents + Insulin)Diet alone35147Data are presented as mean ± standard deviation or median (interquartile range), as appropriateHbA1c: glycated hemoglobin; HDL: high-density lipoprotein; LDL: low-density lipoprotein%CV: coefficient of variation; TIR: time in range (70–180 mg/dL)Macronutrient intake is expressed as percentage contribution to total daily macronutrient intake (grams)CGM: continuous glucose monitoring Baseline demographic, clinical, dietary, and CGM characteristics of the study population (n = 47) Oral hypoglycemic agents Only Insulin Both (Oral hypoglycemic agents + Insulin) Diet alone 35 1 4 7 Data are presented as mean ± standard deviation or median (interquartile range), as appropriate HbA1c: glycated hemoglobin; HDL: high-density lipoprotein; LDL: low-density lipoprotein %CV: coefficient of variation; TIR: time in range (70–180 mg/dL) Macronutrient intake is expressed as percentage contribution to total daily macronutrient intake (grams) CGM: continuous glucose monitoring Associations of meal interval duration pattern with CGM-derived glycemic outcomes Participants level analyses demonstrated no significant associations between the breakfast-lunch interval and CGM-derived glycemic metrics. Longer lunch-dinner intervals demonstrated a positive association with TIR (ρ = 0.339, p = 0.023) and inverse trends with TAR and average glucose although not all associations reached statistical significance. Longer breakfast-dinner intervals were positively associated with higher TIR and not reaching statistical significance with %CV, TBR, TAR or average glucose (Table 2). To improve interpretability beyond correlation coefficients, supplement linear regression analysis was performed at the participant level (Supplementary Table 3). Table 2Spearman correlation between meal timing intervals (hours) and glycemic metrics (n = 46)Glycemic metricsBreakfast-LunchLunch-DinnerBreakfast-Dinnerρ p Ρ p ρ p %CV0.1020.507-0.1990.191-0.0940.540TIR-0.0330.8280.3390.0230.3180.033TBR-0.0640.6750.0160.918-0.1290.397TAR0.0930.545-0.2880.055-0.2170.152Average0.0470.757-0.1920.207-0.1520.319< 54 mg/dL-0.1440.344-0.0920.547-0.2880.05554–69 mg/dL-0.0630.6790.0170.911-0.1270.407181–250 mg/dL0.1210.428-0.2700.073-0.1720.260> 250 mg/dL0.0100.948-0.2100.166-0.2370.117Spearman correlation analysis was performed to evaluate the association between meal timing intervals (breakfast–lunch, lunch–dinner, and breakfast–dinner durations) and glycemic metrics derived from continuous glucose monitoring. Values are presented as Spearman correlation coefficient (ρ) and corresponding p-values Spearman correlation between meal timing intervals (hours) and glycemic metrics (n = 46) Spearman correlation analysis was performed to evaluate the association between meal timing intervals (breakfast–lunch, lunch–dinner, and breakfast–dinner durations) and glycemic metrics derived from continuous glucose monitoring. Values are presented as Spearman correlation coefficient (ρ) and corresponding p-values Associations of macronutrient composition, energy intake and Fibre intake with CGM-derived glycemic outcomes Macronutrient composition demonstrated distinct and contrasting associations with glycemic metrics (Table 3). Higher total carbohydrate intake was associated with higher %CV (ρ = 0.354, p = 0.017), higher TAR (ρ = 0.343, p = 0.021), higher mean glucose (ρ = 0.355, p = 0.017), and lower TIR (ρ = −0.378, p = 0.010). These associations were partially reflected at the meal level, particularly for breakfast carbohydrates and TIR (ρ = −0.363, p = 0.014). Higher fat intake was associated with lower glycemic variability (%CV: ρ = −0.318, p = 0.033), lower TAR (ρ = −0.370, p = 0.012), lower mean glucose (ρ = −0.365, p = 0.014), and higher TIR (ρ = 0.441, p = 0.002). Protein intake showed minimal associations with significant relationship with number of Level-1 hyperglycemic episodes (ρ = −0.346, p = 0.020). Table 3Correlation between glycemic metrics & percentage distribution of macronutrient composition (n = 46)GlycemicMetric%Totalρ (p)%Breakfastρ (p)%Lunchρ (p)%Dinnerρ (p) %Carbohydrates %CV0.354 (0.017)0.181 (0.235)0.229 (0.131)0.301 (0.045)TIR-0.378 (0.01)-0.363 (0.014)-0.156 (0.306)-0.087 (0.568)TBR0.01 (0.946)0.012 (0.936)-0.014 (0.925)0.165 (0.278)TAR0.343 (0.021)0.282 (0.061)0.143 (0.348)0.081 (0.599)Average0.355 (0.017)0.274 (0.069)0.131 (0.391)0.088 (0.563)< 540.063 (0.68)0.13 (0.394)0.14 (0.358)0.132 (0.389)54–690.016 (0.915)0.01 (0.951)-0.007 (0.965)0.172 (0.26)181–2500.519 (< 0.001)0.307 (0.041)0.268 (0.075)0.22 (0.147)> 2500.333 (0.026)0.255 (0.091)0.055 (0.721)0.088 (0.565) % Fats %CV-0.318 (0.033)-0.13 (0.393)-0.269 (0.074)-0.183 (0.23)TIR0.441 (0.002)0.400 (0.006)0.37 (0.012)0.173 (0.255)TBR-0.076 (0.621)0.038 (0.806)0.048 (0.754)-0.039 (0.801)TAR-0.37 (0.012)-0.322 (0.031)-0.333 (0.026)-0.188 (0.215)Average-0.365 (0.014)-0.323 (0.03)-0.304 (0.042)-0.219 (0.148)< 54-0.077 (0.614)-0.09 (0.557)-0.092 (0.548)0.021 (0.893)54–69-0.083 (0.588)0.041 (0.789)0.041 (0.791)-0.045 (0.769)181–250-0.431 (0.003)-0.288 (0.055)-0.377 (0.011)-0.245 (0.104)> 250-0.407 (0.006)-0.325 (0.029)-0.265 (0.079)-0.23 (0.129) %Proteins %CV-0.251 (0.096)-0.178 (0.242)0.005 (0.973)-0.207 (0.173)TIR0.198 (0.191)0.214 (0.158)-0.032 (0.836)-0.046 (0.763)TBR0.054 (0.726)-0.109 (0.474)0.064 (0.676)-0.202 (0.184)TAR-0.186 (0.221)-0.141 (0.356)0.04 (0.795)0.087 (0.57)Average-0.214 (0.158)-0.129 (0.4)0.033 (0.829)0.1 (0.514)< 54-0.06 (0.695)-0.133 (0.383)-0.098 (0.521)-0.212 (0.163)54–690.051 (0.74)-0.107 (0.486)0.058 (0.706)-0.205 (0.177)181–250-0.346 (0.02)-0.213 (0.16)-0.092 (0.55)-0.012 (0.935)> 250-0.158 (0.301)-0.12 (0.432)0.1 (0.512)0.095 (0.535)Values are presented as Spearman’s correlation coefficient (ρ) with corresponding p-values. Macronutrient distribution represents the percentage contribution of carbohydrates, fats, and proteins intake, analyzed overall (%Total) and across individual meals (breakfast, lunch, and dinner). Macronutrient % = proportion of total macronutrient grams. Abbreviations: %CV, coefficient of variation; TIR, time in range (70–180 mg/dL); TBR, time below range (< 70 mg/dL); TAR, time above range (> 180 mg/dL); ρ, Spearman correlation coefficient; CGM, continuous glucose monitoring Correlation between glycemic metrics & percentage distribution of macronutrient composition (n = 46) Values are presented as Spearman’s correlation coefficient (ρ) with corresponding p-values. Macronutrient distribution represents the percentage contribution of carbohydrates, fats, and proteins intake, analyzed overall (%Total) and across individual meals (breakfast, lunch, and dinner). Macronutrient % = proportion of total macronutrient grams. Abbreviations: %CV, coefficient of variation; TIR, time in range (70–180 mg/dL); TBR, time below range (< 70 mg/dL); TAR, time above range (> 180 mg/dL); ρ, Spearman correlation coefficient; CGM, continuous glucose monitoring Fibre intake demonstrated varying associations, with higher intake correlating with higher TAR (ρ = 0.362, p = 0.014), higher mean glucose (ρ = 0.359, p = 0.016), and lower TIR (ρ = −0.295, p = 0.049). Energy intake was not significantly associated with most glycemic metrics; however, breakfast energy intake was inversely associated with hypoglycemia (TBR: ρ = −0.342, p = 0.021), while dinner energy intake showed a positive association with hypoglycemia metrics (Table 4). Supplementary subject-level regression analyses were performed and described in Supplementary Table S5. Table 4Correlation between glycemic metrics with fibre and energy intake across meals (n = 46)GlycemicMetricTotalρ (p)Breakfastρ (p)Lunchρ (p)Dinnerρ (p) Fibre Intake(grams) %CV0.165 (0.278)0.182 (0.230)0.316 (0.034)0.090 (0.554)TIR-0.295 (0.049)-0.253 (0.093)-0.367 (0.013)-0.133 (0.385)TBR-0.042 (0.783)-0.434 (0.003)-0.018 (0.908)0.310 (0.038)TAR0.362 (0.014)0.352 (0.018)0.388 (0.009)0.142 (0.351)Average0.359 (0.016)0.392 (0.008)0.379 (0.010)0.100 (0.514)< 54-0.205 (0.177)-0.251 (0.096)-0.149 (0.328)0.028 (0.855)54–69-0.042 (0.785)-0.438 (0.003)-0.015 (0.922)0.314 (0.036)181–2500.395 (0.007)0.303 (0.043)0.462 (0.001)0.108 (0.481)> 2500.375 (0.011)0.300 (0.045)0.375 (0.011)0.211 (0.164) Energy Intake (Kcal) %CV-0.129 (0.397)-0.034 (0.826)0.098 (0.520)0.056 (0.713)TIR0.048 (0.753)0.033 (0.828)-0.081 (0.596)-0.099 (0.516)TBR-0.155 (0.310)-0.342 (0.021)0.205 (0.176)0.388 (0.008)TAR0.040 (0.795)0.070 (0.648)0.087 (0.570)0.054 (0.724)Average0.037 (0.811)0.106 (0.490)0.073 (0.634)-0.017 (0.910)< 54-0.162 (0.287)-0.236 (0.118)0.056 (0.715)0.251 (0.097)54–69-0.162 (0.287)-0.348 (0.019)0.201 (0.185)0.386 (0.009)181–2500.029 (0.852)0.014 (0.928)0.074 (0.630)0.025 (0.869)> 250-0.045 (0.770)0.009 (0.954)0.105 (0.492)0.056 (0.716)Values are presented as Spearman’s correlation coefficient (ρ) with corresponding p-values. Macronutrient distribution represents the percentage contribution of carbohydrates, fats, and proteins intake, analyzed overall (%Total) and across individual meals (breakfast, lunch, and dinner). Macronutrient % = proportion of total macronutrient grams. Abbreviations: %CV, coefficient of variation; TIR, time in range (70–180 mg/dL); TBR, time below range (< 70 mg/dL); TAR, time above range (> 180 mg/dL); ρ, Spearman correlation coefficient; CGM, continuous glucose monitoring Correlation between glycemic metrics with fibre and energy intake across meals (n = 46) Values are presented as Spearman’s correlation coefficient (ρ) with corresponding p-values. Macronutrient distribution represents the percentage contribution of carbohydrates, fats, and proteins intake, analyzed overall (%Total) and across individual meals (breakfast, lunch, and dinner). Macronutrient % = proportion of total macronutrient grams. Abbreviations: %CV, coefficient of variation; TIR, time in range (70–180 mg/dL); TBR, time below range (< 70 mg/dL); TAR, time above range (> 180 mg/dL); ρ, Spearman correlation coefficient; CGM, continuous glucose monitoring Mixed effects regression analysis of dietary composition and meal timing with glycemic variability Mixed-effects linear regression models incorporating participant-specific random intercepts and a first-order autoregressive [AR(1)] residual correlation structure were used to evaluate associations between dietary composition, meal timing variables, and glycemic variability (%CV) while accounting for repeated daily observations. Time-varying dietary exposures were decomposed into participant-specific means (between-person effects) and day-specific deviations from individual means (within-person effects). Models were adjusted for study day, total energy intake (per 100 kcal/day), BMI, HbA1c, and insulin/insulin-secretagogue use. Across models, marginal R² values ranged from 0.10 to 0.11, whereas conditional R² values ranged from 0.65 to 0.66 (Table 5). Table 5Mixed-effects linear regression models examining associations of dietary composition and meal timing with glycemic variabilityPredictorModel 1A β (95% CI)Model 1B β (95% CI)Model 2 β (95% CI)Model 3 β (95% CI)Model 4 β (95% CI)Model 5 β (95% CI)Carbohydrate (between-person)0.728 (-0.010, 1.465)p = 0.0530.445 (-0.584, 1.474)p = 0.3880.629 (0.065, 1.193)p = 0.0300.625 (0.038, 1.213)p = 0.038Carbohydrate (within-person)-0.017 (-0.138, 0.105)p = 0.788-0.059 (-0.223, 0.105)p = 0.483-0.033 (-0.112, 0.046)p = 0.410-0.029 (-0.108, 0.051)p = 0.481Protein (between-person)0.283 (-1.042, 1.608)p = 0.668Protein (within-person)0.042 (-0.193, 0.277)p = 0.725Fat (between-person)-0.283 (-1.608, 1.042)p = 0.668Fat (within-person)-0.042 (-0.277, 0.193)p = 0.725Breakfast–Dinner interval (between-person)0.066 (-3.054, 3.186)p = 0.966Breakfast–Dinner interval (within-person)0.344 (-0.206, 0.895)p = 0.219Breakfast–Lunch interval (between-person)0.374 (-2.921, 3.669)p = 0.8200.373 (-2.768, 3.515)p = 0.811Breakfast–Lunch interval (within-person)0.484 (-0.230, 1.199)p = 0.1830.466 (-0.252, 1.183)p = 0.203Lunch–Dinner interval (between-person)-0.466 (-4.559, 3.627)p = 0.8190.174 (-3.772, 4.119)p = 0.930Lunch–Dinner interval (within-person)0.317 (-0.279, 0.914)p = 0.2960.303 (-0.296, 0.902)p = 0.321Study day-0.077 (-0.260, 0.106)p = 0.410-0.077 (-0.260, 0.106)p = 0.410-0.078 (-0.260, 0.105)p = 0.403-0.086 (-0.269, 0.096)p = 0.354-0.086 (-0.269, 0.096)p = 0.353-0.074 (-0.257, 0.109)p = 0.426Total energy intake (per 100 kcal/day)0.095 (-0.067, 0.258)p = 0.2500.095 (-0.067, 0.258)p = 0.2500.092 (-0.069, 0.253)p = 0.2630.071 (-0.089, 0.230)p = 0.3830.068 (-0.092, 0.228)p = 0.4060.087 (-0.075, 0.249)p = 0.290BMI0.060 (-0.393, 0.512)p = 0.7960.060 (-0.393, 0.512)p = 0.7960.085 (-0.351, 0.521)p = 0.702-0.211 (-0.587, 0.166)p = 0.272-0.210 (-0.590, 0.170)p = 0.2770.081 (-0.366, 0.529)p = 0.721HbA1c0.178 (-1.178, 1.534)p = 0.7920.178 (-1.178, 1.534)p = 0.7920.240 (-1.071, 1.551)p = 0.7130.312 (-1.085, 1.709)p = 0.6540.334 (-1.081, 1.748)p = 0.6360.251 (-1.099, 1.602)p = 0.709Insulin/ Sulfonyl urea use1.922 (-2.714, 6.559)p = 0.4071.922 (-2.714, 6.559)p = 0.4071.864 (-2.715, 6.442)p = 0.4161.779 (-3.115, 6.673)p = 0.4671.542 (-3.514, 6.599)p = 0.5411.852 (-2.977, 6.681)p = 0.443Values are presented as β (95% confidence interval). All models included a participant-specific random intercept and a first-order autoregressive residual correlation structure [AR(1)] to account for repeated daily observations and serial within-subject dependence. Study day was included to account for temporal trends across CGM monitoring days. Time-varying dietary exposures were decomposed into participant-specific means (between-person effects) and day-specific deviations from individual means (within-person effects). All models were adjusted for total energy intake, BMI, baseline HbA1c, and insulin/insulin-secretagogue use. Energy intake was scaled per 100 kcal/day for interpretability. Macronutrient variables were treated as compositional data; therefore, models were parameterized using two macronutrient components while interpreting the omitted macronutrient as the reference component. Potential multicollinearity was assessed using variance inflation factors (VIFs). Although moderate collinearity was observed for one predictor (maximum VIF = 5.64), model coefficients remained stable and were considered acceptable for interpretationModel 1 A: Marginal R² = 0.100; Conditional R² = 0.653Model 1B: Marginal R² = 0.100; Conditional R² = 0.653Model 2: Marginal R² = 0.101; Conditional R² = 0.649Model 3: Marginal R² = 0.040; Conditional R² = 0.656Model 4: Marginal R² = 0.043; Conditional R² = 0.662Model 5: Marginal R² = 0.100; Conditional R² = 0.660 Mixed-effects linear regression models examining associations of dietary composition and meal timing with glycemic variability 0.728 (-0.010, 1.465) p = 0.053 0.445 (-0.584, 1.474) p = 0.388 0.629 (0.065, 1.193) p = 0.030 0.625 (0.038, 1.213) p = 0.038 -0.017 (-0.138, 0.105) p = 0.788 -0.059 (-0.223, 0.105) p = 0.483 -0.033 (-0.112, 0.046) p = 0.410 -0.029 (-0.108, 0.051) p = 0.481 0.283 (-1.042, 1.608) p = 0.668 0.042 (-0.193, 0.277) p = 0.725 -0.283 (-1.608, 1.042) p = 0.668 -0.042 (-0.277, 0.193) p = 0.725 0.066 (-3.054, 3.186) p = 0.966 0.344 (-0.206, 0.895) p = 0.219 0.374 (-2.921, 3.669) p = 0.820 0.373 (-2.768, 3.515) p = 0.811 0.484 (-0.230, 1.199) p = 0.183 0.466 (-0.252, 1.183) p = 0.203 -0.466 (-4.559, 3.627) p = 0.819 0.174 (-3.772, 4.119) p = 0.930 0.317 (-0.279, 0.914) p = 0.296 0.303 (-0.296, 0.902) p = 0.321 -0.077 (-0.260, 0.106) p = 0.410 -0.077 (-0.260, 0.106) p = 0.410 -0.078 (-0.260, 0.105) p = 0.403 -0.086 (-0.269, 0.096) p = 0.354 -0.086 (-0.269, 0.096) p = 0.353 -0.074 (-0.257, 0.109) p = 0.426 0.095 (-0.067, 0.258) p = 0.250 0.095 (-0.067, 0.258) p = 0.250 0.092 (-0.069, 0.253) p = 0.263 0.071 (-0.089, 0.230) p = 0.383 0.068 (-0.092, 0.228) p = 0.406 0.087 (-0.075, 0.249) p = 0.290 0.060 (-0.393, 0.512) p = 0.796 0.060 (-0.393, 0.512) p = 0.796 0.085 (-0.351, 0.521) p = 0.702 -0.211 (-0.587, 0.166) p = 0.272 -0.210 (-0.590, 0.170) p = 0.277 0.081 (-0.366, 0.529) p = 0.721 0.178 (-1.178, 1.534) p = 0.792 0.178 (-1.178, 1.534) p = 0.792 0.240 (-1.071, 1.551) p = 0.713 0.312 (-1.085, 1.709) p = 0.654 0.334 (-1.081, 1.748) p = 0.636 0.251 (-1.099, 1.602) p = 0.709 1.922 (-2.714, 6.559) p = 0.407 1.922 (-2.714, 6.559) p = 0.407 1.864 (-2.715, 6.442) p = 0.416 1.779 (-3.115, 6.673) p = 0.467 1.542 (-3.514, 6.599) p = 0.541 1.852 (-2.977, 6.681) p = 0.443 Values are presented as β (95% confidence interval). All models included a participant-specific random intercept and a first-order autoregressive residual correlation structure [AR(1)] to account for repeated daily observations and serial within-subject dependence. Study day was included to account for temporal trends across CGM monitoring days. Time-varying dietary exposures were decomposed into participant-specific means (between-person effects) and day-specific deviations from individual means (within-person effects). All models were adjusted for total energy intake, BMI, baseline HbA1c, and insulin/insulin-secretagogue use. Energy intake was scaled per 100 kcal/day for interpretability. Macronutrient variables were treated as compositional data; therefore, models were parameterized using two macronutrient components while interpreting the omitted macronutrient as the reference component. Potential multicollinearity was assessed using variance inflation factors (VIFs). Although moderate collinearity was observed for one predictor (maximum VIF = 5.64), model coefficients remained stable and were considered acceptable for interpretation Model 1 A: Marginal R² = 0.100; Conditional R² = 0.653 Model 1B: Marginal R² = 0.100; Conditional R² = 0.653 Model 2: Marginal R² = 0.101; Conditional R² = 0.649 Model 3: Marginal R² = 0.040; Conditional R² = 0.656 Model 4: Marginal R² = 0.043; Conditional R² = 0.662 Model 5: Marginal R² = 0.100; Conditional R² = 0.660 In analyses evaluating dietary composition, higher between-person carbohydrate intake was associated with higher %CV in Models 2 and 5. In Model 2, the between-person carbohydrate effect demonstrated a β coefficient of 0.629 (95% CI: 0.065 to 1.193), whereas the within-person carbohydrate effect was not clearly associated with %CV [β − 0.0333 (95% CI: −0.114 to 0.044)]. Similar findings were observed in Model 5. Protein and fat intake did not demonstrate clear independent associations with %CV after multivariable adjustment. Single-predictor mixed-effects models were additionally conducted to facilitate comparison between crude and adjusted associations (Supplementary Table 6). The crude analyses demonstrated effect directions comparable to those observed in the corresponding multivariable models. The distribution of meal intervals is summarized in Supplementary Table 7. The breakfast–lunch interval demonstrated a mean duration of 4.72 ± 1.07 h, with a median of 4.75 h (IQR: 4.0–5.5) and a range of 1–9.83 h. The lunch–dinner interval averaged 7.09 ± 1.13 h, with a median of 7.0 h (IQR: 6.5–8.0) and a range of 2.68–9.83 h. The breakfast–dinner interval showed a mean duration of 11.81 ± 1.14 h, with a median of 12.0 h (IQR: 11.25–12.5) and a range of 7–14.5 h. Using coefficients derived from the final adjusted mixed-effects models, the expected absolute change in glycemic variability (%CV) across the observed interquartile range of meal intervals was estimated to facilitate clinical interpretation. Across the IQR, the breakfast–lunch interval corresponded to an estimated 0.56%-point increase in %CV, the lunch–dinner interval to a 0.70%-point decreases in %CV, and the breakfast–dinner interval to a 0.08%-point increase in %CV. Following adjustment for participant-level covariates, meal-timing variables were not independently associated with %CV in the final models; however, these estimates indicate that any potential meal-timing effects across realistic exposure ranges were modest in absolute magnitude. Discussion In this observational study, we evaluated the independent and combined effects of meal intervals, meal frequency and dietary composition on CGM-derived glycemic control metrics in adults with T2DM. The study findings suggest that both time-based eating patterns and dietary composition may contribute to short-term glycemic control. However, after adjustment for participant-level clinical covariates, these associations were weakened or non-significant, suggesting that short-term glycemic variability is influenced by a complex interplay of dietary, temporal, and individual clinical factors. In individuals with type 1 diabetes, a moderate-carbohydrate (30% of total energy) diet is associated with lower mean glucose and time above range and higher time in range without increasing the risk of hypoglycemia [21]. In our study, higher carbohydrate intake was associated with less favourable outcomes in unadjusted analyses, whereas higher fat intake showed associations with lower glycemic variability. In adjusted analyses, associations for several macronutrient measures were weakened; however, habitual dietary carbohydrate composition remained relevant to glycemic variability. Meal composition and timing, particularly macronutrient distribution at breakfast, may influence glycemic exposure in adults with T2DM (Table 3). Park et al. reported that a high-protein breakfast reduces the postprandial glucose levels after breakfast and subsequent meals [22]. In our study, meal-specific exploratory analyses suggested that meals with moderated carbohydrate content and adequate protein and fat intake earlier in the day were associated with improved TIR and reduced hyperglycemic exposure (Table 3). The mechanisms underlying these meal-specific differences may include greater suppression of fasting-derived non-esterified fatty acids, delayed gastric emptying, improved satiety and enhanced peripheral insulin responsiveness [23–25]. Higher fibre intake showed contradictory associations, including higher TAR and mean glucose and lower TIR. These inconsistent results with previous studies [26–28], likely reflect residual confounding co-consumption with high carbohydrate foods, or variability in fibre type and meal context rather than a true adverse effect. Energy intake demonstrated limited associations with glycemic metrics suggesting the minimal role of caloric intake alone in glycemic control within the study population. Therefore, glycemic variability may be influenced by multiple factors beyond dietary intake, including pharmacotherapy, insulin sensitivity, physical activity, and behavioural factors. Existing evidence suggests that alternative eating patterns, including meal consolidation or more frequent smaller meals, may reduce postprandial glycemic burden. Imai S et al., reported that mid-afternoon snacks consumed 3–4 hours after lunch was associated with a reduction in the mean amplitude of glycemic excursions (MAGE) [29]. A randomized crossover study conducted under controlled caloric conditions, found that increasing meal frequency from three to six meals per day had improved glycemic control in individuals with early T2DM [30]. Another study reported that dividing the dinner into two smaller meals has been shown to improve glycemic variability [31]. In our study, descriptive analyses indicated that participants consuming snacks in addition to a traditional three-meal pattern (≥ 4 meals/day) demonstrated numerically higher median TIR and lower median %CV compared with those following a three-meal pattern alone (Supplementary Table S1 & Supplementary Table S2). Skipping lunch or consuming a low-calories lunch may increase the postprandial glucose spikes [32]. The association of longer lunch–dinner and breakfast–dinner intervals with higher TIR and lower glycemic variability raises the possibility, that spacing of meals later in the day could influence short-term glycemic stability. Potential mechanisms may include altered postprandial insulin dynamics, differences in substrate utilization, or behavioural changes in meal size and composition across the day. In contrast, the absence of an association with the breakfast–lunch interval suggests that meal-timing relationships may vary across different periods of the day and could be influenced by circadian metabolic responses and meal context. Mixed-effects analyses accounting for repeated observations highlighted the complexity of the relationship between dietary behaviors and short-term glycemic variability in adults with T2DM. The findings suggest that habitual dietary patterns between individuals may be more relevant to glycemic stability than short-term day-to-day fluctuations in meal timing or macronutrient intake. A marked difference between marginal and conditional R² values further indicated that between-subject heterogeneity explained a considerable proportion of variability in glycemic outcomes. As shown in Supplementary Table 8, sequential inclusion of participant-level covariates, including BMI, HbA1c, and insulin/sulfonylurea use, resulted in slight increases in marginal R² (0.081–0.100), whereas conditional R² remained consistently higher across models (0.646–0.660). These observations suggest that the covariates used in this study did not fully explain the observed heterogeneity and additional individual-level metabolic, therapeutic, and behavioral factors may contribute to glycemic variability. The complex nature of glycemic regulation among patients with T2DM highlights the importance of interpreting dietary composition and meal-timing associations within the broader context of individualized clinical management. Strengths of this study include the use of CGM-derived metrics over multiple consecutive days, prospective dietary recording with daily verification, meal-based nutrient analysis using an Indian food composition database, and the application of mixed-effects models to account for intra-individual variability. The focus on meal-specific macronutrient distribution provides clinically relevant insights beyond total daily intake. The observational design precludes causal inference, and residual confounding cannot be excluded. Dietary intake was self-reported and subject to measurement error, particularly with respect to portion size estimation and recipe variability. The single-centre design may limit generalizability, and the duration of CGM monitoring may not fully capture longer-term glycemic adaptations. The modest participant sample size may limit model stability and increase the risk of overfitting, despite restricting adjustment to a limited number of prespecified clinically relevant covariates. Although key participant-level covariates including BMI, HbA1c, and insulin/sulfonylurea use were incorporated into multivariable models, residual confounding from unmeasured factors cannot be excluded. Physical activity, medication dose/intensity, sleep patterns, and other behavioural factors may have influenced the observed associations. Nutrient estimation relied on database approximations and may introduce systematic measurement error. Future randomized and interventional studies are needed to determine whether targeted modification of meal-specific macronutrient composition and meal timing can causally improve CGM-derived glycemic outcomes. Conclusion Short-term glycemic variability in T2DM reflects a complex interplay between dietary behaviours and individual clinical characteristics. Habitual carbohydrate composition appears more consistently associated with glycemic variability than meal timing after adjustment for clinical factors. Further studies are needed to clarify whether modifying meal timing can improve CGM-derived outcomes. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 2 Abbreviations Type 2 diabetes mellitus Continuous glucose monitoring Glycemic variability Time in range Time above range Time below range Coefficient of variation Glycated hemoglobin Fasting plasma glucose Postprandial plasma glucose Body mass index High-density lipoprotein Low-density lipoprotein Very low-density lipoprotein Standard deviation Interquartile range Confidence interval Regression coefficient Spearman correlation coefficient Coefficient of determination Author contributions MAH: Conceptualization, Methodology, Data curation, Formal analysis, Writing-original draft, AKH: Investigation, Resources, Validation, Writing- Review & Editing, ML: Supervision, project administration, Visualization, Writing- Review & Editing. All authors have read and approved the final version of the manuscript. Funding This study received partial financial support from BioQem Botanicals Pvt. Ltd. Data availability The data that supports the findings of this study are available with corresponding author upon reasonable request. Declarations Ethical approval and consent to participate The study was approved (Ref- No. CDSIMER/MR/0104/IEC2023, dated 26-12-2023) by the Institutional Ethics Committee (CSDISMER-IEC) of Dr. Chandramma Dayananda Sagar Institute of Medical Education and Research (CDSIMER), Bengaluru, India (Regn. No. ECR/1628/Inst/KA/2021). Written informed consent was obtained from all participants prior to their inclusion in the study. Consent for publication Not applicable. Declaration of use of AI and AI-assisted technologies During the preparation of this manuscript, the authors used AI tools like ChatGPT to improve the language and readability. After using this tool, the authors reviewed and edited the content as needed and took full responsibility for the publication’s content. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. References Associated Data Supplementary Materials Supplementary Material 1 Supplementary Material 2 Data Availability Statement The data that supports the findings of this study are available with corresponding author upon reasonable request.