Impact of visceral adiposity on glycemic variability in patients with insulin-treated type 2 diabetes undergoing hemodialysis

Abstract

BACKGROUND

Glycemic variability (GV) is increasingly recognized as a major metabolic risk in patients with type 2 diabetes (T2D) undergoing hemodialysis (HD); however, key contributors remain unclear.

AIM

To identify clinical predictors of GV – measured by differences in time in range (ΔTIR) – between HD and off-HD days in patients with insulin-treated T2D.

METHODS

In this prospective study, 35 patients with insulin-treated T2D on online HD were monitored using continuous glucose monitoring over 20 days. Glucometric variables included mean glucose, TIR, and glucose management indicator. Patients were stratified by ΔTIR into low (< 5%) or high (> 5%) fluctuation groups. Body composition was assessed using multifrequency bioelectrical impedance analysis, including visceral fat area (VFA), skeletal muscle index, and extracellular-to-total body water ratio.

RESULTS

Patients with ΔTIR > 5% showed higher glucose levels (HD: 187.1 ± 44.3 mg/dL; off-HD: 201.9 ± 52.3 mg/dL), reduced TIR, elevated glucose management indicator (8.02% ± 1.14%), and significantly lower VFA (63.4 ± 42.7 cm²vs 127.0 ± 49.5 cm²; P = 0.001). Skeletal muscle index and hydration parameters were not different. VFA independently predicted ΔTIR > 5% (odds ratio = 0.077; P = 0.016), and a VFA threshold of approximately 63 cm² yielded 88% sensitivity and 67% specificity (area under the curve of 0.812; P = 0.003).

CONCLUSION

VFA is a strong independent marker of GV in patients with insulin-treated T2D on HD, supporting body composition profiling in this population.

Key Words: Type 2 diabetes; Hemodialysis; Visceral adiposity; Continuous glucose monitoring; Time in range; Glycemic stability; Body composition; Metabolic fragility

Core Tip: We conducted a prospective study using 20 days of continuous glucose monitoring to better understand glycemic instability in insulin-treated type 2 diabetes patients undergoing hemodialysis. Our results show that low visceral fat area is a strong and independent determinant of larger fluctuations in time-in-range between dialysis and non-dialysis days. Patients with low visceral adiposity had higher mean glucose, lower time-in-range, and more pronounced post-dialysis hyperglycemia. Visceral fat was the only variable that remained significant after multivariate adjustment. These findings suggest that visceral adiposity may act as a metabolic buffer, and that patients with low visceral fat area represent a fragile phenotype who may benefit from closer monitoring and more personalized diabetes management in the hemodialysis.

INTRODUCTION

Type 2 diabetes (T2D) mellitus remains the leading cause of chronic kidney disease and end-stage kidney disease (ESKD) requiring hemodialysis (HD)[1-4]. Among patients on HD, the prevalence of T2D is alarmingly high, exceeding 29%, and is associated with significantly increased cardiovascular morbidity and mortality[1,5-8]. This elevated risk is attributed to chronic inflammation, fluid overload, heart failure, and substantial glycemic instability in seven articles[1,3,4,9-12].

Online hemodiafiltration (OL-HDF), a modality that combines convection and diffusion for solute removal, provides advantages over conventional HD, including improved uremic toxin clearance and better cardiovascular outcomes[13,14]. However, the impact of OL-HDF on glycemic patterns remains under-characterized. While HD generally improves insulin sensitivity and lowers glucose levels[14], off-HD days are often marked by worsened glycemic control and increased insulin resistance[15,16]. Moreover, the process of dialysis itself contributes significantly to glycemic instability by removing both insulin and glucose from circulation[17-20].

The emerging use of continuous glucose monitoring (CGM) has contributed to the understanding of glycemic dynamics in this population. Characteristic patterns, such as intra-HD hypoglycemia followed by post-HD hyperglycemia, have been widely documented[3,16,21]. By contrast, some patients experience prolonged hypoglycemia even after HD[15]. The specific variables accounting for these glycemic excursions remain unclear, and despite the lack of large clinical trials, current guidelines support the use of CGM in patients on dialysis[22,23].

Alongside glucose fluctuations, fluid overload remains a major contributor to cardiovascular complications in patients on HD[11,12]. Bioimpedance analysis (BIA) offers a reliable, non-invasive tool to assess fluid status and body composition, including intracellular water (ICW) and extracellular water (ECW), total body water (TBW), and the ECW/ICW ratio.

Thus, the current study examined the main risk factors, including glycemic parameters assessed by CGM and body composition evaluated by BIA, accounting for glycemic instability between HD and off-HD days in insulin-treated T2D.

MATERIALS AND METHODS

Study population

This prospective study included 35 patients treated with insulin with T2D and ESKD undergoing OL-HDF. Patients older than 18 years were included, whereas those with type 1 diabetes or other forms of diabetes were excluded. We excluded patients with recent hospitalization (< 1 month), active infection, active malignancy, or major limb amputation, since these conditions could influence glycemic control or compromise the accuracy of BIA measurements. Patients with severe acute illness or those unable to complete CGM were also excluded. To minimize eventual confounding factors, patients followed standardized meal patterns during the 20-day CGM period, and all insulin regimens remained unchanged for at least 1 month before enrollment. None of the participants were receiving corticosteroids. A small number of patients were treated with glucagon-like peptide-1 receptor agonist (GLP-1RA) (n = 4) or sodium-glucose cotransporter-2 inhibitor (SGLT2i) (n = 3); these therapies had remained stable for ≥ 3 months and were evenly distributed between differences in time in range (ΔTIR) groups, minimizing their potential confounding effect. Physical activity levels were not formally measured, but activity patterns in patients on HD were characteristically stable and low. Clinical data collected included sex, age, weight, dry weight, and hydration status. The study was approved by the Ethics Committee for Research with Medicines of the Vall d’Hebron University Hospital (Barcelona, Spain). All procedures followed the Declaration of Helsinki and participants provided informed consent.

Laboratory analysis

Blood samples were drawn using Vacutainer™ tubes (Becton Dickinson, NJ, United States), processed in additive-free or EDTA-containing tubes, and centrifuged at 1500 g for 15 minutes at room temperature. The biochemical profile included measurements of glucose, hemoglobin, glycated hemoglobin (HbA1c), parathyroid hormone, ferritin, albumin, calcium (Ca), phosphorus (Pi), and the Ca-phosphorus product. Dialysis adequacy/urea clearance (Kt/V), a measure of dialysis adequacy, was monitored monthly and maintained at > 1.2.

OL-HDF

All participants underwent OL-HDF three times per week at Diaverum Hemodialysis Centers, with each session lasting approximately 4 hours. Dialyzers were single-use and individually selected based on body surface area, utilizing either polysulfone or H-polymer membranes (1.9 m² or 2.1 m²). The dialysis fluid composition included sodium (139 mmol/L), potassium (2 mmol/L), bicarbonate (39 mmol/L), Ca (1.25 mmol/L), and glucose (100 mg/dL). Blood flow rates averaged 381.14 ± 20.85 mL/minute, with a dialysate flow of 600 ± 100 mL/minute. Convective clearance was supported by substitution fluid generated online. Ultrafiltration was set at 4000 ± 200 mL/hour, resulting in an average fluid removal of approximately 2.32 ± 0.68 kg per session.

Multifrequency BIA

Body composition was assessed after HD using the InBody S10 multifrequency BIA (MF-BIA). This analyzer uses a tetrapolar eight-point tactile electrode system, which separately measures impedance of five segments of the body – legs, arms, and trunk – using six different frequencies (1 kHz, 5 kHz, 50 kHz, 250 kHz, 500 kHz, and 1000 kHz). The study used six frequencies to predict the distribution of fluid—including TBW, ECW, ICW, and the ECW/TBW ratio—across various body segments. Normal ECW/TBW ratio is defined as < 0.395 in patients with T2D and < 0.385 in the non-diabetic ESKD control group[24,25]. Visceral fat area (VFA) was calculated based on segmental analysis. VFA is expressed in cm²; values ≥ 100 cm² are considered indicative of visceral obesity, while < 100 cm² is considered non-visceral obesity. This cutoff has been validated against computed tomography and widely adopted in metabolic and cardiovascular research, where multiple studies have confirmed that VFA ≥ 100 cm² identifies individuals with increased visceral adiposity and cardiometabolic risk[26-29]. The abdominal waist (AW) circumference and phase angle were assessed.

These parameters provided insights into fluid balance, nutritional status, and body composition. MF-BIA was performed after CGM initiation in all study participants, and 30 minutes after finalizing HD session.

CGM

All participants underwent CGM using the Dexcom One Plus^® system over a 20-day period, which included both dialysis and non-dialysis days. The CGM device was inserted and calibrated by trained clinical staff according to manufacturer guidelines. Glucose readings were automatically recorded every 5 minutes, allowing for detailed tracking of glycemic excursions throughout the study period. Participants were instructed to maintain their usual dietary patterns and insulin regimens to avoid influencing glucose dynamics. No significant changes in dialysis prescriptions were made during the monitoring period.

Glycemic metrics derived from CGM included mean glucose, TIR (70-180 mg/dL), time above range (TAR) (TAR1 > 180 mg/dL; TAR2 > 250 mg/dL), time below range (TBR) (TBR1 < 70 mg/dL; TBR2 < 54 mg/dL), standard deviation (SD), coefficient of variation, and the glucose management indicator (GMI).

Special attention was given to ΔTIR, defined as the absolute difference in TIR between HD and non-HD days. A ΔTIR > 5% was considered indicative of greater fluctuation in glucose control across the dialysis session and was used as the primary outcome to stratify patient profiles.

Data were exported and analyzed using Dexcom Clarity^® software, and missing CGM data of less than 10% were considered acceptable. Periods of sensor malfunction or detachment were excluded. All patients had at least 14 full days of CGM data with more than 90% data capture to ensure reliability.

Statistical analyses

Statistical analyses were performed using IBM-SPSS Statistics version 27.0 and GraphPad Prism 9.0. Descriptive statistics were presented as the mean ± SD or median (interquartile range) for continuous variables and as percentages for categorical variables. Normality was assessed using the Shapiro-Wilk test. Bivariate analyses were validated with non-parametric methods. Spearman correlation was employed to evaluate relationships between OI-HDF parameters and glycemic metrics. All variables with P < 0.10 in univariate analyses, together with clinically relevant parameters, were entered into the multivariate logistic regression model. Variables were removed sequentially using backward elimination when P > 0.10, resulting in VFA > 100 cm² as the only independent predictor of ΔTIR > 5%. The clinical cutoff of 100 cm² was used solely to define visceral obesity, whereas the receiver operating characteristic (ROC)-derived cutoff of 63 cm² reflected the optimal threshold for predicting ΔTIR > 5% in this cohort.

RESULTS

Baseline characteristics

The baseline characteristics of patients with T2D and ESKD OL-HDF is summarized in Supplementary Table 1. Diabetic kidney disease was the most common cause of ESKD, accounting for 73% of cases, followed by multifactorial etiologies such as hypertension and obesity (10.8%), while the etiology remained unknown in 16.2% of patients.

Regarding insulin regimens, 66.7% of patients were treated with basal-bolus regimen, while 33.3% received premixed insulin. Additionally, 11.4% of patients were treated with a GLP-1RA and 8.6% were receiving a SGLT2i, as part of a combined therapeutic strategy. Only 3 patients were receiving both GLP-1RA and SGLT2i therapies simultaneously.

Interestingly, when assessing visceral adiposity, traditional body mass index (BMI)-based criteria identified obesity in only 24% of the cohort. However, when using waist-to-height ratio (WHtR) and VFA measured by MF-BIA, the prevalence of obesity increased to 44% and 52%, respectively, highlighting the limitations of BMI in detecting central adiposity in this population[30-32].

Glucometric parameters comparing days on HD and days without HD

Table 1 shows that glycemic control was significantly better on HD days compared to off-HD days. Mean glucose levels were lower (172.15 ± 41.53 mg/dL vs 182.49 ± 48.92 mg/dL; P = 0.004), with a higher percentage of TIR (62.68% vs 57.06%; P = 0.008) and reduced TAR1 (22.72% vs 26.63%; P = 0.017). Also, glucose SD tended to be lower in off-HD days (P = 0.004). No differences were observed in hypoglycemic episodes (TBR1 and TBR2) or in extreme hyperglycemia (TAR2) on HD days compared to off-HD days. Additionally, along the study GMI was consistently higher than HbA1c (7.31% ± 0.78% vs 6.98% ± 1.38%; P = 0.001), emphasizing the limitations of HbA1c in capturing short-term glycemic fluctuations in this population.

Table 1 Glucometrics data of type 2 diabetes with end-stage kidney disease between hemodialysis days and off-hemodialysis for 20 days, mean ± SD.

Glucometrics by continuous glucose monitoring	HD	Off-HD	P value
Mean continuous glucose monitoring (mg/dL)	172.15 ± 41.53	182.49 ± 48.92	0.004
SD (mg/dL)	52.93 ± 16.24	49.37 ± 15.59	0.004
Time in range (%)	62.68 ± 19.93	57.06 ± 23.05	0.008
TAR1 (> 180 mg/dL) (%)	22.72 ± 9.15	26.63 ± 10.61	0.017
TAR2 (> 250 mg/dL) (%)	13.41 ± 14.04	14.64 ± 17.97	0.17
TBR1 (< 70 mg/dL) (%)	1.65 ± 2.91	0.95 ± 1.27	0.25
TBR2 (< 54 mg/dL) (%)	0.75 ± 2.71	0.31 ± 0.70	0.60
Glucose management indicator vs glycated hemoglobin (%)	7.31 ± 0.78 vs 6.98 ± 1.38		0.001

HD: Hemodialysis; Off-HD: Non-hemodialysis days; SD: Standard deviation; TAR: Time above range; TBR: Time below range.

Clinical variables associated with changes in TIR between HD and off-HD days

To determinate potential clinical and analytical variables that influence glycemic instability between HD and off-HD days, patients were stratified according to ΔTIR (< 5% vs > 5%). As shown in Table 2, both groups exhibited comparable age, diabetes duration, insulin dose per kg, and antidiabetic treatment regimens, including basal-bolus insulin, GLP1-RA and SGLT2i.

Table 2 Baseline characteristics among differences in time in range (< 5% vs > 5%) type 2 diabetes with end-stage kidney disease groups, n (%)/mean ± SD/median (interquartile range).

Clinical variables	ΔTIR < 5% (n = 17)	ΔTIR > 5% (n = 18)	P value
Age (years)	74.19 ± 7.22	71.65 ± 10.35	0.31
Sex (male/female)	37.5/62.5	72.2/27.8	0.29
Diabetes duration (years)	18.93 ± 7.03	17.87 ± 6.68	0.85
Insulin (UI/kg/day)	0.30 ± 0.14	0.39 ± 0.26	0.71
Insulin regime (basal bolus/insulin pre-mixed) (%)	52.90/47.10	77.77/22.22	0.27
Glucagon-like peptide-1 receptor agonist	2 (11.76)	2 (11.11)	0.73
Sodium-glucose cotransporter-2 inhibitor	2 (11.76)	1 (5.5)	0.46
Diabetic nephropathy (%)	75	83.3	0.70
Diabetic retinopathy (%)	56.3	64.7	0.49
Heart failure (%)	46.7	50	0.47
Cerebrovascular disease (%)	50	29.4	0.61
Peripheral vascular disease (%)	12.5	29.4	0.37
Diabetic polyneuropathy (%)	12.5	23.5	0.67
Abdominal waist (cm)	91.06 ± 12.51	73.50 ± 10.33	< 0.001
Dry weight (kg)	74.61 ± 15.76	64.91 ±15.97	0.06
Body mass index (kg/m2)	29.34 ± 6.58	24.56 ± 3.97	0.011
Waist-to-height ratio	0.56 ± 0.70	0.46 ± 0.60	0.001
Analytical variables
Hemoglobin (g/L)	11.63 ± 0.67	10.98 ± 1.40	0.45
Ferritin (ng/mL)	473.80 ± 218.30	333.24 ± 245.92	0.59
Glycated hemoglobin (%) (mmol/mol)	6.45 ± 0.79 (46.99 ± 8.63)	7.42 ± 1.62 (57.67 ± 17.71)	0.05
Triglycerides (mg/dL)	138.00 (86.00, 197.00)	114.00 (83.50, 164.00)	0.73
Parathyroid hormone (pg/mL)	440.60 ± 324.63	325.29 ± 258.24	0.41
Ca (mg/dL)	8.87 ± 0.56	8.72 ± 0.63	0.47
Pi (mg/dL)	4.50 ± 1.09	4.33 ± 1.30	0.14
Ca-Pi product (product)	40.13 ± 10.33	35.13 ± 10.63	0.24
Albumin (g/dL)	3.77 ± 0.24	3.67 ± 0.42	0.16
Systemic immune-inflammation Index	495.49 (394.86, 636.46)	650.65 (429.96, 1170.21)	0.80
Multifrequency bioelectrical impedance analysis
TBW (L)	36.13 ± 5.78	33.65 ± 4.99	0.43
Intracellular water (L)	21.99 ± 3.63	19.94 ± 3.42	0.27
ECW (L)	14.14 ± 2.16	14.40 ± 3.06	0.81
ECW/TBW ratio	0.391 ± 0.006	0.400 ± 0.018	0.10
Visceral fat area (cm2)	127.02 ± 49.52	63.40 ± 42.73	0.001
Total PA (degrees)	4.70 ± 0.62	4.16 ± 1.05	0.13
Chest PA (degrees)	7.54 ± 2.37	6.63 ± 2.53	0.19
Skeletal muscle index	7.65 ± 0.73	7.55 ± 1.49	0.63
Online hemodiafiltration
Vascular access (arteriovenous fistula/central venous catheter) (%)	37.5/62.5	52.9/47.1	0.63
Dialyzer (polysulfone /H-polymer) (%)	100	100	0.42
Dialyzer (1.9/2.1 m2) (%)	62.5/37.5	78.6/21.4	0.10
Weight pre-post hemodialysis (kg)	2.32 ± 0.66	2.29 ± 0.76	0.94
Substitution volume (L)	28.97 ± 9.24	27.37 ± 5.96	0.33
Arterial pressure (mmHg)	-143.75 ± 77.96	-144.79 ± 91.65	0.74
Venous pressure (mmHg)	157.18 ± 23.60	160.92 ± 21.71	0.86
Blood flow rate (mL/minute)	386.66 ± 16.72	371.11 ± 24.11	0.23
Dialysis adequacy/urea clearance	2.15 ± 0.40	1.98 ± 0.31	0.69

P < 0.05 considered statistically significant. Ca: Calcium; ECW: Extracellular water; PA: Phase angle; Pi: Phosphorus; TBW: Total body water; ΔTIR: Differences in time in range.

However, patients with ΔTIR < 5% presented significantly longer time in dialysis, higher BMI (29.3 ± 6.6 kg/m²vs 24.6 ± 4.0 kg/m², P = 0.011), greater AW (91.1 ± 12.5 cm vs 73.5 ± 10.3 cm; P < 0.001), and higher WHtR (0.56 ± 0.07 vs 0.46 ± 0.06; P = 0.001). In addition, VFA was nearly twice as high in this group (127.0 ± 49.5 cm²vs 63.4 ± 42.7 cm²; P = 0.001), supporting a predominant phenotype of visceral adiposity. HbA1c values were also significantly higher in the high-ΔTIR group (7.42% ± 1.62% vs 6.45% ± 0.79%; P = 0.050), indicating poorer baseline glycemic control among patients with greater TIR variability. No significant differences were observed in ECW/TBW ratio, hemoglobin, parathyroid hormone and albumin. Likewise, dialysis-related parameters, including blood flow rate, Kt/V, or ultrafiltration volume, were similar between groups.

Glycometric comparisons between patients with TIR < 5% vs TIR > 5%

As shown in Table 3, patients with low ΔTIR < 5% maintained stable glycemic profiles between HD and off-HD days. Their mean glucose levels (158.55 ± 33.69 mg/dL on HD vs 161.30 ± 35.80 mg/dL off-HD), SD (48.70 ± 13.40 mg/dL vs 44.66 ± 10.51 mg/dL), and GMI (7.22% ± 0.80%) remained consistent, with only minimal change in TIR (67.16% ± 19.45% vs 64.91% ± 23.91%).

Table 3 Glucometrics between differences in time in range of patients with type 2 diabetes on hemodialysis and off-hemodialysis days, mean ± SD.

	ΔTIR < 5%		ΔTIR > 5%
	HD	Off-HD	HD	Off-HD	P value
Mean continuous glucose monitoring (mg/dL)	158.55 ± 33.69	161.30 ± 35.80	187.09 ± 44.27	201.85 ± 52.26	1
SD (mg/dL)	48.70 ± 13.40	44.66 ± 10.51	60.91 ± 20.04	41.25 ± 9.56	0.53
TIR (%)	67.16 ± 19.44	64.91 ± 23.91	55.30 ± 16.48	48.73 ± 19.88	1
TAR1 (> 180 mg/dL) (%)	21.97 ± 10.55	23.21 ± 11.00	24.07 ± 7.68	30.11 ± 9.57	1
TAR2 (> 250 mg/dL) (%)	9.10± 12.75	9.56 ± 15.86	17.93 ± 14.27	19.96 ± 18.93	1
TBR1 (< 70 mg/dL) (%)	1.79 ± 1.90	1.23 ± 1.33	1.62 ± 3.70	0.76 ± 1.22	0.67
TBR2 (< 54 mg/dL) (%)	0.42 ± 1.0	0.29 ± 0.49	1.09 ± 3.69	0.35 ± 0.87	0.10
Coefficient of variation (%)	31.28 ± 5.79		34.26 ± 8.51		0.10
Glucose management indicator (%)	7.22 ± 0.80		8.02 ± 1.14		2

¹Differences found hemodialysis and off-hemodialysis days in differences in time in range > 5%.

²Differences found between differences in time in range groups.

HD: Hemodialysis; SD: Standard deviation; TAR: Time above range; TBR: Time below range; ΔTIR: Differences in time in range.

By contrast, the ΔTIR > 5% group showed significantly higher glycemia both during HD (187.09 ± 44.27 mg/dL) and off-HD (201.85 ± 52.26 mg/dL; P < 0.001), alongside greater SD (60.91 ± 20.04 mg/dL on HD vs 41.25 ± 9.56 mg/dL off-HD). Their GMI was also significantly elevated (8.02% ± 1.14%; P = 0.0017). Only ΔTIR > 5% group exhibited a significant reduction in TIR between HD and off-HD days (55.30% ± 16.48% vs 48.73% ± 19.88%; P = 0.039), along with a marked increase in TAR 1 (> 180 mg/dL) and TAR2 (> 250 mg/dL) during off-HD (30.11% ± 9.57% vs 24.07% ± 7.68 %; P = 0.04) and (19.96% ± 18.93% vs 17.93% ± 14.27%; P = 0.01).

These results suggest that patients with higher ΔTIR experience more pronounced glucose instability, driven by elevated glycemia, reduced TIR, and increased hyperglycemic episodes during dialysis-free periods. By contrast, patients with lower ΔTIR demonstrate a more consistent and favorable glycemic pattern across dialysis days.

Glycemic patterns between ΔTIR < 5% and ΔTIR > 5% groups across dialysis phases

Finally, we evaluated glycemic profiles in patients stratified by glycemic variability (GV), comparing ΔTIR < 5% vs ΔTIR > 5% across key time points: (1) Before HD; (2) During HD; (3) 3 hours post-HD; and (4) Overnight following HD. Both groups displayed a similar glycemic response during HD; however, significant differences emerged after the dialysis session. Patients with ΔTIR > 5% exhibited a sustained and pronounced rise in post-HD glucose levels that persisted overnight, compared with the ΔTIR < 5% group (Figure 1).

Figure 1 Glycemic profiles during hemodialysis and non-hemodialysis days in patients stratified by glycemic variability. ^aP < 0.001 for differences in time in range (ΔTIR) < 5% vs ΔTIR > 5% at nighttime glycemia on hemodialysis days; ^bP < 0.05 for ΔTIR < 5% vs ΔTIR > 5% at daytime glycemia on non-hemodialysis days; ^cP < 0.05 for ΔTIR < 5% vs ΔTIR > 5% at nighttime glycemia on non-hemodialysis days. CGM: Continuous glucose monitoring; Glyc-HD: Glycemia during hemodialysis; GlycNight-HD: Nighttime glycemia on hemodialysis days; GlycOffHD-Day: Daytime glycemia on non-hemodialysis days; GlycOffHD-Night: Nighttime glycemia on non-hemodialysis days; Glyc3hPost-HD: Glycemia 3 hours post-hemodialysis; GlycPre-HD: Glycemia before hemodialysis; HD: Hemodialysis; ΔTIR: Differences in time in range.

This unfavorable pattern extended to non-HD days, where the ΔTIR > 5% group maintained significantly higher daytime glycemia (daytime glycemia on non-HD days: 201.9 ± 52.3 mg/dL vs 161.3 ± 35.8 mg/dL; P < 0.01) and nighttime glycemia (nighttime glycemia on non-HD days: 189.6 ± 60.5 mg/dL vs 156.6 ± 24.4 mg/dL; P < 0.01) compared with the ΔTIR < 5% group.

Predictors of GV between HD and off-HD days

To explore which variables were associated with greater ΔTIR > 5% between HD and off-HD days, a univariate logistic regression analysis was performed.

As shown in Supplementary Table 2 markers of adiposity such as WHtR, VFA (both as a continuous variable and as a categorical cut-off > 100 cm²) were significantly associated with a lower risk of ΔTIR < 5%. By contrast, just a trend (P = 0.86) without reach statistical significance was observed with BMI.

Based on the results of the univariate analysis, variables with a P < 0.10 were selected for inclusion in the multivariate logistic regression model (Supplementary Table 3). These included VFA, insulin-regimen, PAT, and ECW/TBW ratio. In this adjusted model, only VFA > 100 cm² remained significantly associated with a lower likelihood of glycemic instability (odds ratio = 0.077, 95% confidence interval [CI]: 0.010-0.61; P = 0.016). This finding suggest that increased visceral adiposity, despite often being considered a risk factor, may paradoxically be associated with lower metabolic variability in this particular population.

To identify an optimal cutoff for VFA associated with increased ΔTIR > 5%, ROC curve analysis was performed (Figure 2). The area under the curve was 0.812 (95%CI: 0.645-0.978; P = 0.003), indicating good discriminative capacity. The best VFA cutoff point was 63 cm², which provided a sensitivity of 88% and a specificity of 67% (Supplementary Table 4). Patients with a VFA below this threshold exhibited a higher likelihood of presenting greater glycemic instability between HD and non-HD days.

Figure 2 Receiver operating characteristic curve to determinate the cutoff of visceral fat area and glycemic variability. According to the coordinates of the curve, a visceral fat area threshold of approximately 63 cm² provided the best balance between sensitivity (88%) and specificity (67%). ROC: Receiver operating characteristic.

DISCUSSION

This study provides novel insights into the determinants of glycemic instability in insulin-treated patients with T2D undergoing HD. Notably, differences in glycemic instability between patients with ΔTIR > 5% and ΔTIR < 5% were not attributable to traditional clinical factors such as age, diabetes duration, or skeletal muscle index. Instead, visceral adiposity emerged as a central factor. Patients with lower VFA experienced greater glycemic instability, both during and outside of HD sessions. Importantly, VFA, assessed by MF-BIA, showed a strong predictive value for ΔTIR > 5%, with an area under the ROC curve of 0.812 (95%CI: 0.645-0.978; P = 0.003). The optimal cutoff point identified was approximately 63 cm², which provided a sensitivity of 88% and specificity of 67%. These findings reinforce the clinical utility of VFA as a non-invasive marker to identify patients at higher risk of metabolic instability. Surprisingly, skeletal muscle index was comparable between groups, suggesting that sarcopenia, a common feature in the HD population[33], may not play a major role in glycemic instability in this context. In addition, we did not find that other parameters such as the ECW/TBW ratio – while associated with inflammation and fluid overload[25,34] independently predicted glycemic instability in the multivariate model. Taken together, these observations suggest that, in this specific cohort, body fat distribution, rather than muscle mass or hydration status, predominantly influences glycemic instability.

Recent data highlight that patients undergoing HD are exposed not only to metabolic stress but also to a substantial psychosocial burden that critically affects glycemic stability. Depression, frailty, and poor nutritional status are highly prevalent in this population and have been identified as major determinants of glucose dysregulation. A recent multicenter study demonstrated a strong association between the Geriatric Nutrition Risk Index and depressive symptoms in patients undergoing HD, underscoring the interaction between nutritional reserve, emotional health, and metabolic instability in ESKD populations[35]. These findings support the concept that metabolic phenotyping in HD must integrate body composition, emotional health, and nutritional status to better understand GV.

From a glycometric perspective, patients with ΔTIR < 5% exhibited lower mean glucose levels, higher TIR percentages, and minimal fluctuation between HD and off-HD days. By contrast, the group with ΔTIR > 5% showed significant deterioration in glycemic control during non-dialysis days, characterized by higher glucose means and an increased proportion of time spent in severe hyperglycemia (TAR2 > 250 mg/dL). These findings align with previous descriptions of a biphasic pattern in patients undergoing HD—hypoglycemia during HD followed by rebound hyperglycemia[1,3,17]—but our study extends this knowledge by linking these fluctuations directly to visceral fat reserves.

A plausible explanation for our finding that lower VFA than 63 cm² is associated with greater glucose instability is that at least a minimal amount of visceral adipose tissue is necessary to buffer acute metabolic stress, particularly in insulin-treated patients on HD who are prone to hypoglycemia. Experimental studies have shown that insulin-induced hypoglycemia rapidly activates counterregulatory lipolysis and increases circulating non-esterified fatty acids availability, a response required to support hepatic glucose output and preserve energy balance[36,37]. When visceral fat stores are reduced, this lipid-mobilizing capacity may be blunted, limiting the ability to generate adequate non-esterified fatty acids during hypoglycemia and impairing metabolic compensation through gluconeogenesis. In addition, some studies indicate that individuals with very low VFA exhibit reduced insulin secretion capacity and altered insulin sensitivity independent of BMI[38], a phenotype compatible with the “metabolic frailty” observed in our high-ΔTIR group. Taken together, these findings suggest that insufficient VFA may create a state of impaired counter-regulation, limited lipid buffering, and reduced metabolic flexibility, ultimately amplifying glycemic fluctuations across the dialysis cycle.

Our findings also align with qualitative evidence showing that CGM captures clinically relevant challenges in this population. A recent qualitative study documented that meal avoidance around HD sessions, glucose loss through dialyzers, fear of hypoglycemia, and post-dialysis fatigue contribute substantially to glycemic fluctuations among insulin-treated T2D patients on HD[39]. These patient-reported experiences reinforce the relevance of objective markers, such as VFA, to identify patients vulnerable to glycemic instability.

Moreover, our multivariate analysis identified VFA as the strongest independent predictor of glycemic instability, even after adjusting for potential confounders such as AW circumference, BMI, and dialysis adequacy (Kt/V). These findings are in agreement with prior hypotheses that VAF may exert a buffering metabolic effect – potentially through preserved insulin resistance pathways that avoid wide glucose excursions[40] – as opposed to patients with low visceral adiposity who lack energy reserves.

Finally, we observed that markers of dialysis quality (blood flow rate, Kt/V) were not significant independent predictors of glycemic instability. This support previous reports describing the absence of a direct correlation between dialysis adequacy and glycemic control[41]. Similarly, studies observing negative correlations between blood glucose and the urea reduction ratio support the idea that improved clearance may influence glucose homeostasis[42], though its impact seems limited in our cohort. Overall, considering recent evidence linking nutritional status, mood disorders, and patient-reported burden during HD sessions[35,39], our findings position VFA as a clinically meaningful marker that integrates metabolic resilience with the complex behavioral and physiological stressors of dialysis.

This study had several two major limiting factors. First, it was a single-center study with a relatively modest sample size, which may limit generalizability. However, similar sample sizes are common in CGM studies in HD because patients’ frailty makes challenging a long-term monitoring. Nevertheless, it should be noted that the present reported 20-day CGM dataset follow-up is among the longest published in insulin-treated patients on HD, thus providing valuable real-world evidence. Second, MF-BIA body composition was measured only once. This limitation reflects the fact that repeated BIA assessments in patients undergoing HD are strongly confounded by intradialytic and interdialytic fluid shifts. Thus, a single 30-minute post-dialysis MF-BIA is the validated and standard approach for estimating visceral adiposity in this population. However, future studies incorporating serial BIA measurements may provide complementary information. Third, although our ROC-derived VFA threshold (approximately 63 cm²) identifies patients at higher risk of glucose instability, this value should be considered exploratory. As this was an observational study, we did not evaluate whether modifying body composition or insulin titration based on VFA could improve glycemic stability. Therefore, interventional studies are warranted.

From a clinical perspective, assessing visceral adiposity may offer practical value for identifying patients on HD at higher risk of glucose instability. Individuals with low VFA may benefit from closer CGM surveillance, individualized insulin titration around HD sessions, and targeted nutritional interventions. By incorporating a simple body composition assessment into routine HD care, clinicians could have an additional tool to optimize treatment in this metabolically vulnerable subgroup.

CONCLUSION

In conclusion, we have identified visceral adiposity as a major independent determinant of GV in patients with insulin-treated T2D undergoing HD. Contrary to the traditional perception of visceral fat as purely detrimental, our findings suggest that a minimal threshold of visceral fat reserves may confer metabolic stability by buffering against acute glycemic fluctuations. Patients with low VFA exhibited greater TIR reduction and worse glycemic control between HD and off-HD days, despite having preserved muscle mass and hydration parameters. Therefore, these patients may represent a “metabolic fragile” subgroup that warrants special attention. Furthermore, our results point to incorporating body composition analysis into the routine evaluation of this vulnerable population. Future studies should aim at validating these results in larger, multicenter cohorts are warranted.