Cluster analysis of computed tomography identifies duration-associated abdominal fat deposition in type 2 diabetes mellitus

Abstract

BACKGROUND

Abdominal ectopic fat deposition is closely associated with type 2 diabetes mellitus (T2DM). However, whether there are distinct fat-distribution phenotypes across different organs, and how these phenotypes relate to changes in blood glucose, remains unclear, particularly given the potential impact of disease duration.

AIM

To investigate site specific and duration associated phenotypes of abdominal fat distribution and their associations with glycemic indicators in patients with T2DM of different disease durations, using non-contrast computed tomography (CT) and cluster analysis. The goal was to identify imaging based phenotypes that may inform future risk stratification.

METHODS

Between September 2019 and December 2024, we enrolled 130 patients with prediabetes and 398 patients with T2DM (207 with disease duration < 5 years, and 191 with disease duration ≥ 5 years), along with 150 age- and sex-matched healthy controls. All participants underwent non-contrast abdominal CT. CT attenuation values were measured for the liver, pancreas, kidneys, and psoas muscles, and the areas of visceral adipose tissue, subcutaneous adipose tissue, and renal sinus fat were quantified. K-means clustering was used to identify fat-distribution phenotypes across disease durations, and multiple linear regression was performed to examine their relationships with glycemic indices.

RESULTS

Site-specific and durationassociated heterogeneity. Liver fat was associated with blood glucose throughout the disease course. Pancreatic fat was concentrated predominantly in the tail (P < 0.05). Pancreatic and renal sinus fat were correlated with blood glucose in both the < 5-year and ≥ 5-year groups, whereas psoas muscle fat was evident only in the ≥ 5-year group (P < 0.05 for all). Fat-distribution phenotypes. The prediabetes group comprised “liver-dominant” (higher fasting plasma glucose, 2 hours oral glucose tolerance test, triglyceride, and male predominance) and “pancreas-dominant” phenotypes. The < 5-year T2DM group included “visceral/subcutaneous” (predominantly male, lower blood glucose), “liver-dominant”, and “pancreas-dominant” phenotypes; the latter two showed higher blood glucose, and the liver-dominant phenotype also showed elevated triglyceride, alanine aminotransferase, and aspartate aminotransferase. The ≥ 5-year T2DM group yielded four phenotypes: “Liver-dominant” (dyslipidemia, and elevated liver enzymes), “pancreas-dominant” (highest blood glucose), “renal-dominant” (male predominance, older age, longer disease duration, elevated serum creatinine and blood urea nitrogen, and decreased estimated glomerular filtration rate), and “psoas major-dominant” (no significant clinical differences).

CONCLUSION

Our cross-sectional data show that fat distribution differs across disease durations: pancreatic fat is more prominent in shorter durations, whereas renal and skeletal muscle fat are more prominent in longer durations. CT-based quantification combined with K-means clustering reveals distinct fat-distribution phenotypes. These exploratory findings provide a foundation for future hypothesis-driven studies to assess the clinical utility of CT-based phenotyping.

Key Words: Cluster analysis; Computed tomography; Type 2 diabetes mellitus; Adipose tissue; Abdominal fat; Disease duration; Blood glucose; Body fat distribution

Core Tip: This study identifies computed tomography based abdominal fat distribution phenotypes in type 2 diabetes mellitus that vary with disease duration. Cross sectionally observed pattern shows: Liver fat persists across all disease durations, pancreatic fat appears in shorter durations, and renal and skeletal muscle fat emerge in longer durations. These details are missed by conventional duration based classification. This imaging based approach provides a foundation for phenotype driven risk stratification.

INTRODUCTION

Diabetes mellitus (DM) is among the most common chronic diseases worldwide. According to the 11^th edition of the International Diabetes Federation Diabetes Atlas, 589 million adults aged 20-79 years were living with diabetes globally in 2024[1]. Type 2 DM (T2DM) is the predominant form of DM. Disorders of glucose and lipid metabolism occur throughout the course of T2DM, and promote abnormal fat accumulation in non-adipose tissues, which is an independent risk factor for the onset and progression of T2DM[2]. Fat distribution across abdominal organs and anatomical regions is heterogeneous, and distinct phenotypes of fat deposition are closely associated with insulin resistance (IR)[3]. Moreover, different durations of T2DM may involve interactions among fat deposits in different regions[4]. However, the heterogeneity of fat deposition and its metabolic implications remain incompletely understood.

Percutaneous biopsy is invasive and provides only limited tissue samples. Ultrasonography is susceptible to multiple sources of interference and has limited sensitivity for early or mild fatty degeneration. Magnetic resonance imaging is costly and can be affected by abdominal respiratory motion. By contrast, computed tomography (CT) is widely available, provides rapid image acquisition, and enables comprehensive assessment of whole-body fat distribution[5]. Thus, CT represents a practical tool for largescale evaluation of fat distribution in T2DM.

Cluster analysis is a data-driven approach that requires no predefined grouping variables. Unlike conventional hypothesis-driven methods, it integrates multiple imaging parameters and captures multidimensional joint distribution patterns. Univariate analyses offer no such capability[6,7]. CT-based cluster analysis is feasible for identifying phenotypes of metabolic diseases. For example, Cho et al[8] used abdominal CT combined with hierarchical clustering to predict the prevalence and mortality of metabolic syndrome, osteoporosis, and sarcopenia. However, the application of this approach in diabetes remains limited, particularly with respect to sitespecific and durationassociated phenotypes of abdominal fat distribution and their associations with blood glucose. Although previous studies have described abdominal fat-distribution phenotypes in patients with T2DM[6,9], most focused on single variables[10,11], making it difficult to account for interactions among multiple factors. Here, we used CT and cluster analysis to map abdominal fat distribution in T2DM. We examined fat distribution’s relationship with blood glucose, seeking imaging-based phenotypes for future risk stratification.

MATERIALS AND METHODS

Study population and data collection

This study was approved by the Ethics Committee of Tianjin First Central Hospital, approval No. 2026-114, and the requirement for informed consent was waived for all subjects. Patients with abnormal blood glucose diagnosed at Tianjin First Central Hospital between September 2019 and December 2024 were retrospectively included. The inclusion criteria for patients with prediabetes and type T2DM were as follows: (1) Age 20-65 years; and (2) Meeting the diagnostic criteria for prediabetes or T2DM proposed in the 2024 edition of the “Chinese Guidelines for the Prevention and Treatment of T2DM”[12].

Prediabetes, was defined as meeting either of the following criteria: (1) Impaired fasting glucose [6.1 ≤ fasting plasma glucose (FPG) < 7.0 mmol/L, and 2-hour oral glucose tolerance test (OGTT) < 7.8 mmol/L]; or (2) Impaired glucose tolerance (7.8 ≤ 2-hour OGTT < 11.1 mmol/L and FPG < 7.0 mmol/L).

T2DM was diagnosed according to the following criteria: (1) Typical symptoms of hyperglycemia or hyperglycemic crisis; (2) Fasting blood glucose ≥ 7.0 mmol/L; (3) 2-hour OGTT ≥ 11.1 mmol/L; or (4) Random blood glucose ≥ 11.1 mmol/L. The diagnosis was confirmed when the above blood glucose values were observed again on repeated measurement.

The exclusion criteria were as follows: (1) Other types of diabetes; (2) Acute or chronic liver or kidney diseases caused by other etiologies; (3) Other diseases or a history of diseases that could increase pancreatic fat content or alter bone mineral density; (4) Pregnancy or lactation; and (5) Poor cooperation or poor-quality CT images. During the same period, healthy controls matched for sex and age were enrolled as the healthy control group.

The inclusion criteria for the healthy group were as follows: (1) Age 20-65 years; (2) No diabetes or hypertension, and normal blood glucose test results within the past year; and (3) No history of chronic liver, kidney, pancreatic, or bone disease (Figure 1).

Figure 1 Flow chart for the inclusion of patients with prediabetes and type 2 diabetes mellitus. T2DM: Type 2 diabetes mellitus; FPG: Fasting plasma glucose; IGT: Impaired glucose tolerance; OGTT: Oral glucose tolerance test; CT: Computed tomography; DMa: Type 2 diabetes mellitus with disease duration < 5 years; DMb: Type 2 diabetes mellitus with disease duration ≥ 5 years.

General data, including age, sex, body mass index (BMI), and height, were collected. According to the criteria recommended by the Working Group on Obesity in China for the Chinese adult population, BMI was classified as: Normal weight (18.5 kg/m² ≤ BMI < 24.0 kg/m²), overweight (24.0 kg/m² ≤ BMI < 28.0 kg/m²), and obesity (BMI ≥ 28.0 kg/m²)[13]. Fasting venous blood samples were obtained within one week before CT examination to measure FPG, 2 hours OGTT, glycated hemoglobin (HbA1c), serum creatinine, blood urea nitrogen (BUN), estimated glomerular filtration rate (eGFR), triglyceride (TG), total cholesterol (TC), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), alanine aminotransferase (ALT), and aspartate aminotransferase (AST).

Unenhanced CT scan and image analysis

All subjects underwent abdominal unenhanced CT using a GE Revolution CT, Toshiba Aquilion ONE CT, or Siemens Somatom Definition Flash CT scanner. Before examination, subjects received breathing training, fasted for 6 hours without water intake, and removed metal objects within the scanning range. Subjects were placed in the supine position, and the scanning range extended from the dome to the lower poles of both kidneys. The CT equipment and scanning parameters are shown in Supplementary Table 1.

The CT values of abdominal organs and the fat areas of the abdominal cavity and renal sinus were measured independently on a Siemens Syngo. Via workstation by two physicians with 3 years and 5 years of experience in abdominal imaging diagnosis. Before measurement, all images underwent standardized processing on the workstation. A senior radiologist guided regions of interest (ROI) placement at standardized anatomical landmarks to reduce inter-observer variability: (1) Small ROIs were placed in stable parenchyma, avoiding organ edges and large vessels. ROIs for each organ were manually placed as follows: Three ROIs were placed in the left and right lobes of the liver, the upper, middle, and lower poles of the spleen, the head, body, and tail of the pancreas, the parenchyma of the upper and lower poles of both kidneys, the L3 vertebral body, and the bilateral psoas major muscles at the L3 vertebral body level. During delineation, organ edges and large blood vessels were avoided, and the ROI size was 100 ± 10 mm. The average value was used as the final CT value for each organ; and (2) The fat areas of the abdominal cavity and renal sinus were measured as follows: Visceral adipose tissue (VAT) and subcutaneous adipose tissue (SAT) were measured at the umbilical level[14], and renal sinus fat area was measured at the renal hilum level (Figure 2)[15].

Figure 2 Schematic diagram of the placement and delineation of regions of interests for various abdominal organs, visceral, subcutaneous, and bilateral renal sinus. A: Regions of interests (ROIs) of the left and right lobes of the liver; B-D: ROIs of the upper, middle, and lower poles of the spleen; E: ROIs of the head of the pancreas; F: ROIs of the body, and tail of the pancreas; G: ROIs of the bilateral renal parenchyma; H: ROI of the L3 lumbar vertebra; I: ROIs of the bilateral psoas major muscles; J: The blue area represents the delineation range of the visceral adipose tissue ROI; K: The purple area represents the delineation range of the subcutaneous adipose tissue ROI; L: The yellow area represents the delineation range of the bilateral renal sinus fat ROI. Yellow circle: Regions of interests placement location for the corresponding structures shown in each panel.

Statistical analysis

Statistical analysis was performed using SPSS 28.0 (IBM, United States). Normality was assessed. Normally distributed continuous data are presented as mean ± SD; nonnormally distributed data as median (interquartile range). χ² test and Welch’s analysis of variance (ANOVA) were used to compare general clinical data among the four groups: Healthy control, prediabetes, T2DM with disease duration < 5 years (DMa), and T2DM with disease duration ≥ 5 years (DMb), as well as between adipose distribution phenotypes. The intraclass correlation coefficient assessed interobserver consistency, with values ≥ 0.75 indicating good agreement, 0.4-0.75 moderate, and < 0.4 poor. Within each group, one way ANOVA compared CT values across the pancreatic head, body, and tail; paired t tests compared bilateral kidney and psoas major CT values, as well as bilateral renal sinus fat area. For multiple group comparisons, one way ANOVA or the Kruskal-Wallis test was used, followed by post hoc Least Significant Difference (equal variances) or Games-Howell (unequal variances). K-means cluster analysis of CT values of abdominal organs and regions in each group was performed using R Studio 4.4.2 (New Zealand). The number of clusters (K) was primarily determined using the elbow method (within-cluster sum of squares) and the silhouette coefficient to ensure a robust clustering structure. Multiple linear regression identified factors associated with glycemic parameters in the prediabetes, DMa, and DMb groups across different adipose distribution phenotypes. A two sided was considered statistically significant.

RESULTS

General characteristics

A total of 528 patients with abnormal blood glucose were ultimately included in this study. Among them, 130 patients were in the prediabetes duration, including 77 males and 53 females, with a mean age of 55.73 ± 10.23 years (range, 20-65 years). A total of 398 patients had T2DM and were divided according to disease duration: The < 5-year group (DMa group) included 207 patients, including 106 males and 101 females, with a mean age of 56.13 ± 11.09 years (range, 20-65 years); the ≥ 5-year group (DMb group) included 191 patients, including 100 males and 91 females, with a mean age of 57.65 ± 9.36 years (range, 20-65 years). In addition, 150 healthy controls matched for sex and age were included as the control group, including 75 males and 75 females, with a mean age of 55.20 ± 10.65 years (range, 20-65 years).

There were statistically significant differences in body weight, BMI, FPG, 2 hours OGTT, and HbA1c among the four groups (all P < 0.05). Significant differences were also observed in BUN, eGFR, TG, TC, HDL-C, ALT, and AST among the four groups (P < 0.05). Pairwise comparisons showed that, compared with the healthy group, BMI, FPG, 2 hours OGTT, and HbA1c were increased in the prediabetes, DMa, and DMb groups; TG, TC, ALT, and AST were increased in the DMa and DMb groups; and BUN was increased and eGFR was decreased in the DMb group. Compared with the prediabetes group, FPG, 2 hours OGTT, HbA1c, TG, TC, ALT, and AST were increased in the DMa and DMb groups, and eGFR was decreased in the DMb group. Compared with the DMa group, eGFR and HDL-C were decreased in the DMb group. All the. above differences were statistically significant (all P < 0.05, Table 1).

Table 1 Comparison of demographic and biochemical indicators between groups, mean ± SD.

Item	Healthy group (n = 150)	Prediabetes group (n = 130)	DMa group (n = 207)	DMb group (n = 191)	P value
Gender (male/female)	75/75	77/53	106/101	100/91	0.127
Age (years)	55.20 ± 10.65	55.73 ± 10.23	56.13 ± 11.09	57.65 ± 9.36	0.122
Height (cm)	169.70 ± 9.75	167.8 ± 8.37	168.72 ± 8.54	169.13 ± 8.51	0.091
Weight (kg)	70.81 ± 10.97	70.66 ± 12.69	74.51 ± 16.35a	75.42 ± 52.08a	0.006f
BMI (kg/m2)	24.88 ± 3.37	25.14 ± 3.40a	25.95 ± 4.40b	24.94 ± 3.90a	0.000f
FPG (mmol/L)	5.48 ± 0.98	5.96 ± 1.15a	8.86 ± 3.97b,d	8.43 ± 3.89b,d	0.001f
2 hours OGTT (mmol/L)	7.20 ± 1.19	9.72 ± 2.91b	16.84 ± 5.44b,d	15.17 ± 5.00b,d	0.000f
HbA1c (%)	5.41 ± 0.76	6.36 ± 1.47b	8.64 ± 2.51b,d	8.32 ± 2.03b,d	0.000f
Scr (μmol/L)	80.47 ± 1.47	77.58 ± 1.20	83.37 ± 1.25	90.12 ± 1.28	0.072
BUN (mmol/L)	5.09 ± 2.86	5.76 ± 3.02	5.18 ± 3.31	6.87 ± 4.77b	0.034f
eGFR (mL/minute/1.73 m2)	94.77 ± 3.18	95.51 ± 17.34	98.51 ± 21.99	82.39 ± 27.22a,c,e	0.011f
TG (mmol/L)	1.47 ± 0.17	1.70 ± 0.45	2.43 ± 2.64a,c	2.39 ± 2.27a,c	0.035f
TC (mmol/L)	3.94 ± 0.87	4.17 ± 1.47	4.72 ± 1.44a,c	4.78 ± 1.48a,c	0.036f
HDL-C (mmol/L)	1.29 ± 0.64	1.31 ± 0.81	1.72 ± 0.44	1.11 ± 0.44e	0.041f
LDL-C (mmol/L)	2.91 ± 0.16	2.77 ± 0.91	2.87 ± 1.07	3.54 ± 1.05	0.052
ALT (U/L)	25.79 ± 6.85	30.57 ± 2.54	48.85 ± 7.11b,c	68.06 ± 15.51b,c	0.023f
AST (U/L)	29.72 ± 8.47	30.84 ± 2.01	48.78 ± 4.62b,c	68.27 ± 16.03b,c	0.010f

^aP < 0.05 vs the control group.

^bP < 0.01 vs the control group.

^cP < 0.05 vs the prediabetes group.

^dP < 0.01 vs the prediabetes group.

^eP < 0.05 vs type 2 diabetes mellitus with disease duration < 5 years group.

^fP < 0.05.

DMa: Type 2 diabetes mellitus with disease duration < 5 years; DMb: Type 2 diabetes mellitus with disease duration ≥ 5 years; BMI: Body mass index; FPG: Fasting plasma glucose; OGTT: Oral glucose tolerance test; HbA1c: Glycated hemoglobin; Scr: Serum creatinine; BUN: Blood urea nitrogen; eGFR: Estimated glomerular filtration rate; TG: Triglyceride; TC: Total cholesterol; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase.

Inter-observer agreement analysis

The reliability and reproducibility of the CT values of the liver, spleen, pancreatic head, pancreatic body, pancreatic tail, bilateral kidneys, L3 vertebral body, bilateral psoas major muscles, and the area values of VAT, SAT, and bilateral renal sinus fat measured by the two physicians were relatively high, with intraclass correlation coefficients (95% confidence intervals) ranging from 0.752 to 0.907.

Comparison of CT values in the head, body, and tail of the pancreas

There were statistically significant differences in CT values across different pancreatic regions between the DMa and DMb groups (all P < 0.05), mainly manifested as decreased CT values in the pancreatic tail (Supplementary Table 2).

Comparison of CT values of different abdominal organs and regional fat areas between groups

There were no statistically significant differences in the CT values of the bilateral renal parenchyma or psoas major muscles, or in bilateral renal sinus fat areas, among the four groups (all P > 0.05). Therefore, the average values of bilateral measurements were used as the final results for subsequent statistical analysis (Supplementary Table 3).

There were statistically significant differences in the CT values of the liver, pancreas, renal parenchyma, L3 vertebral body, and psoas major muscles among the healthy, prediabetes, DMa, and DMb groups (P < 0.05 to P < 0.001), whereas no statistically significant difference was observed in the CT value of the spleen (P > 0.05). Pairwise comparisons showed that the CT values of the liver and all pancreatic regions (head, body, and tail) in the prediabetes, DMa, and DMb groups were lower than those in the healthy group (P < 0.01). Among these regions, the CT values of the liver and pancreatic tail in the DMa and DMb groups were further decreased compared with the prediabetes group (P < 0.05), and the CT value of the pancreatic head was lower only in the DMb group than in the prediabetes group (P < 0.01). The CT value of the renal parenchyma was higher in the DMb group than in the healthy group (P < 0.05). The CT value of the L3 vertebral body in the DMa and DMb groups was lower than that in the healthy group (P < 0.001), and the value in the DMb group was lower than that in the prediabetes group (P < 0.001). The CT value of the psoas major muscle in the DMa and DMb groups was lower than that in both the healthy group (P < 0.001) and the prediabetes group (P < 0.05), and the value in the DMb group was lower than that in the DMa group (P < 0.001) (Table 2).

Table 2 Comparison of computed tomography values of different abdominal organs and fat area, mean ± SD.

CT value or fat area values	Healthy group	Prediabetes group	DMa group	DMb group	P value
Liver (HU)	59.57 ± 5.30	54.45 ± 7.63	51.31 ± 8.41	51.22 ± 7.53	0.000a
Spleen (HU)	47.24 ± 4.64	46.15 ± 5.10	46.96 ± 4.95	46.65 ± 4.81	0.263a
Head of pancreas (HU)	42.33 ± 7.31	37.86 ± 9.06	36.64 ± 10.18	35.09 ± 10.30	0.000a
Body of pancreas (HU)	42.65 ± 7.44	38.79 ± 9.03	37.20 ± 9.51	35.98 ± 7.94	0.000a
Tail of pancreas (HU)	42.57 ± 5.87	38.58 ± 8.76	34.71 ± 9.93	33.53 ± 9.74	0.000a
Renal parenchyma (HU)	29.72 ± 4.27	30.38 (25.28, 35.48)	30.89 ± 3.64	31.18 (25.50, 36.86)	0.015a
L3 vertebral body (HU)	140.68 (69.20, 212.16)	120.38 (55.53, 185.23)	123.43 ± 44.14	98.29 (39.83, 156.75)	0.000a
Psoas major (HU)	49.75 ± 5.71	47.97 ± 7.66	45.60 ± 6.99	43.65 (32.44, 54.86)	0.000a
VAT (mm2)	94.80 (66.95, 122.65)	114.78 ± 43.14	119.67 (54.13, 185.21)	111.73 (45.37, 178.09)	0.000a
SAT (mm2)	90.51 (61.66, 119.36)	144.35 ± 73.00	127.70 (18.30, 237.10)	135.40 (52.52, 218.28)	0.000a
Renal sinus fat area (mm2)	1.47 ± 0.56	1.64 ± 0.63	1.79 ± 1.10	2.19 ± 1.16	0.000a

^aP < 0.05.

CT: Computed tomography; DMa: Type 2 diabetes mellitus with disease duration < 5 years; DMb: Type 2 diabetes mellitus with disease duration ≥ 5 years; VAT: Visceral adipose tissue; SAT: Subcutaneous adipose tissue.

There were statistically significant differences in the areas of VAT, SAT, and renal sinus fat among the four groups (all P < 0.001). Pairwise comparisons showed that VAT and SAT areas in the prediabetes, DMa, and DMb groups were significantly higher than those in the healthy group, and renal sinus fat area in the DMa and DMb groups was significantly higher than that in the healthy group. Renal sinus fat area in the DMb group was significantly higher than that in the prediabetes group (all P < 0.05, Table 2).

Fat-distribution phenotypes across different disease durations after K-means clustering

K-means clustering analysis was performed using the CT values of different abdominal organs and fat areas in different regions in the prediabetes, DMa, and DMb groups. The within-cluster sum of squares and silhouette coefficients were compared across different k values. The mean ± SD of each phenotype was used to describe the phenotypic characteristics.

The results showed that the prediabetes group comprised two phenotypes: Type I was mainly characterized by pancreatic fat deposition, and type II was mainly characterized by hepatic fat deposition. The DMa group comprised three phenotypes: Type I was mainly characterized by visceral and subcutaneous fat deposition, type II by hepatic fat deposition, and type III by pancreatic fat deposition. The DMb group comprised four phenotypes: Type I was mainly characterized by hepatic fat deposition, type II by pancreatic fat deposition, type III by psoas major muscle fat deposition, and type IV by renal fat deposition and increased renal sinus fat area (Figure 3).

Figure 3 K-means clustering results of the pre-period group, type 2 diabetes mellitus with disease duration < 5 years group, and type 2 diabetes mellitus with disease duration ≥ 5 years group using R. A-C: When k = 2, the decline rate of within-cluster sum of squares slows down and the average silhouette coefficient is the highest. Set as 2 the final number of clusters, and there is no obvious mixing between the two clusters; D-F: When k = 3, the decline rate of WSS significantly slows down and the average silhouette coefficient is the highest. Set 3 as the final number of clusters, and there is no obvious mixing between the three clusters. G-I: When k = 4, the decline rate of WSS significantly slows down and the average silhouette coefficient is the highest. Set 4 as the final number of clusters, and there is no obvious mixing between the four clusters.

Analysis of clinical and biochemical characteristics of different fat-distribution phenotypes

Clinical and biochemical indicators were further compared among different fat-distribution phenotypes within the prediabetes, DMa, and DMb groups. In the prediabetes group, FPG, 2 hours OGTT, and TG were significantly increased in type II, and the proportion of males was relatively high (Supplementary Table 4). In the DMa group, the proportion of males was relatively high in type I. TG, ALT, and AST were significantly increased in type II, and FPG, 2 hours OGTT, and HbA1c were significantly increased in types II and III (Supplementary Table 5). In the DMb group, TG, LDL-C, ALT, and AST were significantly increased in type I, whereas HDL-C was significantly decreased; FPG, 2 hours OGTT, and HbA1c were significantly increased in type II; and the proportion of males was relatively high in type IV, in which age, disease duration, serum creatinine, and BUN were significantly increased and eGFR was significantly decreased (Supplementary Table 6). All the above differences were statistically significant (all P < 0.05).

Multiple linear regression analysis was performed with FPG, 2 hours OGTT, and HbA1c as dependent variables

Multiple linear regression analyses were performed with FPG, 2 hours OGTT, and HbA1c as dependent variables. Durbin-Watson values ranged from 0.5 to 3.6 and variance inflation factors were all < 10, confirming that the assumptions for multiple linear regression were met. The results showed that, in the prediabetes group, decreased liver CT value had a significant negative effect on FPG and 2 hours OGTT (both P < 0.05). In the DMa group, decreased liver and pancreatic CT values had significant negative effects on FPG, 2 hours OGTT, and HbA1c (all P < 0.05). In the DMb group, decreased liver, pancreatic, and renal parenchymal CT values had significant negative effects on FPG, 2 hours OGTT, and HbA1c; decreased psoas major muscle CT value had significant negative effects on FPG and 2 hours OGTT; and increased renal sinus fat area had significant positive effects on FPG, 2 hours OGTT, and HbA1c (all P < 0.05, Table 3).

Table 3 Comparison of fasting plasma glucose, 2-hour oral glucose tolerance test, and glycated hemoglobin among different groups and multiple linear regression analysis.

Item	FPG	2 hours OGTT	HbA1c
	β (SE, P value)	β (SE, P value)	β (SE, P value)
Prediabetes group
liver (HU)	-0.179 (0.158, 0.049)a	-0.204 (0.288, 0.045)a	-0.404 (0.011, 0.118)
Pancreas (HU)	-0.077 (0.023, 0.671)	-0.461 (0.047, 0.121)	-0.204 (0.019, 0.056)
DMa group
liver (HU)	-0.445 (0.611, 0.017)a	-0.401 (0.117, 0.040)a	-0.340 (0.419, 0.047)a
Pancreas (HU)	-0.420 (0.088, 0.032)a	-0.268 (0.216, 0.034)a	-0.205 (0.069, 0.038)a
VAT (mm2)	0.079 (0.006, 0.137)	0.016 (0.012, 0.117)	0.044 (0.004, 0.136)
SAT (mm2)	0.161 (0.004, 0.067)	0.110 (0.008, 0.074)	0.143 (0.002, 0.110)
DMb group
liver (HU)	-0.025 (0.666, 0.048)a	-0.439 (0.008, 0.044)a	-0.180 (0.004, 0.014)a
Pancreas (HU)	-0.182 (0.109, 0.037)a	-0.317 (0.091, 0.014)a	-0.230 (0.050, 0.022)a
Psoas major (HU)	-0.153 (0.053, 0.026)a	-0.192 (0.077, 0.016)a	-0.017 (0.035, 0.890)
Renal parenchyma (HU)	-0.260 (0.112, 0.009)a	-0.588 (0.194, 0.029)a	-0.022 (0.077, 0.048)a
Renal sinus fat area (mm2)	0.273 (0.419, 0.028)a	0.381 (0.084, 0.038)a	0.285 (0.280, 0.007)a

^aP < 0.05.

Β: Standardized coefficient; SE: Standard error; DMa: Type 2 diabetes mellitus with disease duration < 5 years; DMb: Type 2 diabetes mellitus with disease duration ≥ 5 years; BMI: Body mass index; FPG: Fasting plasma glucose; OGTT: Oral glucose tolerance test; HbA1c: Glycated hemoglobin; Scr: Serum creatinine; BUN: Blood urea nitrogen; eGFR: Estimated glomerular filtration rate; TG: Triglyceride; TC: Total cholesterol; HDL-C: High-density lipoprotein cholesterol; LDL-C: Low-density lipoprotein cholesterol; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase.

DISCUSSION

The global incidence of DM is high, and China accounts for a substantial proportion of the global disease burden[16]. Ectopic abdominal fat deposition is closely related to the onset of T2DM[17]. Most prior studies used a limited set of variables. How fat-deposition phenotypes change with disease duration remains poorly characterized. Here, we used CT quantification to map abdominal fat distribution in patients with T2DM of different disease durations. Their clinical and biochemical profiles were then compared. The results revealed significant site specific and duration associated heterogeneity.

The liver and pancreas are core target organs of insulin metabolism[18]. Intergroup comparisons showed that, except between the DMa and DMb groups, the other group comparisons showed a trend toward decreasing CT values of the liver and pancreas (i.e., increasing fat deposition) with longer disease duration. This suggests that, in the early duration of T2DM, increased blood glucose is accompanied by significant fat deposition in the liver and pancreas. Liver fat deposition can impair gluconeogenesis and insulin responsiveness, thereby increasing the risk of T2DM[19]. Pancreatic fat deposition affects islet cell function and triggers chronic low-grade inflammation, thereby increasing tissue IR[20,21]. However, pancreatic fat deposition is not uniform. In the DMa and DMb groups, the increase was mainly observed in the pancreatic tail, as indicated by CT values that were significantly lower than those of the pancreatic head and body. This likely reflects the tail's distinct anatomy and microenvironment[22]. If so, fat here may be more sensitive to metabolic changes than elsewhere in the gland.

Diabetic kidney injury is one of the most common microvascular complications of T2DM[23]. Renal parenchymal CT attenuation differed between the prediabetes and DMb groups; eGFR was lower in the DMb group. Renal sinus fat has been implicated in early renal injury[24]. In our cohort, sinus fat was already elevated in the DMa group, whereas eGFR declined only in the DMb group. This suggests that chronic hyperglycemia may promote both renal lipid deposition and functional decline[25,26], with sinus fat accumulation preceding visible parenchymal changes on CT. Whether parenchymal and sinus fat each contribute independently to glycemic dysregulation -and how they compare in metabolic impact - remains poorly understood. The elevated renal parenchymal CT attenuation seen in the DMb group warrants separate investigation, as its direction is not consistent with fat accumulation.

Psoas major fat increased after the prediabetes period. L3 vertebral fat, however, differed only between healthy controls and patient groups. These findings indicate that long-term hyperglycemia can affect bone structure and muscle mass[27,28]. As a control organ, spleen fat content remained stable among the groups, providing methodological support for assessing liver fat content using the liver/spleen CT ratio[29]. Abdominal fat accumulation, especially VAT accumulation, is an important background factor in the onset of T2DM[17]. VAT and SAT were elevated across the prediabetes, DMa, and DMb groups. This suggests that abdominal fat area tracks metabolic risk more closely than BMI alone[14].

Conventional intergroup comparisons miss much of the individual heterogeneity. We therefore turned to K-means clustering to characterize fatdistribution phenotypes across disease stages. Some findings echoed prior work[6].

The prediabetes group was clustered into the “liver type” and the “pancreas type” (Figure 3A-C). The former had significantly higher FPG, 2 hours OGTT, male proportion, and TG than the latter. This pattern suggests hepatic steatosis in this subgroup likely reflects IR and enhanced gluconeogenesis[30]. Whether lifestyle intervention can alter their disease course remains unclear[31].

The DMa group was divided into the “visceral/subcutaneous fat type”, the “liver type”, and the “pancreas type” (Figure 3D-F). The “visceral/subcutaneous fat type” was characterized by significant increases in VAT and SAT and was more common in men, which may be related to the influence of androgens on fat distribution[32]. However, its blood glucose indicators were relatively low, suggesting that isolated abdominal fat accumulation has a limited influence on blood glucose. Both the “liver type” and the “pancreas type” had increased fat in their respective organs and higher blood glucose. This suggests ectopic fat in these sites may synergistically impair glycemic control[33]. In the “liver type”, TG, ALT, and AST levels were significantly increased, indicating that liver function impairment should be carefully monitored[34].

The DMb group was clustered into four phenotypes (Figure 3G-I). Patients with the “kidney type” were older, had a longer disease duration, were predominantly male, and showed increased blood creatinine and urea nitrogen levels and decreased eGFR. This phenotype represents a high-risk phenotype for diabetic nephropathy[35], and renal function should be closely monitored. The “pancreas type” had the highest blood glucose indicators, which may be related to cell dysfunction caused by pancreatic fat accumulation[36]. The “liver type” showed dyslipidemia (elevated TG and LDL-C, decreased HDL-C) and elevated liver enzymes (ALT and AST). This points to a close link between non-alcoholic fatty liver disease and T2DM[34]. Differences in the indicators of the “psoas major muscle type” were not significant, and its clinical significance should be clarified in future studies.

TG, TC, and HDL-C levels differed significantly across disease duration groups. TG and HDL-C are independent risk factors for T2DM with metabolic dysfunction-associated steatotic liver disease[37]. Their combination may aid diagnosis. CT-based fat quantification (liver, pancreas, psoas major, and other sites) combined with serological markers such as TG and HDL-C may help build a multidimensional framework. Such a framework may help characterize duration associated phenotypes. However, the clinical utility of this approach requires prospective validation.

This study analyzed the associations between abdominal fat-distribution phenotypes and glycemic indicators (FPG, 2 hours OGTT, and HbA1c) in patients with T2DM with different disease durations (Table 3). Liver fat content was significantly associated with blood glucose across all disease durations. In the DMa group, fat in different pancreatic regions showed significant associations with blood glucose. In the DMb group, psoas major muscle fat, renal parenchymal fat, and renal sinus fat area were also significantly associated with glycemic indicators. This cross-sectional pattern is of hepatic origin and associated with disease duration. It suggests a working hypothesis: The liver may be a primary site of fat deposition; pancreatic fat may be linked to βcell decompensation; and renal and skeletal muscle fat may also reflect multiorgan metabolic changes. However, these hypothesized sitespecific and durationassociated patterns require validation in longitudinal studies, and the clinical implications remain to be determined. Our exploratory findings provide imaging-based evidence for phenotypic characterization across disease durations.

This study has several limitations. Firstly, the sample was small and single-center, without external validation. Cluster phenotyping is sensitive to cohort characteristics and analytical choices, and requires multicenter validation. Secondly, we did not split prediabetes into impaired fasting glucose and impaired glucose tolerance subgroups. Thirdly, this cross-sectional design precludes causal inference. These clusters are snapshots, not fixed subtypes; they shift with disease duration, lifestyle, and treatment. Fourthly, the retrospective design precluded adjustment for treatment and glycemic control; prospective studies should confirm phenotypic stability and clinical relevance. Fifthly, we did not adjust for multiple comparisons or assess cluster stability through bootstrap resampling. These findings require cautious interpretation. Future studies should apply appropriate multiplicity controls and stability assessments, including internal validation and replication in independent cohorts. Finally, ROI-based CT may miss intra-organ fat heterogeneity. Our phenotyping captures mechanistic associations only; it is not a screening tool, and CT radiation and cost preclude routine use in asymptomatic populations.

CONCLUSION

In T2DM, abdominal fat deposition differs by organ and tracks with disease duration. K-means clustering grouped these patterns into phenotypes with characteristic clinical profiles. These findings set the stage for hypothesis-driven studies to test the clinical utility of CT-based phenotyping.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge Dr. Aiyun Sun from GE HealthCare for his critical review of the manuscript and valuable suggestions regarding scientific presentation and overall manuscript quality.