Association of serum bile acid profiles with the risk of gastrointestinal autonomic neuropathy in patients with type 2 diabetes mellitus

Abstract

BACKGROUND

Diabetic gastrointestinal autonomic neuropathy (DGAN) is a common yet underrecognized manifestation of diabetic neuropathy. However, the clinical correlations and associated alterations of serum bile acid (BA) metabolism with DGAN remain unclear.

AIM

To identify potential metabolite biomarkers capable of identifying DGAN among patients with type 2 diabetes mellitus (T2DM).

METHODS

This cross-sectional study included 26 patients with clinically defined DGAN and 69 patients with uncomplicated T2DM. Fifteen individual BAs were quantified in fasting serum using liquid chromatography-tandem mass spectrometry. Gastrointestinal symptoms were scored using the Gastrointestinal Symptom Rating Scale. Spearman’s correlation, multivariable logistic regression, receiver operating characteristic, and decision curve analyses were used to explore the associations and build a predictive nomogram.

RESULTS

Orthogonal partial least squares discriminant analysis of serum BAs revealed clear separation between the T2DM and DGAN groups, identifying taurolithocholic acid (TLCA) as the key discriminator (variable importance in projection > 1, fold change < 0.5). Univariate logistic regression identified age, body mass index, hemoglobin, fasting/2-h C-peptide, albumin, and TLCA levels as protective factors and the urinary albumin-to-creatinine ratio as a risk factor. After multivariate adjustment age, fasting C-peptide levels, and TLCA levels remained independently associated with DGAN. Receiver operating characteristic analysis yielded areas under the curve of 0.651, 0.760, and 0.678 for age, fasting C-peptide, and TLCA, respectively, and the three variables collectively had an area under the curve of 0.970 (95% confidence interval: 0.937-1.000). Decision curve analysis confirmed the clinical net benefit across threshold probabilities of 9%-68%. The derived nomogram displayed excellent calibration and net clinical benefits for the model.

CONCLUSION

These findings highlighted the association between altered BA profiles and DGAN in patients with T2DM. Combining BA profiling with conventional clinical data could facilitate the early identification of DGAN and offer new insights into early screening and BA-targeted interventions. While these findings offer valuable insights, they should nevertheless be viewed as hypothesis-generating and require further validation in larger, multicenter cohorts.

Key Words: Type 2 diabetes mellitus; Diabetic gastrointestinal autonomic neuropathy; Serum bile acids profiles; Gastrointestinal Symptom Rating Scale

Core Tip: This study revealed an association between altered bile acid (BA) profiles and diabetic gastrointestinal autonomic neuropathy. Combining BA profiling with conventional clinical data could facilitate the early identification of DGAN and offer new insights into early screening and BA-targeted interventions.

INTRODUCTION

Diabetic gastrointestinal autonomic neuropathy (DGAN), traditionally represented by diabetic gastroparesis, is one of the most common diabetic complications[1]. Its prevalence increases with disease progression, ultimately affecting 30%-50% of patients[2]. DGAN extends beyond gastroparesis and involves the entire gastrointestinal tract, including the stomach[1], intestines[3,4], gallbladder[5], and esophagus[5]. This disorder is characterized by sensorimotor dysfunction mediated by autonomic neuropathy[5].

Patients with DGAN experience significant distress owing to recurrent gastrointestinal symptoms, including nausea[1], vomiting[1], diarrhea[3], and constipation and diarrhea in alternation[4], particularly in the early stages[6]. As symptoms manifest and progress, existing treatments frequently prove inadequate; consequently, the early identification and diagnosis of DGAN are important.

Gastric emptying scintigraphy, the diagnostic gold standard[7], has limited availability because of the requirement for specialized nuclear medicine equipment. This technique is also associated with radiation exposure and high costs, and it is contraindicated in both children and pregnant females. Moreover, this technique evaluates gastric emptying but not pan-enteric function, and it often reveals a poor correlation between symptom severity and gastric emptying abnormalities[8].

Breath tests and acoustic imaging exhibit limited accuracy because of interference from pulmonary/hepatic dysfunction and medications (e.g., antacids, antibiotics), thus displaying 30%-40% discordance rates compared with scintigraphy[9,10]. Wireless capsule endoscopy and pressure mapping are cost-prohibitive[11], and they lack standardized diagnostic thresholds. International guidelines do not provide established normative values for diabetic populations[7]. Additionally, single-use consumable costs preclude large-scale screening. Current motility assessments quantify pathophysiological outcomes rather than etiological mechanisms. Most techniques detect the downstream effects of muscular/enteric nervous system activity[12] without differentiating the contributions of vagal, sympathetic, or intrinsic enteric neural pathways. Consequently, false-negative results are frequently obtained in early-stage disease (when potentially reversible).

Non-invasive, low-cost diagnostic tools suitable for population-level applications that can detect abnormalities prior to symptom onset are urgently needed. Thus, combining peripheral blood metabolic markers with conventional risk factors could be useful for developing early risk stratification models. Longitudinal studies demonstrated positive correlations among delayed gastric emptying, persistent elevation of glycated hemoglobin (HbA1c), and frequent acute hyperglycemic episodes (e.g., ketoacidosis)[13]. Other risk factors, including peripheral neuropathy, non-proliferative diabetic retinopathy, and microalbuminuria, are positively correlated with DGAN[14]. Female sex[15], perimenopause[15], weight gain[16], and dyslipidemia[17] are recognized as potential risk modifiers.

In diabetes chronic intestinal inflammation, mucosal barrier disruption, and gastrointestinal dysbiosis collectively disrupt bile acid (BA) metabolism and transport[18]. Hepatocyte-synthesized BAs enter the small intestine for digestive functions, undergo ileal reabsorption, and return to the liver via enterohepatic circulation. Physiologically, < 5% of circulating BAs reach the colon. However, excessive ileal BA delivery or reabsorption failure increases colonic BA influx, thereby altering secretion and motility and causing BA malabsorption[19] and BA diarrhea[20]. BA dysregulation induces enteric neuropathy and motility disorders[21,22]. However, the relationship between BA alterations and gastrointestinal motility, specifically in DGAN, remains unclear.

In this hospital-based cross-sectional study, we aimed to characterize the serum BA profile in patients with type 2 diabetes mellitus (T2DM) with or without DGAN and explore its correlation with gastrointestinal motility parameters and clinical indicators, establish an early risk stratification model for DGAN based on altered BA metabolism in patients with T2DM, and identify specific BA species that are highly associated with DGAN risk and elucidate their potential roles in the pathogenesis of gastrointestinal dysmotility.

MATERIALS AND METHODS

Patients and ethics approval

In total 26 patients diagnosed with DGAN and 69 patients with uncomplicated T2DM were selected from the Endocrinology Department of the Second Affiliated Hospital of Nanjing Medical University between May 2023 and March 2024. Additionally, 75 healthy controls (HCs) who underwent health examinations at the Physical Examination Center of the Second Affiliated Hospital of Nanjing Medical University were included in the study. The flow diagram of participant inclusion is illustrated is Figure 1. This study was reviewed and approved by the Ethics Committee of the Second Affiliated Hospital of Nanjing Medical University (2024-KY-232-01).

Figure 1 Flow diagram of participant inclusion in the study. T2DM: Type 2 diabetes mellitus; HC: Healthy control; DGAN: Diabetic gastric autonomic neuropathy; HbA1c: Hemoglobin A1c.

The inclusion criteria were as follows: Age of 18-75 years; meeting the 1999 World Health Organization diagnostic criteria for T2DM; HbA1c < 11.0%; and available complete clinical parameters and metagenomic data. The exclusion criteria were as follows: History of antibiotic use for > 3 consecutive days within 6 weeks prior to recruitment; probiotic/prebiotic/synbiotic supplementation within 12 weeks; presence of clinically significant psychiatric/neurological disorders or epilepsy; comorbid gastrointestinal disorders (e.g., irritable bowel syndrome, ulcerative colitis, Crohn’s disease, gastrointestinal malignancies, thyroid dysfunction, severe malnutrition); previous gastrointestinal surgery (including gastrectomy, fundoplication, and colostomy); current use of metformin, Glucagon-like peptide-1 (GLP-1) receptor agonists, or other glucose-lowering agents with known gastrointestinal side effects; severe diabetic complications, cardiocerebrovascular diseases, or hepatic/renal/hematopoietic system disorders; alcohol abuse, defined as > 5 drinking occasions/week with consumption exceeding 100 g of distilled spirits, 250 g of rice wine, or five bottles ( ≥ 2500 mL) of beer; pregnancy or lactation; physical disability, cognitive impairment, or communication barriers affecting study compliance; and inability to commit to the study schedule.

The diagnostic criteria for DGAN were as follows: Meeting the 1999 World Health Organization diagnostic criteria for T2DM; compliance with either the 2022 American College of Gastroenterology Clinical Guideline: Gastroparesis or the 2021 Chinese Medical Association Expert Consensus on Diagnosis and Treatment of Diabetic Neuropathy; and the presence of ≥ 1 gastrointestinal symptom (dysphagia, regurgitation, nausea/vomiting, early satiety, bloating, diarrhea, constipation, or fecal incontinence), > 60% gastric retention at 2 h or > 10% gastric retention at 4 h post-ingestion of standardized solid-phase test meal, and exclusion of medication-induced or organic causes of gastrointestinal symptoms. Although patients were diagnosed based on gastroparesis criteria, they exhibited a spectrum of gastrointestinal symptoms, reflecting the pan-enteric nature of DGAN.

Physical examination

Height, weight, supine right upper arm blood pressure, and body mass index (BMI, kg/m²) were recorded for all patient groups. Additionally, the Gastrointestinal Symptom Rating Scale (GSRS) questionnaire[23] was administered to patients with DGAN.

Laboratory investigations

Blood samples were collected from the patients upon admission after an overnight fast (≥ 8 h). Biochemical parameters, including fasting blood glucose, fasting C-peptide (FC-P), 2-h postprandial C-peptide (2hC-P), and HbA1c levels, were measured. The duration of diabetes was defined as the interval between the initial diagnosis of diabetes and current hospital admission.

BA profiling analysis

Upon admission fasting conditions were maintained, and 4 mL of venous blood were collected using vacuum blood collection tubes containing coagulation enhancers. Within 2 h the blood samples were centrifuged at 16000 × g for 10 min to separate the serum, which was then stored at -80 °C until further analysis. Thawed serum samples were analyzed by liquid chromatography-tandem mass spectrometry using a Jasper HPLC-Triple Quad 4500MD system (AB SCIEX, Framingham, MA, United States). The analysis included the measurement of six primary Bas [cholic acid, chenodeoxycholic acid, glycocholic acid (GCA), glycochenodeoxycholic acid, taurocholic acid (TCA), and taurochenodeoxycholic acid (TCDCA)] and nine serum BAs [deoxycholic acid (DCA), lithocholic acid, ursodeoxycholic acid (UDCA), glycodeoxycholic acid (GDCA), glycolithocholic acid (GLCA), glycoursodeoxycholic acid (GUDCA), taurodeoxycholic acid (TDCA), taurolithocholic acid (TLCA), and tauroursodeoxycholic acid (TUDCA)].

Statistical analysis

Data are presented as the median (interquartile range). Statistical analysis among the three groups was performed using the unpaired Wilcoxon rank sum test. The ability of different BAs to discriminate between patients with uncomplicated T2DM and those with DGAN was assessed using fold change (FC) with FC ≥ |2|[24] used as the biomarker screening criterion. Orthogonal partial least squares discriminant analysis (OPLS-DA) was performed to determine the classification differences between patients with uncomplicated T2DM and those with DGAN. The discriminatory ability of different BAs was ranked using variable importance in projection (VIP) with VIP ≥ 1[25] used as the biomarker screening criterion. Spearman’s correlation analysis was performed to explore the correlations among clinical indicators, GSRS scores, and BAs. Stepwise simple and multiple regression analyses were performed to optimize diagnostic indicators. Receiver operating characteristic curve analysis was used to evaluate the performance of the diagnostic model by calculating the area under the curve (AUC). Statistical significance was denoted by P < 0.05 (two-tailed). All statistical analyses were conducted using SPSS version 25.0 (IBM, Armonk, NY, United States), GraphPad Prism 9.5 software (GraphPad, Boston, MA, United States), and SIMCA 14.1 (Sartorius, Göttingen, Germany).

RESULTS

Clinical characteristics of the participants

Participants with uncomplicated T2DM and those with DGAN exhibited higher systolic blood pressure, diastolic blood pressure, and fasting blood glucose levels than the HCs (Table 1 and Figure 1) in addition to lower high-density lipoprotein cholesterol levels. Furthermore, hemoglobin and aspartate aminotransferase levels were lower in the DGAN group than in the HC and T2DM groups. Compared with the findings in the T2DM group, the DGAN group had a younger age, lower BMI, HbA1c, FC-P, 2hC-P, and triglyceride (TG) levels, and a higher urinary albumin-to-creatinine ratio (UACR). However, the duration of diabetes did not differ significantly between the DGAN and T2DM groups. Regarding patients’ histories, individuals with DGAN often had associated autonomic and peripheral neuropathies. Many patients presented with ketoacidosis at the time of T2DM diagnosis. Insulin was commonly used as a primary hypoglycemic therapy. However, insulin did not have a clear effect on gastrointestinal motility.

Table 1 Clinical characteristics of the healthy control, type 2 diabetes mellitus, and diabetic gastric autonomic neuropathy groups, n (%).

Variables	HC (n = 75)	T2DM (n = 69)	DGAN (n = 26)	P value	P1 value	P2 value	P3 value
Gender (F, %)	45, 52.30	28, 32.20	9, 34.60	0.951	1.000	0.810	0.813
Age (years)	56.00 (48.50, 65.00)	58.00 (52.00, 64.00)	46.00 (33.00, 63.00)	0.060	0.467	0.053	0.024a
BMI (kg/m2)	22.03 (20.67, 23.28)	25.73 (24.11, 25.73)	23.45 (20.62, 26.54)	< 0.001b	< 0.001b	0.067	0.050a
Duration of diabetes, in months	-	97.65 (24.00, 97.65)	102.00 (36.00, 192.00)	0.490	-	-	0.490
Duration of GI, in months	-	-	1.50 (0.29, 15.00)	-	-	-	-
SBP, mmHg	119.00 (108.00, 127.00)	130.00 (125.00, 143.00)	127.50 (110.00, 150.00)	< 0.001b	< 0.001b	0.003b	0.424
DBP, mmHg	74.00 (67.50, 82.00)	82.00 (78.00, 88.00)	86.50 (76.00, 98.00)	< 0.001b	< 0.001b	< 0.001b	0.224
WBC, × 109/L	6.19 (5.11, 7.56)	6.81 (5.93, 7.44)	5.96 (4.77, 7.30)	0.113	0.088	0.484	0.096
Hb, g/L	134.00 (124.00, 147.00)	137.03 (126.00, 147.00)	129.00 (112.00, 137.00)	0.014a	0.236	0.042a	0.003b
PLT, × 109/L	224.00 (165.50, 257.50)	219.70 (201.00, 229.00)	211.50 (171.00, 262.00)	0.994	0.944	0.825	0.968
CRP, mg/L	0.00 (0.00, 1.28)	1.00 (0.00, 2.04)	2.10 (1.00, 3.60)	0.024a	0.007b	0.746	0.109
FBG, mmol/L	5.24 (4.90. 5.54)	8.49 (6.00, 10.00)	6.50 (5.00, 9.00)	< 0.001b	< 0.001b	0.001b	0.104
HbA1c, %	-	8.93 (7.40, 10.00)	7.93 (6.60, 10.40)	< 0.001b	-	-	< 0.001b
FC-P, nmol/L	-	0.81 (0.55, 0.83)	0.45 (0.25, 0.65)	< 0.001b	-	-	< 0.001b
2hC-P, nmol/L	-	1.84 (1.15, 1.84)	0.73 (0.34, 1.17)	< 0.001b	-	-	< 0.001b
BUN, mmol/L	5.28 (4.32, 6.34)	5.59 (4.60, 6.45)	4.81 (4.28, 6.02)	0.296	0.297	0.435	0.149
SCr, µmol/L	66.10 (52.95, 76.40)	68.90 (58.90, 76.10)	68.45 (49.70, 76.10)	0.528	0.253	0.989	0.571
eGFR, mL/min/1.73 m2	100.10 (100.10, 100.89)	94.83 (93.49, 103.37)	102.36 (88.93, 119.02)	0.011a	0.005b	0.360	0.052
UACR, mg/g	-	32.51 (7.61, 87.32)	66.19 (11.53, 694.73)	0.010a	-	-	0.010a
TG, mmol/L	1.48 (1.05, 1.52)	2.08 (1.29, 2.51)	1.39 (1.08, 2.08)	0.001b	< 0.001b	0.645	0.031a
TC, mmol/L	4.36 (4.04, 4.40)	4.25 (3.64, 4.97)	4.11 (3.59, 4.64)	0.657	0.614	0.361	0.650
LDL-C, mmol/L	2.62 (2.28, 2.70)	2.88 (2.25, 3.32)	2.64 (2.18, 3.37)	0.018a	0.002b	0.528	0.439
HDL-C, mmol/L	1.34 (1.24, 1.34)	1.11 (0.96, 1.20)	1.12 (0.94, 1.22)	< 0.001b	< 0.001b	< 0.001b	0.986
ALB, g/L	43.10 (39.85, 46.85)	42.39 (39.90, 44.20)	38.45 (36.30, 44.10)	0.044a	0.249	0.024a	0.060
ALT, U/L	15.40 (10.00, 19.10)	18.70 (12.50, 24.20)	14.05 (10.00, 21.00)	0.012a	0.004b	0.816	0.061
AST, U/L	17.60 (14.75, 21.95)	19.20 (14.80, 23.05)	14.35 (11.70, 20.10)	0.027a	0.264	0.027a	0.009b
Smoking history	17 (22.70)	17 (24.60)	4 (15.40)	0.640	0.838	0.579	0.414
Drinking history	17 (22.70)	17 (24.60)	7 (26.90)	0.914	0.838	0.791	1.000
Diabetic family history	3 (4.00)	10 (14.50)	2 (7.70)	0.079	0.040a	0.601	0.500
Hypertension	0	25 (36.20)	13 (50.00)	< 0.001b	< 0.001b	< 0.001b	0.247
Cerebral infarction	0	10 (14.50)	3 (11.50)	0.004b	< 0.001b	0.016a	0.750
Coronary atherosclerotic heart disease	0	7 (10.10)	2 (7.70)	0.018a	0.005b	0.064	1.000
Diabetic ketosis	-	1 (1.40)	11 (42.30)	< 0.001b	-	-	< 0.001b
Diabetic nephropathy	-	9 (13.00)	5 (19.20)	0.515	-	-	0.515
Diabetic autonomic neuropathy	-	0 (0.00)	10 (38.50)	< 0.001b	-	-	< 0.001b
Diabetic peripheral neuropathy	-	18 (26.10)	16 (61.50)	0.002b	-	-	0.002b
Diabetic retinopathy	-	3 (4.30)	4 (15.40)	0.086	-	-	0.086
Sulfonylureas	-	4 (5.80)	2 (7.70)	1.000	-	-	1.000
Glinides	-	3 (4.30)	1 (3.80)	1.000	-	-	1.000
Metformin	-	14 (20.30)	9 (34.60)	0.181	-	-	0.181
TZDs	-	0 (0.00)	1 (3.80)	0.275	-	-	0.275
AGI	-	2 (2.90)	1 (3.80)	1.000	-	-	1.000
SGLT-2i	-	9 (13.00)	4 (15.40)	1.000	-	-	1.000
DPP-4i	-	3 (4.30)	2 (7.70)	0.604	-	-	0.604
Insulin	-	10 (14.50)	17 (65.40)	< 0.001b	-	-	< 0.001b
GLP-1Ra	-	6 (8.70)	2 (7.70)	1.000	-	-	1.000

^aP < 0.05.

^bP < 0.01.

Data are presented as median (interquartile range). The groups were compared using Wilcoxon rank sum test. Monte Carlo method for testing exact significance (P). P1: Healthy control vs type 2 diabetes mellitus; P2: Healthy control vs diabetic gastric autonomic neuropathy; P3: Type 2 diabetes mellitus vs diabetic gastric autonomic neuropathy. T2DM: Type 2 diabetes mellitus; HC: Healthy control; DGAN: Diabetic gastric autonomic neuropathy; BMI: Body mass index; GI: Gastrointestinal; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; Hb: Hemoglobin; PLT: Platelets; CRP: C-reactive protein; FBG: Fasting blood glucose; HbA1c: Hemoglobin A1c; FC-P: Fasting C-peptide; 2hC-P: 2-hour c-peptide; BUN: Blood urea nitrogen; SCr: Serum creatinine; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio; TG: Triglycerides; TC: Total cholesterol; LDL-C: Low-density lipoprotein cholesterol; HDL-C: High-density lipoprotein cholesterol; ALB: Albumin; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TZDs: Thiazolidinediones; AGI: Alpha-glucosidase inhibitor; SGLT-2i: Sodium-glucose cotransporter-2 inhibitors; DPP-4i: Dipeptidyl peptidase-4 inhibitor; GLP-1Ra: Glucagon-like peptide-1 receptor agonist; F: Female.

Comparison of serum BA profiles among the participants

We quantified 15 different BAs in the HC, T2DM, and DGAN groups (Table 1 and Figure 2). Compared with the findings in the HC group, the levels of eight of ten conjugated BAs (GCA, GLCA, GDCA, glycochenodeoxycholic acid, GUDCA, TCA, TLCA, and TCDCA) were lower in both the T2DM and DGAN groups. Furthermore, DCA, UDCA, lithocholic acid, GUDCA, GLCA, TCA, TCDCA, TDCA, and TLCA levels were lower in the T2DM group than in the HC group, whereas the levels of these BAs were lower in the DGAN group than in the T2DM group. Further analysis revealed that the levels of GUDCA, GLCA, TCDCA, and TLCA sequentially decreased in the HC, T2DM, and DGAN groups (all P < 0.05).

Figure 2 Differences in bile acid levels among the healthy control, type 2 diabetes mellitus, and diabetic gastric autonomic neuropathy groups. A: Cholic acid; B: Deoxycholic acid; C: Chenodeoxycholic acid; D: Ursodeoxycholic acid; E: Lithocholic acid; F: Glycocholic acid; G: Glycodeoxycholic acid; H: Glycochenodeoxycholic acid; I: Glycoursodeoxycholic acid; J: Glycolithocholic acid; K: Taurocholic acid; L: Taurodeoxycholic acid; M: Taurochenodeoxycholic acid; N: Tauroursodeoxycholic acid; O: Taurolithocholic acid. Data are presented as the median (interquartile range), and significant difference was assessed using Wilcoxon’s rank sum test. ^aP < 0.05; ^bP < 0.01. T2DM: Type 2 diabetes mellitus; HC: Healthy control; DGAN: Diabetic gastric autonomic neuropathy; CA: Cholic acid; DCA: Deoxycholic acid; CDCA: Chenodeoxycholic acid; UDCA: Ursodeoxycholic acid; LCA: Lithocholic acid; GCA: Glycocholic acid; GDCA: Glycodeoxycholic acid; GCDCA: Glycochenodeoxycholic acid; GUDCA: Glycoursodeoxycholic acid; GLCA: Glycolithocholic acid; TCA: Taurocholic acid; TDCA: Taurodeoxycholic acid; TCDCA: Taurochenodeoxycholic acid; TUDCA: Tauroursodeoxycholic acid; TLCA: Taurolithocholic acid.

Spearman’s correlation analysis of associations among clinical indicators, BA levels, and GSRS scores

Regarding the correlations among clinical indicators, BA levels, and GSRS scores in the T2DM and DGAN groups (Figure 3A), we observed predominantly negative correlations between clinical indicators and GSRS scores. Specifically, constipation exhibited a significant positive correlation with TG levels (r = 0.411, P < 0.05; Figure 3B). By contrast diarrhea was negatively correlated with hemoglobin levels (r = -0.518, P < 0.01; Figure 3C). Spearman’s correlation analysis between BA levels and GSRS scores highlighted a strong negative correlation between DCA levels and constipation (r = -0.398,P < 0.05). Furthermore, dyspeptic symptoms were negatively correlated with GCA (r = −0.447, P < 0.05), TCA (r = -0.528,P < 0.01), TUDCA (r = -0.401, P < 0.05), GDCA (r = -0.431, P < 0.05), TLCA (r = -0.458, P < 0.05; Figure 3D), and TDCA levels (r = -0.629, P < 0.01; Figure 3E).

Figure 3 Spearman’s correlation analysis among bile acids, clinical indicators, and Gastrointestinal Symptom Rating Scale scores in the diabetic gastric autonomic neuropathy group. A: Correlations among clinical indicators, bile acid levels, and Gastrointestinal Symptom Rating Scale scores in the type 2 diabetes mellitus and diabetic gastric autonomic neuropathy groups; B-E: Scatter plots depicting the correlations among clinical indicators, bile acids, and the scores of gastrointestinal symptoms. ^aP < 0.05; ^bP < 0.01. GSRS: Gastrointestinal Symptom Rating Scale; eGFR: Estimated glomerular filtration rate; PLT: Platelets; LDL-C: Low-density lipoprotein cholesterol; TC: Total cholesterol; TG: Triglycerides; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; UACR: Urinary albumin-to-creatinine ratio; 2hC-P: 2-hour c-peptide; C-P: C-peptide; HDL-C: High-density lipoprotein cholesterol; T2DM: Type 2 diabetes mellitus; GI: Gastrointestinal; CA: Cholic acid; UDCA: Ursodeoxycholic acid; CDCA: Chenodeoxycholic acid; DCA: Deoxycholic acid; SCr: Serum creatinine; BUN: Blood urea nitrogen; AST: Aspartate aminotransferase; ALT: Alanine aminotransferase; ALB: Albumin; BMI: Body mass index; CRP: C-reactive protein; Hb: Hemoglobin; GUDCA: Glycoursodeoxycholic acid; GCDCA: Glycochenodeoxycholic acid; LCA: Lithocholic acid; HbA1c: Hemoglobin a1c; FBG: Fasting blood glucose; GDCA: Glycodeoxycholic acid; GLCA: Glycolithocholic acid; TLCA: Taurolithocholic acid; TDCA: Taurodeoxycholic acid; TCDCA: Taurochenodeoxycholic acid; GCA: Glycocholic acid; TCA: Taurocholic acid; TUDCA: Tauroursodeoxycholic acid.

OPLS-DA of BAs in the T2DM and DGAN groups

To identify biomarkers that differentiate patients with uncomplicated T2DM from those with DGAN, we employed OPLS-DA using a comprehensive panel of 15 BAs. Using SIMCA 14.1, we normalized the data to ensure comparability. The OPLS-DA model as hypothesis-generating effectively delineated patients with uncomplicated T2DM from those with DGAN (Figure 4). The negative Q² value indicated that the OPLS-DA model lacked independent predictive value; it was used primarily for variable selection (identifying TLCA) rather than for classification. We evaluated the significance of these metabolites in the differentiated groups based on VIP scores and FC (Table 2 and Figure 4C). TLCA had VIP scores exceeding 1, highlighting its potential as a discriminative biomarker. Additionally, FC was lower than 0.5 for TLCA. We then selected clinical indicators and BAs that exhibited statistical significance (P < 0.05, univariate analysis), VIP > 1, and FC < 0.5 between the T2DM and DGAN groups. Only meaningful data are presented in Table 2. These metabolites were deemed differential and subsequently used for logistic regression analysis.

Figure 4 Multivariate analysis of bile acids in the orthogonal partial least squares discriminant analysis model. A: Orthogonal partial least squares discriminant analysis three-dimensional model between uncomplicated type 2 diabetes mellitus and diabetic gastric autonomic neuropathy. R2X = 0.406, R2Y = 0.183, and Q2Y = -0.274; B: The permutation test demonstrated no overfitting in the orthogonal partial least squares discriminant analysis model [Q2 = (0, -0.386), R2 = (0, 0.152)]; C: The contribution of the metabolites distinguishing uncomplicated type 2 diabetes mellitus from diabetic gastric autonomic neuropathy as indicated by variable importance in projection values. T2DM: Type 2 diabetes mellitus; DGAN: Diabetic gastric autonomic neuropathy; TLCA: Taurolithocholic acid; GUDCA: Glycoursodeoxycholic acid; GLCA: Glycolithocholic acid; LCA: Lithocholic acid; UDCA: Ursodeoxycholic acid; CDCA: Chenodeoxycholic acid; GDCA: Glycodeoxycholic acid; TDCA: Taurodeoxycholic acid; CA: Cholic acid; GCA: Glycocholic acid; TCA: Taurocholic acid; TCDCA: Taurochenodeoxycholic acid; GCDCA: Glycochenodeoxycholic acid; DCA: Deoxycholic acid; TUDCA: Tauroursodeoxycholic acid.

Table 2 Variable importance in projection scores and fold change values of bile acids in the type 2 diabetes mellitus and diabetic gastric autonomic neuropathy groups.

Variables	VIP value	FC	Log2FC	Q value
TLCA	1.364	0.248	-2.014	0.027a

^aQ < 0.05.

Q value, P value after false discovery rate adjusted. VIP: Variable importance in projection; FC: Fold change; TLCA: Taurolithocholic acid.

Differential variables for predicting DGAN in patients with T2DM based on logistic analysis

To refine the variable selection process and mitigate the risk of overfitting, we performed penalized regression [least absolute shrinkage and selection operator (LASSO)] regression analysis on the 22 clinical variables. The optimal penalty parameter (λ) was determined via ten-fold cross-validation, which identified 12 variables with non-zero coefficients for inclusion in the subsequent regression models (Supplementary Figure 1). Multivariable logistic regression models were constructed using the 12 variables selected by LASSO regression (Supplementary Table 1). Model 1 included age, BMI, hemoglobin, FC-P, 2hC-P, UACR, albumin, and conjugated BAs (CA, GLCA, TCA, TDCA, TUDCA). Model 2 included all model 1 variables plus TLCA. The odds ratios (ORs) and P values for all variables were highly consistent between the two models, demonstrating remarkable robustness. Several variables, including BMI (model 1, P = 0.078; model 2, P = 0.069), CA (model 1, P = 0.060; model 2, P = 0.058), and 2hC-P (model 1, P = 0.111; model 2, P = 0.094), exhibited near-significant associations, identifying them as potential predictors worthy of further investigation. Although the incorporation of TLCA in model 2 (OR = 1.241, P = 0.553) did not achieve statistical significance, its successful estimation provides a reliable basis and a novel direction for future investigations.

We further evaluated the potential confounding effects of key glucose-lowering medications (insulin and metformin) (Supplementary Table 2). When these medications were included as covariates in the multivariable logistic regression model, insulin use emerged as the strongest independent risk factor for DGAN (OR = 54.73, P = 0.001) while the associations of TLCA and FC-P were no longer statistically significant (both P > 0.5). To address potential confounding effects of glucose-lowering medications, we performed a correlation analysis including the medication status and its interaction with TLCA levels. As presented in Supplementary Table 3, both insulin use (F = 7.777, P = 0.008) and TLCA levels (F = 3.135, P < 0.001) remained significantly associated with the group classification (DGAN vs T2DM) after adjusting for these factors. Notably, metformin use (P = 0.003) and insulin use (P = 0.01) were significantly associated with TLCA levels, suggesting that the relationship between TLCA and DGAN risk can be modulated by medication use.

This result strongly suggested that disease severity and treatment stage were the primary drivers of DGAN risk. Given that insulin use is highly correlated with β cell failure (reflected by low FC-P levels) and overall metabolic dysregulation, we posited that alterations in TLCA levels may be intrinsically linked to this severe disease phenotype rather than acting as an independent causal factor. As the primary aim of this study was to explore the association between BA metabolism and DGAN and considering that including highly collinear clinical treatment variables alongside pathophysiological biomarkers may introduce a risk of overadjustment, we retained the biologically plausible variables (age, FC-P, TLCA) identified in our initial analyses for the subsequent construction of the predictive model.

Univariate regression analysis (Table 3) identified age, BMI, hemoglobin, FC-P, 2hC-P, albumin, and TLCA as protective factors against DGAN, whereas UACR emerged as a risk factor. In the multivariate logistic regression analysis (Table 4), we adjusted for numerous confounding factors (P < 0.05, univariate analysis) in model 1. Age exhibited an independent protective effect against the development of DGAN. Interestingly, FC-P and UACR lost their roles as protective and risk factors. In model 2 the protective factors included FC-P and TLCA in addition to age. Based on VIP scores, univariate analysis results, and logistic regression findings, TLCA was identified as a significant biomarker.

Table 3 Univariate logistic regression analysis of clinical indicators and bile acids in the type 2 diabetes mellitus and diabetic gastric autonomic neuropathy groups.

Variables	Univariate logistic regression analysis
	OR (95%CI)	P value
Clinical indicators
Gender, F, %	1.131 (0.436-2.935)	0.800
Age, years	0.941 (0.904-0.978)	0.002b
BMI, kg/m2	0.759 (0.635-0.907)	0.002b
Duration of diabetes, months	1.002 (0.997-1.008)	0.388
SBP, mmHg	0.999 (0.973-1.025)	0.929
DBP, mmHg	1.029 (0.988-1.071)	0.172
WBC, × 109/L	0.855 (0.650-1.123)	0.260
Hb, g/L	0.962 (0.936-0.989)	0.006b
PLT, × 109/L	1.000 (0.992-1.008)	0.995
CRP, mg/L	1.068 (0.987-1.157)	0.103
FBG, mmol/L	0.874 (0.732-1.044)	0.137
HbA1c, %	0.895 (0.720-1.113)	0.320
FC-P, nmol/L	0.062 (0.011-0.339)	0.001b
2hC-P, nmol/L	0.185 (0.078-0.438)	< 0.001b
BUN, mmol/L	0.873 (0.683-1.117)	0.281
SCr, µmol/L	1.004 (0.991-1.017)	0.555
eGFR, mL/min/1.73 m2	1.013 (0.990-1.036)	0.264
UACR, mg/g	1.001 (1.000-1.003)	0.037a
TG, mmol/L	0.551 (0.327-1.929)	0.065
TC, mmol/L	0.985 (0.677-1.433)	0.937
LDL-C, mmol/L	0.864 (0.540-1.382)	0.542
HDL-C, mmol/L	0.393 (0.052-2.972)	0.365
ALB, g/L	0.895 (0.808-0.992)	0.034a
ALT, U/L	0.960 (0.916-1.006)	0.088
AST, U/L	0.948 (0.887-1.012)	0.111
BAs characteristics
CA, nmol/L	1.001 (1.000-1.002)	0.070
DCA, nmol/L	1.000 (0.999-1.001)	0.813
CDCA, nmol/L	1.000 (1.000-1.001)	0.386
UDCA, nmol/L	0.997 (0.993-1.001)	0.169
LCA, nmol/L	0.996 (0.979-1.014)	0.673
GCA, nmol/L	1.000 (1.000-1.001)	0.276
GLCA, nmol/L	0.948 (0.886-1.014)	0.120
GDCA, nmol/L	1.000 (0.999-1.001)	0.663
GCDCA, nmol/L	1.000 (1.000-1.000)	0.832
GUDCA, nmol/L	0.999 (0.996-1.002)	0.561
TCA, nmol/L	1.003 (0.998-1.007)	0.226
TLCA, nmol/L	0.603 (0.377-0.964)	0.035a
TDCA, nmol/L	0.985 (0.968-1.003)	0.106
TCDCA, nmol/L	1.001 (0.999-1.003)	0.201
TUDCA, nmol/L	1.014 (0.986-1.043)	0.333

^aP < 0.05.

^bP < 0.01.

BMI: Body mass index; SBP: Systolic blood pressure; DBP: Diastolic blood pressure; Hb: Hemoglobin; PLT: Platelets; CRP: C-reactive protein; FBG: Fasting blood glucose; HbA1c: Hemoglobin a1c; FC-P: Fasting C-peptide; 2hC-P: 2-hour C-peptide; BUN: Blood urea nitrogen; SCr: Serum creatinine; eGFR: Estimated glomerular filtration rate; UACR: Urinary albumin-to-creatinine ratio; TG: Triglycerides; TC: Total cholesterol; LDL-C: Low-density lipoprotein cholesterol; HDL-C: High-density lipoprotein cholesterol; ALB: Albumin; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; CA: Cholic acid; DCA: Deoxycholic acid; CDCA: Chenodeoxycholic acid; UDCA: Ursodeoxycholic acid; LCA: Lithocholic acid; GCA: Glycocholic acid; GLCA: Glycolithocholic acid; GDCA: Glycodeoxycholic acid; GCDCA: Glycochenodeoxycholic acid; GUDCA: Glycoursodeoxycholic acid; TCA: Taurocholic acid; TLCA: Taurolithocholic acid; TDCA: Taurodeoxycholic acid; TCDCA: Taurochenodeoxycholic acid; TUDCA: Tauroursodeoxycholic acid; OR: Odds ratios; CI: Confidence interval; F: Female; BA: Bile acid.

Table 4 Multivariate logistic regression analysis of clinical indicators and bile acids in the type 2 diabetes mellitus and diabetic gastric autonomic neuropathy groups.

Variables	Multivariate logistic regression analysis
	Model 1		Model 2
	OR (95%CI)	P value	OR (95%CI)	P value
Age, years	0.919 (0.851-0.993)	0.033a	0.882 (0.779-0.998)	0.047a
BMI, kg/m2	0.961 (0.737-1.252)	0.768	1.127 (0.813-1.562)	0.471
Hb, g/L	0.987 (0.935-1.042)	0.642	0.955 (0.884-1.032)	0.243
FC-P, nmol/L	0.060 (0.001-2.406)	0.135	0.000 (0.000-0.253)	0.017a
2hC-P, nmol/L	0.348 (0.063-1.936)	0.228	0.660 (0.107-4.072)	0.654
UACR, mg/g	1.002 (0.998-1.006)	0.302	1.003 (0.997-1.008)	0.330
ALB, g/L	0.859 (0.694-1.063)	0.163	0.824 (0.601-1.130)	0.231
TLCA, nmol/L	/	/	0.160 (0.040-0.631)	0.009b

^aP < 0.05.

^bP < 0.01.

BMI: Body mass index; Hb: Hemoglobin; FC-P: Fasting C-peptide; 2hC-P: 2-hour C-peptide; UACR: Urinary albumin-to-creatinine ratio; ALB: Albumin; TLCA: Taurolithocholic acid; OR: Odds ratios; CI: Confidence interval.

The results of the receiver operating characteristic analysis of the DGAN risk-prediction model are presented in Table 5. The AUCs for age, FC-P, and TLCA in discriminating between T2DM and DGAN were 0.651, 0.760, and 0.678 (P < 0.05), respectively. The AUC of model 2 for distinguishing T2DM and DGAN was 0.970 (95% confidence interval: 0.937-1.000), indicating greater predictive ability than the individual differentiating parameters and model 1 (AUC = 0.933, 95% confidence interval: 0.878-0.989). Moreover, these independent variables were combined to create a predictive model for identifying DGAN in patients with T2DM (Figure 5A). The clinical validity of this predictive model was assessed using decision curve analysis (Figure 5B), indicating that when the threshold probabilities ranged from 9.0% to 68.0%, a higher net clinical benefit was achieved compared with those of hypothetical testing scenarios. The Hosmer-Lemeshow goodness-of-fit test for the nomogram model (Supplementary Table 4) indicated good calibration: P = 0.096. The Integrated Discrimination Improvement for adding TLCA to the clinical model was 5.1% (P < 0.01), indicating an improvement in predictive performance. Thus, the inclusion of TLCA significantly enhanced the clinical diagnosis of DGAN.

Figure 5 Nomogram model and results of decision curve analysis for quantifying the individual risk of diabetic gastric autonomic neuropathy in patients with type 2 diabetes mellitus. A: The nomogram includes age, fasting C-peptide, and taurolithocholic acid along with their corresponding points and total risk scores; B: Decision curve analysis was performed to evaluate the clinical net benefit across different risk thresholds. TLCA: Taurolithocholic acid; FC-P: Fasting C-peptide; DGAN: Diabetic gastric autonomic neuropathy.

Table 5 Area under the curve of bile acids in the type 2 diabetes mellitus and diabetic gastric autonomic neuropathy groups.

Variables	AUC (95%CI)	P value	Cutoff value	Sensitivity	Specificity
Age, years	0.651 (0.504-0.797)	0.024a	0.450	0.462	0.942
FC-P, nmol/L	0.760 (0.648-0.872)	< 0.001b	0.188	0.885	0.638
TLCA, nmol/L	0.678 (0.565-0.791)	< 0.001b	0.219	0.923	0.435
Model 1	0.933 (0.878-0.989)	< 0.001b	0.277	0.885	0.884
Model 2	0.970 (0.937-1.000)	< 0.001b	0.392	0.923	0.957

^aP < 0.05.

^bP < 0.01.

AUC: Area under the curve; FC-P: Fasting C-peptide; TLCA: Taurolithocholic acid; CI: Confidence interval.

DISCUSSION

This study provided an integrated clinical-metabolic portrait of DGAN. Three observations warrant further emphasis. First, broad suppression of conjugated BA levels was observed in diabetes with this suppression deepening in DGAN. Four species, namely GUDCA, GLCA, TCDCA, and most strikingly TLCA, declined stepwise from HC to uncomplicated T2DM to DGAN. Second, TLCA emerged as an independent protective factor, and its depletion was correlated with higher GSRS scores. The inclusion of TLCA in the multivariable model increased the AUC to 0.97, and its superior clinical net benefit was verified by decision curve analysis. Lastly, several readily available clinical variables distinguished DGAN from T2DM and yielded additional pathophysiological clues.

We observed that patients with T2DM with DGAN were predominantly younger individuals who were insulin-dependent. Moreover, these patients often presented with diabetic ketoacidosis as an initial manifestation of T2DM. This indicated severely impaired pancreatic function. Their younger age, combined with their busy and irregular lifestyles, likely influenced the neglect of their health. This prevented the timely detection of glycemic fluctuations and associated physical changes. Consequently, they often sought medical attention only when symptoms became severe. At this stage prolonged poor glycemic control might have caused hyperglycemia-induced toxicity, leading to gastrointestinal autonomic neuropathy.

Although these patients are relatively young, they exhibit various neurological complications, such as chest tightness, resting palpitations, urinary retention, peripheral numbness, and tingling. This underscores the importance of early glycemic management in preventing diabetes-related complications. Furthermore, HbA1c ≥ 7.0% increases the risk of gastric motility disorders, such as gastroparesis[26]. Counterintuitively, although HbA1c ≥ 7.0% increases gastroparesis risk[16], patients with DGAN had significantly lower HbA1c and C-peptide levels than those with uncomplicated T2DM.

This finding was not associated with DGAN development in univariate analysis, but it can be explained by two non-exclusive mechanisms: A rapidly-progressive phenotype characterized by premature β cell failure, which leads to a catabolic state with significant weight loss prior to glycemic stabilization; and chronic gastrointestinal symptoms, such as reduced appetite, vomiting, and food aversion, causing malnutrition and consequently lower BMI and HbA1c levels. This finding is consistent with that of Yuan et al[27] with corroborating biochemical evidence illustrating that reductions in hemoglobin levels correlating with diarrhea severity (suggesting malabsorption) and a negative constipation-TG association reflecting impaired chylomicron absorption during slowed intestinal transit.

Traditionally, the gold standards for diagnosing DGAN are gastric emptying scintigraphy and the gastric emptying breath test[2,28,29]. However, the administration of specific test meals can exacerbate symptoms in patients with frequent vomiting and severe constipation. Therefore, beyond early clinical identification based on symptoms, potential biomarkers must be explored. Our findings provide novel diagnostic and therapeutic insights. BAs are signaling molecules that influence multiple systems and play a crucial role in gastrointestinal diseases, such as digestive tract tumors, BA diarrhea, and bile reflux[7,27]. However, their relationship with DGAN remains underexplored.

In our study the comparison of serum BA profiles among HCs, patients with T2DM, and patients with DGAN revealed the protective role of TLCA in DGAN development. TLCA levels were negatively correlated with dyspeptic symptoms in patients with DGAN. The inclusion of TLCA improved the diagnostic sensitivity and specificity of the regression model, suggesting its potential as a biomarker for DGAN. TLCA was initially identified in bovine bile and gallstones, and it is a secondary BA derived from the primary BA UDCA. TLCA has also been detected in human bile[30]. As a high-affinity ligand for akeda G-protein-coupled receptor 5 (TGR5)[29,31], TLCA promotes GLP-1 secretion through the TGR5-cAMP-Ca²⁺ pathway in L cells of the mouse intestine[22], resulting in improved glucose metabolism[32,33]. Simultaneously, TLCA levels were lower in patients with diabetes than in HCs[34] as also reflected in our study.

Beyond its role in stimulating gut hormone secretion, TLCA level activation exerts anti-inflammatory and neuroprotective effects. More than 50% of intestinal neurons in the mouse enteric nervous system express TGR5[35]. BAs can act on TGR5 in the intestine to promote defecation in mice[36]. In murine models TGR5 signaling attenuates 5-hydroxytryptamine and nitric oxide release from macrophages, reduces endothelial adhesion molecule expression, and modulates neural inflammation[31]. These actions are particularly relevant in the context of diabetic neuropathy in which chronic low-grade inflammation and oxidative stress contribute to neuronal damage. Moreover, TGR5 is expressed in human enteric neurons and enterochromaffin cells, suggesting a direct pathway through which BAs can influence gut motility and sensory function[36]. This association might be linked to the downstream pathways of gastrointestinal receptors; however, the underlying mechanisms require further validation through animal experiments[22,31-33].

An apparent discrepancy exists between circulating TLCA levels in humans (in the nanomolar range) and the micromolar concentrations typically required to activate TGR5 in vitro, raising questions about its mechanistic relevance in vivo. This paradox can likely be explained by the localized nature of BA signaling. BAs primarily act in specific anatomical compartments, such as the intestinal lumen, biliary tree, and portal venous system, in which their concentrations can reach millimolar levels, far exceeding those in serum and becoming sufficient to robustly activate TGR5[37,38]. The high first-pass clearance (70%-90%) by the liver ensures that systemic BA concentrations reflect downstream “spillover” rather than local activity[37]. Thus, measuring peripheral blood TCLA levels might underestimate its local bioactivity at target sites. Serum TLCA levels likely serve as a surrogate marker reflecting systemic disruption of enterohepatic circulation, rather than directly representing bioactive concentrations at neuronal TGR5 sites.

The strategic distribution of TGR5 in the human digestive tract supports its functional importance. TGR5 is expressed in regions with the highest BA concentration (ileum) or greatest bioactive potential (colon), positioning it as a key sensor for BA signals that regulate metabolism, immunity, and motility[39]. Although its expression pattern in the enteric nervous system might differ from that in mice[35], the functional activity of TGR5 in human gastrointestinal epithelial and immune cells is well established. For instance, TGR5 expression progressively increases during the progression from precancerous lesions to invasive carcinoma in the biliary epithelium[40], and its expression is elevated in gastric cancer in which it promotes metaplasia[41]. These findings unequivocally demonstrate an active TGR5 pathway in human gastrointestinal tissues.

Moreover, direct evidence from human diseases supports the relevance of the TLCA-TGR5 axis. Patients with T2DM exhibit significantly lower serum TLCA levels than HCs[34] and reduced expression of key BA transporters in the gastrointestinal tract[42], consistent with our observation and inversely corroborating our findings in mice.

An important finding of this study is the widespread suppression of conjugated BAs, a pattern highly consistent with ileal dysfunction. DGAN likely impairs ileal reabsorption of BAs via the apical sodium-dependent BA transporter transporter, leading to increased fecal loss and a reduced circulating pool. This is supported by the established link between diabetic and abnormal BA metabolism[43]. Furthermore, intestinal inflammation and barrier defects (common in DGAN) can directly disrupt apical sodium-dependent BA transporter function, promoting colonic BA accumulation and diarrhea[44].

However, the interpretation of our findings must be considered in the context of glucose-lowering medications. A sensitivity analysis adjusting for insulin and metformin use indicated that the association of TLCA and FC-P with DGAN was attenuated while insulin use itself emerged as a powerful predictor (Supplementary Table 3). This result strongly suggests that disease severity and treatment stage are the primary drivers of DGAN risk. We interpret this finding from the following perspective: The serum BA profile, particularly TLCA, may serve as a sensitive biomarker for this rapidly progressive, insulin-dependent, high DGAN-risk diabetic phenotype. In other words, BA metabolic dysregulation may be an intrinsic component of this high-risk phenotype rather than a treatment-independent predictor.

Including therapeutic variables alongside fundamental metabolic biomarkers in the same model may pose a risk of overadjustment as the treatment itself is predicated on the severity of the disease, a state that is likely mediated at least in part by mechanisms such as BA metabolic disruption. Consequently, we propose that for risk stratification models aimed at early identification, utilizing BA biomarkers that reflect underlying physiological states hold greater prospective value and biological relevance than relying on treatment decisions (such as insulin use). Future prospective studies involving repeated measurements of the BA profile prior to the initiation of insulin therapy in patients are warranted to clarify the causal relationship with DGAN risk.

The strengths of this study included the simultaneous assessment of detailed BA spectra, validated symptom scoring, and rigorous multivariate modeling. However, this study had some limitations. These include its cross-sectional design, modest overall sample size, and notably the particularly small sample size of the DGAN group (n = 26) that inherently limited the robustness and generalizability of our predictive model despite the use of LASSO regression. Consequently, this work is exploratory, and the observed discriminatory performance (AUC = 0.970) must be viewed as preliminary. Recruitment from a single center and the lack of direct measurements of the gastrointestinal microbiota or fecal BA output are additional constraints. Therefore, causality remains inferential, and external validation in larger, multicenter cohorts is required before any routine clinical recommendation can be offered; we have planned subsequent studies for this purpose.

Longitudinal cohorts should determine whether declining TLCA precedes symptom onset and predicts responses to BA-centered therapies. Although we adjusted for the use of key anti-diabetic medications (such as metformin and insulin), we did not comprehensively collect information on the use of other glucose-lowering or lipid-lowering drugs. Additionally, although the exclusion of metformin and GLP-1 receptor agonists was methodologically necessary to limit confounding effects on gastrointestinal function and BA metabolism, this might have reduced the generalizability of our findings as these drugs are first-line therapies for T2DM. Furthermore, important confounding factors that can influence BA profiles and disease risk, such as dietary patterns, physical activity, and the specific duration and severity of diabetes, were not included in our analysis. These unmeasured residual confounding factors might have influenced our results. DGAN involves pan-enteric dysfunction, and our gastroparesis-focused diagnostic criterion represents a limitation as the study cohort may not fully reflect colonic or intestinal dysmotility underlying symptoms such as diarrhea or constipation. Future studies with more detailed prospective data collection are warranted to further validate our findings.

CONCLUSION

A simple risk assessment model based on five clinical indicators and serum BAs could reliably predict DGAN risk in patients with T2DM. Timely and effective interventions could prevent gastrointestinal symptoms in patients with T2DM. This predictive model could assist clinicians in assessing the risk in patients with T2DM. Furthermore, it could be used to design future clinical trials aimed at preventing DGAN in individuals with T2DM.

ACKNOWLEDGEMENTS

The authors acknowledge resources and support from the Experimental Center of The Second Affiliated Hospital of Nanjing Medical University.