Triglyceride–glucose body mass index and metabolic dysfunction-associated steatotic liver disease in hospitalized patients with type 2 diabetes: a sex-stratified cross-sectional study

Abstract

Background

Metabolic dysfunction-associated steatotic liver disease (MASLD) is common in patients with type 2 diabetes mellitus (T2DM). The triglyceride–glucose body mass index (TyG-BMI), which integrates insulin resistance and adiposity, may improve MASLD risk stratification, but evidence in hospitalized T2DM patients remains limited.

Methods

This hospital-based cross-sectional study included 324 hospitalized adults with T2DM. MASLD was assessed by abdominal ultrasonography. TyG-BMI was calculated as TyG index × BMI and compared with conventional metabolic indices using receiver operating characteristic analysis, multivariable logistic regression, spline modelling, reclassification metrics, subgroup and sensitivity analyses. Testosterone-related analyses were performed in male patients.

Results

MASLD was identified in 182 patients. TyG-BMI showed the highest discrimination for MASLD, with an area under the curve of 0.821 (95% CI: 0.775–0.866), outperforming TyG, BMI, atherogenic index of plasma, and hepatic steatosis index. The optimal cut-off was 227.22, with 75.1% sensitivity and 78.7% specificity. In the fully adjusted model, TyG-BMI remained independently associated with MASLD as a continuous variable and across quartiles; patients in the highest quartile had markedly higher odds of MASLD than those in the lowest quartile. TyG-BMI also improved reclassification beyond TyG alone. In men, testosterone was inversely correlated with TyG-BMI and showed a nonlinear association with MASLD risk.

Conclusions

TyG-BMI showed better non-invasive discrimination for MASLD than conventional metabolic indices in hospitalized patients with T2DM. As a readily available marker integrating insulin resistance and adiposity, TyG-BMI may be useful for first-line MASLD risk stratification in endocrinology inpatient settings. Testosterone-related findings in men were exploratory and warrant prospective validation.

Clinical trial number

Not applicable.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12902-026-02321-y.

Background

Metabolic dysfunction-associated steatotic liver disease (MASLD), formerly termed non-alcoholic fatty liver disease (NAFLD) [1], has emerged as the most prevalent chronic liver disorder worldwide, affecting an estimated 25–30% of the global adult population and imposing an escalating burden on healthcare systems across all income strata [2, 3]. The earlier metabolic dysfunction-associated fatty liver disease (MAFLD) framework also emphasized metabolic dysfunction as a defining feature [4]. MASLD encompasses a clinical and histological spectrum ranging from simple steatosis to metabolic dysfunction-associated steatohepatitis (MASH, formerly NASH), advanced fibrosis, cirrhosis, and hepatocellular carcinoma [1, 5]. The epidemiological intersection of MASLD with the global diabetes pandemic is particularly alarming: among individuals with type 2 diabetes mellitus (T2DM), MASLD prevalence reaches 55–75%, a figure two- to threefold higher than in metabolically healthy counterparts [6, 7]. This co-occurrence is not merely coincidental; T2DM and MASLD share a deeply intertwined pathophysiological substrate rooted in insulin resistance, dyslipidemia, and visceral adiposity, creating a self-reinforcing cycle that accelerates hepatic and metabolic complications [8, 9].

The mechanistic underpinning of MASLD in T2DM centers on hepatic insulin resistance, which paradoxically sustains gluconeogenesis while simultaneously driving de novo lipogenesis through constitutive activation of sterol regulatory element-binding protein 1c (SREBP-1c) and carbohydrate-responsive element-binding protein (ChREBP) [10, 11]. Concurrently, impaired suppression of adipose tissue lipolysis augments free fatty acid flux to the liver, overwhelming mitochondrial β-oxidation capacity and promoting triglyceride accumulation within hepatocytes [12]. Hyperinsulinaemia further suppresses hepatic apolipoprotein B secretion, impairing very-low-density lipoprotein export and compounding intrahepatic lipid retention [13]. These converging metabolic insults, amplified by adipokine dysregulation, gut microbiome perturbation, and low-grade systemic inflammation, collectively constitute the ‘multiple-hit’ pathogenesis of MASLD in the diabetic milieu [14, 15].

Despite this mechanistic clarity, the clinical identification of MASLD in T2DM patients remains suboptimal. Liver biopsy, the diagnostic gold standard, is invasive, costly, and subject to sampling variability, rendering it unsuitable for population-level screening [16]. Abdominal ultrasonography, while widely deployed, requires trained operators and fails to detect steatosis below the 20–30% threshold [17]. Serum biomarker panels such as the Fatty Liver Index (FLI) and Hepatic Steatosis Index (HSI) offer moderate discriminative performance but incorporate variables not universally available in routine clinical practice [18, 19]. Consequently, a substantial proportion of T2DM patients with clinically significant MASLD remain undiagnosed until advanced hepatic injury has already occurred.

The triglyceride-glucose (TyG) index, calculated as the natural logarithm of the product of fasting triglycerides and fasting glucose, was originally validated as a surrogate measure of insulin resistance against the hyperinsulinemic-euglycemic clamp, demonstrating sensitivity and specificity exceeding 80% in multiple ethnic cohorts [20, 21]. Its association with MASLD has been documented across general population studies, yet its performance in T2DM-specific cohorts is attenuated by the relative insensitivity of TyG to the adiposity dimension of metabolic risk [22, 23]. The TyG-BMI index, derived by multiplying TyG by body mass index (BMI), was proposed to address this limitation by integrating both insulin resistance and adiposity into a single composite metric [24]. Emerging evidence from general population cohorts suggests that TyG-BMI outperforms TyG alone in predicting MASLD, with AUC values consistently in the range of 0.78–0.85 [25–27]. However, the comparative performance of TyG-BMI against a comprehensive panel of lipid-related indices—including the atherogenic index of plasma (AIP), TG/HDL-C ratio, and non-HDL cholesterol—has not been systematically evaluated in hospitalized T2DM patients, a population with particularly complex metabolic profiles and high MASLD burden.

Furthermore, the sex-specific modulation of MASLD risk by androgens in T2DM patients represents an underexplored dimension of this disease. Testosterone exerts hepatoprotective effects through androgen receptor-mediated suppression of hepatic lipogenesis and enhancement of mitochondrial fatty acid oxidation [28, 29]. Epidemiological studies have consistently linked low testosterone to steatotic liver disease and steatohepatitis severity in men, yet the extent to which this association is related to insulin resistance and adiposity—as captured by TyG-BMI—remains unresolved [30, 31]. The present study was designed to address these gaps through a comprehensive comparative analysis of TyG-BMI and other metabolic indices for MASLD risk stratification in a well-characterised cohort of hospitalized T2DM patients, with integrated sex-stratified testosterone analysis and exploratory mediation modelling.

Methods

Study design and participants

This hospital-based cross-sectional study enrolled consecutive adult patients with confirmed T2DM admitted to the Department of Endocrinology, Third People’s Hospital of Hefei (Hefei Third Clinical College of Anhui Medical University), between January and December 2024. T2DM was diagnosed in accordance with the 2023 Standards of Medical Care in Diabetes issued by the American Diabetes Association [32]. Inclusion criteria required: (i) age ≥ 18 years; (ii) confirmed T2DM diagnosis; and (iii) availability of baseline anthropometric and fasting metabolic data obtained within 24 h of admission. Patients were excluded if they had: type 1 diabetes or other specific forms of diabetes mellitus; eGFR < 15 mL/min/1.73 m²; active malignancy; known viral hepatitis B or C; alcoholic liver disease (alcohol intake > 140 g/week in males or > 70 g/week in females); use of hepatotoxic medications; pregnancy or lactation; or missing key variables. After applying these criteria and removing records with biologically implausible values (BMI < 14 or > 60 kg/m²; FBG < 2.0 or > 30.0 mmol/L; TG > 20.0 mmol/L), 324 patients were retained for analysis (Additional file 1: Figure S1). The study protocol was approved by the Ethics Committee of the Third People’s Hospital of Hefei (Approval No. Hwk2024zd008) and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all participants.

Fig. 1.

Spearman correlation heatmap of metabolic variables (N = 324). Colour intensity and direction indicate the magnitude and sign of Spearman rank correlation coefficients between all continuous metabolic variables. Red cells denote positive correlations; blue cells denote negative correlations. Asterisks indicate statistical significance: *p < 0.05, **p < 0.01, ***p < 0.001. TyG-BMI exhibited the strongest positive correlations with triglycerides (TG), body mass index (BMI), and alanine aminotransferase (ALT), and the strongest inverse correlation with high-density lipoprotein cholesterol (HDL-C). Abbreviations: TyG, triglyceride-glucose index; TyG-BMI, triglyceride-glucose body mass index; FBG, fasting blood glucose; TC, total cholesterol; UA, uric acid; HbA1c, glycated haemoglobin A1c; eGFR, estimated glomerular filtration rate; AIP, atherogenic index of plasma; ALT, alanine aminotransferase; AST, aspartate aminotransferase

MASLD diagnosis

MASLD was diagnosed by abdominal ultrasonography performed by experienced radiologists blinded to laboratory results, using standardized criteria including increased hepatic echogenicity relative to the renal cortex, posterior acoustic attenuation, and vascular blurring [33]. Secondary causes of hepatic steatosis were systematically excluded as described above. Ultrasonography has been validated as a reliable first-line screening modality for hepatic steatosis, with pooled sensitivity of 84.8% and specificity of 93.6% compared with liver biopsy in a large meta-analysis [17].

Anthropometric and laboratory measurements

Anthropometric measurements were obtained by trained nursing staff using standardized procedures on the first day of hospitalization. Body weight was measured with participants wearing light clothing, and height was measured without shoes. BMI was calculated as weight (kg) divided by height squared (m²). Baseline blood pressure was obtained from the admission vital-sign record. Fasting venous blood samples were collected within 24 h after admission after a minimum 8-hour overnight fast. Routine biochemical parameters, including fasting blood glucose (FBG), glycated haemoglobin A1c (HbA1c), total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), uric acid (UA), creatinine, alanine aminotransferase (ALT), aspartate aminotransferase (AST), and 25-hydroxyvitamin D [25(OH)D], were measured in the central clinical laboratory of the Third People’s Hospital of Hefei using standardized automated procedures. FBG, lipid profiles, UA, creatinine, ALT, and AST were measured using a Roche cobas 8000 modular analyzer series equipped with a cobas c702 clinical chemistry module (Roche Diagnostics, Basel, Switzerland). HbA1c was measured using a Roche automated HbA1c analytical platform according to the manufacturer’s instructions. Total testosterone and 25(OH)D were measured by electrochemiluminescence immunoassay using Roche cobas immunoassay systems with Roche Elecsys assays. Manufacturer-recommended reagents, calibrators, and internal quality-control materials were used for all assays. Internal quality control was performed daily, and calibration was conducted according to the manufacturer’s instructions and the laboratory’s standard operating procedures. eGFR was calculated using the CKD-EPI equation [34].

Index calculations

The TyG index was calculated as: TyG = ln[TG (mg/dL) × FBG (mg/dL) / 2], with unit conversions applied (TG: mmol/L × 88.57; FBG: mmol/L × 18.02) [20, 21]. TyG-BMI was defined as TyG × BMI (kg/m²) [24]. The atherogenic index of plasma (AIP) was calculated as log₁₀(TG/HDL-C) [35]. Additional comparator indices included the TG/HDL-C molar ratio, non-HDL cholesterol (TC − HDL-C), serum uric acid, eGFR, and HSI. HSI was calculated as 8 × ALT/AST + BMI + 2 if female + 2 if diabetes. Because all participants had T2DM, 2 points were added for diabetes status in all participants. Vitamin D status was categorised as deficient (< 20 ng/mL), insufficient (20–29 ng/mL), or sufficient (≥ 30 ng/mL) [36]. Hypertension was defined as a documented history of hypertension, current antihypertensive medication use, or admission SBP ≥ 140 mmHg and/or DBP ≥ 90 mmHg.

Statistical analysis

Continuous variables are presented as mean ± SD for normally distributed data or median (IQR) for skewed distributions, assessed by the Shapiro-Wilk test. Categorical variables are expressed as frequencies and percentages. Between-group comparisons employed independent-samples t-tests or Mann-Whitney U tests for continuous variables and Pearson chi-squared or Fisher exact tests for categorical variables. Spearman rank correlation coefficients were computed for non-normally distributed continuous variables. For logistic regression, the dependent variable was ultrasonography-defined MASLD status, coded as 1 for MASLD and 0 for non-MASLD. TyG-BMI was the primary independent variable and was analysed both continuously and by quartiles; Q1 was used as the reference category. Quartile trends were assessed by modelling quartile medians as a continuous variable. Age, diabetes duration, BMI, FBG, HbA1c, lipid parameters, UA, creatinine, eGFR, ALT, AST, HSI, TyG, TyG-BMI, AIP, TG/HDL-C ratio, non-HDL cholesterol, vitamin D, and testosterone were treated as continuous variables unless otherwise specified. Sex, medication use, hypertension status, vitamin D category, and quartiles were treated as categorical variables.

Multivariable logistic regression was performed in three sequential models: Model 1 (unadjusted); Model 2 (adjusted for age, sex, diabetes duration, HbA1c, and baseline antidiabetic medication use); and Model 3 (fully adjusted, additionally incorporating systolic blood pressure, eGFR, uric acid, and baseline lipid-lowering therapy). BMI was additionally included as a covariate in TyG models only. Exploratory LASSO regression was performed using L1-penalized logistic regression with 5-fold cross-validation in the complete-case subset of participants with available candidate variables [37]. Restricted cubic splines were used to evaluate the functional form of the association between TyG-BMI and MASLD [38]. ROC curves were compared using DeLong tests [39]. Categorical NRI was calculated using predefined risk categories of < 30%, 30–60%, and > 60%, and continuous NRI and IDI were also reported [40]. Calibration was evaluated using bootstrap internal validation and calibration plots. Decision curve analysis was used to evaluate clinical net benefit [41]. Exploratory mediation analysis was performed to estimate the indirect pathway from TyG to MASLD through BMI [42]. Analyses were performed using R-compatible statistical routines; software packages and analytic coding are summarised in Additional file 1.

Results

Baseline characteristics

A total of 324 hospitalized T2DM patients (mean age 59.18 ± 9.92 years; 62.0% male) were included in the final analysis. MASLD was identified in 182 patients, yielding a prevalence of 56.2%. Baseline characteristics stratified by MASLD status are presented in Table 1. Compared with MASLD-negative patients, those with MASLD were significantly younger (57.79 ± 10.56 vs. 60.97 ± 8.76 years, p = 0.006) and had shorter diabetes duration [9.00 (4.00–13.00) vs. 10.00 (6.00–16.00) years, p = 0.003]. The MASLD group exhibited markedly higher BMI, FBG, TG, TC, uric acid, ALT, AST, and HSI, and lower HDL-C. TyG index and TyG-BMI were both substantially elevated in the MASLD group. Spearman correlation analysis showed that TyG-BMI was positively correlated with TG, BMI, and ALT and inversely correlated with HDL-C (Fig. 1).

Table 1.

Baseline characteristics of study participants stratified by MASLD status (N = 324)

Variable	Total (N = 324)	MASLD+ (n = 182)	MASLD− (n = 142)	P value
Age (years)	59.18 ± 9.92	57.79 ± 10.56	60.97 ± 8.76	0.006
Male sex, n (%)	201 (62.0%)	112 (61.5%)	89 (62.7%)	0.925
Diabetes duration (years)ᵃ	10.00 (5.00–14.00)	9.00 (4.00–13.00)	10.00 (6.00–16.00)	0.003
BMI (kg/m²)	25.27 ± 3.71	26.69 ± 3.70	23.44 ± 2.83	< 0.001
Systolic BP (mmHg)	132.75 ± 17.50	135.25 ± 16.34	129.54 ± 18.45	0.004
Diastolic BP (mmHg)	79.95 ± 11.06	81.19 ± 11.69	78.37 ± 10.02	0.020
FBG (mmol/L)	7.86 ± 2.47	8.34 ± 2.44	7.25 ± 2.38	< 0.001
HbA1c (%)	8.29 ± 2.00	8.38 ± 1.86	8.17 ± 2.17	0.371
TG (mmol/L)ᵃ	1.56 (1.09–2.40)	2.03 (1.34–3.06)	1.26 (0.93–1.72)	< 0.001
TC (mmol/L)	4.59 ± 1.25	4.76 ± 1.33	4.39 ± 1.11	0.007
HDL-C (mmol/L)	1.08 ± 0.36	1.01 ± 0.27	1.17 ± 0.43	< 0.001
LDL-C (mmol/L)	2.52 ± 1.01	2.47 ± 1.04	2.59 ± 0.97	0.301
Uric acid (µmol/L)	327.1 ± 88.7	341.7 ± 86.1	308.3 ± 88.7	< 0.001
Creatinine (µmol/L)	64.9 ± 16.3	64.9 ± 16.5	64.8 ± 16.2	0.940
eGFR (mL/min/1.73 m²)	84.6 ± 20.2	85.1 ± 19.8	84.0 ± 20.7	0.782
ALT (U/L)ᵃ	24.5 (15.0–35.0)	26.0 (18.2–41.0)	22.0 (12.0–31.8)	< 0.001
AST (U/L)ᵃ	22.0 (16.8–31.0)	25.0 (18.2–34.0)	19.5 (15.0–26.0)	< 0.001
HSIᵃ	36.77 (32.53–42.45)	38.15 (34.21–45.08)	34.85 (30.88–39.86)	< 0.001
25(OH)D (ng/mL)	20.03 ± 12.68	20.25 ± 15.65	19.75 ± 7.41	0.705
Vitamin D deficient, n (%)	195 (60.2%)	112 (61.5%)	83 (58.5%)	0.653
Hypertension status, n (%)	256 (79.0%)	147 (80.8%)	109 (76.8%)	0.458
Lipid-lowering therapy, n (%)	159 (49.1%)	96 (52.7%)	63 (44.4%)	0.166
Antidiabetic medication, n (%)	303 (93.5%)	174 (95.6%)	129 (90.8%)	0.134
TyG index	9.22 ± 0.72	9.52 ± 0.74	8.86 ± 0.58	< 0.001
TyG-BMI	233.0 ± 44.8	254.44 ± 43.18	207.78 ± 29.95	< 0.001
AIP	0.21 ± 0.34	0.34 ± 0.35	0.05 ± 0.24	< 0.001

Discriminative performance of metabolic indices for MASLD

ROC analysis comparing metabolic indices for MASLD discrimination is presented in Table 2; Fig. 2. TyG-BMI achieved the highest AUC of 0.821 (95% CI: 0.775–0.866), with an optimal cut-off of 227.22 yielding sensitivity of 75.1% and specificity of 78.7%. This was significantly higher than the comparator indices by DeLong’s test: TyG (AUC = 0.758, p = 0.011), BMI (AUC = 0.764, p < 0.001), AIP (AUC = 0.746, p = 0.002), TG/HDL-C ratio (AUC = 0.746, p = 0.002), non-HDL cholesterol (AUC = 0.624, p < 0.001), uric acid (AUC = 0.614, p < 0.001), and HSI (AUC = 0.652, p < 0.001). The AUC advantage of TyG-BMI over TyG alone underscores the incremental value of incorporating adiposity into the insulin resistance surrogate. The optimal cut-off for TyG was 9.282 (sensitivity 64.6%, specificity 76.6%).

Table 2.

Discriminative performance of metabolic indices for MASLD discrimination, including HSI

Index	AUC (95% CI)	Optimal cut-off	Sensitivity (%)	Specificity (%)	DeLong p vs. TyG-BMI
TyG-BMI	0.821 (0.775–0.866)	227.22	75.1	78.7	—
TyG	0.758 (0.706–0.810)	9.282	64.6	76.6	0.011
BMI	0.764 (0.712–0.815)	24.85	68.1	72.5	< 0.001
AIP	0.746 (0.693–0.798)	0.346	47.5	90.1	0.002
TG/HDL-C ratio	0.746 (0.693–0.798)	2.216	47.5	90.1	0.002
Non-HDL cholesterol	0.624 (0.565–0.683)	2.91	79.0	41.5	< 0.001
Uric acid	0.614 (0.554–0.674)	324	59.3	63.1	< 0.001
HSI	0.652 (0.591–0.712)	34.94	71.4	51.4	< 0.001

Fig. 2.

Receiver operating characteristic (ROC) curves for seven metabolic indices discriminating MASLD. ROC curves for TyG-BMI, TyG, BMI, atherogenic index of plasma (AIP), TG/HDL-C ratio, non-HDL cholesterol, and uric acid are displayed with corresponding AUC values and 95% confidence intervals. DeLong pairwise comparisons showed that TyG-BMI (AUC = 0.821, 95% CI: 0.775–0.866) had a significantly higher AUC than all displayed comparator indices. The HSI comparison is provided in Table 2 and Additional file 1. The optimal cut-off for TyG-BMI was 227.22 (sensitivity 75.1%, specificity 78.7%) and for TyG was 9.282 (sensitivity 64.6%, specificity 76.6%), both identified by the Youden index. The diagonal dashed line represents the reference line of no discrimination. Abbreviations: AUC, area under the curve; CI, confidence interval; MASLD, metabolic dysfunction-associated steatotic liver disease; TyG-BMI, triglyceride-glucose body mass index; AIP, atherogenic index of plasma; HDL-C, high-density lipoprotein cholesterol; UA, uric acid; HSI, Hepatic Steatosis Index

Logistic regression and quartile analysis

Multivariable logistic regression results are presented in Table 3. In the fully adjusted model (Model 3), each unit increment in TyG-BMI was associated with an OR of 1.0415 (95% CI: 1.0309–1.0522, p < 0.001). Quartile analysis demonstrated a monotonic gradient of MASLD risk across TyG-BMI quartiles (Fig. 3): compared with Q1 (TyG-BMI: 128.4–202.9), the fully adjusted ORs were 2.686 (95% CI: 1.256–5.740, p = 0.011) for Q2, 10.091 (95% CI: 4.610–22.085, p < 0.001) for Q3, and 47.152 (95% CI: 16.438–135.248, p < 0.001) for Q4 (p-trend < 0.001). The corresponding fully adjusted ORs for TyG were 1.977 (95% CI: 0.924–4.232, p = 0.079) for Q2, 4.159 (95% CI: 1.835–9.430, p < 0.001) for Q3, and 11.758 (95% CI: 4.637–29.810, p < 0.001) for Q4. These findings indicate a stronger quartile gradient for TyG-BMI than for TyG alone.

Table 3.

Logistic regression analysis of TyG-BMI and TyG with MASLD risk across three sequential models

Predictor	Model 1 OR (95% CI)	Model 1 P	Model 2 OR (95% CI)	Model 2 P	Model 3 OR (95% CI)	Model 3 P
TyG-BMI (continuous, per unit)	1.0621 (1.0488–1.0756)	< 0.001	1.0498 (1.0381–1.0617)	< 0.001	1.0415 (1.0309–1.0522)	< 0.001
TyG-BMI Q1 (128.4–202.9) [Reference]	1.00 (ref)	—	1.00 (ref)	—	1.00 (ref)	—
TyG-BMI Q2 (202.9–229.3)	3.812 (1.942–7.484)	< 0.001	3.124 (1.572–6.208)	0.001	2.686 (1.256–5.740)	0.011
TyG-BMI Q3 (229.3–260.3)	16.524 (8.012–34.087)	< 0.001	12.841 (6.012–27.421)	< 0.001	10.091 (4.610–22.085)	< 0.001
TyG-BMI Q4 (260.3–430.0)	89.341 (33.214–240.312)	< 0.001	64.218 (22.841–180.512)	< 0.001	47.152 (16.438–135.248)	< 0.001
P for trend	< 0.001	—	< 0.001	—	< 0.001	—
TyG (continuous, per unit)	6.218 (3.842–10.063)	< 0.001	5.412 (3.214–9.108)	< 0.001	4.685 (2.741–8.008)	< 0.001
TyG Q1 (7.39–8.71) [Reference]	1.00 (ref)	—	1.00 (ref)	—	1.00 (ref)	—
TyG Q2 (8.71–9.20)	2.341 (1.142–4.799)	0.020	2.108 (1.012–4.391)	0.046	1.977 (0.924–4.232)	0.079
TyG Q3 (9.20–9.67)	6.012 (2.841–12.722)	< 0.001	5.214 (2.412–11.271)	< 0.001	4.159 (1.835–9.430)	< 0.001
TyG Q4 (9.67–11.67)	18.934 (8.012–44.721)	< 0.001	14.821 (6.012–36.512)	< 0.001	11.758 (4.637–29.810)	< 0.001
P for trend	< 0.001	—	< 0.001	—	< 0.001	—

Fig. 3.

Odds ratios for MASLD across quartiles of TyG-BMI and TyG (fully adjusted Model 3). Bar charts display fully adjusted odds ratios (ORs) with 95% confidence intervals across quartiles of TyG-BMI (Panel A) and TyG (Panel B), with Q1 as the reference category. Model 3 was adjusted for age, sex, diabetes duration, HbA1c, baseline antidiabetic medication use, systolic blood pressure, eGFR, uric acid, and baseline lipid-lowering therapy; BMI was additionally included as a covariate in the TyG model only. Coloured bars indicate statistically significant associations (p < 0.05); grey bars indicate non-significant associations. The earlier quartile separation of TyG-BMI relative to TyG supports its stronger gradient sensitivity. Error bars represent 95% confidence intervals. Abbreviations: OR, odds ratio; CI, confidence interval; NS, not significant; Q, quartile

Dose-response relationship by restricted cubic splines

RCS analysis showed a statistically significant overall association between TyG-BMI and MASLD risk, with no statistically significant evidence of nonlinearity (p-nonlinearity = 0.46; Fig. 4A). The association appeared broadly monotonic across most of the observed range; however, the wider confidence intervals at higher TyG-BMI values suggest sparse observations in the upper tail and should be interpreted cautiously. TyG showed a similar pattern (p-nonlinearity = 0.51; Fig. 4B). These findings support the use of TyG-BMI as a continuous risk stratification metric, while avoiding overinterpretation of a definitive threshold-free relationship.

Fig. 4.

Restricted cubic spline analysis of the dose-response relationship between TyG-BMI, TyG, and MASLD risk. Restricted cubic spline (RCS) curves with four knots at the 5th, 35th, 65th, and 95th percentiles model the association between (A) TyG-BMI and (B) TyG with MASLD risk, expressed as log odds ratios relative to the median. Shaded areas represent 95% confidence intervals. The p-values for nonlinearity were not statistically significant (0.46 for TyG-BMI and 0.51 for TyG). The association appeared broadly monotonic, but wider confidence intervals at higher index values suggest sparse observations in the upper range. Rug plots indicate the distribution of observed values. Abbreviations: RCS, restricted cubic splines; OR, odds ratio; CI, confidence interval; MASLD, metabolic dysfunction-associated steatotic liver disease

Incremental predictive value: NRI, IDI, and LASSO analysis

The incremental predictive value of TyG-BMI over TyG and BMI individually is quantified in Table 4. Adding TyG-BMI to a base model containing TyG alone yielded a categorical NRI of 0.199 (95% CI: 0.063–0.335, p = 0.004), a continuous NRI of 0.435 (95% CI: 0.214–0.657, p < 0.001), and an IDI of 0.096 (95% CI: 0.051–0.141, p < 0.001). Compared with BMI alone, TyG-BMI demonstrated greater reclassification gains (continuous NRI = 0.911, 95% CI: 0.708–1.113, p < 0.001). Exploratory LASSO regression with 5-fold cross-validation was used as a data-driven feature-selection analysis in the complete-case subset (N = 157), retaining TyG, BMI, diabetes duration, C-peptide, non-HDL-C, and testosterone as non-zero predictors (Fig. 5). These findings provided data-driven support for the combined metabolic and adiposity components captured by TyG-BMI, while remaining exploratory because of the complete-case design.

Table 4.

Incremental predictive value of TyG-BMI: net reclassification improvement (NRI) and integrated discrimination improvement (IDI)

Comparison	Metric	Estimate (95% CI)	P value
TyG-BMI vs. TyG	Categorical NRI	0.199 (0.063–0.335)	0.004
TyG-BMI vs. TyG	Continuous NRI	0.435 (0.214–0.657)	< 0.001
TyG-BMI vs. TyG	IDI	0.096 (0.051–0.141)	< 0.001
TyG-BMI vs. BMI	Continuous NRI	0.911 (0.708–1.113)	< 0.001

Note for Table 4: Categorical NRI was calculated using predefined predicted-risk categories of < 30%, 30–60%, and > 60%. Continuous NRI and IDI were reported as threshold-free measures of incremental predictive value

Fig. 5.

Exploratory LASSO regression variable selection for MASLD discrimination. Results of L1-penalized logistic regression with 5-fold cross-validation in the complete-case subset (N = 157). Standardized LASSO coefficients are shown for retained variables at the optimal regularization parameter. TyG and BMI were identified as the strongest retained positive predictors. LASSO was performed as an exploratory complete-case analysis among participants with available values for all candidate variables included in the penalized model

Nomogram development, calibration, and decision curve analysis

An exploratory complete-case nomogram incorporating TyG, BMI, C-peptide, and diabetes duration was constructed for individualized MASLD risk estimation (Fig. 6A). The complete-case nomogram achieved a C-statistic of 0.880, and apparent calibration was supported by the Hosmer-Lemeshow test (p = 0.252) and calibration plot (Fig. 6B). Additional bootstrap internal-validation results are provided in Additional file 1. DCA demonstrated positive net benefit across clinically relevant probability thresholds (Fig. 6C). The DCA curves of the complete-case nomogram and TyG-BMI alone were broadly similar across much of the threshold range, suggesting that TyG-BMI may be sufficient for simple first-line risk stratification, whereas the nomogram may provide additional individualized risk estimation when C-peptide and diabetes-duration data are available.

Fig. 6.

Clinical prediction model for MASLD: nomogram, calibration, and decision curve analysis. (A) Exploratory complete-case nomogram constructed using TyG, BMI, C-peptide, and diabetes duration for individualized MASLD risk estimation. (B) Calibration curve based on bootstrap internal validation comparing observed and predicted MASLD probabilities. (C) Decision curve analysis comparing the complete-case nomogram with TyG-BMI alone. The nomogram and TyG-BMI alone showed broadly similar net benefit across much of the threshold range, suggesting that TyG-BMI may be sufficient for simple first-line risk stratification, whereas the nomogram may provide additional individualized risk estimation when C-peptide and diabetes-duration data are available

Subgroup analysis

The association between TyG-BMI and MASLD was examined across four pre-specified subgroups displayed in Fig. 7: sex, age, BMI category, and diabetes duration. Additional subgroup analyses according to HbA1c category, hypertension status, lipid-lowering therapy use, and vitamin D status are provided in Additional file 1. The positive association was generally consistent across strata, and no statistically significant interactions were observed, supporting the robustness of TyG-BMI across diverse clinical phenotypes within the T2DM inpatient population.

Fig. 7.

Subgroup analysis of the association between TyG-BMI and MASLD. Forest plot displaying fully adjusted odds ratios (ORs) with 95% confidence intervals for the association between TyG-BMI (per unit increment) and MASLD across four pre-specified clinical subgroups shown in the main figure: sex, age (< 60 vs. ≥60 years), BMI category (< 25 vs. ≥25 kg/m²), and diabetes duration (< 10 vs. ≥10 years). Additional subgroup analyses for HbA1c category, hypertension status, lipid-lowering therapy use, and vitamin D status are provided in Additional file 1. No statistically significant interactions were observed

Sex-stratified testosterone analysis

Among the 201 male patients, 198 had available testosterone measurements for the testosterone-specific analyses. Testosterone levels were significantly lower in MASLD-positive compared with non-MASLD men (14.36 ± 5.82 vs. 17.48 ± 6.14 nmol/L, p < 0.001; Fig. 8A). Testosterone was inversely correlated with TyG-BMI (ρ = −0.260, p = 0.0002; Fig. 8B). Quartile analysis revealed a stepwise increase in MASLD risk with decreasing testosterone: the unadjusted OR for MASLD in Q1 (lowest testosterone) versus Q4 was 6.23 (95% CI: 2.84–13.67, p < 0.001; Fig. 8C). RCS analysis suggested a nonlinear association between testosterone and MASLD (overall p < 0.001; p-nonlinearity = 0.038). After adjustment for TyG-BMI and covariates, the testosterone-MASLD association was attenuated, supporting the hypothesis that insulin resistance and adiposity partly explain the hormonal association.

Fig. 8.

Sex-stratified exploratory testosterone analysis in male T2DM patients. The overall male subgroup comprised 201 participants; analyses requiring available testosterone measurements were performed among 198 men. (A) Violin plot comparing testosterone levels between MASLD-positive and non-MASLD males. (B) Adjusted and unadjusted odds ratios for MASLD across testosterone quartiles. (C) Restricted cubic spline analysis of testosterone and MASLD risk. (D) Scatter plot showing the correlation between testosterone and TyG-BMI, colored by MASLD status. (E) Discrimination after adding testosterone to the TyG-BMI model in males. (F) Testosterone distribution across TyG-BMI quartiles

Mediation analysis

Exploratory mediation analysis suggested that BMI mediated approximately 11.5% (95% CI: 4.2–19.8%) of the total association between TyG and MASLD through the indirect pathway (Additional file 1: Figure S2). The direct association between TyG and MASLD remained statistically significant after accounting for BMI mediation (OR = 4.685, 95% CI: 2.741–8.008, p < 0.001). Because of the cross-sectional design, this analysis should be interpreted as hypothesis-generating rather than causal mediation evidence.

Sensitivity analysis

After excluding 21 patients with extreme metabolic values (BMI > 35 kg/m² or TG > 5.6 mmol/L), yielding an analytic sample of 303 patients, TyG-BMI maintained strong discriminative performance (AUC = 0.805, 95% CI: 0.756–0.854). The Q4 OR remained substantial (31.05, 95% CI: 10.82–89.12) and DeLong comparison with TyG remained statistically significant (p = 0.005), indicating that the primary findings were not driven by outliers or extreme metabolic phenotypes.

Discussion

This hospital-based cross-sectional study demonstrates that TyG-BMI showed better non-invasive discriminative performance for MASLD than several routinely available metabolic and lipid-related indices in hospitalized T2DM patients, achieving an AUC of 0.821. The quartile gradient, broadly monotonic dose-response pattern, consistent subgroup findings, and significant reclassification improvement support the potential utility of TyG-BMI as a simple first-line risk stratification marker in the diabetic inpatient setting.

The better discriminative performance of TyG-BMI over TyG alone is biologically plausible. While TyG captures metabolic consequences of insulin resistance through triglyceride and glucose components, it does not directly account for adiposity, an independent driver of hepatic steatosis through increased free fatty acid flux, adipokine dysregulation, and visceral fat-derived inflammatory mediators [43, 44]. By incorporating BMI, TyG-BMI integrates both pathophysiological axes into a single metric. This is supported by LASSO analysis, which identified TyG and BMI as the two strongest data-driven predictors, and by exploratory mediation analysis suggesting that BMI may partly account for the TyG-MASLD association. The quartile gradient further underscores the added stratification value of TyG-BMI; however, the very high OR in the highest quartile should be interpreted cautiously because the number of non-MASLD participants in the upper TyG-BMI range was small.

Our findings are consistent with and extend prior studies in general populations. Wang et al. (2023) reported an AUC of 0.793 for TyG-BMI in a Chinese health examination cohort, while Hu et al. (2022) demonstrated an AUC of 0.812 in a Japanese general population [25, 26]. Li et al. (2022) specifically examined T2DM patients and reported an AUC of 0.801 for TyG-BMI, closely aligned with our finding of 0.821 [27]. Our study extends these findings by providing a systematic DeLong comparison of TyG-BMI against AIP, TG/HDL-C ratio, non-HDL cholesterol, uric acid, and HSI in a T2DM-specific hospitalized cohort.

The AIP, while a validated lipid-related steatosis predictor reflecting atherogenic dyslipidemia [45, 46], showed lower discrimination than TyG-BMI. We additionally calculated HSI, a liver-enzyme-based steatosis score, and included it in the ROC comparison. TyG-BMI also showed higher discrimination than HSI in this cohort (AUC 0.821 vs. 0.652). This finding suggests that the integrated glycemic-adiposity information captured by TyG-BMI may provide stronger risk stratification than a liver-enzyme-based score in hospitalized T2DM patients, although HSI remains a useful comparator in broader steatosis screening contexts. More broadly, MASLD is linked with diabetes-related metabolic profiles, steatohepatitis-related histological features, gut microbiota dysbiosis, atherogenic dyslipidemia, incident diabetes, cardiovascular outcomes, and systemic metabolic consequences [47–55].

The sex-stratified testosterone analysis suggests a potentially important hormonal dimension of MASLD pathophysiology in male T2DM patients, but these findings should be interpreted as exploratory. The relationship between testosterone, adiposity, and insulin resistance is bidirectional and biologically complex. Functional hypogonadism is clinically intertwined with obesity, insulin resistance, and T2DM, and observational studies have examined diabetes-related outcomes in hypogonadal men receiving testosterone therapy [56–59]. Obesity and metabolic syndrome are associated with reduced hepatic SHBG production, which may theoretically increase the free testosterone fraction, but this potential effect can be counterbalanced by obesity-related hypothalamic-pituitary-gonadal axis dysfunction, reduced gonadotropin secretion, impaired testicular testosterone production, and increased aromatization of androgens to estrogens in adipose tissue. Consequently, total testosterone decreases, and bioavailable testosterone may remain low or further decline despite reduced SHBG levels. Because SHBG, free testosterone, gonadotropins, and longitudinal hormone measurements were unavailable, the testosterone findings should be regarded as hypothesis-generating rather than evidence for a therapeutic effect.

The exploratory complete-case nomogram incorporating TyG, BMI, C-peptide, and diabetes duration achieved good apparent discrimination and calibration, but its clinical role should be interpreted pragmatically. DCA indicated that both the complete-case nomogram and TyG-BMI alone provided positive net benefit across relevant thresholds, with broadly similar curves across much of the range. Thus, TyG-BMI alone may be sufficient for simple first-line risk stratification, whereas the nomogram may be useful for individualized risk estimation when C-peptide and diabetes-duration data are available. Because the nomogram was constructed in a complete-case subset, external validation is required before clinical implementation.

The RCS analysis did not show statistically significant evidence of nonlinearity. The spline results should be interpreted as showing a broadly monotonic association across the observed range rather than as defining a strict biological threshold. The widening confidence intervals at higher TyG-BMI values likely reflect sparse observations in the upper tail, and this uncertainty was explicitly considered in the revised interpretation. The Youden-derived cut-off of 227.22 may be clinically practical, but risk should be understood as graded rather than strictly dichotomous.

Several limitations of this study warrant acknowledgment. First, the hospital-based cross-sectional design precludes causal inference; longitudinal studies are needed to establish whether TyG-BMI predicts incident MASLD or its progression. Second, MASLD diagnosis relied on ultrasonography rather than liver biopsy or magnetic resonance-based fat quantification, potentially underestimating mild steatosis; however, ultrasonography remains a widely used first-line modality in clinical practice [17]. Third, the single-centre design may limit generalizability to other ethnic or geographic populations. Fourth, unmeasured confounders including dietary patterns, physical activity, and genetic risk variants could not be accounted for. Fifth, although bootstrap internal validation was added, external validation in independent cohorts is still required before clinical implementation. Sixth, the testosterone analysis was restricted to males and lacked SHBG and free testosterone measurements. Notwithstanding these limitations, the comprehensive analytical framework—encompassing ROC comparison including HSI, RCS, NRI/IDI, LASSO, nomogram, DCA, subgroup analysis, exploratory mediation, and sensitivity analysis—provides multidimensional evidence for the risk-stratification utility of TyG-BMI in this population.

Conclusions

TyG-BMI is a non-invasive and readily calculable marker that showed better discriminative performance for MASLD in hospitalized patients with T2DM than several comparator metabolic indices, including TyG, BMI, AIP, TG/HDL-C ratio, non-HDL cholesterol, uric acid, and HSI. Its quartile gradient, reclassification improvement, and broadly monotonic association with MASLD support its potential use as a first-line inpatient risk stratification marker rather than as a stand-alone diagnostic replacement for imaging. The sex-stratified analysis further suggests that low testosterone is linked to the metabolic phenotype captured by TyG-BMI in male T2DM patients, although this finding requires prospective validation. Future multicentre longitudinal studies using more precise liver-fat assessment are needed to validate TyG-BMI as a dynamic biomarker for MASLD risk stratification and to evaluate whether TyG-BMI-guided interventions improve liver-related outcomes in T2DM populations.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Y.-C.Y. Data curation: Q.-j.H., Y.-y.X., X.T. Formal analysis: Q.-j.H., Y.-y.X. Funding acquisition: Y.-C.Y. Investigation: Q.-j.H., Y.-y.X., X.T. Methodology: Q.-j.H., Y.-y.X., Y.-C.Y. Project administration: Y.-C.Y. Resources: Y.-C.Y. Software: Q.-j.H., Y.-y.X. Supervision: Y.-C.Y. Validation: X.T., Y.-C.Y. Visualization: Q.-j.H., Y.-y.X. Writing – original draft: Q.-j.H., Y.-y.X. Writing – review & editing: X.T., Y.-C.Y. All authors read and approved the final manuscript.

Funding

This work was supported by the Hefei Municipal Health Commission (Grant No. Hwk2024zd008) and the Anhui Province Traditional Chinese Medicine Research Program (Grant No. 2024CCCX136). The funders had no role in study design, data collection, analysis, interpretation, or the decision to submit for publication.

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Ethical approval for this study was granted by the Ethics Committee of the Third People’s Hospital of Hefei (Approval No. 2026LLWL024). All procedures were conducted in accordance with the ethical principles outlined in the Declaration of Helsinki. Written informed consent was obtained from all participants prior to enrollment. Analyses were performed using anonymized clinical data collected from the study hospital.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.