Assessment indicators of steatotic liver disease in non-hepatology specialty clinics: A real-world retrospective case-control study

Abstract

BACKGROUND

The escalating global burden of steatotic liver disease (SLD) underscores the critical need for practical, non-invasive screening strategies, particularly in non-specialized clinical settings where advanced imaging may be inaccessible.

AIM

To evaluate the predictive value of anthropometric and metabolic indices for SLD risk stratification and fat quantification in Chinese adults.

METHODS

This retrospective case-control study enrolled 517 adults from nutrition clinics at a tertiary hospital in China over a six-month period. Participants underwent body composition analysis, routine blood tests, and liver imaging (ultrasound, computed tomography, or FibroScan) to assess steatotic changes. Univariate and multivariate logistic regression analyses were utilized to identify independent predictors of SLD. Predictive performance was evaluated using the area under the receiver operating characteristic curve (AUC).

RESULTS

Univariate analysis revealed significant associations between SLD and gender, diabetes status, and seven anthropometric-metabolic indices, including the triglyceride glucose-waist-to-height ratio (TyG-WHtR) and Chinese visceral adiposity index (CVAI) (all P < 0.05). Multivariate logistic regression identified three independent predictors for SLD: male sex [odds ratio (OR) = 4.717, 95%CI: 2.114-10.523, P < 0.001]; the per 1-SD OR of TyG-WHtR = 6.689 (95%CI: 3.227-13.864, P < 0.001), and the per 1-SD OR of CVAI = 3.406 (95%CI: 2.289-5.06, P < 0.001). TyG-WHtR demonstrated robust discriminative capacity (AUC = 0.876; 95%CI: 0.846-0.905), outperforming CVAI (AUC = 0.865; 95%CI: 0.834-0.896). Furthermore, parameter distributions differed significantly across non-SLD, mild SLD, and moderate-to-severe SLD groups (P < 0.0001).

CONCLUSION

TyG-WHtR and CVAI are robust, cost-effective biomarkers for predicting SLD risk and severity in Chinese adults. They optimize primary care screening but require independent prospective cohort validation.

Key Words: Steatotic liver disease; Steatotic degree; Indicators; Non-hepatology specialty clinics; Health check-up populations

Core Tip: The escalating burden of steatotic liver disease (SLD) necessitates accessible screening tools for non-specialized settings. This study demonstrates that the triglyceride glucose-waist-to-height ratio (TyG-WHtR) and the Chinese visceral adiposity index are robust, cost-effective biomarkers for predicting SLD risk and disease severity in Chinese adults. Notably, TyG-WHtR exhibited superior discriminative capacity (area under the curve = 0.876). Implementing these non-invasive indices in primary care can significantly optimize risk stratification, reduce reliance on expensive advanced imaging, and facilitate early therapeutic interventions to mitigate SLD progression.

INTRODUCTION

The prevalence of steatotic liver disease (SLD) worldwide has risen from 25%[1] to over 30%[2,3], and continues to increase[4]. In China, this trend is even more pronounced. Previously, the general prevalence of metabolic dysfunction-associated steatotic liver disease (MASLD) was recorded at 29.88%[5]. However, recent national data reveal that steatosis, severe steatosis, advanced fibrosis, and cirrhosis now affect 44.39%, 10.57%, 2.85%, and 0.87% of Chinese adults, respectively[6]. This surging prevalence anticipates a significant future burden of cirrhosis, hepatocellular carcinoma, and other metabolic diseases.

Many individuals with hepatic steatosis remain unaware of their condition prior to health examinations, often because their body mass index (BMI) does not fall into the overweight category or their routine liver enzymes appear normal. However, non-obese populations also exhibit significant steatosis. For instance, the global prevalence of lean MASLD reaches 40%[7], with significant differences observed in regions and ethnic groups[4,7]. The latest guidelines consensus emphasizes that these incidental findings of steatosis should prompt further assessments of the etiology and extent of SLD and fibrosis[8]. Therefore, a critical clinical need exists in non-hepatology clinics (e.g., nutrition and general practice) to screen patients requiring specialist evaluation and lifestyle or pharmacological interventions. Achieving this via simple, accessible tests is essential for broad implementation across various economic regions, particularly for detecting moderate-to-severe steatosis.

In addition to individual indicators, composite indices derived from simple calculations can be applied in non-hepatology clinics and mobile assessments. The triglyceride glucose (TyG) index, combining fasting plasma glucose (FPG) and triglycerides (TG), effectively diagnoses insulin resistance (IR)[9]. As one of the parameters related to TyG, TyG-waist-to-height ratio (TyG-WHtR) combines waist circumference (WC) and height, making it one of the convenient and alternative indicators of IR in various metabolic diseases[10]. The Visceral Adiposity Index (VAI) was primarily established and validated in Caucasian populations[11], while Chinese VAI (CVAI) developed using Chinese populations is more relevant to metabolic health outcomes in Asian populations[12]. The conicity index (COI) index is a biomarker for centripetal obesity[13]. And the hepatic steatosis index (HSI) and ZJU indices are also good predictors of hepatic steatosis[14,15].

In this study, we aim to explore and compare the feasibility of screening for liver steatosis and its degree through simple laboratory parameters [alanine transaminase (ALT), aspartate aminotransferase (AST), TG, FPG, high density lipoprotein cholesterol (HDL-C), etc.] and physical examination data [neck circumference (NC), thigh circumference, WHtR, ALM%, etc.], as well as their combinations (TyG-WHtR, CVAI, etc.) in non-hepatology specialty clinics. These findings may further guide daily self-assessment strategies for health check-up populations.

MATERIALS AND METHODS

Study design

This retrospective case-control study employed the entire population who received nutritional consultation and intervention, as well as body composition analysis at the Nutrition Department clinic of the Second Affiliated Hospital of Zhejiang University School of Medicine from July 1, 2024 to December 31, 2024. The study was approved by the Ethics Committee of the Second Affiliated Hospital of Zhejiang University School of Medicine (No. 20231215). The data is reliable and covers the entire population who underwent human composition analysis in the outpatient clinic of the nutritional department of a top-tier tertiary hospital over a six-month period, and is representative in terms of demographic characteristics. Information on 3794 participants was obtained from the outpatient system, and all patients underwent body composition analysis using the InBody 720 (Inbody Inc., Korea). We first selected a group of individuals with complete simple laboratory indicators such as ALT, AST, PLT, Hb, TG, FPG, and HDL-C, as well as liver imaging examinations [including any at least one type of examination such as B-mode ultrasound, computed tomography (CT), and FibroScan conducted in our hospital or external hospitals], with a total of 554 individuals. Then, we excluded the following participants: (1) Individuals under 18 years old; (2) Individuals with positive viral hepatitis B or C; (3) Other clearly identified causes of hepatic inflammatory diseases such as autoimmune hepatitis; (4) Patients with liver cancer and other cancers; (5) Alcoholics (with a mean daily alcohol consumption exceeding 50 g for women or 60 g for men); (6) Individuals taking fat-generating drugs for at least 6 months (e.g., prednisone); (7) Pregnant or within 6 months postpartum; and (8) With incorrect data. Finally, 517 participants were included in this study (male = 228; female = 289) (Figure 1).

Figure 1  Flow chart of participants inclusion and exclusion in the survey.

Clinical data and laboratory tests

All variables are from the original database, including gender, age, height, weight, BMI, laboratory indicators, visceral fat area, body fat, WC, NC, and more, rounded to 1-2 decimal places. Basic disease information such as hypertension, diabetes, hyperuricemia, thyroid disease, other liver diseases, and polycystic ovarian syndrome is also collected. Parameter definitions or calculations are as follows[14-18]:

BMI = weight (kg)/height² (m);

WHtR = WC (cm)/height (cm);

TyG = Ln^{[TG (mg/dL) × FPG (mg/dL)/2]};

TyG-WHtR = TyG × WHtR;

CVAI (men) = -267.93 + 0.68 × age + 0.03 × BMI + 4.00 × WC + 22.00 × Log₁₀TG - 16.32 × HDL-C;

CVAI (women) = -187.32 + 1.71 × age + 4.23 × BMI + 1.12 × WC + 39.76 × Log₁₀TG - 11.66 × HDL-C;

COI = 0.109^-1WC (m)[weight (kg)/height (m)]^-1/2;

ALM%: The sum of the muscle mass percentages of right upper limb, right lower limb, left upper limb, and left lower limb, muscle mass percentage = muscle mass/ideal muscle mass × 100;

HSI = 8 × ALT/AST + BMI (+ 2, if female; + 2, if diabetic);

ZJU = BMI (kg/m²) + FPG (mmol/L) + TG (mmol/L) + 3 × ALT (U/L)/AST (U/L) (+ 2, if female).

The definition of SLD

Firstly, the diagnostic criteria for ultrasound are as follows[19,20]: (1) Diffuse enhancement of near field echo in the hepatic region (stronger than in the kidney and spleen region) and gradual attenuation of the far field echo; (2) Unclear display of intra-hepatic lacuna structure; (3) Mild to moderate hepatomegaly with a round and blunt border; (4) Color Doppler ultrasonography shows a reduction of the blood flow signal in the liver or it is even hard to display, but the distribution of blood flow is normal; and (5) Unclear or non-intact display of envelop of right liver lobe and diaphragm. Mild degree of fatty liver displays item 1 and any one of items 2-4; moderate fatty liver displays item 1 and any two items of items 2-4; severe fatty liver displays items 1 and 5 and any two of items 2-4.

The CT image of SLD shows a diffuse decrease in liver density, with a CT value ratio of liver to spleen being less than 1. Among them, a mild fatty liver has a CT ratio of liver/spleen less than 1.0 but greater than 0.7, a moderate fatty liver has a ratio of less than or equal to 0.7 but greater than 0.5, and a severe fatty liver has a ratio of 0.5 or less[21].

The continuous variable of the controlled attenuation parameter (CAP) based on TE may also be utilized to monitor changes in liver fat content. For chronic liver disease patients, optimal cutoff values for CAP measured by the M probe of FibroScan^® to identify significant (≥ S1), moderate-to-severe (≥ S2), and severe hepatic steatosis (S3) are 248 dB/m, 268 dB/m, and 294 dB/m respectively[22].

Statistical analysis

Data were organized using Microsoft Excel and analyzed utilizing SPSS version 22.0 (IBM Corp., Armonk, NY, United States). To ensure methodological clarity, the statistical procedures were conducted and reported in distinct phases.

Descriptive and univariate analyses: Continuous variables were expressed as mean ± SD if normally distributed, or as median (interquartile range) if non-normally distributed. Categorical variables were described as n (%). Differences in baseline characteristics between the SLD and non-SLD groups were compared using the independent sample t-test, χ² test, or Mann-Whitney U test, as appropriate. Non-parametric tests (Kruskal-Wallis H test or Mann-Whitney U test) were utilized to compare parameter distributions across the non-SLD, mild SLD, and moderate-to-severe SLD groups.

Assumption checking: Prior to multivariable modeling, strict assumption diagnostics were performed. The Box-Tidwell method was applied to verify the linear relationship between continuous independent variables and the logit-transformed dependent variable (presence of SLD), with a Bonferroni-corrected significance threshold of P < 0.0025. Multicollinearity among candidate variables was assessed using Tolerance and the variance inflation factor (VIF); variables with a Tolerance > 0.1 and a VIF < 10 were considered to have no significant collinearity.

Multivariable model building and internal validation: Based on a priori clinical relevance and univariate screening (P < 0.05), nine candidate variables (gender, diabetes status, NC, TyG-WHtR, CVAI, HSI, thigh circumference, waist-to-hip ratio, and ALM%) were pre-specified for the binary logistic regression model. BMI, TyG, and WHtR were excluded to prevent information overlap, as they are component variables of CVAI and TyG-WHtR. To ensure the final model was driven by clinical knowledge rather than algorithmic data-driven selection, the "Enter" method was utilized to force all nine variables into the model simultaneously. Gender was coded with female as the reference category (male = 1, female = 0). To facilitate clinical interpretation, the continuous variables TyG-WHtR and CVAI were standardized into Z-scores [Z = (value - mean)/SD] prior to modeling. To assess the risk of overfitting, the events-per-variable (EPV) ratio was calculated. Internal validation was performed using bootstrapping with 1000 replications to yield optimism-corrected coefficients, bootstrap bias, and robust 95% confidence intervals (CIs). Model calibration was evaluated using the Hosmer-Lemeshow goodness-of-fit test.

Predictive performance and clinical utility: Receiver operating characteristic (ROC) curve analysis was performed using MedCalc software (version 23.5.6-64-bit; Ostend, Belgium) to evaluate the discriminative capacity of TyG-WHtR, CVAI, WC, and BMI for SLD. The area under the receiver operating characteristic curve (AUC), SE, and 95%CIs were calculated. Optimal cutoffs were determined by maximizing the Youden index, with the corresponding sensitivity, specificity, positive likelihood ratio (+LR), and negative likelihood ratio (-LR) reported. DeLong's test was employed for pairwise comparisons of AUCs to quantify the incremental value of TyG-WHtR over CVAI, WC, and BMI.

To translate discriminative accuracy into clinically actionable metrics, we calculated the positive predictive value (PPV), negative predictive value (NPV), number needed to screen (NNS), false-positive rate, and false-negative rate for TyG-WHtR and CVAI at their optimal cutoffs under three plausible SLD prevalence scenarios: 20% (general community), 30% (health check-up population, consistent with recent Chinese nationwide data[6]), and 50% (high-risk specialty clinics). PPV and NPV were computed using standard Bayesian formulae: PPV = (prevalence × sensitivity)/[(prevalence × sensitivity) + (1 - prevalence) × (1 - specificity)] and NPV = [(1 - prevalence) × specificity]/{[(1 - prevalence) × specificity] + [prevalence × (1 - sensitivity)]}. NNS, defined as the number of individuals that must be screened to detect one true SLD case, was calculated as NNS = 1/(prevalence × sensitivity). The false-positive rate was defined as 1 - specificity, and the false-negative rate as 1 - sensitivity. Statistical significance was set at a two-sided P < 0.05.

RESULTS

One-way analysis of the association between the presence of SLD and parameters

This study mainly conducted a univariate analysis of categorical variables (Table 1) and continuous variables (Supplementary Table 1) based on basic characteristics (gender, age, hypertension, diabetes, etc.), simple physical measurements (such as NC, WC, hip circumference directly obtained from body composition analysis), and several simple/complex calculated indicators (ALM%, TyG-WHtR, CVAI, COI, etc.). The results showed that there are differences in gender and diabetes prevalence between SLD patients and normal control populations, while there is no significant difference in the prevalence of age and hypertension. BMI, TyG, and WHtR were excluded due to their inclusion in the formula variables of CVAI and TyG-WHtR. The Box-Tidwell method was used to test whether there is a linear relationship between continuous independent variables and the Logit-transformed dependent variable. A total of 20 variables were included in the linearity test model, with a significant level of 0.0025 after Bonferroni correction. The results of the linearity test showed that there is a linear relationship between all continuous independent variables and the Logit-transformed dependent variable. Tolerance (tolerance) or VIF were used to diagnose multicollinearity among independent variables. By gradually eliminating variables, we obtained variables with Tolerance > 0.1 and VIF less than 10, which do not have multicollinearity (Table 2). Based on this, the continuous variable table of univariate analysis was adjusted to Table 3. The results showed that these 7 continuous variables are all related to the presence of SLD (P < 0.0001).

Table 1 Univariate analysis of the association between categorical variables and steatotic liver disease.

Categorical variables	Case group (n = 273)		Control group (n = 244)		χ2	P value
	Number	Composition ratio (%)	Number	Composition ratio (%)
Gender					46.923	< 0.0001
Male	159	58.24	69	28.28
Female	114	41.76	175	71.72
Diabetes					5.480	0.019
Yes	27	9.89	11	4.51
No	246	90.11	233	95.49

Table 2 Variables with no collinearity.

	Tolerance	VIF
Gender	0.426	2.35
Diabetes	0.97	1.031
NC	0.14	7.167
TyG-WHtR1	0.145	6.897
HSI	0.245	4.085
CVAI	0.571	1.75
Thigh circumference	0.206	4.85
ALM%	0.219	4.559
Waist-to-hip ratio	0.197	5.088
	Tolerance > 0.1	VIF < 10

¹By gradually eliminating the variables, the Tolerance > 0.1 and variance inflation factor < 10 was obtained, and there was no collinearity variable.

VIF: Variance inflation factor; TyG-WHtR: Triglyceride glucose-waist-to-height ratio; HSI: Hepatic steatosis index; CVAI: Chinese Visceral Adiposity Index; ALM%: The sum of the muscle mass percentages of right upper limb, right lower limb, left upper limb, and left lower limb, muscle mass percentage = muscle mass/ideal muscle mass × 100; NC: Neck circumference.

Table 3 After removing collinearity, one-way analysis of the association between continuous variables and steatotic liver disease.

Continuous variables	SLD (n = 273)	Non-SLD (n = 244)	P value	T/U
NC, means (SE)	40.03 (3.63)	34.66 (4.27)	< 0.0001	15.445
Waist-to-hip ratio, means (SE)	0.98 (0.06)	0.89 (0.07)	< 0.0001	15.013
TyG-WHtR1, median (IQR)	5.57 (5.09-6.01)	4.34 (3.79-4.95)	< 0.0001	8275.000
HSI, means (SE)	44.11 (7.08)	33.94 (7.66)	< 0.0001	15.497
CVAI1, median (IQR)	145.31 (107.11-191.65)	63.45 (33.44-98.05)	< 0.0001	8993.000
Thigh circumference1, median (IQR)	59.70 (56.03-63.50)	52.28 (48.10-56.98)	< 0.0001	12359.500
ALM%, means (SE)	463.93 (56.28)	405.13 (56.20)	< 0.0001	11.866

¹Because it does not conform to the normal distribution, the Mann-Whitney test is used.

TyG-WHtR: Triglyceride glucose-waist-to-height ratio; SLD: Steatotic liver disease; IQR: Interquartile range; HSI: Hepatic steatosis index; CVAI: Chinese Visceral Adiposity Index; ALM%: The sum of the muscle mass percentages of right upper limb, right lower limb, left upper limb, and left lower limb, muscle mass percentage = muscle mass/ideal muscle mass × 100; NC: Neck circumference.

In addition, ultrasound was the main imaging method in both the non-SLD group and the SLD group (89.8% and 89.0%, respectively), and the overall distribution of imaging methods did not differ significantly between the groups (χ² = 3.320, P = 0.190).

Logistic regression analysis predicts the presence of SLD

A multivariable logistic regression model was constructed using the "Enter" method, incorporating nine pre-specified candidate variables (gender, diabetes status, NC, TyG-WHtR, CVAI, HSI, thigh circumference, waist-to-hip ratio, and ALM%). The model demonstrated excellent fit (χ² = 337.624, P < 0.0001; Nagelkerke R² = 0.651).

Three independent predictors for SLD were identified (Table 4): Male sex [adjusted odds ratio (OR) = 4.717, P < 0.001], TyG-WHtR (OR = 7.536 per 1-unit increase, P < 0.001), and CVAI (OR = 1.018 per 1-unit increase, P < 0.001). ALM% showed nominal significance with a negligible effect size (OR = 1.010, P = 0.036). Conversely, diabetes status, HSI, and thigh circumference lost predictive value after full adjustment. Notably, after adjusting for visceral adiposity indices, NC became negatively correlated with SLD (OR = 0.824, P = 0.013), suggesting its initial univariate association was largely mediated by central obesity.

Table 4 The results of logistic regression analysis showed three parameters that were significantly associated with steatotic liver disease.

	B	SE	Wald	df	P value	Exp (B)	95% EXP (B)
Gender	1.551	0.409	14.354	1	< 0.0001	4.717	2.114	10.523
Diabetes	0.451	0.550	0.675	1	0.411	1.571	0.535	4.612
NC	-0.193	0.077	6.212	1	0.013	0.824	0.708	0.960
TyG-WHtR	2.020	0.395	26.121	1	< 0.0001	7.536	3.473	16.350
CVAI	0.018	0.003	36.531	1	< 0.0001	1.018	1.012	1.024
HSI	0.041	0.029	2.005	1	0.157	1.042	0.984	1.103
Thigh circumference	-0.060	0.043	2.012	1	0.156	0.941	0.866	1.023
Waist-to-hip ratio	1.829	3.820	0.229	1	0.632	6.225	0.003	11108.643
ALM%	0.010	0.005	4.377	1	0.036	1.010	1.001	1.019
Constant	-9.726	2.563	14.396	1	< 0.0001	0.000

TyG-WHtR: Triglyceride glucose-waist-to-height ratio; HSI: Hepatic steatosis index; CVAI: Chinese Visceral Adiposity Index; ALM%: The sum of the muscle mass percentages of right upper limb, right lower limb, left upper limb, and left lower limb, muscle mass percentage = muscle mass/ideal muscle mass × 100; NC: Neck circumference.

Furthermore, the waist-to-hip ratio exhibited a highly unstable CI (OR = 6.225, 95%CI: 0.003-11108.643, P = 0.632), reflecting severe multicollinearity with the component variables of TyG-WHtR and CVAI, thus providing no independent prognostic value.

For standardized clinical comparison, continuous variables were Z-score transformed. Each 1-SD increase in TyG-WHtR and CVAI amplified the SLD risk by 6.689-fold (95%CI: 3.227-13.864) and 3.406-fold (95%CI: 2.289-5.068), respectively (both P < 0.001). Additionally, categorization into quartiles/quantiles revealed a significant stepwise dose-response trend for both indices (Tables 5 and 6).

Table 5 Quartile-specific prevalence of steatotic liver disease according to triglyceride glucose-waist-to-height ratio levels.

Quartile	TyG-WHtR range	n	SLD cases	SLD prevalence (%)	Unadjusted OR (95%CI)
Q1 (lowest)	≤ P25	129	7	5.4	1.00 (Reference)
Q2	P25-P50	129	57	44.2	13.80 (5.97-31.88)
Q3	P50-P75	130	94	72.3	45.51 (19.39-106.80)
Q4 (highest)	> P75	129	115	89.1	143.16 (55.80-366.90)
P for trend					< 0.0001

TyG-WHtR: Triglyceride glucose-waist-to-height ratio; SLD: Steatotic liver disease; OR: Odds ratio; CI: Confidence interval.

Table 6 Quartile-specific prevalence of steatotic liver disease according to Chinese Visceral Adiposity Index levels.

Quartile	CVAI range	n	SLD cases	SLD prevalence (%)	Unadjusted OR (95%CI)
Q1 (lowest)	≤ P25	129	7	5.4	1.00 (Reference)
Q2	P25-P50	129	57	44.2	13.80 (5.97-31.88)
Q3	P50-P75	130	92	70.8	42.20 (18.00-98.70)
Q4 (highest)	> P75	129	117	90.7	169.93 (64.70-446.60)
P for trend					< 0.0001

SLD: Steatotic liver disease; OR: Odds ratio; CVAI: Chinese Visceral Adiposity Index; CI: Confidence interval.

Internal validation via 1000-replicate bootstrapping confirmed the stability of these estimates, yielding negligible bias (e.g., TyG-WHtR bias = 0.067; CVAI bias = 0.001) and the bootstrap-derived P values for both core indices remained < 0.001. The model showed acceptable calibration (Hosmer-Lemeshow P = 0.035). Finally, an EPV ratio of 30.3 (273 events/9 variables) quantitatively confirmed a low risk of overfitting.

The predictive value of two parameters that predict the existence of SLD

To evaluate the discriminative capacity for predicting the presence of SLD, ROC curve analysis was performed. TyG-WHtR and CVAI demonstrated excellent predictive performance, yielding AUCs of 0.876 and 0.865, respectively (Table 7, Figure 2). Based on the maximal Youden index, the optimal cutoff value for TyG-WHtR was 4.787, achieving a high sensitivity of 91.2% and a specificity of 68.9%. For CVAI, the optimal cutoff was 101.9, with a sensitivity of 80.5% and a specificity of 78.7%. Detailed screening metrics under varying background prevalence assumptions are provided in Supplementary Table 2.

Figure 2 Receiver operating characteristic curves for identifying steatotic liver disease. A: Receiver operating characteristic (ROC) curve for identifying steatotic liver disease (SLD) using waist-to-height ratio; B: ROC curves for identifying SLD using Chinese Visceral Adiposity Index. AUC: Area under the receiver operating characteristic curve; CVAI: Chinese Visceral Adiposity Index; TyG-WHtR: Triglyceride glucose-waist-to-height ratio.

Table 7 The predictive power of two associated parameters.

Variable	AUC (95%CI)	Cut-off values	Specificity (%)	Sensitivity (%)	PPV	NPV	P value
TyG-WHtR	0.876 (0.846-0.905)	4.787	68.852	91.209	76.615	87.5	< 0.0001
CVAI	0.865 (0.834-0.896)	101.9	78.689	80.524	76.8	80.48	< 0.0001

TyG-WHtR: Triglyceride glucose-waist-to-height ratio; CVAI: Chinese Visceral Adiposity Index; CI: Confidence interval; PPV: Positive predictive value; NPV: Negative predictive value; AUC: Area under the receiver operating characteristic curve.

To rigorously validate the incremental discriminative value of these composite indices over conventional anthropometric measures, pairwise AUC comparisons were conducted using DeLong’s test (Supplementary Figure 1). Crucially, the AUC of TyG-WHtR (0.876) was statistically superior to both standalone WC (AUC = 0.852; ΔAUC = 0.024, 95%CI: 0.0047-0.0433, P = 0.0146) and BMI (AUC = 0.835; ΔAUC = 0.041, 95%CI: 0.0213-0.0609, P < 0.0001). Although the difference in AUC between TyG-WHtR and CVAI did not reach statistical significance (ΔAUC = 0.011, 95%CI: -0.020 to 0.042, P = 0.4980), TyG-WHtR offers superior clinical utility due to its computational simplicity. While CVAI requires five parameters (age, BMI, WC, TG, and HDL-C), TyG-WHtR relies on a more accessible combination of routine metabolic and basic anthropometric parameters. Furthermore, CVAI showed a marginal statistical advantage over BMI (ΔAUC = 0.030, P = 0.0726), and WC remained significantly superior to BMI (ΔAUC = 0.017, P = 0.0315).

Association between the degree of fat deposition and parameters in SLD

The stratified analysis of SLD fat deposition level showed that there was a significant difference in CVAI median between participants with mild fatty liver and non-fatty liver (75.91 difference), as well as between those with moderate to severe fatty liver and mild fatty liver (23.01 difference), both with P < 0.0001. There was also a significant difference in TyG-WHtR median between participants with mild fatty liver and non-fatty liver (0.3 difference), and between those with moderate to severe fatty liver and mild fatty liver (0.09 difference), both with P < 0.0001. The specific results are shown in Table 8.

Table 8 Numerical distribution of the three correlation parameters in steatotic liver disease with different degrees of fat deposition.

	Moderate to severe SLD (n = 148)	Mild SLD (n = 125)	Non-SLD (n = 224)	P value	U/χ2
CVAI, median (IQR)	162.37 (113.10-205.13)	139.36 (106.95-174.90)	63.45 (33.44-98.05)	< 0.0001	221.283
TyG-WHtR, median (IQR)	4.73 (4.31-5.27)	4.64 (4.23-5.04)	4.34 (3.79-4.95)	< 0.0001	205.628
Gender				< 0.0001	47.393
Male	89	70	69
Female	59	55	175

SLD: Steatotic liver disease; TyG-WHtR: Triglyceride glucose-waist-to-height ratio; CVAI: Chinese Visceral Adiposity Index; IQR: Interquartile range.

DISCUSSION

In this real-world study of a non-hepatology specialty clinic population, we enrolled patients with SLD while excluding those with alcohol-related liver disease (ALD) and SLD of specific etiologies (viral, drug-induced, genetic, etc.). The study population largely corresponded to the combined MASLD and metabolic alcohol-related liver disease groups[23], while also aligning closely with MASLD populations[24]. Significant associations were observed between male sex, TyG-WHtR, CVAI, and SLD risk in Chinese adults. ROC curve analysis demonstrated high predictive accuracy for both continuous variables in determining SLD status, with TyG-WHtR slightly outperforming CVAI. Stratified analysis by hepatic steatosis severity (none, mild, and moderate-to-severe) revealed significant differences in TyG-WHtR and CVAI distributions across steatosis grades.

The diagnosis of SLD relies on the gold standard of invasive pathological biopsy, which involves a risk of postoperative complications and relatively high costs[25]. Therefore, numerous studies have focused on exploring simple and effective alternatives for widespread population health monitoring, to early screen patients with SLD[25,26]. Such a straightforward and feasible indicator has significant clinical value, especially in non-hepatology specialty clinics and physical examination populations.

In our research, a combination of physical examination indicators obtained from standard instruments (body composition analyzers) and simple blood test indices (TyG-WHtR, CVAI) has proven useful for the early screening of SLD. TyG-WHtR has been recognized as a highly reliable indicator for predicting SLD in multiple studies. For example, Kuo et al[27]. found that the AUC for predicting hepatic steatosis in adult MetS (metabolic syndrome) patients in the US was 0.705 using TyG-WHtR. A cross-sectional study in Iran by Forouzesh et al[28]. also showed a significant correlation between TyG-WHtR and the degree of hepatic steatosis, with an AUC > 0.7. A study conducted in West Africa reported a similar result with an AUC of 0.81[29]. In our study, this accuracy metric was 0.876, and the optimal cutoff value derived from the ROC curve was 4.787, which is similar to previous studies (4.94)[30].

Furthermore, this predictive capacity appears superior to the aforementioned previously reported values[27,29]. These discrepancies in discriminative power may be attributed to differences in population characteristics, as our cohort represents mostly health-conscious individuals seeking dietary advice or weight control, without severe underlying diseases. Additionally, differences in the imaging reference standards (e.g., our combination of real-world ultrasound/CT/FibroScan vs strictly standardized protocols) could further influence the diagnostic thresholds across different ethnic cohorts.

However, there are not many studies on whether TyG-WHtR can further predict the degree of hepatic steatosis. Our study provides a comparison of the medians of TyG-WHtR in populations with non-steatotic liver, with mild steatotic liver, and with moderate to severe steatotic liver, and the differences were found to be significant (P < 0.0001). Previous studies have also shown that TyG-WHtR is a potential screening indicator for liver fibrosis associated with MASLD[27,30,31], and a predictor of cardiovascular disease (CVD) and mortality in adult MALSD individuals[32,33]. Therefore, the single indicator TyG-WHtR has the potential to combine multiple predictive functions. Our study provides confirmative and incremental insights into this specific clinical triage scenario.

Another indicator, CVAI, also possesses reliable predictive potential. According to a prospective cohort study conducted by Zhao et al[34] in China, CVAI is independently associated with the risk of MASLD regardless of BMI, with an AUC of 0.720. In another cross-sectional study of an American population, a positive correlation between CVAI and MASLD risk was also observed, with an AUC of 0.759[35]. In our study, this accuracy metric was 0.865, with the optimal cutoff value derived from the ROC curve falling between the two previous studies (79.19-136.654) at 101.9. Additionally, our study presents a comparison of medians for CVAI in individuals with non-steatotic liver, mild steatotic liver, and moderate to severe steatotic liver, showing significant differences (P < 0.0001). Previous studies have also shown that CVAI is a screening indicator for liver fibrosis related to nonalcoholic fatty liver disease (NAFLD)/MASLD[35,36], and is significantly associated with a higher incidence of CVD and stroke[11,12,17]. Like TyG-WHtR, CVAI is a single index that combines multiple predictive functions, although its diagnostic efficacy is slightly lower. Interestingly, CVAI is particularly positively correlated with the prevalence and risk of MASLD in lean (non-obese) adults[37,38], indicating a specific application direction for this index.

In addition, gender also shows significant correlation in predicting SLD. The prevalence of MASLD in males is significantly higher than in females, which is consistent with previous studies[39]. This is likely due to the regulatory role of sex hormones in the development of MASLD. Estrogen, in general, can regulate the occurrence of MASLD through the following mechanisms: Reducing triacylglycerol levels through estrogen receptor α, regulating liver gene expression to reduce hepatic de novo lipogenesis, reducing hepatic lipid accumulation, inhibiting the transport of free fatty acids to the liver, and suppressing the occurrence of IR[40]. The effects of androgen still need further demonstration. Current research focuses on two aspects: In males, a decrease in endogenous total testosterone is closely related to IR, obesity, and hepatic lipid accumulation; however, in females, the effects caused by hyperandrogenism are the opposite[41]. Interestingly, studies have shown that TyG-WHtR and CVAI are better at identifying metabolic associated fatty liver disease/NAFLD/MASLD in females[34,42], making these two indicators more balanced for diagnosing the condition across the entire population.

The present cohort was derived from a nutrition clinic, which may raise concerns about selection bias and the enrichment of metabolic disease. To clarify this, we quantified the prevalence of key metabolic risk factors: Diabetes was 7.35% (compared to approximately 11% in the general Chinese adult population)[43], obesity was 72.92% (vs approximately 50% combined overweight/obesity nationally)[44], and hypertension was 9.09% (vs approximately 25% generally)[45]. These figures suggest that our cohort largely mirrors the metabolic profile of community health-screening populations rather than a highly-selected inpatient metabolic cohort. To further address the applicability of these cut-offs in lower-risk groups, we conducted a sensitivity analysis excluding participants with known diabetes (n = 479); both TyG-WHtR and CVAI remained robust independent predictors, with TyG-WHtR demonstrating an AUC of 0.877 (95%CI: 0.846-0.907), and CVAI demonstrating an AUC of 0.866 (95%CI: 0.834-0.897). By standardizing the continuous variables, the per 1-SD increase in TyG-WHtR yielded an adjusted OR of 7.453 (95%CI: 3.474-15.988, P < 0.001), and the per 1-SD OR for CVAI was 3.562 (95%CI: 2.353-5.392, P < 0.001). This robustness indicates that their predictive utility is not solely driven by extreme metabolic dysfunction, supporting their extension to primary care settings.

Regarding screening algorithm performance and clinical trade-offs, interpreting the diagnostic performance requires evaluating the balance between sensitivity (91.21%) and specificity (68.85%). According to established principles conceptually supported by the EASL-EASD-EASO Guidelines[8], a first-line index in non-hepatology settings must prioritize high sensitivity and high NPV. In a hypothetical health check-up population with 30% SLD prevalence, TyG-WHtR yields an NPV of 94.8%, a PPV of 55.7%, and a NNS of approximately 3.7 to detect one true case (Supplementary Table 2). This ensures that low-risk patients genuinely have minimal probability of harboring SLD, while the clinical "cost" of its lower specificity merely results in a non-invasive ultrasound referral. Conversely, CVAI (cutoff ≥ 101.9) provides higher specificity (78.7%), yielding a PPV of 61.8%, an NPV of 90.4%, and an NNS of 4.1 in the same 30% prevalence setting. Consequently, we propose a highly efficient two-step algorithm for mass screening: Step 1, screen all adults with TyG-WHtR (cutoff ≥ 4.787) to broadly capture cases; Step 2, apply CVAI (cutoff ≥ 101.9) to test-positive individuals to filter out false positives before specialist referral. In high-risk clinics (prevalence > 50%), either index alone achieves a PPV > 74% and an NNS < 2.5. Future prospective studies should formally compare this two-step strategy against the current practice of unselected ultrasound referrals.

The limitations of this study mainly lie in six aspects: (1) This is a retrospective study, and there is a lack of necessary indicators for too many samples, resulting in a small proportion of samples actually included in the analysis; (2) The absolute sample size is not large, and there are diverse evaluation indicators, which may lead to biased results. For example, hypertension, which has been shown to be related to SLD in other studies[46], did not show a significant correlation in this study. Additionally, although TyG-WHtR showed statistically significant incremental value compared with simpler indicators (BMI and WC), the absolute AUC difference is small, and CVAI did not show a significant difference; larger samples are needed; (3) Compared with the source population, included participants were slightly older and had higher BMIs, potentially limiting generalizability to younger, leaner populations. However, since age and BMI were covariates in all multivariable models, internal validity remains robust; (4) The roughness of outpatient records may lead to deviations, and unmeasured confounders (e.g., dietary habits, physical activity, specific medications) could cause residual confounding; and (5) Diagnostic modality heterogeneity existed (ultrasound/CT/FibroScan); however, our sensitivity analysis limited to ultrasound-diagnosed participants (n = 462) yielded almost identical ORs and AUCs (Supplementary Tables 3 and 4), confirming minimal impact on our findings[6]. While we provided standard performance metrics, we did not perform complex clinical utility analyses like decision-curve analysis or net reclassification improvement due to retrospective data constraints. Incorporating these mathematical metrics in future prospective research will be essential to further quantify the net clinical benefit of adopting these indices. Furthermore, the clinical applicability of cutoffs (e.g., TyG-WHtR = 4.787) requires external validation.

In the future, we can further expand the sample size and conduct prospective research based on the results of this study, enrolling unselected community populations with protocolized data collection to validate and recalibrate the diagnostic thresholds. Meanwhile, it is observed that many simple measurement data, such as NC and ALM%, show good predictive performance in univariate analysis. A previous study involving 253 children has found a close relationship between NC measurement and childhood MASLD[47]. If a strong correlation can be found between a single measurement and SLD in the population, it will facilitate a wider group of people to self-assess this highly prevalent disease through simple methods at home, which is more beneficial for early detection and early intervention of the disease under different economic and medical conditions.

CONCLUSION

In the present study, TyG-WHtR and CVAI were reliable predictors of SLD and the degree of hepatic steatosis in the Chinese adult population, while also reaffirming that the incidence of SLD was higher in males than in females. TyG-WHtR and CVAI, as inexpensive and convenient indicators, can confirm with incremental insight for this specific clinical triage scenario, guide early intervention to control SLD in non-hepatology specialty clinics population and physical examination population. At the same time, the results of this preliminary study also point out a certain direction for subsequent research. It is necessary to optimize the risk prediction model for a broader population (with different economic and medical conditions), and it needs to be validated in an independent large-scale prospective cohort.