The prognostic significance of cholesterol, high-density lipoprotein and glucose (CHG) index in evaluating all-cause mortality risk in metabolic dysfunction-associated steatotic liver disease (MASLD) individuals: evidence from two cohort studies

Abstract

Background

The prognostic assessment of metabolic dysfunction-associated steatotic liver disease (MASLD) is of critical importance. Although previous international studies have suggested an association between the cholesterol, high-density lipoprotein, and glucose (CHG) index and mortality in MASLD patients, its generalizability in the Chinese population remains unclear, and there is a lack of predictive tools directly applicable for individual risk stratification.

Methods

A total of 6936 participants with MASLD were sourced from the China Health and Retirement Longitudinal Study (CHARLS) database from 2011 to 2020. Survival differences were visualized using the Kaplan-Meier method. Multivariate Cox regression was employed to assess the association between the CHG index and mortality risk. Smooth curve fitting analysis was conducted to examine the nonlinear relationship between them. Mediation analysis was performed to explore the mediating roles of C-reactive protein (CRP) and triglyceride glucose-weight-adjusted waist index (TyG-WWI). To evaluate the predictive value of the CHG index, we implemented a set of eight distinct machine learning (ML) models. Subgroup and sensitivity analyses were conducted to verify the robustness of the results. Concurrently, the National Health and Nutrition Examination Survey (NHANES) database (1999–2018) was utilized for external validation.

Results

The median follow-up duration was 7.44 years. A total of 615 mortality events were recorded. After adjusting for multiple confounding factors, an elevated CHG index was significantly associated with an increased risk of mortality (highest vs. lowest tertile, HR = 1.49, 95% CI [1.21–1.84], P = 0.0002). Kaplan-Meier curves demonstrated that higher CHG index levels were significantly associated with reduced survival (P < 0.0001). Smooth curve fitting and threshold effect analysis revealed a nonlinear relationship between them. Mediation analysis indicated that CRP and TyG-WWI mediated 13.36% and 45.33% of the effect of CHG index on mortality, respectively (all P < 0.05). Logistic Regression model showed superior discrimination. Decision curve analysis confirmed that the Logistic Regression model provided significant clinical net benefit across a wide range of risk thresholds. The primary analyses were replicated using data from the representative NHANES cohort, revealing a significant positive association between CHG index and mortality.

Conclusions

Elevated CHG index is significantly associated with an increased risk of mortality in the Chinese MASLD population, demonstrating a nonlinear relationship. Furthermore, indicators related to inflammation and insulin resistance (IR) significantly mediate these associations. The CHG index serves as an important tool for predicting long-term adverse outcomes in this population.

Graphical abstract

Research Insights

What is currently known about this topic?

1. International studies have suggested that the CHG index is associated with mortality in patients with MASLD.

2. However, its generalizability in Chinese populations remains unclear, and there is a lack of prediction tools that can be directly applied for individual risk stratification.

What is the key research question?

1. How does the CHG index influence mortality risk in Chinese MASLD individuals?

What is new?

1. This is the first Chinese cohort study to investigate the relationship between CHG index and mortality risk in MASLD individuals.

2. A prediction model was constructed using ML.

3. Mediation analysis revealed the mediating role of inflammatory and IR-related indicators for the first time.

How might this study influence clinical practice?

1. This research may promote the integration of the CHG index into routine risk assessment for MASLD patients. By providing a low-cost risk-stratification tool, it could facilitate early intervention. The identified threshold effect may offer evidence-based support for establishing personalized management cut-off values.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12933-026-03171-7.

Introduction

MASLD has become the most prevalent chronic liver disease worldwide [1]. Its disease spectrum can progress to liver fibrosis, cirrhosis, and hepatocellular carcinoma, and it significantly increases the risk of cardiovascular events and all-cause mortality [2, 3]. Accurate risk stratification and prognosis assessment are crucial for optimizing the clinical management of MASLD patients and enabling early intervention. However, existing non-invasive fibrosis markers (such as the fibrosis-4 index) primarily reflect structural changes in the liver and have limited ability to capture the dynamic nature of metabolic disorders and to assess individualized absolute risk [4, 5]. Therefore, exploring novel biomarkers that can integrate metabolic abnormalities and are easily accessible in clinical practice, and developing predictive tools capable of quantifying individual risk, hold profound clinical significance [6].

In recent years, composite indices have garnered attention due to their ability to comprehensively reflect multiple pathophysiological processes. The CHG index, an indicator integrating lipid and glucose metabolism, has demonstrated promising predictive value in fields such as cardiovascular disease [7, 8]. A study based on the United States population first reported the association between the CHG index and mortality in patients with MASLD (metabolic dysfunction-associated steatotic liver disease), providing preliminary evidence for its clinical application [9]. However, the generalizability of this conclusion to the Chinese population remains unknown [7, 8]. The Chinese MASLD population exhibits distinct characteristics in genetic background, lifestyle, and comorbidity profiles, and directly applying conclusions derived from Western data may introduce bias. Therefore, independent validation of the prognostic value of the CHG index using large-scale, local Chinese data is a necessary prerequisite for its application in clinical practice in China [10].

Although the association between the CHG index and mortality risk in patients with MASLD has been preliminarily established, untangling the underlying biological pathways is crucial for understanding disease mechanisms. This study aims to investigate the key roles of inflammation and IR through mediation analysis. To this end, CRP was selected as a marker of inflammation [11]. Synthesized by the liver, CRP is a classic indicator of low-grade inflammatory status, closely associated with adverse outcomes in metabolic diseases [12]. Concurrently, TyG-WWI and TyG-WHtR were introduced as more direct and stable surrogate measures for assessing IR [13, 14]. Substantial evidence indicates that TyG-related indices hold significant value in predicting MASLD progression and cardiovascular outcomes [15]. These indicators are deeply involved in the onset and progression of MASLD, forming a biologically plausible network of potential mediating pathways that link the CHG index to mortality risk. In-depth analysis of these pathways will help untangle the core mechanisms through which the CHG index influences MASLD prognosis.

Given these research gaps, this study primarily utilized the CHARLS database to examine the association between the CHG index and all-cause mortality in the Chinese MASLD population. The NHANES database was further analysed to validate the robustness of our findings. Mediation analysis was introduced to explore the mediating roles of inflammatory markers and IR indicators in the association between the CHG index and mortality. Concurrently, by integrating the CHG index with other readily available clinical parameters, a dynamic and personalized prediction model was developed using ML. This aims to offer novel insights for the clinical management of the MASLD population.

Methods

Data source and study population

The data for this study were derived from the CHARLS, a nationally representative large-scale longitudinal survey. CHARLS is designed to collect information on the socioeconomic status, health status, biomarkers, and healthcare utilization of Chinese residents aged 45 years and older. Its complex sampling design ensures the representativeness of the sample for the middle-aged and older adult population in China. This analysis utilized baseline survey data from CHARLS 2011 and 2015 wave (serving as the baseline for this study) and follow-up data on mortality events up to the latest available wave (with follow-ups conducted in 2013, 2015, 2018, and 2020, respectively). Participants were selected using a multi-stage stratified probability sampling design. Twenty-eight provinces in mainland China served as primary sampling units. Using the probability proportional to size sampling method, 150 counties (districts) were selected as secondary sampling units based on the proportion of the resident population size. Sampling proceeded stepwise to select townships (streets), administrative villages (neighborhood committees), and eligible survey respondents, ensuring that both densely and sparsely populated areas had an inclusion probability proportional to their size. This design achieved a high initial survey response rate of 80.5%, effectively reducing the risk of selection bias and significantly enhancing the sample’s representativeness of the national population and the external validity of the study conclusions [16]. To strengthen the robustness of the findings, sensitivity analyses were conducted using data from the NHANES (1999–2018).

To evaluate the association between the CHG index and all-cause mortality in participants with MASLD, only individuals diagnosed with MASLD were included. A total of 24,229 participants were reviewed across five interview cycles from 2011 to 2020. (1) Participants lacking data on blood glucose, cholesterol, and high-density lipoprotein (HDL) (n = 10949) were excluded, leaving 13,280 matched individuals. (2) Screening based on the lipid accumulation product (LAP) score led to the exclusion of 6120 participants (males with LAP < 30.5 and females with LAP < 23), resulting in 7160 matched data entries. (3) Participants of 82 lacking data on cardiometabolic risk factors were excluded, leaving 7078 matched data entries. (4) Participants of 142 with missing follow-up data were excluded. Ultimately, 6936 participants were included for statistical analysis (corresponding to the second step in the graphical abstract).

The CHARLS study was approved by the Peking University Institutional Review Board (IRB00001052-11015, IRB00001052-11014) and conducted in accordance with the ethical principles of the Declaration of Helsinki and the STROBE guidelines for reporting observational studies. Informed consent was obtained from all participants before their involvement.

Definition of MASLD

MASLD was defined according to the latest international expert consensus. The diagnosis of MASLD requires the simultaneous fulfillment of the following two core criteria.

Assessment of hepatic steatosis: based on previous literature reports [17, 18], the LAP index was employed in this study as a non-invasive tool to infer the presence of steatotic liver disease (SLD). LAP, which integrates information on central obesity and dyslipidemia, has been validated as an effective indicator for screening SLD in the general population. It is calculated using the following formulas: for males: LAP = (WC-65) × TG; for females: LAP = (WC-58) × TG. Based on established gender-specific cut-off points, participants with an LAP score ≥ 30.5 (males) or ≥ 23.0 (females) were inferred to have SLD.

The fatty liver index (FLI) is a simple tool for assessing the risk of fatty liver disease. In the NHANES cohort, FLI ≥ 60 was used as an alternative criterion for fatty liver assessment. Sensitivity analyses were conducted to examine whether the robustness of the core findings was affected by the choice of different indicators.

ALT, alanine aminotransferase; AST, aspartate aminotransferase; BMI: body mass index (kg/m²); TG, triglyceride (mg/dL for FLI); TG, triglyceride (mmol/L for LAP); GGT, γ-glutamyl transferase (IU/L); WC, waist circumference (cm).

Identification of cardiometabolic risk factors [8, 13]: based on the aforementioned presumption of the presence of SLD, participants were required to meet at least one of the following cardiometabolic risk factors to be ultimately defined as having MASLD: (1) BMI ≥ 25 kg/m² or WC ≥ 94 cm for males or ≥ 80 cm for females; (2) Fasting blood glucose (FBG) ≥ 100 mg/dL or 2-hour post-load glucose levels ≥ 140 mg/dL or glycated hemoglobin A1c (HbA1c) ≥ 5.7% or a diagnosis of diabetes mellitus (DM) or current treatment for DM; (3) Systolic blood pressure ≥ 130 mmHg or diastolic blood pressure ≥ 85 mmHg, or a diagnosis of hypertension, or current antihypertensive therapy; (4) TG ≥ 150 mg/dL or current lipid-lowering treatment; (5) HDL-C < 40 mg/dL for males or < 50 mg/dL for females or current lipid-lowering treatment.

CHG index assessment

The CHG index was calculated using the following specific formula: CHG index = Ln [Total cholesterol (TC, mg/dL) × FBG (mg/dL) / (2 × HDL-C (mg/dL))] [19].

Assessment of endpoint events

All-cause mortality was selected as the outcome for patients with MASLD. Deaths occurring from the second to the fifth wave were confirmed via death certificates, medical records, or interviews with relatives. The exact time of death was only available for waves two through five. The time-to-event was determined by measuring the duration from baseline to the last wave of interview. Follow-up time was calculated from the date of the baseline survey until the date of death, the date of the last successful follow-up, or the end of the study follow-up period, whichever occurred first.

Covariates

Demographic characteristics of participants with MASLD were collected from the CHARLS database. Sociodemographic information included: age, gender, household registration (hukou), marital status, education, WC, and BMI. Laboratory measurements included: platelet count (PLT), white blood cell count (WBC), hemoglobin (HGB), hematocrit (HCT), mean corpuscular volume (MCV), TG, low-density lipoprotein cholesterol (LDL-C), TC, HDL-C, GLU, HbA1c, and estimated glomerular filtration rate (eGFR). Lifestyle information included smoking, alcohol consumption habits, and physical activity. Comorbidities included: cancer, hypertension, DM, chronic kidney disease (CKD), lung disease, and digestive disease.

Mediating variable evaluation

To explore the potential mechanisms through which the CHG index influences mortality risk, this study evaluated the roles of inflammation and IR as potential mediating variables. The triglyceride-glucose (TyG) index and its derived metrics [(triglyceride glucose-waist circumference, TyG-WC), (triglyceride glucose-body mass index, TyG-BMI), (triglyceride glucose-waist-to-height ratio, TyG-WHtR), and TyG-WWI] are important markers reflecting IR. The specific calculation formulas are as follows: TyG index = Ln [TG (mg/dL) × FBG (mg/dL)/2]; TyG-BMI index = TyG index × BMI; TyG-WHtR index = TyG index × WC/height; TyG-WWI index = TyG index × WC/√weight; TyG-WC index = TyG index × WC. All raw data required for these calculations (TG, FBG, WC, height, weight) were obtained from baseline measurements [20].

Statistical analyses

Continuous variables are summarized as mean ± standard deviation (SD), and categorical variables are summarized as frequency (%). Analysis of variance (ANOVA) or the Kruskal-Wallis test was used to compare differences in continuous variables between groups, while the chi-square test or Fisher’s exact test was used to compare differences in categorical variables between groups. Missing data were handled using multiple imputation by chained equations. The proportion of missing data for each variable is presented in Table S1.

The association between the CHG index and the risk of all-cause mortality in the MASLD population was assessed using the Cox proportional hazards regression model. In the multivariable Cox regression analysis, Model 1 was unadjusted; Model 2 was adjusted for age and gender; and Model 3 was the fully adjusted model, which included age, gender, smoking, drinking, physical activity, and eGFR. In the survival analysis, survival time was defined as the duration from the baseline survey to either the occurrence of death or the end of follow-up. Kaplan-Meier curves were plotted for the tertile groups of the CHG index, and differences in survival between groups were compared using the log-rank test.

We employed smooth curve fitting and threshold effect analysis to explore the dose-response relationship between the CHG index and mortality risk. Based on the fully adjusted model (Model 3), a multivariate-adjusted curve depicting the relationship between the CHG index and the probability of mortality risk was plotted. A two-piecewise Cox regression model was used for the threshold effect analysis. The optimal inflection point for the association between the CHG index and mortality risk was determined using a recursive algorithm and the maximum likelihood method. Hazard ratio (HR) and the 95% confidence interval (CI) were calculated on both sides of the inflection point. The statistical significance of the threshold effect was assessed using the likelihood ratio test.

To evaluate the predictive value of the CHG index for mortality risk in the MASLD population, we first performed feature selection using LASSO regression and the Boruta algorithm. LASSO regression achieves variable compression and selection through penalty coefficients, while the Boruta algorithm identifies relevant features based on permutation importance derived from Random Forest. The intersection of the results from both methods was taken to determine the final set of predictor variables for model inclusion. Following feature selection, the variance inflation factor (VIF) was employed to assess multicollinearity, and variables with a VIF > 10 were excluded to avoid interference from collinearity. Subsequently, the dataset was randomly divided into a development set (70%) and a test set (30%). Eight ML models were constructed: Logistic Regression, K-Nearest Neighbors (KNN), Decision Tree, Random Forest, XGBoost, LightGBM, Support Vector Machine (SVM), and Neural Network. The models were optimized through hyperparameter tuning. Their performance was evaluated using the receiver operating characteristic (ROC) curve and the area under the curve (AUC). Additionally, metrics including accuracy, sensitivity, specificity, precision, and the F1 score were calculated to assess model performance. To further evaluate the clinical utility of the predictive models, decision curve analysis (DCA) was conducted to quantify the net benefit at different threshold probabilities. This net benefit was compared against the strategies of “treat all” and “treat none”. A calibration curve was plotted to verify the reliability and accuracy of the models.

To explore the potential mechanisms through which the CHG index influences mortality risk, a causal mediation analysis was conducted. Using CRP and TyG-WWI as potential mediators, the proportions of the total effect attributable to the indirect effect and the direct effect were calculated. The statistical significance of the mediation effects was assessed by calculating 95% CI using the bootstrap method with 1000 resamples.

To verify the robustness of the results, a comprehensive subgroup analysis was conducted. Stratified analyses were performed based on age, gender, marital status, education, hypertension, CKD, smoking, and BMI. The likelihood ratio test was used to compare the model fit before and after the inclusion of interaction terms, thereby assessing whether significant interactions existed between each variable and the CHG index.

In addition, twelve sensitivity analyses were conducted: (1) exclusion of elderly participants with a baseline age ≥ 80 years; (2) exclusion of participants who died early during follow-up (follow-up time ≤ 24 months) to mitigate reverse causation; (3) additional adjustment for the use of antihypertensive drugs; (4) additional adjustment for the use of anti-hyperglycemic medications; (5, 6) external validation using data from participants with MASLD (screened by LAP and FLI respectively) across ten cycles of NHANES 1999–2018 to assess the generalizability of the findings; (7) excluding participants with BMI exceeding 75; (8) association of CHG index with mortality risk among MASLD patients with T2DM; (9) association of CHG index with mortality risk among MASLD patients with dyslipidemia; (10) association of simplify-CHG index with mortality risk among MASLD patients; (11) association of CHG index with mortality risk among MASLD patients before multiple imputation; (12) association of CHG index with mortality risk among MASLD patients during different follow-up periods. All tests were two-sided, and a P-value < 0.05 was considered statistically significant. Statistical analyses were performed using R software.

Results

Baseline characteristics

A total of 6936 participants were included in this study. After stratifying the participants into tertiles based on the CHG index, significant differences in baseline characteristics were observed among the groups. Compared with the low CHG group (T1), participants in the high CHG index group (T3) were older (58.91 ± 10.16 vs. 54.76 ± 10.70 years) and had a higher proportion of males (43.45% vs. 22.70%). Laboratory parameters demonstrated a clear trend of metabolic deterioration: levels of TG (250.83 ± 170.28 mg/dL), TC (212.20 ± 43.08 mg/dL), LDL-C (121.10 ± 43.36 mg/dL), GLU (139.50 ± 56.66 mg/dL), and HbA1c (5.92 ± 1.41%) were significantly elevated in the T3 group, while HDL-C levels were significantly lower (37.24 ± 9.37 mg/dL). Furthermore, WC was larger in the high CHG index group, and WBC, HGB, and HCT also showed an increasing trend. Regarding clinical comorbidities, the prevalence of hypertension (54.04%), DM (49.11%), and CKD (20.40%) was significantly higher in the T3 group than in the T1 group (36.82%, 6.22%, and 18.24%, respectively). A history of smoking, drinking, and the proportion of participants with urban household registration were also higher in the T3 group (Table 1).

Table 1.

Baseline characteristics of study individuals according to CHG index

Characteristics	Total (n = 6936)	Low CHG index (T1, n = 2312)	Middle CHG index (T2, n = 2312)	High CHG index (T3, n = 2312)	P-value
Age (Years)	56.88 ± 10.46	54.76 ± 10.70	56.89 ± 10.12	58.91 ± 10.16	< 0.0001
PLT (109/L)	212.80 ± 74.35	210.47 ± 81.17	212.47 ± 70.08	215.37 ± 71.57	0.0808
TG (mg/dL)	182.68 ± 123.81	129.45 ± 57.20	165.66 ± 75.10	250.83 ± 170.28	< 0.0001
LDL-C (mg/dL)	114.40 ± 35.62	102.65 ± 26.81	118.94 ± 31.43	121.10 ± 43.36	< 0.0001
WC (cm)	91.40 ± 8.67	89.96 ± 8.48	91.31 ± 8.30	92.86 ± 8.99	< 0.0001
BMI (kg/m2)	26.96 ± 50.45	29.17 ± 88.47	25.68 ± 4.40	26.12 ± 4.36	0.0396
WBC (103/uL)	6.28 ± 1.80	6.02 ± 1.80	6.23 ± 1.75	6.59 ± 1.81	< 0.0001
HGB (g/dL)	14.17 ± 2.11	13.76 ± 2.12	14.23 ± 2.06	14.51 ± 2.09	< 0.0001
HCT (%)	41.35 ± 5.95	40.34 ± 5.79	41.60 ± 5.91	42.07 ± 6.02	< 0.0001
MCV (fl.)	89.78 ± 7.80	89.60 ± 8.44	89.98 ± 7.34	89.76 ± 7.61	0.2588
TC (mg/dL)	196.94 ± 38.80	180.67 ± 31.71	197.29 ± 33.92	212.20 ± 43.08	< 0.0001
GLU (mg/dL)	112.68 ± 39.78	93.60 ± 12.87	104.20 ± 13.49	139.50 ± 56.66	< 0.0001
HDL-C (mg/dL)	45.41 ± 11.91	53.99 ± 11.25	45.37 ± 8.56	37.24 ± 9.37	< 0.0001
HbA1c (%)	5.56 ± 0.99	5.35 ± 0.60	5.40 ± 0.61	5.92 ± 1.41	< 0.0001
Hukou (%)					< 0.0001
Agricultural	60.82	63.02	58.66	60.88
Non-agricultural	27.35	21.52	28.57	31.71
Other	11.83	15.46	12.77	7.41
Gender (%)
Male	34.14	22.7	35.79	43.45	< 0.0001
Female	65.86	77.3	64.21	56.55
Marital status (%)
Married/partner	88.65	89.33	88.8	87.85	0.2769
Widowed/divorced/separated/never married	11.35	10.67	11.2	12.15
Education (%)
Below primary school	38.42	37.2	39.12	38.89	< 0.0001
Primary school	30.59	34.55	30.59	26.79
Middle school	19.15	19.07	16.8	21.58
High school and above	11.84	9.18	13.49	12.73
Cancer (%)
No	96.16	96.46	96.13	95.9	0.6107
Yes	3.84	3.54	3.87	4.1
Smoking (%)
Never	70.15	79.46	68.76	62.62	< 0.0001
Former	8.43	5.43	8.32	11.43
Now	21.42	15.11	22.92	25.95
Drinking (%)
Never	65.32	67.88	63.98	64.21	0.0001
Former	7.48	6.58	6.67	9.16
Now	27.2	25.54	29.35	26.63
Hypertension (%)
No	53.27	63.18	51.11	45.96	< 0.0001
Yes	46.73	36.82	48.89	54.04
Diabetes (%)
No	76.67	93.78	85.99	50.89	< 0.0001
Yes	23.33	6.22	14.01	49.11
CKD (%)
No	82.29	81.76	85.46	79.6	< 0.0001
Yes	17.71	18.24	14.54	20.4
Lung disease (%)
No	87.71	87.36	88.2	87.54	0.6546
Yes	12.29	12.64	11.8	12.46
Digestive disease (%)
No	78.5	77.14	80.18	78.03	0.074
Yes	21.5	22.86	19.82	21.97
eGFR	93.11 ± 15.96	95.43 ± 15.30	92.98 ± 15.03	90.93 ± 17.16	< 0.001
Physical activity (%)
No	10.67	9.6	10.68	11.72	0.066
Yes	89.33	90.4	89.32	88.28
CHG index	5.48 ± 0.44	5.05 ± 0.18	5.42 ± 0.09	5.94 ± 0.37	< 0.0001
Follow-up time (months)	86.85 ± 25.98	82.66 ± 25.90	87.19 ± 25.50	90.53 ± 25.95	< 0.0001

The continuous variables are expressed as Mean ± standard and tertiles, and the categorical variables are expressed as numbers (%).T1, Tertile 1; T2, Tertile 2; T3, Tertile 3; PLT, platelet; TG, triglyceride; LDL-C, low-density lipoprotein cholesterol; WC, waist circumference; BMI, body mass index; WBC, white blood cell count; HGB, hemoglobin; HCT, hematocrit; MCV, mean corpuscular volume; TC, total cholesterol; GLU, glucose; HDL-C, high-density lipoprotein cholesterol; HbA1c, glycated hemoglobin A1c; CKD, chronic kidney disease; CHG index, cholesterol, high-density lipoprotein, and glucose index; eGFR, estimated glomerular filtration rate

Association between CHG index and mortality in MASLD participants

After multivariate adjustment (Table 2), a significant positive association was observed between the CHG index and all-cause mortality risk. In the continuous variable analysis, each one-unit increase in the CHG index was associated with a 67% (HR 1.67, 95% CI 1.44–1.94, P < 0.0001), 56% (HR 1.56, 95% CI 1.33–1.83, P < 0.0001), and 58% (HR 1.58, 95% CI 1.34–1.86, P < 0.0001) increase in mortality risk before and after adjustment for confounders, respectively. In the categorical analysis, using the low CHG index group as the reference, the high CHG index group exhibited a significantly elevated mortality risk across all three models, with adjusted HRs of 1.77 (95% CI 1.44–2.17), 1.52 (95% CI 1.24–1.87), and 1.49 (95% CI 1.21–1.84), respectively. In contrast, the moderate CHG index group showed a statistically non-significant association in the univariate and multivariate adjustment model (P > 0.05). Trend tests indicated a significant linear increase in mortality risk with ascending CHG index categories (P < 0.0001 for all models). In addition to univariate analyses, the number of death events, incidence, and HR (95% CI) for mortality were analysed according to the quartiles of the CHG index (Table S2-S4).

Table 2.

The relationship of CHG index with mortality by Cox regression analysis in the MASLD individuals

CHG index	HR (95% CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.67 (1.44, 1.94) < 0.0001	1.56 (1.33, 1.83) < 0.0001	1.58 (1.34, 1.86) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	1.21 (0.97, 1.51) 0.0878	1.12 (0.90, 1.40) 0.3114	1.12 (0.90, 1.40) 0.3072
High CHG index	1.77 (1.44, 2.17) < 0.0001	1.52 (1.24, 1.87) < 0.0001	1.49 (1.21, 1.84) 0.0002
P for trend	< 0.0001	< 0.0001	< 0.0001

Model 1 adjusted for no variables. Model 2 adjusted for age and gender. Model 3 adjusted for age, gender, smoking, drinking, physical activity and eGFR. eGFR, estimated glomerular filtration rate; CHG index, cholesterol, high-density lipoprotein, and glucose index

A total of 6936 patients with MASLD were followed up for a median duration of 89 months. A total of 615 deaths (8.87%) were observed. The CHG index was categorized into quartiles, and the number of death events and occurrence rates were briefly reported for each group. A significant increasing trend in all-cause death events and occurrence rates was observed with ascending CHG index quartiles (Table S3). Furthermore, Cox regression analysis was performed according to quartile groups to examine the association between the CHG index and mortality in the MASLD population (Table S4). Smooth curve fitting analysis (Fig. 1A) revealed a significant nonlinear positive correlation between the CHG index and the risk of all-cause mortality. As the CHG index increased, the probability of death showed a gradual upward trend. Notably, the slope of the curve increased markedly when the CHG index exceeded 5.17, suggesting an accelerated increase in mortality risk. Kaplan-Meier survival curve analysis (Fig. 1B) demonstrated a statistically significant difference in survival probability among the tertile groups stratified by the CHG index (Log-rank P < 0.0001). The high CHG index group exhibited the lowest survival rate, while the low CHG index group showed the highest survival rate. The medium CHG index group fell between the two. Furthermore, the survival curves for the three groups gradually diverged over the course of follow-up.

Fig. 1.

A Nonlinear relationship between CHG index and all-cause mortality in MASLD individuals. B Kaplan-Meier curves of the survival rate of participants with CHG index tertiles

Threshold effect analysis

To further elucidate the dose-response relationship between the CHG index and all-cause mortality risk, a two-piecewise linear regression model was employed for threshold effect analysis (Table 3). The results indicated an inflection point at 5.17, with the two-piecewise model demonstrating a better goodness-of-fit compared to the standard linear model (likelihood ratio test P = 0.024). Specifically, no significant association with mortality risk was observed when the CHG index was < 5.17 (HR 0.64, 95% CI 0.32–1.27, P = 0.1979). Conversely, when the CHG index was > 5.17, mortality risk increased significantly with higher CHG index values (HR 1.71, 95% CI 1.44–2.04, P < 0.0001). A significant difference in effect magnitude was noted on either side of the inflection point (HR 2.69, 95% CI 1.28–5.69, P = 0.0094). These findings suggest a distinct threshold effect of the CHG index on all-cause mortality risk in patients with MASLD, with a significant positive association observed only when the index exceeds 5.17.

Table 3.

Threshold effect analysis of CHG index on all-cause mortality in MASLD individuals

All-cause mortality	Adjusted HR (95% CI), P-value
Fitting by the standard linear model	1.58 (1.34, 1.86) < 0.0001
Fitting by the two-piecewise linear model
Inflection point	5.17
CHG index < 5.17	0.64 (0.32, 1.27) 0.1979
CHG index > 5.17	1.71 (1.44, 2.04) < 0.0001
HR between < 5.17 and > 5.17	2.69 (1.28, 5.69) 0.0094
P for Log-likelihood ratio test	0.024

The analysis was adjusted for age, gender, smoking, drinking, physical activity and eGFR. eGFR, estimated glomerular filtration rate; CHG index, cholesterol, high-density lipoprotein, and glucose index

Associations of the CHG index with inflammation and TyG-related indicators in MASLD patients

The CHG index was significantly positively correlated with the inflammatory marker CRP and IR-related indices (TyG, TyG-WHtR, TyG-WWI, TyG-BMI, and TyG-WC) (Table 4). After multivariate adjustment (Model 3), each unit increase in the CHG index was associated with an average increase of 1.74 mg/L in CRP levels (β = 1.74, 95% CI 1.42–2.07, P < 0.0001). Categorical analysis revealed that, compared to the low CHG index group, the high CHG index group exhibited significantly elevated levels of both inflammatory and IR markers. Specifically, CRP was elevated by 1.43 units (β = 1.43, 95% CI 1.07–1.78, P < 0.0001), whereas the association between the medium CHG index group and CRP was not statistically significant (P = 0.9094). For TyG and its composite indices, analyses using both continuous and categorical variables demonstrated a significant linear increasing trend in the levels of these indices with ascending CHG index categories (all P for trend < 0.05). Specifically, the TyG level in the high CHG index group was 0.94 units higher than that in the reference group (β = 0.94, 95% CI 0.91–0.97, P < 0.0001), and these associations persisted after adjustment for confounding factors.

Table 4.

Associations of the CHG index with inflammation and TyG-related indicators in MASLD patients

CHG index	β (95%CI) P-value
	Model 1	Model 2	Model 3
C-reactive protein
Continuous	1.91 (1.60, 2.23) < 0.0001	1.77 (1.44, 2.09) < 0.0001	1.74 (1.42, 2.07) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	0.16 (-0.19, 0.50) 0.3644	0.03 (-0.32, 0.38) 0.8526	0.02 (-0.33, 0.37) 0.9094
High CHG index	1.67 (1.32, 2.01) < 0.0001	1.46 (1.11, 1.82) < 0.0001	1.43 (1.07, 1.78) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001
TyG
Continuous	1.09 (1.07, 1.11) < 0.0001	1.09 (1.07, 1.11) < 0.0001	1.09 (1.07, 1.11) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	0.35 (0.33, 0.38) < 0.0001	0.34 (0.32, 0.37) < 0.0001	0.35 (0.32, 0.37) < 0.0001
High CHG index	0.95 (0.92, 0.98) < 0.0001	0.94 (0.91, 0.97) < 0.0001	0.94 (0.91, 0.97) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001
TyG-WHtR
Continuous	0.70 (0.66, 0.75) < 0.0001	0.71 (0.66, 0.76) < 0.0001	0.71 (0.66, 0.76) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	0.19 (0.14, 0.24) < 0.0001	0.19 (0.13, 0.24) < 0.0001	0.19 (0.13, 0.24) < 0.0001
High CHG index	0.60 (0.55, 0.65) < 0.0001	0.60 (0.54, 0.65) < 0.0001	0.60 (0.55, 0.65) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001
TyG-WWI
Continuous	12.81 (12.39, 13.24) < 0.0001	12.73 (12.32, 13.14) < 0.0001	12.77 (12.37, 13.18) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	3.84 (3.34, 4.35) < 0.0001	3.61 (3.13, 4.09) < 0.0001	3.65 (3.17, 4.13) < 0.0001
High CHG index	11.12 (10.62, 11.63) < 0.0001	10.75 (10.26, 11.24) < 0.0001	10.84 (10.36, 11.33) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001
TyG-BMI
Continuous	26.44 (12.18, 40.70) 0.0003	27.79 (13.17, 42.40) 0.0002	27.50 (12.85, 42.15) 0.0002
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	-2.21 (-17.74, 13.33) 0.7808	-1.33 (-17.04, 14.39) 0.8686	-1.59 (-17.32, 14.13) 0.8427
High CHG index	16.70 (1.16, 32.24) 0.0352	18.11 (2.10, 34.11) 0.0266	17.36 (1.30, 33.42) 0.0341
P for trend	0.0352	0.0258	0.0332
TyG-WC
Continuous	125.42 (121.05, 129.78) < 0.0001	116.76 (112.44, 121.08) < 0.0001	116.85 (112.53, 121.18) < 0.0001
Categories
Low CHG index	Reference	Reference	Reference
Middle CHG index	43.42 (38.35, 48.48) < 0.0001	36.94 (31.99, 41.89) < 0.0001	37.16 (32.22, 42.10) < 0.0001
High CHG index	113.40 (108.33, 118.46) < 0.0001	102.89 (97.85, 107.93) < 0.0001	103.25 (98.20, 108.29) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001

Model 1 adjusted for no variables. Model 2 adjusted for age and gender. Model 3 adjusted for age, gender, smoking, drinking, physical activity and eGFR. eGFR, estimated glomerular filtration rate; TyG-WHtR, triglyceride-glucose-waist-to-height ratio; TyG-WWI, triglyceride-glucose-weight-adjusted waist index; TyG-BMI, triglyceride-glucose-body mass index; TyG-WC, triglyceride-glucose-waist circumference; TyG, triglyceride-glucose index; CHG index, cholesterol, high-density lipoprotein, and glucose index

The associations of inflammation and TyG-related indicators with all-cause mortality in MASLD patients

Further analysis revealed that CRP, TyG, TyG-WHtR, and TyG-WWI were all significantly positively associated with the risk of all-cause mortality in patients with MASLD. After multivariate adjustment, each unit increase in CRP was associated with a 3% increase in mortality risk (HR 1.03, 95% CI 1.03–1.04, P < 0.0001). The high-level CRP group had a 40% higher mortality risk compared to the low-level group (HR 1.40, 95% CI 1.15–1.70, P = 0.0009), while the moderate-level group showed no significant association with mortality risk (P > 0.05). TyG, TyG-WHtR, and TyG-WWI exhibited similar patterns. Notably, the high-level TyG-WWI group demonstrated the most pronounced increase in mortality risk (HR 1.70, 95% CI 1.35–2.13, P < 0.0001). Significant positive linear trends were observed for all these indices (P for trend < 0.05). However, TyG-BMI and TyG-WC showed no significant independent association with mortality risk across the models (Model 3, P > 0.05) (Table 5).

Table 5.

The associations of inflammation and TyG-related indicators with all-cause mortality in MASLD patients

C-reactive protein	HR (95%CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.04 (1.03, 1.04) < 0.0001	1.03 (1.03, 1.04) < 0.0001	1.03 (1.03, 1.04) < 0.0001
Categories
Low C-reactive protein	Reference	Reference	Reference
Middle C-reactive protein	1.26 (1.02, 1.56) 0.0310	1.09 (0.89, 1.35) 0.4011	1.08 (0.87, 1.33) 0.4863
High C-reactive protein	1.83 (1.50, 2.23) < 0.0001	1.46 (1.20, 1.78) 0.0002	1.40 (1.15, 1.70) 0.0009
P for trend	< 0.0001	< 0.0001	0.0006

TyG	HR (95%CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.31 (1.17, 1.47) < 0.0001	1.34 (1.18, 1.51) < 0.0001	1.34 (1.18, 1.52) < 0.0001
Categories
Low TyG	Reference	Reference	Reference
Middle TyG	1.33 (1.08, 1.63) 0.0072	1.21 (0.98, 1.48) 0.0740	1.18 (0.96, 1.45) 0.1181
High TyG	1.53 (1.25, 1.87) < 0.0001	1.55 (1.27, 1.90) < 0.0001	1.50 (1.22, 1.84) 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001

TyG-WHtR	HR (95% CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.06 (1.03, 1.09) 0.0004	1.07 (1.02, 1.12) 0.0054	1.07 (1.02, 1.12) 0.0095
Categories
Low TyG-WHtR	Reference	Reference	Reference
Middle TyG-WHtR	1.28 (1.04, 1.59) 0.0213	1.03 (0.83, 1.27) 0.8005	1.03 (0.83, 1.27) 0.8208
High TyG-WHtR	1.71 (1.40, 2.08) < 0.0001	1.29 (1.05, 1.58) 0.0137	1.29 (1.05, 1.58) 0.0134
P for trend	< 0.0001	0.0084	0.0081

TyG-WWI	HR (95%CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.04 (1.03, 1.05) < 0.0001	1.02 (1.01, 1.03) < 0.0001	1.02 (1.01, 1.03) < 0.0001
Categories
Low TyG-WWI	Reference	Reference	Reference
Middle TyG-WWI	1.47 (1.16, 1.88) 0.0016	1.18 (0.92, 1.50) 0.1847	1.20 (0.94, 1.52) 0.1465
High TyG-WWI	2.86 (2.30, 3.55) < 0.0001	1.70 (1.35, 2.13) < 0.0001	1.70 (1.35, 2.13) < 0.0001
P for trend	< 0.0001	< 0.0001	< 0.0001

TyG-BMI	HR (95%CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.00 (1.00, 1.00) 0.8105	1.00 (1.00, 1.00) 0.5310	1.00 (1.00, 1.00) 0.6326
Categories
Low TyG-BMI	Reference	Reference	Reference
Middle TyG-BMI	0.92 (0.76, 1.11) 0.4011	1.06 (0.87, 1.29) 0.5487	1.07 (0.88, 1.30) 0.5007
High TyG-BMI	0.82 (0.67, 0.99) 0.0405	1.06 (0.87, 1.30) 0.5552	1.06 (0.86, 1.29) 0.5953
P for trend	0.041	0.5411	0.5770

TyG-WC	HR (95%CI) P-value
	Model 1	Model 2	Model 3
All-cause mortality
Continuous	1.00 (1.00, 1.00) 0.0014	1.00 (1.00, 1.00) 0.1294	1.00 (1.00, 1.00) 0.1468
Categories
Low TyG-WC	Reference	Reference	Reference
Middle TyG-WC	1.34 (1.09, 1.64) 0.0051	1.11 (0.90, 1.36) 0.3519	1.09 (0.88, 1.34) 0.4289
High TyG-WC	1.42 (1.17, 1.74) 0.0005	1.17 (0.95, 1.44) 0.1375	1.16 (0.94, 1.43) 0.1696
P for trend	0.0006	0.1415	0.1706

Model 1 adjusted for no variables. Model 2 adjusted for age and gender. Model 3 adjusted for age, gender, smoking, drinking, physical activity and eGFR. eGFR, estimated glomerular filtration rate; TyG-WHtR, triglyceride-glucose-waist-to-height ratio; TyG-WWI, triglyceride-glucose-weight-adjusted waist index; TyG-BMI, triglyceride-glucose-body mass index; TyG-WC, triglyceride-glucose-waist circumference; TyG, triglyceride-glucose index; CHG index, cholesterol, high-density lipoprotein, and glucose index

Mediating role of inflammation and IR-related indicators

Mediation analysis revealed that the association between CHG index and all-cause mortality risk in MASLD participants was partially mediated by CRP and TyG-WWI. The mediation proportion were 13.36% and 45.33%, respectively (all P < 0.05) (Fig. 2 and Table S5).

Fig. 2.

The mediating effects of TyG-WWI (A), and C-reactive protein (B) on the relationship of CHG index with all-cause mortality in MASLD individuals. The mediating analysis adjusted for age, gender, smoking, drinking, physical activity and eGFR. TyG-WWI: triglyceride-glucose-weight-adjusted waist index; CHG index, cholesterol, high-density lipoprotein, and glucose index; eGFR, estimated glomerular filtration rate

Variable selection for the model

We employed two complementary feature selection methods, LASSO regression (Fig. 3A, B) and the Boruta algorithm (Fig. 3C), to reduce dimensionality and eliminate collinearity among variables. The following nine features were selected using LASSO regression: BLU, age, smoking, hypertension, CKD, gender, CHG index, marital status, and education. The following fifteen features were selected using the Boruta algorithm: MCV, TC, GLU, HbA1c, age, cancer, smoking, drinking, hypertension, diabetes, CKD, gender, CHG index, marital status, and education. Through a cross-analysis of the results from both algorithms, nine common feature variables selected by both methods were identified: BLU, age, smoking, hypertension, CKD, gender, CHG index, marital status, and education (Fig. 3D). These variables were ultimately used to construct the model. The absence of multicollinearity among these predictors was confirmed by examining the VIF (Table S6).

Fig. 3.

LASSO regression and Boruta algorithm are employed during the variable selection phase. Coefficient profiles (A) and optimal lambda selection (B) in the LASSO regression. (C) The Boruta algorithm was employed to assess the importance of potential risk factors for mortality. Important variables (green boxes), unimportant variables (red boxes). (D) Venn diagram illustrating the overlap of features selected by LASSO regression and the Boruta algorithm

Development and validation of predictive models

Figure 4 presents the ROC curves and their corresponding AUC values for 8 ML models (Logistic Regression, Decision Tree, Random Forest, KNN, SVM, Neural Network, XGBoost, and LightGBM) on both the train and test sets. In the train set (Fig. 4B), all models demonstrated excellent fitting performance. On the test set (Fig. 4A), the AUC values for the models were as follows: LightGBM (0.808), XGBoost (0.803), Logistic Regression (0.810), Neural Network (0.801), Random Forest (0.798), KNN (0.698), Decision Tree (0.692), and SVM (0.616). For a comprehensive evaluation, the performance metrics of these models were compared, as shown in Table S7. Since AUC is a key criterion for selecting the most effective ML method, the Logistic Regression model was identified as optimal. The calibration curve (Fig. 5A) and DCA (Fig. 5B) further indicated that the Logistic Regression model exhibited strong clinical reliability and accuracy while providing a significant net benefit.

Fig. 4.

The ROC of the eight machine learning models in the test (A) and train (B) sets

Fig. 5.

The calibration curve assessed the reliability of the predicted probabilities by comparing the eight models (A). DCA displaying net benefit across different risk thresholds across eight models (B)

To facilitate clinical translation, a nomogram was constructed based on the key predictors identified in the ML model (Figure S1). This tool allows the conversion of values for each predictor into corresponding points on an axis, enabling the direct estimation of an individual’s 1-year, 3-year, and 5-year survival probabilities through the summation of these points.

Subgroup and sensitivity analyses

To verify the robustness of the results, a comprehensive subgroup analysis was conducted (Fig. 6). Stratified analyses were performed based on age, gender, marital status, education, hypertension, CKD, smoking, and BMI. A significant interaction between age and the CHG index was identified among these variables by comparing model fits with and without the interaction term using the likelihood ratio test. Specifically, the risk increase associated with an elevated CHG index was more pronounced in individuals aged < 60 years (HR 1.95, 95% CI 1.47–2.58, P < 0.0001), whereas a smaller increase in risk was observed for the same elevation in the CHG index among those aged ≥ 60 years (HR 1.33, 95% CI 1.08–1.63, P = 0.0062).

Fig. 6.

Subgroup analysis of the correlation between CHG index and all-cause mortality in MASLD patients. Each subgroup analysis was adjusted for age, gender, smoking, drinking, physical activity and eGFR, except for stratified variables. BMI, body mass index; CKD, chronic kidney disease; eGFR, estimated glomerular filtration rate; CHG index, cholesterol, high-density lipoprotein, and glucose index

Furthermore, twelve sensitivity analyses were conducted: (1) exclusion of elderly participants aged ≥ 80 years at baseline (Table S8); (2) exclusion of participants who died early during follow-up (follow-up time ≤ 24 months) (Table S9) to mitigate reverse causality; (3) additional adjustment for the use of antihypertensive drugs (Table S10); (4) additional adjustment for the use of anti-hyperglycemic medications (Table S11); (5) external validation using data from participants with MASLD across ten cycles (1999–2018) of NHANES, selecting FLI ≥ 60 as the new diagnostic criterion for SLD to evaluate the impact of the CHG index on mortality risk in different MASLD populations. Specifically, in Models 1–3, the CHG index was positively associated with all-cause mortality in the MASLD population (Model 1: HR = 1.59, 95% CI 1.43–1.77; Model 2: HR = 1.42, 95% CI 1.26–1.60; Model 3: HR = 1.40, 95% CI 1.22–1.60) (Table S12). Threshold analysis and smoothed curve fitting (Table S13 and Figure S2) revealed a significant U-shaped association between CHG index and all-cause mortality. The risk of all-cause mortality decreased when the CHG index value was below 5.12. However, when the CHG index value exceeded 5.12, the risk of all-cause mortality increased significantly. The likelihood ratio test for the model yielded an overall p-value < 0.001; (6) selecting LAP score ≥ 30.5 (males) or ≥ 23.0 (females) as diagnostic criterion for SLD using the NHANES database (Table S14); (7) excluding participants with BMI exceeding 75 (outliers) (Table S15); (8) association of CHG index with mortality risk among MASLD patients with T2DM (Table S16); (9) association of CHG index with mortality risk among MASLD patients with dyslipidemia (Table S17); (10) association of simplify-CHG index with mortality risk among MASLD patients (Table S18); (11) association of CHG index with mortality risk among MASLD patients before multiple imputation (Table S19); (12) association of CHG index with mortality risk among MASLD patients during different follow-up periods (Table S19). The results of the above 12 sensitivity analyses were consistently consistent with the main conclusions. These results confirm the robustness of the study findings.

Discussion

This is the first cohort study primarily based on a Chinese population to investigate the impact of the CHG index on all-cause mortality in individuals with MASLD. Utilizing prospective cohort data from 6936 MASLD patients in the CHARLS database, this study systematically evaluated the predictive value of the CHG index for all-cause mortality risk and its potential underlying mechanisms over a median follow-up period of 7.44 years. We found that after comprehensive adjustment for multiple confounding factors, an elevated CHG index was significantly and independently associated with an increased risk of all-cause mortality in MASLD patients. The highest tertile group exhibited a 49% higher mortality risk compared to the lowest group (HR = 1.49, 95% CI: 1.21–1.84). Kaplan-Meier survival curves further confirmed that higher CHG index levels were significantly associated with reduced survival rates (P < 0.0001). Dose-response relationship analysis revealed a non-linear association between the CHG index and mortality risk, suggesting the potential existence of specific thresholds for risk stratification. Mediation analysis indicated that CRP and TyG-WWI partially mediated the effect of the CHG index on mortality risk, with mediation proportions of 13.36% and 45.33%, respectively. This suggests that inflammation and IR-related metabolic disorders may be key pathophysiological pathways through which the CHG index influences adverse outcomes. Furthermore, this study developed prediction models incorporating eight ML algorithms. DCA confirmed that the Logistic Regression model provided significant clinical net benefit across a wide range of risk thresholds. Importantly, these core findings were successfully replicated in an external validation cohort from NHANES, further strengthening the robustness and generalizability of the study conclusions.

The findings of this study are generally consistent with previous research on the prognostic value of the CHG index, further validating its ability to predict adverse outcomes in populations with metabolic diseases. Specifically, studies by Zhu et al. [9] in MASLD populations based on the NHANES and HRS cohorts, Wei et al. [21] in patients with metabolic syndrome, and Guo et al. [7] in patients with calcific aortic valve stenosis (CAVS) have all confirmed that an elevated CHG index is significantly associated with an increased risk of all-cause and cardiovascular mortality. The present study, conducted within the CHARLS cohort, observed a 53% increased mortality risk (HR = 1.53) in the highest tertile compared to the lowest tertile, which aligns with the trends reported in the aforementioned studies. Furthermore, similar to the research by Zhu et al. [9], this study also employed mediation analysis to untangle the potential mediating role of inflammatory status and IR-related indicators in the association between the CHG index and mortality risk. Although the specific mediator variables selected differed, both findings point towards the metabolic disorders-inflammation pathway as a potential key biological mechanism through which the CHG index influences prognosis [22, 23]. These consistent findings across diverse populations and ethnicities strengthen the evidence level for the CHG index as a universal prognostic biomarker [24].

However, several noteworthy differences exist between the findings of this study and those of previous research. First, regarding the shape of the dose-response relationship, both Zhu et al. [9] and Wei et al. [21] reported a U-shaped nonlinear association between the CHG index and mortality risk, suggesting that both excessively low and high CHG index levels may increase the risk of death. In contrast, although a nonlinear association was observed in this study, it primarily manifested as a monotonic trend of increasing risk with higher CHG index levels, with no clear evidence of a loss of protective effect or risk reversal at low CHG index levels. This discrepancy may stem from differences in the characteristics of the study populations: the CHARLS cohort focuses on middle-aged and older adults, whose baseline nutritional status and metabolic phenotypes may fundamentally differ from the all-age population of NHANES or specific disease populations (such as CAVS) [25]. The adverse prognostic risk associated with low cholesterol or low blood glucose may be less apparent in an older population with relatively balanced nutrition. Second, this study innovatively constructed a prediction model based on ML algorithm and performed DCA, a methodological extension not addressed in prior studies, providing more direct evidence to support the clinical translation of the CHG index. Third, in the mediation analysis, this study employed composite indices (such as TyG-WWI) as proxy variables for IR and central obesity, differing from the weight-adjusted waist index (WWI) and estimated glucose disposal rate (eGDR) used by Zhu et al. [9]. This difference in indicator selection may reflect varying emphases among studies regarding the core metabolic phenotypes of MASLD. This study places greater emphasis on the impact of the interaction between the TyG axis and obesity, whereas Zhu et al. [9] focused more on central obesity alone and glucose disposal capacity. Finally, the effect size in this study (HR = 1.53) was slightly higher than the risk ratio reported by Zhu et al. [9] for Q4 vs. Q1 (HR ≈ 1.30). This may be related to the higher baseline metabolic risk profile of the CHARLS population, the longer follow-up period (7.44 years), or the increased between-group differences resulting from tertile grouping [26, 27]. It may also suggest that Asian MASLD populations are more susceptible to IR-related metabolic damage compared to American populations [26].

The CHG index, as a composite metabolic indicator integrating TC, FBG, and HDL-C, embodies a core biological significance that extends beyond the simple mathematical summation of its glycemic and lipid components. It represents a holistic quantification of glucolipotoxicity. This is the fundamental pathological basis for its superior accuracy in predicting adverse outcomes in patients with MASLD, compared to isolated hyperglycemia or isolated dyslipidemia. Isolated hyperglycemia or hypercholesterolemia merely reflects an abnormality in a single metabolic pathway. In contrast, an elevated CHG index signifies a synergistic disorder characterized by hypercholesterolemia, hyperglycemia, and HDL-C. This combination creates a pathological closed loop of “amplified glucolipotoxicity interaction”, whose damaging effects on target organs far exceed the simple additive effects of individual metabolic abnormalities [27, 28]. Within the MASLD population, this metabolic imbalance may contribute to increased mortality risk through multiple pathophysiological pathways [7]. Firstly, the coexistence of hypercholesterolemia and low HDL-C can directly promote atherosclerotic plaque formation and cardiovascular events [7, 29], while persistent hyperglycemia exacerbates vascular endothelial dysfunction and target organ damage via oxidative stress and the accumulation of advanced glycation end products [30]. These factors have been confirmed as primary causes of mortality in MASLD patients [31]. Secondly, the mediation analysis in this study indicated that systemic inflammation and IR-related metabolic disorders (TyG-WWI) mediated 13.36% and 45.33% of the effect, respectively. This finding aligns with previous research. Zhu et al. [9] demonstrated that WWI and eGDR act as mediators in this process. Chronic low-grade inflammation driven by central obesity and perihepatic fat deposition may be the key pathological links connecting the CHG index to adverse outcomes [32, 33]. Specifically, visceral adipose tissue releases pro-inflammatory cytokines that exacerbate hepatic IR and fibrosis while inducing systemic metabolic disorders, ultimately forming a vicious cycle within the liver-heart-metabolism axis [34–36]. Furthermore, although this study primarily observed a positive association between elevated CHG index and mortality risk, considering the U-shaped relationship reported in previous studies, extremely low CHG index levels may reflect a state of metabolic exhaustion associated with severe malnutrition or end-stage liver disease [13]. This metabolic depletion could similarly increase mortality risk through immunosuppression and diminished organ repair capacity [37]. The CHG index influences the long-term survival of MASLD patients through a complete pathological pathway: glucolipotoxicity-cascade activation of inflammation and IR-damage to hepatic and cardiovascular target organs. Compared to isolated glycemic or lipid parameters, or the binary diagnosis of diabetes mellitus or hyperlipidemia, the CHG index provides a more comprehensive reflection of the severity and pathological progression of metabolic disorders [38]. Early intervention targeting this pathway may help improve the prognosis of this population [15]. CHG index greater than 5.17 can be established as a high-risk threshold for adverse outcomes in patients with MASLD, identifying them for inclusion in a high-risk management cohort and warranting intensified dynamic monitoring throughout the disease course. Clinicians should consider initiating personalized interventions, including standardized lifestyle modification, precise lipid-lowering and glycemic control therapies. Additionally, based on the patient’s inflammatory status and degree of liver injury, targeted treatments such as anti-inflammatory and hepatoprotective therapy may be initiated as appropriate to reduce the risk of adverse clinical outcomes.

This study possesses the following strengths. First, the large-sample prospective cohort design based on CHARLS, with a median follow-up duration of 7.44 years, provides robust longitudinal evidence for assessing the association between the CHG index and long-term mortality risk in MASLD patients. Second, this study innovatively integrates ML algorithms to construct a prediction model, and its clinical net benefit is confirmed through DCA, offering direct methodological support for translating the CHG index from a research metric to clinical application. Third, mediation analysis systematically elucidates the mediating roles of inflammatory status and IR-related metabolic disorders in the association between the CHG index and mortality risk, deepening the understanding of the underlying pathophysiological mechanisms. Fourth, the core findings are replicated and confirmed in the independent external validation cohort from NHANES, enhancing the generalizability and reliability of the study conclusions across different ethnicities and geographical regions.

This study has several limitations that warrant consideration. First, as an observational study, despite multivariable adjustment, the possibility of residual confounding and reverse causality cannot be entirely ruled out, which limits the ability to infer causality. Second, the CHG index and all biochemical indicators were based on a single baseline measurement. The potential time-dependent effects of their long-term dynamic changes on mortality risk were not assessed. Future studies should employ repeated-measures designs to capture cumulative exposure effects. Third, the diagnosis of MASLD relied on non-invasive surrogate indices such as LAP and FLI. Although these indices have been reasonably validated in epidemiological research, they were not confirmed by liver biopsy (the diagnostic gold standard). This may introduce some degree of misclassification bias, particularly in the stratification of disease severity. Fourth, important factors such as genetic predisposition, specific dietary patterns, medication adherence, and MASLD disease progression stages were not comprehensively recorded in the database used. These factors represent potential sources of confounding that may not have been adequately controlled for.

Conclusion

The results of this study indicate that in individuals with MASLD, the CHG index is independently and positively associated with the risk of all-cause mortality. Considering the observational design and related limitations of this study, these findings provide preliminary evidence for the potential application of the CHG index in risk stratification among MASLD patients. However, its value as a routine monitoring indicator in clinical practice still requires further validation through prospective studies and health-economic evaluations.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MASLD

Metabolic dysfunction-associated steatotic liver disease

CHG index

Cholesterol, high-density lipoprotein, and glucose index

CRP

C-reactive protein

TyG-WWI

Triglyceride glucose-weight-adjusted waist index

TyG-WHtR

Triglyceride glucose-waist-to-height ratio

TyG

Triglyceride-glucose

TyG-BMI

Triglyceride glucose-body mass index

TyG-WC

Triglyceride glucose-waist circumference

IR

Insulin resistance

ML

Machine learning

CHARLS

China Health and Retirement Longitudinal Study

NHANES

National Health and Nutrition Examination Survey

HDL

High-density lipoprotein

LAP

Lipid accumulation product

SLD

Steatotic liver disease

FLI

Fatty liver index

ALT

Alanine aminotransferase

AST

Aspartate aminotransferase

BMI

Body mass index

TG

Triglyceride

GGT

γ-glutamyl transferase

WC

Waist circumference

FBG

Fasting blood glucose

HbA1c

Glycated hemoglobin A1c

DM

Diabetes mellitus

TC

Total cholesterol

PLT

Platelet count

WBC

White blood cell count

HGB

Hemoglobin

HCT

Hematocrit

MCV

Mean corpuscular volume

LDL-C

Low-density lipoprotein cholesterol

CKD

Chronic kidney disease

HR

Hazard ratio

CI

Confidence interval

VIF

Variance inflation factor

KNN

K-Nearest Neighbors

SVM

Support Vector Machine

ROC

Receiver operating characteristic

AUC

Area under the curve

DCA

Decision curve analysis

CAVS

Calcific aortic valve stenosis

WWI

Weight-adjusted waist index

eGDR

Estimated glucose disposal rate

eGFR

Estimated glomerular filtration rate

Author contributions

J.W., G.T., and Q.T. : Designed the research plan, collected and analysed the data, explained the research results, and wrote the initial and final drafts. R.L. and H.Z. contributed to data collection, statistical analysis, and result interpretation. S.H. was responsible for the data results visualization. X.S. and G.Z. supervised the research project, read and revised the manuscript. All authors read and approved the final version of the manuscript.

Funding

This work was supported by Shenzhen Fundamental Research Program (JCYJ20220530144404010 & JCYJ20220530144404011) and Futian Healthcare Research Project No. FTWS2023037, FTWS050, and FTJCYJ20220530144404011WS049; Futian District key specialty funding No. QZDZK-202413 and Outstanding Medical Innovation Talent Program of The Eighth Affiliated Hospital of Sun Yat-sen University No. YXYXCXRC202414 and Guangdong Foundation for Basic and Applied Research Enterprise Joint Fund No. 2023A1515220186.

Data availability

The datasets supporting the conclusions of this article are available at https://charls.pku.edu.cn/ and https://wwwn.cdc.gov/nchs/nhanes/Default.aspx.

Declarations

Ethics approval and consent to participate

The CHARLS study was conducted in line with the principles stated in the Declaration of Helsinki and received approval from the Institutional Review Board of Peking University (IRB00001052-11015). Before their involvement in the CHARLS study, all participants gave their written informed consent. The research design and implementation of NHANES follow the ethical principles outlined in the Helsinki Declaration. This study utilized data from the NHANES project that were publicly available and approved by the National Center for Health Statistics (NCHS) Ethics Review Board (ERB). NHANES requires informed consent from participants and ensures their privacy and information security. All of the data used in this study are deidentified, and most of the data are publicly available. The research adhered to the STROBE guidelines for reporting observational studies in epidemiology.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.