Abstract
Objectives
To assess long-term global trends in the burden of metabolic dysfunction-associated steatotic liver disease–related cirrhosis and develop a simplified, waist circumference–based prediction model for early identification of individuals at risk of advanced liver fibrosis.
Design
This study combined (i) ecological time-series analysis using Global Burden of Disease 2021 data (1990–2021) and (ii) cross-sectional analysis of population-based survey data (NHANES 2021–2023). A waist circumference–based logistic regression model was developed and validated to predict metabolic dysfunction-associated steatotic liver disease–related advanced fibrosis, with subgroup evaluation and model calibration.
Setting
Global analyses were based on Global Burden of Disease data across all World Health Organization regions and income groups. U.S.-based analyses used NHANES, a nationally representative health and nutrition survey, with data collected in community and outpatient settings across multiple states.
Participants
The Global Burden of Disease analysis included global population-level estimates from 204 countries. The NHANES component involved 11,933 adults aged ≥20 years, with subgroups defined by metabolic dysfunction-associated steatotic liver disease status, age, sex, and race/ethnicity. Exclusion criteria included viral hepatitis and excessive alcohol use.
Interventions
This was an observational study with no interventions. The waist circumference–based predictive model served as an analytic tool rather than a clinical intervention.
Primary and secondary outcome measures: The primary outcomes were temporal trends in disability-adjusted life years attributable to metabolic dysfunction-associated steatotic liver disease–related cirrhosis and the performance of a waist circumference–based prediction model for advanced fibrosis (≥9.5 kPa by transient elastography). Secondary outcomes included subgroup-specific model performance (area under the curve, calibration) and comparison of metabolic profiles between metabolic dysfunction-associated steatotic liver disease and non-metabolic dysfunction-associated steatotic liver disease participants.
Introduction
Metabolic dysfunction-associated steatotic liver disease (MASLD) has emerged as the most prevalent chronic liver disorder worldwide, paralleling the rise in obesity and metabolic syndrome.1,2 It encompasses a spectrum from simple steatosis to metabolic dysfunction-associated steatohepatitis (MASH), fibrosis, cirrhosis, and ultimately hepatocellular carcinoma.3,4 Among these outcomes, MASLD-related cirrhosis has emerged as a leading cause of liver-related morbidity and mortality, imposing substantial clinical and socioeconomic burdens.5,6 In line with the recent multisociety consensus nomenclature, nonalcoholic fatty liver disease (NAFLD) and nonalcoholic steatohepatitis (NASH) are termed MASLD and MASH, respectively. In the present study, the historical terms are retained where necessary to maintain consistency with source datasets and prior literature. 7
Global epidemiological estimates suggest that the prevalence of MASLD exceeds 25% of the adult population, with marked variation across regions and ethnic groups.8,9 However, despite its high prevalence, the long-term burden attributable to MASLD-related cirrhosis remains incompletely understood. Recent evidence indicates increasing disability-adjusted life years (DALYs) and premature deaths due to MASLD cirrhosis;10,11 however, systematic assessments of temporal trends and regional disparities are limited. Furthermore, the relative contribution of years of life lost (YLLs) versus years lived with disability (YLDs) has not been comprehensively quantified at the global scale. This gap is particularly important because cirrhosis represents the advanced and clinically most consequential stage of steatotic liver disease, and a clearer understanding of its burden trajectory is essential for prioritizing prevention, surveillance, and resource allocation. A population-level evaluation of DALY patterns may therefore provide important insight into whether the overall burden is driven predominantly by premature mortality or by long-term disability, with distinct implications for health policy.12,13
In parallel, MASLD has strong metabolic underpinnings, with obesity, insulin resistance, dyslipidemia, and chronic inflammation recognized as major risk factors.14,15 Identifying high-risk individuals remains challenging, as traditional anthropometric indicators such as body mass index (BMI) are insufficiently sensitive for MASLD detection. 16 Waist circumference and other metabolic markers have been suggested as better predictors, 17 but predictive models often suffer from complexity or lack of validation across diverse populations.18,19 Importantly, MASLD is increasingly regarded not merely as a liver-specific disorder but as a systemic manifestation of metabolic dysfunction. 20 In this context, central adiposity may be more informative than overall body mass, because it more closely reflects visceral fat accumulation and metabolic risk. Accordingly, simplified and accessible indicators such as waist circumference could offer practical value for early risk stratification, especially in large-scale screening settings or resource-limited clinical environments where laboratory- or imaging-based tools are less feasible. 21
To address these gaps, we integrated data from the Global Burden of Disease (GBD) 2021 study and the most recent National Health and Nutrition Examination Survey (NHANES) cycles. Using GBD, we quantified the temporal and regional burden of MASLD-related cirrhosis over three decades. 19 Using NHANES, we evaluated baseline metabolic differences, constructed prediction models, and assessed their generalizability across demographic subgroups. Our dual-track approach aimed to provide a comprehensive picture of both the population-level burden and individual-level risk stratification of MASLD, thereby informing public health policy and clinical decision-making. Specifically, we sought not only to characterize the magnitude and structure of the global burden of MASLD-related cirrhosis but also to determine whether a simplified model centered on readily obtainable anthropometric information could provide clinically meaningful discrimination for identifying individuals at increased risk.
Materials and methods
Ethics declaration
This study was based on secondary analyses of two publicly available and de-identified datasets: the GBD 2021 study and the U.S. NHANES 2017–2020 and 2021–2023. All methods were conducted in accordance with the Declaration of Helsinki and relevant institutional guidelines. Ethical approval and informed consent were not required, as both datasets were anonymized and publicly accessible.
Data sources
Two complementary datasets were used. The GBD 2021 dataset provided estimates of cirrhosis burden attributable to MASLD from 1990 to 2021, including DALYs, mortality, and age-standardized rates, stratified by sex and region. In addition, individual-level data were obtained from the 2021–2023 cycles of the NHANES, a nationally representative survey of the U.S. population.
GBD 2021 data are publicly available through the Global Health Data Exchange (GHDx) portal maintained by the Institute for Health Metrics and Evaluation (IHME): https://vizhub.healthdata.org/gbd-results/
NHANES data are publicly accessible via the U.S. Centers for Disease Control and Prevention (CDC) website: https://wwwn.cdc.gov/nchs/nhanes/
Study population
For the NHANES analysis, adults aged ≥20 years were included. Although adulthood is often defined as age ≥18 years in epidemiological studies, we restricted the analytic sample to individuals aged ≥20 years to align with the commonly used NHANES adult anthropometric reporting framework and better align the model for risk screening in the general adult population. Pregnant individuals and those with missing key variables were excluded. Survey weights, clustering, and stratification were incorporated in all analyses to ensure population representativeness.
Definition of MASLD
MASLD was defined using validated noninvasive indices, including the Fatty Liver Index (FLI), Hepatic Steatosis Index (HSI), and vibration-controlled transient elastography (VCTE) where available. Participants with significant alcohol intake (>30 g/day for men and >20 g/day for women) or evidence of viral hepatitis were excluded to minimize confounding.
Covariates
Demographic factors included age, sex, and race/ethnicity. Anthropometric variables included BMI and waist circumference. Metabolic measures included fasting glucose, glycated hemoglobin (HbA1c), triglycerides, high-density lipoprotein (HDL) cholesterol, and uric acid. Lifestyle factors such as alcohol consumption and physical activity were also recorded. All laboratory values were obtained according to standardized NHANES protocols.
Model development and interpretation
The full and simplified prediction models were developed using multivariable logistic regression. Variable importance was assessed using a permutation-based importance approach, and partial dependence plots were generated to illustrate the effects of key predictors on model-predicted risk. These importance measures were used to reflect the relative contribution of predictors to model performance within this analytical framework.
Statistical analysis
For GBD data, age-standardized DALY and mortality rates were calculated, and estimated annual percentage changes (EAPCs) were derived using log-linear regression. For NHANES, baseline differences between MASLD and non-MASLD participants were assessed using standardized mean differences (SMDs). Machine learning models based on gradient boosting were developed to predict MASLD. Variable importance rankings were derived, and partial dependence plots were generated to visualize nonlinear associations. Discrimination was assessed using the area under the receiver operating characteristic curve (AUC), calibration was evaluated using calibration slope and intercept, and clinical utility was tested using decision curve analysis. Subgroup analyses were performed by age, sex, and race/ethnicity. All statistical analyses were conducted using R (version 4.3.2). A two-sided P value <0.05 was considered statistically significant.
Results
Study population and analytical framework
The overall study design integrated two complementary data sources: the GBD 2021 dataset (1990–2021) and the most recent NHANES cycles (2021–2023) (Figure 1). From GBD, we extracted estimates of DALYs attributable to MASLD-related cirrhosis, stratified by region, sex, and time. From NHANES, we included adults aged ≥20 years after excluding pregnant individuals and those with missing key variables. Standardized definitions of MASLD were applied using validated indices (FLI, HSI, and VCTE), while survey weights were incorporated to ensure representativeness. Key metabolic, inflammatory, and lifestyle variables, including BMI, waist circumference, homeostasis model assessment of insulin resistance, high-sensitivity C-reactive protein, uric acid, physical activity, and dietary factors, were harmonized through imputation and weighting procedures. Participants were classified into MASLD and non-MASLD groups for further comparative analyses.

Study flowchart of the dual-track design combining GBD and NHANES data. GBD: Global Burden of Disease; NHANES: National Health and Nutrition Examination Survey; IHME: Institute for Health Metrics and Evaluation; MASLD: metabolic dysfunction-associated steatotic liver disease; DALYs: disability-adjusted life years; EAPC: estimated annual percentage change; SDI: Sociodemographic Index; BMI: body mass index; HOMA-IR: homeostasis model assessment of insulin resistance; hs-CRP: high-sensitivity C-reactive protein; US-FLI: United States Fatty Liver Index; VCTE: vibration-controlled transient elastography; WTMEC2YR: two-year MEC examination sample weight; SDMVPSU: masked variance pseudo-primary sampling unit; SDMVSTRA: masked variance pseudo-stratum.
Global and regional trends in MASLD-related cirrhosis burden
From 1990 to 2021, the age-standardized DALY rate attributable to MASLD-related cirrhosis increased substantially worldwide. Globally, DALYs rose from approximately 22.6 per 100,000 in 1990 to 33.9 per 100,000 in 2021, reflecting an average EAPC of +1.14% (Figure 2(a)). The U.S. demonstrated the steepest rise, from 28.9 to 44.4 per 100,000 (EAPC: +1.74%/year) (Figure 2(a)). In contrast, the increase in China was relatively modest, from 10.9 to 11.1 per 100,000 (EAPC: +0.08%/year), with trends remaining comparatively stable over the three decades (Figure 2(a)). Parallel to DALYs, mortality rates from MASLD-related cirrhosis escalated, although with notable geographic variation. In the U.S., the mortality rate nearly doubled from ∼0.9 to 1.6 per 100,000, while the global rate rose more gradually from 0.8 to 1.2 per 100,000 (Figure 2(b)). China displayed a slower upward trend, with mortality increasing from 0.3 to 0.4 per 100,000 during the same period (Figure 2(b)). The share of MASLD among total cirrhosis-related DALYs has grown consistently. In 1990, MASLD accounted for ∼1.5% of cirrhosis DALYs in China, rising to ∼3% by 2021 (Figure 2(c)). Globally, the fraction increased from 4.0% to 4.6%, while the U.S. exhibited the most pronounced growth, reaching >5% by 2021 (Figure 2(c)). These findings indicate a growing etiological contribution of MASLD to the cirrhosis burden, particularly in high-income settings. Decomposition of DALYs revealed that YLLs dominated the overall burden, far exceeding YLDs. Between 1990 and 2021, YLL rates attributable to MASLD-related cirrhosis increased substantially in both global and U.S. populations, whereas YLDs contributed only a minor fraction and remained relatively stable (Figure 2(d)). In China, the total DALY rate was consistently lower, with YLLs still constituting the major driver of disease burden.

Global and regional trends in MASLD-related cirrhosis burden, 1990–2021. MASLD: metabolic dysfunction-associated steatotic liver disease; DALYs: disability-adjusted life years; YLL: years of life lost; YLD: years lived with disability; NAFLD: nonalcoholic fatty liver disease.
Baseline characteristics of participants with and without MASLD
Compared with non-MASLD participants, individuals with MASLD displayed markedly adverse metabolic profiles (Figure 3). Obesity indices were substantially higher in the MASLD group, with both BMI and waist circumference showing the largest standardized mean differences (SMD >2.0, P < 1 × 10−4). Markers of glycemic control, including fasting glucose and HbA1c, were also significantly elevated in the MASLD group (P < 0.001), consistent with a higher prevalence of insulin resistance. Similarly, triglyceride levels were markedly increased in MASLD, whereas HDL cholesterol was significantly reduced (P < 0.001), reflecting a typical atherogenic dyslipidemia pattern.

Baseline differences between metabolic dysfunction-associated steatotic liver disease (MASLD) and non-MASLD participants.
In contrast, no significant difference was observed in the mean age (P = 0.826) or alcohol intake (P = 0.111), indicating that the observed disparities are unlikely attributable to demographic or alcohol-related factors. Taken together, these findings highlight the clustering of central obesity, impaired glucose homeostasis, and dyslipidemia among MASLD individuals, reinforcing its strong metabolic underpinning.
Model performance for MASLD prediction
We compared the predictive performance of the full multivariable model with a simplified model containing a reduced set of predictors (Figure 4). Both models demonstrated nearly identical discrimination ability, with AUC values close to 0.95 in cross-validation (Figure 4(a) and (b)). Calibration analysis indicated strong agreement between predicted and observed probabilities in both models, as reflected by calibration plots closely aligned with the diagonal line (Figure 4(c)). The corresponding calibration slope and intercept values were near unity and zero, respectively, confirming minimal overfitting.

Model performance comparison between the full and simplified prediction models for metabolic dysfunction-associated steatotic liver disease (MASLD). (a–b) Discrimination ability of the full and simplified models, both achieving high area under the curve values (∼0.95). (c) Calibration plots showing close concordance between predicted and observed risks. (d) Decision curve analysis indicating similar net clinical benefit across threshold ranges.
Decision curve analysis further revealed comparable net benefit between the two approaches across a wide range of threshold probabilities (Figure 4(d)). Collectively, these findings suggest that the simplified model retains equivalent predictive accuracy while reducing complexity, thereby offering a more efficient tool for MASLD risk stratification.
Variable importance and nonlinear effects on MASLD risk
Machine learning–based variable importance analysis identified waist circumference as the single most influential predictor of MASLD, far outweighing other factors. Sex, age, and race also contributed to risk discrimination, although to a much lesser extent (Figure 5(a)). Traditional metabolic indicators such as HDL cholesterol, triglycerides, fasting glucose, and HbA1c ranked lower in relative importance, while alcohol consumption exerted minimal influence.

Machine learning analysis of variable importance and nonlinear associations with MASLD. (a) Global importance ranking of predictors in the classification model, showing waist circumference as the strongest determinant. (b–e) PDPs illustrating the marginal effects of waist circumference, fasting glucose, HbA1c, and BMI on predicted MASLD risk. MASLD: metabolic dysfunction-associated steatotic liver disease; PDP: partial dependence plot; FPG, fasting plasma glucose; HbA1c: hemoglobin A1c; BMI: body mass index; HDL: high-density lipoprotein cholesterol.
The partial dependence analysis provided further insights into nonlinear associations. A sharp, dose-dependent increase in MASLD risk was observed with waist circumference, with the predicted probability rising steeply between 90 and 110 cm before plateauing (Figure 5(b)). In contrast, fasting glucose (Figure 5(c)) and HbA1c (Figure 5(d)) showed weak inverse associations, although effect sizes were small and unlikely to be clinically meaningful. Interestingly, BMI demonstrated an almost flat relationship with predicted MASLD risk (Figure 5(e)), indicating that central adiposity rather than overall body mass plays a dominant role in driving disease risk.
Collectively, these results emphasize the importance of waist circumference as a determinant of MASLD, highlighting the clinical relevance of central obesity beyond BMI.
Subgroup performance of the MASLD prediction model
The simplified prediction model demonstrated robust discrimination across demographic subgroups, with AUC values consistently close to or exceeding 0.95 (Figure 6). Performance was highly stable across age categories, with AUCs of 0.963 for participants aged <50 years and 0.965 for those aged ≥50 years. Sex-specific analyses showed no appreciable difference, with AUCs of 0.964 in men and 0.966 in women.

Subgroup performance of the simplified metabolic dysfunction-associated steatotic liver disease prediction model.
Across racial and ethnic groups, model performance remained uniformly high. Non-Hispanic White (AUC = 0.964) and non-Hispanic Black (AUC = 0.963) groups showed nearly identical discrimination. Mexican American (AUC = 0.962), Non-Hispanic Asian (AUC = 0.959), and Other Hispanic (AUC = 0.957) participants also achieved excellent predictive accuracy, although slightly lower than in the larger racial subgroups, likely reflecting smaller sample sizes. The “Other race” group similarly maintained strong performance (AUC = 0.960).
These findings underscore the generalizability and equity of the simplified model across major demographic subgroups, suggesting minimal bias related to age, sex, or race/ethnicity.
Subgroup calibration of the prediction model
Calibration analysis across demographic subgroups confirmed that predicted probabilities from the simplified model were well aligned with observed outcomes (Figure 7(a)). For both age categories (<50 and ≥50 years) and sex (male and female), the calibration curves closely followed the diagonal line, indicating reliable probability estimates across risk strata.

Calibration performance of the simplified metabolic dysfunction-associated steatotic liver disease prediction model across subgroups and risk reduction heatmap.
Similarly, subgroup calibration was consistent across racial and ethnic groups, including non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, Mexican American, other Hispanic, and other race participants. Minor deviations were observed at the extremes of predicted probabilities, particularly in smaller subgroups such as “other Hispanic” and “other race” (Figure 7(a)), likely due to sample size limitations. Nevertheless, overall calibration remained satisfactory, supporting the model’s applicability across diverse populations.
In the risk reduction heatmap (Figure 7(b)), waist circumference showed a pronounced effect on lowering MASLD risk. Incremental reductions in waist circumference were associated with progressively larger decreases in predicted risk, with the delta risk reaching approximately 0.3 at a relative reduction of 0.2. These findings underscore waist circumference as the dominant modifiable factor for MASLD risk mitigation.
Discussion
This study provides an integrated assessment of the global burden of MASLD-related cirrhosis and its metabolic underpinnings using complementary evidence from the GBD dataset and NHANES survey. We observed a substantial increase in DALYs and mortality rates attributable to MASLD-related cirrhosis over the past three decades, with the most rapid growth occurring in the United States, while China exhibited comparatively modest changes. Decomposition analyses demonstrated that years of life lost contributed disproportionately to the total burden, indicating that premature mortality, rather than disability, remains the predominant driver of disease impact. These findings underscore the urgent need for early identification and intervention to prevent MASLD progression before cirrhosis develops.
At the individual level, MASLD was characterized by profound metabolic disturbances, including significantly higher body mass index, waist circumference, fasting glucose, HbA1c, and triglycerides, alongside lower HDL cholesterol.19,22 Although these findings are consistent with established knowledge, they provide useful clinical context for interpreting the prediction analyses, particularly the prominent role of waist circumference in the simplified model. Moreover, the absence of significant differences in age and alcohol consumption between MASLD and non-MASLD groups further highlights the metabolic nature of the disease, distinguishing it from traditional alcohol-related liver injury. 23 These results are consistent with the concept of MASLD as a systemic manifestation of metabolic dysfunction24,25 and support the prioritization of metabolic health optimization in its prevention and management. 26
In developing predictive models, we found that a simplified approach centered on waist circumference achieved excellent discrimination and calibration, with performance comparable to that of a more complex full model. The model was highly robust across age, sex, and racial or ethnic subgroups, with area under the curve values consistently near 0.95. Importantly, waist circumference emerged as the most influential predictor, while body mass index contributed little independent predictive value. This observation reinforces the growing recognition that central adiposity, rather than overall body mass, is the dominant risk factor for hepatic steatosis and progression to advanced disease. The apparent inverse patterns of fasting plasma glucose and HbA1c in the partial dependence plots should be interpreted cautiously. These plots reflect model-based marginal effects rather than crude associations. After conditioning on dominant predictors such as waist circumference and other correlated metabolic features, the marginal effects of glucose-related variables became attenuated and slightly non-monotonic. Therefore, this pattern is more likely to reflect predictor interdependence and model structure rather than a true inverse biological relationship. Such evidence supports a paradigm shift toward incorporating waist circumference into standard risk assessments for MASLD in both clinical and public health contexts. Taken together, these findings establish the waist circumference-based predictive model as a major contribution of the present study.
The strengths of our study include the integration of global burden estimates with individual-level data, providing a comprehensive perspective on MASLD from both population and clinical viewpoints. The use of machine learning methods enhanced the robustness of the prediction models, 27 and the consistent performance across subgroups suggests wide generalizability. Nonetheless, several limitations should be acknowledged. The present study included only participants aged ≥20 years. Although this choice helped align the analysis with the commonly used NHANES adult anthropometric reporting framework and simplified the interpretation of the model in a general adult setting, it may reduce comparability with studies that define adulthood as 18 years and older. MASLD in NHANES was defined using noninvasive surrogate indices rather than histological confirmation, which may introduce some degree of misclassification. 28 GBD estimates are based on modeled data and may not fully capture regional heterogeneity, particularly in countries with limited epidemiological surveillance. 29 In addition, external validation in prospective cohorts outside NHANES is still needed to establish broader applicability. 30 External validation was not performed in the present study because an independent dataset with sufficiently comparable disease definitions, outcome assessment, and covariate structure was not available. Therefore, although the model showed good performance in internal evaluation, its generalizability should be interpreted with caution and requires further validation in independent cohorts.
Taken together, our findings highlight the dual challenges posed by MASLD-related cirrhosis at both the population and individual levels. The rising global burden, dominated by premature mortality, necessitates urgent preventive strategies. Furthermore, the identification of waist circumference as a powerful and parsimonious predictor of MASLD provides a practical tool for risk stratification. Future efforts should focus on validating simplified prediction models in diverse healthcare settings, improving access to noninvasive diagnostics, and developing targeted interventions to mitigate the metabolic drivers of MASLD progression.
Footnotes
Acknowledgements
We thank the Institute for Health Metrics and Evaluation for access to the Global Burden of Disease 2021 data and the U.S. CDC/NCHS for the NHANES datasets. We are grateful to colleagues at the Basic Medicine Research Innovation Center for Cardiometabolic Diseases, Ministry of Education, Southwest Medical University, for valuable discussions and technical support.
Author contributions
Xuli Wang contributed to the study design, data acquisition, statistical analysis, figure preparation, and manuscript drafting. Bingwei Wang supervised the study, guided the interpretation of results, reviewed and revised the manuscript critically for important intellectual content, and provided project administration and technical support. Both authors read and approved the final manuscript.
Declaration of conflicting interests
The authors declare that they have no competing interests.
Competing interests
The authors declare that they have no competing interests.
Funding
This work was supported by grants from the Science and Technology Strategic Cooperation Project of Southwest Medical University (Grant No. 2024PZXNYD01 to BW), Science and Technology Strategic Cooperation Programs of Luzhou Municipal People's Government and the Southwest Medical University (Grant No. 2024LZXNYDJ088 to BW), Southwest Medical University Technology Program (Grant No. 2024ZKZ014 to BW), and Research Start-up Foundation of Southwest Medical University (Grant No. 00170071 to BW).
