ABSTRACT
-
Background/Aims
Liver stiffness measurement (LSM) is a key tool for risk stratification in metabolic dysfunction-associated steatotic liver disease (MASLD), yet static thresholds fail to capture dynamic transition across risk strata. We aimed to characterize LSM-risk transitions and develop a time-updated, individualized model for predicting state transitions, liver-related events and death (LREs/death).
-
Methods
In a real-world MASLD cohort, we applied a multi-state, time-homogeneous Markov model to quantify annual transition probabilities and mean state occupancy times across LSM-defined low-, intermediate-, and high-risk strata. A Markov model incorporating age, sex, type 2 diabetes (T2D), hypertension was used to generate individualized, time-updated risk trajectories and probabilities of LREs/death.
-
Results
Among 11,514 MASLD individuals with ≥2 vibration-controlled transient elastography assessments, the low-risk category demonstrated notable stability, with 92% remaining unchanged at 1 year and a mean occupancy time of 8.43 years (95% confidence interval [CI] 7.94–8.95). Contrarily, the intermediate-risk category was highly dynamic, with only 39% remaining unchanged after 1 year and a mean occupancy time of 0.92 years (95% CI 0.88–0.96). T2D, hypertension, and obesity substantially shorten low-risk occupancy time, whereas antidiabetic medication was associated with more favorable transitions. Finally, we developed a dynamic, multi-state Markov model integrating longitudinal LSM-defined risk states with relevant covariates to generate individualized predictions of state transitions and risks of LREs/death.
-
Conclusions
LSM-based strata in MASLD represent distinct and meaningful dynamic trajectories. In particular, the marked instability of the intermediate-risk state supports more frequent reassessment. By quantifying transition pathways and time-updated risks of LREs/death, this model may inform the personalized surveillance intervals and risk-adapted management.
-
Keywords: Vibration-controlled transient elastography; Liver stiffness measurement; Metabolic dysfunction-associated steatotic liver disease; Markov model
Study Highlights
• In MASLD, liver stiffness measurement (LSM) offers a practical alternative for fibrosis assessment and risk stratification. A Markov model revealed high stability in the LSM-defined low-risk category but pronounced dynamism in intermediate-risk individuals. Type 2 diabetes, hypertension, and obesity most strongly influenced state transitions. These findings support guideline recommendations to reassess low-risk patients every 2–3 years and suggest shorter 1–2 years intervals for intermediate-risk cases, with metabolic comorbidities prompting closer monitoring. Finally, our DYNAMO model, incorporating LSM status and clinical covariates, enables personalized surveillance.
Graphical Abstract
INTRODUCTION
Metabolic dysfunction-associated steatotic liver disease (MASLD) imposes a heavy global health burden, affecting approximately 38% of the adult population worldwide with risk of progressing to metabolic dysfunction-associated steatohepatitis (MASH), liver fibrosis, cirrhosis, and ultimately liver-related events (LREs) [
1–
6]. Given its high prevalence, long sub-clinical course, and potential for progression, careful longitudinal monitoring of liver health in individuals with MASLD is essential.
A precise understanding of the natural history of MASLD is critical for informing individualized surveillance strategies and optimizing follow-up intervals. Biopsy-based studies examining the natural history of liver fibrosis, which is the strongest predictor of clinical adverse events [
7], have reported a pooled fibrosis progression rate of 0.07–0.14 stages per year, corresponding to an estimated state occupancy time of approximately 7–14 years for MASLD patients without baseline fibrosis, depending on the presence or absence of underlying MASH [
8]. However, real-world evidence indicates that liver biopsy is performed in only ~4% of patients with MASLD, even in large healthcare databases encompassing more than 500,000 individuals [
9]. Moreover, patients undergoing biopsy are typically selected because of suspected advanced disease, rendering biopsy-derived estimates highly susceptible to selection bias and limiting their generalizability to the broader MASLD population [
9].
Non-invasive liver stiffness measurement (LSM) via vibration-controlled transient elastography (VCTE) has therefore emerged as a pragmatic alternative for fibrosis assessment in routine clinical practice. LSM is now widely implemented for risk stratification and has recently been proposed as a reasonably likely surrogate endpoint in MASLD drug development programs [
10–
12]. A commonly adopted three-tiered LSM-based classification system broadly parallels histological fibrosis staging, with LSM values <8 kPa generally excluding significant fibrosis and values >12 kPa strongly suggestive of advanced fibrosis [
13–
16]. Based on these LSM-defined risk strata, current guidelines from the American Gastroenterological Association (AGA) recommend reassessment every 2–3 years for individuals at low or intermediate risk, and specialist referral for those at high risk [
13].
Despite the central role of LSM in contemporary MASLD management, the longitudinal dynamics of transitions between LSM-defined risk categories remain poorly characterized. Characterizing these transitions using real-world, non-invasive measurements allows for a more representative reconstruction of MASLD natural history than estimates derived from biopsy-based cohorts, and provides evidence for rational, risk-adapted reassessment intervals. Importantly, there is a lack of models to dynamically predict individual-level transitions across LSM risk strata over time or link these evolving trajectories to the subsequent risk of adverse outcomes, which hinder personalized surveillance and management.
To address these gaps, we applied a dynamic, multistate Markov modelling framework to a large real-world cohort of individuals with MASLD undergoing serial VCTE assessments. This approach enables quantification of annual transition probabilities between LSM-based risk categories, estimation of state-specific occupancy times, and evaluation of the impact of key clinical characteristics on disease trajectories. Building on this framework, we further developed the dynamic, multi-state Markov model (DYNAMO), which integrates clinical variables to provide individualized predictions of LSM risk strata trajectory and the subsequent risk of hard clinical outcome, thereby supporting risk-adapted surveillance and prognostic stratification.
MATERIALS AND METHODS
Ethics statements
The study adhered to the Declaration of Helsinki and was approved by the ethics committees of each participating center. The ethics approval number for each center is as follows: Virginia Commonwealth University (No. RV00024398), University of Yokohama (No. 2022-0315), University of Torino (No. 2018/CS2/880/78404), Royal Free Hospital (IRAS 254793, REC 07/Q0501/50), Singapore General Hospital (No. 2022/2313), Karolinska University Hospital (No. 2018/880-31), Angers University Hospital (No. CB2010-01), Hospital Puerta de Hierro (No PI 205/25), University of Seville (No. 0156-N-23), First Affiliated Hospital of Wenzhou Medical University (No. KY2016-246), University of Birmingham (No. 13/WA/0385), University of Palermo (“Comitato Etico Palermo 1” ID 2014), Yonsei University School of Medicine (No. 4-2023-1687), University of Bordeaux (No. 1629487), Chinese University of Hong Kong (No. 2022.255), University of Malaya (No. 20168124134).
Participants
As described elsewhere, the international VCTEPrognosis cohort enrolled patients with MASLD from 16 centers across the United States, Europe, and Asia [
17–
19]. MASLD was diagnosed based on liver biopsy, imaging, or VCTE. Participants were excluded if they had chronic viral hepatitis, significant alcohol consumption (>30 g/day for men or >20 g/day for women), other causes of steatotic liver disease, or a history of hepatocellular carcinoma (HCC), hepatic decompensation, solid organ transplantation, or extrahepatic malignancy prior to enrollment. For the present analysis, eligible participants were required to be ≥18 years old, free of LREs at baseline, have a follow-up duration of at least 6 months, and underwent a minimum of two VCTE measurements (
Fig. 1A). The study was conducted in accordance with the Declaration of Helsinki and approved by the institutional review boards of all participating centers; written informed consent was waived.
Demographic, clinical, and laboratory data, along with VCTE parameters, were collected at each study visit. VCTE assessments, including LSM and controlled attenuation parameter, were performed using FibroScan® (Echosens, France) by trained personnel following standardized protocols. A valid VCTE examination required at least 10 measurements with an interquartile range-to-median ratio (IQR/M) ≤30%. Participants were followed from the date of their baseline VCTE assessment until the occurrence of an LRE, death, or the date of last clinical follow-up, whichever occurred first.
Definition of states and transition pathways in the Markov model
Four states were defined in the Markov model, including three LSM-based risk states according to recent guidelines: low risk (LSM <8 kPa), intermediate risk (LSM 8–12 kPa), and high risk (LSM >12 kPa) [
13,
14], and one absorbing state of LREs or death. LREs were defined as a composite of hepatic decompensation, HCC, liver transplantation and liver-related mortality. Since VCTE measurements are unreliable following LREs and unavailable after death, and to avoid overcomplicating the model, these two outcomes were combined into a single absorbing state. The constructed transition pathways are illustrated in
Figure 1B. Bi-directional transitions between adjacent LSM-based states were permitted, whereas no transitions occurred once the absorbing state was reached.
Categorical variables were summarized as counts and percentages, and continuous variables as medians with IQRs. Comparisons among low-, intermediate-, and high-risk categories were performed using the Kruskal–Wallis test or χ2 test, as appropriate.
To model disease trajectories and transitions between LSM-defined states, a time-homogeneous continuous-time Markov model was employed with the assumption that transition rates depend solely on the current state (details provided in the
Supplementary Methods). From the Markov model, we estimated annual transition probabilities and mean occupancy times based on the transition intensity matrix, which contained the instantaneous rate of each transition pathway in
Figure 1B. To validate the occupancy times derived from the Markov model, Kaplan–Meier curves were used to depict the timing of the first transition to the high-risk category among participants initially classified as low- or intermediate-risk, followed until either the last VCTE measurement or the first observation of high-risk status. Cumulative incidences of high-risk status at time points of 1, 2, 3 and 5 years were also assessed to inform optimal screening intervals for participants initially at low or intermediate risk.
We further evaluated the impact of clinically relevant covariates on state transitions, including age (<55, ≥55), sex, type 2 diabetes (T2D), hypertension, and obesity (defined by body mass index [BMI] ≥25 kg/m
2 for Asians and ≥30 kg/m
2 for non-Asians). In addition, antidiabetic medications with reported hepatoprotective effects, including sodium-glucose cotransporter 2 inhibitors (SGLT-2i) and glucagon-like peptide-1 receptor agonists (GLP-1RAs), were also examined [
20,
21]. Among covariates, age group, obesity state, and the use of SGLT-2i and GLP-1RAs were modeled as time-updated variables, with values updated at each follow-up visit.
Several sensitivity analyses were performed: (i) restricting the interval between tests to >6 months to ensure sufficient time for LSM change; (ii) including only participants with at least three LSM measurements to minimize bias from measurement frequency; (iii) redefining low-, intermediate-, and high-risk categories using alternative cutoffs of 10 and 15 kPa; (v) restricting the analysis to individuals with a follow-up duration of more than two years to exclude early events and minimize potential misclassification or measurement error associated with VCTE; (iv) requiring a relative LSM change >20% to classify a stage transition, to account for minor fluctuations that cross category thresholds (e.g., 7.9 kPa to 8.1 kPa); (vi) reconstructing the transition pathways by either modeling LREs as the sole absorbing state, or treating non-liver-related mortality as a distinct competing absorbing state.
Moving beyond discrete and static approaches such as multinomial regression, we developed DYNAMO, a dynamic, multi-state Markov model that provides a framework for MASLD disease evolution. By leveraging all available longitudinal VCTE assessments and time-updating covariates, DYNAMO enables simultaneous forecasting of LSM risk strata evolution and LREs/death across any specified time horizon. For model development, the cohort was randomly partitioned into training and testing sets in a 7:3 ratio. Candidate demographic, clinical, and laboratory variables, including those updated at each visit, were evaluated within the training set. A two-step variable selection strategy was applied. Firstly, associations with longitudinal LSM trajectories were assessed using linear mixed-effects models. Secondly, associations with LREs/death were evaluated using Cox proportional hazards regression models. Only variables demonstrating significant and consistent associations across both steps were retained in the final model. Given the low incidences of LREs/death, we evaluated the model performance for LREs/death using the 5-year area under the time-dependent precision-recall curves (AUPRC), a metric better suited to rare events. Model performance was also evaluated using the area under the receiver operating characteristic curve (AUROC), the ratio of the AUPRC to the event rate, and decision curve analysis (DCA). To further assess the model’s ability to capture transitions across LSM risk strata, we compared the observed and predicted state occupancy across all states at the final time point. Model adequacy was evaluated using a χ2 test, providing a population-level assessment of agreement between observed and predicted state distributions.
All statistical analyses were conducted using R (version 4.3.1). Two-sided P-values <0.05 were considered statistically significant.
RESULT
Participant characteristics
A total of 11,514 adult participants with at least two VCTE assessments between February 2004 and January 2023 were included in the analysis (
Fig. 1A). Based on baseline LSM values, 8,372 (72.7%), 1,825 (15.9%), and 1,317 (11.4%) individuals were categorized into the low-, intermediate-, and high-risk groups, respectively. As summarized in
Table 1, age, BMI, aspartate aminotransferase, fasting glucose, glycated hemoglobin, and the prevalence of T2D, hypertension, MASH, and at-risk MASH increased progressively across ascending LSM-based risk strata (all
P<0.001), whereas platelet count and the proportions of male and Asian participants demonstrated an inverse trend (all
P<0.001).
Transition counts across LSM-based risk states are summarized in
Supplementary Table 1. In total, 1,208 and 217 transitions were observed from low to intermediate and high risk; 1,798 and 553 from intermediate to low and high risk; and 290 and 668 transitions from high to low and intermediate risk, respectively. The transition intensity matrix reflected the instant likelihood of switching between risk states (
Supplementary Table 2). As expected, the intensity of transition to the absorbing state of LREs/death increased stepwise across risk categories (low vs. intermediate vs. high: 0.002 vs. 0.004 vs. 0.04). Notably, among individuals classified as intermediate risk, the transition intensity toward regression to low risk was 2.38-fold greater than that of progression to high risk (0.76 vs. 0.32).
Yearly transition probabilities are illustrated in
Figure 2. Of individuals currently categorized as low risk, 92% were predicted to remain in the same state after one year, while 7% and 1% were estimated to transition to intermediate and high risk, respectively. In contrast, the intermediate-risk group was notably dynamic: only 39% were expected to remain stable after one year, whereas 45% would regress to low risk and 16% would progress to high risk. Among those in the high-risk category, the probability of transitioning to the LREs/death state over one year was highest at 4%, with recovery to low and intermediate risk observed in 9% and 20% of individuals, respectively.
In the overall cohort, the estimated state occupancy time for the low-risk category was 8.43 years (95% confidence interval [CI] 7.94–8.95) (
Fig. 3A). In contrast, the high-risk category demonstrated a markedly reduced occupancy time of 2.27 years (95% CI 2.12–2.43), while the intermediate-risk category exhibited the shortest duration, with an estimated occupancy time of only 0.92 years (95% CI 0.88–0.96), reflecting rapid cycling and substantial instability within this state. Given that the shorter occupancy time in the intermediate-risk state could potentially be explained by its greater number of permissible transitions (i.e., bidirectional movements) compared with the low- and high-risk states, we conducted two counterfactual simulations to assess the robustness of this observation: (1) disabling the backward transition from the intermediate state to the low state; and (2) disabling the forward transition from the intermediate state to the high state. In the first scenario, the estimated occupancy time for the intermediate state was 3.07 years, which remained substantially shorter than the 8.43 years observed for the low state under the same forward-transition constraint. In the second scenario, the intermediate state exhibited an occupancy time of 1.30 years, still markedly lower than that of the high state. These findings indicate that the relative instability of the intermediate state persists even after accounting for differences in the number of available transition pathways.
Kaplan-Meier analysis confirmed a significantly higher probability of transitioning to high risk among individuals initially categorized as intermediate risk compared with those at low risk (
P<0.001,
Fig. 3B). Among participants starting in the low-risk category, 182 progressed to high risk, with a median transition interval of 28.60 months, and cumulative incidences of 0.48%, 1.17%, 1.85%, and 3.76% at 1, 2, 3, and 5 years, respectively (
Supplementary Table 3). In contrast, individuals with an initial intermediate-risk classification demonstrated substantially higher cumulative probabilities of progressing to high risk: 5.40%, 12.58%, 18.16%, and 28.45% at 1, 2, 3, and 5 years, respectively, with a shorter median transition interval of 21.33 months for 357 patients who progressed to high risk.
Older age (>55 years), T2D, hypertension and obesity were consistently associated with a higher yearly probability of liver stiffness progression and a lower likelihood of regression, whereas sex-related differences were minimal (
Supplementary Figs. 1–
5). The greatest between-group variability was observed in the transition from intermediate to low risk: 49% vs. 40% for age <55 vs. ≥55, 51% vs. 39% for individuals without vs. with T2D, and 51% vs. 39% for those without vs. with hypertension, while the transition from low- to intermediate-risk was most pronounced according to obesity status, with a higher transition probability observed in obese compared with non-obese individuals (14% vs. 5%). Among individuals with T2D, SGLT-2i use was associated with more favorable transition patterns, including a higher probability of regression from intermediate to low risk (37% vs. 42% for those not using vs. using SGLT-2i). In contrast, GLP-1 receptor agonists were predominantly linked to enhanced reverse transitions from the high-risk state, including regression to low risk (7% vs. 10%) and to intermediate risk (20% vs. 25%), as shown in
Supplementary Figures 6,
7.
The effects of these factors were further quantified using transition-specific hazard ratios (
Supplementary Table 4). Older age, T2D, and hypertension were each associated with reduced probabilities of regression and an increased risk of progression to LREs/death. Notably, both T2D and hypertension significantly increased the likelihood of upward risk reclassification, conferring 1.32–1.39-fold higher risk for progression from low to intermediate risk and 1.20– 1.40-fold higher risk for transition from intermediate to high risk. Obesity was also significantly associated with disease progression, with effect estimates ranging from 1.36–3.20-fold across different transitions.
Differences in state occupancy time were primarily driven by variation in the duration spent within the low-risk category, with T2D, hypertension and obesity having the greatest impact (
Table 2). Individuals without T2D remained in the low-risk category for an estimated 9.07 years (95% CI 8.38–9.82), which shortened to 6.81 years (95% CI 6.19– 7.49) in those with T2D. A similar attenuation was observed for hypertension, with estimated low-risk occupancy times of 9.32 years and 6.64 years in those without and with hypertension, respectively. Obese patients with MASLD had a significantly shorter low-risk occupancy time (3.58 years) compared with those without obesity (11.07 years).
A series of sensitivity analyses was performed to assess the robustness of our results. Restricting the test interval to over 6 months, including only participants with at least three LSM measurements, or redefining risk categories using cutoffs of 10 kPa and 15 kPa all yielded similar occupancy time and yearly transition probabilities (
Supplementary Figs. 8–
10). A sensitivity analysis restricted to participants with more than two years of follow-up, thereby excluding early events, confirmed the robustness of the main findings (
Supplementary Fig. 11). When considering state transitions only if changes exceeded 20% simultaneously, the results were identical to those of the main analyses (
Supplementary Fig. 12). Besides, models that separated LREs and non-liver-related mortality into distinct absorbing states, as well as models including only LREs as the absorbing state, showed similar estimates and predictive performance (
Supplementary Figs. 13,
14).
To further enable individualized prediction of risk category transitions and the probability of LREs or death, we developed a DYNAMO that integrated liver stiffness—defined risk categories with clinically relevant covariates. Using a two-step selection strategy based on linear mixed-effects models for longitudinal LSM trajectories and Cox proportional hazards models for clinical outcomes, age, sex, BMI, T2D, hypertension, and platelet count were retained as model covariates (
Supplementary Table 5). The 5-year AUPRC was 0.18 (95% CI 0.12–0.28) in the training set and 0.15 (95% CI 0.09–0.29) in the testing set, corresponding to 9.96- and 8.15-fold improvements over the baseline event rate, respectively. Consistent findings were observed for AUROC and DCA (
Supplementary Table 6;
Supplementary Fig. 15). Furthermore, the model showed high concordance between observed and predicted state occupancy across all states at the final time point, as indicated by a non-significant χ
2 test (
Supplementary Table 7).
Using DYNAMO, for example, a 50-year-old MASLD patient without T2D or hypertension, starting from an intermediate LSM risk state, with a BMI of 24 kg/m
2 and platelet count of 200 (10
9/L) would have a 5-year probability of LREs/death of 2.42%, with probabilities of 10.24% for remaining in the intermediate-risk state, 76.30% for transitioning to low risk, and 11.03% for progressing to high risk (
Fig. 4A). If the patient were re-evaluated by VCTE after one year and found to have shifted to the low-risk category, the model predicts an 89.40% probability of staying in the low-risk state at 5 years (
Fig. 4B). Conversely, if the patient progressed to the high-risk category, the 5-year predicted probability of LREs/death would rise to 5.35%, with only a 43.49% likelihood of reverting to low risk (
Fig. 4C). Furthermore, the model would update predictions based on changes in clinical factors such as the onset of T2D and BMI increasing to 27 kg/m
2 after one year. In this case, the risk of LREs or death at 5 years would increase to 2.65%, and the probability of reverting to low risk would decrease to 61.32% (
Fig. 4D). The DYNAMO predictions could be accessed through a dedicated webpage at
https://models-webpage.shinyapps.io/dynamo/.
DISCUSSION
In this multicenter longitudinal cohort encompassing a large population of individuals with MASLD, we delineated for the first time the dynamic transitions across LSM-based risk categories derived from VCTE. Our findings highlight substantial stability within the low-risk category, where 92% of participants remained unchanged over one year and the mean occupancy time exceeded 8 years. In contrast, transitions occurred far more rapidly among individuals in intermediate and high-risk categories. Notably, the intermediate-risk stratum demonstrated marked instability, with only 39% remaining in the same category at one year and a mean occupancy time of less than one year. T2D and hypertension emerged as the strongest determinants for transition patterns, markedly shortening occupancy time in the low-risk state and reshaping transition pathways. Building on these transition dynamics, the DYNAMO multi-state Markov model, which integrates the current LSM-defined state with patient-specific clinical characteristics, dynamically estimates future risk-state evolution and LREs over time.
Although few studies have explicitly examined transitions between VCTE-derived LSM categories, our findings are highly consistent with prior work on the natural history and histological evolution of MASLD. Meta-analyses of paired-biopsy cohorts have estimated annual fibrosis progression rates of 0.07–0.14 stages for individuals with baseline F0, equating to 7–14 years before a one-stage progression [
8]. Another comprehensive meta-analysis, integrating both biopsy and imaging-based studies, similarly reported that 59–82% of those with F0 remained stable, with 7% per year progressing predominantly to F1–F2 [
22]. This corresponds to an occupancy time of 8.4–11.6 years at F0 before progression. Our observation of an 8.43-year occupancy time in the low-risk LSM stratum aligns closely with these histology-derived trajectories. Such concordance reinforces the validity of VCTE-LSM as a robust surrogate marker for monitoring fibrosis progression in MASLD and supports its utility in risk stratification and longitudinal disease assessment.
The intermediate-risk category warrants particular clinical attention. Its inherent instability, mirroring the 31–49% stability rate reported for histologic F2 [
22], underscores its role as a diagnostic “gray zone.” Clinically, this category encompasses a heterogeneous population that some individuals resemble low-risk patients and may remain stable or regress, whereas others behave more like high-risk patients and are prone to rapid progression. In our data, the distribution of LSM values in the intermediate-risk category exhibited a trimodal pattern on the density plot (data not shown), with peaks centered at approximately 8.8, 10.2, and 11.7 kPa. Therefore, small fluctuations in stiffness could easily shift patients across category thresholds, contributing to the shortened occupancy time observed. This highlights the urgent need for improved classification frameworks, potentially incorporating serial LSM trends, biomarkers, or metabolic profiles, to identify true progressors versus regressors within this pivotal group.
Our quantified transition probabilities and occupancy times supplemented empirical test interval by guideline and provide real-world evidence to refine risk-stratified surveillance strategies. For low-risk individuals, the long occupancy time and the low 2–3-year cumulative incidence of newly developed high-risk status (1.17–1.85%) strongly supported the current recommendation of repeating VCTE every 2–3 years. In contrast, the rapid transition dynamics in the intermediate-risk stratum, which characterized by a one-year occupancy time and a 5.40–12.58% incidence of progression to high risk within 1–2 years, indicate that a shorter monitoring interval of 1–2 years may be more appropriate for these patients. Decisions must weigh clinical benefit against resource utilization, and future cost-effectiveness analyses are needed to confirm feasibility in broader practice.
An additional key finding is the profound effect of metabolic comorbidities on transition dynamics. T2D and hypertension consistently accelerated deterioration and impeded regression across categories. These observations are fully aligned with evidence demonstrating that T2D affects 30– 75% of MASLD patients and is associated with substantially higher risks of fibrosis progression, decompensation, and HCC [
23,
24]. Hypertension, similarly prevalent, has been linked to fibrosis progression and LREs in large prospective cohorts [
8,
25]. Our results reaffirm the central role of metabolic dysfunction in driving MASLD progression and highlight the need for tailored surveillance strategies, including shorter intervals for patients with T2D or hypertension. Therefore, by integrating high-risk comorbidities and LSM status, our DYNAMO model enables an individualized and quantitative prediction of state transitions and LREs. Unlike conventional approaches that rely on a single VCTE measurement to estimate the risk of LREs, DYNAMO models annual bidirectional transitions across liver stiffness by defining risk states, capturing both disease progression and regression, and thereby provides individualized, quantitative predictions that directly inform surveillance intensity and follow-up strategies across the full spectrum of MASLD.
However, several limitations merit consideration. First, although risk states were assessed at discrete clinical visits, the exact timing of transitions was unknown, and test intervals varied based on clinical need. However, the uncertain timing and flexible intervals reflect real-world practice, and the sensitivity analyses restricting intervals to ≥6 months yielded consistent findings. Second, the Markov model assumes that future transitions depend only on the current state. Therefore, the model assigns identical transition probabilities to individuals who share the same current state and covariate profile. This assumption holds regardless of whether the state reflects long-term stability or recent rapid progression. Similarly, the model does not differentiate between a first-time transition into a high-risk state and repeated prior entries into that state. This simplifying assumption may partially overlook trajectory-dependent heterogeneity and thus not fully capture all clinically relevant variation in progression risk. Furthermore, due to the inherent assumptions of the Markov model, our approach does not distinguish between rapid progressors, slow regressors, and individuals who remain stable within the same LSM-based risk stratum, even though these subgroups may have distinct clinical characteristics and disease trajectories. Third, categorizing continuous LSM values may reduce granularity and may obscure clinically meaningful within-state variations. Finally, although our internal validation demonstrated satisfactory model performance, external validation in independent cohorts is necessar y to confirm its generalizability and clinical applicability. Particularly, our findings should be further validated in populations with other hepatic etiologies, especially Met-ALD and ALD, where concomitant alcohol consumption may influence disease trajectories and introduce complexity beyond the scope of the current model. Furthermore, the absence of repeated alcohol consumption questionnaires during follow-up precludes reclassification based on longitudinal alcohol exposure. which may lead to misclassification of a small subset of patients initially categorized as MASLD.
In conclusions, our study demonstrates that LSM-defined risk categories represent distinct clinical trajectories: lowrisk individuals exhibit substantial stability with an average occupancy time exceeding 8 years, whereas intermediaterisk individuals show pronounced dynamism and rapid transitions. These findings provide strong empirical support for current guideline recommendations to reassess low-risk individuals every 2–3 years and suggest that a shorter interval of 1–2 years may be warranted for those in the intermediate-risk category. Metabolic comorbidities, particularly T2D and hypertension, significantly accelerate disease progression and should trigger closer clinical monitoring. The DYNAMO multi-state Markov model, which enables simultaneous estimation of future risk-state transitions and LREs/death, provides a tool for personalized surveillance strategies and follow-up planning in MASLD. Future research should focus on optimizing cost-effective surveillance intervals for specific subgroups and on developing refined classification frameworks to resolve the diagnostic ambiguity of the intermediate-risk category.
FOOTNOTES
-
Authors’ contribution
Guarantor of the article: Zheng is identified as the guarantor. Shi and Zheng had full access to all of the data in the study and took responsibility for the integrity of the data and the accuracy of the data analysis. Concept and design: Shi, Yip, Petta, Bugianesi, Yoneda, Calleja, Newsome, Fan, Xiao-Dong Zhou, Kim, V. Wong. Acquisition, analysis, or interpretation of data: Shi, Zhou, Yip, Tsochatzis, Bugianesi, Yoneda, Zheng, Hagström, Boursier, Calleja, Goh, W. Chan, Sanyal, De Lédinghen, Newsome, Castéra, Lai, Fan, Romero-Gomez, Kim, V. Wong. Drafting of the manuscript: Shi, Ruo-qi Zhou, and Zheng. Critical review of the manuscript for important intellectual content: All authors. Statistical analysis: Shi, Ruo-qi Zhou, and Zheng. Administrative, technical, or material support: Shi, Bugianesi, Yoneda, Zheng, Goh, Newsome, V. Wong, Zheng. Supervision: Shi, Yip, Tsochatzis, Bugianesi, Yoneda, W. Chan, Newsome, Kim, V. Wong, Zheng.
-
Acknowledgements
This paper was funded by Key R&D Program of Zhejiang (2026C02A1172), the National Natural Science Foundation of China (82570717, 82070588, 82370577), and State Key Laboratory for Diagnosis and Treatment of Infectious Diseases (202503).
Artificial Intelligence Generated Content (AIGC) tools were only used to improve spelling and grammar.
VCTE-Prognosis Study Group: Seung Up Kim, Terry Cheuk-Fung Yip, Salvatore Petta, Elisabetta Bugianesi, Masato Yoneda, Ming-Hua Zheng, Manuel Romero-Gomez, Emmanuel Tsochatzis, Philip Newsome, Hannes Hagström, George Goh, Wah Kheong Chan, José-Luis Calleja, Jerome Boursier, Arun J. Sanyal, Jian-Gao Fan, Michelle Lai, Laurent Castéra, Victor de Lédinghen, Hye Won Lee, Vincent Wai-Sun Wong, Grazia Pennisi, Angelo Armandi, Atsushi Nakajima, Wen-Yue Liu, Carmen Lara, Mirko Zoncape, Sara Mahgoub, Ying Shang, Kevin Teh, Elba Llop, Marc de Saint-Loup, Amon Asgharpour, Huapeng Lin, Grace Lai-Hung Wong, Xiao-Dong Zhou, Rocio Gallego-Durán, Racio Macias, Adrien Lannes.
Data sharing statement: Data will be available upon request and approval from the corresponding author.
Patient and public involvement: Patients or the public WERE NOT involved in the design, or conduct, or reporting, or dissemination plans of our research.
-
Conflicts of Interest
Dr Yip reported serving as an advisory committee member and a speaker for Gilead Sciences outside the submitted work. Dr Tsochatzis reported receiving personal fees as an advisory board member for Boehringer, Novo Nordisk, Pfizer, and Siemens; receiving speaker fees from Echosens, Novo Nordisk, and AbbVie outside the submitted work. Dr Hagström reported personal fees from Astra-Zeneca, personal fees from Bristol Myers-Squibb, personal fees from MSD, personal fees from Novo Nordisk, personal fees from Boehringer Ingelheim, personal fees from KOWA, and personal fees from GW Phara outside the submitted work, and grants from AstraZeneca, grants from Echosens, grants from Gilead Sciences, grants from Intercept, grants from MSD, grants from Novo Nordisk, and grants from Pfizer outside the submitted work. Dr Boursier reported receiving grants and personal fees from Echosens outside the submitted work. Dr Calleja reported receiving other from Echosens Clinical Trials during the conduct of the study; grants from Roche Pharma and other from Gilead Advisory Board outside the submitted work. Dr WK. Chan reported serving as consultant or advisory board member for Abbott, AbbVie, Boehringer Ingelheim, IPSEN, Kowa, Novo Nordisk, Roche and Zuellig Pharma; a speaker for Abbott, Echosens, Hisky Medical, Novo Nordisk, Roche and Viatris; and receiving grants from Abbott and Roche. Dr Sanyal reported receiving grants from Intercept, personal consulting fees from Gilead, grants from Merck, personal consulting fees from Pfizer, grants and personal consulting fees from Eli Lilly, grants and personal consulting fees from Novo Nordisk, Boehringer Ingelheim, Novartis, Histoindex, and stock options from Genfit, Tiziana, Durect, Inversago, and personal consulting fees from Genentech, ALnylam, Regeneron, Zydus, LG chem, Hanmi, Madrigal, Path AI, 89 Bio, and stock options from Galmed outside the submitted work. Dr De Lédinghen reported receiving nonfinancial support from Echosens during the conduct of the study. Dr Newsome reported receiving grants from Novo Nordisk, advisory board and personal consulting fees, honoraria for lectures and travel expenses from Novo Nordisk, personal consulting and advisory board fees from Boehringer Ingelheim, Gilead, Intercept, Poxel Pharmaceuticals, Bristol-Myers Squibb, Pfizer, MSD, Sun Pharma, Eli Lilly, Madrigal, GSK, and nonfinancial support for educational events from AiCME outside the submitted work. Dr Castéra reported receiving personal fees for consulting and speakers bureau from Echosens during the conduct of the study; personal consultancy fees from Boston pharmaceutical and Gilead, speaker bureau and consultancy personal fees from GSK, personal speaker bureau fees from Inventiva, personal consultancy fees from Madrigal, personal Consultancy fees from MSD and Novo Nordisk, personal consultancy fees from Pfizer, Sagimet, and Siemens Healthineers outside the submitted work. Dr. Yoneda reported receiving grants from Gilead Sciences and speaker fees from KOWA outside the submitted work. Dr Romero-Gomez reported receiving personal fees from Alpha-sigma, Astra-Zeneca, Bausch Health, BMS, Boehringer-Ingelheim, Exo-Biologics, Gilead, Ipsen, MSD, Novo-Nordisk, Pfizer, Prosciento, Resolution Therapeutics, Roche, Rubió, Sagimet, Siemens, UCB pharma, and research grants from Gilead, Intercept, Siemens, Theratechnologies, Novo-Nordisk, Echosens. Dr Kim reported personal fees from Gilead Sciences, personal fees from GSK, personal fees from Bayer, personal fees from Eisai, personal fees from AbbVie, personal fees from Echosens, personal fees from MSD, personal fees from Bristol-Myers Squibb, and personal fees from AstraZeneca outside the submitted work, and grants from AbbVie, grants from Bristol-Myers Squibb, and grants from Gilead Sciences outside the submitted work. Dr V. Wong reported receiving personal speaker fees from Abbott, consultant and speaker fees from AbbVie, personal consultant fees from Boehringer Ingelheim, Echosens, Gilead Sciences, grants from Gilead Sciences, personal consultant fees from Intercept, Inventiva, Novo Nordisk, personal consultant fees from Pfizer, Sagimet Biosciences, TARGET PharmaSolutions, personal speaker fees from Unilab, personal consultant fees from Visirna, and being a cofounder of Illuminatio outside the submitted work.
SUPPLEMENTARY MATERIAL
Supplementary material is available at Clinical and Molecular Hepatology website (
http://www.e-cmh.org).
Supplementary Table 3.
Cumulative incidence of the first transition to the high-risk category among participants with an initial low- or intermediate-risk state
cmh-2026-0279-Supplementary-Table-3.pdf
Supplementary Figure 3.
Yearly probability of state transitions by type 2 diabetes. (A) Without T2D; (B) With T2D. LREs, liver-related events; T2D, type 2 diabetes.
cmh-2026-0279-Supplementary-Fig-3.pdf
Supplementary Figure 4.
Yearly probability of state transitions by hypertension. (A) Without hypertension; (B) With hypertension. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-4.pdf
Supplementary Figure 5.
Yearly probability of state transitions by obesity status. (A) Without obesity; (B) With obesity. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-5.pdf
Supplementary Figure 6.
Yearly probability of state transitions by SGLT-2i usage in those with type 2 diabetes. (A) Without SGLT-2i usage; (B) With SGLT-2i usage. LREs, liver-related events; SGLT-2i, sodium-glucose cotransporter 2 inhibitors (SGLT-2i).
cmh-2026-0279-Supplementary-Fig-6.pdf
Supplementary Figure 7.
Yearly probability of state transitions by GLP-1RAs usage in those with type 2 diabetes. (A) Without GLP-1RAs usage; (B) With GLP-1RAs usage. LREs, liver-related events; GLP-1RAs, glucagon-like peptide-1 receptors agonists.
cmh-2026-0279-Supplementary-Fig-7.pdf
Supplementary Figure 8.
Occupancy time and yearly probability of state transitions by restricted test interval to over 6 months. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-8.pdf
Supplementary Figure 9.
Occupancy time and yearly probability of state transitions by including those with at least 3 VCTE measurements. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events; VCTE, vibration-controlled transient elastography.
cmh-2026-0279-Supplementary-Fig-9.pdf
Supplementary Figure 10.
Occupancy time and yearly probability of state transitions by using cutoffs of 10 and 15 kPa. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-10.pdf
Supplementary Figure 11.
Occupancy time and yearly probabilities of state transitions restricted to follow-up over 2 years. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-11.pdf
Supplementary Figure 12.
Occupancy time and yearly probability of state transitions by restricting state transitions to cases where the changes were simultaneously over 20%. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-12.pdf
Supplementary Figure 13.
Occupancy time and yearly probabilities of state transitions by separating LREs and non-liver mortality as a single state. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-13.pdf
Supplementary Figure 14.
Occupancy time and yearly probabilities of state transitions by only including LREs as absorbing state. (A) Occupancy time at each LSM-based risk category; (B) Yearly probability of state transitions. LREs, liver-related events.
cmh-2026-0279-Supplementary-Fig-14.pdf
Figure 1Flowchart of participant selection (A) and transition pathways (B). LREs, liver-related events; MASLD, metabolic dysfunction-associated steatotic liver disease; VCTE, vibration-controlled transient elastography.
Figure 2Yearly probabilities of state transitions in the general population. LREs, liver-related events.
Figure 3Occupancy time at each LSM-based risk category and cumulative incidence of first transition to high-risk category. (A) Occupancy time at each LSM-based risk category. (B) Kaplan–Meier curve showing cumulative incidence of first transition to high-risk category among participants with an initial low-or intermediate-risk state. LSM, liver stiffness measurement.
Figure 4Individualized 5-year risk predictions for LREs or death. The stacked area plot illustrates dynamic 5-year state predictions for a patient with MASLD, with the probabilities of low-, intermediate-, and high-risk states, as well as LREs/death, summing to 100%. Panel (A) shows predicted probabilities based on baseline LSM state and covariates. Panels (B–D) display updated predictions following changes in LSM state or covariates at 1 year. (A) The 5-year probability of the MASLD individual aged 50, having a BMI of 24 kg/m2, without T2D and hypertension, having a platelet count of 200 (109/L); (B) the probability updated when turn to low-risk state after 1 year; (C) the probability updated when turn to high-risk state after 1 year; (D) the probability updated when develop T2D and having a BMI of 27 kg/m2 after 1 year. BMI, body mass index; LREs, liver-related events; LSM, liver stiffness measurement; MASLD, metabolic dysfunction-associated steatotic liver disease; T2D, type 2 diabetes.
Table 1
Table 1
|
Characteristic |
Low (< 8 kPa) |
Intermediate (8–12 kPa) |
High (>12 kPa) |
P-value |
|
Number (%) |
8,372 (72.7) |
1,825 (15.9) |
1,317 (11.4) |
|
|
Age (yr) |
53 (43, 61) |
55 (44, 63) |
58 (48, 66) |
<0.001 |
|
Male sex (n [%]) |
4,999 (59.7) |
991 (54.3) |
689 (52.3) |
<0.001 |
|
BMI (kg/m2) |
26.3 (24.2, 28.9) |
29.1 (26.2, 32.1) |
29.7 (26.3, 33.9) |
<0.001 |
|
Asian (n [%]) |
7,683 (91.8) |
1,430 (78.4) |
913 (69.3) |
<0.001 |
|
T2D (n [%]) |
2,644 (31.6) |
801 (43.9) |
746 (56.6) |
<0.001 |
|
Hypertension (n [%]) |
2,640 (31.5) |
794 (43.5) |
706 (53.6) |
<0.001 |
|
ALT (IU/L) |
32.9 (21.0, 54.0) |
51.0 (30.0, 82.0) |
49.0 (31.0, 82.0) |
<0.001 |
|
AST (IU/L) |
28.0 (21.0, 39.0) |
44.0 (30.0, 63.0) |
50.0 (33.0, 75.7) |
<0.001 |
|
Platelet (109/L) |
244.0 (207.0, 286.0) |
232.0 (192.0, 277.0) |
198.0 (152.5, 249.0) |
<0.001 |
|
Total bilirubin (μmol/L) |
11.9 (9.0, 15.4) |
11.7 (8.6, 15.4) |
12.0 (8.6, 16.6) |
<0.001 |
|
Fasting glucose (mmol/L) |
5.7 (5.2, 6.7) |
6.1 (5.4, 7.3) |
6.3 (5.4, 7.8) |
<0.001 |
|
HbA1c (%) |
6.1 (5.6, 6.9) |
6.5 (5.8, 7.3) |
6.7 (6.0, 7.6) |
<0.001 |
|
Total cholesterol (mmol/L) |
4.8 (4.1, 5.6) |
4.7 (3.9, 5.4) |
4.4 (3.8, 5.2) |
<0.001 |
|
HDL-c (mmol/L) |
1.2 (1.0, 1.4) |
1.2 (0.9, 1.4) |
1.1 (0.9, 1.3) |
<0.001 |
|
LDL-c (mmol/L) |
2.9 (2.2, 3.6) |
2.7 (2.1, 3.6) |
2.5 (1.9, 3.2) |
<0.001 |
|
Triglycerides (mmol/L) |
1.5 (1.1, 2.1) |
1.6 (1.2, 2.2) |
1.6 (1.2, 2.2) |
<0.001 |
|
SGLT-2i use (n [%])*
|
996 (37.8) |
286 (35.7) |
217 (29.1) |
0.004 |
|
GLP-1RAs use (n [%])*
|
324 (12.3) |
138 (17.2) |
126 (16.9) |
<0.001 |
|
LSM (kPa) |
5.2 (4.3, 6.2) |
9.4 (8.7, 10.6) |
17.0 (14.0, 23.3) |
<0.001 |
|
CAP (dB/m) |
297 (270, 327) |
320 (289, 347) |
319 (283, 353) |
<0.001 |
|
Liver biopsy (n [%]) |
887 (10.6) |
620 (34.0) |
595 (45.2) |
<0.001 |
|
MASH |
514 (57.9) |
445 (71.8) |
456 (76.6) |
<0.001 |
|
At-risk MASH |
155 (17.5) |
266 (42.9) |
383 (64.4) |
<0.001 |
Table 2Occupancy time across LSM-based risk categories stratified by relevant factors
Table 2
|
Characteristics |
Occupancy time |
|
Low (8 kPa) |
Intermediate (8–12 kPa) |
High (>12 kPa) |
|
Age |
|
<55 |
8.49 (7.76, 9.28) |
0.85 (0.80, 0.91) |
2.06 (1.86, 2.28) |
|
≥55 |
8.26 (7.61, 8.97) |
0.99 (0.93, 1.06) |
2.43 (2.21, 2.67) |
|
Sex |
|
Female |
7.91 (7.22, 8.67) |
0.91 (0.85, 0.98) |
2.55 (2.29, 2.82) |
|
Male |
8.82 (8.13, 9.56) |
0.92 (0.87, 0.98) |
2.05 (1.87, 2.25) |
|
T2D |
|
No |
9.07 (8.38, 9.82) |
0.83 (0.78, 0.89) |
2.35 (2.13, 2.60) |
|
Yes |
6.81 (6.19, 7.49) |
1.00 (0.93, 1.07) |
2.15 (1.95, 2.37) |
|
Hypertension |
|
No |
9.32 (8.59, 10.11) |
0.87 (0.82, 0.93) |
2.48 (2.26, 2.73) |
|
Yes |
6.64 (6.06, 7.28) |
0.95 (0.88, 1.01) |
2.03 (1.83, 2.25) |
|
Obesity |
|
No |
11.07 (10.29, 11.91) |
0.99 (0.94, 1.05) |
2.40 (2.19, 2.62) |
|
Yes |
3.58 (3.21, 4.00) |
0.76 (0.70, 0.83) |
2.07 (1.85, 2.32) |
|
SGLT-2i |
|
No |
6.69 (5.95, 7.53) |
1.03 (0.94, 1.13) |
2.27 (2.01, 2.56) |
|
Yes |
7.70 (6.50, 9.12) |
1.06 (0.94, 1.20) |
2.22 (1.86, 2.66) |
|
GLP-1RAs |
|
No |
7.51 (6.75, 8.35) |
1.08 (1.00, 1.17) |
2.41 (2.16, 2.70) |
|
Yes |
4.92 (3.89, 6.22) |
0.89 (0.75, 1.04) |
1.74 (1.37, 2.20) |
Abbreviations
aspartate aminotransferase
controlled attenuation parameter
compensated advanced chronic liver disease
glucagon-like peptide-1 receptor agonists
high-density lipoprotein cholesterol
interquartile range/median
low-density lipoprotein cholesterol
liver stiffness measurement
metabolic dysfunction-associated steatohepatitis
metabolic dysfunction-associated steatotic liver disease
sodium-glucose cotransporter 2 inhibitors
vibration-controlled transient elastography
REFERENCES
- 1. Miao L, Targher G, Byrne CD, Cao YY, Zheng MH. Current status and future trends of the global burden of MASLD. Trends Endocrinol Metab 2024;35:697-707.
- 2. Younossi ZM, Kalligeros M, Henry L. Epidemiology of meta-bolic dysfunction-associated steatotic liver disease. Clin Mol Hepatol 2025;31:S32-s50.
- 3. Hagström H, Shang Y, Hegmar H, Nasr P. Natural history and progression of metabolic dysfunction-associated steatotic liver disease. Lancet Gastroenterol Hepatol 2024;9:944-956.
- 4. Tilg H, Petta S, Stefan N, Targher G. Metabolic dysfunction-associated steatotic liver disease in adults: a review. Jama 2026;335:163-174.
- 5. Yang Z, Liu Z, Liu W, Dong XC. Steatotic liver disease and cancer: from pathogenesis to therapeutic targets. eGastroenterology 2025;3:e100218.
- 6. Zhou R, Zhang X, Liu X, Huang R, Wang Y, Xia D, et al. Association between dietary amino acid intake and the risk of metabolic dysfunction-associated steatotic liver disease. J Adv Res 2026;80:893-903.
- 7. Taylor RS, Taylor RJ, Bayliss S, Hagström H, Nasr P, Schattenberg JM, et al. Association Between fibrosis stage and outcomes of patients with nonalcoholic fatty liver disease: a systematic review and meta-analysis. Gastroenterology 2020;158:1611-1625e2.
- 8. Singh S, Allen AM, Wang Z, Prokop LJ, Murad MH, Loomba R. Fibrosis progression in nonalcoholic fatty liver vs nonalcoholic steatohepatitis: a systematic review and meta-analysis of paired-biopsy studies. Clin Gastroenterol Hepatol 2015;13:643-654e1. 9 quiz e39–40.
- 9. Zhang X, Kam LY, Barnett SD, Henry L, Cheung R, Nguyen MH. Disease progression for histologic diagnosis of metabolic dysfunction-associated steatotic liver disease in the real-world: a nationwide US study. Dig Dis 2025;43:36-45.
- 10. Anania FA, Hager R, Higgins K, Makar GA, Siegel J, Tran TT. Non-invasive tests: establishing efficacy for metabolic dys-function-associated steatohepatitis beyond the biopsy-current perspectives from the division of hepatology and nutrition, US Food and Drug Administration. Hepatology 2026;84:321-327.
- 11. Selvaraj EA, Mózes FE, Jayaswal ANA, Zafarmand MH, Vali Y, Lee JA, et al. Diagnostic accuracy of elastography and magnetic resonance imaging in patients with NAFLD: a systematic review and meta-analysis. J Hepatol 2021;75:770-785.
- 12. Mózes FE, Lee JA, Selvaraj EA, Jayaswal ANA, Trauner M, Boursier J, et al. Diagnostic accuracy of non-invasive tests for advanced fibrosis in patients with NAFLD: an individual patient data meta-analysis. Gut 2022;71:1006-1019.
- 13. Kanwal F, Shubrook JH, Adams LA, Pfotenhauer K, Wai-Sun Wong V, Wright E, et al. Clinical care pathway for the risk stratification and management of patients with nonalcoholic fatty liver disease. Gastroenterology 2021;161:1657-1669.
- 14. European Association for the Study of the Liver (EASL); European Association for the Study of Diabetes (EASD); European Association for the Study of Obesity (EASO). EASL-EASD-EASO Clinical Practice Guidelines on the management of metabolic dysfunction-associated steatotic liver disease (MASLD). J Hepatol 2024;81:492-542.
- 15. Park H, Yoon EL, Kim M, Kwon SH, Kim D, Cheung R, et al. Cost-effectiveness study of FIB-4 followed by transient elastography screening strategy for advanced hepatic fibrosis in a NAFLD at-risk population. Liver Int 2024;44:944-954.
- 16. Papatheodoridi M, Hiriart JB, Lupsor-Platon M, Bronte F, Boursier J, Elshaarawy O, et al. Refining the Baveno VI elastography criteria for the definition of compensated advanced chronic liver disease. J Hepatol 2021;74:1109-1116.
- 17. Lin H, Lee HW, Yip TC, Tsochatzis E, Petta S, Bugianesi E, et al. Vibration-controlled transient elastography scores to predict liver-related events in steatotic liver disease. Jama 2024;331:1287-1297.
- 18. Shi Y, Zhou R, Kim SU, Yip TC, Tsochatzis E, Petta S, et al. Association between longitudinal weight change and clinical outcome in individuals with MASLD. Hepatology 2025 Oct 9;10.1097/HEP.0000000000001557.
- 19. Yip TC, Lee HW, Lin H, Tsochatzis E, Petta S, Bugianesi E, et al. Prognostic performance of the two-step clinical care pathway in metabolic dysfunction-associated steatotic liver disease. J Hepatol 2025;83:304-314.
- 20. Shi Y, Kim SU, Yip TC, Tsochatzis E, Petta S, Nakajima A, et al. Effect of antidiabetic drug classes on the risk of liver-related events in individuals with T2D and MASLD. Clin Gastroenterol Hepatol 2026;24:723-732e10.
- 21. Newsome PN, Ambery P. Incretins (GLP-1 receptor agonists and dual/triple agonists) and the liver. J Hepatol 2023;79:1557-1565.
- 22. Le P, Payne JY, Zhang L, Deshpande A, Rothberg MB, Alkhouri N, et al. Disease state transition probabilities across the spectrum of NAFLD: a systematic review and meta-analysis of paired biopsy or imaging studies. Clin Gastroenterol Hepatol 2023;21:1154-1168.
- 23. Huang DQ, Noureddin N, Ajmera V, Amangurbanova M, Bettencourt R, Truong E, et al. Type 2 diabetes, hepatic decompensation, and hepatocellular carcinoma in patients with non-alcoholic fatty liver disease: an individual participant-level data meta-analysis. Lancet Gastroenterol Hepatol 2023;8:829-836.
- 24. Huang DQ, Wilson LA, Behling C, Kleiner DE, Kowdley KV, Dasarathy S, et al. Fibrosis progression rate in biopsy-proven nonalcoholic fatty liver disease among people with diabetes versus people without diabetes: a multicenter study. Gastroenterology 2023;165:463-472e5.
- 25. Zhou XD, Lian LY, Chen QF, Kim SU, Cheuk-Fung Yip T, Petta S, et al. Effect of hypertension on long-term adverse clinical outcomes and liver fibrosis progression in MASLD. J Hepatol 2026;84:254-265.
, Ming-Hua Zheng22,23,24
