Introduction

Metabolic dysfunction-associated steatohepatitis (MASH), formerly known as non-alcoholic steatohepatitis (NASH)1,2, is currently the most prevalent chronic liver disease in the United States, affecting an estimated 3–5% of adults3. Metabolic risk factors, including insulin resistance, dyslipidemia, central obesity, and hypertension, influence disease progression2. Patients with untreated MASH are at risk of developing significant fibrosis, compensated cirrhosis, and decompensated cirrhosis, potentially necessitating liver transplantation. Notably, liver transplants due to MASH, which have increased by 52% over the past decade, now rank as a leading indication for liver transplantation, along with alcohol-associated liver disease4,5.

The management of MASH has historically focused on lifestyle modifications, including dietary changes and weight loss6. However, MASH management has radically changed in the last few years with the development and approval of a novel thyroid hormone receptor beta agonist for adults with non-cirrhotic MASH and moderate-to-advanced fibrosis7,8. Notably, no novel therapeutics have been approved for patients with compensated cirrhosis, defined as the presence of cirrhosis, but without overt clinical symptoms or decompensation. Therefore, patients with compensated cirrhosis are uniquely vulnerable; often, their disease process is silent, they remain undiagnosed, and have few therapeutic options to reverse or stabilize their disease progression. This highlights the importance of effective screening strategies to detect compensated cirrhosis9.

Real-world evidence of disease trajectories in patients with compensated MASH cirrhosis remains limited. Prior studies have suggested that these patients are subject to disparities along lines of socioeconomic status and demography, with Hispanic/Latino patients at greater risk of adverse outcomes10,11. However, detailed data identifying the drivers of these disparities are scarce10.

To address this gap, this study aimed to characterize the demographic and clinical features of patients with compensated MASH cirrhosis, examine progression rates to decompensation and other key clinical outcomes, assess baseline predictors of adverse events, and compare clinical outcomes between Hispanic/Latino versus non-Hispanic/Latino subgroups, utilizing data from a major tertiary care center with a diverse patient population.

Results

Descriptive Characteristics

A total of 493 patients with compensated MASH, with a median follow-up of 3.4 years, were identified (Table 1). At baseline, the median age of the cohort was 56.1 years, 61.5% (N = 303) were women, 33.9% (N = 167) were of Hispanic/Latino ethnicity, and 57.0% (N = 281) had Medicare or Medicaid insurance. The median distance traveled for care was 130.9 kilometers. The cohort had a relatively high burden of comorbid conditions, with a median Charlson Comorbidity Index score of 5. The most common comorbidities were hypertension (64.9%), hyperlipidemia (52.1%), and diabetes (45.8%). The median body mass index (BMI) was 32 kg/m², with 74.4% (N = 367) having a BMI of more than 30.

Table 1 Baseline characteristics for the overall patient cohort and subgroups based on ethnicity
Full size table

The study population experienced significant adverse effects during the follow-up period. All-cause mortality was reported in 14.2% (N = 70) of patients (annual incidence of 34 per thousand person-years), and decompensation symptoms were reported in 19.9% (n = 98) of patients (annual incidence of 50 per thousand person-years). The temporal relationship between study outcomes, as shown in Fig. S1, revealed that elevated liver indices most frequently preceded the onset of decompensation symptoms. In total, 32.5% of the patients experienced at least one of the study outcomes (i.e., composite outcome) (annual incidence of 96 per thousand person-years). The incidence rates of the individual components of decompensation symptoms and composite outcomes are presented in Table 2.

Table 2 Unadjusted Outcomes in Overall Cohort and Subgroups
Full size table

Multivariate Analyses

Multivariate Cox regression analysis was performed using an L1-penalized (lasso) regression to enable data-driven variable selection, retaining only the most informative predictors while reducing overfitting. Missing data were addressed using multiple imputation to preserve sample size and minimize bias, ensuring more stable and reliable estimates in the final model (Table 3). Several key variables showed significant associations with outcomes; Higher MELD and FIB-4 scores were consistently associated with an increased risk of all-cause mortality, composite outcomes, and decompensation events. Older age and abnormal alkaline phosphatase (ALP) levels further contributed to higher all-cause mortality, whereas oral diabetes medication use was negatively associated with the composite outcome (Fig. 1). Finally, recent cohort entry was associated with a reduced risk of composite outcomes and decompensation symptoms.

Fig. 1: Multivariate L1-penalized (LASSO) Cox regression hazard-ratio plots depicting significant predictors for.
Fig. 1: Multivariate L1-penalized (LASSO) Cox regression hazard-ratio plots depicting significant predictors for.
Full size image

A Composite outcome—earliest occurrence of any follow-up event (decompensation symptoms, hepatocellular carcinoma, liver transplant, MELD-3 ≥ 15, or eCPT ≥ 7). B Decompensation symptoms—earliest occurrence of ascites, esophageal varices hemorrhage, spontaneous bacterial peritonitis, or encephalopathy. C All-cause mortality.

Table 3 Multivariate L1-Penalized Cox Regression Analysis
Full size table

Subgroup Analysis Based on Hispanic/Latino Ethnicity

We observed significant differences in the Hispanic/Latino subgroups (n = 167) compared with other patients (n = 326). Hispanic/Latino patients were younger (mean age 54.3 vs. 57.1 years), had a longer median driving distance from home to UCSF Medical Center (142.0 vs. 96.4 km), and a higher median ADI (7.7 vs. 4.9). In addition, they had higher median MELD scores, ALPALP levels, and insulin use at baseline, while the use of antihypertensives and other medications was consistently lower. Notably, Hispanic/Latino patients had higher proportions of Medicaid coverage and lower proportions of commercial insurance coverage (Table 1).

The unadjusted analyses of the study outcomes in the Hispanic/Latino subgroups are summarized in Table 2. Hispanic/Latino patients had higher measures of liver dysfunction (MELD Score ≥ 15 and eCPT score ≥ 7) and were at a greater risk of ascites and mortality. After adjusting for baseline clinical characteristics (CCI and MELD) and SDOH (ADI, driving distance, and insurance status) in the Cox proportional hazards model, only MELD ≥ 15 remained statistically significant (Table 4). These results suggest that baseline clinical and SDOH factors largely accounted for most of the observed disparities in the outcomes between the two groups.

Table 4 Unadjusted and Adjusted Cox Regression Analysis: Association of Hispanic/Latino Ethnicity with Study Outcomes
Full size table

Discussion

We studied a real-world cohort of compensated MASH patients at a tertiary academic medical center. Our cohort showed a substantial comorbidity burden; the 5-year all-cause mortality rate was 14.2%, which was significantly higher than the age- and sex-adjusted mortality risk in the general population estimated at 5%12. This cohort had a high incidence of liver-related complications, with a 5-year incidence of 32.5% for at least one of these adverse outcomes. Although the current literature on the long-term outcomes of real-world cohorts of patients with MASH is limited, this study demonstrated similar rates of adverse outcomes compared to other studies13,14. Time-to-event analysis revealed that several commonly used risk prediction metrics, notably the MELD and FIB-4 scores, were predictive of adverse outcomes at the index visit. Moreover, higher baseline ALP levels were linked to worse survival, aligning with reports that ALP reflects cholestatic injury and advanced fibrosis, both of which predict adverse outcomes in MASH15,16Moreover, higher baseline ALP levels were linked to worse survival, aligning with reports that ALP reflects cholestatic injury and advanced fibrosis, both of which predict adverse outcomes in MASH.

Compared with other etiologies, compensated MASH shares long-term risk but differs in key ways: alcohol-associated cirrhosis tends to present at younger ages and decompensate more frequently17,18. In HCV, direct-acting antivirals have reshaped the natural history, reducing the 5-year risk of decompensation from approximately 15–37% when untreated to about 8–10% after cure, although the risk of hepatocellular carcinoma persists19,20. For MASH, emerging agents (GLP-1–based therapies and FGF21 analogs) improve intermediate endpoints; however, no drug has yet demonstrated benefit on hard clinical outcomes21,22,23. Until such evidence emerges, early detection, risk stratification, and lifestyle modification remain central to care.

Given the high proportion of Hispanic/Latino patients at the UCSF Medical Center, differences in outcomes by race and ethnicity were also investigated. Our findings indicated that Hispanic/Latino patients lived in areas with higher social deprivation, relied more on public insurance, traveled farther to UCSF facilities, and had a higher incidence of adverse outcomes. These patients also had a higher baseline MELD score, implying a later stage at diagnosis, which has also been observed in previous studies24. While hereditary and genetic factors, such as the PNPLA3 genetic variant, have been linked to liver disease progression in this population25,26, other factors, including lifestyle choices, healthcare access, and socioeconomic conditions, can also contribute to the observed disparities in outcomes27,28. Furthermore, previous literature suggests that screening and risk-stratification disparities are observed in Hispanic/Latino patients with MASLD. For instance, non-invasive test cutoff points were derived primarily from non-Hispanic/Latino populations and may not accurately represent the risk for Hispanic/Latino individuals. A study by Tincopa et al. showed that standard FIB-4 and VCTE cut-points miss a substantial proportion of Hispanic patients with biopsy-proven advanced fibrosis, which can result in considerable delays in establishing an accurate diagnosis29. The adjusted analysis showed that baseline SDOH and clinical factors were the primary contributors to this heightened risk of adverse outcomes. This underscores the critical need for earlier diagnosis and enhanced healthcare strategies to address the specific challenges experienced by Hispanic/Latino patients with liver diseases.

These findings highlight several areas where clinical practice and health-system planning may benefit from further attention. The strong influence of baseline disease severity and social determinants of health on outcomes underscores the need to evaluate more effective strategies for earlier detection of compensated MASH, including the potential role of automated EHR/NLP tools and routine risk stratification approaches using MELD and FIB-4. Our results also highlight significant gaps in healthcare accessibility, including social deprivation, long travel distances, and reliance on public insurance, which may limit timely engagement with specialty liver care. Addressing these barriers will likely require a combination of improved referral pathways, greater availability of supportive services, and efforts to expand access to hepatology care in underserved settings. At the policy level, the attenuation of ethnic disparities after adjusting for social and access factors suggests that community-based screening efforts, enhanced linkage to care, and targeted outreach in high-risk Hispanic/Latino communities warrant further study. Overall, the findings reinforce the need to develop and rigorously evaluate more effective screening and management protocols for compensated MASH, alongside strategies to promote equitable access to specialty liver care across diverse populations.

This study included patients from UCSF, a major tertiary center in the Bay Area. While our cohort may not fully represent the nationwide demographic distribution, the underlying pathophysiology, natural history, and clinical progression of MASH are consistent across geographic and demographic groups. Thus, despite regional differences in patient composition, the biological mechanisms driving disease development and progression support the broader applicability and generalizability of our findings to the U.S. population as a whole30. We used routinely collected electronic health record (EHR) data. Despite some variables having varying degrees of missingness, this limitation was addressed using imputation methods. In addition, the limited sample size and follow-up duration resulted in certain endpoints (e.g., liver transplantation and hepatocellular carcinoma) being relatively rare, thereby reducing the statistical power for those outcomes. Although our NLP method does not entirely replicate the outcomes of manual chart reviews, it achieves acceptable performance metrics in direct comparison. Furthermore, we utilized L1 penalized regression to select variables, an approach that, although widely used, may not yield the most suitable subset of predictors for our dataset. Finally, other causes of liver disease, such as high-risk genetic variants or exposure to liver-toxic drugs, could have contributed to the study outcomes beyond MASH.

In conclusion, this study of a real-world cohort of patients with MASH cirrhosis highlights the substantial disease burden and the incidence of poor outcomes in the natural history of the disease. The findings of this study suggest a strong association between specific baseline characteristics, such as elevated baseline MELD and FIB-4 scores, and adverse outcomes. In addition, Hispanic/Latino patients have higher crude incidence rates and more severe disease at diagnosis, which could be potentially driven by SDOH and baseline disease status. These findings underscore the need for earlier screening and improved access to healthcare in this population. Although therapeutic advancements show promise in reducing MASH progression and complications, early identification and risk stratification remain crucial for improving outcomes in diverse patient populations.

Methods

Design and Setting

This retrospective observational cohort study was conducted at the University of California, San Francisco (UCSF), using electronic health records (EHR) from January 1, 2012, to January 1, 2024. Patient data were obtained from the de-identified Clinical Data Warehouse at UCSF. This resource contains comprehensive patient data, including demographics, clinical variables, and encounter records31. The study was conducted in accordance with both the Declaration of Helsinki and the Istanbul Declaration, and the UCSF Institutional Review Board approved this study (Study #21-33854). The need for informed consent was waived by the UCSF ethics committee because the study used only de-identified data.

Study Population

The study inclusion criteria consisted of adult patients (age ≥ 18 years) with at least one clinical encounter within the study timeframe who had MASH as determined by a previously validated natural language processing algorithm (NLP) called NASHDetection32. The NASHDetection algorithm was applied to identify clinical notes authored by gastroenterologists and hepatologists. To identify the cohort with compensated cirrhosis, patients who were flagged by the NASHDetection and had at least one ICD-9 or ICD-10 code corresponding to compensated cirrhosis were selected.

The exclusion criteria consisted of patients with ICD-9 or ICD-10 codes indicating the presence or history of decompensated cirrhosis within 90 days of the initial diagnosis of compensated cirrhosis or other chronic liver diseases, such as hepatitis B or C, primary sclerosing cholangitis, primary biliary cholangitis, alcohol-associated liver disease, or autoimmune hepatitis. As post-liver transplant MASH is a clinical entity distinct from native MASH, patients who were diagnosed with MASH after liver transplantation were also excluded. The index date was defined as the earliest recorded cirrhosis code in the absence of any decompensation within the subsequent 90 days. Patients who developed a decompensation code within 90 days may have been misclassified at the initial diagnosis of compensated cirrhosis. Patients were monitored from this index date until an outcome event or until loss to follow-up, which was defined as death, the data extraction date, or an absence of clinical encounters for 18 months or more. Subgroup analyses based on Hispanic/Latino ethnicities were also performed to understand the influence of social determinants of health on racial and ethnic differences in outcomes.

Variables

Baseline demographic characteristics, comorbidities, medication use, and laboratory results were extracted to characterize the patients and evaluate factors predicting disease progression. Demographic variables included age, sex, race and ethnicity, insurance status, area deprivation index (ADI), and driving distance to San Francisco. Laboratory indices used to assess liver disease included the Model for End-Stage Liver Disease 3.0 (MELD) score33, the Fibrosis-4 (FIB-4) score34, and the electronic Child-Pugh-Turcotte (eCPT) score35, which were previously validated for extraction from EHR data.

Outcomes of Interest

Major outcomes of interest included decompensation (defined as the incidence of ascites, spontaneous bacterial peritonitis, variceal hemorrhage, and hepatic encephalopathy), hepatocellular carcinoma, disease progression defined by MELD score ≥ 15 or eCPT score ≥ 7, liver transplantation, all-cause mortality, and a composite outcome defined as the first occurrence of the following: decompensation symptoms, hepatocellular carcinoma, liver transplant, MELD-3 ≥ 15 or eCPT ≥ 7.

Statistical Analyses

Descriptive statistics were used to characterize the baseline features of the study population, with continuous variables reported as means, standard deviations, medians, interquartile ranges, and categorical variables as counts and proportions. Chi-square tests were used to compare categorical variables, and t-tests or Mann-Whitney tests were used to compare continuous variables, depending on distributional characteristics. Kaplan-Meier estimation was used to report descriptive survival analyses. Time-to-event analysis was performed using Cox proportional hazards regression, which was initially employed to evaluate univariate associations between baseline factors and each clinical outcome. All variables were subsequently incorporated into Cox proportional hazards models with L1 regularization applied to the regression coefficients to mitigate overfitting and identify the most predictive subset of covariates. A grid search was performed to determine the optimal lambda parameter for regularization, ensuring the best model performance.

For subgroup comparisons between Hispanic/Latino patients and non-Hispanic/Latino patients, baseline characteristic differences were summarized using descriptive statistics. Subgroup-specific Kaplan–Meier curves and unadjusted Cox models were constructed to identify potential disparities in outcomes. Adjusted Cox models were fitted, controlling for demographic factors, social determinants of health (SDOH), and baseline clinical status, to elucidate the underlying causes of these discrepancies and compare the adjusted results with those from the unadjusted analysis.

Missing values for any variable with more than 30% missing data were excluded from the multivariable analyses, whereas those with fewer than 30% missing data were imputed using single-imputation methods. A comprehensive list of variables, their definitions, and associated electronic health record (EHR) codes is available in the supplementary materials. All data and queries were extracted, processed, and analyzed using PySpark and Python 3.8.1736.