We conducted a retrospective cohort study of patients diagnosed with MASLD in California, USA from 2005 to 2019. The study was approved by the California Health and Human Services Agency Institutional Review Board in Sacramento, California (project number: 14-10-1748).

Data were collected from the California Department of Health Care Access and Information (HCAI), which is a longitudinal database of 45 million inpatient hospital records (>98% of all civilian inpatients) from approximately 450 healthcare facilities throughout California, USA. Patient records were at the event-level with unique identification numbers to link patients across visits. Mortality data were obtained from the California State Death Statistical Master file records, containing death certificates for all Californians. Hospital characteristics data were gathered from HCAI and American Hospital Association databases. Mortality and hospital characteristics data were linked to inpatient records via patient and facility identification numbers, respectively.

Patients eligible for inclusion were adults (≥18 and ≤75 years) with one or more Internal Classification of Diseases, Ninth Revision and Tenth Revision, Clinical Modification (ICD-9-CM and ICD-10-CM) MASLD and cirrhosis codes (Supplementary Appendix) at baseline. We excluded patients with HIV, bariatric surgery prior to index MASLD diagnosis, liver transplant prior to bariatric surgery, or inconsistent death records from manual review.

For treated patients, the index date was defined as the first date of bariatric surgery. For non-bariatric controls, the index date was defined as the first date of MASLD diagnosis, formerly known as non-alcoholic fatty liver disease. Baseline characteristics were obtained for 1 year prior to index date.

Demographic information collected included age, sex, race and ethnicity, and insurance types (Medicare, MediCal, private health maintenance organization [HMO], private traditional, and other). Self-reported race and ethnicity data were presented according to current reporting guidelines [22]. Baseline Charlson Comorbidity index (CCI) was calculated as a composite of various baseline comorbidities (myocardial infarction, congestive heart failure, peripheral vascular disease, cerebrovascular disease, dementia, chronic pulmonary disease, rheumatic disease, peptic ulcer disease, mild liver disease, diabetes with or without chronic complication, hemiplegia or paraplegia, renal disease, any malignancy, moderate or severe liver disease, and metastatic solid tumor) via ICD-9-CM and ICD-10-CM codes shown in the supplement. A weighted index was calculated with a maximum possible score of 24 [23,24]. Compensated cirrhosis was defined by the diagnosis of cirrhosis and/or portal hypertension manifestations, including thrombocytopenia, splenomegaly, and varices. Hepatic decompensation was defined by diagnosis of jaundice, ascites, hepatic encephalopathy, or variceal bleeding. Hospital characteristics, including academic center, setting (urban or rural), and size (<100, 100 to 399, ≥400 beds) were identified for each patient’s baseline visit.

Primary study end points included (1) all-cause, (2) liver-related, and (3) non-liver-related mortality. Liver-related mortality was indicated by an ICD-10-CM cause of death code for chronic hepatitis, liver fibrosis and cirrhosis, malignant neoplasm of the liver, hepatic failure, and other inflammatory liver diseases and was further stratified into mortality without liver transplant and mortality with and without hepatocellular carcinoma (HCC) at any point over study follow-up. Non-liver-related mortality was classified as any other cause of death not identified above and was further stratified into mortality from cardiovascular disease (CVD), chronic kidney disease (CKD), and non-liver cancer. Secondary study end points were the incidence of specific liver events (decompensation, HCC, and transplantation). The ICD-9-CM and ICD-10-CM codes used for end point identification are in the supplement (Supplementary Appendix).

To reduce the effects of confounding variables, each patient in the bariatric surgery group was matched to 4 nonsurgical controls using propensity score matching (PSM) on age, sex, race and ethnicity, health insurance type, CCI, presence of liver decompensation, and hospital characteristics at baseline. 1:4 PSM was performed via optimal matching in SAS^® version 9.4, in which controls were matched to each treated patient by minimizing the total absolute difference in propensity score across all matches, with a caliper of 0.1 of the standard deviation of the logit propensity scores. As P-values are sensitive to sample size, standardized mean difference (SMD) values were used to evaluate the balancing of potential confounding variables between the two study groups. As per the PSM algorithm, patients with missing data for any of the matched variables were not considered eligible for matching and were excluded. Of the 91,708 patients evaluated for matching, only 1,530 (1.67%) were excluded for missing values.

Survival curves were plotted via the Kaplan-Meier method and the log-rank test was used to compare the differences between the bariatric and non-bariatric groups. Curves for cumulative incidence of specific liver events were plotted via the Fine-Gray method with overall mortality as a competing risk; the Gray’s test was used to compare the differences between bariatric and non-bariatric groups.

The Cox proportional hazards regression model was used to estimate hazard ratios (HRs) and 95% confidence intervals (CIs) for the risk of mortality, and Fine-Gray model was used to estimate subdistribution HRs for the risk of liver complication events with overall mortality as competing risk. The multivariable models were adjusted for age, sex, and baseline CCI. Surgical subgroup analyses also adjusted for year of visit to account for changes in post-operative care over time.

Subgroup analysis explored whether the association between bariatric surgery and mortality differed by surgical approach and across demographic and clinical characteristics. P-values for interaction were calculated to compare the effect of surgery across demographic subgroups.

To confirm the robustness of our findings, we also performed two separate sensitivity analyses by estimating mortality risk after excluding (1) patients with viral hepatitis or alcohol use disorder and (2) patients without cardiometabolic comorbidities (obesity, diabetes, prediabetes, hypertension, hyperlipidemia, hypercholesterolemia, hypertriglyceridemia, or metabolic syndrome).

Statistical significance was defined as a 2-tailed P<0.05 and the SMD threshold was set as 0.10. Continuous variables were reported as mean and standard deviation; categorical variables were reported as percentages. A student’s t-test was used to compare mean age, while a nonparametric Wilcoxon test was used to compare baseline CCI. Chi-square tests were used for descriptive analyses of baseline categorical variables. All statistical analyses were performed using SAS^® version 9.4 (SAS Institute) and R version 4.3.2 (The R Foundation) from July to December 2023. Further details regarding the diagnostic and procedural codes for patient selection, cohort classification, and outcome identification are in the supplement (Supplementary Appendix).

Table 1 describes patient demographics and clinical characteristics at baseline before and after PSM. Before PSM, the patients who underwent bariatric surgery were nearly 5 years younger on average than those who did not (49.93 vs. 54.37 years; P<0.001). Compared to non-bariatric controls, patients who underwent bariatric surgery were more likely to be female (69.90% vs. 48.77%; P<0.0001) and Non-Hispanic (NH) White (55.90% vs. 50.88%; P<0.0001). Treated patients were also more likely to use private or HMO insurance (46.85% vs. 28.11%; P<0.0001) and receive care at a large, academic center in an urban setting (all P<0.05). Clinically, patients who underwent bariatric surgery had a lower CCI (2.18 vs. 3.11; P<0.0001) and were less likely to have liver decompensation at baseline (67.94% vs. 87.33%; P<0.0001). After PSM, the two study groups were similar across demographic and clinical characteristics (all SMD<0.10).

In the PSM cohort, overall mortality was significantly lower for the bariatric group compared to controls at 5 years (24.9% vs. 37.1%; P<0.0001) and 10 years (49.1% vs. 56.1%; P<0.0001; Fig. 2A). Notably, this difference was even more pronounced for liver-related mortality at 5 years (3.3% vs. 14%; P<0.0001) and 10 years (8.3% vs. 22.4%; P<0.0001; Fig. 2B). Consistent findings for liver-related mortality were observed among patients who did not undergo liver transplantation as well as patients with or without HCC at any point during follow-up (all P<0.05; Fig. 2D–F). Overall, non-liver-related mortality risk varied over the course of study follow-up with treated patients at lower risk at 5 years (22.3% vs. 26.9%; P=0.046) and similar risk at 10 years (44.4% vs. 43.5%; P=0.055; Fig. 2C). Such trends were consistent for CVD-related and CKD-related mortality (Fig. 2G, H). Patients who underwent bariatric surgery were observed to have a higher risk of non-liver cancer-related mortality than non-bariatric controls over the study follow-up (10 years: 14.2% vs. 10.8%; P=0.003; Fig. 2I).

With all-cause mortality as a competing risk, the incidence of liver decompensation among treated patients was significantly lower than non-bariatric controls throughout study follow-up (10 years: 45.2% vs. 65.0%; P<0.001; Supplementary Fig. 1A). Although rare, incidence of HCC and liver transplantation were both lower relative to non-bariatric controls over study follow-up (HCC: P<0.001; liver transplant: P=0.006; Supplementary Fig. 1B, C).

In multivariable analysis, bariatric surgery was independently associated with decreased risk of all-cause and cause-specific mortality (Table 2). After adjusting for age, sex, and baseline CCI, bariatric surgery was associated with a 37% reduction in all-cause (adjusted HR [aHR] 0.63, 95% CI 0.56–0.72; P<0.0001) and 76% reduction in liver-related (aHR 0.24, 95% CI 0.17–0.34; P<0.0001) mortality. Reduced liver-related mortality risk was consistently observed among patients who did not undergo liver transplantation and those with or without HCC at any time during follow-up (all P<0.05). Bariatric surgery was also associated with a 19% reduction in non-liver-related mortality (aHR 0.81, 95% CI 0.70–0.93; P=0.0026). Specifically, bariatric surgery was independently associated with a 30% reduction in CVD-related mortality (aHR 0.70, 95% CI 0.53–0.94; P=0.019), 30% reduction in CKD-related mortality (aHR 0.70, 95% CI 0.37–1.32; P=0.269), and 18% elevation in non-liver cancer-related mortality (aHR 1.18, 95% CI 0.92–1.51; P=0.189).

Bariatric surgery was also independently associated with decreased risk of adverse liver outcomes with overall mortality as a competing risk (Supplementary Table 1). After adjusting for age, sex, and baseline CCI, bariatric surgery was associated with a 47% reduction in liver decompensation risk (aHR 0.53, 95% CI 0.43–0.66; P<0.0001), 47% reduction in HCC risk (aHR 0.53, 95% CI 0.27–1.05; P=0.071), and 60% reduction in liver transplantation risk (aHR 0.40, 95% CI 0.21–0.77; P=0.006).

On multivariable analysis adjusting for age, sex, CCI, and year of visit, laparoscopic bariatric surgeries were associated with a 61% reduction in overall mortality (aHR 0.39, 95% CI 0.32–0.47; P<0.0001), but open surgeries were associated with a 24% elevation in overall mortality (aHR 1.24, 95% CI 1.03–1.49; P=0.022; Fig. 3A). Specifically, laparoscopic gastric banding was associated with the largest reduction in all-cause mortality (aHR 0.23, 95% CI 0.12–0.43; P<0.0001; Supplementary Fig. 2A). Both laparoscopic and open surgeries were associated with reduced liver-related mortality, but laparoscopic approaches (aHR 0.16, 95% CI 0.10–0.26; P<0.0001) were associated with greater risk reduction than open approaches (aHR 0.54, 95% CI 0.32–0.90; P=0.019; Fig. 3B).

Additionally, the association between bariatric surgery and reduced overall mortality was consistent across subgroups by age, sex, and race and ethnicity except for groups with small sample sizes (Non-Hispanic Asian and Native American patients; younger age (≤51 years), female, and NH White patients experienced the greatest risk reduction (all P for interaction <0.05; Supplementary Fig. 3A). Patients with a CCI above the average of 2.17 were also at an elevated risk of mortality (aHR 1.26, 95% CI 1.07–1.49) compared to those with CCIs below average (aHR 0.41, 95% CI 0.34–0.49; P for interaction <0.0001). The association between bariatric surgery and reduced liver-related mortality was also consistent across subgroups, but the association only differed significantly across race and ethnicity (P for interaction=0.005) and across CCI (P for interaction=0.031; Supplementary Fig. 3B).

Results of the sensitivity analyses excluding patients with viral hepatitis or alcohol use disorder and without cardiometabolic comorbidities were consistent with the results from the primary analysis showing reduced overall, liver-related, and non-liver-related mortality among the bariatric group compared to non-bariatric controls (Supplementary Tables 2–3). Specifically, among the 8,046 patients analyzed after exclusion for viral hepatitis or alcohol use disorder, bariatric surgery was associated with reduced risk of overall mortality (aHR 0.66, 95% CI 0.57–0.77; P<0.0001). Among the 9,356 patients with a diagnosis of cardiometabolic co-morbidities, bariatric surgery was also associated with reduced risk of overall mortality (aHR 0.64, 95% CI 0.56–0.73; P<0.0001).

There are several strengths of this study. First, this study utilized data from a longitudinal, population-based patient sample that includes nearly all inpatients from the State of California, which offers an ethnically and socioeconomically diverse sample of about 40 million people. Such large-scale data obtained from inpatient records without selection bias of tertiary care center are more generalizable, better informing clinical decision-making. Second, linkage of individual inpatient records to death certificates enabled evaluation of long-term mortality risk. Third, this study utilized PSM to balance the background risks of treated patients and non-bariatric controls, enabling more robust outcome comparison between the two study cohorts. Additionally, we performed extensive subgroup analysis and several sensitivity analyses to confirm that our results are robust.

There are also some limitations to our study. First, administrative records containing ICD codes are prone to miscoding and misclassification, but the HCAI database is validated and maintained by the State of California to minimize reporting errors. Second, because cirrhosis may elevate risk factors that contraindicate bariatric surgery, selection bias may exist. A lack of anthropometric and laboratory data in the HCAI inpatient records also limited the scope of variable adjustment, so residual confounding may exist. However, our study utilized PSM to successfully match patients across all available and relevant demographic and clinical characteristics, including CCI and liver decompensation, in order to reduce selection bias between the two study groups. Additionally, the index dates differed between bariatric recipients and non-bariatric controls, which may have biased the results. However, because patients who underwent bariatric surgery prior to their first MASLD diagnosis were excluded from the study, we would anticipate that the current methodology may bias towards an underestimated benefit of bariatric surgery as index dates for bariatric surgery recipients would be further along in time and thus a potentially shorter survival. There may also be some loss of follow-up if patients died outside of California but there is probably no reason to expect one group to be more likely to die outside of California versus the other. The loss of follow-up is likely minimal given that our study database includes hospitalizations for greater than 98% of all civilian inpatients from all hospitals in California (except for military and federal hospitals) and is linked to the California State Death Statistical Master file, providing mortality data for nearly all individuals residing in California as the vast majority of people will have died in the hospital. Also, it is possible that, compared to non-surgical controls, patients experiencing substantial weight loss from bariatric surgery were more likely to engage in behavioral and lifestyle modifications to sustain this weight loss, which may have biased the results. However, this potential confounder can also be considered an indirect benefit of bariatric surgery. Lastly, the present study concluded in 2019, prior to the adoption of novel and minimally invasive bariatric surgery mechanisms, such as endoscopic sleeve gastroplasties. Additionally, a recent uptick in the consumption of pharmacotherapies, such as GLP-1 agonists, warrants further investigation as such therapies have demonstrated the ability to induce significant weight loss both independently and following bariatric surgery [32,33]. As sufficient time passes, further research including patient-reported outcomes, such as quality of life assessment, is needed for the long-term evaluation of such novel surgical approaches and pharmacotherapies.

Our population-based study with long-term follow-up showed that patients with MASLD and cirrhosis who underwent bariatric surgery through laparoscopic approaches had significantly lower overall, liver-related, and non-liver-related mortality than their non-surgical counterparts. Open approaches to bariatric surgery were associated with reduced risk of liver-related mortality, yet elevated risk of overall mortality. The present study suggests that bariatric surgery should be considered for eligible patients with MASLD and cirrhosis, favoring the use of laparoscopic surgical approaches. Future prospective studies with patient laboratory data are needed to validate our findings.