Comparative Outcomes of Glucagon-Like Peptide-1 Receptor Agonists versus Sodium-Glucose Cotransporter-2 Inhibitors in Patients with Cirrhosis

Introduction

Cirrhosis represents the end stage of chronic hepatic injury and remains a leading contributor to global morbidity and mortality. Worldwide, liver cirrhosis accounted for over 1.4 million deaths in 2019, and the U.S. burden has risen markedly over the last decade, with an estimated 2.2 million American adults living with cirrhosis as of 2023 [¹]. Once patients progress from compensated to decompensated disease, defined by development of ascites, spontaneous bacterial peritonitis (SBP), variceal hemorrhage, hepatic encephalopathy, or hepatorenal syndrome, survival and quality of life deteriorate substantially [²]. Despite continued refinement of supportive care and complication management, pharmacotherapies that meaningfully alter the natural history of cirrhosis are limited. Therefore, identifying agents that can both reduce the risk of decompensation and prolong survival remains a clinical priority.

Metabolic dysfunction is now established as a central driver of cirrhosis epidemiology. Insulin resistance and type 2 diabetes mellitus (T2DM), core features of the metabolic syndrome, intersect with adiposity and steatosis to fuel progressive liver injury along the metabolic dysfunction – associated steatotic liver disease (MASLD)/steatohepatitis (MASH) continuum [³,⁴,⁵]. The clinical overlap between diabetes and advanced liver disease is striking: roughly one-third of all patients with cirrhosis carry a T2DM diagnosis, and this prevalence rises above 50% among MASLD- and cryptogenic-related cirrhosis [⁶]. Conversely, MASLD is documented in the majority of patients with T2DM, with a meaningful subset showing fibrotic MASH [⁷]. This bidirectional relationship has elevated the question of which glucose-lowering therapies are safest and most beneficial in patients with established liver disease.

Two emerging drug classes have drawn particular attention in this regard. Glucagon-like peptide-1 receptor agonists (GLP-1RAs) lower body weight, improve glycemia, and have shown improvements in MASH-related histologic and biomarker endpoints in non-cirrhotic populations [⁸]; observational data in patients with both cirrhosis and T2DM further suggest reductions in decompensation events compared with other antidiabetic therapies [⁹,¹⁰,¹¹]. Sodium-glucose cotransporter-2 inhibitors (SGLT2i), through insulin-independent glycemic effects and pleiotropic cardiometabolic actions, have been linked to favorable changes in hepatic enzymes and steatosis-related measures, and observational analyses have suggested reduced rates of MASLD progression and liver-related complications versus alternative regimens [¹²,¹³,¹⁴]. Despite these parallel signals, head-to-head data comparing the two classes within an established cirrhosis population are limited, and clinicians often choose between them based on extrapolated cardiometabolic indications rather than direct hepatic evidence.

To address this gap, we performed a multicenter retrospective analysis within the TriNetX Research Network, evaluating 1-year clinical outcomes among adults with cirrhosis who initiated either a GLP-1RA or an SGLT2i after their cirrhosis diagnosis. The objective was to inform pharmacotherapeutic decision-making in a high-risk population in which both efficacy and safety considerations carry particular weight.

Methods

Results

Discussion

In this large multicenter real-world analysis of adults with cirrhosis, initiation of a GLP-1 receptor agonist was associated with more favorable 1-year outcomes than initiation of an SGLT2 inhibitor following 1:1 propensity score matching across demographics, comorbidities, concomitant medications, and laboratory parameters. Compared with the SGLT2i arm, GLP-1RA users had lower 1-year all-cause mortality and a lower composite cirrhosis decompensation event rate, with the latter signal driven principally by fewer ascites-related events (ascites diagnoses and paracentesis) and fewer SBP episodes. Other decompensation phenotypes (variceal bleeding, hepatic encephalopathy, and hepatorenal syndrome) did not differ meaningfully between groups. AKI was less frequent with GLP-1RAs, while rates of hypoglycemia and pancreatitis were comparable. Together, these observations are consistent with, but do not establish an incremental short-term advantage for GLP-1RAs over SGLT2i for selected patients with cirrhosis, particularly with respect to portal-hypertension-related events and renal safety. A sensitivity analysis restricted to patients with concomitant conditions yielded directionally concordant and numerically stronger results, further supporting the robustness of the primary findings.

These findings align with an expanding observational literature linking GLP-1RA exposure to fewer major liver events and improved survival in chronic liver disease, including cirrhosis. In a population-based cohort of patients with T2DM and cirrhosis, GLP-1RA use was associated with lower mortality and lower decompensation risk relative to nonuse [¹⁵]. In MASLD-related cirrhosis specifically, GLP-1RA initiation has been associated with lower risks of adverse liver outcomes (including decompensation events) and other clinically important endpoints such as liver transplantation [¹⁶]. Beyond strictly cirrhosis-only cohorts, a Scandinavian registry-based analysis demonstrated lower risk of serious liver events (a composite of incident compensated/decompensated cirrhosis and HCC) among GLP-1RA initiators with T2DM [¹⁷].

The direct head-to-head architecture of our analysis adds to this literature. Most prior work has compared each class with nonuse or with unrelated antidiabetic agents, leaving uncertain how GLP-1RAs and SGLT2i compare with each other in cirrhosis, a setting in which both are increasingly considered. Our results suggest that the GLP-1RA signal for reduced mortality and decompensation reported in earlier observational work persists when the comparator is another cardiometabolic class with its own emerging hepatic profile.

The observed pattern of benefit, concentrated in less ascites, paracentesis, and SBP, is mechanistically plausible. Ascites-related events reflect the severity of portal hypertension and systemic inflammation, and SBP commonly arises from immune dysfunction and gut – barrier compromise in decompensated cirrhosis [¹⁸]. The separation favoring GLP-1RAs may relate to improvements in insulin resistance, modulation of metabolic inflammation [¹⁹], and downstream effects on the gut – liver axis that could reduce bacterial translocation and infection-related complications [²⁰]. Although our dataset cannot directly probe these mechanisms (we do not have weight trajectories, hepatic venous pressure gradient, or microbiome data), the outcome profile is more consistent with a portal-hypertension-dominant benefit than with a uniform effect across all decompensation phenotypes.

We did not observe meaningful differences in variceal bleeding or hepatic encephalopathy. These outcomes are influenced by factors that are unlikely to be modified rapidly by metabolic therapy, including endoscopic surveillance and banding practice patterns, precipitating events such as infection or gastrointestinal bleeding, and concomitant psychoactive medication use.

The lower AKI incidence among GLP-1RA users is clinically important in this population, where AKI is a major driver of morbidity and mortality and is typically multifactorial, encompassing hypovolemic injury, infection-associated injury, nephrotoxin exposure, and hepatorenal physiology [²¹]. Although SGLT2 inhibitors are widely favored for kidney protection in non-cirrhotic diabetes populations [²²], the cirrhosis setting is physiologically distinct: frequent diuretic use, reduced effective arterial volume, and circulatory dysfunction may attenuate or alter the SGLT2i renal-protective profile in this group [²³]. Our findings imply that direct extrapolation of diabetic-kidney-disease paradigms to cirrhosis may be incomplete and warrants prospective evaluation.

With respect to safety, similar 1-year rates of hypoglycemia and acute pancreatitis are reassuring but should be interpreted cautiously. EHR-based ascertainment may miss milder events managed in the outpatient setting, and we could not reliably capture class-specific concerns such as gastrointestinal intolerance, gallbladder disease, volume depletion symptoms, or diabetic ketoacidosis when these are not consistently coded.

5. Strengths and Limitations

The principal strengths of this study include the large sample size, the multicenter EHR base, the new-user active-comparator architecture, and 1:1 propensity score matching across more than 50 covariates spanning demographics, comorbidities, concomitant medications, procedures, and laboratory parameters. Together, these features improve comparability between cohorts and enhance generalizability to routine U.S. practice populations.

Several limitations require explicit acknowledgment, and many relate directly to features of the TriNetX platform.

First, residual confounding is unavoidable in any observational analysis of this kind. Despite matching on more than 50 covariates, confounding by indication and prescribing channeling (preferential selection of one class based on frailty, obesity severity, heart failure status, or clinician comfort) likely persist. The strong pre-match imbalances we observed in heart failure, CKD, ischemic heart disease, anticoagulation, diuretics, and obesity make this concern explicit. Although matching reduced these imbalances to SMD <0.1 for the great majority of covariates, residual imbalance persisted for BMI (SMD 0.154). Because BMI is mechanistically linked to both treatment selection and liver-related outcomes, the BMI signal and its potential contribution to the observed differences cannot be fully ruled out by matching alone.

Second, formal measures of liver disease severity, such as Child-Pugh class, MELD, hepatic venous pressure gradient, and endoscopic findings, are not consistently available within TriNetX as discrete structured variables and could not be incorporated as covariates. However, we indirectly measured these variables by balancing our cohorts on components of the Child-Pugh class and MELD scores to overcome this limitation.

Third, the cirrhosis case definition relies on ICD-10-CM K74 (which encompasses both fibrosis and cirrhosis) plus K70.3 (alcoholic cirrhosis). The K74 umbrella may include some patients with non-cirrhotic fibrosis, which would dilute rather than amplify between-cohort differences. We did not perform a sensitivity analysis restricted to more specific cirrhosis-only codes (e.g., K74.6x, K70.3, K71.7, K72.x), and recommend such analyses in future replication work.

Fourth, exposure was defined by prescription records rather than confirmed dispensing or adherence. Dose, titration, discontinuation, and post-index switching between classes are not fully verifiable in the platform. Our analysis approach is intention-to-treat with respect to the index class. Because some patients may have discontinued or crossed over to the comparator class during follow-up, the comparison may underestimate true on-treatment effect sizes.

Fifth, we did not implement formal competing-risk methods for non-fatal outcomes. Higher mortality in one arm could in principle alter the apparent incidence of nonfatal complications such as ascites, SBP, HRS, and AKI, since a patient who dies cannot subsequently develop these events. Although the absolute mortality difference between arms ($\approx 1$ percentage point at 1 year) is modest. The secondary-outcome separations are larger and in the same direction, secondary results should be interpreted with this consideration in mind.

Sixth, etiology heterogeneity is real. Although baseline etiology was captured and matched, the cohort combines MASLD/MASH-related, alcohol-associated, viral, and autoimmune cirrhosis, and the mechanistic rationale for GLP-1RA hepatic benefit is strongest in metabolic phenotypes. A class-level effect observed in a heavily metabolic population should not be uncritically generalized to all cirrhosis. Etiology-stratified analyses (separating MASLD/MASH cirrhosis from non-metabolic etiologies) are an appropriate next step.

Seventh, our laboratory analyses use the most recent value within the follow-up window, which captures point estimates rather than longitudinal trajectories. Several of these distributions (notably AST and ALT) are right-skewed, so means and t-tests should be regarded as descriptive.

Finally, the TriNetX "Deceased" indicator captures in-network mortality recorded in the EHR and is known to have lag and incompleteness for patients who die outside of contributing health systems; the absolute mortality rates reported here may therefore underestimate true mortality, although this misclassification should be roughly non-differential between cohorts after matching.

Implications and Future Directions

Despite these constraints, the consistent directionality of effect across mortality and ascites-related decompensation events supports the hypothesis that GLP-1RAs may offer incremental short-term benefit over SGLT2i for selected patients with cirrhosis, particularly those at high risk for portal-hypertension-related complications. The concordance between the primary analysis and the T2DM-restricted sensitivity analysis strengthens confidence in these associations, but we emphasize that the findings remain hypothesis-generating rather than definitive. Future work should ideally: (i) incorporate validated severity scoring (MELD/Child-Pugh) and hepatic venous pressure gradient where available; (ii) perform etiology-stratified analyses, especially separating MASLD/MASH-related cirrhosis from non-metabolic etiologies; (iii) pursue prospective cirrhosis-specific comparative trials, ideally stratified by compensated versus decompensated status, to determine causality and to identify the patient phenotypes most likely to benefit.

Conclusion

In this large real-world retrospective cohort of adults with cirrhosis, GLP-1 receptor agonist initiation was associated with lower 1-year all-cause mortality and fewer ascites-related decompensation and AKI events than SGLT2 inhibitor initiation. These observational signals require prospective confirmation and should not on their own redirect clinical practice. Cirrhosis-specific trials and well-designed comparative-effectiveness studies that incorporate severity scoring and etiology stratification are needed to determine whether the differences observed here reflect a true class effect.