Dear Editor,
We read with great interest the recent study by Mao et al., published in Clinical and Molecular Hepatology, which emulated a target trial to evaluate the comparative effectiveness of GLP-1 receptor agonists (GLP-1RAs) versus DPP-4 inhibitors in patients with type 2 diabetes and metabolic dysfunction-associated steatotic liver disease (MASLD) [
1]. This timely investigation addresses a clinically important gap in the management of MASLD—a condition for which no approved therapies currently exist that target both hepatic and extrahepatic outcomes. Notably, the authors leveraged a large U.S. administrative claims dataset and employed a robust new-user design with inverse probability of treatment weighting (IPTW) to mitigate confounding, enhancing the internal validity of their findings. While the results are compelling and will likely inform future therapeutic decision-making, we believe that several methodological and interpretive issues merit further discussion to contextualize the study’s conclusions.
First, reliance on administrative claims data to define MASLD and liver-related outcomes undermines diagnostic precision and introduces a high risk of misclassification, particularly in the absence of imaging, histology, or laboratory biomarkers. Although the authors applied a comprehensive exclusion algorithm to minimize the inclusion of patients with other chronic liver diseases, diagnosing MASLD using claims data, particularly in the absence of confirmatory imaging, histology, or serum biomarkers, is inherently limited. Previous studies have shown that International Classification of Diseases codes for MASLD have moderate sensitivity and specificity, particularly in popula-tions with overlapping metabolic comorbidities [
2]. Similarly, ascertaining cirrhosis and hepatocellular carcinoma (HCC) using diagnosis codes alone may miss subclinical or undiagnosed cases. Given the importance of precise disease classification in emulated trials, future studies that incorporate laboratory measures, imaging reports, or natural language processing from electronic health records would improve phenotype accuracy and internal validity.
Second, although the authors employed inverse probability of treatment weighting to balance a wide range of baseline covariates, residual confounding remains a significant concern. For example, GLP-1RA users were more likely to be younger, female, obese, and to have used SGLT-2i or statins. These baseline differences reflect underlying disparities in health behavior and treatment selection, indicating a substantial risk of confounding by indication. Although inverse probability of treatment weighting reduced standardized mean differences to acceptable levels (<0.1), unmeasured factors such as diabetes duration, socioeconomic status, medication adherence, frailty, and provider preference may still introduce bias into the observed associations. Additionally, while the study employed a new-user design, it is unclear if the timing and sequencing of prior treatments were considered, such as the duration of metformin use or insulin exposure. These nuances are important because GLP-1RAs are often prescribed to patients with obesity and suboptimal glycemic control, which could influence liver disease progression or cancer risk independently [
3]. The incorporation of negative control exposures or outcomes—such as non-systemic medications—would provide a more rigorous test of the extent and direction of residual confounding.
Third, the relatively short median follow-up time of approximately two years limits the interpretability of cancerrelated outcomes, which often require longer latency to manifest. As a result, claims-based analyses may inadvertently capture prevalent rather than truly incident cases. Moreover, GLP-1RA users may have experienced more frequent clinical encounters and diagnostic evaluations, increasing the likelihood of cancer detection and introducing potential surveillance bias. To distinguish true pharmacologic effects from differential screening intensity, future studies should consider stratifying cancer incidence by time since treatment initiation, adjusting for healthcare utilization patterns, and employing landmark analyses. Furthermore, the biological plausibility of GLP-1RAs’ anticancer effects may be supported not only by preclinical studies—such as modulation of the PI3K/AKT pathway, apoptosis induction, and immune modulation—but also by emerging clinical evidence linking GLP-1RAs to histologic or imaging-based improvements in hepatic steatosis and fibrosis. For instance, randomized trials using magnetic resonance imaging–proton density fat fraction and elastography have demonstrated that GLP-1RA treatment reduces hepatic fat content and may attenuate fibrosis progression in patients with MASLD [
4,
5]. Given that hepatic steatosis and fibrotic remodeling are known risk factors for hepatocarcinogenesis, these findings lend mechanistic support to the observed reduction in HCC incidence. However, the cancer-specific effects of GLP-1RAs and the durability of these hepatic improvements remain to be fully elucidated in prospective, biomarker-integrated studies.
Fourth, the differential effects of GLP-1RA across subgroups imply that patient selection and disease stage meaningfully influence therapeutic benefit. Notably, the protective associations for HCC and non-liver cancer were stronger among individuals without cirrhosis at baseline. This implies that earlier intervention may be more effective before advanced liver injury occurs. This is biologically plausible given that decompensated cirrhosis can limit drug metabolism and reduce responsiveness to metabolic interventions. Furthermore, subgroup analyses indicated a greater benefit of GLP-1RAs for cardiovascular disease outcomes among statin users, suggesting the possibility of additive or synergistic effects of cardiometabolic therapies. These findings raise important clinical questions. For example, should GLP-1RAs be prioritized for MASLD patients without advanced fibrosis, given the stronger associations observed in this subgroup? It also remains uncertain whether combining agents such as GLP-1 receptor agonists and SGLT-2 inhibitors could further enhance the mitigation of extrahepatic outcomes. In addition, treatment allocation may be optimized through biomarker-guided strategies that incorporate measures of hepatic fat fraction or fibrosis severity, potentially enabling a more personalized approach to therapy.
Lastly, although the authors conducted extensive sensitivity analyses, including per-protocol analyses, competing risk models, and 1:1 propensity score matching, the interpretation of absolute risk reduction and clinical significance requires further clarification. For instance, although the hazard ratios for HCC and chronic kidney disease were below 0.7, the absolute event rates were relatively low, especially for HCC (0.3% over approximately two years). Presenting absolute risk differences or the number needed to treat would provide more clinically meaningful insights into the public health impact of GLP-1RAs. Similarly, although chronic obstructive pulmonary disease was used as a negative control outcome, the lack of association does not completely eliminate the possibility of residual bias, especially since chronic obstructive pulmonary disease risk may not correlate with most baseline confounders that were adjusted for in the main model. To strengthen the assessment of systematic bias, future analyses could incorporate additional negative control outcomes with more comparable pathophysiological mechanisms, such as osteoarthritis or urinary tract infection.
In conclusion, this study provides valuable real-world evidence supporting the potential of GLP-1RAs to reduce both hepatic and systemic complications in patients with type 2 diabetes and MASLD. The use of a new-user design, IPTW adjustment, and multiple sensitivity analyses represents a methodological strength that enhances confidence in the observed associations. At the same time, future prospective studies with more granular phenotyping and longer follow-up are warranted to validate these findings and support biomarker-guided treatment strategies in this population.
FOOTNOTES
-
Authors’ contributions
Chunyan Wang and Xu Han wrote the manuscript, Yinyan Li provided methodological and revised the manuscript.
-
Conflicts of Interest
The authors have no conflicts to disclose.
Abbreviations
inverse probability of treatment weighting
metabolic dysfunction-associated steatotic liver disease
REFERENCES
- 1. Mao X, Zhang X, Lai R, Cheung KS, Yuen MF, Cheung R, et al. Glucagon-like peptide 1 receptor agonist and reduced liver and non-liver complications in adults with type 2 diabetes and metabolic dysfunction-associated steatotic liver disease: a target trial emulation study. Clin Mol Hepatol 2025;31:1084-1099.
- 2. Corey KE, Kartoun U, Zheng H, Shaw SY. Development and validation of an algorithm to identify nonalcoholic fatty liver disease in the electronic medical record. Dig Dis Sci 2016;61:913-919.
- 3. Davies MJ, Aroda VR, Collins BS, Gabbay RA, Green J, Maruthur NM, et al. Management of hyperglycemia in type 2 diabetes, 2022. A Consensus Report by the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD). Diabetes Care 2022;45:2753-2786.
- 4. Newsome PN, Buchholtz K, Cusi K, Linder M, Okanoue T, Ratziu V, et al. A placebo-controlled trial of subcutaneous semaglutide in nonalcoholic steatohepatitis. N Engl J Med 2021;384:1113-1124.
- 5. Armstrong MJ, Gaunt P, Aithal GP, Barton D, Hull D, Parker R, et al. Liraglutide safety and efficacy in patients with non-alcoholic steatohepatitis (LEAN): a multicentre, double-blind, randomised, placebo-controlled phase 2 study. Lancet 2016;387:679-690.
Citations
Citations to this article as recorded by
