Methods
The protocol for this systematic review and network meta-analysis has been registered with PROSPERO (CRD42024625681). This study followed the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) 2020 statement and the extension for network meta-analyses (PRISMA-NMA).5,6
Data sources and search strategy
We searched PubMed, Web of Science, the Cochrane Central Register of Controlled Trials (CENTRAL), Scopus, and Embase for randomized controlled trials (RCTs) in adults with MASLD/MASH (previously known as NAFLD/MAFLD/NASH) from inception to November 20, 2024, as well as abstracts presented at the annual meetings of the American Association for the Study of Liver Diseases (AASLD) and the European Association for the Study of the Liver (EASL) between 2020 and 2024. Six paired reviewers (WJN, XYW, SSZ, XB, ML, and ZRJ) independently conducted the first-round screening based on abstracts and titles, cross-checking their results sequentially. Any studies with uncertainty or inconsistencies were retained for further review. Subsequently, the second-round screening was conducted independently by the same six reviewers using full texts and/or trial protocols, with results also checked in sequence. Any remaining inconsistencies were resolved by a seventh reviewer (JL). Comprehensive search strategies for each of the five databases are provided in the Supplementary File 1.
Study selection
The inclusion criteria were as follows: (1) RCTs enrolling participants aged 18 years or older; (2) MASLD diagnosed by liver histology and/or magnetic resonance imaging proton density fat fraction (MRI-PDFF); (3) interventions targeting THR-β agonists, GLP-1 RAs, GLP-1/glucose-dependent insulinotropic polypeptide (GIP) dual RAs, GLP-1/GIP/glucagon (GCG) triple RAs, GLP-1/GCG dual RAs, farnesoid X receptor (FXR) agonists, FGF21 analogues, FGF19 analog, acetyl-coenzyme A carboxylase (ACC) inhibitors, and PPAR agonists or PPAR-related isomers.
The exclusion criteria included: (1) review articles, meta-analyses, letters, comments, editorials, animal studies, practice guidelines, consensus statements, case reports, protocols, or other forms of gray literature, including preprints not formally published, unpublished reports, regulatory documents, theses, or dissertations. Conference proceedings were not routinely included, except for abstracts presented at the annual meetings of AASLD and EASL between 2020 and 2024; (2) observational, cross-sectional, case-control, or single-arm studies; (3) phase 1 trials; (4) trials enrolling a majority of participants with cirrhosis (F4); (5) trials using lifestyle interventions as the comparator; (6) trials reporting treatments involving traditional Chinese medicine, probiotics, or lifestyle interventions (diet and/or physical exercise); (7) trials conducted in pediatric or adolescent populations (younger than 18 years); and (8) trials written in languages other than English; or (9) studies without available data for extraction and analysis.
Data extraction
Data extraction from eligible studies included study characteristics (title, authors, year of publication, trial registration number, and country/region), participant baseline characteristics, including age, sex, sample size, body mass index (BMI), glycated hemoglobin (HbA1c), LFC assessed by MRI-PDFF, liver stiffness measurement (LSM) assessed by magnetic resonance elastography, prevalence of diabetes, hypertension, and dyslipidemia, distributions of fibrosis stages (F1–F3) diagnosed by liver biopsy, intervention details (medication name, therapeutic target, duration, dose, dosing frequency, and route of administration), relevant liver histological outcomes at the end of treatment, other outcome changes from baseline, and prespecified safety outcomes. Continuous variables were extracted only when each arm included sample size, mean, and standard deviation (SD), or when SD could be calculated. For each intervention, the dose and frequency associated with the most favorable estimate for the corresponding outcome were selected. If multiple published reports were identified from the same clinical trial, the most recent and most complete dataset was used for data extraction and analysis. When certain outcomes were not available in the most recent publication, relevant data from earlier reports of the same trial were used, ensuring each outcome was included only once per trial in the meta-analysis. Plot Digitizer was used to extract values when data were presented only in figures.7,8
Outcomes
The primary outcomes were: (1) fibrosis improvement of at least one stage without worsening of steatohepatitis; and (2) steatohepatitis resolution without worsening of fibrosis. Secondary outcomes included relative and absolute changes from baseline in LFC assessed by MRI-PDFF; absolute changes in liver enzymes (alanine aminotransferase [ALT] and aspartate aminotransferase [AST]); body weight; lipid and glucose parameters (triglycerides, low-density lipoprotein cholesterol [LDL-C], high-density lipoprotein cholesterol [HDL-C], total cholesterol [TC], HbA1c, and fasting blood glucose [FBG]); as well as non-invasive tests (NITs), including the enhanced liver fibrosis (ELF) score, N-terminal type III collagen propeptide (PRO-C3), and LSM assessed by vibration-controlled transient elastography (VCTE). Safety outcomes included any adverse events (AEs), serious adverse events (SAEs), discontinuation due to AEs, three common gastrointestinal (GI) events (nausea, diarrhea, and vomiting), cardiovascular events, and pruritus (specifically for FXR agonists).
Quality assessment
The Cochrane Risk of Bias tool for randomized trials (version 2.0, RoB 2) was used to assess risk of bias in eligible studies.9 RoB 2 includes five domains: randomization process, deviations from intended interventions, missing outcome data, measurement of the outcome, and selection of the reported result. Overall risk of bias was categorized as low, some concerns, or high. Two investigators (WJN and SSZ) independently assessed each domain, and inconsistencies were resolved through discussion. Funnel plots were used to evaluate publication bias. The Confidence in Network Meta-Analysis (CINeMA) framework was used to assess certainty of evidence.10,11 This framework evaluates six domains: within-study bias, reporting bias, indirectness, imprecision, heterogeneity, and inconsistency. Certainty of evidence was classified as high, moderate, low, or very low.
Data synthesis and analysis
For lipid and glucose variables, mean and SD of absolute changes were converted from millimoles per liter (mmol/L) to milligrams per deciliter (mg/dL). When studies reported separate results for participants with and without diabetes for glucose-related variables, the formula provided below was used to calculate combined results according to Cochrane Handbook guidance12:
(N1-1)SD12+(N2-1)SD22+N1N2N1+N2(M12+M22-2M1M2)N1+N2-1
Network meta-analyses of RCTs were conducted using Stata version 17 (StataCorp LLC) within a frequentist framework. Liver histological and safety outcomes, treated as binary variables, were measured using odds ratios (ORs). Absolute risk differences (ARDs) for histological outcomes were derived from ORs estimated in the network meta-analysis. The pooled placebo event risk, calculated as the weighted mean event rate across placebo arms of included trials for each outcome, was used as the assumed comparator risk (ACR).13 The corresponding intervention risk (Riskintervention) was calculated as:
Corresponding intervention risk=OR×ACR1-ACR+(OR×ACR)
The ARD was then calculated as: ARD = Riskintervention − ACR. A positive ARD indicated a higher event probability in the intervention group than in the placebo group, representing an absolute increase in treatment benefit. The 95% confidence intervals (CIs) for ARDs were derived by applying the same transformation to the upper and lower limits of the OR 95% CIs. Secondary continuous outcomes were calculated as mean differences with 95% CIs. Subgroup analyses by treatment duration and leave-one-out sensitivity analyses were performed to evaluate liver histological improvement. Network diagrams and surface under the cumulative ranking curve (SUCRA) plots were generated for ranking assessments. SUCRA values range from 0% to 100%, with higher values indicating a greater probability of ranking among more effective interventions, and lower values indicating less effective interventions. Rankograms were also generated to display full ranking probability distributions across all possible ranks for both primary outcomes and for relative and absolute changes in LFC, with rank 1 indicating the most favorable outcome. Global inconsistency was assessed using an inconsistency model when network geometry allowed. Inconsistency testing was not performed when insufficient indirect evidence was available. Evidence contribution matrices were generated for the two primary outcomes and for relative and absolute changes in LFC to quantify the contribution of each direct comparison to network estimates. Meta-regression was conducted to explore the impact of treatment duration on histological outcomes. Forest plots were generated using R software version 4.3.1 (http://www.R-project.org ) and GraphPad Prism version 10.1.2.
Role of the funding source
The funding agencies had no role in the design, data collection, data analysis, data interpretation, or writing of the study or manuscript, nor in the decision to publish.
Results
Characteristics of eligible studies
A total of 18,237 records were retrieved from five databases, with 1,038 studies selected for full-text screening. Finally, 44 studies (11,410 randomized participants, 11 targets, 33 medications) that met the study eligibility criteria were included (Fig. 1). A summary of these eligible studies is provided in Supplementary Table 1. At baseline, the mean age of participants across the included RCTs ranged from 37.0 to 59.4 years, indicating that most study populations were middle-aged. The reported mean or median BMI generally ranged from 28.4 to 39.7 kg/m2, except for one study that enrolled participants with a mean BMI of 18–22 kg/m2.14 Within individual RCTs, the distributions of key baseline characteristics, including sex, metabolic comorbidities, and fibrosis stage, were generally balanced across randomized treatment arms. However, between-trial variations were observed in baseline BMI, metabolic profiles, fibrosis stage distributions, and intervention durations. A summary of key baseline characteristics of the study populations is provided in Supplementary Table 2.
Analysis for PPAR agonists included the most medications (pioglitazone, pemafibrate, lanifibranor, elafibranor, rosiglitazone, saroglitazar, and chiglitazar). PXL065, a deuterium-stabilized R-pioglitazone (d-(R)-pioglitazone), was included as a separate target for analysis, as it is developed with deuterium added to its chiral center and does not activate PPAR.15 Analysis of THR-β agonists included two medications (resmetirom and VK2809). Analysis of FXR agonists included five medications (obeticholic acid, EDP-305, vonafexor, cilofexor, and FXR314), while analysis of the FGF19 analogue included only aldafermin (also known as NGM282 or M70). Analysis of FGF21 included four medications (pegozafermin, pegbelfermin, efruxifermin, and efimosfermin alfa [BOS-580]), and analysis of ACC inhibitors included two medications (PF-05221304 and firsocostat [GS-0976]). As for GLP-1 RA, dulaglutide, semaglutide, and liraglutide were included. Analysis of GLP-1/GCG dual RA included four medications (efinopegdutide, pemvidutide, survodutide, and cotadutide); analysis of GLP-1/GIP dual RA included tirzepatide; and analysis of GLP-1/GCG/GIP triple RA included retatrutide.
Regarding route of administration, PPAR agonists, d-(R)-pioglitazone, THR-β agonists, FXR agonists, and ACC inhibitors were administered orally once daily, whereas FGF19 analogue, FGF21 analogue, GLP-1 RA, GLP-1/GIP dual RA, and GLP-1/GCG/GIP triple RA were administered subcutaneously. The duration of intervention in the included studies ranged from 12 to 72 weeks. Specifically, the intervention durations of the included RCTs contributing to histological outcomes ranged from 16 to 72 weeks.
Risk of bias, publication bias, and global consistency
The details of the risk of bias assessment for each included study are presented in Supplementary Figure 1 and Supplementary Table 3. Briefly, 84.0% of included studies had a low risk of bias for the randomization process. The majority of studies [34 (77.2%), 41 (93.1%), 42 (95.4%), and 42 (95.4%)] were considered at low risk for deviations from intended interventions, missing outcome data, measurement of outcomes, and selection of the reported result, respectively. Funnel plots showed no significant asymmetry, suggesting no evidence of publication bias (Supplementary Figs. 2 and 3). Global inconsistency tests were largely not applicable for most outcomes due to insufficient indirect evidence, and no inconsistency was detected where testing was possible for relative change from baseline in LFC, liver enzymes, and certain safety outcomes (Supplementary Table 4).
Liver histology improvement
Fibrosis improvement at least one stage without steatohepatitis worsening
A total of 17 studies with 3,300 participants reported the outcome of fibrosis improvement of at least one stage without steatohepatitis worsening. Nine targets comprising 14 medications were included for comparison (Fig. 2A). The network was predominantly placebo-centered, with interventions connected through placebo and no direct head-to-head comparisons among active interventions. Compared with placebo, d-(R)-pioglitazone (OR: 4.75, 95% CI: 1.21 to 18.58), FGF21 analogue (OR: 3.52, 95% CI: 1.81 to 6.85), GLP-1/GCG dual RA (OR: 3.40, 95% CI: 1.53 to 7.56), and GLP-1/GIP dual RA (OR: 3.01, 95% CI: 1.29 to 7.02) showed favorable signals, with ORs greater than 3 for this outcome (Fig. 3A). The SUCRA scores showed that d-(R)-pioglitazone (SUCRA: 79.3), FGF21 analogue (SUCRA: 71.9), and GLP-1/GCG dual RA (SUCRA: 69.5) ranked higher than the other interventions, followed by PPAR agonists (SUCRA: 61.3), GLP-1/GIP dual RA (SUCRA: 59.7), and FXR agonists (SUCRA: 55.5) (Supplementary Table 5.1). Because SUCRA summarizes ranking probabilities rather than the magnitude of effect, rankings should be interpreted together with effect estimates, 95% CIs, and the certainty of evidence. The rankograms showed a ranking pattern broadly aligned with the SUCRA estimates, while also illustrating uncertainty in ranking probabilities across targets (Supplementary Fig. 4A).
In the CINeMA assessment, the certainty of evidence for most target comparisons was rated as low or very low, mainly due to small sample sizes, single-trial evidence, imprecision, and limited direct comparative evidence. Comprehensive comparisons between targets and the CINeMA assessment are provided in Supplementary Table 6.1. In addition, given the placebo-centered network structure, an evidence contribution matrix was generated. The results showed that active-versus-placebo estimates were mainly informed by the corresponding direct placebo-controlled comparisons, whereas most active-versus-active estimates were primarily derived from indirect evidence through the common placebo comparator (Supplementary Table 7.1).
We further evaluated the ARDs across targets. Using the pooled placebo event risk of 17.41% as the ACR, the estimated ARDs across targets ranged from 12.30 to 32.30 percentage points versus placebo, indicating an absolute increase in the probability of achieving the histological outcome (Supplementary Table 8.1).
As for medications, 30 mg/day pioglitazone16 ranked highest, followed by 44 mg every two weeks of pegozafermin17 and 50 mg/week efruxifermin18 (Supplementary Fig. 5A). However, liver biopsy results from efruxifermin were flagged for high risk of bias due to missing outcome data caused by the COVID-19 pandemic and patient-specific issues.18
Steatohepatitis resolution without fibrosis worsening
A total of 21 studies evaluating 9 targets in 3,768 participants were analyzed for the outcome of steatohepatitis resolution without fibrosis worsening (Fig. 2B). Compared with placebo, GLP-1/GIP dual RA, the highest-ranking target according to SUCRA (91.7), had an OR of 14.33 (95% CI: 3.39 to 60.57) for achieving this outcome, followed by GLP-1/GCG dual RA (OR: 10.67, 95% CI: 3.07 to 37.04) (Supplementary Table 5.2). All of these targets, except for d-(R)-pioglitazone, showed favorable efficacy signals compared with placebo (Fig. 3B and Supplementary Table 6.2). In addition, because the LEAN trial enrolled a small proportion of patients with cirrhosis, as assessed by the Kleiner scoring system,19 a sensitivity analysis was conducted excluding liraglutide. The ranking order remained unchanged, and a similar SUCRA value (59.7) was observed for GLP-1 RA compared with placebo (OR: 4.35, 95% CI: 1.91–9.87). In the CINeMA assessment, many comparisons were downgraded due to imprecision, as reflected by wide 95% CIs and limited sample sizes. Detailed comparisons and certainty of evidence are provided in Supplementary Table 6.2. The rankogram was generally consistent with the SUCRA-based ranking pattern (Supplementary Fig. 4B). The contribution matrix showed that most active-versus-active estimates were primarily informed by indirect evidence through placebo, highlighting the limited direct comparative evidence within the network and suggesting caution when comparing active targets (Supplementary Table 7.2).
In addition, using the pooled placebo event risk of 15.59% as the ACR, the estimated ARDs across targets ranged from 5.10 to 57.00 percentage points, suggesting varying absolute increases in the probability of achieving steatohepatitis resolution (Supplementary Table 8.2).
For specific medications, 15 mg/week pegozafermin17 and 15 mg/week tirzepatide20 showed favorable signals compared with placebo (Supplementary Fig. 5B).
LFC as assessed by MRI-PDFF
There were 21 studies involving 2,296 participants, and 16 studies comprising 1,127 participants were analyzed to evaluate the relative and absolute changes in LFC from baseline, respectively. For relative change in LFC from baseline, 11 targets were compared with placebo in double-blind trials (Fig. 2C). GLP-1/GIP/GCG triple RA demonstrated a favorable effect estimate (mean difference: −81.40%, 95% CI: −109.87 to −52.93, SUCRA: 97.5) (Fig. 3C and Supplementary Table 5.3). GLP-1 RA (semaglutide) and GLP-1/GCG dual RA (efinopegdutide) were compared in an open-label RCT,21 with low to very low certainty of evidence (Supplementary Table 6.3). Regarding absolute change in LFC from baseline, GLP-1/GCG dual RA was associated with a relatively larger estimated treatment effect (mean difference: −14.50, 95% CI: −22.12 to −6.88, SUCRA: 94.3) (Fig. 3D and Supplementary Table 5.4).
For both outcomes, the rankograms showed patterns generally aligned with the corresponding SUCRA estimates (Supplementary Figs. 4C and D). The contribution matrix indicated that the evidence structure was not purely star-shaped, as a few active-comparator trials contributed to selected network estimates. However, the network remained predominantly placebo-centered, and most active-versus-active estimates were still primarily informed by indirect evidence via placebo (Supplementary Tables 7.3 and 7.4).
Among specific medications, 12 mg/week retatrutide,22 70 mg/week efruxifermin,18 and 1.8 mg/week pemvidutide23 exhibited favorable signals in reducing relative change in LFC from baseline (Supplementary Fig. 5C), and 70 mg/week efruxifermin,18 1.8 mg/week pemvidutide,23 and 27 mg/week pegozafermin17,24 ranked higher for reducing absolute change in LFC from baseline (Supplementary Fig. 5D).
Liver enzyme improvement
The effects of different targets on ALT improvement were analyzed using data from 23 studies with 3,176 participants. Combination therapy of GLP-1 RA + ACC inhibitor + FXR agonist yielded a favorable point estimate for reducing ALT levels (mean difference: −46.31, 95% CI: −67.96 to −24.67; SUCRA: 91.7) (Fig. 4A and Supplementary Table 6.5). For AST reduction, data from 3,058 participants in 21 studies were analyzed. GLP-1 RA + ACC inhibitor had a comparatively favorable point estimate (mean difference: −30.54, 95% CI: −49.48 to −11.61; SUCRA: 89) (Fig. 4B and Supplementary Table 6.6). Additional details of target and medication comparisons for ALT and AST are provided in Supplementary Tables 6.5–6.6 and Supplementary Figures 6A–B.
Body weight
A total of 11 targets in 21 studies involving 3,375 participants were analyzed to evaluate the effects of interventions on body weight. GLP-1/GCG dual RA showed a favorable point estimate for body weight reduction (mean difference: −12.51, 95% CI: −15.30 to −9.72; SUCRA: 99.4) (Fig. 4C and Supplementary Table 5.7). Comparisons of targets for body weight reduction are detailed in Supplementary Table 5.7. These findings were generally aligned with the results for specific interventions (Supplementary Fig. 6C).
Metabolic profiles
Lipid parameters
TG, LDL-C, HDL-C, and TC were selected as key parameters for assessing improvements in lipid profiles. Among these targets, PPAR agonists showed a favorable point estimate for reducing TG levels (mean difference: −53.36, 95% CI: −72.98 to −33.74; SUCRA: 86.2). For LDL-C reduction, GLP-1 RA displayed the highest SUCRA value (78.7). However, the reduction was not statistically significant compared with placebo (mean difference: −21.13, 95% CI: −52.64 to 10.38). The FGF21 analogue showed a favorable point estimate for increasing HDL-C (mean difference: 7.66, 95% CI: 3.13 to 12.20; SUCRA: 94.5) and for lowering TC (mean difference: −34.33, 95% CI: −68.15 to −0.16; SUCRA: 86.4) (Figs. 5A–D and Supplementary Tables 5.8–5.11). Comparisons of lipid profiles across different targets and specific medications are provided in Supplementary Tables 6.8–6.11 and Supplementary Figures 6D–G.
Glucose parameters
HbA1c and FBG were selected as representative parameters for evaluating improvements in glucose profiles. Although GLP-1 RA + ACC inhibitor had the highest SUCRA score (77.5) for lowering HbA1c (Supplementary Table 6.12), its effect did not reach statistical significance (mean difference: −0.73, 95% CI: −1.60 to 0.13) (Fig. 5E). For FBG, GLP-1 RA + ACC inhibitor showed a relatively larger estimated effect (mean difference: −24.92, 95% CI: −41.98 to −7.86; SUCRA: 84.1) (Fig. 5F and Supplementary Table 6.13). Comparisons of targets and medications for glucose profiles are provided in Supplementary Tables 6.12–6.13 and Supplementary Figures 6H–I.
Three representative NITs
We analyzed ELF score, PRO-C3, and LSM-VCTE. GLP-1/GIP dual RA showed favorable point estimates for decreasing both ELF score (mean difference: −0.66, 95% CI: −1.06 to −0.26; SUCRA: 87.50) and PRO-C3 (mean difference: −50.90, 95% CI: −69.57 to −32.23; SUCRA: 100) (Figs. 6A–B and Supplementary Tables 5.14–5.15), with tirzepatide also ranking higher in analyses of individual medications (Supplementary Fig. 6J). For LSM-VCTE, the FGF21 analogue showed a more favorable point estimate (Fig. 6C and Supplementary Table 5.16). Detailed comparisons among targets are provided in Supplementary Tables 6.14–6.16.
Subgroup analyses of intervention duration in liver histology outcomes
In the included trials, intervention duration ranged from 16 to 72 weeks; therefore, subgroup analyses were conducted according to intervention duration (≤36 or >36 weeks) across targets. For trials with intervention durations ≤36 weeks, five targets from nine trials were analyzed. d-(R)-pioglitazone and the FGF21 analogue ranked higher for fibrosis improvement (Supplementary Table 9.1), and FGF21 and FGF19 analogues showed favorable signals for steatohepatitis resolution (Supplementary Table 9.3).
For trials with durations >36 weeks, six and seven targets were available for analyzing fibrosis improvement and steatohepatitis resolution, respectively. GLP-1/GCG dual RA appeared to have a favorable profile for fibrosis improvement (Supplementary Table 9.2), while GLP-1/GIP dual RA showed a favorable signal for steatohepatitis resolution (Supplementary Table 9.4).
Sensitivity analyses and meta-regression analyses of intervention duration in liver histology outcomes
We conducted leave-one-out sensitivity analyses to assess the robustness of target hierarchy and effect estimates for liver histological outcomes. For fibrosis improvement by ≥1 stage without worsening of steatohepatitis, d-(R)-pioglitazone, GLP-1/GCG dual RA, and the FGF21 analogue generally remained among the higher-ranked interventions based on SUCRA values, including when data from the 16-week efruxifermin trial were omitted (Supplementary Table 10.1).18 Because data for d-(R)-pioglitazone were based on only a single 36-week trial,25 omitting this trial removed d-(R)-pioglitazone from the network, and it could not be ranked. When the 24-week lanifibranor trial was omitted,26 the SUCRA value for PPAR agonists increased and ranked highest, while the OR changed from 2.94 (95% CI: 1.66 to 5.24) to 7.00 (95% CI: 1.98 to 24.75) compared with placebo. Estimates for the other targets showed no substantial changes (Supplementary Table 10.2).
For steatohepatitis resolution without worsening of fibrosis, GLP-1/GIP dual RA, GLP-1/GCG dual RA, GLP-1 RA, and the FGF21 analogue consistently ranked among the higher-ranked interventions based on SUCRA values, with only minor changes in ORs, consistent with the primary results (Supplementary Tables 10.3 and 10.4).
In addition, meta-regression using intervention duration as a continuous variable was performed to assess whether intervention duration modified relative treatment effects on histological outcomes. The results did not detect a statistically significant modifying effect of intervention duration for either histological outcome, with coefficients of 0.0007 (95% CI: −0.0429 to 0.0443, P = 0.974) and 0.0021 (95% CI: −0.0283 to 0.0326, P = 0.891) for fibrosis improvement without worsening of steatohepatitis and steatohepatitis resolution without fibrosis worsening, respectively (Supplementary Table 11). However, these findings should not be interpreted as definitive evidence that intervention duration does not modify treatment effects, given the limited number of trials and potential lack of statistical power.
Safety outcomes assessment
We analyzed selected safety outcomes among different pharmacologic targets. For any AEs during treatment, GLP-1 RA and GLP-1/GCG dual RA were associated with a significantly higher risk compared with placebo (GLP-1 RA: OR: 1.61, 95% CI: 1.21 to 2.14; GLP-1/GCG dual RA: OR: 3.07, 95% CI: 1.53 to 6.15; both P < 0.001) (Fig. 7A). For SAEs, none of the targets showed a statistically significant increase in risk compared with placebo (Fig. 7B). Numerically, FXR agonists and GLP-1/GCG dual RA were associated with higher risks of treatment discontinuation due to AEs compared with placebo (FXR agonists: OR: 1.63, 95% CI: 1.28 to 2.07; GLP-1/GCG dual RA: OR: 5.67, 95% CI: 1.98 to 16.23; both P < 0.001) (Fig. 7C). Specifically, we evaluated pruritus, a target-specific and commonly reported AE for FXR agonists. The OR for pruritus was 3.61 (95% CI: 1.15 to 11.34, P = 0.03), significantly higher than that of placebo. Given the clinical relevance of GI tolerability, we assessed three common GI-related AEs, including nausea, diarrhea, and vomiting. The results showed that GLP-1 RA, GLP-1/GCG dual RA, and FGF21 analogues all had significantly higher risks of these GI-related AEs compared with placebo (Figs. 7D–F). Furthermore, considering the clinical importance of cardiovascular safety, we also analyzed cardiovascular events in trials with clearly reported data. No statistically significant increase in cardiovascular event risk was observed for any of these targets compared with placebo (Fig. 7G).
Discussion
This network meta-analysis evaluated 11 therapeutic targets across 33 medications to assess their efficacy in improving hepatic histology, liver function, body weight, metabolic profiles, NITs, and certain safety outcomes in MASLD/MASH. Based on SUCRA rankings, d-(R)-pioglitazone and FGF21 analogues ranked higher for fibrosis improvement, whereas GLP-1/GIP dual agonists ranked higher for steatohepatitis resolution. Across secondary outcomes, incretin-based co-agonists and the FGF21 analogue showed favorable effects on LFC and metabolic profiles. Subgroup analyses based on treatment duration and leave-one-out sensitivity analyses for liver histological outcomes were largely consistent with the main findings. For safety outcomes, GLP-1 RA and GLP-1/GCG dual RA were associated with higher risks of any AEs during treatment, whereas evidence from available trials did not suggest a significant increase in SAEs or cardiovascular events for any pharmacologic target. FXR agonists and GLP-1/GCG dual agonists were associated with higher risks of medication discontinuation due to AEs, and incretin-based agents showed higher risks of nausea, diarrhea, and vomiting. Overall, our findings highlight the importance of balancing potential efficacy with tolerability in clinical decision-making.
In this study, SUCRA was applied to assess the hierarchy of different treatments. While a key advantage of SUCRA is that it makes complex ranking information easier to interpret, SUCRA does not capture the magnitude of treatment effects, does not incorporate the certainty of evidence, and rankings may be highly sensitive to the precision of estimates and the specific geometry of the treatment network.27 In our analyses, THR-β agonists did not rank among the top interventions for the prespecified outcomes, but this should not be interpreted as a lack of clinical value. Resmetirom (Rezdiffra), a THR-β agonist, was the first therapy to receive accelerated approval from the U.S. Food and Drug Administration (FDA) for adults with MASH with moderate-to-advanced fibrosis (F2–F3),3,28 supported by large phase 2–3 trials. Thus, resmetirom represents an approved, evidence-based option for non-cirrhotic MASH, whereas several higher-ranked targets in our analyses remain earlier in clinical development and require confirmation in larger and longer-term studies.
More recently, semaglutide (Wegovy), a GLP-1 RA, received accelerated approval from the FDA for the treatment of MASH in adults with moderate-to-advanced fibrosis (F2–F3).29 Incretin-based medications have demonstrated pleiotropic benefits for MASLD/MASH, with several trials showing the glycemic and weight-reducing efficacy of GLP-1 RA in T2DM and overweight/obesity,8,30–32 while also reducing LFC and resolving steatohepatitis, as shown for liraglutide in the LEAN trial in 2016.19,33,34 A series of clinical trials investigating long-acting GLP-1 RA, particularly semaglutide, further highlighted its efficacy in improving liver histology.35 GLP-1 RA can stimulate insulin secretion to regulate glucose levels and interact with GLP-1 receptors located in circumventricular organs to enhance satiety and reduce food intake.36 Although no evidence has confirmed GLP-1 receptor expression on hepatocytes, these agents are believed to exert their effects through indirect mechanisms. In this study, GLP-1 RA showed favorable signals in lowering LDL-C levels and promoting steatohepatitis resolution. Of note, because the combination of semaglutide and resmetirom has not yet been studied for safety or efficacy, no specific recommendation can be made for their concomitant use in MASH treatment according to the updated AASLD guidance for the management of MASH.29
Advancements in GCG biology have led to the development of GLP-1 co-agonists targeting GIP or GCG receptors, as well as GLP-1/GIP/GCG triple RAs. These novel agents have shown promising results in the treatment of T2DM, obesity, and MASLD.20,37–39 We found that GLP-1/GIP dual RA was associated with favorable effects on steatohepatitis resolution, along with reductions in ELF score and PRO-C3. GIP can stimulate insulin secretion in the pancreas, improve insulin sensitivity in white adipose tissue, suppress appetite by acting on the central nervous system, and reduce ectopic lipid accumulation in the liver and skeletal muscle.40 However, the evidence base for these multi-agonist strategies in histology-proven MASLD trials remains relatively limited, and the ranking hierarchy may be sensitive to relatively small sample sizes and the small number of contributing trials. Therefore, these findings should be interpreted with caution.
GLP-1/GCG dual RA showed promise in reducing the absolute change from baseline in LFC and body weight, while also showing favorable signals for improvements in liver histology, glucose, and lipid profiles. Evidence indicates that GCG enhances hepatic lipolysis, facilitates clearance of circulating cholesterol, and contributes to appetite suppression.41 However, it is important to note that trials of GLP-1 dual or triple co-agonists remain limited, and results from ongoing studies are still emerging. Confirmation in larger-sample and longer-duration trials will be essential before firm conclusions can be drawn for routine practice.
FGF21 analogue demonstrated potential for improving liver fibrosis or steatohepatitis resolution, reducing LFC, and improving lipid and glucose metabolism, particularly HDL-C and TG. FGF21 exerts hepatoprotective effects by promoting fatty acid oxidation, inhibiting de novo lipogenesis, and through other mechanisms.42 d-(R)-pioglitazone is a unique agent, as despite lacking PPARγ activity, it demonstrated a favorable signal in reducing liver fibrosis, likely via its effects on the mitochondrial pyruvate carrier and acyl-CoA synthetase long chain family member 4, as well as the non-genomic effects of pioglitazone.15,43 In the DESTINY-1 trial, 50% of individuals in the 15 mg PXL065 group achieved fibrosis improvement without steatohepatitis worsening.25 However, this analysis was conducted in a small subset of patients who completed serial liver biopsies, which may have introduced selection bias, limited generalizability, and increased imprecision. Importantly, estimates based on small biopsy subsets tend to be less precise, and this imprecision can translate into more variable ranking probabilities. Consequently, SUCRA values may appear inflated even when the underlying evidence is uncertain. Indeed, although d-(R)-pioglitazone had a high SUCRA ranking for fibrosis improvement, the certainty of evidence assessed by CINeMA was very low. Therefore, interpretation of d-(R)-pioglitazone requires caution, and SUCRA should be considered alongside effect estimates and their uncertainty rather than as a standalone measure. Meanwhile, leave-one-out sensitivity analyses showed that omitting the d-(R)-pioglitazone trial removed this node from the network, while the overall hierarchy of the remaining targets remained largely unchanged, supporting the robustness of our primary findings.
Treatment duration is an important consideration when interpreting histological outcomes in MASLD/MASH trials. Fibrosis regression is a slow biological process and may require longer treatment exposure than changes in biochemical markers. Therefore, shorter-duration trials may have limited ability to capture meaningful fibrosis or steatohepatitis improvement. In our current analysis, trials contributing to histological outcomes ranged from 16 to 72 weeks. Although subgroup analyses, leave-one-out sensitivity analyses, and network meta-regression using treatment duration as a continuous covariate did not detect statistically significant evidence that the main findings were driven by intervention duration, these negative findings should not be interpreted as evidence of the absence of duration-related effect modification. The limited number of included trials, sparse network structure, and potential correlation between intervention duration and trial design may have reduced statistical power to detect such effects. Currently, the optimal treatment duration for achieving and sustaining liver histological benefits remains uncertain.44 Longer-duration trials and longitudinal follow-up studies are needed to better clarify whether and how treatment duration modifies histological responses and to inform treatment strategies.
For safety assessment, we focused on a prespecified set of clinically relevant safety outcomes, including any AEs during treatment, SAEs, discontinuation due to AEs, three common GI-related AEs across different targets, pruritus for FXR agonists, and cardiovascular safety. Overall, most pharmacologic targets did not show a statistically significant increase in AEs compared with placebo based on available data. However, some incretin-based therapies were associated with increased risks of any AEs during treatment and GI events, highlighting the importance of tolerability in clinical decision-making. Although most reported AEs in the included trials were not considered treatment-related, longer-term safety and tolerability beyond the intervention period still require further investigation.
The findings of our study may inform clinical decision-making by providing a comparative overview of the potential benefits and safety outcomes of these emerging pharmacologic targets. However, treatment selection in clinical practice should not be based solely on ranking metrics or isolated efficacy outcomes. Instead, clinicians should consider clinical phenotype, including fibrosis stage, presence of T2DM, obesity, and other comorbidities, cardiometabolic risk profiles, long-term tolerability, contraindications, and individual preferences. In this context, given the limited direct head-to-head evidence and the low or very low certainty of evidence for histological comparisons, our findings should be viewed as supportive evidence for individualized treatment considerations rather than direct guidance for treatment selection.
The strengths of this study lie in the comprehensive and updated nature of the therapeutic interventions included and the endpoints analyzed. The study evaluated several promising therapeutic targets currently under investigation in high-quality RCTs with comparative data not only for the major therapeutic target classes but also for individual medications within each class. The current study evaluated fibrosis reduction and steatohepatitis resolution as primary endpoints, along with several other metabolic parameters and safety profiles as secondary endpoints. The primary endpoints analyzed in this study are highly relevant outcomes, as liver fibrosis is a well-recognized independent factor associated with both liver-related and non–liver-related mortality among people with MASLD, with steatohepatitis considered its “prodrome”.45,46 It is also highly relevant to evaluate the therapeutic impacts of MASH interventions on metabolic parameters, as these are underlying factors in the pathogenesis of MASLD.47,48
We also note several limitations. First, the primary efficacy outcomes relied heavily on liver biopsy results, which may have led to the inclusion of participants with more severe disease states. Second, although our predefined eligibility criteria and the use of placebo-controlled RCTs with broadly accepted histological endpoints provide partial support for the clinical plausibility of the transitivity assumption, transitivity should not be considered fully established across all treatment nodes. Residual clinical heterogeneity remained. The included trials varied in baseline fibrosis stage distributions, metabolic comorbidity burden, BMI, and other key clinical factors, all of which may influence histological treatment responses. We therefore added a summary of key baseline characteristics to facilitate assessment of comparability across trials. These characteristics showed variability, although we did not observe an obvious separation of target nodes by markedly distinct patient populations. However, the possibility of residual intransitivity cannot be fully excluded. Therefore, indirect active-versus-active comparisons should be interpreted cautiously, particularly when estimates are mainly derived from a common placebo comparator rather than direct head-to-head evidence. Accordingly, these findings should be interpreted in light of the observed clinical heterogeneity and the structure of the evidence network. Future studies with arm-level subgroup-specific events and denominators, or preferably individual participant data, are warranted to better evaluate whether baseline characteristics modify treatment responses.
Third, although SUCRA was used to summarize treatment rankings, these rankings should be interpreted cautiously, as some data were extracted from single trials with relatively small sample sizes. SUCRA does not account for the magnitude or certainty of treatment effects and may be unstable in sparse or placebo-centered networks. In our analysis, several higher-ranked targets were supported by limited trial evidence, wide 95% CIs, or low to very low CINeMA certainty ratings. Therefore, SUCRA-based rankings and active-versus-active estimates should be interpreted cautiously, particularly when supported primarily by indirect evidence, and should be considered together with evidence contribution and certainty of evidence.
Fourth, not all medications and interventions were included for comparison. Due to the large number of treatments in the development pipeline, we focused on the most promising targets. Furthermore, our analyses focused on absolute changes from baseline for body weight, metabolic profiles, and NITs. The lack of separate analysis of percent change outcomes due to inconsistent reporting across trials is also a limitation. In addition, this study did not evaluate the efficacy of different treatments in MASH-related cirrhosis (F4), which remains a major unmet medical need. Considering that there are substantial differences in trial design, patient risk profiles, stratification factors, and endpoint settings between trials conducted in individuals with cirrhosis and those in the non-cirrhotic MASLD/MASH population (F1–F3), we excluded cirrhotic MASLD/MASH clinical trials to minimize clinical heterogeneity and improve comparability across studies. Therefore, our findings should not be extrapolated to patients with cirrhosis. Future studies should evaluate whether the treatment effects observed in the F1–F3 population translate into clinically meaningful benefits in patients with cirrhosis.
Finally, although our safety analyses focused on a prespecified set of safety outcomes, the assessment of safety remained limited by the availability and consistency of reported data. In particular, longer-term tolerability could not be formally evaluated because follow-up durations varied across trials and safety outcomes beyond the trial period were rarely reported. Therefore, our safety findings should not be interpreted as definitive evidence of longer-term safety or tolerability.