Comparing the effectiveness of long-term use of daily and weekly glucagon-like peptide-1 receptor agonists treatments in patients with nonalcoholic fatty liver disease and type 2 diabetes mellitus: a network meta-analysis

A search for all treatments in NAFLD was conducted across the PubMed databases from the date of inception until December 2022 using the following search strategy: (Liraglutide OR Dulaglutide OR Semaglutide OR Albiglutide OR Lixisenatide OR Exenatide OR glucagon-like peptide-1 agonists OR glucagon like peptide OR GLP-1 receptor agonists OR glp-1) AND (Non-Alcoholic Fatty Liver Disease OR Nonalcoholic Fatty Liver Disease OR Nonalcoholic Fatty Liver OR NAFLD OR Nonalcoholic Steatohepatitis OR NASH) AND (liver enzymes OR alanine aminotransferase OR aspartate aminotransferase OR γ-glutamyl transferase OR ALT OR AST OR γGGT OR intrahepatic fat content OR liver fat content OR intrahepatic content of lipid OR hepatic lipid content OR hepatic fat content OR LFC OR IHF OR IHCL OR HFC) in all fields without other limitations.

And search strategies for PubMed, Cochrane and Embase databases were shown in Table 1.

The paper inclusion criteria were as follows: (1) Subjects: clinically diagnosed as NAFLD or NASH with T2DM; (2) Drug interventions: patients in the experimental group were treated with GLP-1RAs; (3) Study type: randomized controlled trials (RCTs);

The paper exclusion criteria were as follows: (1)Animal models; (2)Duplicate articles; (3) Subjects were aged <18 years; (4) Study duration <12weeks. (5) The outcomes: liver fat content (LFC), aspartate aminotransferase (AST), alanine aminotransferase (ALT), γ-glutamyl transferase (γGGT) and body weight were not clearly reported. (6)The interventions were not GLP-1RAs versus placebo or blank control; (7)Data outcomes could not be extracted.

Study selection and data extraction were conducted separately by two individuals. Two reviewers initially selected the relevant studies by reading the title and abstract and then selected the studies for NMA based on the inclusion and exclusion criteria and after reading the full text. Next, any disagreements were resolved by discussion or by a third researcher.

The extracted data included: 1) the baseline information: the last name of the first author, publication year, intervention and control, sample size (female/male), dose (frequency of application), duration, baseline age (mean ± standard deviation [SD]), T2DM, with or without NASH, and the countries of study population, the characteristics of included studies were listed in Table 2; 2) the data used for analysis: mean and SD changes from the baseline to the end of each outcome, and sample size (n); 3) the information for quality assessment; 4) the items of evidence certainty assessment.

The Cochrane Risk of Bias tool (26) was used to assess the risk of bias of the included studies. The following seven items were included: 1) “random sequence generation”: describes how the sequence was generated, such as by using a random table of numbers or a computer for generating a random sequence of numbers; 2) “allocation concealment”: whether the subjects and researchers were aware of group assignments, such as through assignment hiding via telephone and Internet; 3) “blinding of the participants and personnel”: whether subjects, researchers, and all participants were blinded; 4) “blinding of outcome assessment”: describe whether an outcome assessor was blinded, but objective outcomes, such as serological outcomes, were unlikely to be affected by the lack of blinding; 5) “incomplete outcome data”: whether there was any missing data, such as loss to follow-up and exclusion of data from analysis; 6) “selective reporting”: whether all outcomes were reported; 7) “other bias”: each study was considered to have a “high”, “low”, or “unclear” risk of bias. The judgment of risk of bias was conducted by two authors separately in Review Manager (Version 5.4).

And then, we used GRADE (Grades of Recommendation, Assessment, Development and Evaluation) model to assess the evidence certainty (27). Since all the included studies were RCTs, we evaluated the following five items: 1) risk of bias: such as allocation concealment, blinding and loss to follow-up, and so on; 2) inconsistency: the results heterogeneity, and whether the authors give a reasonable explanation for its high heterogeneity; 3) indirectness; 4) imprecision: whether the confidence interval (CI) was wide and the sample size was large; 5) publication bias: the number of included studies. This assessment was performed in GRADEprofiler (version 3.6).

First, we constructed network plots of the outcomes to demonstrate all available evidence for each outcomes (Figure 1). Second, the outcomes we selected were all continuous variables, and therefore the mean and standard deviation (SD) changes from the baseline to the end and the sample size (n) were extracted for statistical analysis. The existing evidence only involved indirect comparison; therefore, the network graph had no closed loop and there was no need to examine the inconsistency of the outcomes. We employed SUCRA to evaluate the ranking of each intervention in each outcome (Figure 2). The higher the SUCRA value, the more likely the corresponding intervention to be regarded as the best treatment. “Zero” indicated that the treatment was the worst. The forest plots for each outcome were depicted in Figure 3, which shown the comparison between each intervention. The forest plots visually demonstrated the 95% confidence interval (CI) of the results of the pairwise comparison of interventions and whether they had any statistical significance. Finally, league plots were drawn based on SUCRA and the forest plots (Figure 3). The league plots ranked the effect of the intervention in each outcome from the best to the worst (Table 3). The results with statistical significance were highlighted in bold. The league plots more intuitively exhibited the effectiveness of each intervention. All of the abovementioned analyses were conducted by Stat17.0.

Then, we divided all studies with included outcomes into two subgroups of daily and weekly preparations, drew forest plots (Figure 4) using a random effects model to compared the mean difference (MD) between the two subgroups, and to observe which one was better in each outcome. The above analysis was performed by RevMan (version 5.4).

According to the search strategy, 1055 studies were searched from the following databases: PubMed, 224 studies; Embase, 649 studies; and Cochrane, 182 studies, and 310 duplicate references were removed. According to the inclusion and exclusion criteria, 14 RCTs were finally included in this NMA. The experimental group included five RCTs (12, 14–17) of weekly GLP-1RAs and nine RCTs (9, 18–25) of daily agents. The detailed literature selection process was shown in Figure 5.

As Table 2 shown, the female to male ratio in the study population was approximately 1.19:1. Subjects from all over the world.

The quality of the included studies was assessed by the risk assessment of Cochrane review items. The following aspects were considered during the assessment: random sequence generation, allocation hiding, the blindness of participants and personnel, the blindness of result evaluation, incomplete result data, selective reporting, and other biases. The specific evaluation results were presented in Figure 6.

Using the GRADEprofiler to assess overall quality of evidence. The evaluation results were as follows: two outcomes were assessed as “low”, three outcomes were assessed as “moderate”. The assessment results were shown in Table 4.

All experiments were included in this NMA, and the network evidence graphs of each outcome were shown in Figure 1. Among them, weekly GLP-1RA drugs in the treatment of patients with NAFLD mainly include semaglutide (qw), dulaglutide and exenatide (qw) and daily drugs include liraglutide, semaglutide (qd) and exenatide (bid). However, studies on other GLP-1RAs are scarce. The main outcomes we evaluated were LFC, ALT, and AST. Four drugs (except exenatide (bid) and semaglutide (qw)) showed the AST and five drugs (except exenatide (bid)) showed the ALT outcomes, and four drugs (except semaglutide (qd) and exenatide (qw)) showed the LFC outcome. The secondary outcomes were γGGT and body weight, whereas only three drugs (liraglutide, dulaglutide and exenatide (qw)) showed γGGT outcome, and all drugs, except exenatide (bid) and semglutide (qd), showed body weight outcome.

The SUCRA curves of interventions for outcomes were shown in Figure 2. Among five interventions (exenatide (bid), liraglutide, dulaglutide, semaglutide (qw) and placebo) evaluated for improving LFC, exenatide (bid) was the best (SUCRA = 66.8%, 59.1%, 59.1%, 60%, and 5.1%, respectively). Among five interventions (semaglutide (qd), liraglutide, dulaglutide, exenatide (qw), and placebo) evaluated for AST outcome, and six interventions (semaglutide (qd), liraglutide, dulaglutide, semaglutide (qw), exenatide (qw) and placebo) evaluated for ALT outcome, semaglutide (qd) was the most effective drug (SUCRA (AST) = 100%, SUCRA (ALT) = 95.6%). For AST, followed by liraglutide and dulaglutide (SUCRA (AST) = 69.9%, and 40.2%, respectively); For ALT, followed by semaglutide (qw) and liraglutide (SUCRA (ALT) = 78.8%, and 49.7%, respectively). Finally, the effects of four interventions (liraglutide, dulaglutide, exenatide (qw), and placebo) on γGGT were compared, and liraglutide was the most effective treatment (SUCRA (γGGT) = 70%), and the effects of five interventions (liraglutide, dulaglutide, semaglutide (qw), exenatide (qw), and placebo) on body weight were compared, semaglutide (qw) seemed better than liraglutide (SUCRA (body weight) = 99.8% vs 63.9%).

The forest plots were shown that in all outcomes except γGGT, the daily preparations seemed more effective than weekly ones. The result of LFC in daily GLP-1RAs group was MD = -3.66, 95% CI [-5.56, -1.76] and in weekly GLP-1RAs group, it was MD = -3.51, 95% CI [-4, -3.02], p=0.88. As to AST and ALT, the results in daily GLP-1RAs group versus weekly GLP-1RAs group were AST: MD = -7.45, 95% CI [-14.57, -0.32] versus MD = -0.58, 95% CI [-3.18, 2.01], p=0.08 and ALT: MD = -11.12, 95% CI [-24.18, 1.95] versus MD = -5.62, 95% CI [-15.25, 4], p=0.51. The result of Daily GLP-1RAs group also was better than weekly one in body weight (MD = -3.32, 95% CI [-4.61, -2.03] vs MD = -1.72, 95% CI [-2.31, -1.13], p=0.03). However, the result of γGGT showed contrary to other outcomes (MD _daily = -4.83, 95% CI [-15.5, 5.83] vs MD _weekly = -6.16, 95% CI [-14.13, 1.81], p=0.85).

GLP-1RAs have been a popular hypoglycemic drug in recent years. Apart from hypoglycemic and weight loss effects, GLP-1RAs are also of great research value in NAFLD. However, no studies have compared the efficacy of daily and weekly GLP-1RA treatments for NAFLD with T2DM yet. Therefore, we adopted the NMA method to comprehensively analyze the effect of several commonly used GLP-1RAs on the reduction of LFC, liver enzymes, and body weight in patients with NAFLD and T2DM and to obtain an optimal treatment. However, we included only five studies on the weekly agents, which was limited in number and may lead to weak evidence, thus RCTs including more studies on weekly agents vs. placebo are needed to validate the present results. Moreover, due to the lack of direct comparative studies of the two GLP-1RAs, we cannot analyze inconsistent. The league plots showed a comparison between liraglutide and the placebo, showing that the major outcome, LFC, was statistically significant; however, the rest of the results were not statistically significant, which might be because of the small sample size. And there is only one study of semaglutide(qw), thus more studies of weekly semaglutide are needed to compare with daily exenatide.to assess which is superior in LFC. In the future, more large-sample, head-to-head RCTs are required to confirm these findings.