To identify all relevant studies on the effectiveness and safety of semaglutide and dulaglutide for glycemic control and weight loss in adult patients with T2DM, a thorough literature search was conducted using the Embase (https://www.embase.com), Google Scholar (https://scholar.google.com/), MEDLINE (https://www.nlm.nih.gov/medline) and SCOPUS (https://www.scopus.com/) databases from their inception until July 2024. Specific search terms were used, including semaglutide, dulaglutide and T2DM. The search included studies in all languages but was limited to human subjects.

Studies that met the following criteria were only considered: i) Comparing the effectiveness and safety of various doses of semaglutide with dulaglutide in achieving glycemic control and weight loss in patients with T2DM; ii) were conducted in patients who were at least 18 years old; iii) diagnosed T2DM according to specific criteria from either the World Health Organization (8) or the American Diabetes Association (9); and iv) were published in a peer-reviewed journal. Meta-analyses, reviews, case reports, editorials, letters, commentaries, expert opinions and experimental studies were excluded.

Two authors reviewed the titles and abstracts of the included publications to determine whether they met specific requirements for inclusion or exclusion. If a study met the criteria, the full text was examined. In cases of disagreement between the two researchers, a third researcher was consulted to make a conclusive decision. For all included studies, the following information was extracted: Authorship, year of publication, study design, study duration, randomization process, intervention method, sample size, study location, population studied, therapy duration and follow-up period, patients' body weight or body mass index (BMI) and adverse events.

The primary outcome of effectiveness was the difference in HbA1c levels from the initial measurement between the groups. The secondary outcome was the effect of the medications on body weight loss (change in body weight from baseline) during the treatment period.

The quality of citations included in the study was evaluated using the Cochrane Collaboration's Risk of Bias Tool (10) and Review Manager. The evaluation considered several factors: Allocation concealment, blinding of participants and personnel, random sequence generation, blinding of outcome assessors, selective outcome reporting, incomplete outcome data and other biases. The studies were categorized into three levels of risk: high, unclear and low. Two authors conducted the quality appraisal and consulted a third reviewer in case of any discrepancies.

Comprehensive Meta-Analysis (CMA) 2 was utilized for all the analyses. Data were expressed as the mean ± standard deviation (SD). The effect size statistic for continuous measurement data was the standardized mean difference (SMD). In cases where SD was not provided, the following formula was used to convert standard error to SD, where N represents the sample size (11):

In studies where confidence intervals (CIs) were provided, the following formula was used to convert CIs to SDs. As before, N represents the sample size (11):

In the present meta-analysis, a random-effects model was employed to consider the differences in effect size estimates. This enables us to make inferences at the population level and is more rigorous than a fixed-effects model (12). The authors did not rely on the I² statistic to indicate the level of heterogeneity in effect size because it is not designed for that purpose. Unless the I² value is zero, it cannot provide information on the extent of heterogeneity. While the I² statistic measures the proportion of variability due to heterogeneity, it does not provide insight into the absolute variation or its practical significance. By contrast, prediction intervals indicate the range within which future study results are likely to fall, making them directly interpretable in real-world contexts. This allows for an improved understanding of how markedly study results might vary and whether that variation is meaningful for clinical or practical decisions. Additionally, prediction intervals are not as sensitive to the number and size of included studies, addressing some of the limitations of the I² statistic. Therefore, prediction intervals were used to report heterogeneity in the data. Prediction intervals were calculated using a random-effects model, which considers both within-study and between-study variability. The interval was derived from the overall effect size estimate, accounting for the standard error and variability between studies (13).

A funnel plot was used to determine if there was any bias in the published material. A balanced funnel plot indicated no bias, while an unbalanced one suggested that there was bias. Egger's and Begg's tests were also used to confirm the presence of bias in the publications because the interpretation of the funnel plot can be subjective.

Moderator analysis, performed through meta-regression, explores how study-level characteristics influence the effect sizes across different studies. In this approach, potential moderators, such as treatment duration, were identified and data were collected for each included study. The meta-regression model was then set up using CMA, with the effect size as the dependent variable and the moderator as the independent variable. The analysis estimated a regression line to describe the relationship, and a significant P-value (typically <0.05) indicated that the moderator significantly impacts the effect size (14). If the P-value of either test was less than 0.05, it indicated bias. P<0.05 was deemed to indicate a statistically significant difference.

The search for relevant literature yielded 127 publications. After eliminating duplicates and examining the titles and abstracts of the sources, 20 were selected for a thorough review. Out of these 20, only six met the criteria for inclusion (6,15-19). A total of 3 of these studies had several subgroups and comparisons, resulting in 10 comparisons in the present meta-analysis. A total of 2 studies were non-randomized (6,18), while 4 were randomized clinical trials. A flowchart in Fig. 1 shows the steps involved in including and excluding the studies. The duration of the studies ranged between 12-52 weeks. There were 1,038 patients in the semaglutide group and 847 patients in the dulaglutide group. In the included trials, semaglutide and dulaglutide doses varied between 0.5-14 mg and between 0.75-1.5 mg, respectively. The study characteristics are included in Table I.

The analysis included 10 different comparisons. The initial global examination of all glycemic control measures using a CMA required the inclusion of each study only once in a nested analysis. The results demonstrated no difference between semaglutide and dulaglutide in achieving glycemic control in patients with T2DM, with an SMD of 0.613 (95% CI, -0.164-1.391; P=0.122) (Fig. 2A).

Similar to the previous analysis, this examination included 10 different comparisons. It was found that semaglutide was more effective than dulaglutide in inducing weight loss in patients with T2DM, with an SMD of 2.45 (95% CI, 0.467-4.434; P=0.015) (Fig. 2B).

The analysis was conducted 10 times, with each iteration excluding one study to demonstrate the impact of that particular study on the outcome. It was found that the exclusion of no single study affects the outcome of glycemic control efficacy data (Fig. 3A). However, the exclusion of one of the comparisons conducted by Pratley et al (15) caused a change in the initial assumption that semaglutide had higher efficacy than dulaglutide in inducing weight loss in patients with T2DM. This means that these results should be interpreted with caution (Fig. 3B).

The prediction interval analysis for glycemic control data found that the mean effect size was 0.61, with a 95% CI of -0.164-1.39. In 95% of similar populations, the true effect size falls between -2.37-3.60 (Fig. 4A). This indicates a large amount of heterogeneity among the studies analyzed, as the prediction interval is wider than the CI and suggests a broader range of potential treatment effects. The presence of a statistically significant effect is not confirmed here, as all values in the CI and prediction interval do not fall on one side of the null. The mean effect size for the weight loss data was found to be 2.45, with a 95% CI of 0.47-4.43. The true effect size in 95% of all comparable populations fell from -0.07-4.97. Again, the data heterogeneity was considered high because the prediction interval was wider than the CI. All CI values on one side of the null confirm the presence of a statistically significant effect. At the same time, the prediction interval suggests the possibility of values on both sides of the null (Fig. 4B).

For the first set of comparisons, the funnel plot was reasonably symmetric. However, one study was found to be missing to the right of the mean, while no study was missing to the left of the mean (Fig. 5A). The Begg's and Mazumdar rank correlation showed no publication bias (2-tailed P=0.05 in Kendall's tau with continuity correction). Similarly, Egger's regression revealed no sign of publication bias (two-tailed P=0.22).

For the second set of comparisons, the shape of the funnel plot was also symmetric. Nevertheless, two studies were missing to the right of the mean in the trim-and-fill analysis. No study was found missing to the left of the mean (Fig. 5B). The Begg's and Mazumdar rank correlation showed no publication bias (2-tailed P=0.05 in Kendall's tau with continuity correction). Similarly, Egger's regression revealed no sign of publication bias (two-tailed P=0.22).

No correlation was found between treatment duration or semaglutide/dulaglutide doses and glycemic control (slope P>0.05 for all comparisons; data not shown). However, a correlation was found between treatment duration and weight loss, indicating that a longer treatment duration was associated with higher weight loss (slope P=0.02; Fig. 6).

Based on the Cochrane risk of bias assessment tool, the general quality of the included studies was low. Only one of the studies was blinded (18), and randomization was performed in four of the studies. None of the studies implemented allocation concealment (Fig. 7).

No serious adverse events were reported in the study of Takahashi et al (19). During the study period, three patients experienced hypoglycemia (one with dulaglutide and two with semaglutide). However, none of these incidents required the patients to stop their treatment (19). In line with that Iacobellis and Villasante Fricke (18) demonstrated no adverse events or need for hospitalization. Iijima et al (17) reported treatment discontinuation in one patient due to intractable vomiting and weight loss. In the course of the study by Pratley et al (15), six participants lost their lives: One and two individuals succumbed while using semaglutide and dulaglutide, respectively. The incidence of serious adverse effects was similar across all treatment groups. Gastrointestinal problems were the most common adverse effects (6,16), with similar rates in all groups except for those receiving dulaglutide 0.75 mg, where fewer patients experienced gastrointestinal issues (15). Seijas-Amigo et al (6) reported one cardiovascular-related death during dulaglutide treatment.

Previous studies have shown mixed results regarding the efficacy of these medications in achieving glycemic control in patients with T2DM, with some indicating that there is no significant difference between the two medications (6), while others show that semaglutide has improved efficacy in reducing HbA1c levels compared with dulaglutide (15,16). When semaglutide was compared with other GLP-1 RAs in a meta-analysis, it was identified that it lowers HbA1c and fasting plasma glucose levels more effectively and increases the likelihood of achieving both targeted and intense HbA1c level reductions by >2 times. Although the likelihood of significant HbA1c reductions with semaglutide was comparable to that of liraglutide, the absolute reduction in HbA1c and fasting plasma glucose levels obtained with semaglutide was not larger than that attained with dulaglutide (20). In another meta-analysis, despite substantial variation in the length of the trials and the therapy regimens employed, semaglutide was shown to be more efficient than dulaglutide in terms of glycemic control, with a significant decrease in HbA1c by 0.47% (21). Another meta-analysis revealed that semaglutide 1 mg significantly outperformed other GLP-1 RAs, such as dulaglutide, in decreasing HbA1c, with a similar drop of 0.38% (22). In a similar vein, a network meta-analysis showed that semaglutide 1 mg was more substantially associated with a decrease in HbA1c among individuals with T2DM than dulaglutide (23).

Several factors could cause the variations in the results provided. These could include differences in study design, sample sizes and patient populations. Additionally, variations could be caused by differences in dosage, administration route (oral vs. subcutaneous) and formulation. Other factors, such as patient adherence, treatment duration and lifestyle modifications, could also contribute to the differences observed in the results. It is thus important to carefully consider these factors when interpreting and comparing the results of different studies.

Moderate weight reduction can improve blood glucose control, decrease the need for hypoglycemic medications, and is linked to a substantial decrease in the risk of various obesity-connected complications in overweight or obese patients with T2DM. Therefore, the impact on weight should be considered when glucose-lowering medications are prescribed for overweight or obese patients with T2DM. GLP-1 RAs exhibit both the effectiveness of weight reduction and a favorable hypoglycemic impact (24). The results concerning the efficacy of semaglutide vs. dulaglutide in inducing weight loss are more consistent. According to the findings of Patoulias et al (20), semaglutide was found to be more effective than dulaglutide when looking at absolute weight loss, probabilities for attaining weight loss between 5-10%, and BMI decrease. Similarly, in two other aforementioned meta-analyses, semaglutide was substantially more effective than dulaglutide in reducing body weight. In direct connection with these findings, Andreadis et al (25) found that compared with dulaglutide, semaglutide 1 mg was more effective at lowering body weight.

Due to a lack of data, a safety analysis could not be performed in the present study. It was found that the most common side effect of treatment in these therapies was gastrointestinal complications such as nausea and vomiting. However, a direct comparison between medications was not possible. Mishriky et al (22) showed that, compared with other GLP-1 RAs, semaglutide was considerably more likely to cause nausea and vomiting. On the other hand, Shi et al's (21) meta-analysis of semaglutide and other GLP-1 RAs (for example, dulaglutide) failed to detect a higher incidence of gastrointestinal side effects in the former as opposed to the latter (21). In addition, a meta-analysis by Mishriky et al (22) found no statistically significant difference between semaglutide and other GLP-1 RAs in the incidence of any serious adverse events.

Heterogeneity is a common issue in meta-analyses, arising from differences in study design, patient characteristics, interventions and outcome measures among included studies. In the present study, high heterogeneity was found among the studies included in the meta-analysis, as evidenced by prediction interval analysis, which suggested significant between-study variability. This heterogeneity is important to acknowledge, as it can affect the precision, validity and generalizability of the meta-analysis results.

The I² statistic is commonly used in meta-analyses to quantify the percentage of total variation across studies due to heterogeneity rather than chance. It measures the inconsistency or diversity among studies and can range from 0-100%, with higher values indicating greater heterogeneity (26). However, it has been recently argued that the I² statistic does not adequately reveal the degree of variation in effect size in a meta-analysis (27). In that light, prediction interval analysis was used instead. Prediction interval analysis is a statistical method used to calculate CIs for the true effect size based on the distribution of the data and the sampling error. Given the observed data and sampling variability, it provides a range of values within which the true effect size is likely to fall. Prediction interval analysis is particularly useful when estimating small or large effect sizes, as CIs are narrower for larger samples and effect sizes and wider for smaller samples and effect sizes.

To assess the potential impact of individual studies on the meta-analysis results, a sensitivity analysis was conducted by sequentially removing one study at a time and recalculating the effect size and pooled estimate. The sensitivity analysis results showed that removing one study (15) had a substantial effect on the overall effect size of the weight loss data and the statistical significance of the meta-analysis. This underscores the importance of cautious interpretation of meta-analysis results and the potential impact of individual studies on the outcome.

The present study has several limitations that must be acknowledged. Despite the comprehensive search strategy used, it is still possible that some eligible studies may have been missed, which may influence the precision of the meta-analysis results. Besides, the number of studies available for this meta-analysis was limited which might be attributed to several factors. Firstly, semaglutide and dulaglutide are relatively new therapeutic options for the management of T2DM, and head-to-head comparisons between these two drugs have only recently become the focus of clinical research. This has resulted in a smaller pool of studies that directly compare their efficacy and safety. Additionally, our strict inclusion criteria, which required randomized controlled trials with specific outcome measures such as glycemic control and weight loss, further narrowed the range of eligible studies. As a result, some potentially relevant studies may have been excluded due to differences in study design, populations, or intervention protocols that did not align with the objectives of the current analysis.

Additionally, the quality of the included studies was variable, with some studies being of lower quality. This variability could have contributed to the observed heterogeneity and the sensitivity analysis results. The heterogeneity of the included data was also high. Although random-effects model was used to account for heterogeneity, it is important to acknowledge that some residual heterogeneity may remain, affecting the precision of the estimates. The high degree of heterogeneity observed in the present meta-analysis is likely influenced by differences in study design, patient populations, treatment regimens and dosage levels among the included studies. Specifically, variations in baseline HbA1c levels, treatment duration and dosing strategies (for example, semaglutide doses ranging from 0.5-14 mg vs. dulaglutide doses between 0.75-1.5 mg) may have contributed to the observed variability in outcomes. While subgroup analyses could provide further insights into the impact of these factors, the limited number of studies and variability in study characteristics made such analyses impractical. Instead, using moderator analysis allowed us to identify a significant relationship between longer treatment duration and greater weight loss, suggesting that duration may be an important factor to consider when choosing between these treatments. The broad prediction intervals indicate that, while semaglutide generally shows an advantage in weight loss, results may vary across different clinical settings and patient populations, highlighting the need for individualized treatment decisions. Future studies should aim for more consistent study designs and larger sample sizes to better delineate the comparative effectiveness of these therapies. The current sensitivity analysis identified that the exclusion of one study could significantly impact the overall results of weight loss data. This suggests that the present conclusions depend on including all studies and that one or more studies may have an undue influence on the overall estimate. Finally, meta-analyses are inherently limited by the quality and availability of the primary studies and the limitations of the meta-analytic methods used. While it was aimed to minimize potential biases and increase transparency in our review process, the present results may be subject to publication bias, incomplete reporting of results and other methodological limitations.

The present meta-analysis compared the efficacy of two important GLP-1 RAs, semaglutide and dulaglutide, in achieving glycemic control and weight loss in patients with T2DM. The results revealed that while there was no significant difference in glycemic control between the two drugs, semaglutide was more effective in inducing weight loss than dulaglutide. It is important to note that although random-effects model was used to account for heterogeneity, some residual heterogeneity may have remained, which could affect the precision of the estimates. Despite this, the present findings suggested that semaglutide may be a more favorable treatment option for patients with T2DM requiring weight management and glycemic control. Further research is needed to investigate the long-term anti-diabetic and cardiovascular benefits of GLP-1 RAs, the optimal dosages and regimens, and individual factors that may influence treatment response and outcomes. Nevertheless, the present meta-analysis provides valuable insights for clinicians and patients when making treatment decisions for T2DM.