Subcutaneously administered tirzepatide vs semaglutide for adults with type 2 diabetes: a systematic review and network meta-analysis of randomised controlled trials

We included RCTs published in English that assessed s.c. tirzepatide at maintenance doses of 5 mg, 10 mg or 15 mg once weekly, or s.c. semaglutide at maintenance doses of 0.5 mg, 1.0 mg or 2.0 mg once weekly for a minimum duration of 12 weeks. Eligible trials compared any of the specified doses of tirzepatide and semaglutide against each other, placebo or other glucose-lowering drugs. For a glucose-lowering drug to be included as a comparator, it was required to have been evaluated in at least one trial comparison against tirzepatide and one trial comparison against semaglutide. This approach was adopted to prevent unconnected networks, ensuring that each comparator served as a link for indirect comparisons between tirzepatide and semaglutide. We included trials recruiting adults with type 2 diabetes regardless of their background glucose-lowering treatment, defined as the glucose-lowering therapy used both in the intervention and control arms after the randomisation.

We searched PubMed and Cochrane databases from inception until 11 November 2023. Our search strategy included both free-text and Medical Subject Headings (MeSH) terms, utilising the keywords ‘tirzepatide,’ ‘ly3298176,’ ‘semaglutide’ and ‘nn9535’ (electronic supplementary material [ESM] Table). 1

After deduplication, search results were screened at title and abstract level, and potentially eligible records were examined in full text with reasons for exclusion being recorded. Two independent reviewers performed the study selection process and any disagreements were resolved by a third reviewer. For the deduplication and the screening process we used the Systematic Review Accelerator (SRA) web application [11].

Using predesigned forms, we extracted information on study characteristics, participants’ baseline characteristics and outcome data. Given the aggregated data format of the included RCTs in our meta-analysis, direct information on how sex or gender was determined in the individual studies was beyond the scope of our analysis. Our two primary outcomes were the change from baseline in HbA_1c and in body weight. Secondary efficacy outcomes were the proportion of participants attaining an HbA_1c target of ≤48 mmol/mol (≤6.5%) or <53 mmol/mol (<7.0%), and those achieving a minimum of 10% body weight loss. Safety outcomes included the incidence (no. of participants with at least one outcome event) of nausea, vomiting, diarrhoea, treatment discontinuation due to gastrointestinal events, severe adverse events and severe hypoglycaemia (a hypoglycaemic event requiring assistance). Data were extracted from the intention-to-treat population, which included all randomly assigned participants who received at least one dose of the study medication. For eligible trials identified through our database searches, we utilised ClinicalTrials.gov, using their respective National Clinical Trial (NCT) identifiers, to retrieve additional information when outcome data were absent or incomplete in the published articles. Data extraction was conducted by two independent reviewers, with discrepancies resolved by a third reviewer.

We used version 2 of the Cochrane risk-of-bias tool for randomised trials (ROB 2) to assess the risk of bias for the two primary outcomes [12]. Following the tool’s algorithms, each trial’s overall risk of bias was classified as low if all domains were at low risk, and high if any domain was at high risk. If none of the domains were classified as high risk but one or more were deemed to have some concerns, the overall risk of bias for that trial was categorised as ‘of some concern’. This assessment was conducted independently by two reviewers, with a third reviewer resolving any disagreements. We evaluated the presence of small-study effect (publication bias) by means of comparison-adjusted funnel plots [13].

We explored the transitivity assumption by comparing the distribution of potential effect modifiers (baseline HbA_1c and body weight) across treatment comparisons [14]. We conducted frequentist random-effects network meta-analyses and calculated mean differences (MDs) for the two primary outcomes and risk ratios for dichotomous outcomes, alongside 95% CIs [15]. We evaluated heterogeneity for the primary outcomes based on the agreement between CIs and prediction intervals in relation to the null effect and the clinically important effect on the opposite direction to the point estimate [16, 17]. We assumed a minimum reduction in HbA_1c of 5.5 mmol/mol (0.5%) and in body weight of 4.5 kg (5% of mean body weight value at baseline across all trials) as clinically important [18]. We addressed incoherence (inconsistency) both locally by comparing directly with indirect evidence using the Separating Indirect from Direct Evidence (SIDE) method [19] and globally using the design-by-treatment interaction model [20]. Moreover, we used P-scores, ranging from 0 to 1, to rank treatments; these can be interpreted as the average degree of certainty for a treatment to be better than the other treatments in the network [21]. Statistical analyses were performed in R (R Core Team 2019, R Foundation for Statistical Computing, Vienna, Austria) using the R packages ‘meta’ and ‘netmeta’ [22], and in NMAstudio (version 2.0) web application [23, 24].

We evaluated Confidence In Network Meta-Analysis (CINeMA) effect estimates for the primary outcomes utilising the CINeMA methodological framework and application [17, 25]. The six domains evaluated were within-study bias (risk of bias), across-study bias (small-study effect/publication bias), indirectness, imprecision, heterogeneity and incoherence (inconsistency). We assigned judgements at three levels (no concerns, some concerns and major concerns) to each domain and summarised judgements across domains to an overall assessment ranging across very low, low, moderate or high level of confidence [17, 25].

The search retrieved 2798 records, of which 28 RCTs [6, 7, 26–51] with 23,622 participants were included in the systematic review and network meta-analysis (ESM Fig. 1). Study and participant characteristics are presented in Table 1. Only two trials directly compared tirzepatide with semaglutide, with one of these also including a placebo arm [6, 7]. Sixteen trials compared semaglutide with placebo, other GLP-1 RAs, basal insulin, prandial insulin or varying doses of semaglutide. The remaining ten trials compared tirzepatide with placebo, GLP-1 RA (other than semaglutide), basal insulin, prandial insulin or varying doses of tirzepatide. All trials had a parallel-group design and 15 were open-label (Table 1). Most trials were multinational, except for five that recruited exclusively Japanese participants [39–41, 48, 49]. The intervention duration ranged from 24 to 28 weeks in five trials and from 30 to 56 weeks in 21 trials. The remaining two trials, a trial with tirzepatide in people with obesity and type 2 diabetes (SURMOUNT-2) [50] and a cardiovascular outcomes trial with semaglutide (SUSTAIN 6) [31], had a duration of 72 and 104 weeks, respectively. The background glucose-lowering therapy, referring to the common treatment received by all trial groups post-randomisation, varied across the trials. However, the predominant background treatment was metformin, used either as monotherapy or in combination with other medications. Across all trials, 10,442 participants (44.2%) were female, participants’ mean HbA_1c at baseline was 66.6 mmol/mol (8.3%), mean body weight was 88.8 kg and mean age was 57.8 years (Table 1). The distribution of potential effect modifiers (HbA_1c and body weight at baseline) was deemed sufficiently similar across all treatment comparisons to assume that a network meta-analysis was appropriate (ESM Figs 2 and 3).

Figure 1 shows the network of comparisons used in the meta-analysis. Risk of bias for the change in HbA_1c was assessed as low in all trials except for one that was at high risk of bias and one with some concerns (ESM Table 2). For the change in body weight, seven trials were at high risk of bias and one trial had some concerns; all other trials were at low risk of bias (ESM Table 3). Comparison-adjusted funnel plots did not suggest the presence of small-study effect (ESM Figs 4 and 5). There was presence of heterogeneity in some comparisons, particularly those involving semaglutide 2.0 mg (ESM Tables 4 and 5). In terms of incoherence, the design-by-treatment interaction model did not identify global inconsistency in the analyses for both primary outcomes (ESM Tables 4 and 5), while local inconsistency was also low.

Compared with placebo, tirzepatide 15 mg was the most efficacious treatment in reducing HbA_1c (MD [95% CI]: −21.61 mmol/mol [−23.26 to −19.97] [−1.96% (−2.11 to −1.82)]), followed by tirzepatide 10 mg (−20.19 mmol/mol [−21.89 to −18.48] [−1.84% (−1.99 to −1.69)]), semaglutide 2.0 mg (−17.74 mmol/mol [−22.03 to −13.45] [−1.59% (−1.95 to −1.22)]), tirzepatide 5 mg (−17.60 mmol/mol [−19.36 to −15.84] [−1.60% (−1.75 to −1.44)]), semaglutide 1.0 mg (−15.25 mmol/mol [−16.73 to −13.77] [−1.39% (−1.52 to −1.26)]) and semaglutide 0.5 mg (−12.00 mmol/mol [−13.74 to −10.26] [−1.09% (−1.24 to −0.94)]) (Fig. 2 and ESM Fig. 6). In comparisons between tirzepatide and semaglutide, when HbA_1c was measured in mmol/mol, all tirzepatide doses were comparable with semaglutide 2.0 mg and superior to semaglutide 1.0 mg and 0.5 mg (ESM Table 6). Specifically, effect estimates (MD [95% CI]) for tirzepatide 15 mg vs semaglutide 2.0 mg, tirzepatide 10 mg vs semaglutide 1.0 mg, and tirzepatide 5 mg vs semaglutide 0.5 mg were, respectively, as follows: −3.87 mmol/mol (−8.22 to 0.48); −4.94 (−6.65 to −3.23); and −5.60 mmol/mol (−7.60 to −3.60) (ESM Table 6). When HbA_1c was measured in %, tirzepatide at doses of 15 mg, 10 mg and 5 mg demonstrated greater efficacy than semaglutide at doses of 2.0 mg (MD = −0.38% [95% CI −0.75% to −0.01%]), 1.0 mg (MD = −0.45% [95% CI −0.60% to −0.31%]) and 0.5 mg (MD = −0.51% [95% CI −0.68% to −0.33%]), respectively (ESM Table 7). The confidence in estimates for comparisons between tirzepatide and semaglutide was high to moderate, except for comparisons vs semaglutide 2.0 mg, where the confidence was generally low (ESM Table 8). Consistently with meta-analysis findings, tirzepatide 15 mg held the highest probability (P-score = 0.99) of being the most efficacious treatment in reducing HbA_1c (ESM Fig. 7).

Compared with placebo, semaglutide 2.0 mg (risk ratio = 7.73 [95% CI 5.62, 10.63]) and tirzepatide 15 mg (risk ratio = 7.01 [95% CI 5.73, 8.57]) were the most efficacious in achieving an HbA_1c target of ≤48 mmol/mol (≤6.5%) (ESM Table 9). In between-drug comparisons, tirzepatide 15 mg and 10 mg outperformed semaglutide 1.0 mg and 0.5 mg and tirzepatide 5 mg was superior to semaglutide 0.5 mg, while no differences were found between semaglutide 2.0 mg and any of the tirzepatide doses (ESM Table 9). Similarly, semaglutide 2.0 mg (risk ratio = 4.01 [95% CI 3.24, 4.95]) and tirzepatide 15 mg (risk ratio = 3.70 [95% CI 3.26, 4.20]) were the most efficacious in achieving an HbA_1c target of <53 mmol/mol (<7%) as compared with placebo (ESM Table 10). No differences were found when any of the tirzepatide doses were compared with semaglutide 2.0 mg or 1.0 mg, while all tirzepatide doses were superior to semaglutide 0.5 mg (ESM Table 10).

In comparisons vs placebo, tirzepatide was the most efficacious medication for lowering body weight, resulting in reductions ranging from 9.57 kg (95% CI 8.36, 10.78) with tirzepatide 15 mg to 5.27 kg (95% CI 3.98, 6.56) with tirzepatide 5 mg (Fig. 3). Semaglutide showed a less pronounced effect, with reductions ranging from 4.97 kg (95% CI 1.68, 8.26) with semaglutide 2.0 mg to 2.52 kg (95% CI 1.26, 3.78) with semaglutide 0.5 mg (Fig. 3). In between-drug comparisons, tirzepatide at doses of 15 mg, 10 mg and 5 mg demonstrated greater efficacy than semaglutide at doses of 2.0 mg (MD = −4.60 kg [95% CI −7.94, −1.26]), 1.0 mg (MD = −3.53 kg [95% CI −4.80, −2.25]) and 0.5 mg (MD = −2.75 kg [95% CI −4.23, −1.28]), respectively (ESM Table 11). The confidence in estimates for comparisons between tirzepatide and semaglutide was high to moderate, except for comparisons vs semaglutide 2.0 mg, where the confidence was low (ESM Table 12). Tirzepatide 15 mg was ranked highest (P-score = 1.00) among all treatments in terms of weight reduction (ESM Fig. 8).

All doses of tirzepatide and semaglutide were superior to placebo in achieving at least a 10% body weight reduction, with tirzepatide 15 mg (risk ratio = 10.51 [95% CI 7.55, 14.64]) and tirzepatide 10 mg (risk ratio = 8.84 [95% CI 6.35, 12.32]) being the most efficacious treatments (ESM Table). In between-drug comparisons, tirzepatide at both the 15 mg and 10 mg doses outperformed all doses of semaglutide, while tirzepatide at the 5 mg dose was more efficacious than semaglutide 0.5 mg (ESM Table). 13 13

Compared with placebo, all doses of tirzepatide and semaglutide demonstrated an increase in the risk for nausea (ESM Fig. 9), vomiting (ESM Fig. 10) and diarrhoea (ESM Fig. 11). Specifically, the risk ratios for nausea ranged from 2.07 to 3.51 across different doses of tirzepatide, and from 2.45 to 2.84 for semaglutide (ESM Table 14). For vomiting, the risk ratios ranged from 2.39 to 4.36 with tirzepatide, and from 2.33 to 3.62 with semaglutide (ESM Table 15). For diarrhoea, the risk ratios ranged from 1.81 to 2.18 with tirzepatide, and from 1.66 to 1.80 with semaglutide (ESM Table 16). In comparisons between tirzepatide and semaglutide, all doses of tirzepatide had similar risk profiles for gastrointestinal adverse events when compared with semaglutide 2.0 mg. However, tirzepatide 15 mg and 10 mg generally exhibited an increased risk compared with semaglutide 1.0 mg and 0.5 mg (ESM Tables 14–16). Discontinuation of treatment due to gastrointestinal adverse events was more frequent with any dose of tirzepatide (risk ratios ranging from 6.39 to 10.65) or semaglutide (risk ratios ranging from 4.99 to 8.91) compared with placebo (Fig. 4). No differences were observed when comparing tirzepatide with semaglutide, except for tirzepatide 15 mg vs semaglutide 0.5 mg (ESM Table 17).

Neither tirzepatide nor semaglutide were associated with an increased risk for serious adverse events when compared with placebo (ESM Fig. 12), and no differences were observed in the comparisons between tirzepatide and semaglutide (ESM Table 18). We did not conduct a meta-analysis for severe hypoglycaemia due to the absence of events in most treatment arms across all trials. In particular, in the overall population, 107 participants experienced an episode of severe hypoglycaemia, with 30 of the cases occurring in a single trial arm wherein participants were randomised to prandial insulin [46].

Our network meta-analysis of 28 RCTs found that s.c. tirzepatide generally had a more pronounced effect than s.c. semaglutide in reducing HbA_1c and body weight in people with type 2 diabetes. Notably, both drugs, particularly the higher doses of tirzepatide, were associated with an increased incidence of gastrointestinal adverse events. These findings can inform clinical decisions and optimising treatment strategies in the management of type 2 diabetes.

The datasets generated and/or analysed during the current study are available from the corresponding author on reasonable request.