Efficacy and safety of once‐weekly insulins in type 2 diabetes: A systematic review and meta‐analysis

Abstract

Aims

Once‐weekly (OW) insulins represent a promising strategy to reduce injection burden and improve adherence in type 2 diabetes management. This meta‐analysis aimed to quantify pooled effect sizes for key biological and clinical outcomes, integrating new findings from the five large trials published in 2025.

Materials and Methods

We systematically searched PubMed, WoS, ClinicalTrials.gov and EU Clinical Trials databases up to June 2025 for randomized controlled trials (RCTs) evaluating OW insulins in type 2 diabetes. Primary outcomes included efficacy (HbA1c, fasting plasma glucose (FPG), time‐in‐range (TIR), time‐above‐range) and safety outcomes (body weight, hypoglycaemia, time‐below‐range). Data were analysed using both frequentist and Bayesian approaches.

Results

Among 435 screened records, 16 RCTs were included. OW insulins achieved a greater reduction in HbA1c compared to once‐daily insulin or semaglutide (pooled mean difference: −0.12%, 95% CI: −0.19 to −0.04, p = 0.007). TIR was significantly longer in the OW insulin group (+2.41% of total time, 95% CI: 1.13 to 3.69, p = 0.002), while FPG changes were comparable (−3.37 mg/dL, 95% CI: −7.95 to 1.21 mg/dL, p = 0.14). Body weight changes were generally similar, but excluding trials using IcoSema revealed a modest weight gain (+0.33 kg, 95% CI: 0.04 to 0.62, p = 0.030). The incidence of clinically significant or severe hypoglycaemia did not differ between groups (pooled incidence rate ratio: 1.04, 95% CI: 0.79 to 1.28, p = 0.75). Meta‐regression confirmed consistent effect sizes across OW insulin types, treatment durations and prior insulin regimens.

Conclusions

OW insulins modestly improve glycaemic control without increasing hypoglycaemia risk, though slight weight gain may occur.

1. INTRODUCTION

The primary objective of glucose‐lowering medications in the management of type 2 diabetes is to achieve and maintain optimal glycaemic and weight control, while also preventing cardiorenal complications in high‐risk individuals. ¹ Insulin therapy remains a cornerstone in the treatment regimen for many patients with type 2 diabetes due to its high efficacy in glycaemic control.^{2, 3} It may be considered when other antidiabetic medications fail, which occurs over 5 years in approximately 21% and 15% of patients treated with metformin and rosiglitazone monotherapy, respectively. ⁴ According to the American Diabetes Association 2024 guidelines, insulin therapy is typically initiated in patients presenting with random blood glucose levels ≥300 mg/dL (≥16.7 mmol/L), glycated haemoglobin (HbA1c) >10% (>86 mmol/mol), or if those experiencing symptoms of hyperglycaemia or evidence of catabolism. ¹ Insulin is also recommended when glycaemic targets are not achieved despite the use of GLP‐1 receptor agonists or dual glucose‐dependent insulinotropic polypeptide (GIP) and GLP‐1 receptor agonists. ¹ Around 20–25% of patients discontinue treatment with GLP‐1 receptor agonists or SGLT2 inhibitors 1 year after initiation. ⁵ As a result, approximately one in three individuals with type 2 diabetes will require insulin therapy within 7 years of diagnosis. ² However, traditional insulin regimens, which require daily injections, can be burdensome, potentially compromising patient adherence and satisfaction. ⁶

Recent advancements in pharmaceutical research have led to the development of long‐acting insulin formulations, such as insulin efsitora alfa (Eli Lilly™) and insulin icodec alone or in a fixed combination with semaglutide as IcoSema (Novo Nordisk™). ⁷ These novel formulations use chemical modifications to reduce insulin receptor affinity, resulting in slower clearance and enabling a weekly dosing schedule. These once‐weekly (OW) insulins potentially allow for improving adherence and quality of life. However, the efficacy and safety of OW insulins—particularly in comparison to traditional daily insulin regimens—remain key areas of ongoing investigation.

Several meta‐analyses have assessed the efficacy and safety of OW insulins in type 2 diabetes.^{8, 9, 10, 11, 12, 13, 14, 15, 16} However, we sought to update such a meta‐analysis given the availability of new findings in this population from four large trials published in June 2025, comparing OW insulins to daily insulins.^{17, 18, 19, 20} By aggregating data from randomized controlled trials (RCTs), this meta‐analysis aims to provide an updated, comprehensive evaluation of the potential benefits and risks associated with this emerging treatment modality.

2. MATERIALS AND METHODS

This meta‐analysis follows the PRISMA 2020 guidelines and has been registered in the PROSPERO database (CRD420251087354). ²¹

2.1. Eligibility criteria

We searched for RCTs evaluating OW insulin therapies in patients with type 2 diabetes. Table S1 presents the inclusion and exclusion criteria of the records. The primary outcome was the change in HbA1c levels from baseline. Secondary efficacy outcomes included time‐in‐range (TIR; percentage of time glucose was 70–180 mg/dL) and time‐above‐range (TAR; >180 mg/dL) via continuous glucose monitoring (CGM) and changes from baseline in fasting plasma glucose (FPG). Secondary safety outcomes were changes from baseline in body weight, CGM‐based time‐below‐range (TBR) and combined level 2 (<54 mg/dL, <3 mmol/L) and level 3 (severe cognitive impairment requiring assistance) hypoglycaemic events, as these pose the most clinically relevant immediate and long‐term risks. ²²

2.2. Search strategy

We systematically searched records from the following databases up to June 30, 2025: PubMed, Web of Science, ClinicalTrials.gov and the European Union Clinical Trials Register. We used the following search terms: ‘once‐weekly insulin’, ‘icodec’, ‘148‐0287‐A’, ‘AWIQLI’, ‘efsitora’, ‘basal insulin Fc’, ‘LY3209590’, ‘icosema’, ‘long‐acting insulin’ and ‘ultra‐long insulin’.

2.3. Data collection process

Data were collected from each report by the two authors, and disagreements were resolved by consensus. The collected data included baseline values for both treatment groups and between‐group effect estimates for changes from baseline in HbA1c, FPG and body weight. For TIR, TAR and TBR, data were extracted as the percentage of time measured by CGM along with the between‐group effect estimates. For hypoglycaemic events, we collected the percentage of participants experiencing level 2 and level 3 hypoglycaemia, as well as the incidence rate ratio.

2.4. Study risk of bias assessment

The Risk of Bias 2 (RoB‐2) tool was used to assess risk of bias in the included studies. ²³

2.5. Effect measures

For the continuous outcomes, we collected the between‐group estimated mean difference in change from baseline with the corresponding confidence interval (CI). For the incidence rate of hypoglycaemic events, we collected incidence rate ratios with CI. When not directly available, standard errors were calculated from 95% CI as (upper limit‐lower limit)/3.92 or from 90% CI as (upper limit‐lower limit)/3.29.

2.6. Synthesis methods

The meta‐analysis was conducted with JASP version 0.19.3 (University of Amsterdam), using both frequentist and Bayesian models. Forest plots were used to visually display results. Further details on the models are provided in the S1.

2.7. Investigation of heterogeneity and subgroup analysis

In the frequentist approach, the heterogeneity between the different studies was assessed using the I² statistic, with the interpretation scale proposed by the Cochrane Statistical Methods Group. ²⁴ In the Bayesian approach, the heterogeneity was assessed by the BF_rf for the standard deviation τ.

Subgroups analyses were conducted using three covariates: (1) nature of OW insulin; (2) previous insulin regimen: insulin‐naïve versus previously treated with insulin; and (3) treatment duration. Meta‐regression was used for testing for differences between subgroups using the Restricted Maximum Likelihood mixed‐effects model, including all three covariates together. To mitigate the risk of type I error associated with the limited number of included studies, a Bonferroni adjustment was applied, defining statistical significance as p < 0.017 (0.05/3).

2.8. Publication bias assessment

Publication bias was assessed visually using funnel plots and statistically using Egger's regression test for outcomes that included at least 10 studies in the meta‐analysis. ²⁵

3. RESULTS

3.1. Study selection

As shown in the PRISMA flow diagram for study selection (Figure S1), a total of 545 records were identified using our systematic approach, and a total of 16 RCTs were included after assessment for eligibility.^{17, 18, 19, 20, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36} Records from the COMBINE 1 trial (IcoSema vs. icodec) and the single‐arm ONWARDS 9 trial (icodec) were excluded as they did not meet our inclusion criteria related to trial design. Further details on the study selection are provided in the Supplementary Appendix.

3.2. Study characteristics and data selection

Among the 16 included RCTs, the OW insulin evaluated was icodec in eight trials, efsitora alfa in six and IcoSema in two (Table S2). Titration protocols in all RCTs were based on fasting self‐monitored blood glucose for both treatment groups (Table S3). All RCTs enrolled participants who had received stable doses of antidiabetic drugs for at least 90 days and excluded those with renal impairment (Table S4). Further details are provided in the Supplementary Appendix.

3.3. Risk of bias in studies

The assessment of risk of bias revealed a low risk in two RCTs and some concerns in the remaining 14 RCTs (Figure S2). The open‐label design used in these 14 RCTs was identified as the primary potential source of deviations from intended interventions.

3.4. Results of syntheses

3.4.1. Change in HbA1c

All 16 selected trials reported the between‐group difference in the change in HbA1c levels from baseline, including 5203 and 4530 participants in the intervention and control groups, respectively (Table S5). Mean HbA1c decreased in all 16 trials during the treatment period, ranging from −0.5% in Bajaj et al. (2021) to −1.68% in ONWARDS 5—the latter exhibiting the highest mean baseline HbA1c level (8.96%). OW insulins resulted in more reduction in HbA1c levels compared to comparators using both the frequentist (pooled mean difference: −0.12%, 95% CI: −0.19 to −0.04, p = 0.007; Figure 1) and Bayesian meta‐analysis approaches (averaged posterior estimated effect size: −0.12%, 95% credible interval: −0.20 to −0.03; Figure S3). The presence of an effect on HbA1c was 1.7 times more likely than the absence of an effect (BF₁₀ = 1.7), suggesting anecdotal evidence.

FIGURE 1.

Forest plot of between‐group differences in changes in HbA1c levels. The x‐axis represents the effect size, which corresponds to the between‐group difference in changes in HbA1c level from baseline (expressed as %). The y‐axis lists the individual studies included in the analysis, sorted by publication date. The right column displays the effect size along with the 95% confidence interval in brackets, followed by the weight of the studies. The black boxes indicate the effect size for each study, with the area of each box proportional to the weight of the study in the meta‐analysis.

Substantial heterogeneity was observed across studies with the frequentist approach (I² = 73%, 95% CI: 47–89%, p < 0.001), and the Bayesian approach concluded to extreme evidence for heterogeneity (BF_rf for τ = 17 287). The COMBINE 2 trial likely contributed to the observed heterogeneity, given the distinct nature of the comparator arm (semaglutide). Upon its exclusion from the meta‐analysis, heterogeneity was substantially reduced (I² = 41%, 95% CI: 0–83%, p = 0.07; BF_rf for τ = 0.52). The pooled effect size continued to favour OW insulins in both the frequentist (−0.09%, 95% CI: −0.15 to −0.03, p = 0.007) and Bayesian (−0.09%, 95% credible interval: −0.14 to −0.05, BF₁₀ = 7.4) frameworks (Figure S4).

Subgroup analysis using meta‐regression showed that the effect size did not significantly differ based on the previous insulin regimen, type of OW insulin or treatment duration (Table S6).

3.4.2. Time‐in‐range and time‐above‐range by continuous glucose monitoring

Of the 16 selected RCTs, 12 reported TIR values assessed by CGM and expressed as a percentage of total time (Table S7). However, TIR data from Bue‐Valleskey et al. (2023) and Frias et al. (2023) were excluded due to missing distributional data (CI or standard errors).^{29, 37} Data on TIR from Rosenstock et al. (2020) were excluded due to an inconsistent glucose target range (70–180 mg/dL vs. 70–140 mg/dL in other studies). ²⁶

In the remaining 9 studies that were suitable for inclusion in the meta‐analysis (representing 2977 participants in each group), CGM was conducted during the last 2 weeks of the treatment period for studies lasting 16 weeks. For trials with longer durations, CGM was performed during the final 5 weeks.

Employing the frequentist approach for the meta‐analysis, TIR was significantly longer in participants receiving OW insulins compared to those in the comparator arms, with a pooled estimate of the mean difference of 2.41% (95% CI: 1.13 to 3.69; p = 0.002; Figure 2). The Bayesian meta‐analysis approach produced an averaged posterior estimated effect size of 1.87% (95% credible interval: 0.91 to 2.80) (Figure S5). The evidence of the presence of an effect on TIR was considered as extreme (BF₁₀ = 181). Heterogeneity across the studies was deemed unimportant (I² = 11.6%, 95% CI: 0–84%, p = 0.34; BF_rf = 0.85).

FIGURE 2.

Forest plot of between‐group differences in time‐in‐range. The x‐axis represents the effect size, which corresponds to the between‐group difference in TIR (expressed as % of total time). The y‐axis lists the individual studies included in the analysis, sorted by publication date. The right column displays the effect size along with the 95% confidence interval in brackets, followed by the weight of the studies. The black boxes indicate the effect size for each study, with the area of each box proportional to the weight of the study in the meta‐analysis.

Subgroup analyses showed a comparable effect size for TIR regardless of the type of OW insulin, previous insulin regimen and treatment duration (Table S6).

Regarding TAR, 6 studies reported the mean difference in the percentage of total time, while only the estimated ratio of time was reported in COMBINE 3, precluding its inclusion in the meta‐analysis (Table S8). TAR was significantly shorter in participants receiving OW insulins compared to those in the comparator arms in both the frequentist (pooled estimate of the mean difference: −2.71%; 95% CI: −4.19 to −1.23; p = 0.005) and Bayesian (averaged posterior estimated effect size: −1.95%; 95% credible interval: −2.99 to −0.79; BF₁₀ = 58) approaches (Figure S6). Heterogeneity across the studies was deemed unimportant (I² = 0%, 95% CI: 0–82%, p = 0.42; BF_rf = 1.08). Subgroup analyses demonstrated that the previous insulin regimen, the type of OW insulin and treatment duration did not significantly predict the pooled effect size for TAR (Table S6).

3.4.3. Change in fasting plasma glucose

Of the 16 selected RCTs, 13 reported the between‐group mean difference in the change in FPG from baseline, including 3605 participants in each of the intervention and control groups (Table S9). FPG decreased in all 13 trials during the treatment period, ranging from −60 mg/dL in ONWARDS 1 to −10 mg/dL in the study by Bajaj et al.^{27, 32} OW insulins resulted in a similar reduction in FPG levels from baseline compared to comparators, using both the frequentist approach with a pooled mean difference of −3.37 mg/dL (95% CI: −7.95 to 1.21, p = 0.14; Figure 3) and the Bayesian approach with an averaged posterior estimated effect size of −0.61 95% credible interval: −2.40 to 1.19) (Figure S7). The evidence for an effect on FPG was anecdotal (BF₁₀ = 1.19).

FIGURE 3.

Forest plot of mean difference in changes in fasting plasma glucose. The x‐axis represents the effect size, which corresponds to the between‐group difference in changes in FPG from baseline (expressed as mg/dL). The y‐axis lists the individual studies included in the analysis, sorted by publication date. The right column displays the effect size along with the 95% confidence interval in brackets, followed by the weight of the studies. The black boxes indicate the effect size for each study, with the area of each box proportional to the weight of the study in the meta‐analysis.

Substantial heterogeneity was observed across studies using the frequentist approach (I² = 80%, 95% CI 63–94%, p < 0.001). Accordingly, the Bayesian approach indicated extreme evidence for heterogeneity (BF_rf for τ = 3.6 × 10⁶). The COMBINE 2 trial contributed significantly to this heterogeneity due to the nature of the comparator (semaglutide). Upon its exclusion from the meta‐analysis, the heterogeneity was reduced (I² = 61%, 95% CI 26–90%, p = 0.003; BF_rf for τ = 3.0). The pooled effect size remained similar between OW insulins and comparators using both the frequentist (−2.08 mg/dL, 95% CI: −5.71 to 1.54, p = 0.23) and Bayesian (−0.96 mg/dL, 95% credible interval: −2.49 to 0.78, BF₁₀ = 1.73) approaches (Figure S8).

Subgroup analyses demonstrated that the type of OW insulin, previous insulin regimen and treatment duration did not significantly predict the effect size between OW insulin and comparators in terms of the change in FPG (Table S6).

3.4.4. Change in body weight

All 16 selected RCTs reported the mean difference in body weight change from baseline, including 5203 and 4530 participants in the intervention and control groups, respectively (Table S10). The majority of studies (n = 15, 94%) observed weight gain during the treatment period, with increases ranging from +0.9 kg (Lingvay et al., 2021) to +3.9 kg (QWINT‐1).^{17, 28} In contrast, only the COMBINE 2 and 3 trials—which used semaglutide either as a comparator or in combination with icodec—reported a mean weight loss of approximately −3 kg.^{20, 36}

Using the frequentist approach, OW insulins resulted in a change in body weight similar to that of the comparators, with a pooled effect size of +0.19 kg (95% CI: −0.98 to 1.35; p = 0.74; Figure 4). The Bayesian meta‐analysis approach yielded a comparable conclusion, with an averaged posterior estimated effect size for body weight of +0.13 kg (95% credible interval: −0.80 to 1.06) for OW insulins compared with the comparator arms. This analysis was accompanied by a BF₁₀ of 0.49, indicating anecdotal evidence against a meaningful effect (Figure S9).

FIGURE 4.

Forest plot of mean difference in changes in body weight. The x‐axis represents the effect size, which corresponds to the between‐group difference in changes in body weight from baseline (expressed as kg). The y‐axis lists the individual studies included in the analysis, sorted by publication date. The right column displays the effect size along with the 95% confidence interval in brackets, followed by the weight of the studies. The black boxes indicate the effect size for each study, with the area of each box proportional to the weight of the study in the meta‐analysis.

A considerable degree of heterogeneity was observed across the studies (I ²  = 96%, 95% CI 95–99%, p < 0.001). Accordingly, the Bayesian approach indicated extreme evidence for heterogeneity (BF_rf for τ = 1.4 × 10⁷⁶). The COMBINE 2 and 3 trials likely contributed to this heterogeneity due to the well‐established effect of semaglutide on body weight. ³⁸ Upon their exclusion, the heterogeneity was markedly reduced (I ²  = 42%, 95% CI 0–85%, p = 0.048; BF_rf for τ = 0.82). The pooled effect size showed a higher increase in body weight with the use of OW insulins, using both the frequentist (+0.33 kg, 95% CI: 0.04 to 0.62; p = 0.030) and Bayesian (+0.30 kg, 95% credible interval: 0.10 to 0.52, BF₁₀ = 17) approaches (Figure S10).

Subgroup analyses demonstrated that the previous insulin regimen, the type of OW insulin and treatment duration did not significantly predict the pooled effect size between OW insulin and comparators in terms of body weight change (Table S6).

3.4.5. Incidence rate of clinically significant and severe hypoglycaemia

Of the 16 selected RCTs, 15 reported the incidence rate ratio of combined clinically significant (level 2) or severe (level 3) hypoglycaemic episodes, including 4945 participants in each of the intervention and control groups (Table S11). Eleven of the sixteen studies did not report the incidence rate ratio for severe hypoglycaemia, most often because no participants experienced such episodes.

Clinically significant or severe hypoglycaemic episodes were reported in a proportion of participants ranging from 0% in Lingvay et al. ²⁸ to 56% in QWINT‐4. Using the frequentist approach, the incidence of combined level 2 or level 3 hypoglycaemia was comparable between OW insulins and comparator groups, with a pooled incidence rate ratio of 1.03 (95% CI: 0.78 to 1.28; I² = 92%, 95% CI 59–93%, p = 0.82; Figure 5). The Bayesian meta‐analysis approach further supported these findings, yielding an averaged posterior estimated effect size for the rate ratio of 0.98 (95% credible interval: 0.68 to 1.32; BF_rf for τ = 2.1 × 10 ²⁸ ), indicating no meaningful difference in hypoglycaemia risk between OW insulins and comparators (Figure S11).

FIGURE 5.

Forest plot of the incidence rate ratio of combined level 2 or level 3 hypoglycaemic episodes. The x‐axis shows the effect size, transformed using the standardized mean difference (SMD) to log odds (logistic transformation), providing a logarithmic scale. The y‐axis lists the studies and their corresponding log‐transformed effect sizes with confidence intervals. The black boxes denote point estimates, with their areas proportional to the weight of the studies. The vertical line at an x‐value of 1.813 signifies the threshold corresponding to a rate ratio of 1.0 after de‐transformation. This line serves as a reference point, indicating where the incidence rate of hypoglycaemic episodes would be equal between the compared groups.

The incidence rate ratio was not computable in the study by Lingvay et al. ²⁸ due to a zero count in the control group. ²⁸ To address this, a continuity correction of 0.5 was applied in both groups. ³⁹ Upon integrating this study into the meta‐analysis, the pooled incidence rate ratio was recalculated to be 1.04 (95% CI: 0.79 to 1.28, p = 0.75).

Subgroup analyses demonstrated that the previous insulin regimen, the type of OW insulin and treatment duration did not significantly predict the pooled incidence rate ratio of level 2 or level 3 hypoglycaemia between OW insulin and comparators (Table S6).

3.4.6. Time‐below‐range

Ten studies reported data on TBR, all using the threshold of 0.70 mg/dL and eight using 0.54 mg/dL (Table S12). Only seven studies provided the difference in effect size, all referring to the 0.54 mg/dL threshold. Among these, three studies reported the mean difference as a percentage of total time, whereas four expressed it as the estimated ratio of time, precluding data pooling. As shown in Figure S12, TBR (<54 mg/dL) was comparable between OW insulins and comparators in both the frequentist and Bayesian approaches.

3.4.7. Sensitivity analyses

We tested various combinations of priors to assess the robustness of the Bayesian meta‐analysis. Overall, the pooled estimate of the effect size remained consistent across the different combinations of priors for all outcomes (Table S13).

3.4.8. Publication bias

Funnel plots and the corresponding statistical tests indicated no significant evidence of publication bias (Figure S13 and Table S14).

4. DISCUSSION

To our knowledge, this meta‐analysis is the first systematic evaluation of the efficacy and safety of OW insulins in type 2 diabetes by including all five large RCTs published in 2025 to reach a total of 16 RCTs. Furthermore, it is also the first to employ both frequentist and Bayesian approaches in the meta‐analysis, thereby offering a comprehensive and robust assessment of the available evidence. The findings from both statistical approaches converged, demonstrating that OW insulins provide significant benefits in glycaemic control relative to traditional daily insulin regimens, while demonstrating comparable effects on hypoglycaemia risk and a warning sign on body weight. Using meta‐regression, we demonstrated no significant impact of the type of OW insulins and treatment duration on the effect size on glycaemic control and hypoglycaemia. Similarly, the effect size was statistically equal in insulin‐naive participants and in those previously treated by daily insulins.

With regard to glycaemic control, our meta‐analysis demonstrated a modest yet statistically significant greater reduction in HbA1c levels with OW insulins compared to comparator groups (mean difference of −0.12%), based on aggregated data from 16 trials. While statistically significant, the difference of 0.12% in HbA1c reduction between OW insulins and comparators falls below the 0.3–0.5% threshold typically considered the minimally clinically important difference. ⁴⁰ A meta‐analysis by Saleem et al. reported a similar difference of −0.12% with icodec, though this result did not reach statistical significance. ⁴¹ However, their analysis included only four trials with insulin icodec published up to early 2023. Two other meta‐analyses, encompassing five and seven trials, respectively, observed a significant difference of −0.17% and −0.16% favouring icodec over once‐daily insulins. ¹⁵ Conversely, a recent meta‐analysis found no significant difference in HbA1c reduction, but this conclusion was based on only three trials comparing efsitora to degludec in type 2 diabetes. ⁹

The analysis of TIR in our meta‐analysis of 9 trials revealed a slight but significant increase of 2.41% with OW insulins compared to comparator groups, indicating better glycaemic stability. This corresponded to an estimated HbA1c reduction of 0.19%. ⁴² Four meta‐analyses reported a significant mean difference between 2.42% and 4.68% with icodec compared to once‐daily insulins in type 2 diabetes, but their analysis included only between three and seven trials, published up to 2023.^{12, 15, 41, 43} Another recent meta‐analysis found a non‐significant difference in TIR of 0.56%, but focused only on five trials comparing efsitora to degludec. ⁴⁴

OW insulins exhibited an effect on FPG similar to comparator groups in our meta‐analysis of 13 trials. This finding aligns with the findings of three prior meta‐analyses, which similarly reported no significant difference in FPG when analysing data obtained with icodec from seven trials for one analysis and four trials in the other two analyses.^{12, 15, 41} However, the meta‐analysis conducted by Saleem et al. identified a significant reduction in FPG favouring icodec over glargine, also based on the four oldest trials. ⁴¹

Regarding body weight, our meta‐analysis of all 16 RCTs revealed no significant difference between OW insulins and comparator groups. This overall finding was substantially influenced by the COMBINE 2 and 3 trials, in which the fixed‐ratio combination of icodec and semaglutide (IcoSema) resulted in weight loss, consistent with the well‐documented effects of semaglutide on body weight. ³⁸ Upon exclusion of these two trials, a modest yet statistically significant weight gain (+0.33 kg) was observed with OW insulins compared to once‐daily insulins. However, the difference of 0.33 kg between OW insulins and comparators falls below the 5 kg threshold typically considered the minimally clinically important difference. ⁴⁰ This result was consistent with three earlier meta‐analyses, which similarly reported a significant but modest (<1 kg) pooled weight gain associated with icodec.^{12, 15, 43} In contrast, two other meta‐analyses found no significant difference in body weight, but these analyses were based on a limited number of trials.^{9, 41} Overall, our findings underscore a notable consideration: the use of OW insulins may be associated with a slight increase in body weight, a factor that clinicians should weigh when selecting therapeutic regimens, particularly for patients at risk of obesity‐related complications. Further research is warranted to elucidate the long‐term metabolic effects of OW insulins and to identify strategies to mitigate potential weight gain.

The incidence of clinically significant or severe hypoglycaemia was comparable between OW insulins and comparators in our meta‐analysis of 15 trials (pooled incidence rate ratio: 1.04). This aligns with three meta‐analysis of trials testing icodec and with two meta‐analyses of trials testing efsitora.^{12, 15, 41} In contrast, a participant‐level meta‐analysis of the ONWARDS 1–5 trials found a + 51% higher incidence in level 2 and level 3 hypoglycemia. ⁴³

Our meta‐analysis has several strengths, including the inclusion of the largest number of trials to date, incorporating all five major RCTs published in 2025 for a total of 16 studies. Additionally, the dual application of frequentist and Bayesian statistical approaches provides a comprehensive and robust evaluation of the available evidence, enhancing the reliability of our conclusions. However, certain limitations must be considered. First, studies predominantly enrolled participants from high‐income countries with baseline HbA1c levels ≥7%, excluding those with recent hypoglycaemic episodes or advanced renal impairment. Consequently, the generalizability of these results to broader populations—such as those with well‐controlled diabetes (HbA1c at target), chronic kidney disease, or higher hypoglycaemia risk—remains uncertain. Including subgroup analyses in future trials could help assess the efficacy and safety in high‐risk populations, such as those with renal impairment or a history of severe hypoglycaemia. In particular, the safety and efficacy of transitioning from once‐daily to OW insulin in patients already at glycaemic target has yet to be thoroughly investigated and warrants further study. Second, the majority of included studies employed an open‐label design, which may introduce risk of detection and performance bias, particularly regarding subjective or behavioural outcomes. Awareness of treatment allocation could influence outcome reporting or assessment, potentially biasing treatment effect estimates. However, distinct injection devices and differing dosing frequencies between OW and once‐daily insulin regimens made blinding logistically difficult.

To conclude, this meta‐analysis confirms that OW insulins slightly improve glycaemic control without increasing hypoglycaemia risk relative to daily regimens. These findings offer the prospect of enhancing treatment adherence and patients' quality of life relative to daily regimens while maintaining comparable efficacy and safety in glycaemic management. However, several critical gaps with OW insulins remain to be addressed. First, long‐term studies extending beyond 2 years are essential to evaluate sustained efficacy and safety, particularly with respect to diabetic complications and mortality. Second, real‐world evidence on adherence, patient satisfaction and quality of life is needed to fully assess their clinical utility. Adherence to daily insulin therapy in type 2 diabetes is suboptimal in real‐world settings, generally ranging from 50 to 80%, and good adherence is crucial as it directly influences glycaemic outcomes.^{45, 46} Although real‐world data are lacking, improvements in adherence using OW insulins can reasonably be expected, similar to those observed with OW GLP‐1 receptor agonists compared with their daily formulations. ⁴⁷ Third, direct head‐to‐head comparisons between efsitora alfa and icodec are lacking, which would inform optimal treatment selection. Finally, the cost‐effectiveness of OW insulins and their coverage by healthcare systems represent key considerations for widespread adoption in the management of type 2 diabetes. Addressing these areas will be crucial to optimizing the integration of OW insulins into routine clinical practice.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Data S1: Supporting Information.

DATA AVAILABILITY STATEMENT

The data that support the findings of this study are available from the corresponding author upon reasonable request.