This Bayesian NMA was conducted based on 23 RCTs, involving 11 SGLT2is, 32 interventions and 9144 individuals. This study compared the efficacy and safety of SGLT2is in patients with T2DM inadequately controlled with metformin monotherapy.

The findings indicated that, compared with metformin monotherapy, the addition of most SGLT2is to metformin showed favourable antidiabetic efficacy in reducing HbA1c, FPG and total BW, and acceptable safety regarding the incidence of AEs, SAEs, hypoglycaemia, UTI and GI. Combined SGLT2is could be an ideal option for T2DM patients inadequately controlled with metformin.

Regarding efficacy, 15 mg of ertugliflozin demonstrated the best efficacy in HbA1c reduction, while 300 mg of canagliflozin is the best option for FPG reduction and WL. The addition of some SGLT2is to metformin might be related to inferior safety outcomes, increasing the incidence of total AEs (eg, 10 mg of henagliflozin), hypoglycaemia (eg, 15 mg of ertugliflozin) and GI (eg, 10 mg of empagliflozin). According to the SUCRA values (online supplemental table S6), the interventions displaying the best safety were 400 mg of sotagliflozin in total AEs, 10 mg of ertugliflozin and 150 mg of ipragliflozin in SAEs, 300 mg of canaglilflozin in hypoglycaemia, 12.5 mg of ipraglilflozin in UTI and 25 mg of ertugliflozin in GI. Generally, the included SGLT2i exhibited significant difference in the incidence of AEs, SAEs, hypoglycaemia and GI. Among them, 5 mg of luseoglilfozin showed the greatest safety considering the incidence of total AEs, SAEs, hypoglycaemic events and UTI (ranked second, fourth, sixth, fifth, respectively). 300 mg of canagliflozin appeared to exhibit superior efficacy, since it ranked first in FPG reduction and WL and third in HbA1c reduction. Among the dual therapy regimens, 300 mg of canagliflozin demonstrated acceptable incidence of hypoglycaemia and GI, but was simultaneously related to high risks of AEs and SAEs. By contrast, 5 mg of luseogliflozin was relatively safe while being significantly inferior in glucose control.

Previous studies49–51 assessed the efficacy and safety of SGLT2is in adults with T2DM. Shyangdan et al 49 reported that canagliflozin and empagliflozin as an add-on to metformin were significantly more effective than placebo combined with metformin in HbA1c reduction and WL, which is consistent with our findings. This could be attributed to the dual action arising from combination therapy in suppressing hepatic glucose output and promoting renal glycosuria.52 Meanwhile, they also found that 300 mg of canagliflozin showed the greatest efficacy in reducing HbA1c and BW among the four interventions (ie, 100 mg and 300 mg of canagliflozin, 10 mg and 25 mg of empagliflozin). Zaccardi et al 50 indicated that the highest dose (ie, 300 mg) of canagliflozin reduced HbA1c and FPG to a greater extent than dapagliflozin and empagliflozin at any doses. Our study found that there was significant difference among SGLT2is as add-on therapy in WL. However, Zaccardi et al 50 and Tsapas et al 51 arrived at a similar conclusion that no differences of efficacy in WL were found between individual SGLT2is. This discrepancy may arise from variations in study design, since their studies included both SGLT2is monotherapy and combination therapies, while ours only included combination therapies. Additionally, only 3 and 4 SGLT2is were involved in their studies, respectively, but 11 SGLT2is were included in ours. Meanwhile, the differences between our study and previous studies may also be related to disparities in the racial composition of the included patients, as efficacy of SGLT2is in WL may be different between Asians and non-Asians.53

In terms of safety, there was no significant difference in the incidence of total AEs between most of SGLT2is and placebo. The difference across SGLT2is was also not significant. It is worth noting that sotagliflozin, a dual inhibitor of SGLT2 and SGLT1,54 showed relatively greater safety considering total AEs. As for hypoglycaemia, we found that the addition of most of the SGLT2is did not increase the risk compared with placebo. This could be attributed to the fact that the glucosuric effect of SGLT2is is bound to the filtered glucose load. When the filtered load is less than 80 g/d, SGLT2is become ineffective at further lowering blood glucose.55 However, our findings indicated that different SGLT2is exhibited varying degrees of risks of hypoglycaemia, for example, 5 mg and 50 mg of empagliflozin and 25 mg janagliflozin were observed to have higher risks, while 12.5 mg of ipragliflozin and 50 mg and 300 mg of canagliflozin were relatively safe.

The present Bayesian NMA showed that none of the combination therapies increased the risk of UTI compared with placebo (online supplemental table S5). Our findings differed from a previous meta-analysis,56 but were consistent with a recent real-world study.57 Moreover, the association between SGLT2is and the risk of UTI is not clear, with prior studies reporting conflicting findings.56–58 It should be noted that the safety profiles of SGLT2is may depend on races/ethnicities. For example, a meta-analysis found that, compared with placebo, the risk of UTI with SGLT2is treatment was not increased in Asians, but was significantly increased in non-Asians.53

SGLT2i-induced glucosuria plays a facilitating role in elevating the risk of GI.59 Accordingly, SGLT2is appear to easily lead to GI in T2DM patients.60–63 Nevertheless, in the current study, most of SGLT2is did not increase the risk of GI compared with placebo (online supplemental table S5). Moreover, the SUCRA values (online supplemental table S6) showed that 1 mg of ertugliflozin (93%) and 300 mg of ipragliflozin (92%) were safe, with significantly fewer GI. It is important to note that the data regarding the two interventions were obtained from individual studies, with a relatively small number of patients receiving these interventions (54 and 72, respectively). Therefore, the results should be interpreted with caution. The subgroup analysis of a previous research found that ipragliflozin was not associated with an increased risk of reproductive tract infections in either drug-naïve or metformin-based patients.64 Similarly, we also found that ipragliflozin at any doses combined with metformin was not related to a higher risk of GI (online supplemental table S5). Besides, 300 mg of ipragliflozin was safer than most of the other SGLT2is (online supplemental table S6). A real-world retrospective cohort study61 reported that, although the use of SGLT2is was associated with an increased risk of GI, no meaningful difference among individual SGLT2is was identified. Two studies65 66 suggested that SGLT2is may increase the risk of diabetic ketoacidosis (DKA). However, only a few cases of DKA were reported in the included studies. Accordingly, the risk of DKA was not assessed in this study.

The differences in efficacy and safety among SGLT2is may be attributed to population, study design, doses, outcome definition and/or pharmacological selectivity. Two studies67 68 investigated the impact of pharmacological selectivity of SGLT2is on efficacy and safety outcomes and found that the extent of selectivity of SGLT2is to SGLT2 over SGLT1 might be clinically relevant.

The present study systematically evaluated the safety and efficacy of 11 marketed SGLT2is and 32 interventions, including both newer agents (eg, henagliflozin) and relatively older ones (eg, canagliflozin) by synthesising evidence from RCTs. The results of the current study can provide clinicians with a reference to assess the advantages and weaknesses of SGLT2is comprehensively for T2DM patients inadequately controlled with metformin and also serve as valuable evidence for updating clinical practice guidelines. Our findings indicated that, among the dual therapy regimens, 300 mg of canagliflozin demonstrated optimal efficacy and acceptable incidence of hypoglycaemia and GI, but simultaneously increased the risks of AEs and SAEs. 5 mg of luseogliflozin was relatively safe while being significantly inferior in glucose control. Clinical decisions often involve multiple relevant outcomes and should take into account patient-specific factors such as comorbidities, tolerability and individual treatment goals. For example, for patients prioritising WL, 300 mg of canagliflozin might be an ideal option, but the high risk of AEs should be noted. The American Diabetes Association guideline emphasises the risks of GI when using SGLT2is.69 Based on the results of the present study, for patients at a high risk of GI, 300 mg of ipragliflozin may be a more suitable choice.

Some new SGLT2is (eg, 5 mg and 10 mg of henagliflozin) also showed promising efficacy and acceptable safety profiles. However, the evaluation of novel SGLT2is was limited by the availability of data, and some evidence included in this study exhibited moderate quality. Additionally, most of the included RCTs were placebo-controlled, and only two RCTs were a head-to-head study. Therefore, more large-scale, high-quality and active-controlled RCTs are needed to further validate our conclusions.

SUCRA rankings provide some guidance in clinical practice but come with certain limitations in their interpretation and application.70 For instance, differences in relative treatment effects might not be clinically meaningful.71 Therefore, in this study, to avoid the potential misinterpretation of SUCRA rankings, we used the MCID value for continuous outcomes.72 Nevertheless, although some differences among SGLT2is were statistically significant, many differences did not surpass MCID threshold, which may limit their practical impact in clinical practice.

We should acknowledge several limitations of this study. First, most of the data for novel drugs each were derived from only one trial with a limited sample size, which could introduce bias. Second, characteristics of placebo, races of included participants, baseline glycaemic control, dose of metformin and diabetes duration differ across RCTs. The variation in metformin dosage, baseline glycaemic control and diabetes duration could potentially influence the efficacy and safety outcomes. Some of the included studies involved patients from specific countries or regions, which may limit the generalisability of our findings to broader T2DM populations. Third, a degree of statistical heterogeneity was found in the NMA for some outcome measures (eg, GI). Hence, a sensitivity analysis was conducted and the results did not change substantially when excluding studies significantly influenced the heterogeneity of the model. Despite conducting a comprehensive literature search and including unpublished reports from ClinicalTrials.gov, the results still indicated the presence of publication bias in the analysis of GI probably due to missing data. The limited number of studies available for each comparison restricted our ability to investigate potential sources of publication bias, which could lead to highly misleading of the safety of competing interventions.73 Besides, local inconsistency was identified in the network for hypoglycaemia analysis, which may be attributed to differences in participant characteristics, such as ethnicity, across the included studies. We provided results of direct and indirect comparisons and excluded these studies in the main analysis. Considering these limitations, the results of the study should be interpreted cautiously.