Safety of sodium-glucose transporter 2 (SGLT-2) inhibitors in patients with type 2 diabetes: a meta-analysis of cohort studies

Aims: This study aimed to investigate the association between the use of sodium-glucose transporter 2 inhibitors (SGLT-2i) and the risk of diabetic ketoacidosis (DKA), lower limb amputation (LLA), urinary tract infections (UTI), genital tract infections (GTI), bone fracture, and hypoglycemia in cohort studies. Methods: A systematic search was conducted in the PubMed and Embase databases to identify cohort studies comparing the safety of SGLT-2i versus other glucose-lowering drugs (oGLD) in patients with type 2 diabetes mellitus (T2DM). The quality of the studies was assessed using the Newcastle-Ottawa Scale. Primary endpoints were DKA and LLA, while secondary endpoints included UTI, GTI, bone fracture, and hypoglycemia. Hazard ratios (HR) with 95% confidence intervals (CI) were calculated. Results: A total of 9,911,454 patients from 40 cohort studies were included in the analysis. SGLT-2i use was associated with a higher risk of DKA (HR: 1.21, 95% CI: 1.07–1.38, p = 0.003) and GTI (HR: 2.72, 95% CI: 2.48–2.98, p < 0.01). However, it was not associated with an increased risk of LLA (HR: 1.06, 95% CI: 0.92–1.23, p = 0.42), UTI (HR: 0.99, 95% CI: 0.89–1.10, p = 0.83), or bone fracture (HR: 0.99, 95% CI: 0.94–1.04, p = 0.66). Furthermore, SGLT-2i was associated with a reduced risk of hypoglycemia. Furthermore, compared to dipeptidyl peptidase 4 inhibitors, SGLT-2i as a class and individually was associated with an increased risk of DKA. Canagliflozin specifically increased the risk of LLA (HR: 1.19, 95% CI: 1.04–1.36, p = 0.01). The subgroup analysis suggested that SGLT-2i increased the risk of LLA among patients with a history of cardiovascular disease. Conclusion: SGLT-2i versus oGLD was associated with a similar occurrence of LLA, UTI, and bone fracture. However, SGLT-2i was associated with a higher risk of DKA and GTI than oGLD. These findings provide valuable information on the safety profile of SGLT-2i in patients with T2DM and can help inform clinical decision-making.


Introduction
Individuals diagnosed with type 2 diabetes mellitus (T2DM) face an increased risk of mortality and cardiovascular problems.SGLT2 inhibitors (SGLT-2i) have received approval to reduce blood glucose levels in adults with T2DM when used in conjunction with diet and exercise.In addition to lowering glucose, SGLT-2i offers additional benefits in addressing various aspects of metabolic syndrome, such as improving blood pressure, weight management, and lipid profiles.Furthermore, they have shown promising outcomes in reducing cardiovascular events and overall mortality.SGLT-2i is effective in lowering cardiovascular risk in individuals with T2DM by interrupting or mitigating several processes involved in atherosclerosis development, resulting in fewer cardiovascular complications.Additionally, promising outcomes in reducing cardiovascular events and overall mortality have been reported in studies conducted by (Hu et al., 2020;Li et al., 2021;Andreadi et al., 2023).Despite their apparent advantages, a degree of controversy has arisen regarding the potential link between SGLT-2i and an elevated risk of adverse events, including diabetic ketoacidosis (DKA), lower limb amputations (LLA), genital tract infections (GTI), and others.
Several meta-analyses and cohort studies have investigated the risk of DKA, urinary tract infections (UTI), GTI, and hypoglycemia in individuals with T2DM who received SGLT-2i treatment.Multiple meta-analyses and cohort studies have been performed to evaluate the risk of DKA, UTI, or GTI in T2DM patients receiving SGLT-2i treatment.The findings have exhibited some degree of inconsistency.Concerning the risk of DKA, three meta-analyses that synthesized data from randomized controlled trials (RCTs) (Tang et al., 2016;Monami et al., 2017;Donnan et al., 2019) indicated that SGLT-2i did not increase the risk of DKA in T2DM patients.In contrast, two additional meta-analyses (Liu et al., 2020;Salah et al., 2021) reported an elevated risk of DKA in these individuals due to SGLT-2i use.Likewise, three cohort studies (Wang et al., 2017;Kim et al., 2018;Han et al., 2021) found no increased risk of DKA associated with SGLT-2i use, while three different cohort studies (Douros et al., 2020;Fralick et al., 2021a;Patorno et al., 2021) identified an increased risk of DKA in T2DM patients treated with SGLT-2i.A study by Mantovani suggested that newer hypoglycemic medications with reduced potential for drug-induced hypoglycemia can mitigate the occurrence of severe hypoglycemia and related adverse effects, particularly among more susceptible patients (Mantovani et al., 2016).
Concerning UTI and GTI, several meta-analysis of RCTs (Storgaard et al., 2016;Wu et al., 2016;Tang et al., 2017;Zhang et al., 2018) revealed that SGLT-2i increased the risk of these infections when compared to the placebo group.In contrast, two additional meta-analyses (Zhang et al., 2018;Donnan et al., 2019) reported that the incidence of UTI and GTI in the SGLT-2i group closely resembled that in the placebo group.
Multiple meta-analyses of RCTs have explored the relationship between SGLT-2i and the risk of LLA, demonstrating that SGLT-2i use elevated the LLA risk compared to the placebo group (Zaccardi et al., 2016;Rådholm et al., 2018).In contrast, additional meta-analyses (Dorsey-Treviño et al., 2020;Kumar et al., 2020) did not identify a statistically significant increase in LLA risk associated with SGLT-2i use compared to control groups.The evidence of observational studies that examine the relationship between SGLT-2i and the risk of LLA presents a broader range of findings.Seven observational studies (Chang et al., 2018;Ryan et al., 2018;Udell et al., 2018;Ueda et al., 2018;Yuan et al., 2018;Dawwas et al., 2019;Yang et al., 2019) have generated varied outcomes concerning the link between SGLT-2i use and the risk of LLA.
Observational studies using real-world data offer valuable insight into the safety and efficacy of medications in clinical practice.However, when evaluating the risk of adverse events linked to SGLT-2i, the results of these studies have displayed inconsistency.While one study (Caparrotta et al., 2021) systematically assessed the safety of SGLT-2i based on observational studies, it is important to note that the studies included were published before 2020.Because medical practices and patient demographics can evolve, the applicability of older studies in capturing current safety patterns may be constrained.Another concern is that the baseline characteristics of participants in the included studies were not balanced through propensity score matching (PSM), a statistical technique used to mitigate confounding in observational research.In the absence of PSM, there is the possibility of imbalances in the baseline characteristics between the SGLT-2i and control groups, which introduces confounding variables that might impact the study's outcomes.Moreover, in some cases, participants in the included studies may not have been accurately classified as individuals with T2DM.Such misclassification could introduce additional confounding variables, potentially compromising the accuracy and reliability of the study's findings.Given these constraints, conducting updated meta-analyses of real-world data and employing more robust methodologies is imperative to gain a deeper understanding of the safety profile of SGLT-2i in individuals with T2DM.

Methods
This meta-analysis followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (Kumpf et al., 2023).The protocol was registered in the International Prospective Register of Systematic Review (PROSPERO, registration number CRD42021235831).

Data sources
Cohort studies investigating the safety or adverse events of SGLT-2i in patients with T2DM were identified by searching the PubMed and Embase databases.All eligible studies in English published up to 11 July 2023, were considered.Literature searches used specific keywords related to SGLT-2i, such as DKA, LLA, UTI, GTI, bone fracture, hypoglycemia, safety, and adverse effects.RCTs were excluded from the search strategy.The customized search strategy for each database is shown in Supplementary Appendix S2.

Study selection 2.2.1 Criteria for inclusion
Studies that met the following criteria were included in this meta-analysis: 1. Types of study: Prospective or retrospective cohort studies.2. Study populations: Patients with T2DM without restrictions on age, sex, or ethnicity.3. Study design: For retrospective cohort studies, baseline information from the observation and control groups had to be essentially the same, achieved through PSM.For prospective cohort studies, the comparability between baseline information from the observation and control groups was necessary.4. Interventions: The observation group received treatment with SGLT-2i as a class or as individual agents, while the control group received other glucose-lowering drugs (oGLD).5. Sample size: Studies with a sample size of 1,000 or more were included to minimize the heterogeneity of the pooled study arising from small sample size studies.
The primary outcomes were the occurrences of DKA and LLA.Secondary outcomes were the occurrences of GTI, UTI, hypoglycemia, and bone fractures.Studies were included if they reported at least one of these outcome measures.

Criteria for exclusion
The following were the exclusion criteria for this meta-analysis: 1. RCTs, reviews, systematic reviews, meta-analyses, case reports, case studies, case series, letters, opinions, audits, protocols, and methodologies.2. Studies in which the intervention did not meet the specified outcome measures.
Two investigators (T.T.L. and Q.Z.) independently screened the literature.In cases of discrepancies or uncertainties, consensus was reached through discussion with the other author (C.X.L.).

Data extraction
A standardized extraction form was used to collect the study data.The following data were independently extracted by three authors (T.T.L., Q. Z, and Q.X.): first author, year of publication, country, study population, age, sex, number of patients, intervention measure, follow-up time, and outcome measures.The data extraction forms were cross-checked to verify the accuracy and consistency of the extracted data.The third author (C.X.L.) checked all data and disagreements were resolved by discussion.

Quality assessment of the study
The quality assessment of the studies was conducted independently by three authors (T.T.L., Q.Z., and Q.X.) using the Newcastle-Ottawa Scale (Yu et al., 2018).This scale evaluates studies based on three main domains: selection, comparability, and exposure.Within the selection and exposure categories, each numbered item can be awarded a maximum of 1 point, while comparability can receive a maximum of 2 points.The total score ranges from 0 to 9, with higher scores indicating higher study quality.Studies were classified into low quality (scores 0-3), moderate quality (scores 4-6), and high quality (scores 7-9).

Statistical analysis
Meta-analysis was performed using Stata 16.0 software (StataCorp, College Station, TX, United States).Statistical heterogeneity between studies was assessed using the Cochran chi-square test complemented with the I2 statistic.I 2 values of 25%, 50%, and 75% indicate low, moderate, and high heterogeneity, respectively (Higgins et al., 2003).The randomeffects model was used for the analysis.The hazard ratio (HR) and the 95% confidence interval (95% CI) were used to describe categorical variables, where the p-value < 0.05 is considered significantly different.
To examine the sources of heterogeneity, we performed a metaregression analysis with intervention, study region, study year, sex proportion, follow-up time, and sample size as independent variables, and DKA, LLA, UTI, GTI, bone fracture, and hypoglycemia as dependent variables, respectively.To assess the stability of the results, a sensitivity analysis was systematically performed, excluding one study at a time to examine its impact on overall findings.
We utilized Confunnel plots for qualitative evaluation and Egger's test for quantitative analysis to assess publication bias.Confunnel plots display areas of statistical significance on a funnel plot, and it help distinguish publication bias from other causes of asymmetry.If the missing studies are in areas of non-significant, the asymmetry observed in the confunnel plot is caused by publication bias.In contrast, if the absent studies are statistically significant, the observed asymmetry is more likely due to factors other than publication bias based on statistical significance (e.g., variable study quality or non-statistical significancebased publication bias mechanisms) (Peters et al., 2008).P < 0.05 indicates a possible publication bias.

Study characteristics and quality assessment
The included studies spanned from 2017 to 2023, 22 published in 2020 or later.Among these studies, 23 were conducted in the US, 7 in northern Europe, three in Canada, and 3 in Asia.The proportion of women in the included studies ranged from 33.6% to 64.7%.The median duration of follow-up was 10.8 months (3.3-58.8).Baseline glycosylated hemoglobin (HbA1c) was reported in 13 studies, while the baseline prevalence of peripheral vascular diseases (PVD) was reported in 26 studies.A comprehensive summary of the characteristics of the included studies is presented in Table 1.
Studies have indicated that preexisting LLA can increase the risk of future LLA (Arnott et al., 2020).In the studies we included, researchers matched the baseline incidence of LLA between the groups using PSM or directly excluded patients with preexisting LLA.This approach helps minimize the influence of this confounding factor on our results.However, it is crucial to acknowledge that publication bias could potentially affect estimates of the association between the use of SGLT2i and the risk of LLA.
This study showed that SGLT-2i did not increase the risk of UTI or bone fractures.Several meta-analyses of RCTs have consistently shown that the risk of UTI did not increase with SGLT-2i compared to the placebo or active comparator groups (Storgaard et al., 2016;Wu et al., 2016;Puckrin et al., 2018).However, a subgroup analysis revealed a 6% increased risk of UTI when SGLT-2i was compared to DPP-4i.It is essential to interpret this result with caution, as it was mainly influenced by a study of a Korean study conducted in 2021 (Han et al., 2021).This study included older patients over 65 years of age, with a significant proportion of women (57.5%).T2DM itself is a known risk factor for UTI, regardless of the treatment regimen (Nitzan et al., 2015), and older women are particularly susceptible to UTI (Nicolle et al., 2014;Fioretto et al., 2016).Additionally, the study did not match the baseline incidence of UTI.Therefore, we must consider these specific factors and potential confounders when interpreting the association between SGLT-2i use and the risk of UTI in our study.
Our study revealed that SGLT-2i did not increase the risk of bone fractures, even in patients with T2DM over 65 years of age.These findings are consistent with the results of a previous study (Wu et al., 2016).However, concerning GTI, SGLT-2i use was associated with a three-fold increased risk, regardless of sex.This result remained consistent across different subgroup analyses based on various active comparators.Several meta-analyses have also reported an increased risk of GTI with SGLT-2i (Jabbour et al., 2018;Zhang et al., 2018).Fortunately, GTI is considered an adverse event of lesser severity than other safety outcomes and can be mitigated by educating patients to increase their water intake during medication.Furthermore, the hypoglycemic effect of SGLT-2i is independent of insulin, resulting in a 14% lower risk of hypoglycemia than other medications.

Strength and limitation of the study
Our investigation has a large cohort of 9,911,454 patients with T2DM and included data from 40 cohort studies conducted in various countries.Rigorous inclusion and exclusion criteria were used to ensure the robustness of the study.The PSM matched the baseline information of included studies and some excluded patients with preexisting LLA, further reducing potential confounding variables and increasing the reliability of our results.To address the study heterogeneity, we used meta-regression and subgroup analysis.Fortunately, our sensitivity analyses did not show significant differences compared to our primary analyses, reinforcing the reliability of our findings.We also assessed potential publication bias using Egger's test, which indicated no evidence of bias except for LLA and DKA.
However, our study has the following limitations.First, although all included studies had comparable demographic characteristics between treatment groups through PSM, there may still be residual confounding from some unmeasured or not fully measured factors (e.g., HbA1c level, diabetes duration, prior insulin use) that cannot be completely ruled out.Second, certain pooled studies showed high heterogeneity.Third, the discussion of the safety of SGLT-2i as an individual agent was limited due to the availability of limited data.Furthermore, the subgroup analysis included only a few studies included, and inconsistencies in the results of some subgroup analyses must be verified by more high-quality studies with larger sample sizes.

Conclusion
In a comprehensive population-based cohort study comprising 9,911,454 patients with T2DM, the use of SGLT-2i compared to oGLD showed a similar incidence of LLA, UTI, and bone fractures.However, SGLT-2i was associated with a higher risk of DKA and GTI than oGLD.The subgroup analysis indicated that the use of SGLT-2i was associated with an increased risk of LLA among patients with a history of CVD.Specifically, canagliflozin, empagliflozin, and dapagliflozin increased the risk of DKA compared to DPP-4i.Canagliflozin was associated with an elevated risk of LLA.

TABLE 1
Baseline characteristics of included studies.

TABLE 1 (
Continued) Baseline characteristics of included studies.

TABLE 1 (
Continued) Baseline characteristics of included studies.

TABLE 1 (
Continued) Baseline characteristics of included studies.

TABLE 1 (
Continued) Baseline characteristics of included studies.

TABLE 2
Results of secondary outcomes.