Use of SGLT2i Versus DPP-4i as an Add-on Therapy and the Risk of PAD-Related Surgical Events (Amputation, Stent Placement, or Vascular Surgery): A Cohort Study in Veterans With Diabetes

An estimated 37 million Americans have diabetes, of whom >90% have type 2 diabetes. Diabetes is increasing rapidly (1) and independently confers a risk for chronic conditions, including kidney disease, heart failure, blindness, nerve damage, cardiovascular disease (CVD), peripheral artery disease (PAD), and amputation (2,3).

Between 2013 and 2019, multiple placebo-controlled randomized trials evaluated the safety and efficacy of sodium–glucose cotransporter 2 inhibitors (SGLT2is) among participants with underlying CVD (4–9). In the Canagliflozin Cardiovascular Assessment Study (CANVAS) and CANVAS-Renal (CANVAS-R) trials, investigators reported an increased risk of amputation among SGLT2i users compared with placebo (6.3 vs. 3.4 events per 1,000 person-years). As a result, the U.S. Food and Drug Administration (FDA) issued a black box warning regarding the risk of amputation with canagliflozin in 2017 (10). Additional randomized placebo-controlled trials published in 2018–2019 included the Canagliflozin and Renal Outcomes in Type 2 Diabetes and Nephropathy (CREDENCE) and the Dapagliflozin Effect on Cardiovascular Events trial (DECLARE-TIMI 58). DECLARE-TIMI 58 was larger (N = 17,160 participants) with a broader study population, including patients at risk for or with established CVD, and CREDENCE was comparable in size to CANVAS but among those with kidney disease (6,11). Both studies found no statistical difference in amputation among SGLT2i users compared with placebo. In 2020, the FDA revised its guidance and removed the black box warning but retained the amputation risk in the warnings and precautions section of the prescribing information (4). Given this conflicting evidence, uncertainty remains about the use of SGLT2is in populations with an elevated risk of PAD.

This study tested the hypothesis that among U.S. veterans at high risk for PAD and revascularization, use of SGLT2is compared with DPP-4is is associated with an increased cause-specific hazard for the composite of surgical PAD outcomes (amputations and peripheral artery revascularization procedures). DPP-4i was chosen as the reference for multiple reasons: 1) it is proposed to have cardioneutral effects, 2) it is a contemporary comparator for SGLT2i in that both medications were available and newer medication classes during the study period, and 3) both are oral formulations (12–14).

We assembled a retrospective national cohort of patients who used the Veterans Health Administration (VHA) for their health care. Veterans were identified as having received their first prescription for a glucose-lowering medication between 1 October 2000 and 31 December 2016. Additional cohort follow-up data extended through 31 December 2021. VHA data included demographic, diagnostic, and procedure information from inpatient and outpatient encounters, laboratory results, and vital signs from clinical sources. Pharmacy data included medication dispensed, date filled, days supplied, and number of pills dispensed. Data included enrollment, claims files, and prescription (Part D) data for eligible veterans who were enrolled in Medicare or Medicaid. National Death Index and vital status files provided dates and causes of death. The institutional review board of the VA Tennessee Valley Healthcare System approved this study with a waiver of informed consent.

The source population included veterans aged ≥18 years who were regular VHA users, defined as an encounter or prescription filled at least once every 365 days for ≥2 years before cohort entry. We identified an inception cohort of patients at the start of new type 2 diabetes treatment, defined as patients who filled a first prescription for metformin, any insulin (long/short-acting or mixed), or sulfonylurea without any glucose-lowering drug fill in the prior 180 days. The earliest qualifying prescription fill represented the cohort entry date. From this inception cohort, we further identified new episodes of SGLT2i or DPP-4i use.

The index date was the day of new SGLT2i or DPP-4i prescription fill. SGLT2i medications included empagliflozin, dapagliflozin, and canagliflozin. DPP-4i medications included alogliptin, linagliptin, saxagliptin, and sitagliptin. We defined a new episode of use as a prescription fill for one of the study drug classes without prior use of SGLT2is or DPP-4is in the past 180 days. We allowed the use of metformin, any insulin or insulin combination, or sulfonylurea as cotherapies (or combinations) in the 180 days before the index date.

We then created a washout period in the 90 days prior to the index date by restricting the use of any medication class, other than the cotherapies of metformin, insulin, or sulfonylureas. Thus, any use of SGLT2is, DPP-4is, or glucagon-like peptide 1 receptor agonists (GLP-1RAs) in the 90 days prior to the index date was prohibited (Supplementary Fig. 1). This allowed for the evaluation of the new medication without contamination or withdrawal or switching from a different medication class. For example, a patient would qualify for a new episode of SGLT2i if the medication was not filled in the prior 180 days (new user) and there was no concurrent prescription for either DPP-4i or GLP-1RA in the preceding 90 days. We further excluded patients who used all three cotherapy medications (metformin plus sulfonylurea plus insulin), another drug class as cotherapy (e.g., acarbose, thiazolidinedione), dialysis, organ transplant, or hospice care within the 2 years before the index date.

Study exposures included persistent use of any study SGLT2i or DPP-4i. DPP-4i was chosen as the comparator due to its safety, weight-neutral and cardioneutral effects, and length of time in the market (FDA approved in 2006) (15). During the time frame of the study, empagliflozin was the preferred formulary SGLT2i added in 2016. There were two preferred formulary DPP-4is: Saxagliptin was the preferred agent between 2014 and 2019, and alogliptin was the preferred agent between 2019 and 2024.

Medication exposure ascertainment, covariate measurement, and follow-up started at the index date through the study outcome or censoring. Follow-up ended at the earliest of a study outcome, death, end of the study (31 December 2021), loss to follow-up (181st day of no VHA contact either inpatient, outpatient, or pharmacy use), or nonpersistence of that drug class (the 91st day without medication fill). Persistence was defined as the days’ supply of the medication plus an additional 90 days as a grace period prior to the persistence window ending on day 91 of no medication supply. Additional censoring criteria included crossover to another medication class, including GLP-1RA. Switching within medication class was allowed. A patient who finished follow-up could reenter the cohort as a new episode if all entry criteria were satisfied, including new-user status (180 days without prescription fill).

The primary outcome was the time to the first surgical event for PAD. This was considered the first surgical date for amputation, peripheral revascularization and bypass, or peripheral vascular stent. For patients who experienced more than one event during follow-up, we only considered the first event within each 30-day window to be the outcome date. Events separated by >30 days were considered two events; however, these analyses evaluated time-to-first event and did not include subsequent or recurring events. For example, a toe amputation would be viewed as the first event even if there was a forefoot amputation 25 days later. Secondary outcomes were associations of each exposure with the individual outcome.

The outcome definition was adapted from the Veterans Aging Cohort and the Million Veteran Program for the PAD and amputation phenotype (16,17). Each phenotype outcome was identified using ICD-9, ICD-10, and Current Procedural Terminology surgical procedure codes (16,17) (Supplementary Table 1).

Amputations were categorized as minor (toe, partial foot, forefoot, midfoot) and major (below the knee, above the knee). Revascularization procedures were defined as bypass, endarterectomy, and atherectomy. Peripheral stents were defined as angioplasty with stent placement. Prior studies have shown that compared with a review of medical records, the specificity and sensitivity of these definitions were very high at 88% and 100%, respectively (16,17).

Study covariates were measured on the index date and included age, sex, self-reported race derived from VHA and Medicare, index date, and a surrogate for diabetes duration (years from cohort entry to index date) (8). We accounted for diabetes cotherapies (metformin, sulfonylurea, and insulin) and the Veterans Integrated Service Network, a geographic designation used by VHA that allows for a more granular estimation of geographic variation of diabetes care. Physiologic variables were defined as the most recent measure on or up to 720 days before the index date. They included BMI, blood pressure, HbA_1c, LDL, hemoglobin, proteinuria, ejection fraction on echocardiogram, and creatinine. Creatinine was used to calculate the estimated glomerular filtration rate using the Chronic Kidney Disease Epidemiology Collaboration 2021 equation without race adjustment (7,18). Indicators of health care use included hospitalization, nursing home, number of outpatient visits or medications, and Medicare or Medicaid insurance use. Categorical data also included smoking or a history of alcohol misuse, which was noted as positive if there was ever a diagnosis code for these conditions. Definitions for comorbidities are noted in Supplementary Table 2.

The primary analysis used Cox proportional hazards models to compare time to first event between SGLT2i and DPP-4i users in a propensity score–weighted cohort. The unit of analysis was the episode of medication use. A propensity score modeled the probability of SGLT2i use given baseline covariates using a logistic regression model. Missing covariates were mean imputed, and an indicator for imputed covariates was included in the model. We used matching weights to balance the distribution of observed baseline covariates between exposure groups (19) (Supplementary Methods and Supplementary Figs. 2 and 3). DPP-4i was the reference, and all outcome models were adjusted for covariates and used robust SEs to account for patients who contributed multiple episodes of medication use. Missing covariates in the outcome models were handled using 30 iterations of chained imputations (9). The secondary analysis used a multistate model to estimate cumulative incidence in the presence of competing risks for all-cause mortality (20,21).

We report cause-specific adjusted hazard ratios (aHRs) and 95% CIs using pool SEs (22). For each comparison, statistical significance was set at a two-sided P < 0.05. The proportional hazards assumptions were verified through examination of Schoenfeld residuals over time, and follow-up was truncated at 4 years because of sparse data yielding uncertainty in the proportional hazards assumptions (23). When calculating the adjusted risk differences (aRDs), the aHRs are used as approximate estimates of relative risk, and the reference rate was calibrated to DPP-4i users. The inverse Kaplan-Meier estimates of the survival function were used to generate the cumulative incidence curves for the time to event in the weighted cohorts. Pointwise CIs for the cumulative incidence curves were generated using 5,000 bootstraps (24). Nonparametric estimates of the cumulative incidence of the primary outcome, surgical PAD, accounting for the competing risks of death were generated using the Aalen-Johansen estimator. Secondary analysis evaluated the incidence of each individual component outcome.

We conducted a sensitivity analysis that extended the persistence window grace period and time of follow-up for 360 days after the patient prescription fill ended. This allowed us to capture any PAD events that may have begun while on medications and resulted in the provider stopping the medication. In this analysis, the surgical outcome could have occurred up to 360 days after medication nonpersistence or changing medication class. Subgroup analyses included prior history of PAD and osteomyelitis (yes, no), along with insulin use at baseline (yes, no). We used an interaction term to test for an effect measure modification between each exposure and each subgroup to assess whether there were differential prescribing preferences for patients with underlying PAD or insulin use at baseline. We assessed covariate imbalance to assess for an impact of confounding by indication. Finally, we assessed the robustness of results to unmeasured confounding using an E-value, which estimates the minimum strength of an unmeasured confounder needed to nullify the observed effect (25). All analyses were performed using R version 4.1.2 (R Foundation) (17).

The protocol, statistical code, and deidentified and anonymized data set are available from the corresponding author upon written request.

We identified 657,247 new episodes of the two medication classes under investigation (Fig. 1). We excluded 109,821 episodes because of the use of other agents during the washout period. Additional exclusions were for data errors (n = 114); no VHA use in the prior 2 years (n = 60,968); cotherapy not metformin, insulin, or sulfonylureas (n = 117,695); prescription outside the study time frame (n = 251); missing key data (n = 79); hospice care (n = 1,702); organ transplant (n = 2,079); and dialysis use in the past 2 years (n = 3,067). We also excluded DPP-4i users (n = 33,252) with a prescription before April 2013, when the first SGLT2i became FDA approved, because there is a lack of propensity score overlap in prescription fill index date for those patients. The unweighted sample included 115,124 SGLT2i and 213,095 DPP-4i episodes. After propensity score weighting, the cohort included 76,072 SGLT2i and 75,833 DPP-4i episodes. SGLT2i episodes were 94% empagliflozin, 4% canagliflozin, and 2% dapagliflozin. DPP-4i episodes were 45% saxagliptin, 34% alogliptin, 15% sitagliptin, and 6% linagliptin.

The unweighted and weighted characteristics of SGLT2i and DPP-4i users are described in Table 1. After weighting, patient characteristics were similar between each exposure group as demonstrated by standardized mean differences of <0.1 (Supplementary Fig. 3). Within the DPP-4i and SGLT2i weighted cohort, the median HbA_1c was 8.4% (interquartile range [IQR] 7.5–9.4%), median age was 69 years, and median diabetes duration of 10.1 (IQR 6.6–14.6) years. Approximately 12,013 individuals (8%) in the weighted cohort had a prior history of PAD, and 6,255 (4%) had a diagnosis of osteomyelitis in the previous 2 years.

The median follow-up for the weighted cohort was 0.69 (IQR 0.34–1.43) years for SGLT2i and 0.71 (IQR 0.38–1.42) years for DPP-4i. Censoring occurred for the following reasons among SGLT2i and DPP-4i episodes: end of study, 37.0% vs. 33.7%; stopped study drug, 33.0% vs. 33.3%; addition of other diabetes medication, 21.6% vs. 23.2%; no VHA care, 2.9% vs. 3.1%; and all-cause death, 2.4% vs. 3.6%.

There were 874 PAD events for SGLT2i users and 780 PAD events for DPP-4i users. The event rate was 11.2 (95% CI 10.5–11.9) per 1,000 person-years for SGLT2i users and 10.0 (95% CI 9.4–10.6) per 1,000 person-years for DPP-4i users, with an aRD of 1.8 (95% CI 0.8–2.9) events per 1,000 person-years. In the weighted covariate-adjusted analyses, SGLT2i was associated with an increased cause-specific hazard of PAD events compared with DPP-4i (aHR 1.18 [95% CI 1.08–1.29]) (Table 2). The cumulative probability of a PAD event at 4 years was 4.0% for SGLT2i vs. 2.8% for DPP-4i. Figure 2A considers the outcome of PAD events alone, and Fig. 2B considers the cause-specific incidence of PAD and death as a competing risk (Supplementary Table 4). The Aalen-Johansen plots demonstrate divergence in all-cause mortality, with the cumulative incidence of death lower among SGLT2i versus DPP-4i users (5,6,11,26).

When evaluating individual outcomes, SGLT2i users had 499 amputations and 508 revascularizations or stents and an event rate of 6.4 events per 1,000 person-years for amputations and 6.5 events per 1,000 person-years for surgical revascularization (stent and bypass combined). DPP-4i users had 451 amputations and 438 revascularizations or stents for an event rate of 5.7 events per 1,000 person-years for amputations and 5.6 events per 1,000 person-years for a revascularization procedure. Results were consistent in the association of SGLT2i with amputation events (aHR 1.15 [95% CI 1.02–1.29]) and revascularization events (aHR 1.25 [95% CI 1.11–1.41]).

In a sensitivity analysis, we allowed for events to occur up to 360 days after the medication fill ended to account for the chronicity of PAD and recognition that providers may discontinue medications when a PAD event, such as a toe or foot ulcer, begins. Results were consistent with the main analysis (Fig. 2C and D and Supplementary Tables 3 and 4).

The test for an interaction between exposure and PAD and osteomyelitis was not statistically significant (P = 0.73). The test for an interaction effect between medication exposures and insulin use at baseline was also nonsignificant (P = 0.12), suggesting that there was no differential effect of SGT2i across subgroups. The characteristics of SGLT2i and DPP-4i users by insulin subgroups are detailed in the Supplementary Tables 5 and 6 and Supplementary Figs. 4 and 5. Subgroup event rates among those without underlying PAD and osteomyelitis and among nonusers of insulin were low but consistent with the primary results (Supplementary Table 3).

The unweighted cohort had 1,326 revascularization procedures and 1,115 amputations among DPP-4i users. Of these, 903 were minor (foot or toe) and 212 were major (above- or below-the-knee amputations). There were 115,124 unweighted SGLT2i users among whom there were 687 revascularizations and 658 amputations (canagliflozin n = 18, empagliflozin n = 635, dapagliflozin n = 5). Of these, 576 were minor amputations and 82 were major amputations.

In an analysis that assessed the sensitivity to unmeasured confounding, the E-value was 1.37. A hypothetical confounder would need to have a relative risk of 1.37 between exposure and the composite PAD-related surgical outcome to render results inconclusive (Supplementary Results).

In a national cohort of high-risk veterans with type 2 diabetes, SGLT2i use as an add-on to common treatment regimens was associated with an increased cause-specific hazard of a composite PAD surgical outcome. This finding was consistent when evaluating amputations and revascularization procedures separately. Additionally, results were consistent among prespecified sensitivity analyses evaluating outcomes at 90 and 360 days and when accounting for the competing risk of death.

Our findings provide important information for clinicians and patients using SGLT2is by evaluating a more inclusive PAD outcome in a high-risk group of veterans with diabetes. The results from CANVAS and CANVAS-R show an HR of 1.97 (95% CI 1.41–2.75) for amputations among canagliflozin users versus placebo (5). Similar to the FDA response, the European Medicines Agency issued an official warning in 2017 for all SGLT2i medications for lower-limb amputation risk (27). The cause-specific risk for PAD should be considered in light of the competing risk of a reduced all-cause mortality. The extended survival time for SGLT2i users emphasizes the established cardiovascular and renal protective effects and contribute to the reduction in mortality seen in Fig. 2. Clinicians must consider whether a patient will likely benefit from an SGLT2i (including death reduction) versus being more likely to experience an adverse effect (e.g., PAD).

When looking at recent literature, multiple placebo-controlled trials, including CREDENCE, Dapagliflozin in Patients With Chronic Kidney Disease (DAPA-CKD), and DECLARE-TIMI 58, found no effect of SGLT2i use on the risk of lower-limb amputations (6,11,26). Meta-analyses pooling data from randomized trials and retrospective cohort studies have mixed results. Recent meta-analyses, such as those by Kaze et al. (28) and See et al. (29), reported a relative risk of 1.21 (95% CI 0.85–1.72) and HR of 1.33 (95% CI 0.92–1.92) for amputations, with wide CIs in both evaluations. Li et al. (30) and Qui et al. (31) found increased point estimates and wide CIs for SGLT2i use and lower-limb amputations among populations with a baseline prevalence of PAD >10%. All meta-analyses report large confidence bounds, likely reflecting imprecision related to sample size and follow-up.

Other studies report risk specifically with canagliflozin rather with SGLT2i as a class. An observational study using the Taiwan National Health Insurance Research Database found a reduction in lower-limb ischemia requiring revascularization (HR 0.73 [95% CI 0.54–0.98]) or amputation (HR 0.43 [95% CI 0.30–0.62]) for first-line SGLT2i empagliflozin and dapagliflozin users versus second-line DPP-4i users (32). A meta-analysis of 39 randomized trials reported an increased risk of amputation and PAD among canagliflozin-treated patients with an odds ratio (OR) of 1.23 (95% CI 1.08–1.40) versus non-SGLT2i users. Notably, the risk of events was higher for >52 weeks duration compared with non-SGLT2i (amputation, OR 1.22 [95% CI 1.07–1.39]; PAD, OR 1.22 [95% CI 1.03–1.44]) (33,34).

The mechanism by which SGLT2i may contribute to worsening PAD and amputations remains unknown. Tanaka and Node (35) speculated that the increased risk of amputations may be seen in canagliflozin-treated populations during the course of major adverse cardiovascular events reduction. They suggested that stopping SGLT2i may increase the risk of amputation through worsened glycemic control and progression of arteriosclerosis, negatively impacting perfusion. Lin et al. (33) hypothesized that hemodynamic instability occurs during the early stages of SGLT2i use and contributes to intravascular damage.

Given the differences seen in the trials and observational studies, along with the lack of a firm mechanism by which SGLT2i contributes to PAD, an expert panel reviewed the available clinical trials data through 2019. They concluded that amputation risk increased with canagliflozin only and was not an SGLT2i class effect. However, the panel acknowledged that more data are needed before definitive conclusions are made (36).

Our study builds on previous studies and includes a large, high-risk group of veterans with diabetes. Notably, our study predominantly evaluated empagliflozin (>94% of the users) and accounted for 95% of the SGLT2i PAD events vs. 4% canagliflozin, accounting for 4% of outcome events; therefore, we are unable to examine the association of individual SGLT2i with surgical PAD outcomes. Amputations are relatively rare clinical events; thus, it is possible that studies with a short follow-up and relatively small sample sizes did not have enough time and power to fully capture outcomes. PAD is a chronic disease process with a long time from disease incidence to the terminal amputation surgery. A major strength of this study is the broad outcome definition that includes revascularization procedures. This serves to capture events occurring earlier in the disease process of PAD than amputation. Other strengths of this study are the use of a trial emulation approach and real-world methodology encompassing medication use and tracking, the large data set available from the combination of VHA and Centers for Medicare & Medicaid data sources, study of a high-risk cohort of veterans with long-standing diabetes, and extensive control of covariates.

Several limitations should also be noted. First, this evaluation excluded patients whose initial treatment of diabetes was not metformin, insulin, or sulfonylurea. Approximately 57% of the cohort used SGLT2is or DPP-4is as third-line treatment, indicating a longer duration of diabetes. For patients newly diagnosed with type 2 diabetes, the American Diabetes Association continues to recommend lifestyle modification as first-line therapy, followed by individual comorbidity and risk factor assessment for determination of pharmacologic therapies (37). Recently, SGLT2is have been used more often, including as first- and second-line therapy, given their benefit in heart failure and chronic kidney disease (38). Second, while some patients had up to 4 years of follow-up, the median follow-up for both groups was ∼0.7 years. This limited follow-up time could impact the number of amputations and revascularization events. There also remains the possibility of outcomes incompletely captured from veterans seeking care outside of the VHA, Medicare, or Medicaid. However, we would not expect this undercount to be differential between the two medications. Third, an E-value of 1.37 indicates that a moderate confounder would render the study findings inconclusive. The plausibility of moderate confounders depends on the thoroughness of the covariates. We extensively accounted for confounding variables using propensity score weighting and direct covariate adjustment; however, there may still be residual confounding, including confounding by indication. Providers may have prescribed SGLT2i preferentially over DPP-4i for patients with a higher risk of CVD. Finally, our study population mainly consisted of White men. Therefore, results may not generalize to women and other groups not strongly represented within this cohort.

In conclusion, among a national cohort of older veterans with diabetes, we found that the addition of SGLT2is was associated with an increased risk of amputation and peripheral revascularization surgical events compared with the addition of DPP-4is. These results were robust when evaluating each outcome and strengthened by prespecified sensitivity analyses. Further studies and evidence are needed on the effect of SGLT2is on broad PAD outcomes in higher-risk and more diverse populations, particularly in light of the variation of findings in the existing literature. These results underscore the need to determine the safety of SGLT2i use among patients with diabetes who remain at very high risk for PAD. We believe that a more risk-stratified approach to SGLT2i prescribing with respect to risks of PAD and cardiovascular benefit is warranted (39,40).

Funding. This project was funded by VA Clinical Science Research and Development investigator-initiated grant CX000570-12 (to C.L.R.). C.L.R. and A.H. were supported in part by Center for Diabetes Translation Research grant P30DK092986. Support for all investigators was also provided in part by the VETWISE-LHS Center of Innovation (CIN 24-128). Veterans Affairs/Centers for Medicare & Medicaid Services data were provided by the Department of Veterans Affairs, Veterans Affairs Health Services Research and Development Service, and Veterans Affairs Information Resource Center (project nos. SDR 02-237 and 98-004).

The VA Clinical Science Research and Development had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The contents do not represent the views of the U.S. Department of Veterans Affairs or the U.S. government.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. K.E.G., K.S., A.H.J., A.H., R.G., and C.L.R. contributed to the study conduct and data collection. K.E.G., K.S., and C.L.R. drafted the manuscript. K.E.G., A.H., R.G., C.G.G., and C.L.R. contributed to the study design. A.H.J., A.H., and R.G. contributed to the data analysis. All authors contributed to the critical revision of the manuscript. A.H.J. and C.L.R. are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. Parts of this study were presented in abstract form at the 84th Scientific Sessions of the American Diabetes Association, Orlando, FL, 21–24 June 2024.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Elizabeth Selvin and M. Sue Kirkman.