Use of Sodium–Glucose Cotransporter 2 Inhibitors in Patients With Type 1 Diabetes and Rates of Diabetic Ketoacidosis

Inhibition of sodium–glucose cotransporter 2 (SGLT2) in the proximal tubules suppresses renal glucose reabsorption, resulting in urinary glucose excretion and lowering of blood glucose in patients with diabetes. Canagliflozin (1), dapagliflozin (2), empagliflozin (3), and ertugliflozin (4) are SGLT2 inhibitors indicated as adjuncts to diet and exercise to improve glycemic control in adults with type 2 diabetes. Because of proven cardiovascular benefits (5,6), canagliflozin is additionally indicated to reduce the risk of major adverse cardiovascular events and empagliflozin is indicated to reduce the risk of cardiovascular death in adult patients with type 2 diabetes and established cardiovascular disease. To date, no SGLT2 inhibitor has been approved by the U.S. Food and Drug Administration (FDA) for the treatment of type 1 diabetes.

Diabetic ketoacidosis (DKA), a life-threatening complication of diabetes, has been observed after exposure to approved SGLT2 inhibitors in patients with type 2 diabetes and during off-label use in patients with type 1 diabetes (7,8). Clinical trials of sotagliflozin, an investigational dual sodium–glucose cotransporter 1 and SGLT2 inhibitor, demonstrated a dose-dependent increased risk of DKA compared with placebo in patients with type 1 diabetes treated with insulin (9–11). Across these trials, the DKA risks ranged from 2.3 to 4.2% among patients randomized to sotagliflozin 200 or 400 mg, and from 0 to 0.6% among patients randomized to placebo. The sotagliflozin clinical trials, referred to as trials 309 (9), 310 (10), and 312 (11), included patients with type 1 diabetes from North America, Europe and Israel, and worldwide, respectively, who were followed for 52 weeks (trials 309 and 310) or 24 weeks (trial 312) to determine the efficacy and safety of sotagliflozin. Because of the controlled trial environment and the use of patient and provider instructions to mitigate the risk for DKA, it is unclear whether DKA rates observed in these trials can serve as a basis for projecting DKA risk associated with SGLT2 inhibitor use to treat type 1 diabetes in the real world.

In the U.S., the Centers for Disease Control and Prevention estimated an age-adjusted annual DKA hospitalization rate of 3.0 per 100 persons with diabetes in 2014, which did not account for diabetes type (12). In the U.K., hospital admissions for DKA occurred at a rate of 3.6 per 100 person-years between 1998 and 2013 among patients with type 1 diabetes and 0.085 per 100 person-years among patients with type 2 diabetes (13). In the U.S., in-hospital case-fatality rates following DKA, regardless of diabetes type, decreased from 1.1% in 2000 to 0.4% in 2014 (12), but a mortality rate as high as 4.1% was found in Israel (14).

The frequency of off-label SGLT2 inhibitor use in patients with type 1 diabetes and the associated risk of DKA in the real world are unknown. The objectives of this descriptive study were to estimate the extent of real-world off-label use of SGLT2 inhibitors in patients with type 1 diabetes and to estimate real-world rates of DKA following exposure to SGLT2 inhibitors among U.S. patients with type 1 diabetes. Additionally, using data from sotagliflozin clinical trials as the reference, we contrasted observed and expected counts of DKA during off-label use of SGLT2 inhibitors in patients with type 1 diabetes. These data were presented before the FDA’s Endocrinologic and Metabolic Drugs Advisory Committee on 17 January 2019, which discussed the risks and benefits of sotagliflozin when used in patients with type 1 diabetes.

This analysis was conducted in 17 data partners of FDA’s Sentinel System, which includes administrative claims data from >100 million patients who are enrolled in commercial or public health insurance plans (15). The study period began 1 March 2013 (month of canagliflozin approval) and ended between 31 May 2015 and 30 June 2018, depending on the data partner.

Using outpatient pharmacy dispensing data, we identified all new users of canagliflozin, dapagliflozin, or empagliflozin, of any age, with continuous medical and pharmacy benefits (allowing coverage gaps up to 45 days) during a 365-day baseline period prior to their index dispensing of a study drug, without a dispensing of a member of the same drug class. We created exposure episodes for each study drug using days of supply, permitting a gap between dispensings of ≤10 days and applying a 10-day extension at the end of the episode. Exposure episodes were censored at the end of SGLT2 inhibitor supply (+10 days), first DKA event, disenrollment, death, or the end of available data, whichever occurred first.

Using adaptations of validated algorithms (16–18), we categorized patients with type 1 diabetes based on two definitions. Type 1 diabetes–broad required that a majority (>50%) of each patient’s diabetes diagnosis codes during the baseline period be specific to type 1 diabetes (see Supplementary Data). This definition was of primary interest for the drug-use analysis, because its expected high sensitivity maximizes capture of off-label use, providing an upper-bound estimate of use. Type 1 diabetes–narrow additionally required at least one dispensing for short- or rapid-acting insulin and no oral antidiabetic drug dispensing (except metformin) during the baseline period. This definition was of primary interest for the DKA analysis because its expected high positive predictive value minimizes the inclusion of falsely categorized type 2 diabetes patients, who are at a lower risk for DKA. For both definitions, we disregarded diabetes codes within 5 days prior to the index dispensing due to concerns that type 2 diabetes codes may be used in patients with type 1 diabetes to facilitate reimbursement for SGLT2 inhibitors.

DKA was defined using an inpatient or emergency department diagnosis with an ICD-9-CM code 250.1x or an ICD-10 code E1x.1x in any position (19).

This study was descriptive. To quantify off-label use among initiators of each study drug, we calculated the proportion of SGLT2 inhibitor users who met type 1 diabetes criteria. Analyses of DKA rates were conducted among patients age ≥25 years due to few DKA events being observed among younger patients and a requirement by one of the data partners that cell sizes of 1 to 10 events not be displayed. Incidence rates of DKA were calculated using counts of the first DKA event observed during exposure episodes in the numerator and cumulative person-years of exposure in the denominator. Analyses were stratified by age and sex. We extracted cohort-specific baseline patient characteristics including age, sex, history of use of antidiabetic drug(s) and continuous subcutaneous insulin infusion (herein referred to as insulin pumps), and diagnosis for DKA during the baseline period. In a secondary analysis, we included sitagliptin, a dipeptidyl peptidase 4 (DPP-4) inhibitor, as a separate exposure of interest, primarily to gauge performance of the diabetes algorithms. Because of its lack of efficacy in type 1 diabetes (20), sitagliptin use among type 1 diabetes patients was expected to be infrequent. Thus, observing minimal type 1 diabetes prevalence among sitagliptin users would provide additional validation of the algorithms. Additional secondary analyses for all study drugs were performed in patients with type 2 diabetes.

We calculated age- and sex-adjusted standardized incidence ratios (SIRs) for DKA, comparing the Sentinel population of type 1 diabetes patients exposed to an SGLT2 inhibitor with subjects randomized to sotagliflozin in trials 309, 310, and 312. For these trials, we obtained age- and sex-specific follow-up duration and event counts for treatment-emergent, positively adjudicated DKA events, considering the first DKA event within each patient. We calculated expected age- and sex-specific DKA event counts in Sentinel using the distribution of age- and sex-specific follow-up time in Sentinel and the clinical trials’ age- and sex-specific DKA rates. The SIR was calculated as the number of observed events (in Sentinel) divided by the number of events that would be expected if the Sentinel population experienced DKA at the rate observed in the clinical trials.

Analyses were performed in SAS, version 9.4 (SAS Institute, Cary, NC). We used OpenEpi to calculate 95% CIs of DKA rates and SIR (21). Sentinel is a public health surveillance activity that does not require the approval of institutional review boards (22,23).

This analysis was conducted using distributed methods. Each participating organization executed identical programs on their own data, behind their individual firewalls. The Sentinel Operations Center aggregated data returned by each participating organization to form the study data set.

The study sample included 297,633 new users of canagliflozin, 79,311 new users of dapagliflozin, and 98,583 new users of empagliflozin, regardless of diabetes status or type (Supplementary Fig. 1). The mean episode duration for each drug ranged from 4 to 5 months, during which patients received on average four dispensings of their cohort-defining drugs.

Among new users of SGLT2 inhibitors, between 0.47% (empagliflozin) and 0.51% (canagliflozin) of patients met the criteria for type 1 diabetes–narrow (Fig. 1A) and between 0.74% (empagliflozin) and 0.98% (canagliflozin) met the criteria for type 1 diabetes–broad. The proportion of new users who met the criteria for type 1 diabetes–narrow (Fig. 1B) or type 1 diabetes–broad (Fig. 1C) was highly age dependent. For instance, among patients who initiated canagliflozin between ages 12 and 18 years, 14.0% met the criteria for type 1 diabetes–broad and 10.8% met the criteria for type 1 diabetes–narrow. In contrast, among canagliflozin initiators age ≥65 years, 0.72% of patients met the criteria for type 1 diabetes–broad and 0.24% met the criteria for type 1 diabetes–narrow. Among 667,468 new users of sitagliptin, 0.44% and 0.12% met the criteria for type 1 diabetes–broad and type 1 diabetes–narrow, respectively.

Supplementary Table 1 displays baseline characteristics of SGLT2 inhibitor and sitagliptin users who met the criteria for type 1 diabetes. The mean (±SD) age ranged from 51.0 years (±11.9) to 52.3 years (±12.9) among users of SGLT2 inhibitors with type 1 diabetes–narrow. Initiators of sitagliptin were, on average, ∼10 years older. Across all SGLT2 inhibitors, females comprised 53% of those who met the criteria for type 1 diabetes–narrow. Patients who initiated canagliflozin or dapagliflozin tended to enter the study cohort in earlier calendar years, compared with patients who initiated empagliflozin, reflecting the drugs’ order of approval. Demographics of patients with type 1 diabetes–narrow or type 1 diabetes–broad were comparable.

Within each study drug cohort, baseline use of metformin was more prevalent among patients defined by type 1 diabetes–broad compared with type 1 diabetes–narrow, while the use of injectable glucagon-like peptide 1 (GLP-1) analogs was largely comparable. Compared with sitagliptin users, SGLT2 inhibitor initiators were more likely to have used metformin and sulfonylureas during the baseline period but were less likely to have used GLP-1 analogs. Consistent with type 1 diabetes–narrow criteria, the use of short- and rapid-acting insulin (i.e., lispro, regular, glulisine, aspart, lispro protamine) was more common among patients who met the criteria for type 1 diabetes–narrow than type 1 diabetes–broad. Among initiators of SGLT2 inhibitors, the use of long- or intermediate-acting insulin tended to be comparable between patients who met the criteria for type 1 diabetes–narrow and type 1 diabetes–broad.

Among initiators of SGLT2 inhibitors, insulin pump use was present during the baseline period in 30.6–41.2% of patients who met type 1 diabetes–narrow criteria and 17.3–27.7% of those who met the criteria for type 1 diabetes–broad. In contrast, 3.8% of sitagliptin users who met criteria for type 1 diabetes–narrow and 1.6% of sitagliptin users who met type 1 diabetes–broad criteria had prior insulin pumps use.

During the baseline year, DKA events occurred among 4.1% (empagliflozin) to 5.6% (canagliflozin) of SGLT2 inhibitor initiators with type 1 diabetes–narrow and in 6.7% of sitagliptin initiators with type 1 diabetes–narrow. Baseline characteristics for patients with type 2 diabetes are shown in Supplementary Table 2.

The study sample included 2,304 SGLT2 inhibitor initiators age ≥25 years who met the criteria for type 1 diabetes–narrow and 4,288 initiators who met the criteria for type 1 diabetes–broad (Table 1). Among these patients, rates of DKA during exposure to SGLT2 inhibitors ranged from 6.09 (95% CI 3.20–10.59) per 100 person-years for empagliflozin to 7.84 (95% CI 5.97–10.13) per 100 person-years for canagliflozin among patients with type 1 diabetes–narrow, with lower DKA rates observed among patients with type 1 diabetes–broad (Fig. 2). Among sitagliptin users, DKA rates were 5.28 (95% CI 3.18–8.29) in patients with type 1 diabetes–narrow and 1.70 (95% CI 1.09–2.53) per 100 person-years in patients with type 1 diabetes–broad.

Rates of DKA among initiators of any SGLT2 inhibitor were 8.47 (95% CI 6.23–11.27) among females and 5.87 (95% CI 4.06–8.22) among males with type 1 diabetes–narrow (Table 1). DKA was most frequent in patients aged 25–44 years, and especially among females aged 25–44 years, who experienced DKA at a rate of 19.74 (95% CI 13.27–28.32) per 100 person-years. DKA rates tended to be lower among patients with type 1 diabetes–broad but followed a similar pattern in subgroups by age and sex.

In contrast, among patients ≥25 years who met the criteria for type 2 diabetes, DKA rates ranged from 0.32 (95% CI 0.26–0.40) for dapagliflozin to 0.44 (95% CI 0.40–0.48) per 100 person-years for canagliflozin (Supplementary Fig. 2 and Supplementary Tables 2–4). Among SGLT2 inhibitor users with type 2 diabetes, rates of DKA were higher among those with baseline exposure to insulin (0.69 per 100 person-years) compared with those not on insulin (0.31 per 100 person-years) (Supplementary Table 4).

In the Sentinel sample, SGLT2 inhibitor initiators who met the criteria for type 1 diabetes–narrow and were ≥25 years old experienced 75 DKA events during 1,025.5 person-years at risk, resulting in a rate of 7.31 per 100 person-years. On the basis of age- and sex-specific incidence rates from the sotagliflozin clinical trials, 41 DKA events were expected in the Sentinel population (Fig. 3), resulting in an SIR of 1.83 (95% CI 1.45–2.28). Age-specific, sex-standardized SIRs decreased with increasing age, suggesting a larger discrepancy in observed versus expected events in younger patients. Patients with type 1 diabetes–narrow who were between 25 and 44 years old at the time of SGLT2 inhibitor initiation experienced a 2.6-fold higher rate of DKA compared with what would be expected based on clinical trial data (SIR = 2.57; 95% CI 1.85–3.45).

This analysis of almost 500,000 new users of SGLT2 inhibitors found that off-label use among patients who met study criteria for type 1 diabetes accounted for a small proportion of overall use. Among young patients, the proportion of SGLT2 inhibitor initiators with type 1 diabetes was numerically higher, which was possibly related to the epidemiology of type 1 diabetes and type 2 diabetes (24). Young patients also experienced the highest rates of DKA. In addition, for patients who met the type 1 diabetes–narrow criteria, DKA rates observed in Sentinel were higher than expected based on the sotagliflozin clinical trials, with the largest discrepancy among patients 25–44 years old at the time of drug initiation.

Placebo-controlled clinical trials showed an increased risk of DKA for canagliflozin (25,26), dapagliflozin (27,28), empagliflozin (29), and sotagliflozin (9–11) in patients with type 1 diabetes. A meta-analysis of type 1 diabetes clinical trials found a combined risk ratio for DKA of 2.66 (95% CI 1.32–5.35) in trials that compared canagliflozin, dapagliflozin, or empagliflozin with placebo, while the risk ratio was 5.80 (95% CI 2.11–15.88) for sotagliflozin (30). Several observational studies found higher rates of DKA associated with SGLT2 inhibitors in patients with type 2 diabetes when compared with DPP-4 inhibitors (31), GLP-1 analogs (32), or nonuse of SGLT2 inhibitors (33). However, another study comparing SGLT2 inhibitors with non-SGLT2 inhibitor antidiabetic drugs found an increased risk for DKA that was not statistically significant (34), and a Korean study did not find a higher risk for DKA comparing SGLT2 inhibitors with DPP-4 inhibitors (35).

As expected, DKA rates observed in this study among SGLT2 inhibitor users with type 1 diabetes exceeded those observed among type 2 diabetes patients by an order of magnitude, but they also exceeded DKA rates observed in other type 1 diabetes populations (12,13). Patients with type 1 diabetes who were between the ages of 25 and 44 years, and especially females in this age group, were at the highest risk for DKA after exposure to SGLT2 inhibitors, compared with older age groups. This is consistent with prior research findings supporting that female sex and younger age are associated with an increased risk for DKA in adults with type 1 diabetes (36,37), which may be precipitated by lower insulin adherence and reduced glycemic control (38,39). In addition to these factors, a reduction in insulin concentrations during SGLT2 inhibitor therapy may contribute to the development of DKA in patients with type 1 diabetes, regardless of age (7). The high risk of DKA following SGLT2 inhibitor use in all patients with type 1 diabetes indicates the need for the development, evaluation, and dissemination of risk mitigation strategies (40).

We detected, as expected, less off-label use of sitagliptin compared with SGLT2 inhibitors. In addition, we found that DKA rates among patients with type 1 diabetes–broad tended to be lower for patients who initiated sitagliptin compared to SGLT2 inhibitors (Fig. 2). Yet, DKA rates were comparable between initiators of sitagliptin, dapagliflozin, and empagliflozin in patients who met the criteria for type 1 diabetes–narrow. It is unclear whether the DKA rates observed among sitagliptin users who met the criteria for type 1 diabetes–narrow reflect pharmacologic properties, different patient characteristics, or random error. Because our study was not designed to formally test a hypothesis of a differential DKA risk between SGLT2 inhibitors and sitagliptin, it does not support conclusions about comparative safety.

Strengths of this analysis include a large sample size, a geographically diverse patient population within the U.S., and the wide age range captured by the Sentinel database. However, because Sentinel primarily includes large commercial data partners and Medicare, our study underrepresents patients covered by Medicaid and uninsured patients. Because the relative prevalence of type 1 diabetes and type 2 diabetes varies by age (24), overall rates of off-label use of SGLT2 inhibitors may vary between databases that differ in their age distributions. Consequently, because age is a determinant for DKA risk, our age-specific analyses are important with regard to generalizability.

Possible limitations arise from use of diagnostic codes and prescription claims to classify patients with type 1 diabetes and to ascertain DKA events. First, even though the type 1 diabetes–broad algorithm had a positive predictive value of 96.4% in a validation study (17), its performance may not be consistent across age categories or in patients who initiate SGLT2 inhibitors or sitagliptin. For this reason, we created the type 1 diabetes–narrow algorithm that included additional criteria that have shown high positive predictive values (16). The stricter definition is expected to have reduced the inadvertent inclusion of patients with type 2 diabetes in the type 1 diabetes–narrow cohort compared with the type 1 diabetes–broad cohort. Because type 2 diabetes is associated with lower rates of DKA, the inadvertent inclusion of such patients may explain the lower DKA rates observed in the type 1 diabetes–broad cohort. Conversely, because the type 1 diabetes–narrow and, to a lesser degree, the type 1 diabetes–broad cohorts likely missed some patients with type 1 diabetes, we potentially underestimated off-label use. Second, our algorithm to detect DKA was validated with a positive predictive value of 88.9% in patients up to age 24 years (19), but we are not aware of a validation study in adults. Regardless of age, any claims-based DKA algorithm can only detect DKA events that presented for care and were properly coded. Thus, we may have missed some DKA events while possibly including false-positives.

The estimated SIRs should be interpreted with caution. The clinical trial investigators trained patients to recognize symptoms that may lead to DKA, which may have prevented some DKA events. However, the extent of such training in clinical practice is unclear. Also, DKA events that occurred in the clinical trials are likely to be accurately ascertained through active screening and adjudication processes. Conversely, detection of DKA events in Sentinel relies on the accuracy and completeness of billing data. Other factors that may differentially affect rates of DKA in Sentinel, compared with clinical trials, include the following: a potentially higher risk for DKA associated with sotagliflozin at the doses tested in the clinical trials compared with approved SGLT2 inhibitors (30), different patient characteristics based on differences in inclusion/exclusion criteria, international settings of the trials compared with the U.S. focus of the Sentinel data partners, and shorter average exposure episodes in Sentinel compared with the clinical trials.

The proportion of patients using SGLT2 inhibitors who met the criteria for type 1 diabetes was low in the overall study population but higher in younger patients. Among patients who used SGLT2 inhibitors off-label for type 1 diabetes, the risk for DKA was notable, especially among patients under the age of 45 years and especially for females in this age group. For patients who met the type 1 diabetes–narrow criteria, DKA rates observed in Sentinel were higher than expected based on the sotagliflozin clinical trials, especially among younger patients.

Acknowledgments. The authors thank William Chong, Mitra Rauschecker, and Lisa Yanoff, Division of Metabolism and Endocrinology Products, Office of New Drugs, Center for Drug Evaluation and Research, FDA, Silver Spring, MD, for input in study design and critical review of the manuscript. Data partners who provided data used in the analysis: Aetna, Blue Bell, PA; Blue Cross Blue Shield of Massachusetts, Boston, MA; Duke University School of Medicine, Department of Population Health Sciences, Durham, NC, through the Centers for Medicare & Medicaid Services, which provided data; Harvard Pilgrim Health Care Institute, Boston, MA; HealthCore, Inc., Translational Research for Affordability and Quality, Alexandria, VA; HealthPartners Institute, Minneapolis, MN; Humana, Inc., Healthcare Research, Miramar, FL; Kaiser Permanente Colorado Institute for Health Research, Denver, CO; Kaiser Permanente Center for Health Research Hawai’i, Honolulu, HI; Kaiser Permanente Mid-Atlantic States, Mid-Atlantic Permanente Research Institute, Rockville, MD; Kaiser Permanente Northern California, Division of Research, Oakland, CA; Kaiser Permanente Northwest Center for Health Research, Portland, OR; Kaiser Permanente Washington Health Research Institute, Seattle, WA; Marshfield Clinic Research Institute, Marshfield, WI; Meyers Primary Care Institute, Worcester, MA; OptumInsight Life Sciences Inc., Boston, MA; and Vanderbilt University Medical Center, Department of Health Policy, Nashville, TN, through the TennCare Division of the Tennessee Department of Finance & Administration, which provided data.

Funding. This project was supported by task orders HHSF22301001T and HHSF22301012T under master agreement HHSF223201400030I from the U.S. Food and Drug Administration.

The content reflects the views of the authors and should not be construed to represent the U.S. Food and Drug Administration’s views or policies.

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

Author Contributions. C.Ha. participated in all steps of the study and wrote the manuscript. R.S.S., Y.Q., and S.K.D. helped design the analysis, interpreted the data, and provided critical review of the manuscript. C.Ho., E.D., A.P., R.C.T., and J.C.M. participated in data analysis and interpretation of results and reviewed the manuscript. C.A.P. helped design the analysis, participated in data analysis, interpreted the data, and provided critical review of the manuscript. C.Ha. and C.A.P. 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 study data set and the accuracy of the data analysis.

Prior Presentation. Data included herein were presented before the FDA’s Endocrinologic and Metabolic Drugs Advisory Committee on 17 January 2019. An abstract related to this study was presented at the 35th International Conference on Pharmacoepidemiology and Therapeutic Risk Management, Philadelphia, PA, 24–28 August 2019.