Diabetes Care 1

Sodium–Glucose Cotransporter Oriana Hoi Yun Yu,1,2 Sophie Dell’Aniello,1 Baiju R. Shah,3,4,5 Vanessa C. Brunetti,1,6 2 Inhibitors and the Risk of Below- Jean-Marc Daigle,7 Michael Fralick,3,8 Antonios Douros,1,9,10,11 Nianping Hu,12 Knee Amputation: A Multicenter Silvia Alessi-Severini,13,14 Anat Fisher,15 Shawn C. Bugden,13,16 Paul E. Ronksley,17 Observational Study Kristian B. Filion,1,9 Pierre Ernst,1,9 and Lisa M. Lix,18 for the Canadian Network https://doi.org/10.2337/dc20-0267 for Observational Drug Effect Studies (CNODES) Investigators* PDMOOYHAT EVCSRESEARCH SERVICES EPIDEMIOLOGY/HEALTH

OBJECTIVE Reports of amputations associated with sodium–glucose cotransporter 2 (SGLT2) 1Center for Clinical Epidemiology, Lady Davis inhibitors have been inconsistent. We aimed to compare the risk of below-knee Institute, Jewish General Hospital, Montreal,´ amputation with SGLT2 inhibitors versus dipeptidyl peptidase 4 (DPP-4) inhibitors Quebec,´ Canada among patients with . 2Division of Endocrinology, Department of Med- icine, Jewish General Hospital, Montreal,´ Quebec,´ RESEARCH DESIGN AND METHODS Canada 3Department of Medicine, University of Toronto, This multicenter observational study used administrative health care databases Toronto, Ontario, Canada from seven Canadian provinces and the U.K. Incident SGLT2 inhibitor users were 4Institute for Clinical Evaluative Sciences (ICES), matched to DPP-4 inhibitor users using a prevalent new-user design and time- Toronto, Ontario, Canada 5Sunnybrook Health Sciences Centre, Toronto, conditional propensity scores. Cox proportional hazards models were used to Ontario, Canada estimate site-specific adjusted hazard ratios (HR) and corresponding 95% CIs of 6Department of Epidemiology, Biostatistics and incident below-knee amputation for SGLT2 inhibitor versus DPP-4 inhibitor users. Occupational Health, McGill University, Mon- Random effects meta-analyses were used to pool the site-specific results. treal,´ Quebec,´ Canada 7Institut national d’excellence en santeeten´ ´ ´ RESULTS services sociaux (INESSS), Quebec City, Quebec, Canada The study cohort included 207,817 incident SGLT2 inhibitor users matched to 8Sinai Health System, Toronto, Ontario, Canada 207,817 DPP-4 inhibitor users. During a mean exposed follow-up time of 11 months, 9Departments of Medicine and of Epidemiology, the amputation rate was 1.3 per 1,000 person-years among SGLT2 inhibitor users Biostatistics, and Occupational Health, McGill University, Montreal,´ Quebec,´ Canada and 1.5 per 1,000 person-years among DPP-4 inhibitor users. The adjusted HR of 10Institute of Clinical Pharmacology and Toxicol- below-knee amputations associated with SGLT2 inhibitor use compared with DPP-4 ogy, Charite-Universit´ atsmedizin¨ Berlin, corporate inhibitor use was 0.88 (95% CI 0.71–1.09). Similar results were obtained in stratified member of Freie Universitat¨ Berlin, Humboldt- ¨ analyses by specific SGLT2 inhibitor molecule. Universitat zu Berlin, Berlin, Germany 11Berlin Institute of Health, Berlin, Germany 12The Health Quality Council, Saskatoon, Sas- CONCLUSIONS katchewan, Canada In this large multicenter observational study, there was no association between 13College of Pharmacy, Rady Faculty of Health SGLT2 inhibitor use and incident below-knee amputations among patients with Sciences, University of Manitoba, Winnipeg, Man- type 2 diabetes compared with DPP-4 inhibitor use. While these findings provide itoba, Canada 14Manitoba Centre for Health Policy, Rady Fac- some reassurance, studies with a longer duration of follow-up are needed to assess ulty of Health Sciences, University of Manitoba, potential long-term effects. Winnipeg, Manitoba, Canada 15Department of Anesthesiology, Pharmacology and Therapeutics, University of British Columbia, Sodium–glucosecotransporter 2(SGLT2) inhibitors are the newest antidiabetic agents Vancouver, Canada for type 2 diabetes management (1). They inhibit the SGLTs on renal proximal tubules, 16School of Pharmacy, Memorial University of leading to glucosuria. This effect not only lowers glycemia but also induces weight loss Newfoundland, St John’s, Newfoundland and Lab- rador, Canada and blood pressure reduction (2). Indeed, randomized placebo-controlled trials have 17Department of Community Health Sciences, shown that SGLT2 inhibitors also decrease the risk of cardiovascular outcomes (3,4) Cumming School of Medicine, University of Cal- and heart failure (3–5). Current guidelines from the American Diabetes Association gary, Calgary, Alberta, Canada Diabetes Care Publish Ahead of Print, published online August 5, 2020 2 SGLT-2 Inhibitors and Risk of Amputation Diabetes Care

recommend the use of SGLT2 inhibitors between SGLT2 inhibitor use and the risk of present. Hospital Episodes Statistics link- as second- or third-line treatment in ad- amputation, with most of the studies using age for this study was from 1 April 1997 dition to in the management of thesameemploymentinsurancedatabase to 31 December 2017. type 2 diabetes (1). Owing to the cardio- in the U.S.; however, their findings have The study protocol was registered at vascular and renal benefits conferred by been inconsistent (7,12–18). ClinicalTrials.gov (NCT04017221). Ethics these agents, SGLT2 inhibitors are recom- In light of the findings from the CAN- approval was obtained at each partici- mended as one of the preferred second- VAS Program trial and the inconsistent pating site. The study protocol received line agents for patients who have high risk results of previous observational studies, ethical and scientific approval from the factors or known cardiovascular disease, further studies are needed to address Independent Advisory Scientific Commit- heart failure, and chronic renal disease whether SGLT2inhibitor useis associated tee of the CPRD (protocol number: (i.e., estimated glomerular filtration rate withan increased riskof amputation.This 19_007A2). [eGFR] of 30–60 mL/min/1.73 m2 or urine study used data from Canada and the Individuals aged $18 years with type 2 albumin-to-creatinine ratio of 30 mg/g). U.K. with the aim of determining whether diabetes were identified by prescrip- Given the cardiovascular benefits of SGLT2 inhibitor use compared with DPP-4 tions for an antidiabetic medication (a- SGLT2 inhibitors and the emphasis on inhibitor use, is associated with an in- glucosidase inhibitors, DPP-4 inhibitors, these benefits in recent treatment guide- creased risk of below-knee amputation glucagon-like peptide [GLP]-1 receptor lines, the use of SGLT2 inhibitors has among patients with type 2 diabetes in a agonists, , , metfor- increased substantially among patients real-world setting. min, SGLT2 inhibitors, , thia- with type 2 diabetes (6). zolidinediones, or combinations of these Despite the cardiovascular benefits as- RESEARCH DESIGN AND METHODS drugs) between 1 January 2006 and 30 June sociated with SGLT2 inhibitor treatment, 2018 or the most recent date of data Source Population availability at each site. Potentially eligi- there are several safety concerns asso- We used administrative health care data- ble subjects were identified as of 2006 to ciated with their use, including a reported bases from the Canadian provinces of increased risk of below-knee amputations Alberta,BritishColumbia,Manitoba,Nova cover the period of general availability of (3,7). Concerns regarding this adverse event Scotia, Ontario, Quebec,´ and Saskatch- DPP-4 inhibitors and SGLT2 inhibitors. In fl stem from the Canagli ozin Cardiovas- ewan, and the U.K. Clinical Practice Re- Nova Scotia, due to limitations in pre- cular Assessment Study (CANVAS) Program search Datalink (CPRD). The Canadian scription drug data availability, individuals trial, in which participants randomized to databases contain population-level data who received antidiabetic medications fl the SGLT2 inhibitor canagli ozin had a on physician claims, hospitalization re- between 1 November 2017 and 30 June twofold increased risk compared with cords, and prescription drug claims. Pre- 2018 were eligible to be included in the participants randomized to placebo (haz- scription drug data are only available for study cohort. ard ratio [HR] 1.97; 95% CI 1.41–2.75) (3). individuals aged $18 years in Alberta, From this source population, we con- As a result of this finding, the U.S. Food aged $65 years in Ontario, and those structed the study cohort of users of and Drug Administration issued a black- $65 years, receiving social assistance SGLT2 inhibitors and users of DPP-4 box warning of amputation risk for can- and without access to private drug in- inhibitors with a prescription after the agliflozin (8). However, there was no surance in Quebec.´ date of introduction of SGLT2 inhib- increased risk of amputation associated The CPRD is a large primary care da- itors at each study site (Supplementary with canagliflozin use among patients tabase containing medical information Table 1). Identifiable information was not with type 2 diabetes and chronic renal documented by primary care physicians accessible, and we complied with all disease compared with patients treated on ;13 million patients enrolled in .680 privacy requirements of the data custo- with placebo in the Canagliflozin and general practices in the U.K. (19,20). This dians at each site. A prevalent new-user Renal Events in Diabetes with Established database documents demographic char- design (22) was used to match each user Nephropathy Clinical Evaluation (CRE- acteristics, diagnoses, laboratory test re- of an SGLT2 inhibitor to a user of a DPP-4 DENCE) trial, which assessed canagliflo- sults, procedures, prescriptions, medical inhibitor. Study cohort entry among zin use and renal outcomes (HR 1.11; history, administrative information, and SGLT2 inhibitor users was defined by 95% CI 0.79–1.56) (9). Nevertheless, two clinical data, including smoking, BMI, and thedateofthefirst SGLT2 dispensing pharmacovigilance analyses found an in- alcoholuse.The CPRD is regularly audited, (or prescription in CPRD). Study cohort creased reporting of amputation among and the data have been shown to be valid entry for DPP-4 inhibitor users was the individuals treated with SGLT2 inhibitors and of high quality (19,21). CPRD data date of the matched (see below) DPP-4 compared with other antidiabetic agents were linked to the Hospital Episodes inhibitor prescription during the pe- (10,11). Several observational studies have Statistics database, which contains full riod defined by the first prescription been conducted to assess the association hospitalization data from 1997 to the of SGLT2 inhibitors and 30 June 2018.

18Department of Community Health Sciences, Uni- Clinical trial reg. no. NCT04017221, clinicaltrials © 2020 by the American Diabetes Association. versity of Manitoba, Winnipeg, Manitoba, Canada .gov Readers may use this article as long as the work is Corresponding author: Lisa M. Lix, lisa.lix@ This article contains supplementary material properly cited, the use is educational and not for fi umanitoba.ca online at https://doi.org/10.2337/figshare.12652454. pro t, and the work is not altered. More infor- mation is available at https://www.diabetesjournals Received 6 February 2020 and accepted 6 July *A list of the CNODES investigators is provided in .org/content/license. 2020 the supplementary material online. care.diabetesjournals.org Yu and Associates 3

Individuals with a prior history of ampu- with the closest TCPS and in chrono- or vildagliptin) alone or in combination tation at any time before or on study logical order. However, in five sites, with non-SGLT2 inhibitor antidiabetic cohort entry were excluded. there was a loss of .10% of exposure agents. Exposure was defined using an sets after trimming the areas of non- as-treated approach whereby exposure Matching overlap of the TCPS distribution and was time-fixed and defined by the cohort We created exposure sets that were de- matching. In these sites, matching with entry drug. Patients were monitored fined by user type (incident vs. prevalent), replacement was performed using a cal- until they discontinued treatment, de- level of antidiabetic treatment, prior use iper width of 60.2 SDs of log TCPS. finedasagapof$30 days after the end of GLP-1 agonists, and calendar time (DPP- of a prescription (grace period) or the 4 inhibitor prescription within 120 days Covariates initiation of an SGLT2 inhibitor for users of the SGLT2 inhibitor initiation). Incident The following covariates, defined a priori, of DPP-4 inhibitors, or censored due to users were defined as using SGLT2 in- were used to construct the TCPS in all death, end of health care coverage, or hibitororDPP-4inhibitorforthefirst time study cohorts: age, sex, calendar year at end of study period, whichever occurred (i.e., new users). Incident SGLT2 inhibitor cohort entry, and diabetes duration (,1 first. DPP-4 inhibitor use was the com- users were matched to incident DPP-4 year, 1–4.9 years, 5–10 years, and .10 parator because this class of antidiabetic inhibitor users. Patients treated with DPP- years). We included comorbidities iden- agent is also a second- or third-line treat- 4 inhibitors who switched to or added an tified during the 3 years before study ment and has not been shown to be SGLT2 inhibitor to their treatment regi- cohort entry such as alcohol-related dis- associated with increased amputation men (prevalent new users) were matched orders, cancer, cerebrovascular disease, risk (23). to patients treated with DPP-4 inhibitors cirrhosis, coronary artery disease, dia- for the same duration but who remained betic nephropathy, diabetic neuropathy, Study Outcomes on DPP-4 inhibitor treatment (Supplemen- diabetic retinopathy, dialysis, hyperten- The primary study outcome was incident tary Fig. 1). We also matched, on the level sion, ischemic stroke, myocardial infarc- below-knee amputation, defined as hav- of antidiabetic treatment, a three-level tion,otherkidneydiseases,andperipheral ing transtibial amputations or amputa- categorical variable created as a proxy for arterial disease. We also included medica- tions involving the ankle and foot using severity of diabetes. Three levels of an- tion use in the year before study cohort procedure codes documented during tidiabetic treatment were created to entry, such as acetylsalicylic acid, aldo- hospitalization or physician claims data mirror the severity of type 2 diabetes sterone antagonists, a-glucosidase inhib- (24). Below-knee amputation was as- based on the type and number of dif- itors, angiotensin II receptor blockers, sessed, given that 71% of amputations ferent antidiabetic agents in the prior ACE inhibitors, b-blockers, calcium chan- that occurred in the CANVAS Program 365 days. The first level was defined as nel blockers, digitalis-like agents, direct trial were at the toe and metatarsal level patients treated with only one antidia- renin inhibitors, GLP-1 receptor agonists, (3,25). Furthermore, given that this study betic agent or treated with lifestyle mod- insulin, loop diuretics, meglitinides, met- excluded individuals with a prior history ifications (i.e., they did not receive an formin, nonacetylsalicylic acid antiplate- of amputation, it would be clinically antidiabetic agent during the prior 365 days). let drugs, nonsteroidal anti-inflammatory unlikely for individuals to develop in- The second level included patients who drugs, oral anticoagulants, other diuretics, cident above-knee amputation associ- required at least two noninsulin antidi- other lipid-lowering therapy, statins, ated with SGLT2 inhibitor use. The diagnostic abetic agents. Finally, the third level in- sulfonylureas, thiazide diuretics, and and procedure codes used to define this cluded patients who received insulin . Finally, we included outcome are reported in Supplementary treatment (alone or in combination with covariates that are indicators of health Table 2. other antidiabetic agents). Time condi- care use in the year before study cohort tional propensity scores (TCPS) were entry, including the number of inpatient then constructed separately for inci- hospitalizations (0, 1–2, and $3) and Statistical Analyses dent and prevalent new users by using number of physician visits (0–2, 3–5, Patient characteristics were summarized conditionallogisticregression stratified and $6). in each study cohort using frequencies by exposure set to estimate the pro- In the CPRD study cohort, additional a and percentages for categorical variables pensity of receiving a SGLT2 inhibitor priori defined covariates were included and means 6 SD for continuous varia- versus a DPP-4 inhibitor using the co- in the TCPS models. These covariates in- bles. Multivariable Cox proportional haz- variates shown in the COVARIATES section. cluded BMI, smoking status (never, ever, ards models were used to estimate the Scores were computed for each indi- and unknown), race, blood pressure, es- site-specific adjusted HR and the corre- vidual in each exposure set; hence, an timated glomerular filtration rate, and sponding 95% CI for the risk of incident individual may have different scores for glycated hemoglobin A1c. below-knee amputation among SGLT2 exposure sets they enter, depending inhibitor users versus DPP-4 inhibitor on the time of entry (i.e., time condi- Exposure Assessment users.Coxmodelswereadjustedforage, tional). Additional covariates were in- Patients were classified as being current sex, diabetes duration, and deciles of cluded in the TCPS in the CPRD cohort users of SGLT2 inhibitors (canagliflozin, TCPS. In secondary analyses, we strat- (please see COVARIATES section). SGLT2 dapagliflozin, or empagliflozin) alone or ified by age to determine whether the inhibitor users were matched 1:1 with- combined with other antidiabetic agents, risk of amputation associated with SGLT2 out replacement to users of DPP-4 or current users of DPP-4 inhibitors (alog- inhibitor use was higher among the el- inhibitors from their exposure set liptin, , , , derly population ($70 and ,70 years), 4 SGLT-2 Inhibitors and Risk of Amputation Diabetes Care

sex, prior insulin use (defined as insulin # 19_007A2) of the CPRD, and the ap- Statistics Deaths, V2, BC Ministry of use in the previous year), and SGLT2 proved protocol was made available to Health (publisher). Parts of this material inhibitor molecule. journal reviewers. The British Columbia are based on data and information com- Finally, we completed five sensitivity (BC) Ministry of Health approved access piled and provided by the Ontario Ministry analyses. First, we repeated the primary to and use of BC data for this study. of Health and Long-Term Care. This study analysis using grace periods of 0, 60, and Data sources were as follows (https:// was supported by Institute for Clinical 365 days to define continuous use of the www.popdata.bc.ca/data): BC Minis- Evaluative Sciences, which is funded by an study drug. Second, we stratified the try of Health (creator) (2018): Medical annual grant from the Ministry of Health primary analysis by incident and preva- Services Plan (MSP) Payment Informa- and Long-Term Care. Parts of this material lent new user status among SGLT2 in- tion File, BC Ministry ofHealth(publisher). are based on data and/or information hibitor users. Third, we defined exposure Ministry of Health (2018), BC Ministry of compiled and provided by the Canadian using an intention-to-treat approach in Health (creator) (2018): Consolidation Institute for Health information. which exposure was defined at cohort File (MSP Registration & Premium Bill- entry and patients were monitored until ing), BC Ministry of Health (publisher). RESULTS occurrence of the outcome or censored Ministry of Health (2018), BC Ministry of The study cohort included 207,817 SGLT2 due to death, end of health care cover- Health (creator) (2018): PharmaNet, BC inhibitor users matched to 207,817 DPP- age, end of study period, entry into the Ministry of Health (publisher). Data Stew- 4 inhibitor users (Fig. 1). Among the users SGLT2 inhibitor cohort for DPP-4 inhib- ardship Committee (2018) and Canadian of SGLT2 inhibitors, 102,263 were clas- itor users, or a maximum of 1 year of Institute for Health Information (creator) sified as incident new users and 105,554 follow-up, whichever occurred first. (2018): Discharge Abstract Database as prevalent new users. Baseline char- Fourth, given that SGLT2 inhibitor use is (Hospital Separations), BC Ministry of acteristics were well balanced after TCPS not indicated for patients on dialysis, we Health (publisher). Ministry of Health matching (Table 1 and Supplementary repeated the primary analysis after ex- (2018). BC Ministry of Health (publisher). Table 4). In the CPRD, the number of cluding individuals with a prior history of Ministry of Health (2018), BC Vital Sta- patients with renal insufficiency, defined dialysis using the study cohort from tistics Agency (creator) (2018): Vital as having an eGFR ,60 mL/min/1.73 m2 Ontario because this cohort contained the highest number of individuals with a prior history of dialysis. Kaplan-Meier curves for each site were visually eval- uated to assess potential departures from the assumption of proportional hazards. This assessment revealed no indication that the assumption was vi- olated at any of the sites. Finally, we repeated the primary analysis using a fixed-effects model.

Meta-analysis We pooled the adjusted HRs from each site using DerSimonian and Laird random- effects meta-analysis with inverse vari- ance weighting (26). Inclusion in the meta-analysis was restricted to sites with at least five events in each exposure group (Supplementary Table 3). Between- site heterogeneity was assessed using the I2 statistic. Analyses were conducted using RevMan version 5.3 software.

Data Source This study was made possible through data sharing agreements between the Canadian Network for Observational Drug Effect Studies member research centers and the respective provincial governments of Alberta, British Columbia, Manitoba Figure1—Flowchartdescribing construction of thestudy cohort. Notes: Numbers maynot add (HIPC # 2018/2019-58), Nova Scotia, On- up because small cells were suppressed due to privacy restrictions. Patients ,19 years in , ´ Alberta and 66 years in Ontario were included to ensure that patients had at least 1 year of tario, Quebec, and Saskatchewan. This data before cohort entry. Patients were eligible to enter the study cohort a maximum of two study was approved by the Independent times, a first time with a DPP-4 inhibitor prescription and a second time with an SGLT2 Scientific Advisory Committee (protocol inhibitor prescription. care.diabetesjournals.org Yu and Associates 5

Table 1—Baseline characteristics of users of SGLT2 inhibitors and their matched DPP-4 users* SGLT2 inhibitors (n 5 207,817) DPP-4 inhibitors (n 5 207,817) Age (years) 63.8 6 9.5 64.0 6 9.6 18–35 3,479 (1.7) 3,612 (1.7) 36–45 12,288 (5.9) 11,915 (5.7) 46–55 30,845 (14.8) 30,105 (14.5) 56–65 47,711 (23.0) 48,079 (23.1) 66–75 89,410 (43.0) 87,993 (42.3) 76–85 22,074 (10.6) 23,920 (11.5) .85 2,010 (1.0) 2,193 (1.1) Female sex 86,360 (41.6) 87,030 (41.9) Calendar year at cohort entry 2013 320 (0.2) 343 (0.2) 2014 6,954 (3.3) 7,322 (3.5) 2015 51,464 (24.8) 50,921 (24.5) 2016 66,242 (31.9) 66,422 (32.0) 2017 61,291 (29.5) 61,013 (29.4) 2018 21,546 (10.4) 21,796 (10.5) New user status Incident users 102,263 (49.2) 102,263 (49.2) Prevalent users 105,554 (50.8) 105,554 (50.8) SGLT2 inhibitor molecule Canagliflozin 87,922 (42.3) – Dapagliflozin 63,792 (30.7) – Empagliflozin 56,103 (27.0) – Diabetes duration (years) 12.6 6 6.6 12.5 6 6.6 ,1 7,166 (3.4) 7,341 (3.5) 1–4.9 25,204 (12.1) 25,766 (12.4) 5–10 52,543 (25.3) 52,758 (25.4) .10 122,904 (59.1) 121,952 (58.7) Comorbidities† Alcohol-related disorders 3,626 (1.7) 3,658 (1.8) Cancer 21,692 (10.4) 21,937 (10.6) Cerebrovascular disease 9,892 (4.8) 10,156 (4.9) Cirrhosis 3,621 (1.7) 3,606 (1.7) Coronary artery disease 44,710 (21.5) 43,939 (21.1) Diabetic nephropathy 7,476 (3.6) 7,478 (3.6) Diabetic neuropathy 3,807 (1.8) 3,844 (1.8) Diabetic retinopathy 5,266 (2.5) 5,296 (2.5) Dialysis 277 (0.1) 315 (0.2) Hypertension 111,130 (53.5) 111,332 (53.6) Ischemic stroke 2,448 (1.2) 2,535 (1.2) Myocardial infarction 5,326 (2.6) 5,113 (2.5) Other kidney diseases 10,222 (4.9) 10,850 (5.2) Peripheral arterial disease 4,472 (2.2) 4,471 (2.2) Use of medications† Acetylsalicylic acid 36,875 (17.7) 36,792 (17.7) Aldosterone antagonists 6,146 (3.0) 6,182 (3.0) a-Glucosidase inhibitors 3,057 (1.5) 2,949 (1.4) Angiotensin II receptor blockers 66,747 (32.1) 66,301 (31.9) ACE inhibitors 94,489 (45.5) 94,092 (45.3) b-Blockers 58,854 (28.3) 58,371 (28.1) Calcium channel blockers 63,281 (30.5) 63,671 (30.6) Digitalis-like agents 2,586 (1.2) 2,624 (1.3) Direct renin inhibitors 104 (0.1) 92 (0.0) GLP-1 receptor agonists 8,464 (4.1) 8,464 (4.1) Insulin 57,143 (27.5) 57,143 (27.5) Loop diuretics 21,314 (10.3) 21,559 (10.4) Meglitinides 4,680 (2.3) 4,707 (2.3) Metformin 180,662 (86.9) 180,828 (87.0) Nonacetylsalicylic acid antiplatelet drugs 14,034 (6.8) 13,655 (6.6) Nonsteroidal anti-inflammatory drugs 40,470 (19.5) 40,263 (19.4) Oral anticoagulants 13,393 (6.4) 13,359 (6.4) Other diuretics 18,497 (8.9) 18,406 (8.9)

Continued on p. 6 6 SGLT-2 Inhibitors and Risk of Amputation Diabetes Care

Table 1—Continued SGLT2 inhibitors (n 5 207,817) DPP-4 inhibitors (n 5 207,817)

Other lipid-lowering therapy 23,524 (11.3) 22,937 (11.0) Statins 159,742 (76.9) 159,061 (76.5) Sulfonylureas 108,451 (52.2) 108,327 (52.1) Thiazide diuretics 45,019 (21.7) 44,788 (21.6) Thiazolidinediones 5,175 (2.5) 4,863 (2.3) Different classes of non-antidiabetic drugs, n‡ 0–1 8,465 (4.1) 8,666 (4.2) 2–5 65,919 (31.7) 66,729 (32.1) $6 133,433 (64.2) 132,422 (63.7) Health care use† Inpatient hospitalizations, n 0 176,833 (85.1) 176,723 (85.0) 1–2 28,687 (13.8) 28,697 (13.8) $3 2,296 (1.1) 2,398 (1.2) Physician visits, n 0–2 14,963 (7.2) 15,117 (7.3) 3–5 31,883 (15.3) 32,206 (15.5) $6 160,971 (77.5) 160,494 (77.2) Data are presented as n (%) or mean 6 SD. *Patients on SGLT2 inhibitors were matched to patients on DPP-4 inhibitors from their exposure set (defined on level of antidiabetic therapy, prior use of GLP-1 receptor agonists, time on DPP-4 inhibitors for prevalent new users, and calendar time) on TCPS. Because of privacy restrictions, values of ,6 were replaced by 3 before pooling. †Comorbidities were assessed in the 3 years before study cohort entry, andmedicationsandhealthcareusewereassessedintheyearbeforestudycohortentry.‡InSaskatchewan,thenumberofnon-antidiabeticdrugclasses was defined using the list of medication covariates rather than Anatomical Therapeutic Chemical-defined classes due to unavailability of Anatomical Therapeutic Chemical codes. In Quebec´ and the CPRD, drug classification was performed using the American Hospital Formulary Service and the British National Formulary, respectively.

was higher among DPP-4 inhibitor users In secondary analyses, the risk of in- CONCLUSIONS compared with SGLT2 inhibitor users cident below-knee amputation associated In this large, multicenter observational (Supplementary Table 4). with SGLT2 inhibitor use versus DPP-4 study, using administrative data from The mean exposed follow-up time for inhibitor use did not differ according to seven Canadian provinces and the U.K. the matched cohort was 11 6 9 months, age ($70 vs. ,70 years), sex, or history of CPRD, we found no increased risk of generating 369,458 person-years of ob- prior insulin use or SGLT2 inhibitor mol- incident below-knee amputation asso- servation.Therateofincidentbelow-knee ecule (Table 3). The results remained ciated with SGLT2 inhibitor versus DPP- amputation was similar among SGLT2 in- consistent across most sensitivity anal- 4 inhibitor use among patients with type hibitor users (1.3 per 1,000 person-years) yses, including among patients with no 2 diabetes (HR 0.88; 95% CI 0.71–1.09). versus DPP-4 inhibitor users (1.5 per 1,000 prior history of dialysis (Supplementary Results were consistent across subgroups person-years) in the matched cohort. Table 5). However, there was a trend defined by age, sex, and prior insulin use. There was no significant increased risk toward an increased risk of incident Similarly, there was no increased risk of of incident below-knee amputation asso- below-knee amputation among preva- below-knee amputation associated with ciated with SGLT2 inhibitor use compared lent new users of SGLT2 inhibitors ver- individual SGLT2 inhibitor use, including with DPP-4 inhibitor use (HR 0.88; 95% CI sus DPP-4 inhibitor users (HR 1.29; 95% canagliflozin, compared with DPP-4 in- 0.71–1.09; I2 5 18%) (Table 2 and Fig. 2). CI 0.97–1.70) (Table 3). hibitor use.

Table 2—Crude and adjusted HR for the association between the use of SGLT2 inhibitors and the risk of below-knee amputation among patients with type 2 diabetes Crude incidence Adjusted models‡ Treatment Patients Events Mean follow-up Person- rate (per 1,000 Crude HR group (n) (n) time (years) years person-years)* (95% CI)† HR (95% CI) I2 (%) SGLT2 207,817 253 0.90 187,641 1.3 0.87 0.88 18 inhibitors (0.69–1.10) (0.71–1.09) DPP-4 inhibitors 207,817 281 0.88 181,817 1.5 Reference Reference *Incidence rate was calculated using all study cohorts. †Patients on SGLT2 inhibitors were matched to patients on DPP-4 inhibitors from their exposure set (defined on the level of antidiabetic therapy, time on DPP-4 inhibitors [for prevalent new users only], prior use of GLP-1 receptor agonists, and within 120 days of the SGLT2 inhibitor prescription) on TCPS. HR estimation was restricted to sites with at least five events in each exposure group. ‡Outcome models were adjusted for age (continuous), sex, diabetes duration (continuous), and deciles of TCPS. care.diabetesjournals.org Yu and Associates 7

Figure 2—HR (95% CI) of below-knee amputation associated with SGLT2 inhibitors use compared with DPP-4 inhibitors use among patients with type 2 diabetes.* *Nova Scotia was not included in this analysis because where were fewer than five events in one of the two exposure groups. IV, inverse variance.

Concerns regarding an increased risk hoc analyses must be interpreted with Program trial was supported by pharma- of amputation associated with the use of caution. In the Dapagliflozin Effect on covigilance analyses conducted using the SGLT2 inhibitors arose when the CANVAS Cardiovascular Events–Thrombolysis in U.S. Food and Drug Administration Ad- Program trial found a nearly twofold Myocardial Infarction 58 trial (DECLARE- verse Event Reporting System, which increased risk associated with canagli- TIMI 58) trial, there was no increased found an increased reporting of ampu- flozin use versus placebo (HR 1.97; 95% CI risk of amputation associated with ran- tation specifically among canagliflozin 1.41–2.75) (3). In light of this finding, domization to dapagliflozin versus pla- users (proportional reporting ratio [PRR] further analyses were conducted in the cebo (HR 1.09; 95% CI 0.84–1.40) (5). 5.33; 95% CI 4.04–7.04) compared with BI 10773 (Empagliflozin) Cardiovascular Recently, the CREDENCE trial, which en- users of non-SGLT2 inhibitor antidiabetic Outcome Event Trial in Type 2 Diabetes rolled patients with type 2 diabetes agents (10). Subsequently, another phar- Mellitus Patients (EMPA-REG OUTCOME) and chronic renal disease to study renal macovigilance analysis conducted using trial to assess amputation risk (27). These outcomes associated with canagliflozin the World Health Organization global analyses revealed no increased risk of use, did not find an increased risk of database of individual case safety re- amputation with empagliflozin (HR 1.00; amputation among patients treated with ports (Vigibase) found that the PRR was 95% CI 0.70–1.44) (27). However, as ac- canagliflozin (HR 1.11; 95% CI 0.79–1.56) increased for all available SGLT2 inhib- knowledged by the authors, the ascer- (9). itors compared with other antidiabetic tainment of amputations may have been Thesafetysignalregardinganincreased medications (canagliflozin: PRR 7.09; inaccurate, and the results of such post risk of amputation identified in the CANVAS 95% CI 5.25–9.57; empagliflozin: PRR

Table 3—Summary of results of stratified and sensitivity analyses of pooled adjusted HR (95% CI) for below-knee amputation for SGLT2 inhibitor use versus DPP-4 inhibitor use among patients with type 2 diabetes Number of sites included Adjusted HR (95% CI)* I2 (%) Main analysis 7 0.88 (0.71–1.09) 18 Age $70 years 3 1.13 (0.81–1.56) 0 ,70 years 6 0.80 (0.58–1.12) 46 Sex Females 3 1.04 (0.64–1.70) 17 Males 6 0.87 (0.65–1.15) 38 Prior insulin use† Yes 6 0.69 (0.46–1.03) 51 No 4 1.14 (0.87–1.50) 0 SGLT2 inhibitor molecule Canagliflozin 5 0.98 (0.77–1.25) 0 Dapagliflozin 4 0.69 (0.45–1.06) 0 Empagliflozin 3 1.08 (0.70–1.68) 0 Varying grace period 0 day 4 1.12 (0.74–1.69) 0 60 days 7 0.90 (0.73–1.12) 28 365 days 7 0.90 (0.72–1.11) 46 New user status Incident users 6 0.68 (0.53–0.88) 0 Prevalent users 3 1.29 (0.97–1.70) 0 Intention-to-treat approach‡ 6 0.81 (0.63–1.05) 35 No prior history of dialysis 1 1.13 (0.81–1.57) NA Fixed-effects model analysis 7 0.90 (0.75–1.08) 18 Inclusion in each meta-analysis was restricted to sites with at least five events in each exposure group. NA, not applicable. *Outcome models were adjusted for age (continuous), sex, diabetes duration (continuous), and deciles of TCPS. †Prior insulin use was defined as a prescription for insulin in the year prior. ‡In the intention-to-treat approach, maximum follow-up was 1 year. 8 SGLT-2 Inhibitors and Risk of Amputation Diabetes Care

4.96; 95% CI 2.89–8.50; and for dapagli- for their study population (i.e., excluding no prior history of dialysis (i.e., during the flozin: PRR for toe-amputations 2.62; patients with prior history of amputation, 3 years before study cohort entry). 95% CI 1.33–5.14) (11). Nevertheless, insulin use, renal insufficiency, and base- Third, the databases used in this study there are well-recognized limitations line cardiovascular disease) and when capture dispensing of medications (i.e., to using adverse event reporting data, different comparators were used in the Canadian databases) or prescriptions which include difficulties with adverse analyses (7). (i.e., CPRD) without any guarantee that event recognition, underreporting of ad- Our study has several strengths, in- these medications were taken by the verse events, absence of a denominator, cluding the use of DPP-4 inhibitors as patient. biases that affect event reporting, and comparators because they are prescribed Fourth, we did not study the specific variations in report quality (28). at a similar stage of type 2 diabetes as site of below-knee amputation because To date, there have been eight obser- SGLT2 inhibitors (i.e., as second- or third- we anticipated insufficient events per vational studies, to our knowledge, that line treatment). The use of the prevalent site. have assessed the risk of amputation new-user design allowed inclusion of Fifth, we included users of SGLT2 in- associated with SGLT2 inhibitor use; patients who switched to or added an hibitors who had (prevalent new users) these studies have produced heteroge- SGLT2 inhibitor, which is reflective of and had not (incident new users) used neous results (7,12–17). Six of these clinical practice and allows our findings DPP4 inhibitors previously, which may studies used the MarketScan Commer- to be more generalizable to the realities consist of individuals that differ in the cial Claims and Encounters Database, of clinical practice. Indeed, we noted severity of diabetes and risk for compli- fi with ve studies showing no association ;50% of SGLT2 inhibitor users had pre- cations associated with diabetes. In strat- fi between SGLT2 inhibitor use and am- viously used DPP-4 inhibitors. Finally, i ed analyses, the risk of below-knee – putation (12 15) and one study showing with data from eight databases across amputation associated with SGLT2 in- an increased risk of amputation with two countries, our study is the largest hibitor use compared with DPP-4 inhib- SGLT2 inhibitor use compared with DPP- observational study conducted examin- itor use was higher among prevalent new – 4 inhibitor use (HR 1.69; 95% CI 1.20 ing this safety issue to date, increasing users of SGLT2 inhibitors, but this did not fi 2.38) (7). Interestingly, in that study, the precision and generalizability of our reach statistical signi cance. SGLT2 inhibitor use was not associated results. Finally, the duration of follow-up was with an increased risk of lower-extremity Ourstudyalsohas potentiallimitations. modest, and future studies with a longer amputation compared with First, residual confounding is possible duration of follow-up are needed to use (7). Udell et al. (16) performed an determine whether long-term use of given that this is an observational study. observational study using the U.S. De- SGLT2 inhibitors is associated with below- However, we used various approaches to partment of Defense Health System and knee amputation risk, because the in- minimize confounding by using an active found that patients initiating an SGLT2 creased risk of amputation associated comparator and extensive matching. inhibitor had a nearly twofold increased with canagliflozin treatment in the CAN- Second, there is potential confounding risk of lower-extremity amputation com- VAS Program trial was observed near the by contraindication as physicians may pared with patients treated with non- end of the study period (mean follow-up have been less likely to prescribe SGLT2 SGLT2 inhibitor antidiabetic agents (HR of ;3.6 years) (3). Furthermore, our inhibitors to patients who were at higher 1.99;95%CI1.12–3.51). However,most study focused on patients with no prior risk for amputation. Furthermore, there is of the amputations in the SGLT2 inhibitor history of amputation, whereas the CAN- an imbalance in the proportion of SGLT2 group were among patients treated with VAS Program trial and the CREDENCE trial canagliflozin. This finding was also ob- inhibitor versus DPP-4 inhibitor users with involved patients with a prior history of renal insufficiency (eGFR ,60 mL/min/ served in the study performed by Ueda 2 amputation. Thus, the patients in these et al. (17), using the Swedish/Danish 1.73 m ) noted in the CPRD cohort, be- trials have an underlying higher risk of National Register, which found that cause use of SGLT2 inhibitors is not rec- amputation compared with the popula- fi SGLT2 inhibitor users had an increased ommended for patients with signi cant tion cohort of our study (the rate of fi , risk of incident amputation compared renal insuf ciency (i.e., eGFR 45 mL/ amputation in our study was 1.3 per 2 with GLP-1 receptor agonist users (HR min/1.73 m ) (1,29). Although the eGFR 1,000 person-years vs. 6.3 per 1,000 2.48; 95% CI 1.14–5.40). The heteroge- findings from the CPRD comprised a small person-years in the CANVAS Program neity of the findings of these previous percentage of the total cohort (3.3% of the trial and 12.3 per 1,000 person-years in studies may be due to a number of weight of the meta-analysis), there may the CREDENCE trial) (3,9). factors, including differences in the be residual confounding because patients In conclusion, in this multicenter ob- populations assessed, methodologies with renal insufficiency have a higher risk servational study, we found no evidence used, comparator drug used, differen- of amputation compared with patients of an association between SGLT2 inhib- ces in the extent of amputation (i.e., with normal renal function (30). As such, itor use and incident below-knee ampu- some studies included above-knee am- DPP-4 inhibitor users may have an un- tation compared with DPP-4 inhibitor use putations) (14,18), duration of follow-up derlying higher risk of amputation com- among patients with type 2 diabetes. time, and the inclusion of patients with pared with SGLT2 inhibitor users, and this Similarly, there was no increased risk of prior history of amputation (7,13,17,18). could mask the higher risk of amputation below-knee amputation associated with Some of these inconsistencies were ob- associated with SGLT2 inhibitor use. How- specific SGLT2 inhibitor molecule use served within the same study when the ever, in our sensitivity analysis, our results compared with DPP-4 inhibitor use. Fu- researchers varied the exclusion criteria remained consistent among patients with ture studies will be needed to further care.diabetesjournals.org Yu and Associates 9

address whether SGLT2 inhibitor use 2. Lee PC, Ganguly S, Goh SY. Weight loss as- 2 inhibitors in patients with type 2 diabetes. increases the risk of incident below-knee sociated with sodium-glucose cotransporter-2 Diabetes Obes Metab 2019;21:28–36 amputation over the longer term. inhibition: a review of evidence and underlying 16. Udell JA, Yuan Z, Rush T, Sicignano NM, Galitz mechanisms. Obes Rev 2018;19:1630–1641 M, Rosenthal N. Cardiovascular outcomes and 3. Neal B, Perkovic V, Mahaffey KW, et al.; CAN- risks after initiation of a sodium glucose co- VAS Program Collaborative Group. Canagliflozin transporter 2 inhibitor: results from the EASEL Acknowledgments. The authors thank Audray and cardiovascular and renal events in type 2 population-based cohort study (evidence for St-Jean and Corine Mizrahi (Jewish General diabetes. N Engl J Med 2017;377:644–657 cardiovascular outcomes with sodium glucose Hospital) at the CNODES Coordinating Center 4. Zinman B, Wanner C, Lachin JM, et al.; EMPA- cotransporter 2 inhibitors in the real world). for their important contributions to this work. REG OUTCOME Investigators. Empagliflozin, car- Circulation 2018;137:1450–1459 The authors also acknowledge the programming diovascular outcomes, and mortality in type 17. UedaP,SvanstromH,MelbyeM,etal.Sodium¨ andanalyticalsupportof theanalystsat eachsite: 2 diabetes. N Engl J Med 2015;373:2117– glucose cotransporter 2 inhibitors and risk of Greg Carney and Jason Kim (University of British 2128 serious adverse events: nationwide register Columbia; British Columbia), Zhihai Ma (Univer- 5. Wiviott SD, Raz I, Bonaca MP, et al.; DECLARE– based cohort study. BMJ 2018;363:k4365 sity of Calgary; Alberta), Matthew Dahl (Univer- TIMI 58 Investigators. Dapagliflozin and cardio- 18. Adimadhyam S, Lee TA, Calip GS, Smith sity of Manitoba; Manitoba), Yan Wang and Steve vascular outcomes in type 2 diabetes. N Engl J Marsh DE, Layden BT, Schumock GT. Risk of Doucette (Dalhousie University; Nova Scotia), C. Med 2019;380:347–357 amputations associated with SGLT2 inhibitors Fangyun Wu (Institute for Clinical Evaluative Sci- 6. Dennis JM, Henley WE, McGovern AP, et al.; compared to DPP-4 inhibitors: a propensity- ences; Ontario), Jean-Marc Daigle (Institut na- MASTERMIND Consortium. Time trends in pre- matched cohort study. Diabetes Obes Metab tional d’excellence en sante´ et en services scribing of type 2 diabetes drugs, glycaemic re- 2018;20:2792–2799 sociaux; Quebec), and Hui Yin and Christopher sponse and risk factors: a retrospective analysis of 19. Garc´ıa Rodr´ıguez LA, Perez´ Gutthann S. Use Filliter (Jewish General Hospital; CPRD). The primary care data, 2010-2017. Diabetes Obes of the UK General Practice Research Database authors thank also thank Hala Tamim (York Metab 2019;21:1576–1584 for pharmacoepidemiology. Br J Clin Pharmacol University) for her contributions to this study. 7. Yang JY, Wang T, Pate V, et al. Sodium-glucose 1998;45:419–425 Funding. The Canadian Network for Observa- co-transporter-2 inhibitor use and risk of lower- 20. Herrett E, Gallagher AM, Bhaskaran K, et al. tional Drug Effect Studies, a collaborating center extremityamputation:evolvingquestions,evolv- Data resource profile: Clinical Practice Research of the Drug Safety and Effectiveness Network ing answers. Diabetes Obes Metab 2019;21: Datalink (CPRD). Int J Epidemiol 2015;44:827– (DSEN), is funded by the Canadian Institutes of 1223–1236 836 HealthResearch (grant # DSE-146021). This study 8. U.S. Food and Drug Administration. FDA Drug 21. Jick SS, Kaye JA, Vasilakis-Scaramozza C, et al. was supported by the Institute for Clinical Eval- Safety Communication: FDA confirms increased Validity of the general practice research data- uative Sciences, which is funded by an annual risk of leg and foot amputations with the diabe- base. Pharmacotherapy 2003;23:686–689 grant from the Ministry of Health and Long-Term tes medicine canagliflozin (Invokana, Invokamet, 22. Suissa S, Moodie EE, Dell’Aniello S. Prevalent Care. O.H.Y.Y. receives salary support from the Invokamet XR), 2017. Accessed 6 January 2020. new-user cohort designs for comparative drug Fonds de recherche du Quebec´ –sante´ (FRQS). Available from https://www.fda.gov/Drugs/ effect studies by time-conditional propensity V.C.B. holds a doctoral award from the FRQS. DrugSafety/ucm557507.htm scores. Pharmacoepidemiol Drug Saf 2017;26: K.B.F. receives salary support from the FRQS 9. Perkovic V, Jardine MJ, Neal B, et al.; CRE- 459–468 and William Dawson Scholar award from McGill DENCE Trial Investigators. Canagliflozin and renal 23. Chang CC, Chen YT, Hsu CY, et al. Dipeptidyl University. L.M.L. is supported by a Tier I Canada outcomes in type 2 diabetes and nephropathy. peptidase-4 inhibitors, peripheral arterial disease, Research Chair. N Engl J Med 2019;380:2295–2306 and lower extremity amputation risk in diabetic The opinions, results, and conclusions reported 10. Fadini GP, Avogaro A. SGLT2 inhibitors and patients. Am J Med 2017;130:348–355 in this article are those of the authors. No endorse- amputations in the US FDA Adverse Event Re- 24. Huseynova K, Sutradhar R, Booth GL, Huang ment by the provinces, data stewards, the Institute porting System. Lancet Diabetes Endocrinol 2017; A, Ray JG. Risk of contralateral lower limb am- for Clinical Evaluative Sciences, the Canadian In- 5:680–681 putation and death after initial lower limb am- stitute for Health Information, or the Institut na- 11. Khouri C, Cracowski JL, Roustit M. SGLT-2 putation- a population-based study.Heliyon2018; tional d’excellence en sante´ et en services sociaux inhibitors and the risk of lower-limb amputation: 4:e00836 is intended or should be inferred. is this a class effect? Diabetes Obes Metab 2018; 25. Carbone S, Dixon DL. The CANVAS Program: Duality of Interest. S. SA. has research funding 20:1531–1534 implications of canagliflozin on reducing cardio- from Pfizer and Merck for studies not involving 12. Yuan Z, DeFalco FJ, Ryan PB, et al. Risk of vascular risk in patients with type 2 diabetes SGLT2 inhibitors or DPP-4 inhibitors. No other lower extremity amputations in people with mellitus. Cardiovasc Diabetol 2019;18:64 potential conflicts of interest relevant to this type 2 diabetes mellitus treated with sodium- 26. DerSimonian R, Laird N. Meta-analysis in article were reported. glucose co-transporter-2 inhibitors in the USA: clinical trials. Control Clin Trials 1986;7:177– Author Contributions. O.H.Y.Y. drafted the man- a retrospective cohort study. Diabetes Obes 188 uscript. O.H.Y.Y., S.D’A., B.R.S., V.C.B., JM.D., Metab 2018;20:582–589 27. Inzucchi SE, Iliev H, Pfarr E, Zinman B. Empa- M.F., A.D., N.H., S. SA., A.F., S.C.B., P.E.R., K.B.F., 13. Ryan PB, Buse JB, Schuemie MJ, et al. Com- gliflozin and assessment of lower-limb amputa- P.E., and L.M.L. were involved in the study design, parative effectiveness of canagliflozin, SGLT2 tions in the EMPA-REG OUTCOME trial. Diabetes interpretation of results, and critically reviewed inhibitors and non-SGLT2 inhibitors on the risk Care 2018;41:e4–e5 the manuscript for important intellectual con- of hospitalization for heart failure and ampu- 28. Goldman SA. Limitations and strengths of tent. L.M.L. conducted the meta-analyses. All tation in patients with type 2 diabetes mellitus: spontaneous reports data. Clin Ther 1998;20(Suppl. authors approved the final version of the man- a real-world meta-analysis of 4 observational C):C40–C44 uscript. L.M.L. is the guarantor of this work and, databases (OBSERVE-4D). Diabetes Obes Metab 29. Diabetes Canada Clinical Practice Guidelines as such, had full access to all the data in the study 2018;20:2585–2597 Expert Committee. Diabetes Canada 2018 clinical and takes responsibility for the integrity of the 14. Chang HY, Singh S, Mansour O, Baksh S, practice guidelines for the prevention and man- data and the accuracy of the data analysis. Alexander GC. Association between sodium- agement of diabetes in Canada. Can J Diabetes glucose cotransporter 2 inhibitors and lower ex- 2018;42(Suppl. 1):S1–S325 References tremity amputation among patients with type 2 30. Morbach S, Quante C, Ochs HR, Gaschler F, 1. American Diabetes Association. 9. Pharma- diabetes. JAMA Intern Med 2018;178:1190– Pallast JM, Knevels U. Increased risk of lower- cologic approaches to glycemic treatment: Stand- 1198 extremity amputation among Caucasian diabetic ards of Medical Care in Diabetesd2020. Diabetes 15. Dawwas GK, Smith SM, Park H. Cardiovas- patients on dialysis. Diabetes Care 2001;24: Care 2020;43(Suppl. 1):S98–S110 cular outcomes of sodium glucose cotransporter- 1689–1690