Safety of Sodium-Glucose Cotransporter-2 Inhibitors in Patients with CKD and Type 2 Diabetes: Population-Based US Cohort Study

Introduction

Sodium-glucose cotransporter-2 inhibitors (SGLT2i) are recommended as first-line therapy in patients with type 2 diabetes and CKD who have an eGFR ≥20 ml/min per 1.73 m2.1,2 Although randomized controlled trials have shown the cardiovascular and renoprotective effects of SGLT2i in patients with diabetic kidney disease,3,4 their uptake in routine clinical practice has been slow: Recent studies show that as few as 6% of patients with CKD and type 2 diabetes are currently prescribed SGLT2i in the United States.5,6 This is especially concerning because these patients are at high risk for cardiovascular disease and kidney disease progression.7,8

The slow clinical adoption of SGLT2i may partly be due to concerns about potential adverse effects, including diabetic ketoacidosis (DKA), fractures, amputations, and urogenital infections.9–11 These safety events are especially important as patients with CKD have a higher baseline risk of fractures and lower limb amputations than the non-CKD population, due to disorders in mineral and bone metabolism and high prevalence of risk factors for foot ulceration.12–15

Currently, there is a paucity of safety data on SGLT2i in patients with CKD and type 2 diabetes. Clinical trials are generally underpowered to assess rare but potentially severe side effects.3,4 They also include highly selected patient populations with different characteristics from those who receive SGLT2i in routine care16,17 and apply monitoring protocols to lower the risk of adverse effects which may not be adopted in routine practice. We therefore aimed to comprehensively investigate the safety profile of SGLT2i in routinely cared patients with CKD and type 2 diabetes using three nationwide US databases.

Methods Data Source

We used data from three large US health insurance databases: Optum's deidentified Clinformatics Data Mart Database (CDM), IBM MarketScan, and Medicare Fee-for-Service Parts A, B, and D. CDM and IBM MarketScan include a national commercially insured US population. Medicare is a federal health insurance program providing health care coverage for US residents 65 years or older or younger than 65 years with disabilities. The databases contain deidentified, longitudinal, individual-level data on health care use, inpatient and outpatient diagnoses, diagnostic tests and procedures, outpatient laboratory results (approximately 45% of patients in CDM and 5%–10% in IBM MarketScan), and pharmacy dispensing of drugs. This study was approved by the Mass General Brigham Institutional Review Board, and signed data license agreements were in place for all data sources.

Study Design and Study Population

We conducted an active comparator, new-user cohort study of patients 18 years or older (65 years or older for Medicare) who newly initiated an SGLT2i (i.e., empagliflozin, dapagliflozin, canagliflozin, and ertugliflozin) or a glucagon-like peptide-1 receptor agonist (GLP-1RA) (i.e., liraglutide, dulaglutide, semaglutide, exenatide, albiglutide, and lixisenatide) between April 2013, when the first SGLT2i was released in the United States, and the end of available data (April 2021 in CDM, December 2019 in Medicare Fee-for-Service, and December 2020 in IBM MarketScan) (Supplemental Figure 1). New initiation was defined as a filled prescription for SGLT2i or GLP-1RA, with no previous filled prescriptions of either drug in the previous 365 days. We used GLP-1RA as the active comparator18–20 because GLP-1RA has similarly been shown to reduce cardiovascular events in randomized trials.21 Both drugs had similar clinical indications during the study period (i.e., second-line or third-line glucose-lowering drugs for patients at high cardiovascular risk) and similar temporal trends in use.5,22–24 To be eligible, patients were required to have at least 12 months of continuous health insurance enrollment preceding the cohort entry date as well as diagnoses for CKD and type 2 diabetes. CKD was defined as at least one inpatient or two outpatient diagnosis codes for CKD stages 3–5, and no data on eGFR were used (Supplemental Table 1); the definition was based on a previously validated algorithm that showed sufficient accuracy to identify a population with CKD stages 3–5 (positive predictive value >80%).25 We excluded individuals with a history of type 1 diabetes, secondary or gestational diabetes, kidney failure, nursing home admission, organ transplant, pancreatitis, cirrhosis, acute hepatitis, or multiple endocrine neoplasia type 2 (Supplemental Table 1).

Drug Exposure and Follow-Up

The study exposure was filled prescription for SGLT2i or GLP-1RA. Follow-up began on the day after cohort entry and continued in an “as-treated” approach until the earliest of treatment discontinuation or switch to a drug in the comparator class, outcome occurrence, death, end of continuous health plan enrollment, or end of available data, whichever occurred first. Discontinuation was defined as no prescription refill for the index exposure within 30 days after the termination of the last prescription's supply. We chose an as-treated follow-up approach as primary analysis to address the high rate of treatment discontinuation in routine care,26 which reduces the exposure misclassification that occurs when intention-to-treat analyses are applied in observational studies.20

Study Outcomes

Safety outcomes included DKA, nonvertebral fractures, lower limb amputations, genital infections, hypovolemia, severe hypoglycemia, AKI, and severe urinary tract infections (UTI). We selected these outcomes on the basis of potential safety signals of SGLT2i previously identified in randomized trials or observational studies. The outcomes were identified using validated International Classification of Diseases-9/10-CM procedural and diagnosis codes (Supplemental Table 2). Validation studies for the claims-based algorithms for DKA, nonvertebral fractures, severe hypoglycemia, and AKI have shown positive predictive values >80%.27–31 We adapted definition codes from previous studies for safety outcomes without validation studies (genital infections, lower limb amputations, and severe UTI).32–34

Covariates

Baseline characteristics were measured during 365 days before or on cohort entry. These included demographics, comorbid conditions, diabetes-specific complications, use of diabetes and non–diabetes-related drugs, and measures of health care use (Supplemental Table 1). To address potential confounding by frailty, we also used a claims-based frailty index.35 Laboratory results were available for approximately 15% of the overall population (approximately 45% of patients in CDM and 5%–10% in IBM MarketScan). Race was self-reported in the claims data sources and not specifically collected for research purposes. There were no missing data for the other covariates (the absence of a diagnosis or procedure code was interpreted as the absence of a particular condition).

Statistical Analyses

We used 1:1 propensity score matching using the nearest neighbor method with a caliper of 0.01 of the propensity score to adjust for confounding.36 We estimated the probability of receiving SGLT2i versus GLP-1RA as a function of >120 pre-exposure covariates using multivariable logistic regression. All covariates listed in Supplemental Table 1 were included in the propensity score model except for laboratory results, which were only available in a subset of patients. Covariate balance before and after matching was assessed using standardized mean differences.37,38 Balance in laboratory results was also inspected to assess potential residual confounding by unmeasured factors because laboratory results were not included in the propensity score. For all outcomes, we calculated propensity score–matched numbers of events, incidence rates, incidence rate differences, and hazard ratios (HRs). The HRs and incidence rate differences with their 95% confidence intervals (CIs) were estimated in each data source and then pooled using a fixed effects meta-analysis. HRs were estimated using cause-specific Cox regression, and incidence rate differences using generalized linear regression with identity link function and normal error distribution.39 We constructed cumulative incidence function plots with the Aalen–Johansen estimator, which does not overestimate risks in the presence of the competing risk of death.40 Analyses were performed using R version 3.6.2 and Aetion Evidence Platform v4.53.

Subgroup, Sensitivity, and Post Hoc Analyses

We performed subgroup analyses in the following prespecified strata: age (65–74 versus 75 years or older), sex, cardiovascular disease, heart failure, metformin use, insulin use, and sulfonylurea use. Within each subgroup, we re-estimated the propensity score and reperformed 1:1 propensity score matching. We also performed multiple sensitivity analyses. First, we defined treatment discontinuation as no prescription refill within 60 days (instead of 30 days). Second, to investigate the influence of informative censoring, we applied an intention-to-treat follow-up approach, where follow-up was continued for a maximum of 6 and 12 months regardless of treatment discontinuation or switch. We also performed three post hoc analyses. First, to address the potential for unmeasured confounding associated with risk for recurrence, we excluded individuals with prior nonvertebral fractures and lower limb amputations. Second, to investigate a potential effect of GLP-1RA on some of the study outcomes, we also compared SGLT2i with dipeptidyl peptidase-4 inhibitor (DPP4i). Third, we investigated the association between SGLT2i versus GLP-1RA with different types of nonvertebral fractures (hip and femur, humerus, pelvis, radius, and ulna).

Results Study Population

We included a total of 96,128 individuals with CKD and type 2 diabetes, of whom 32,192 initiated SGLT2i and 63,936 GLP-1RA (Supplemental Figure 2). Although reasonably well balanced in baseline characteristics before propensity score matching, compared with GLP-1RA initiators, the SGLT2i group was slightly older, more likely to be male, and less likely to have obesity or CKD stage 4 (Table 1, Supplemental Table 3). They were also more likely to use metformin and DPP4i, but less likely to use insulin. After 1:1 propensity score matching 28,847 SGLT2i initiators to 28,847 GLP-1RA initiators, differences in patient characteristics were well balanced across treatment groups (see Supplemental Tables 4–6 for baseline characteristics in each database, Supplemental Figure 3 for propensity score distributions before and after propensity score matching). Laboratory results were also well balanced, except for a small difference in eGFR (2.6 ml/min per 1.73 m2 higher for SGLT2i users) among the subset with available data. In the matched cohort, the mean age was 72 years, 56% were men, and 64% were White. Furthermore, 25% had heart failure, 90% had stage 3 CKD, 41% used metformin, and 27% used insulin. Among SGLT2i agents, empagliflozin was most commonly used (51%), followed by canagliflozin (35%) and dapagliflozin (14%) (Supplemental Table 7). The most used GLP-1RA agents were dulaglutide (40%), liraglutide (33%), and semaglutide (13%).

Table 1 - Selected baseline characteristics of patients with CKD and type 2 diabetes initiating treatment with SGLT2i versus GLP-1RA, before and after 1:1 propensity score matching in the pooled cohort Characteristic Before Propensity Score Matching After 1:1 Propensity Score Matching SGLT2i GLP-1RA SMD SGLT2i GLP-1RA SMD Total 32,192 63,936 28,847 28,847 Demographics  Age, mean (SD) 73 (7) 71 (7) −0.16 72 (7) 72 (7) 0.00  Men, n (%) 18,452 (57) 32,061 (50) −0.14 16,080 (56) 16,070 (56) 0.00  Race/ethnicity,an (%)    Asian 1719 (5) 1842 (3) −0.12 1202 (4) 1162 (4) −0.01    Black 3184 (10) 6442 (10) 0.01 2840 (10) 2874 (10) 0.01    Hispanic 2207 (7) 3650 (6) −0.05 1888 (7) 1836 (6) 0.00    Other 1497 (5) 2528 (4) −0.03 1280 (4) 1276 (4) 0.00    White 20,135 (63) 42,214 (66) 0.07 18,444 (64) 18,506 (64) 0.01 Burden of comorbidities  Combined comorbidity score, mean (SD)b 4.07 (2.48) 4.14 (2.38) 0.03 4.03 (2.45) 4.06 (2.43) 0.01  Frailty score, mean (SD)c 0.20 (0.06) 0.21 (0.06) 0.17 0.21 (0.06) 0.21 (0.06) 0.00 Comorbidities, n (%)  Hypertension 30,996 (96) 61,926 (97) 0.03 27,797 (96) 27,758 (96) −0.01  Hyperlipidemia 28,753 (89) 57,081 (89) 0.00 25,732 (89) 25,678 (89) −0.01  Cardiovascular diseased 19,204 (60) 37,233 (58) −0.03 16,935 (59) 16,962 (59) 0.00  Acute myocardial infarction 1541 (5) 2395 (4) −0.05 1222 (4) 1249 (4) 0.00  Coronary atherosclerosis 13,546 (42) 25,324 (40) −0.05 11,823 (41) 11,955 (41) 0.01  Heart failure 8332 (26) 16,428 (26) 0.00 7236 (25) 7267 (25) 0.00  Ischemic stroke 4814 (15) 9054 (14) −0.02 4218 (15) 4217 (15) 0.00  Peripheral arterial disease 6058 (19) 12,059 (19) 0.00 5361 (19) 5425 (19) 0.01  AKI 5362 (17) 12,292 (19) 0.07 4835 (17) 4896 (17) 0.01  CKD stage 3 29,339 (91) 53,125 (83) 0.24 26,086 (90) 26,104 (91) 0.00  CKD stage 4 2853 (9) 10,811 (17) 0.24 2761 (10) 2743 (10) 0.00  Urinary tract infection 5683 (18) 13,376 (21) 0.08 5228 (18) 5208 (18) 0.00  Kidney and urinary stone 2315 (7) 4982 (8) 0.02 2095 (7) 2093 (7) 0.00  Edema 6601 (21) 16,047 (25) 0.11 6102 (21) 6107 (21) 0.00  COPD 5554 (17) 11,274 (18) 0.01 4951 (17) 5028 (17) 0.01  Asthma 3080 (10) 6889 (11) 0.04 2809 (10) 2848 (10) 0.01  Fractures 655 (2) 1590 (3) 0.03 606 (2) 623 (2) 0.01  Falls 2175 (7) 4725 (7) 0.02 1943 (7) 1972 (7) 0.00 Diabetes-related conditions, n (%)  Diabetic kidney disease 21,765 (68) 45,445 (71) 0.08 19,501 (68) 19,570 (68) 0.00  Diabetic retinopathy 5737 (18) 13,610 (21) 0.09 5265 (18) 5225 (18) −0.01  Diabetic neuropathy 11,280 (35) 25,859 (40) 0.11 10,324 (36) 10,452 (36) 0.01  Diabetes with peripheral circulatory disorders 796 (3) 1684 (3) 0.01 716 (3) 732 (3) 0.00  Diabetic foot 1389 (4) 3677 (6) 0.07 1300 (5) 1309 (5) 0.00  Lower limb amputation 352 (1) 998 (2) 0.04 327 (1) 337 (1) 0.01  Hypoglycemia 5770 (18) 11,834 (19) 0.02 5089 (18) 5137 (18) 0.01  Diabetic ketoacidosis 121 (0.4) 259 (0.4) 0.00 104 (0.4) 100 (0.3) −0.02 No. of distinct medications, mean (SD) 16 (6) 17 (7) 0.16 16 (6) 16 (6) 0.01 Diabetes medications on day of entry to cohort  No. of antidiabetes drugs, mean (SD) 2 (1) 2 (1) −0.11 2 (1) 2 (1) 0.00  Metformin, n (%) 14,063 (44) 18,617 (29) −0.31 11,887 (41) 11,455 (40) −0.03  Sulfonylureas, n (%) 12,374 (38) 22,349 (35) −0.07 11,033 (38) 11,377 (39) 0.02  DPP-4 inhibitors, n (%) 9783 (30) 14,164 (22) −0.19 8404 (29) 7757 (27) −0.05  Insulin, n (%) 7488 (23) 26,224 (41) 0.39 7378 (26) 7934 (28) 0.04 Other medication use, n (%)  ACE inhibitors or angiotensin II receptor blockers 26,282 (82) 51,546 (81) −0.03 23,536 (82) 23,508 (82) 0.00  Beta blockers 19,572 (61) 39,597 (62) 0.02 17,482 (61) 17,501 (61) 0.00  Calcium channel blockers 14,467 (45) 29,410 (46) 0.02 12,985 (45) 12,880 (45) −0.01  Loop diuretics 10,711 (33) 25,734 (40) 0.14 9780 (34) 9828 (34) 0.00  Statins 27,347 (85) 54,022 (85) −0.01 24,414 (85) 24,383 (85) 0.00  Antiplatelets 6285 (20) 11,366 (18) −0.04 5439 (19) 5425 (19) 0.00  Anticoagulants 4948 (15) 9767 (15) 0.00 4395 (15) 4381 (15) 0.00  Oral corticosteroids 6459 (20) 13,124 (21) 0.01 5771 (20) 5858 (20) 0.01  Antiosteoporosis agents 1754 (5) 3017 (5) −0.03 1475 (5) 1480 (5) 0.00  Opioids 11,387 (35) 25,951 (41) 0.11 10,486 (36) 10,651 (37) 0.01 Health care utilization markers, mean (SD)  No. of hospital days 1.66 (4.99) 1.72 (5.31) −0.01 1.66 (5.41) 1.64 (5.19) 0.00  No. of emergency department visits 0.86 (1.99) 0.92 (2.04) 0.03 0.86 (1.97) 0.87 (2.09) 0.00  No. of internist visits 22.42 (26.93) 21.43 (26.10) −0.04 22.03 (26.52) 22.18 (26.73) 0.01  No. of cardiologist visits 5.85 (10.80) 5.16 (9.65) −0.07 5.50 (10.14) 5.52 (10.43) 0.00  No. of endocrinologist visits 1.70 (7.10) 2.46 (7.40) 0.10 1.81 (7.39) 1.80 (6.00) −0.00  No. of nephrologist visits 1.90 (5.82) 2.51 (6.32) 0.10 1.97 (6.02) 1.97 (5.32) 0.00  No. of HbA1c tests ordered 3.00 (1.57) 3.12 (1.57) 0.08 3.02 (1.57) 3.02 (1.55) 0.00  No. of metabolic or creatinine tests ordered 4.86 (3.78) 5.22 (3.79) 0.10 4.88 (3.72) 4.90 (3.57) 0.01  No. of microalbuminuria/proteinuria tests ordered 1.58 (1.58) 1.70 (1.62) 0.07 1.60 (1.59) 1.59 (1.54) −0.01

SGLT2i, sodium-glucose cotransporter-2 inhibitor; GLP-1RA, glucagon-like peptide-1 receptor agonist; SMD, standardized mean difference; n, number of patients; COPD, chronic obstructive pulmonary disease; No., number of; DPP-4, dipeptidyl peptidase-4; ACE, angiotensin-converting enzyme; HbA1c, hemoglobin A1c.

aPooled across Clinformatics Data Mart and Medicare databases.

dCardiovascular disease was defined as a composite of myocardial infarction, stable angina, acute coronary syndrome, coronary atherosclerosis, history of coronary procedure, heart failure, ischemic stroke, and peripheral vascular disease.


Safety of SGLT2i versus GLP-1RA

The mean on-treatment follow-up time was 7.5 months (median, 4.0 months). Most patients were censored due to treatment discontinuation (62%) or end of study period (23%) (Supplemental Table 8). In the 1:1 propensity score–matched cohort, SGLT2i compared with GLP-1RA were associated with a higher risk of nonvertebral fractures (HR, 1.30 [95% CI, 1.03 to 1.65]; incidence rate difference, 2.13 [95% CI, 0.28 to 3.97] events per 1000 person-years), lower limb amputations (HR, 1.65 [95% CI, 1.22 to 2.23]; incidence rate difference, 2.46 [95% CI, 1.00 to 3.92]), and genital infections (HR, 3.08 [95% CI, 2.73 to 3.48]; incidence rate difference, 41.26 [95% CI, 37.06 to 45.46]) (Figure 1). Similar risks of DKA (HR, 1.07 [95% CI, 0.74 to 1.54]; incidence rate difference, 0.29 [95% CI, −0.89 to 1.46]), hypovolemia (HR, 0.99 [95% CI, 0.86 to 1.14]; incidence rate difference, 0.20 [95% CI, −2.85 to 3.25]), hypoglycemia (HR, 1.08 [95% CI, 0.92 to 1.26]; incidence rate difference, 1.46 [95% CI, −1.31 to 4.23]), and severe UTI (HR, 1.02 [95% CI, 0.87 to 1.19]; incidence rate difference, 0.35 [95% CI, −2.51 to 3.21]) were observed, and a lower AKI risk (HR, 0.93 [95% CI, 0.87 to 0.99]; incidence rate difference, −6.75 [95% CI, −13.69 to 0.20]) was observed. Cumulative incidence curves (Figure 2) showed that the divergence for lower limb amputations, genital infections, and nonvertebral fractures occurred within the first 6 months of follow-up.

fig1Figure 1:

Number of events, incidence rates, incidence rate differences, and hazard ratios for safety outcomes, comparing SGLT2i versus GLP-1RA after 1:1 propensity score matching. CI, confidence interval; GLP-1RA, glucagon-like peptide-1 receptor agonist; HR, hazard ratio; IR, incidence rate; PY, person-year; SGLT2i, sodium-glucose cotransporter-2 inhibitors; UTI, urinary tract infection.

fig2

留言 (0)

沒有登入
gif