To evaluate the comparative cardiovascular effectiveness and safety of sodium–glucose cotransporter 2 inhibitors (SGLT-2is), glucagon-like peptide 1 receptor agonists (GLP-1RAs), and dipeptidyl peptidase 4 inhibitors (DPP-4is) in older adults with type 2 diabetes (T2D) across different frailty strata.
We performed three 1:1 propensity score–matched cohort studies, each stratified by three frailty strata, using data from Medicare beneficiaries (2013–2019) with T2D who initiated SGLT-2is, GLP-1RAs, or DPP-4is. In time-to-event analyses, we assessed the primary cardiovascular effectiveness composite outcome of acute myocardial infarction, ischemic stroke, hospitalization for heart failure, and all-cause mortality. The primary safety outcome was a composite of severe adverse events that have been linked to SGLT-2i or GLP-1RA use.
Compared with DPP-4is, the overall hazard ratio (HR) for the primary effectiveness outcome associated with SGLT-2is (n = 120,202 matched pairs) was 0.72 (95% CI 0.69–0.75), corresponding to an incidence rate difference (IRD) of −13.35 (95% CI −15.06 to −11.64). IRD ranged from −6.74 (95% CI −8.61 to −4.87) in nonfrail to −27.24 (95% CI −41.64 to −12.84) in frail people (P for interaction < 0.01). Consistent benefits were observed for GLP-1RAs compared with DPP-4is (n = 113,864), with an overall HR of 0.74 (95% CI 0.71–0.77) and an IRD of −15.49 (95% CI −17.46 to −13.52). IRD in the lowest frailty stratum was −7.02 (95% CI −9.23 to −4.81) and −25.88 (95% CI −38.30 to −13.46) in the highest (P for interaction < 0.01). Results for SGLT-2is versus GLP-1RAs (n = 89,865) were comparable. Severe adverse events were not more frequent with SGLT-2is or GLP-1RAs than DPP-4is.
SGLT-2is and GLP-1RAs safely improved cardiovascular outcomes and all-cause mortality, with the largest absolute benefits among frail people.
Introduction
One in every four older adult with type 2 diabetes (T2D) has frailty, which refers to a clinically detectable state of decreased physiological reserve and increased vulnerability to stressors and poor clinical outcomes (1,2). T2D increases the risk of frailty by its effect on sarcopenia, mobility, cognitive impairment, and exhaustion or through microvascular and macrovascular complications, such as neuropathy or cardiovascular dysautonomia (3,4). While frailty differentiates between chronological and biological age (5), frail people are at higher risk for poor outcomes related to T2D, in part because of poor treatment tolerance and low adherence (6). Multimorbidity, polypharmacy, and unfavorable behavioral factors add to the complexity and challenges in their diabetes management.
Addressing frailty is now increasingly recognized as part of routine clinical assessment and a goal in the holistic management of diabetes care (7). Because of the greater risk of hypoglycemia and lower life expectancy, guidelines recommend less tight glycated hemoglobin targets among older and frail people with diabetes (7,8); however, it is still debated how these recommendations should be applied and whether certain antidiabetic medications are more favorable than others (9). Many clinicians may be leery of initiating a new antidiabetic drug because of concern about adverse effects, especially since risk-benefit information is widely lacking in frail people (9). Although older people with frailty are usually excluded from clinical trials, newer antidiabetic drug classes with known cardiovascular benefits, such as sodium–glucose cotransporter 2 inhibitors (SGLT-2is) and glucagon-like peptide 1 receptor agonists (GLP-1RAs), are expected to be increasingly prescribed to them. To better direct the clinicians’ approach to setting priorities in clinical diabetes care and to address a gap in knowledge recently recognized by the American Diabetes Association and European Association for the Study of Diabetes (10), we sought to assess the comparative cardiovascular effectiveness and safety of SGLT-2is and GLP-1RAs versus dipeptidyl peptidase 4 inhibitors (DPP-4is) in older people with T2D according to frailty status using nationwide Medicare claims data.
Research Design and Methods
Data Source and Study Design
This was a retrospective cohort study using the entire Medicare fee-for-service data from Parts A (inpatient coverage), B (outpatient coverage), and D (prescription drug coverage). The institutional review board of Mass General Brigham approved the study (#2021P001784), and a data use agreement was in place.
To emulate three frailty-stratified randomized trials comparing SGLT-2is with DPP-4is, GLP-1RAs with DPP-4is, and SGLT-2is with GLP-1RAs, we created three new-user cohorts of Medicare beneficiaries with T2D who filled a prescription of one of the three antidiabetic drug classes of interest between April 2013 (first month after the approval of the first SGLT-2i canagliflozin on 29 March 2013) and December 2019.
Study Population
The eligible study population included patients aged ≥65 years who initiated (cohort entry) an SGLT-2i (i.e., canagliflozin, dapagliflozin, empagliflozin, ertugliflozin), a GLP-1RA (i.e., exenatide, lixisenatide, albiglutide, dulaglutide, liraglutide, semaglutide), or a DPP-4i (i.e., sitagliptin, saxagliptin, linagliptin, alogliptin), with no exposure to that specific class or the comparator class, depending on the pairwise comparison of interest, in the previous 365 days. In the 365 days before cohort entry, eligible patients were required to have at least one inpatient or outpatient diagnosis of T2D and remain continuously enrolled in Medicare. Patients were excluded if they had a diagnosis of type 1 or secondary diabetes, end-stage renal disease, solid organ transplant, or a skilled nursing home admission in the previous 365 days or if they concomitantly started a second antidiabetic drug within the same drug class of interest on cohort entry date. To address the potential for unmeasured confounding associated with the high risk for recurrence of a 1) cardiovascular or 2) safety outcome, we excluded patients with 1) a hospitalization for acute myocardial infarction, ischemic stroke, heart failure, or coronary or cerebrovascular revascularization and 2) a lower-limb amputation, nonvertebral fracture, hospital or emergency department admission for hypoglycemia, or hospitalization for acute kidney injury, diabetic ketoacidosis (DKA), severe genital or urinary tract infection, acute pancreatitis, or nonmalignant biliary event (i.e., cholelithiasis, cholecystitis, cholangitis, nonmalignant gallbladder and bile duct disorders) in the 90 days before cohort entry (Supplementary Table 1 and Supplementary Fig. 1).
Patient Characteristics and Measurement of Frailty
Patient characteristics at baseline were measured during the 12 months before and including the date of cohort entry. Covariates of interest included demographic characteristics (age, sex, race, census region), calendar time (in semesters), comorbidities, diabetes-specific complications, use of diabetes drugs, use of other medications, indicators of health care use as proxy for overall disease state, surveillance, and intensity of care. The overall burden of comorbidity was quantified using a combined comorbidity score (11). A complete list of baseline patient characteristics, overall and stratified by frailty status, is reported in Supplementary Tables 2–10.
Frailty was measured using a validated claims-based frailty index (CFI) (12), which estimates a deficit accumulation frailty index (range 0–1) using 93 variables defined by diagnoses, health services, and durable medical equipment codes in the 365 days before cohort entry. Details on the CFI and its clinical application have been previously described (13,14). Using clinically meaningful cut points (14–16), we defined the status of nonfrailty as a CFI <0.15, prefrailty as a CFI of 0.15–0.24, and frailty as a CFI of ≥0.25.
Follow-up and Outcomes
Study follow-up for outcome assessment began on the day after cohort entry and continued in an as-treated approach until treatment discontinuation (defined as a treatment gap [grace period] >60 days), switch to or augmentation with a drug in the comparator class, the occurrence of a specific study outcome (either cardiovascular or safety event), death, end of continuous health plan enrollment, or end of the study period, whichever came first. The as-treated approach was chosen as a primary statistical analysis to address the risk of exposure misclassification caused by treatment nonadherence, as frequently observed in a real-world setting. The primary cardiovascular effectiveness outcome was time to the first occurrence of a composite of hospitalization for acute myocardial infarction or ischemic stroke, hospitalization for heart failure (HFF), or all-cause mortality. In prior studies, the positive predictive values of claims-based algorithms for acute myocardial infarction, ischemic stroke, and HHF were at least 84% (17–20) (Supplementary Table 11). Secondary effectiveness outcomes included the individual components of the primary effectiveness outcome. These effectiveness outcomes were selected to mirror a four-point major adverse cardiac event outcome commonly explored in cardiovascular outcome trials. Since Medicare data do not include any information on cause of death, we were not able to include cardiovascular mortality in our primary effectiveness outcome. As a substitute, however, we included overall mortality as a fourth component of the primary effectiveness outcome.
The primary safety outcome was time to the first occurrence of a composite of the following nine events that could be potentially linked to the use of antidiabetic drugs of interest: lower-limb amputation, nonvertebral fracture, hospital or emergency department admission for hypoglycemia, or hospitalization for acute kidney injury, DKA, severe genital or urinary tract infection, acute pancreatitis, or nonmalignant biliary event. Definitions were either validated against medical records or previously used in pharmacoepidemiologic studies (Supplementary Table 11). Secondary safety outcomes included the individual components of the primary safety outcome, except for hospitalizations for severe genital infection and DKA, which were not individually assessed owing to a small number of events and the Centers for Medicare & Medicaid Services Cell Size Suppression Policy (21). Instead, we separately examined genital infections in any clinical setting and hospitalization with a discharge diagnosis of DKA in any position (Supplementary Table 11).
Statistical Analysis
For each of the three cohorts, the analyses were conducted both in the entire study population and within each frailty stratum. Baseline characteristics were compared by exposure groups and stratified by frailty status. To reduce imbalance in patient characteristics and to address confounding, for each cohort-specific frailty stratum, we estimated a propensity score (PS) using a logistic regression that modeled the probability of newly initiating SGLT-2is versus DPP-4is, GLP-1RAs versus DPP-4is, or SGLT-2is versus GLP-1RAs. All prespecified covariates (>170) were included in the PS model. Within each frailty stratum, patients were 1:1 PS matched without replacement using the nearest neighbor method with a maximum caliper of 0.01 on the PS scale (22). Thus, our estimand was the average treatment effect in the treated for whom comparator matches could be found within the frailty strata. The total PS-matched population was created by pooling the three matched frailty subgroups to mimic stratified randomization by frailty status (23). A standardized mean difference <0.10 was considered adequate for covariate balance after PS matching. The density of the PSs for each of the cohorts showed a good area of common support before PS matching and an excellent area of common support after PS matching (Supplementary Figs. 2–4). In the matched cohort, we summarized the rates of all primary and secondary outcomes, estimated hazard ratios (HRs) using a stratified (by frailty strata) Cox proportional hazards model, and calculated incidence rate differences (IRDs) with their respective 95% CIs. We evaluated for heterogeneity in HR and IRD estimates across frailty strata using the Wald test for homogeneity. Number needed to treat (NNT) or to number need to harm were calculated at 1-year of follow-up (24). We performed several sensitivity and subgroup analyses, which are described in detail in the Supplementary Information on Statistical Analyses. All analyses were performed using SAS 9.4 statistical software (SAS Institute, Cary, NC).
Results
Characteristics of Study Cohorts
Of 744,310 eligible patients for PS matching, we identified three 1:1 PS-matched new-user cohorts comparing SGLT-2is versus DPP-4is (n = 120,202 matched pairs), GLP-1RAs versus DPP-4is (n = 113,864 pairs), and SGLT-2is versus GLP-1RAs (n = 89,865 pairs) (Supplementary Fig. 5). In total, 250,028 patients were not frail, 359,562 were prefrail, and 38,272 were frail. For the SGLT-2i versus DPP-4i comparison, we successfully matched between 96.0% (nonfrail) and 97.0% (frail) of patients treated with SGLT-2is, for the GLP-1RA versus DPP-4i comparison, we matched between 83.1% (nonfrail) and 86.2% (frail) of patients treated with GLP-1RAs, and for the SGLT-2i versus GLP-1RA comparison, we matched between 65.3% (nonfrail) and 84.1% (frail) of patients treated with SGLT-2is. For DPP-4i after PS matching, all patient characteristics were balanced within each cohort and frailty stratum (Supplementary Tables 2–10). Number and percentage of individual SGLT-2i, GLP-1RA, and DPP-4i agents included in the analysis are shown in Supplementary Table 12. Among the three PS-matched cohorts, the SGLT-2i versus DPP-4i cohort had older age and a higher proportion of male patients, lower mean CFI, lower mean comorbidity score, lower burden of comorbidities, lower medication use, and lower number of emergency department visits (Table 1).
Characteristic . | SGLT-2i vs. DPP-4i . | GLP-1RA vs. DPP-4i . | SGLT-2i vs. GLP-1RA . | |||
---|---|---|---|---|---|---|
SGLT-2i (n = 120,202) . | DPP-4i (n = 120,202) . | GLP-1RA (n = 113,864) . | DPP-4i (n = 113,864) . | SGLT-2i (n = 89,865) . | GLP-1RA (n = 89,865) . | |
Age, years, mean (SD) | 72.3 (5.4) | 72.3 (5.5) | 72.0 (5.3) | 71.9 (5.3) | 71.8 (5.1) | 71.8 (5.2) |
Male sex, n (%) | 62,830 (52.3) | 62,334 (51.9) | 51,555 (45.3) | 51,319 (45.1) | 43,888 (48.8) | 44,031 (49.0) |
Race, n (%) | ||||||
Asian | 4,268 (3.6) | 4,318 (3.6) | 2,091 (1.8) | 1,923 (1.7) | 1,711 (1.9) | 1,842 (2.1) |
Black | 9,078 (7.6) | 9,140 (7.6) | 10,141 (8.9) | 10,136 (8.9) | 7,003 (7.8) | 7,036 (7.8) |
White | 97,364 (81.0) | 97,161 (80.8) | 94,006 (82.6) | 94,269 (82.8) | 75,035 (83.5) | 74,808 (83.2) |
Other | 9,492 (7.9) | 9,583 (8.0) | 7,626 (6.7) | 7,536 (6.6) | 6,116 (6.8) | 6,179 (6.9) |
Geographic region, n (%) | ||||||
Midwest | 24,839 (20.7) | 24,705 (20.6) | 25,950 (22.8) | 26,126 (22.9) | 20,028 (22.3) | 19,894 (22.1) |
Northeast | 21,444 (17.8) | 21,401 (17.8) | 18,595 (16.3) | 18,032 (15.8) | 15,071 (16.8) | 15,094 (16.8) |
South | 51,837 (43.1) | 51,961 (43.2) | 49,466 (43.4) | 49,920 (43.8) | 38,902 (43.3) | 38,992 (43.4) |
West | 21,903 (18.2) | 21,969 (18.3) | 19,719 (17.3) | 19,667 (17.3) | 15,758 (17.5) | 15,781 (17.6) |
Unknown | 179 (0.2) | 166 (0.1) | 134 (0.1) | 119 (0.1) | 106 (0.1) | 104 (0.1) |
Frailty index, mean (SD) | 0.16 (0.04) | 0.16 (0.04) | 0.17 (0.05) | 0.17 (0.05) | 0.17 (0.05) | 0.17 (0.05) |
Not frail (CFI <0.15) | 50,843 (42.3) | 50,843 (42.3) | 39,675 (34.8) | 39,675 (34.8) | 34,496 (38.4) | 34,496 (38.4) |
Prefrail (CFI 0.15–0.24) | 63,649 (53.0) | 63,649 (53.0) | 65,715 (57.7) | 65,715 (57.7) | 50,417 (56.1) | 50,417 (56.1) |
Frail (CFI ≥0.25) | 5,710 (4.8) | 5,710 (4.8) | 8,474 (7.4) | 8,474 (7.4) | 4,952 (5.5) | 4,952 (5.5) |
Combined Comorbidity Score, mean (SD) | 1.94 (2.4) | 1.94 (2.4) | 2.40 (2.6) | 2.40 (2.6) | 2.13 (2.5) | 2.13 (2.5) |
Lifestyle factors, n (%) | ||||||
Overweight | 12,078 (10.1) | 12,076 (10.1) | 9,723 (8.5) | 9,949 (8.7) | 8,330 (9.3) | 8,329 (9.3) |
Obese | 43,336 (36.1) | 43,459 (36.2) | 51,249 (45.0) | 51,571 (45.3) | 39,713 (44.2) | 39,454 (43.9) |
Smoking | 27,843 (23.2) | 27,838 (23.2) | 28,454 (25.0) | 28,379 (24.9) | 22,308 (24.8) | 22,284 (24.8) |
Diabetes-related conditions, n (%) | ||||||
Nephropathy | 15,783 (13.1) | 15,576 (13.0) | 22,301 (19.6) | 22,255 (19.6) | 14,090 (15.7) | 14,106 (15.7) |
Neuropathy | 30,975 (25.8) | 30,829 (25.7) | 35,016 (30.8) | 34,909 (30.7) | 26,342 (29.3) | 26,307 (29.3) |
Retinopathy | 6,900 (5.7) | 6,799 (5.7) | 7,928 (7.0) | 7,869 (6.9) | 5,404 (6.0) | 5,394 (6.0) |
Hypoglycemia | 14,887 (12.4) | 14,804 (12.3) | 16,502 (14.5) | 16,368 (14.4) | 12,076 (13.4) | 12,143 (13.5) |
Hyperglycemia | 45,798 (38.1) | 45,808 (38.1) | 45,686 (40.1) | 45,599 (40.1) | 37,949 (42.2) | 37,847 (42.1) |
DKA | 316 (0.3) | 314 (0.3) | 397 (0.4) | 394 (0.4) | 263 (0.3) | 260 (0.3) |
Medical history, n (%) | ||||||
Hypertension | 110,878 (92.2) | 110,859 (92.2) | 106,339 (93.4) | 106,297 (93.4) | 83,390 (92.8) | 83,414 (92.8) |
Hyperlipidemia | 102,261 (85.1) | 102,202 (85.0) | 97,412 (85.6) | 97,342 (85.5) | 76,493 (85.1) | 76,613 (85.3) |
Acute myocardial infarction | 2,391 (2.0) | 2,343 (2.0) | 2,276 (2.0) | 2,302 (2.0) | 1,809 (2.0) | 1,822 (2.0) |
Coronary atherosclerosis | 39,975 (33.3) | 39,324 (32.7) | 37,335 (32.8) | 37,264 (32.7) | 29,675 (33.0) | 29,787 (33.2) |
Heart failure | 14,522 (12.1) | 14,263 (11.9) | 16,605 (14.6) | 16,705 (14.7) | 11,689 (13.0) | 11,745 (13.1) |
Atrial fibrillation | 15,248 (12.7) | 14,947 (12.4) | 14,885 (13.1) | 14,896 (13.1) | 11,468 (12.8) | 11,558 (12.9) |
Ischemic stroke | 14,081 (11.7) | 13,999 (11.7) | 13,626 (12.0) | 13,620 (12.0) | 10,449 (11.6) | 10,426 (11.6) |
Peripheral artery disease | 15,312 (12.7) | 15,226 (12.7) | 16,142 (14.2) | 16,147 (14.2) | 11,741 (13.1) | 11,873 (13.2) |
Chronic obstructive pulmonary disease | 16,086 (13.4) | 16,118 (13.4) | 17,604 (15.5) | 17,893 (15.7) | 12,832 (14.3) | 12,820 (14.3) |
Acute kidney injury | 3,627 (3.0) | 3,625 (3.0) | 5,913 (5.2) | 5,945 (5.2) | 3,196 (3.6) | 3,247 (3.6) |
Chronic kidney disease | 17,819 (14.8) | 17,701 (14.7) | 28,137 (24.7) | 28,128 (24.7) | 15,977 (17.8) | 15,980 (17.8) |
Urinary tract infection | 18,254 (15.2) | 18,310 (15.2) | 21,494 (18.9) | 21,505 (18.9) | 14,691 (16.4) | 14,727 (16.4) |
Osteoporosis without fracture | 9,092 (7.6) | 9,032 (7.5) | 9,098 (8.0) | 9,144 (8.0) | 6,734 (7.5) | 6,730 (7.5) |
Depression | 18,737 (15.6) | 18,753 (15.6) | 22,802 (20.0) | 22,947 (20.2) | 16,486 (18.4) | 16,435 (18.3) |
Dementia | 5,585 (4.7) | 5,672 (4.7) | 6,591 (5.8) | 6,509 (5.7) | 4,396 (4.9) | 4,382 (4.9) |
Diabetes treatment at cohort entry, n (%) | ||||||
Metformin | 76,879 (64.0) | 77,331 (64.3) | 62,042 (54.5) | 62,505 (54.9) | 53,207 (59.2) | 53,438 (59.5) |
Sulfonylurea | 41,883 (34.8) | 42,453 (35.3) | 36,578 (32.1) | 37,331 (32.8) | 29,558 (32.9) | 29,469 (32.8) |
Glitazone | 8,134 (6.8) | 8,150 (6.8) | 7,375 (6.5) | 7,539 (6.6) | 6,250 (7.0) | 6,227 (6.9) |
Insulin | 22,036 (18.3) | 21,837 (18.2) | 32,490 (28.5) | 32,181 (28.3) | 23,602 (26.3) | 23,700 (26.4) |
Other medication use, n (%) | ||||||
ACE inhibitors | 54,182 (45.1) | 54,146 (45.1) | 51,583 (45.3) | 51,713 (45.4) | 40,559 (45.1) | 40,515 (45.1) |
ARBs | 43,225 (36.0) | 43,184 (35.9) | 41,447 (36.4) | 41,330 (36.3) | 32,917 (36.6) | 33,001 (36.7) |
β-Blockers | 58,875 (49.0) | 58,670 (48.8) | 57,628 (50.6) | 57,532 (50.5) | 44,403 (49.4) | 44,540 (49.6) |
Calcium channel blockers | 40,590 (33.8) | 40,655 (33.8) | 39,710 (34.9) | 39,579 (34.8) | 30,300 (33.7) | 30,401 (33.8) |
Thiazides | 42,302 (35.2) | 42,514 (35.4) | 42,084 (37.0) | 42,170 (37.0) | 32,971 (36.7) | 32,816 (36.5) |
Loop diuretics | 21,310 (17.7) | 21,158 (17.6) | 27,490 (24.1) | 27,531 (24.2) | 18,466 (20.6) | 18,565 (20.7) |
Antiplatelets | 16,209 (13.5) | 15,955 (13.3) | 15,007 (13.2) | 14,966 (13.1) | 11,846 (13.2) | 11,912 (13.3) |
Statins | 93,938 (78.2) | 93,757 (78.0) | 88,571 (77.8) | 88,459 (77.7) | 70,183 (78.1) | 70,176 (78.1) |
Measures of health care use, n (%) | ||||||
Emergency department visits | 33,629 (28.0) | 34,050 (28.3) | 36,317 (31.9) | 36,658 (32.2) | 26,628 (29.6) | 26,562 (29.6) |
Hospitalization 30 days prior to cohort entry | 887 (0.7) | 859 (0.7) | 811 (0.7) | 826 (0.7) | 586 (0.7) | 598 (0.8) |
Internist visit 30 days prior to cohort entry | 75,269 (62.6) | 75,347 (62.7) | 67,919 (59.7) | 68,870 (60.5) | 54,808 (61.0) | 53,839 (59.9) |
Cardiologist visit 30 days prior to cohort entry | 11,742 (9.8) | 11,235 (9.4) | 9,120 (8.0) | 9,123 (8.0) | 7,951 (8.9) | 7,860 (8.8) |
Endocrinologist visit 30 days prior to cohort entry | 13,388 (11.1) | 12,689 (10.6) | 17,181 (15.1) | 16,389 (14.4) | 12,944 (14.4) | 13,129 (14.6) |
Glycated hemoglobin testing | 116,395 (96.8) | 116,278 (96.7) | 110,217 (96.8) | 110,114 (96.7) | 87,165 (97.0) | 87,057 (96.9) |
Creatinine testing | 13,266 (11.0) | 13,276 (11.0) | 14,053 (12.3) | 14,220 (12.5) | 10,480 (11.7) | 10,466 (11.7) |
Lipid testing | 109,686 (91.3) | 109,252 (90.9) | 102,315 (89.9) | 102,178 (89.7) | 81,379 (90.6) | 81,232 (90.4) |
Microalbuminuria testing | 72,352 (60.2) | 72,294 (60.1) | 68,747 (60.4) | 68,539 (60.2) | 54,894 (61.1) | 55,058 (61.3) |
Electrocardiogram | 60,255 (50.1) | 59,537 (49.5) | 57,709 (50.7) | 57,713 (50.7) | 45,335 (50.5) | 44,869 (49.9) |
Echocardiogram | 33,800 (28.1) | 33,551 (27.9) | 32,999 (29.0) | 33,258 (29.2) | 25,626 (28.5) | 25,679 (28.6) |
Characteristic . | SGLT-2i vs. DPP-4i . | GLP-1RA vs. DPP-4i . | SGLT-2i vs. GLP-1RA . | |||
---|---|---|---|---|---|---|
SGLT-2i (n = 120,202) . | DPP-4i (n = 120,202) . | GLP-1RA (n = 113,864) . | DPP-4i (n = 113,864) . | SGLT-2i (n = 89,865) . | GLP-1RA (n = 89,865) . | |
Age, years, mean (SD) | 72.3 (5.4) | 72.3 (5.5) | 72.0 (5.3) | 71.9 (5.3) | 71.8 (5.1) | 71.8 (5.2) |
Male sex, n (%) | 62,830 (52.3) | 62,334 (51.9) | 51,555 (45.3) | 51,319 (45.1) | 43,888 (48.8) | 44,031 (49.0) |
Race, n (%) | ||||||
Asian | 4,268 (3.6) | 4,318 (3.6) | 2,091 (1.8) | 1,923 (1.7) | 1,711 (1.9) | 1,842 (2.1) |
Black | 9,078 (7.6) | 9,140 (7.6) | 10,141 (8.9) | 10,136 (8.9) | 7,003 (7.8) | 7,036 (7.8) |
White | 97,364 (81.0) | 97,161 (80.8) | 94,006 (82.6) | 94,269 (82.8) | 75,035 (83.5) | 74,808 (83.2) |
Other | 9,492 (7.9) | 9,583 (8.0) | 7,626 (6.7) | 7,536 (6.6) | 6,116 (6.8) | 6,179 (6.9) |
Geographic region, n (%) | ||||||
Midwest | 24,839 (20.7) | 24,705 (20.6) | 25,950 (22.8) | 26,126 (22.9) | 20,028 (22.3) | 19,894 (22.1) |
Northeast | 21,444 (17.8) | 21,401 (17.8) | 18,595 (16.3) | 18,032 (15.8) | 15,071 (16.8) | 15,094 (16.8) |
South | 51,837 (43.1) | 51,961 (43.2) | 49,466 (43.4) | 49,920 (43.8) | 38,902 (43.3) | 38,992 (43.4) |
West | 21,903 (18.2) | 21,969 (18.3) | 19,719 (17.3) | 19,667 (17.3) | 15,758 (17.5) | 15,781 (17.6) |
Unknown | 179 (0.2) | 166 (0.1) | 134 (0.1) | 119 (0.1) | 106 (0.1) | 104 (0.1) |
Frailty index, mean (SD) | 0.16 (0.04) | 0.16 (0.04) | 0.17 (0.05) | 0.17 (0.05) | 0.17 (0.05) | 0.17 (0.05) |
Not frail (CFI <0.15) | 50,843 (42.3) | 50,843 (42.3) | 39,675 (34.8) | 39,675 (34.8) | 34,496 (38.4) | 34,496 (38.4) |
Prefrail (CFI 0.15–0.24) | 63,649 (53.0) | 63,649 (53.0) | 65,715 (57.7) | 65,715 (57.7) | 50,417 (56.1) | 50,417 (56.1) |
Frail (CFI ≥0.25) | 5,710 (4.8) | 5,710 (4.8) | 8,474 (7.4) | 8,474 (7.4) | 4,952 (5.5) | 4,952 (5.5) |
Combined Comorbidity Score, mean (SD) | 1.94 (2.4) | 1.94 (2.4) | 2.40 (2.6) | 2.40 (2.6) | 2.13 (2.5) | 2.13 (2.5) |
Lifestyle factors, n (%) | ||||||
Overweight | 12,078 (10.1) | 12,076 (10.1) | 9,723 (8.5) | 9,949 (8.7) | 8,330 (9.3) | 8,329 (9.3) |
Obese | 43,336 (36.1) | 43,459 (36.2) | 51,249 (45.0) | 51,571 (45.3) | 39,713 (44.2) | 39,454 (43.9) |
Smoking | 27,843 (23.2) | 27,838 (23.2) | 28,454 (25.0) | 28,379 (24.9) | 22,308 (24.8) | 22,284 (24.8) |
Diabetes-related conditions, n (%) | ||||||
Nephropathy | 15,783 (13.1) | 15,576 (13.0) | 22,301 (19.6) | 22,255 (19.6) | 14,090 (15.7) | 14,106 (15.7) |
Neuropathy | 30,975 (25.8) | 30,829 (25.7) | 35,016 (30.8) | 34,909 (30.7) | 26,342 (29.3) | 26,307 (29.3) |
Retinopathy | 6,900 (5.7) | 6,799 (5.7) | 7,928 (7.0) | 7,869 (6.9) | 5,404 (6.0) | 5,394 (6.0) |
Hypoglycemia | 14,887 (12.4) | 14,804 (12.3) | 16,502 (14.5) | 16,368 (14.4) | 12,076 (13.4) | 12,143 (13.5) |
Hyperglycemia | 45,798 (38.1) | 45,808 (38.1) | 45,686 (40.1) | 45,599 (40.1) | 37,949 (42.2) | 37,847 (42.1) |
DKA | 316 (0.3) | 314 (0.3) | 397 (0.4) | 394 (0.4) | 263 (0.3) | 260 (0.3) |
Medical history, n (%) | ||||||
Hypertension | 110,878 (92.2) | 110,859 (92.2) | 106,339 (93.4) | 106,297 (93.4) | 83,390 (92.8) | 83,414 (92.8) |
Hyperlipidemia | 102,261 (85.1) | 102,202 (85.0) | 97,412 (85.6) | 97,342 (85.5) | 76,493 (85.1) | 76,613 (85.3) |
Acute myocardial infarction | 2,391 (2.0) | 2,343 (2.0) | 2,276 (2.0) | 2,302 (2.0) | 1,809 (2.0) | 1,822 (2.0) |
Coronary atherosclerosis | 39,975 (33.3) | 39,324 (32.7) | 37,335 (32.8) | 37,264 (32.7) | 29,675 (33.0) | 29,787 (33.2) |
Heart failure | 14,522 (12.1) | 14,263 (11.9) | 16,605 (14.6) | 16,705 (14.7) | 11,689 (13.0) | 11,745 (13.1) |
Atrial fibrillation | 15,248 (12.7) | 14,947 (12.4) | 14,885 (13.1) | 14,896 (13.1) | 11,468 (12.8) | 11,558 (12.9) |
Ischemic stroke | 14,081 (11.7) | 13,999 (11.7) | 13,626 (12.0) | 13,620 (12.0) | 10,449 (11.6) | 10,426 (11.6) |
Peripheral artery disease | 15,312 (12.7) | 15,226 (12.7) | 16,142 (14.2) | 16,147 (14.2) | 11,741 (13.1) | 11,873 (13.2) |
Chronic obstructive pulmonary disease | 16,086 (13.4) | 16,118 (13.4) | 17,604 (15.5) | 17,893 (15.7) | 12,832 (14.3) | 12,820 (14.3) |
Acute kidney injury | 3,627 (3.0) | 3,625 (3.0) | 5,913 (5.2) | 5,945 (5.2) | 3,196 (3.6) | 3,247 (3.6) |
Chronic kidney disease | 17,819 (14.8) | 17,701 (14.7) | 28,137 (24.7) | 28,128 (24.7) | 15,977 (17.8) | 15,980 (17.8) |
Urinary tract infection | 18,254 (15.2) | 18,310 (15.2) | 21,494 (18.9) | 21,505 (18.9) | 14,691 (16.4) | 14,727 (16.4) |
Osteoporosis without fracture | 9,092 (7.6) | 9,032 (7.5) | 9,098 (8.0) | 9,144 (8.0) | 6,734 (7.5) | 6,730 (7.5) |
Depression | 18,737 (15.6) | 18,753 (15.6) | 22,802 (20.0) | 22,947 (20.2) | 16,486 (18.4) | 16,435 (18.3) |
Dementia | 5,585 (4.7) | 5,672 (4.7) | 6,591 (5.8) | 6,509 (5.7) | 4,396 (4.9) | 4,382 (4.9) |
Diabetes treatment at cohort entry, n (%) | ||||||
Metformin | 76,879 (64.0) | 77,331 (64.3) | 62,042 (54.5) | 62,505 (54.9) | 53,207 (59.2) | 53,438 (59.5) |
Sulfonylurea | 41,883 (34.8) | 42,453 (35.3) | 36,578 (32.1) | 37,331 (32.8) | 29,558 (32.9) | 29,469 (32.8) |
Glitazone | 8,134 (6.8) | 8,150 (6.8) | 7,375 (6.5) | 7,539 (6.6) | 6,250 (7.0) | 6,227 (6.9) |
Insulin | 22,036 (18.3) | 21,837 (18.2) | 32,490 (28.5) | 32,181 (28.3) | 23,602 (26.3) | 23,700 (26.4) |
Other medication use, n (%) | ||||||
ACE inhibitors | 54,182 (45.1) | 54,146 (45.1) | 51,583 (45.3) | 51,713 (45.4) | 40,559 (45.1) | 40,515 (45.1) |
ARBs | 43,225 (36.0) | 43,184 (35.9) | 41,447 (36.4) | 41,330 (36.3) | 32,917 (36.6) | 33,001 (36.7) |
β-Blockers | 58,875 (49.0) | 58,670 (48.8) | 57,628 (50.6) | 57,532 (50.5) | 44,403 (49.4) | 44,540 (49.6) |
Calcium channel blockers | 40,590 (33.8) | 40,655 (33.8) | 39,710 (34.9) | 39,579 (34.8) | 30,300 (33.7) | 30,401 (33.8) |
Thiazides | 42,302 (35.2) | 42,514 (35.4) | 42,084 (37.0) | 42,170 (37.0) | 32,971 (36.7) | 32,816 (36.5) |
Loop diuretics | 21,310 (17.7) | 21,158 (17.6) | 27,490 (24.1) | 27,531 (24.2) | 18,466 (20.6) | 18,565 (20.7) |
Antiplatelets | 16,209 (13.5) | 15,955 (13.3) | 15,007 (13.2) | 14,966 (13.1) | 11,846 (13.2) | 11,912 (13.3) |
Statins | 93,938 (78.2) | 93,757 (78.0) | 88,571 (77.8) | 88,459 (77.7) | 70,183 (78.1) | 70,176 (78.1) |
Measures of health care use, n (%) | ||||||
Emergency department visits | 33,629 (28.0) | 34,050 (28.3) | 36,317 (31.9) | 36,658 (32.2) | 26,628 (29.6) | 26,562 (29.6) |
Hospitalization 30 days prior to cohort entry | 887 (0.7) | 859 (0.7) | 811 (0.7) | 826 (0.7) | 586 (0.7) | 598 (0.8) |
Internist visit 30 days prior to cohort entry | 75,269 (62.6) | 75,347 (62.7) | 67,919 (59.7) | 68,870 (60.5) | 54,808 (61.0) | 53,839 (59.9) |
Cardiologist visit 30 days prior to cohort entry | 11,742 (9.8) | 11,235 (9.4) | 9,120 (8.0) | 9,123 (8.0) | 7,951 (8.9) | 7,860 (8.8) |
Endocrinologist visit 30 days prior to cohort entry | 13,388 (11.1) | 12,689 (10.6) | 17,181 (15.1) | 16,389 (14.4) | 12,944 (14.4) | 13,129 (14.6) |
Glycated hemoglobin testing | 116,395 (96.8) | 116,278 (96.7) | 110,217 (96.8) | 110,114 (96.7) | 87,165 (97.0) | 87,057 (96.9) |
Creatinine testing | 13,266 (11.0) | 13,276 (11.0) | 14,053 (12.3) | 14,220 (12.5) | 10,480 (11.7) | 10,466 (11.7) |
Lipid testing | 109,686 (91.3) | 109,252 (90.9) | 102,315 (89.9) | 102,178 (89.7) | 81,379 (90.6) | 81,232 (90.4) |
Microalbuminuria testing | 72,352 (60.2) | 72,294 (60.1) | 68,747 (60.4) | 68,539 (60.2) | 54,894 (61.1) | 55,058 (61.3) |
Electrocardiogram | 60,255 (50.1) | 59,537 (49.5) | 57,709 (50.7) | 57,713 (50.7) | 45,335 (50.5) | 44,869 (49.9) |
Echocardiogram | 33,800 (28.1) | 33,551 (27.9) | 32,999 (29.0) | 33,258 (29.2) | 25,626 (28.5) | 25,679 (28.6) |
ARB, angiotensin receptor blocker; CFI, claims-based frailty index; DPP-4i, dipeptidyl peptidase 4 inhibitor; GLP-1RA, glucagon-like peptide 1 receptor agonist; SGLT-2i, sodium–glucose cotransporter 2 inhibitor.
Comparative Cardiovascular Effectiveness and Safety by Frailty Status
SGLT-2is Versus DPP-4is
Over a mean follow-up time on treatment of 10.6 months (SD 11.3 months), the overall incidence rate of the primary cardiovascular effectiveness outcome per 1,000 person-years was 34.3 for SGLT-2i initiators and 47.7 for DPP-4i initiators (HR 0.72 [95% CI 0.69–0.75], IRD −13.35 [95% CI −15.06 to −11.64]). There were larger absolute rate reductions among frailer people, with the IRD ranging from −6.74 (95% CI −8.61 to −4.87) for nonfrail, to −16.58 (95% CI −19.26 to −13.90) for prefrail, to −27.24 (95% CI −41.64 to −12.84) for frail people (P for heterogeneity < 0.01). The NNT was 159, 66, and 39, respectively (Table 2). On the relative rate scale, there was no evidence for effect heterogeneity among the frailty strata. Cumulative incidence plots of the primary effectiveness outcome were consistent with these findings across all frailty strata (Fig. 1A and B and Supplementary Fig. 6).
Outcome . | Overall . | Not frail (CFI <0.15) . | Prefrail (CFI 0.15–0.24) . | Frail (CFI ≥0.25) . | P for heterogeneity . | ||||
---|---|---|---|---|---|---|---|---|---|
SGLT-2i vs. DPP-4i | SGLT-2i (n = 120,202) | DPP-4i (n = 120,202) | SGLT-2i (n = 50,843) | DPP-4i (n = 50,843) | SGLT-2i (n = 63,649) | DPP-4i (n = 63,649) | SGLT-2i (n = 5,710) | DPP-4i (n = 5,710) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 3,447 | 5,393 | 865 | 1,276 | 2,131 | 3,426 | 451 | 691 | |
Person-years | 100,452 | 113,146 | 46,180 | 50,102 | 50,167 | 58,004 | 4,105 | 5,040 | |
Event rate per 1,000 person-years (95% CI) | 34.31 (33.18 to 35.47) | 47.66 (46.40 to 48.95) | 18.73 (17.52 to 20.02) | 25.47 (24.11 to 26.91) | 42.48 (40.71 to 44.32) | 59.06 (57.12 to 61.07) | 109.86 (100.17 to 120.48) | 137.10 (127.25 to 147.71) | |
HR (95% CI) | 0.72 (0.69 to 0.75) | 0.74 (0.68 to 0.81) | 0.72 (0.69 to 0.76) | 0.79 (0.70 to 0.89) | 0.266 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −13.35 (−15.06 to −11.64) | −6.74 (−8.61 to −4.87) | −16.58 (−19.26 to −13.90) | −27.24 (−41.64 to −12.84) | <0.001 | ||||
NNT at 1 year | 80 | 159 | 66 | 39 | |||||
Primary safety outcome | |||||||||
Events, n | 3,642 | 5,064 | 972 | 1,310 | 2,234 | 3,164 | 436 | 590 | |
Person-years | 99,609 | 111,835 | 45,987 | 49,715 | 49,623 | 57,229 | 3,999 | 4,891 | |
Event rate per 1,000 person-years (95% CI) | 36.56 (35.39 to 37.77) | 45.28 (44.05 to 46.54) | 21.14 (19.85 to 22.51) | 26.35 (24.96 to 27.82) | 45.02 (43.19 to 46.93) | 55.29 (53.40 to 57.25) | 109.03 (99.26 to 119.76) | 120.62 (111.27 to 130.76) | |
HR (95% CI) | 0.81 (0.77 to 0.84) | 0.80 (0.74 to 0.87) | 0.81 (0.77 to 0.86) | 0.89 (0.78 to 1.01) | 0.258 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −8.72 (−10.44 to −7.00) | −5.21 (−7.16 to −3.26) | −10.27 (−12.95 to −7.59) | −11.59 (−25.71 to 2.53) | 0.004 | ||||
NNT at 1 year | 123 | 197 | 103 | 83 | |||||
GLP-1RA vs. DPP-4i | GLP-1RA (n = 113,864) | DPP-4i (n = 113,864) | GLP-1RA (n = 39,675) | DPP-4i (n = 39,675) | GLP-1RA (n = 65,715) | DPP-4i (n = 65,715) | GLP-1RA (n = 8,474) | DPP-4i (n = 8,474) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 4,178 | 6,450 | 726 | 1,085 | 2,612 | 4,144 | 840 | 1,221 | |
Person-years | 94,836 | 108,316 | 35,506 | 39,492 | 52,881 | 61,005 | 6,448 | 7,820 | |
Event rate per 1,000 person-years (95% CI) | 44.06 (42.74 to 45.42) | 59.55 (58.11 to 61.02) | 20.45 (19.02 to 21.99) | 27.47 (25.88 to 29.15) | 49.39 (47.53 to 51.32) | 67.93 (65.89 to 70.03) | 130.27 (121.75 to 139.38) | 156.15 (147.63 to 165.16) | |
HR (95% CI) | 0.74 (0.71 to 0.77) | 0.75 (0.68 to 0.82) | 0.73 (0.70 to 0.77) | 0.83 (0.76 to 0.91) | 0.027 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −15.49 (−17.46 to −13.52) | −7.02 (−9.23 to −4.81) | −18.54 (−21.34 to −15.74) | −25.88 (−38.30 to −13.46) | <0.001 | ||||
NNT at 1 year | 69 | 162 | 60 | 42 | |||||
Primary safety outcome | |||||||||
Events, n | 4,858 | 6,135 | 922 | 1,149 | 3,047 | 3,947 | 889 | 1,039 | |
Person-years | 93,526 | 106,732 | 35,200 | 39,143 | 52,060 | 60,083 | 6,266 | 7,506 | |
Event rate per 1,000 person-years (95% CI) | 51.94 (50.50 to 53.42) | 57.48 (56.06 to 58.94) | 26.19 (24.55 to 27.94) | 29.35 (27.70 to 31.10) | 58.53 (56.49 to 60.65) | 65.69 (63.67 to 67.77) | 141.87 (132.84 to 151.51) | 138.42 (130.25 to 147.10) | |
HR (95% CI) | 0.90 (0.87 to 0.94) | 0.89 (0.82 to 0.97) | 0.89 (0.85 to 0.93) | 1.01 (0.92 to 1.10) | 0.021 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −5.54 (−7.59 to −3.49) | −3.16 (−5.56 to −0.76) | −7.16 (−10.08 to −4.24) | 3.45 (−9.11 to 16.01) | 0.048 | ||||
NNT at 1 year | 187 | 339 | 152 | 804 | |||||
SGLT-2i vs. GLP-1RA | SGLT-2i (n = 89,865) | GLP-1RA (n = 89,865) | SGLT-2i (n = 34,496) | GLP-1RA (n = 34,496) | SGLT-2i (n = 50,417) | GLP-1RA (n = 50,417) | SGLT-2i (n = 4,952) | GLP-1RA (n = 4,952) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 2,627 | 2,839 | 574 | 586 | 1,676 | 1,826 | 377 | 427 | |
Person-years | 72,421 | 71,746 | 30,362 | 29,571 | 38,556 | 38,643 | 3,503 | 3,531 | |
Event rate per 1,000 person-years (95% CI) | 36.27 (34.91 to 37.68) | 39.57 (38.14 to 41.05) | 18.91 (17.42 to 20.52) | 19.82 (18.28 to 21.49) | 43.47 (41.44 to 45.60) | 47.25 (45.13 to 49.47) | 107.61 (97.28 to 119.04) | 120.91 (109.97 to 132.94) | |
HR (95% CI) | 0.92 (0.87 to 0.97) | 0.96 (0.85 to 1.07) | 0.92 (0.86 to 0.98) | 0.89 (0.78 to 1.02) | 0.744 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −3.30 (−5.31 to −1.29) | −0.91 (−3.14 to 1.32) | −3.78 (−6.78 to −0.78) | −13.30 (−29.10 to 2.50) | 0.067 | ||||
NNT at 1 year | 343 | 1,425 | 289 | 86 | |||||
Primary safety outcome | |||||||||
Events, n | 2,944 | 3,223 | 692 | 736 | 1,858 | 2,022 | 394 | 465 | |
Person-years | 71,707 | 70,954 | 30,231 | 29,347 | 38,081 | 38,144 | 3,394 | 3,463 | |
Event rate per 1,000 person-years (95% CI) | 41.06 (39.60 to 42.57) | 45.42 (43.88 to 47.02) | 22.89 (21.25 to 24.66) | 25.08 (23.33 to 26.96) | 48.79 (46.62 to 51.06) | 53.01 (50.75 to 55.37) | 116.08 (105.17 to 128.13) | 134.29 (122.62 to 147.07) | |
HR (95% CI) | 0.91 (0.86 to 0.95) | 0.92 (0.83 to 1.02) | 0.92 (0.87 to 0.98) | 0.86 (0.75 to 0.99) | 0.708 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −4.36 (−6.52 to −2.20) | −2.19 (−4.68 to 0.30) | −4.22 (−7.42 to −1.02) | −18.21 (−34.95 to −1.47) | 0.086 | ||||
NNT at 1 year | 257 | 532 | 250 | 60 |
Outcome . | Overall . | Not frail (CFI <0.15) . | Prefrail (CFI 0.15–0.24) . | Frail (CFI ≥0.25) . | P for heterogeneity . | ||||
---|---|---|---|---|---|---|---|---|---|
SGLT-2i vs. DPP-4i | SGLT-2i (n = 120,202) | DPP-4i (n = 120,202) | SGLT-2i (n = 50,843) | DPP-4i (n = 50,843) | SGLT-2i (n = 63,649) | DPP-4i (n = 63,649) | SGLT-2i (n = 5,710) | DPP-4i (n = 5,710) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 3,447 | 5,393 | 865 | 1,276 | 2,131 | 3,426 | 451 | 691 | |
Person-years | 100,452 | 113,146 | 46,180 | 50,102 | 50,167 | 58,004 | 4,105 | 5,040 | |
Event rate per 1,000 person-years (95% CI) | 34.31 (33.18 to 35.47) | 47.66 (46.40 to 48.95) | 18.73 (17.52 to 20.02) | 25.47 (24.11 to 26.91) | 42.48 (40.71 to 44.32) | 59.06 (57.12 to 61.07) | 109.86 (100.17 to 120.48) | 137.10 (127.25 to 147.71) | |
HR (95% CI) | 0.72 (0.69 to 0.75) | 0.74 (0.68 to 0.81) | 0.72 (0.69 to 0.76) | 0.79 (0.70 to 0.89) | 0.266 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −13.35 (−15.06 to −11.64) | −6.74 (−8.61 to −4.87) | −16.58 (−19.26 to −13.90) | −27.24 (−41.64 to −12.84) | <0.001 | ||||
NNT at 1 year | 80 | 159 | 66 | 39 | |||||
Primary safety outcome | |||||||||
Events, n | 3,642 | 5,064 | 972 | 1,310 | 2,234 | 3,164 | 436 | 590 | |
Person-years | 99,609 | 111,835 | 45,987 | 49,715 | 49,623 | 57,229 | 3,999 | 4,891 | |
Event rate per 1,000 person-years (95% CI) | 36.56 (35.39 to 37.77) | 45.28 (44.05 to 46.54) | 21.14 (19.85 to 22.51) | 26.35 (24.96 to 27.82) | 45.02 (43.19 to 46.93) | 55.29 (53.40 to 57.25) | 109.03 (99.26 to 119.76) | 120.62 (111.27 to 130.76) | |
HR (95% CI) | 0.81 (0.77 to 0.84) | 0.80 (0.74 to 0.87) | 0.81 (0.77 to 0.86) | 0.89 (0.78 to 1.01) | 0.258 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −8.72 (−10.44 to −7.00) | −5.21 (−7.16 to −3.26) | −10.27 (−12.95 to −7.59) | −11.59 (−25.71 to 2.53) | 0.004 | ||||
NNT at 1 year | 123 | 197 | 103 | 83 | |||||
GLP-1RA vs. DPP-4i | GLP-1RA (n = 113,864) | DPP-4i (n = 113,864) | GLP-1RA (n = 39,675) | DPP-4i (n = 39,675) | GLP-1RA (n = 65,715) | DPP-4i (n = 65,715) | GLP-1RA (n = 8,474) | DPP-4i (n = 8,474) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 4,178 | 6,450 | 726 | 1,085 | 2,612 | 4,144 | 840 | 1,221 | |
Person-years | 94,836 | 108,316 | 35,506 | 39,492 | 52,881 | 61,005 | 6,448 | 7,820 | |
Event rate per 1,000 person-years (95% CI) | 44.06 (42.74 to 45.42) | 59.55 (58.11 to 61.02) | 20.45 (19.02 to 21.99) | 27.47 (25.88 to 29.15) | 49.39 (47.53 to 51.32) | 67.93 (65.89 to 70.03) | 130.27 (121.75 to 139.38) | 156.15 (147.63 to 165.16) | |
HR (95% CI) | 0.74 (0.71 to 0.77) | 0.75 (0.68 to 0.82) | 0.73 (0.70 to 0.77) | 0.83 (0.76 to 0.91) | 0.027 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −15.49 (−17.46 to −13.52) | −7.02 (−9.23 to −4.81) | −18.54 (−21.34 to −15.74) | −25.88 (−38.30 to −13.46) | <0.001 | ||||
NNT at 1 year | 69 | 162 | 60 | 42 | |||||
Primary safety outcome | |||||||||
Events, n | 4,858 | 6,135 | 922 | 1,149 | 3,047 | 3,947 | 889 | 1,039 | |
Person-years | 93,526 | 106,732 | 35,200 | 39,143 | 52,060 | 60,083 | 6,266 | 7,506 | |
Event rate per 1,000 person-years (95% CI) | 51.94 (50.50 to 53.42) | 57.48 (56.06 to 58.94) | 26.19 (24.55 to 27.94) | 29.35 (27.70 to 31.10) | 58.53 (56.49 to 60.65) | 65.69 (63.67 to 67.77) | 141.87 (132.84 to 151.51) | 138.42 (130.25 to 147.10) | |
HR (95% CI) | 0.90 (0.87 to 0.94) | 0.89 (0.82 to 0.97) | 0.89 (0.85 to 0.93) | 1.01 (0.92 to 1.10) | 0.021 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −5.54 (−7.59 to −3.49) | −3.16 (−5.56 to −0.76) | −7.16 (−10.08 to −4.24) | 3.45 (−9.11 to 16.01) | 0.048 | ||||
NNT at 1 year | 187 | 339 | 152 | 804 | |||||
SGLT-2i vs. GLP-1RA | SGLT-2i (n = 89,865) | GLP-1RA (n = 89,865) | SGLT-2i (n = 34,496) | GLP-1RA (n = 34,496) | SGLT-2i (n = 50,417) | GLP-1RA (n = 50,417) | SGLT-2i (n = 4,952) | GLP-1RA (n = 4,952) | |
Primary cardiovascular effectiveness outcome | |||||||||
Events, n | 2,627 | 2,839 | 574 | 586 | 1,676 | 1,826 | 377 | 427 | |
Person-years | 72,421 | 71,746 | 30,362 | 29,571 | 38,556 | 38,643 | 3,503 | 3,531 | |
Event rate per 1,000 person-years (95% CI) | 36.27 (34.91 to 37.68) | 39.57 (38.14 to 41.05) | 18.91 (17.42 to 20.52) | 19.82 (18.28 to 21.49) | 43.47 (41.44 to 45.60) | 47.25 (45.13 to 49.47) | 107.61 (97.28 to 119.04) | 120.91 (109.97 to 132.94) | |
HR (95% CI) | 0.92 (0.87 to 0.97) | 0.96 (0.85 to 1.07) | 0.92 (0.86 to 0.98) | 0.89 (0.78 to 1.02) | 0.744 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −3.30 (−5.31 to −1.29) | −0.91 (−3.14 to 1.32) | −3.78 (−6.78 to −0.78) | −13.30 (−29.10 to 2.50) | 0.067 | ||||
NNT at 1 year | 343 | 1,425 | 289 | 86 | |||||
Primary safety outcome | |||||||||
Events, n | 2,944 | 3,223 | 692 | 736 | 1,858 | 2,022 | 394 | 465 | |
Person-years | 71,707 | 70,954 | 30,231 | 29,347 | 38,081 | 38,144 | 3,394 | 3,463 | |
Event rate per 1,000 person-years (95% CI) | 41.06 (39.60 to 42.57) | 45.42 (43.88 to 47.02) | 22.89 (21.25 to 24.66) | 25.08 (23.33 to 26.96) | 48.79 (46.62 to 51.06) | 53.01 (50.75 to 55.37) | 116.08 (105.17 to 128.13) | 134.29 (122.62 to 147.07) | |
HR (95% CI) | 0.91 (0.86 to 0.95) | 0.92 (0.83 to 1.02) | 0.92 (0.87 to 0.98) | 0.86 (0.75 to 0.99) | 0.708 | ||||
Difference in event rate per 1,000 person-years (95% CI) | −4.36 (−6.52 to −2.20) | −2.19 (−4.68 to 0.30) | −4.22 (−7.42 to −1.02) | −18.21 (−34.95 to −1.47) | 0.086 | ||||
NNT at 1 year | 257 | 532 | 250 | 60 |
The primary cardiovascular effectiveness outcome was time to the first occurrence of a composite of hospitalization for acute myocardial infarction or ischemic stroke, HHF, or all-cause mortality. The primary safety outcome was time to the first hospitalization for acute kidney injury, occurrence of hospital or emergency department admission for hypoglycemia, DKA, severe genital or urinary tract infection, lower-limb amputation, nonvertebral fracture, acute pancreatitis, or nonmalignant biliary event.
For secondary effectiveness outcomes, the initiation of SGLT-2is was associated with lower overall rates of acute myocardial infarction, HHF, all-cause mortality, and a trend toward a reduced rate of stroke. There was evidence for effect heterogeneity by frailty on both relative and absolute rate scales for HHF and all-cause mortality, with larger absolute rate reductions among frail people (Supplementary Table 13).
The overall incidence rate of the primary safety outcome was 36.6 for SGLT-2i initiators and 45.3 for DPP-4i initiators (HR 0.81 [95% CI 0.77–0.84], IRD −8.72 [95% CI −10.44 to −7.00]), with larger absolute rate reductions for frail people (P for heterogeneity < 0.01). The NNT was 197 for nonfrail, 103 for prefrail, and 83 for frail people (Table 2). Cumulative incidences of the primary safety outcome were consistent with these findings across all frailty strata (Fig. 2A and B and Supplementary Fig. 7).
For secondary safety outcomes, SGLT-2i initiators were associated with lower overall rates of acute kidney injury, hypoglycemia, severe urinary tract infection, and nonvertebral fracture compared with DPP-4i initiators. However, overall rates of DKA, genital infections, and lower-limb amputations associated with the initiation of SGLT-2is versus DPP-4is were higher. We found relevant effect heterogeneity by frailty for genital infections with a larger rate increase on the absolute scale among frail people (Supplementary Table 14).
GLP-1RAs Versus DPP-4is
Over a mean follow-up time on treatment of 10.7 months (SD 11.5 months), the overall incidence rate of the primary effectiveness outcome was 44.1 for GLP-1RA initiators and 59.6 for DPP-4i initiators (HR 0.74 [95% CI 0.71–0.77], IRD −15.49 [95% CI −17.46 to –13.52]). There was effect heterogeneity by frailty status on both relative (P for heterogeneity = 0.03) and absolute scales, with an IRD of −7.02 (95% CI −9.23 to –4.81) for nonfrail, −18.54 (95% CI −21.34 to –15.74) for prefrail, and −25.88 (95% CI −38.30 to –13.46) for frail people (P for heterogeneity < 0.01). The NNT was 162, 60, and 42, respectively (Table 2). Cumulative incidences were consistent with these findings across all frailty strata (Fig. 1C and D and Supplementary Fig. 8).
For secondary effectiveness outcomes, the initiation of GLP-1RAs was associated with lower rates of acute myocardial infarction, ischemic stroke, HHF, and all-cause mortality. Frailty was an effect modifier for HHF and all-cause mortality, with larger absolute rate reductions among frail people (Supplementary Table 13).
The overall rate of the primary safety outcome was 51.9 for GLP-1RA initiators and 57.5 for DPP-4i initiators (HR 0.90 [95% CI 0.87–0.94], IRD −5.54 [95% CI −7.59 to –3.49]), with rate reductions for nonfrail and prefrail people but a neutral finding for those with frailty (Table 2). Cumulative incidences of the primary safety outcome were consistent with these findings across all frailty strata (Fig. 2C and D and Supplementary Fig. 9). For secondary safety outcomes, compared with DPP-4i initiators, GLP-1RA initiators had lower overall rates of acute kidney injury and hypoglycemia, while there were no differential rates of DKA, acute pancreatitis, and biliary events with no evidence for effect heterogeneity by frailty (Supplementary Table 14).
SGLT-2is Versus GLP-1RAs
Over a mean follow-up time on treatment of 9.6 months (SD 10.3 months), the overall incidence rate of the primary effectiveness outcome was 36.3 for SGLT-2i initiators and 39.6 for GLP-1RA initiators (HR 0.92 [95% CI 0.87–0.97], IRD −3.30 [95% CI −5.31 to –1.29]), with no evidence of effect heterogeneity by frailty on either relative or absolute scale (Table 2 and Supplementary Fig. 10). For secondary outcomes, the use of SGLT-2is was associated with a lower risk for HHF, with the largest absolute rate reduction among frail people (Supplementary Table 13).
The overall incidence rate of the primary safety outcome was 41.1 for SGLT-2i initiators and 45.4 for GLP-1RA users (HR 0.91 [95% CI 0.86–0.95], IRD −4.36 [95% CI −6.52 to –2.20]) with no effect heterogeneity by frailty on either scale (Table 2 and Supplementary Fig. 11). For secondary safety outcomes, SGLT-2i initiators were associated with lower overall rates of acute kidney injury and hypoglycemia, while they had higher rates of DKA and genital infections. Frailer people had a larger rate increase for genital infections on the absolute scale (Supplementary Table 14).
Sensitivity and Subgroup Analyses
Overall, results for effectiveness and safety outcomes remained consistent across all sensitivity analyses performed (Supplementary Tables 15–19). A detailed description of the results is provided in the Supplementary Information on Results of Sensitivity Analyses.
Subgroup analyses by history of cardiovascular disease showed that compared with DPP-4is or GLP-1RAs among those initiating an SGLT-2i, the relative and absolute effectiveness benefits were larger in patients with cardiovascular disease. Again, absolute effectiveness differences increased with higher severity of frailty (Supplementary Tables 15–16). Subgroup analyses by age showed larger absolute benefits among people aged ≥75 years, though even in younger patients, there was evidence of a trend toward effectiveness benefit with increasing severity of frailty. For the primary safety outcome, absolute rate reductions were largest among frail patients aged ≥75 years with SGLT-2i use (Supplementary Table 18).
Conclusions
This cohort study of >740,000 Medicare beneficiaries with T2D found that both SGLT-2is and GLP-1RAs were associated with a greater cardiovascular effectiveness than DPP-4is. While relative risk reductions for the effectiveness outcomes remained stable across frailty strata, absolute reductions were larger in frailer patients at higher risk for these outcomes. Severe adverse events, although more frequent in frailer patients, were not more common in those using SGLT-2is or GLP-1RAs compared with DPP-4is.
The consideration of frailty for diabetes care management has not yet become routine in clinical practice (25). Because many clinical trials have not focused on participants with T2D who are older and frail, data on effectiveness and safety of newer antidiabetic medications are still scant in these clinically complex populations, and it is unclear whether therapeutic recommendations should be adapted accordingly (26,27).
In patients with heart failure, however, two recent studies have evaluated the efficacy (mainly driven by the incidence of acute heart failure) and safety of dapagliflozin according to frailty and found a sustained effect across the spectrum of frailty with more pronounced absolute risk reductions in frailer patients (28,29). Of note, despite the lower proportion of people with T2D in the DAPA-HF (Study to Evaluate the Effect of Dapagliflozin on the Incidence of Worsening Heart Failure or Cardiovascular Death in Patients With Chronic Heart Failure) (30) and DELIVER (Dapagliflozin Evaluation to Improve the Lives of Patients With Preserved Ejection Fraction Heart Failure) (31) trial, >70% of the most frail patients had T2D (28,29). Nonetheless, these findings derived from patients selected through narrow eligibility criteria are not powered to assess less frequent safety events and do not provide sufficient information generalizable to a population with T2D in clinical practice.
In this study with large representation of prefrail (n >350,000) and frail (n >35,000) people, we observed evidence of treatment effect modification by frailty status on an absolute rate scale. Specifically, we found that frailer people experienced larger benefits from SGLT-2i or GLP-1RA treatment than those without frailty, as explicitly shown by the NNT for the primary effectiveness outcome (SGLT-2i vs. DPP-4i cohort: 39 [frail] vs. 159 [not frail] after 1 year; GLP-1RA vs. DPP-4i cohort: 42 [frail] vs. 162 [not frail]). Given the stable HRs across the spectrum of frailty, these differences in NNT between frail and nonfrail people are primarily driven by the higher number of events among frail patients, which highlights the greater vulnerability of this population. Accordingly, the number of severe adverse events was also higher in frail people. A potential explanation for the heterogeneity in number of events among the frailty strata could also be based on the difference in mean baseline comorbidity score. This is a likely scenario, since frailty not only is a critical contributing factor to the complexity of care for aging populations but also describes a dynamic state of vulnerability to stressors resulting from decreased physiological reserve and cumulating comorbidity.
Our results for secondary effectiveness and safety outcomes in the overall cohort were mostly consistent with findings from previous clinical trials, although these trials were comparing active drugs with placebo. In agreement with a recent meta-analysis of five placebo-controlled, double-blind outcome trials of SGLT-2is in patients with T2D (32), we found lower rates of acute myocardial infarction, HHF, and all-cause mortality associated with the use of SGLT-2is versus DPP-4is.
Regarding safety outcomes, we observed higher rates of DKA, genital infections, and lower-limb amputations among SGLT-2i users, which was also found in a recent meta-analysis (32). For genital infections, there was evidence of effect modification by frailty, with larger harm among more frail people. While the rate increase for DKA and genital infection is known for all approved SGLT-2is, it still remains unclear whether the association with lower-limb amputations is caused by the class of SGLT-2is per se or canagliflozin only (33). In contrast to many clinical trials, we included a larger and more high-risk study population, which could explain this association for lower-limb amputations. Event rates for hypoglycemia and severe urinary tract infections were lower in patients initiating SGLT-2is and in line with findings from previous observational studies (34,35).
For people using GLP-1RAs, our results for all effectiveness outcomes were consistent with a meta-analysis of seven GLP-1RA trials in patients with T2D, even though point estimates were closer to the null in clinical trials (36), in particular for HHF. However, findings from the recent GRADE (A Comparative Effectiveness Study of Major Glycemia-Lowering Medications for Treatment of Type 2 Diabetes) study also showed a strong beneficial effect of liraglutide versus an active comparator regarding HHF, with a relative risk reduction of 51% (37). For HHF and all-cause mortality, we observed effect modification by frailty with a larger rate reduction in more frail people. For safety outcomes, we did not find any concerns, overall and by frailty status, when comparing GLP-1RAs with DPP-4is. Explicitly, there was no rate increase for hypoglycemia, acute pancreatitis, or nonmalignant biliary events, in accordance with observational and clinical data, when comparing GLP-1RAs with placebo or an active comparator (36,38,39).
Compared with GLP-1RAs, the initiation of SGLT-2is was associated with a rate reduction of HHF, with larger benefit among frailer people on an absolute rate scale. Otherwise, effectiveness outcomes were comparable between the two groups. Although the overall safety profile was similar, SGLT-2i users had a lower rate of hospitalization for acute kidney injury but higher rates of DKA and genital infections compared with GLP-1RA users.
Given the broad patient spectrum in clinical routine settings, which includes older and frail patients typically excluded from clinical trials, and the robustness of our results shown by different sensitivity analyses, this nationwide study fills an important gap in the current evidence on the effectiveness and safety of newer antidiabetic medications in high-risk people with frailty. Providing a comprehensive analysis not only on effectiveness but also on safety outcomes, this study adds critical information across the spectrum of frailty that is needed in clinical decision making. While cardiovascular effectiveness might play a lesser role in people with a limited life expectancy, safety is a major consideration in deciding which antidiabetic medication might be most appropriate.
This study has limitations. First, although we used PS matching to adjust for a large number of baseline characteristics, residual confounding by unmeasured or not fully measured characteristics in claims (e.g., laboratory values, vital signs, diabetes duration, severity of disease, nonbiometric patient factors driving physician prescribing) cannot be entirely excluded. However, a previous new-user active comparator cohort study with use of claims data linked to electronic health records showed sufficient balance in many of these characteristics after adjustment for claims-based proxies of diabetes severity and duration (40). Second, the assessment of a safety composite outcome to provide an overall measure for severe adverse events could dilute the interpretation of safety outcomes. However, to avoid a misleading message, we addressed each safety outcome individually in secondary analyses. Third, we were unable to evaluate cardiovascular death because the Medicare database does not contain information on cause of death. Fourth, the mean follow-up time on treatment was shorter compared with cardiovascular outcome trials because this study was based on the routine use of SGLT-2is, GLP-1RAs, and DPP-4is. Nevertheless, all sensitivity analyses addressing potential informative censoring showed robust findings, and a time-stratified analysis found consistent findings among people who had ≥1 year of follow-up. Moreover, the generous size of our study population generated results with high precision, despite a shorter duration of time on treatment. Given no time-varying hazards as shown in the follow-up time-stratified analysis, these results should be generalizable to longer-term findings. Nevertheless, future long-term cardiovascular clinical trials in older people with T2D should prospectively assess frailty to confirm the risk-benefit profiles shown in this study.
In conclusion, in this population-based cohort study of patients with T2D receiving routine care, the initiation of an SGLT-2i or a GLP-1RA, compared with a DPP-4i, was associated with lower cardiovascular events and death without increasing the overall incidence rate of severe safety outcomes, regardless of frailty status. Importantly, absolute rate reductions for all effectiveness outcomes were larger with increasing severity of frailty. These data can inform clinical guidelines and support more individualized decision making in frail patients.
This article contains supplementary material online at https://doi.org/10.2337/figshare.23992563.
This article is featured in a podcast available at diabetesjournals.org/care/pages/diabetes_care_on_air.
Article Information
Funding. This study was supported by Patient-Centered Outcomes Research Institute grant DB-2020C2-20326 by the Brigham and Women’s Hospital and Harvard Medical School. A.K. was supported by Swiss National Science Foundation postdoctoral mobility grant P400PM_194479/1 and by a Novo Nordisk educational grant. E.P. was supported by National Institute on Aging career development grant K08AG055670.
Duality of Interest. D.H.K. has been supported by grants from the National Institute on Aging for unrelated work and receives personal fees from Alosa Health and VillageMD. D.J.W. serves on data monitoring committees for Novo Nordisk. S.S. is participating in investigator-initiated grants to the Brigham and Women’s Hospital from Boehringer Ingelheim and UCB unrelated to the topic of this study; is a consultant to Aetion Inc., a software manufacturer in which he owns equity; and is an advisor to Temedica GmbH, a patient-oriented data generation company. R.J.G. has been supported by the grants from AstraZeneca, Kowa, Novartis, and Pfizer not related to the topic of this work. E.P. is an investigator of a research grant to the Brigham and Women’s Hospital from Boehringer Ingelheim not related to the topic of this work and was supported by a research grant (5U01FD007213) from the U.S. Food and Drug Administration for unrelated work. No other potential conflicts of interest relevant to this article were reported.
Author Contributions. A.K. wrote the first draft of the manuscript. A.K., D.H.K., D.J.W., J.L., S.S., R.J.G., and E.P. contributed to the analysis and interpretation of data and critically revised the manuscript for important intellectual content and approved the final version of the manuscript. A.K., D.H.K., and E.P. were responsible for the concept and design of the study. A.K., J.L., R.J.G., and E.P. performed the statistical analysis. A.K. and E.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 data and the accuracy of the data analysis.
Prior Presentation. Parts of this study were presented in poster form at the 82nd Scientific Sessions of the American Diabetes Association, New Orleans, LA, 3–7 June 2022; in oral form at the 38th International Society for Pharmacoepidemiology Annual Conference, Copenhagen, Denmark, 24–28 August 2022; and in oral and podcast form at the 58th Annual Meeting of the European Association for the Study of Diabetes, Stockholm, Sweden, 19–23 September 2022.