Clinical Outcomes With Metformin and Sulfonylurea Therapies Among Patients With Heart Failure and Diabetes
The authors sought to characterize associations between initiation of metformin and sulfonylurea therapy and clinical outcomes among patients with comorbid heart failure (HF) and diabetes (overall and by ejection fraction [EF] phenotype).
Metformin and sulfonylureas are frequently prescribed to patients with diabetes for glycemic control. The impact of these therapies on clinical outcomes among patients with comorbid HF and diabetes is unclear.
The authors evaluated Medicare beneficiaries hospitalized for HF in the Get With The Guidelines–Heart Failure Registry between 2006 and 2014 with diabetes and not prescribed metformin or sulfonylurea before admission. In parallel separate analyses for metformin and sulfonylurea, patients with newly prescribed therapy within 90 days of discharge were compared with patients not prescribed therapy. Multivariable models landmarked at 90 days evaluated associations between prescription of therapy, and mortality and hospitalization for HF (HHF) at 12 months. Negative control (falsification) endpoints included hospitalization for urinary tract infection, hospitalization for gastrointestinal bleed, and influenza vaccination. Prespecified subgroup analyses were stratified by EF ≤40% versus >40%.
Of 5,852 patients, 454 (7.8%) were newly prescribed metformin and 504 (8.6%) were newly prescribed sulfonylurea. After adjustment, metformin prescription was independently associated with reduced risk of composite mortality/HHF (HR: 0.81; 95% CI: 0.67-0.98; P = 0.03), but individual components were not statistically significant. Findings among patients with EF >40% accounted for associations with mortality/HHF (HR: 0.68; 95% CI: 0.52-0.90) and HHF (HR: 0.58; 95% CI: 0.40-0.85) endpoints (all P for interaction ≤0.04). After adjustment, sulfonylurea initiation was associated with increased risk of mortality (HR: 1.24; 95% CI: 1.00-1.52; P = 0.045) and HHF (HR: 1.22; 95% CI: 1.00-1.48; P = 0.050) with nominal statistical significance. Associations between sulfonylurea initiation and endpoints were consistent regardless of EF (all P for interaction >0.11). Neither metformin initiation nor sulfonylurea initiation were associated with negative control endpoints.
In this population of older U.S. adults hospitalized for HF with comorbid diabetes, metformin initiation was independently associated with substantial improvements in 12-month clinical outcomes, driven by findings among patients with EF >40%. By contrast, sulfonylurea initiation was associated with excess risk of death and HF hospitalization, regardless of EF.
Several randomized clinical trials of antidiabetic therapies have shown varying effects on heart failure (HF) risk (1). For example, thiazolidinediones and the dipeptidyl peptidase-4 inhibitor saxagliptin have each been shown to increase risk for HF events (2-4). By contrast, sodium glucose co-transporter 2 (SGLT2) inhibitors are effective in preventing incident HF among patients with type 2 diabetes mellitus (DM), and in improving clinical outcomes among patients with established HF with reduced ejection fraction (HFrEF) or preserved ejection fraction (HFpEF) (5-7).
Despite mounting evidence supporting broad use of SGLT2 inhibitors, metformin and sulfonylureas continue to remain the most commonly used oral medications for DM, including among patients with comorbid HF (8). Likewise, although newer guidelines recognize the importance of SGLT2 inhibitor therapy for patients with type 2 DM (particularly in cases of comorbid HF or chronic kidney disease), they continue to endorse metformin as general first-line therapy because it is less costly, readily available, and widely used as baseline therapy in trials of other antidiabetic medications (9,10). Nonetheless, to date, the impact of both metformin and sulfonylurea therapies on cardiovascular and HF outcomes remains largely unknown. This uncertainty is particularly present for patients with established HF, a high-risk population for which safety with select oral antidiabetic therapies has historically proven a concern. Likewise, given differences in respective pathophysiology of HFrEF vs HFpEF, it is possible that differences in safety and cardiovascular risk reduction with metformin and sulfonylurea could vary by ejection fraction (EF).
With the increasing recognition SGLT2 inhibitors as an alternative oral medication proven to reduce cardiovascular risk, understanding the effects of metformin and sulfonylurea therapies on clinical outcomes among patients with HF and DM is increasingly relevant given their continued dominant use in routine practice. In this context, the GWTG-HF (Get With The Guidelines–Heart Failure) registry offers a novel opportunity to evaluate the comparative effectiveness of metformin and sulfonylurea therapies on clinical outcomes in a large nationally representative cohort of older U.S. adults hospitalized for HF with comorbid DM. In addition, the registry offers the ability to assess whether associations between use of each therapy and clinical outcomes differ among patients with reduced versus preserved EF.
GWTG-HF is an ongoing, observational, hospital-based, quality improvement registry formed in 2005 by the American Heart Association. The registry includes patients hospitalized for HF, and those who develop symptoms attributable to HF during hospitalization such that HF was the primary diagnosis. IQVIA serves as the data collection and coordinating center for the American Heart Association’s GWTG programs, and the Duke Clinical Research Institute serves as the data analytic center. As the primary purpose of the registry is for quality improvement, patient informed consent is waived under the Common Rule. The GWTG-HF protocol was approved by the institutional review boards at each participating center. For the current study, participants 65 years of age and older with fee-for-service Medicare coverage were linked to Medicare data using a validated technique (11).
We identified Medicare beneficiaries hospitalized for HF in the GWTG-HF registry between January 1, 2006, and December 31, 2014, who were discharged with a diagnosis or history of DM. To further confirm history of DM, a medication fill for any antidiabetic medication from January 1, 2006, to December 31, 2015, was additionally required for study inclusion. If multiple HF hospitalizations existed for a patient, the first hospitalization was considered the index hospitalization, and other hospitalizations were excluded.
Key reasons for patient exclusion from this study were: 1) left against medical advice, transferred to an acute care facility, or discharged to hospice care; 2) significant kidney disease, defined as chronic or in-hospital dialysis therapy, estimated glomerular filtration rate (eGFR) <45 mL/min/1.73 m2 at hospital discharge, or history of chronic kidney disease (CKD) with serum creatinine >2.0 mg/dL if eGFR data unavailable; and 3) died within 90 days of discharge (Supplemental Figure 1). The rationale for the kidney function exclusion stemmed from contraindications and cautions against use of metformin (among patients with eGFR <30 mL/min/1.73 m2 and 30 to 44 mL/min/1.73 m2, respectively) and recognition that patients with CKD who are potentially less likely to receive metformin may confound associations between metformin and clinical outcomes. Identical kidney function selection criteria were prespecified for use in the sulfonylurea initiation analyses to mitigate potential differences in confounding driving differences between metformin and sulfonylurea results. We also excluded patients who had a medication fill for thiazolidinedione therapy within 90 days of discharge, given the established adverse effects of thiazolidinedione medications among HF patients. In addition, to capture patients newly initiating metformin or sulfonylurea therapy and likely naive to either medication, our primary analysis excluded patients with prescription fills for either therapy in the 12 months before index hospitalization.
The effectiveness of metformin and sulfonylurea therapies were separately evaluated in identical parallel analyses. For each therapy, exposure was defined as a medication fill for the respective medication within 90 days post-discharge. Patients with a medication fill were then compared with patients who did not have a medication fill for the respective therapy.
Prespecified clinical outcomes were all-cause mortality, hospitalization for HF, and the composite of all-cause mortality or HF hospitalization. Outcomes were assessed over a 12-month period, beginning at day 91 post-discharge (ie, landmark analysis with outcome assessment period beginning immediately after the 90-day exposure period detailed in the preceding text). Under the landmark analysis approach, hospitalizations that occurred within 90 days postdischarge were excluded from outcome analyses. Outcomes were identified using the Medicare Master Beneficiary Summary File and inpatient claims files (Supplemental Table 1).
In separate analyses for metformin and sulfonylurea, baseline patient characteristics were compared between patients initiated with the respective therapy versus those not initiated. Groups were compared using standardized mean differences, with a difference >10 indicating imbalance between groups.
To align with the early postdischarge period used to define exposure to metformin or sulfonylurea therapy, all outcome analyses were landmarked at 90 days postdischarge. Thus, the 12-month outcome assessment period extended from 3 months to 15 months following index hospital discharge. The unadjusted cumulative incidence of each outcome over 12-month follow-up was compared among patients initiating vs not initiating the respective therapy.
For metformin and sulfonylurea respectively, unadjusted and adjusted Cox proportional hazards models were used to derive HRs and assess associations between initiation of the therapy and 12-month clinical outcomes. All outcomes were assessed as time-to-first event. To help address potential selection bias or confounding, adjusted models included 29 prespecified covariates chosen by clinical judgment that spanned patient demographics, EF, vital signs, laboratories, medical history, HF medical and device therapies at time of discharge, and time from index hospitalization discharge to metformin/sulfonylurea medication fill (Supplemental Table 2). All models were weighted by the inverse probability of receiving metformin or sulfonylurea. If the propensity scores were extreme, overlap weighting was used. All models accounted for clustering of patients within hospitals by using a robust sandwich variance via the generalized estimating equation method. Cause-specific hazards were used to account for competing risk of mortality. The proportional hazards assumption was verified for each cause-specific model using plots of Schoenfeld residuals and the log-cumulative hazard. In addition, for all clinical endpoints, associations between initiation of each treatment and outcomes were assessed among subgroups of patients with EF ≤40% and EF >40%, and interaction testing was performed.
To assess risk of residual confounding, unadjusted and adjusted models were used to separately test associations between initiation of metformin or sulfonylurea and 3 falsification endpoints (ie, negative controls) chosen based on the lack of biologically plausible associations with either metformin or sulfonylurea (12). Falsification endpoints were hospitalization for urinary tract infection, hospitalization for gastrointestinal bleed, and vaccination for influenza.
As sensitivity analyses, the preceding analyses (with the exception of the subgroup analyses by EF group) for metformin and sulfonylurea were each repeated in the larger population of patients including those with use of the respective medication in the prior 12 months. Thus, populations examined in the parallel sensitivity analyses were a combination of new and prevalent users of each medication.
Missing patient-level covariates were imputed before entering models (Supplemental Table 3). Missing discharge values were imputed with admission values when available, otherwise multiple imputation methods were used with 10 data sets. Hospital characteristics were not imputed. Rates of missing data for most variables were <8% with few exceptions (eGFR 31.1%, heart rate 20.5%, systolic blood pressure 19.2%). All statistical tests were 2-sided, with P < 0.05 considered statistically significant. All analyses were performed using SAS software version 9.4 (SAS Institute).
After all exclusions, 5,852 patients deemed naive to metformin and sulfonylurea therapy across 367 U.S. hospitals were included in the primary analysis (Supplemental Figure 1). Of 5,852 patients, 454 patients (7.8%) were newly prescribed metformin and 504 (8.6%) were newly prescribed sulfonylurea within 90 days of hospital discharge. Overall, 133 patients (2.3%) were newly initiated on both therapies. The median (interquartile range) age was 75 (69 to 82) years, 51.6% were women, 73.0% were White. Median ejection fraction was 45%, with 41.9% of patients having an EF ≤40%. Median eGFR at time of hospital discharge was 60 mL/min/1.73 m2.
Compared with patients not treated with metformin, patients initiated on metformin therapy tended to be younger, more likely to be treated with an angiotensin-converting enzyme inhibitor/angiotensin receptor blocker, and less likely to have ischemic heart disease and an implantable cardioverter-defibrillator or cardiac resynchronization therapy device (Table 1). Metformin initiators were also more likely to have received care at a hospital in the Western United States, and less likely to be treated with insulin. Otherwise, patients initiating and not initiating metformin were generally similar with respect to sex, household income, education level, EF, blood pressure, eGFR, natriuretic peptide level, hemoglobin A1c, and index hospital characteristics.
|Metformin (n = 454)||No Metformin (n = 5,398)||Absolute Std. Diff, %|
|Age, y||74 (68-81)||76 (70-82)||17.10|
|Female||243 (53.5)||2,776 (51.4)||4.20|
|White||320 (72.1)||3,846 (73.1)|
|Black||61 (13.7)||860 (16.3)|
|Other||63 (14.2)||554 (10.5)|
|Median household income (indexed using CPI-U 2014 average), $||51,186 (44,489-59,313)||51,186 (45,220-60,288)||3.02|
|LVEF ≤40%a||193 (43.7)||2,197 (41.8)||3.80|
|Ejection fraction||45 (30-56)||45 (30-58)||4.35|
|Length of stay, days||4 (3-6)||4 (3-6)||2.80|
|Vital sign and laboratory data at discharge|
|Heart rate, beats/min||78 (67-89)||74 (66-84)||22.17|
|Systolic blood pressure, mm Hg||124 (113-138)||126 (113-141)||2.99|
|Body mass index, kg/m2||30.0 (26.4-36.3)||30.1 (26.0-35.7)||4.61|
|Potassium, mEq/L||4.0 (3.7-4.3)||4.0 (3.7-4.3)||0.95|
|BUN, mg/dL||22 (17-28)||22 (17-30)||3.12|
|Serum Cr, mg/dL||1.0 (0.8-1.1)||1.1 (0.9-1.3)||9.41|
|eGFR, MDRD formula||67 (55-81)||60 (51-72)||3.97|
|BNP, pg/mL||459 (194-813)||434 (199-861)||8.36|
|NT-BNP, pg/mL||4,160 (1,734-6,140)||2,775 (1,270-6,428)||9.84|
|Hemoglobin A1C||7.1 (6.4-8.5)||7.1 (6.4-8.1)||9.53|
|Hypertension||380 (83.7)||4,484 (83.1)||1.66|
|Hyperlipidemia||79 (17.4)||777 (14.4)||3.44|
|Prior myocardial infarction||88 (19.4)||1,185 (22.0)||6.36|
|Prior cerebrovascular disease/TIA||66 (14.5)||997 (18.5)||10.62|
|Peripheral vascular disease||52 (11.5)||807 (15.0)||10.35|
|Smoker||68 (15.0)||519 (9.6)||16.37|
|Chronic obstructive pulmonary disease or asthma||149 (32.8)||1,763 (32.7)||0.33|
|Atrial fibrillation||138 (30.4)||1,850 (34.3)||8.31|
|Ischemic heart disease||249 (54.8)||3,471 (64.3)||19.38|
|ICD or CRT device||36 (7.9)||617 (11.4)||11.87|
|Heart failure medical therapy at discharge|
|ACE inhibitor/ARB||355 (78.7)||3,806 (71.8)||16.82|
|Beta-blocker||368 (81.1)||4,413 (82.2)||3.92|
|MRA||83 (18.7)||972 (18.6)||8.99|
|Hydralazine nitrate||27 (6.1)||612 (11.9)||22.36|
|Loop diuretic||253 (95.5)||2,854 (95.5)||0.06|
|Background diabetes therapies|
|Insulin||114 (25.1)||2,228 (41.3)||34.84|
|DPP-4 inhibitor||45 (9.9)||248 (4.6)||20.61|
|Sulfonylurea||133 (29.3)||271 (6.9)||60.90|
|Northeast||138 (30.4)||1,805 (33.4)|
|Midwest||96 (21.1)||1,351 (25.0)|
|South||155 (34.1)||1,779 (33.0)|
|West||65 (14.3)||463 (8.6)|
|Academic/teaching hospital||323 (73.4)||3,859 (74.1)||1.53|
|No. of beds||348 (226-483)||375 (243-556)||8.19|
|Rural location||20 (4.8)||261 (5.2)||1.94|
Compared with patients not treated with sulfonylurea, patients initiated on therapy tended to have modestly longer index hospital lengths of stay, lower body mass index, higher level of N-terminal pro–B-type natriuretic peptide, and lower rates of comorbid atrial fibrillation (Table 2). Sulfonylurea initiators were also more likely to have received care at a hospital in the Western United States, and less likely to be treated with insulin. Patients initiating and not initiating sulfonylurea were generally similar with respect to age, sex, household income, education level, EF, vital signs, renal function, and HF medications.
|Sulfonylurea (n = 504)||No Sulfonylurea (n = 5,348)||Absolute Std. Diff, %|
|Age, y||76 (70-83)||75 (69-82)||8.97|
|Sex, female||244 (48.4)||2,775 (51.9)||6.96|
|White||351 (71.3)||3,815 (73.2)|
|Black||71 (14.4)||850 (16.3)|
|Other||70 (14.2)||547 (10.5)|
|Median household income (indexed using CPI-U 2014 average), $||51,268 (44,024-61,120)||51,186 (45,220-60,225)||1.90|
|LVEF ≤40%||218 (44.5)||2,172 (41.7)||5.66|
|Ejection fractiona||45 (30-58)||47 (30-58)||8.61|
|Length of stay, days||5 (3-7)||4 (3-6)||13.36|
|Vital sign and laboratory data at discharge|
|Heart rate, beats/min||74 (66-84)||74 (66-84)||4.20|
|Systolic blood pressure, mm Hg||126 (111-142)||126 (113-141)||0.71|
|Body mass index, kg/m2||29.2 (25.4-34.0)||30.2 (26.1-36.0)||22.67|
|Potassium, mEq/L||4.0 (3.7-4.3)||4.0 (3.7-4.3)||1.28|
|BUN, mg/dL||23 (17-30)||22 (17-29)||2.53|
|Serum Cr, mg/dL||1.0 (0.8-1.2)||1.1 (0.9-1.2)||7.54|
|eGFR, MDRD formula||60 (52-73)||60 (52-72)||2.17|
|BNP, pg/mL||541 (229-947)||428 (197-843)||3.98|
|NT-BNP, pg/mL||3,706 (1,534-7,000)||2,781 (1,262-6,237)||14.54|
|Hemoglobin A1C||7.3 (6.4-8.5)||7.1 (6.4-8.1)||7.15|
|Hypertension||423 (83.9)||4,441 (83.1)||2.35|
|Hyperlipidemia||74 (14.7)||782 (14.6)||11.58|
|Prior myocardial infarction||96 (19.0)||1,177 (22.0)||7.35|
|Prior cerebrovascular accident/TIA||83 (16.5)||980 (18.3)||4.91|
|Peripheral vascular disease||68 (13.5)||791 (14.8)||3.74|
|Smoker||59 (11.7)||528 (9.9)||5.95|
|Chronic obstructive pulmonary disease or asthma||168 (33.3)||1,744 (32.6)||1.53|
|Atrial fibrillation||148 (29.4)||1,840 (34.4)||10.85|
|Ischemic heart disease||302 (59.9)||3,418 (63.9)||8.25|
|ICD or CRT device||48 (9.5)||605 (11.3)||5.87|
|Heart failure medical therapy at discharge|
|ACE inhibitor/ARB||375 (75.5)||3,786 (72.1)||7.74|
|Beta-blocker||400 (79.5)||4,381 (82.3)||7.48|
|MRA||104 (21.3)||951 (18.4)||8.20|
|Hydralazine nitrate||52 (10.9)||587 (11.5)||3.75|
|Loop diuretic||275 (95.5)||2,832 (95.5)||0.02|
|Background diabetes therapies|
|Insulin||121 (24.0)||2,221 (41.5)||38.00|
|DPP-4 inhibitor||42 (8.3)||251 (4.7)||14.79|
|Metformin||133 (26.4)||321 (6.0)||57.59|
|Northeast||163 (32.3)||1,780 (33.3)|
|Midwest||116 (23.0)||1,331 (24.9)|
|South||157 (31.2)||1,777 (33.2)|
|West||68 (13.5)||460 (8.6)|
|Academic/teaching hospital||356 (73.9)||3,826 (74.0)||0.43|
|No. of beds||367 (238-537)||375 (243-539)||2.35|
|Rural location||20 (4.4)||261 (5.3)||4.36|
Outcomes with metformin initiation
Unadjusted rates of HF hospitalization (19.5 vs 25.8; P = 0.005) and composite of all-cause mortality or HF hospitalization (30.7 vs 39.7; P = 0.004) at 12 months were significantly lower among patients who were prescribed metformin compared with those who were not (Table 3, Figure 1). There was no significant difference in the unadjusted risk of all-cause mortality between groups (18.5 vs 20.5; P = 0.24).
|Study Endpoint||Cumulative Incidence, %||HR (95% CI); P Value|
|Metformin (n = 454)||No Metformin (n = 5,398)||Unadjusted||Adjusteda|
|All-cause mortality||18.5||20.5||0.88 (0.70-1.10); 0.25||1.06 (0.85-1.33); 0.59|
|HF hospitalization||19.5||25.8||0.72 (0.58-0.89); 0.002||0.80 (0.63-1.02); 0.072|
|All-cause mortality/HF hospitalization||30.7||39.7||0.73 (0.61-0.87); 0.001||0.81 (0.67-0.98); 0.032|
|Falsification (negative control) endpoints|
|Urinary tract infection||1.4||2.8||0.48 (0.22-1.06); 0.068||0.67 (0.31-1.48); 0.33|
|Gastrointestinal bleed||1.2||2.4||0.46 (0.16-1.33); 0.15||0.69 (0.24-2.03); 0.50|
|Vaccination for influenza||9.5||10.7||0.87 (0.63-1.20); 0.39||0.90 (0.61-1.32); 0.59|
In adjusted analysis, metformin initiation remained significantly associated with lower risk of composite all-cause mortality or HF hospitalization at 12 months (HR: 0.81 [95% CI: 0.67-0.98]; P = 0.032). After adjustment, associations between metformin initiation and all-cause mortality (HR: 1.06 [95% CI: 0.85-1.33]; P = 0.59) and HF hospitalization (HR: 0.80 [95% CI: 0.63-1.02]; P = 0.072) were not statistically significant. Metformin initiation was not significantly associated with any falsification endpoint in unadjusted or adjusted analyses.
In subgroup analysis stratified by EF ≤40% vs >40%, the association between metformin initiation and risk of HF hospitalization (P for interaction = 0.007) and composite all-cause mortality or HF hospitalization (P for interaction = 0.040) was significantly different between EF groups. For both endpoints, associations with substantially reduced risk were confined to patients with EF >40% (HF hospitalization HR: 0.58 [95% CI: 0.40-0.85]; composite endpoint HR: 0.68 [95% CI: 0.52-0.90]), with no suggestion of reduced risk among patients with EF ≤40% (Central Illustration). There was no evidence for significant subgroup interaction (P for interaction = 0.28) for all-cause mortality.
Outcomes with sulfonylurea initiation
The unadjusted rates of HF hospitalization (27.1 vs 25.1; P = 0.35) and composite of all-cause mortality or hospitalization for HF (42.1 vs 38.7; P = 0.12) were not significantly different among patients who were initiated on sulfonylurea therapy compared with those who were not (Table 4, Figure 2). The unadjusted risk of all-cause mortality was significantly higher among patients who were initiated on sulfonylurea therapy (23.8 vs 20.1; P = 0.04).
|Study Endpoint||Cumulative Incidence, %||HR (95% CI); P Value|
|Sulfonylurea (n = 504)||No Sulfonylurea (n = 5,348)||Unadjusted||Adjusteda|
|All-cause mortality||23.8||20.1||1.22 (1.00-1.48); 0.045||1.24 (1.00-1.52); 0.045|
|HF hospitalization||27.1||25.1||1.11 (0.92-1.34); 0.27||1.22 (1.00-1.48); 0.050|
|All-cause mortality/HF hospitalization||42.1||38.7||1.12 (0.97-1.30); 0.12||1.17 (1.00-1.37); 0.047|
|Falsification (negative control) endpoints|
|Urinary tract infection||2.1||2.8||0.78 (0.41-1.48); 0.45||0.96 (0.50-1.84); 0.89|
|Gastrointestinal bleed||3.0||2.3||1.34 (0.77-2.34); 0.31||1.29 (0.68-2.44); 0.44|
|Vaccination for influenza||11.0||10.6||1.07 (0.82-1.39); 0.64||1.19 (0.88-1.60); 0.25|
After adjustment, sulfonylurea initiation remained associated with a statistically significant excess risk of all-cause mortality (Table 4). (HR: 1.24 [95% CI: 1.00-1.52]; P = 0.045). In adjusted analyses, initiation of sulfonylurea therapy was also associated with higher risk of composite all-cause mortality or HF hospitalization (HR: 1.17 [95% CI: 1.00-1.37]; P = 0.047) and HF hospitalization (HR: 1.22 [95% CI: 1.00-1.48]; P = 0.050), both with marginal statistical significance. Sulfonylurea initiation was not significantly associated with any falsification endpoint in unadjusted or adjusted analyses.
In subgroup analysis stratified by EF, associations between sulfonylurea initiation and all clinical endpoints were consistent among patients with EF ≤40% and >40%, with no statistical evidence of heterogeneity (all P for interaction ≥0.11) (Central Illustration).
Outcomes with metformin and sulfonylurea among combined prevalent and new users (sensitivity analyses)
After including patients who had a medication fill for metformin or sulfonylurea in the 12 months before index hospitalization, the size of the analytic cohort increased to 11,570 patients. Of these patients, within 90 days of discharge, 2,888 (25.0%) were newly treated or continued treatment with metformin, and 3,114 (26.9%) were newly treated or continued treatment with sulfonylurea. Overall, 1,073 patients (9.3%) were treated with both therapies within 90 days of discharge.
The unadjusted cumulative incidence of all 12-month clinical outcomes was lower among patients treated with metformin than those who were not (all-cause mortality 17.8 vs 22.9; P < 0.001; HF hospitalization 23.0 vs 25.9; P = 0.001; all-cause mortality or HF hospitalization 34.6 vs 41.4; P < 0.001) (Supplemental Table 4, Supplemental Figure 2). The unadjusted cumulative incidence of hospitalization for urinary tract infection was also significantly less (2.0 vs 2.7; P = 0.033). After adjustment, treatment with metformin was independently associated with reduced risk of all-cause mortality (HR: 0.88 [95% CI: 0.79-0.97]; P = 0.013) and all-cause mortality or HF hospitalization (HR: 0.89 [95% CI: 0.82-0.97]; P = 0.007), but not HF hospitalization (HR: 0.94 [95% CI: 0.85-1.04]; P = 0.25). There were no significant associations between metformin treatment and falsification endpoints.
The unadjusted cumulative incidence of HF hospitalization was higher among patients treated with sulfonylurea therapy than those who were not (26.5 vs 24.6; P = 0.039), whereas rates of all-cause mortality (22.1 vs 21.5; P = 0.49) and composite all-cause mortality or HF hospitalization (40.5 vs 39.4; P = 0.30) were similar (Supplemental Table 5, Supplemental Figure 3). After adjustment for clinical factors, sulfonylurea treatment was associated with higher risk of HF hospitalization (HR: 1.13 [1.03-1.23]; P = 0.012), whereas there were no statistically significant associations with other effectiveness endpoints or falsification endpoints.
In this large cohort of older patients hospitalized for HF with comorbid DM, after adjustment for several domains of potential confounders, metformin initiation was independently associated with a lower risk of composite mortality and HF hospitalization. This association was exclusively driven by risk reduction among patients with EF >40%. By contrast, after adjustment, there was no signal of potential benefit or harm towards mortality or HF hospitalization with metformin initiation in HFrEF. Second, sulfonylurea initiation was independently associated with increased risk of mortality and HF hospitalization outcomes. These associations with excess clinical events were consistent among patients with EF ≤40% and >40%.
To our knowledge, we present the most comprehensive comparative effectiveness analysis of metformin and sulfonylurea therapy among patients with HF and DM in U.S. clinical practice. In this regard, several strengths and novel features warrant mention. First, as compared with other clinical practice analyses limited to use of diagnoses codes alone, this analysis used patient-level clinical data (eg, vital signs, laboratories) from the GWTG-HF registry. Access to these patient-level data facilitates more accurate selection of patients eligible for treatment (eg, eGFR criteria), more comprehensive risk adjustment, and improved generalizability of findings through more detailed description of patient profile. Importantly, patient-level data included precise recording of EF (as compared with reliance on diagnoses codes for systolic or diastolic HF) and allowed for subgroup analyses defined by the clinically relevant EF cutpoint of 40%. Second, although GWTG-HF involves voluntary hospital participation, prior studies have supported its national representativeness (13). Likewise, by virtue of its primary purpose for quality improvement, the registry does not require patient informed consent, thus circumventing this form of selection bias and supporting the registry’s representation of routine clinical practice. Third, recognizing that observational studies are potentially subject to bias despite sophisticated statistical adjustment, this analysis included 3 falsification endpoints, none of which were significantly associated with metformin or sulfonylurea therapy. Fourth, this analysis assessed associated outcomes among new users of therapy, an approach that may strengthen causal inference by obviating potential prevalent user bias. Nonetheless, to augment statistical power, this incident user approach was coupled with sensitivity analyses among the larger cohort of patients who were either continuing or initiating therapy. Results of both approaches were overall concordant, suggesting reduced clinical risk with metformin and excess clinical risk with sulfonylurea, and thus further supporting the robustness of the present findings.
A post hoc analysis of the multinational SAVOR-TIMI 53 (Does Saxagliptin Reduce the Risk of Cardiovascular Events When Used Alone or Added to Other Diabetes Medications-Thrombolysis in Myocardial Infarction 53) trial of patients with type 2 DM found metformin to be independently associated with lower all-cause mortality, but noted this association only among patients without prior HF (14). Moreover, EF data were not available in that analysis, and thus the distribution of HFrEF versus HFpEF in the study population and associations with metformin within each EF phenotype could not be characterized (14). In this context, the current study among a large U.S. cohort with established HF offers new perspective, with metformin associated with meaningful improvements in clinical outcomes. Likewise, the varying relationship between metformin initiation and clinical outcomes by EF phenotype warrants further consideration. Notably, these findings differ from an older study of patients with HF and DM receiving care from 1998 to 2001 where metformin was independently associated with reduced mortality and HF hospitalization, irrespective of EF (15). Yet, this same analysis also found thiazolidinediones associated with reduced mortality and did not include falsification endpoints, raising suspicion for residual confounding. More recently, an observational study by Richardson et al (16) from the Veterans Health Administration among patients with type 2 DM and CKD found metformin initiation associated with lower risk of HF hospitalization, compared with sulfonylurea, driven by findings among patients with HFrEF. However, given 88% of patients in the study by Richardson et al (16) did not have HF at baseline, those findings are most relevant to the primary prevention of HF, as compared with treatment of patients with DM and established HF in the current study. Specifically, the present analysis supports the need for future randomized clinical trials testing the effects of metformin therapy among patients with HFpEF and DM.
Understanding effectiveness of metformin in HFrEF has become increasingly important in the context of SGLT2 inhibitors becoming standard of care therapy for HFrEF (irrespective of type 2 DM status). At the same time, current American Diabetes Association guidelines still recommend metformin as the first-line pharmacological treatment for patients with type 2 DM, unless contraindicated or not tolerated (9). In a population most closely resembling the current study, the SOLOIST-WHF (Effect of Sotagliflozin on Cardiovascular Events in Patients With Type 2 Diabetes Post Worsening Heart Failure) trial showed significant reductions in cardiovascular death and HF hospitalization with in-hospital or early postdischarge initiation of the dual SGLT1/SGLT2 inhibitor sotagliflozin among patients hospitalized for HF with type 2 DM, among which 48% were not receiving background metformin therapy (17). Thus, with data from SOLOIST-WHF and other outcome trials proving substantial clinical benefits with SGLT2 inhibitors, the lack of beneficial association with metformin among patients with HFrEF in the current study adds further support for prioritizing SGLT2 inhibitors before metformin as first-line therapy for type 2 DM in the setting of HFrEF.
Our findings regarding the higher risk of adverse outcomes associated with sulfonylurea prescription is consistent with other reports among patients with type 2 DM who were initiated on sulfonylureas who had higher risk of cardiovascular death and HF compared with similar patients who were initiated on metformin (18,19). The current data extend these findings to a high-risk older population of HF with reduced and preserved EF. These results are especially relevant in the context of clinicians often choosing sulfonylureas as the initial choice of drug for patients with DM due to the lack of gastrointestinal side effects, ease of titration, low cost, and physician comfort level (8). In the setting of multiple antidiabetic therapies proven to have favorable (eg, SGLT2 inhibitors) or neutral effects on cardiovascular outcomes in both HF and type 2 DM populations, the suggestion of harm with sulfonylurea in the current study supports prioritizing use of alternative antidiabetic medications in patients with HF, if at all possible.
Despite rigorous statistical adjustment with prespecified covariates, this observational study cannot definitively prove cause–effect relationships and residual and/or unmeasured confounding may remain. Second, this study was not structured to assess postdischarge adherence or persistence of metformin or sulfonylurea therapy, or the role of these factors in clinical outcomes. Thus, it is unclear whether patients initially filling prescriptions for metformin or sulfonylurea continued to have high exposure to the medication during follow-up. However, although adherence and persistence of therapy or the relative degree of downstream cross-over between study groups are unknown, defining study group by initial exposure is most consistent with the “intention to treat” principle used in randomized clinical trials, where adherence to therapy is not considered in determination of efficacy. Although some degree of suboptimal persistence is likely for metformin, sulfonylurea, and other medications in clinical practice, the current study estimates the benefits/harms associated with metformin and sulfonylurea under the conditions of routine clinical practice. Third, although prior data suggest that patients with HF and mildly reduced EF 41% to 49% vs EF ≥50% may respond differently to medical therapies, sample size limitations prevented dedicated analysis of these separate EF subgroups in the current study. Fourth, covariate data were not entirely complete, and imputation was used, although rates of missing data were generally low. Fifth, this analysis excluded patients with severe CKD, and thus outcomes with metformin and sulfonylureas in such patients were not addressed. Lastly, this analysis was limited to Medicare beneficiaries age 65 years or older, and the degree to which these results generalize to younger patients with HF and DM is uncertain.
In this US cohort of older patients hospitalized for HF with DM, initiation of metformin therapy was associated with decreased 12-month mortality and HF hospitalization, driven by substantial associated benefits among patients with EF >40%. By contrast, sulfonylurea prescription was associated with excess risk of mortality and HF hospitalization regardless of EF. These findings support the potential use of metformin among patients with HFpEF and DM, and the need for confirmatory randomized clinical trials. In addition, in the context of multiple alternative antidiabetic medications being available with favorable or neutral effects on cardiovascular outcomes, these findings suggest that sulfonylureas may best be avoided among patients with HF and DM.
COMPETENCY IN PATIENT CARE: These observational clinical practice data suggest that metformin therapy is a safe and potentially effective therapy for patients with adequate kidney function who have HF and DM, particularly among patients with EF >40%. By contrast, sulfonylurea therapy was associated with significant harm, irrespective of EF. In the context of multiple alternative antidiabetic therapies with favorable or neutral effects on cardiovascular outcomes, these data suggest that sulfonylurea therapy may best be avoided among patients with comorbid HF and DM.
TRANSLATIONAL OUTLOOK: These data support the need for confirmatory randomized clinical outcome trials to definitively assess the efficacy and safety of metformin as a therapy for patients with HFpEF and DM. With regard to sulfonylureas, given the frequent use of this therapy in clinical practice and the associated harms identified in the current study, future quality improvement initiatives may target deprescribing this therapy among patients with HF and DM, and replacement with an evidence-based therapy such as an SGLT2 inhibitor.
Funding Support and Author Disclosures
This work was supported in part by an American Heart Association (AHA) grant award #16SFRN30180010 (to Dr DeVore). The GWTG-HF program is provided by the AHA and sponsored, in part, by Novartis, Boehringer Ingelheim and Eli Lilly Diabetes Alliance, Novo Nordisk, Sanofi, AstraZeneca, and Bayer. Dr Sharma has received support from the Fonds de Recherche Santé Quebec (FRSQ) Junior 1 Clinical Research Scholar, Canada Institute for Health Research grant #175095, European Society of Cardiology young investigator grant, Roche Diagnostics, Boehringer Ingelheim, Novartis, AstraZeneca, and Takeda. Dr DeVore has received research funding through the Duke Clinical Research Institute from the AHA, Amgen, AstraZeneca, Bayer, Intra-Cellular Therapies, American Regent, Inc, the National Heart, Lung, and Blood Institute (NHLBI), Novartis, and the Patient-Centered Outcomes Research Institute; and consulting for AstraZeneca. Dr Felker has received research grants from National Heart, Lung, and Blood Institute (NHLBI), AHA, Amgen, Bayer Merck, Cytokinetics, and Myokardia; has been a consultant to Novartis, Amgen, BMS, Cytokinetics, Medtronic, Cardionomic, Innolife, Boehringer Ingelheim, American Reagent, Abbott, Eidos Therapeutics, Reprieve, and Sequana; and has served on clinical endpoint committees/data safety monitoring boards for Amgen, Merck, Medtronic, EBR Systems, V-Wave, LivaNova, Siemens, and Rocket Pharma. Dr Hernandez has received consulting fees from AstraZeneca, Bayer, Boston Scientific, Merck, Novartis, and Sanofi; and research support from AstraZeneca, GlaxoSmithKline, Luitpold, Merck, and Novartis. Dr Butler has been a consultant to Abbott, Adrenomed, Arena Pharma, Array, Amgen, Applied Therapeutics, AstraZeneca, Bayer, Boehringer Ingelheim, Cardior, CVRx, Eli Lilly, G3 Pharma, Imbria, Impulse Dynamics, Innolife, Janssen, LivaNova, Luitpold, Medtronic, Merck, Novartis, Novo Nordisk, Sequana Medical, V-Wave Limited, and Vifor. Dr Peterson has received research funding from the NHLBI; and personal fees from American Heart Association outside the submitted work. Dr Fonarow has received research funding from the National Institutes of Health; and has been a consultant for Amgen, Bayer, Medtronic, and Novartis. Dr Greene has received research support from the Duke University Department of Medicine Chair’s Research Award, the AHA, NHLBI, Amgen, AstraZeneca, Bristol Myers Squibb, Cytokinetics, Merck, Novartis, Pfizer, and Sanofi; has served on advisory boards for Amgen, AstraZeneca, Bristol Myers Squibb, and Cytokinetics; and has served as a consultant for Amgen, Bayer, Bristol Myers Squibb, Merck, and Vifor. All other authors have reported that they have no relationships relevant to the contents of this paper to disclose.
Abbreviations and Acronyms
chronic kidney disease
estimated glomerular filtration rate
hospitalization for heart failure
heart failure with reduced ejection fraction
heart failure with preserved ejection fraction
sodium glucose co-transporter 2
1. "Prevalent and incident heart failure in cardiovascular outcome trials of patients with type 2 diabetes". J Am Coll Cardiol 2018;71:1379-1390.
2. "Secondary prevention of macrovascular events in patients with type 2 diabetes in the PROactive Study (PROspective pioglitAzone Clinical Trial In macroVascular Events): a randomised controlled trial". Lancet 2005;366:1279-1289.
3. "Rosiglitazone evaluated for cardiovascular outcomes in oral agent combination therapy for type 2 diabetes (RECORD): a multicentre, randomised, open-label trial". Lancet 2009;373:2125-2135.
4. "Saxagliptin and cardiovascular outcomes in patients with type 2 diabetes mellitus". N Engl J Med 2013;369:1317-1326.
5. "Empagliflozin, cardiovascular outcomes, and mortality in type 2 diabetes". N Engl J Med 2015;373:2117-2128.
6. "Cardiovascular and renal outcomes with empagliflozin in heart failure". N Engl J Med 2020;383:1413-1424.
8. "Contemporary treatment patterns and clinical outcomes of comorbid diabetes mellitus and HFrEF: the CHAMP-HF registry". J Am Coll Cardiol HF 2020;8:469-480.
9. "American Diabetes Association. 9. Pharmacologic approaches to glycemic treatment: Standards of Medical Care in Diabetes-2021". Diabetes Care 2021;44:suppl 1: S111-S124. https://doi.org/10.2337/dc21-S009.
10. "2019 ESC guidelines on diabetes, pre-diabetes, and cardiovascular diseases developed in collaboration with the EASD". Eur Heart J 2020;41:255-323.
11. "Linking inpatient clinical registry data to Medicare claims data using indirect identifiers". Am Heart J 2009;157:995-1000.
12. "Negative control outcomes: a tool to detect bias in randomized trials". JAMA 2016;316:2597-2598.
13. "Representativeness of a national heart failure quality-of-care registry: comparison of OPTIMIZE-HF and non-OPTIMIZE-HF Medicare patients". Circ Cardiovasc Qual Outcomes 2009;2:377-384.
14. "Metformin use and clinical outcomes among patients with diabetes mellitus with or without heart failure or kidney dysfunction: observations from the SAVOR-TIMI 53 trial". Circulation 2019;140:1004-1014.
15. "Thiazolidinediones, metformin, and outcomes in older patients with diabetes and heart failure: an observational study". Circulation 2005;111:583-590.
16. "Hospitalization for heart failure among patients with diabetes mellitus and reduced kidney function treated with metformin versus sulfonylureas: a retrospective cohort study". J Am Heart Assoc 2021;10:8: e019211. https://doi.org/10.1161/JAHA.120.019211.
17. "Sotagliflozin in patients with diabetes and recent worsening heart failure". N Engl J Med 2021;384:117-128.
19. "Association of second-line antidiabetic medications with cardiovascular events among insured adults with type 2 diabetes". JAMA Netw Open 2018;1: e186125-e186125.
John Teerlink, MD, served as the Guest Editor of this paper.
The authors attest they are in compliance with human studies committees and animal welfare regulations of the authors’ institutions and Food and Drug Administration guidelines, including patient consent where appropriate. For more information, visit the Author Center.