Abstract
Background
Metformin and sulfonylurea are commonly prescribed oral medications for type 2 diabetes mellitus. The association of metformin and sulfonylureas on heart failure outcomes in patients with reduced estimated glomerular filtration rate remains poorly understood.
Methods and Results
This retrospective cohort combined data from National Veterans Health Administration, Medicare, Medicaid, and the National Death Index. New users of metformin or sulfonylurea who reached an estimated glomerular filtration rate of 60 mL/min per 1.73 m2 or serum creatinine of 1.5 mg/dL and continued metformin or sulfonylurea were included. The primary outcome was hospitalization for heart failure. Echocardiogram reports were obtained to determine each patient's ejection fraction (EF) (reduced EF <40%; midrange EF 40%–49%; ≥50%). The primary analysis estimated the cause‐specific hazard ratios for metformin versus sulfonylurea and estimated the cumulative incidence functions for heart failure hospitalization and competing events. The weighted cohort included 24 685 metformin users and 24 805 sulfonylurea users with reduced kidney function (median age 70 years, estimated glomerular filtration rate 55.8 mL/min per 1.73 m2). The prevalence of underlying heart failure (12.1%) and cardiovascular disease (31.7%) was similar between groups. There were 16.9 (95% CI, 15.8–18.1) versus 20.7 (95% CI, 19.5–22.0) heart failure hospitalizations per 1000 person‐years for metformin and sulfonylurea users, respectively, yielding a cause‐specific hazard of 0.85 (95% CI, 0.78–0.93). Among heart failure hospitalizations, 44.5% did not have echocardiogram information available; 29.3% were categorized as reduced EF, 8.9% as midrange EF, and 17.2% as preserved EF. Heart failure hospitalization with reduced EF (hazard ratio, 0.79; 95% CI, 0.67–0.93) and unknown EF (hazard ratio, 0.84; 95% CI 0.74–96) were significantly lower in metformin versus sulfonylurea users.
Conclusions
Among patients with type 2 diabetes mellitus who developed worsening kidney function, persistent metformin compared with sulfonylurea use was associated with reduced heart failure hospitalization.
Keywords: chronic kidney disease, heart failure hospitalization, type 2 diabetes mellitus
Subject Categories: Risk Factors, Quality and Outcomes, Complications, Heart Failure
Nonstandard Abbreviations and Acronyms
- T2D
type 2 diabetes mellitus
- VHA
Veterans Health Administration
Clinical Perspective
What Is New?
Persistent use of metformin after kidney function decline is associated with reduced risk of heart failure hospitalizations compared with sulfonylurea drugs.
The reduced risk of heart failure hospitalization was statistically significant in those hospitalized with reduced ejection fraction and unknown ejection fraction.
What Are the Clinical Implications?
This study adds evidence to the growing body of literature demonstrating metformin's association with beneficial cardiovascular effects compared with sulfonylurea in patients with mild to moderate renal dysfunction.
More healthcare resources are spent on diabetes mellitus than any other medical condition in the United States. 1 In 2015, the Centers for Disease Control and Prevention noted that about 9.4% of the population, including >30 million Americans, were living with type 2 diabetes mellitus (T2D). 2 , 3 Patients with T2D often develop secondary complications, including impaired kidney function and heart failure. Heart failure is the most common cause of hospitalization among veterans and patients with Medicare. Furthermore, a diagnosis of T2D is associated with a 33% increase in the odds of a hospitalization for heart failure. 4 , 5
Among veterans, metformin and sulfonylureas remain the most commonly used oral medications for T2D, even as kidney function declines. 6 However, limited data about heart failure outcomes exist for these older medications, particularly among those patients with reduced kidney function. Metformin and sulfonylureas exert their glucose lowering effects via distinct mechanisms; metformin improves insulin sensitivity and is typically weight neutral, whereas the sulfonylurea class increases endogenous insulin secretion. Hyperinsulinemia exerts many end‐organ and systemic effects, including weight gain and fluid retention. Physiologically, these changes in hemodynamics can lead to or worsen hypertensive heart disease, left ventricular hypertrophy, and subsequent remodeling, thus increasing the risk of decompensated heart failure. 7
Before 2016, metformin was restricted for patients with serum creatinine levels ≥1.5 mg/dL in men and ≥1.4 mg/dL in women. In 2016, the Food and Drug Administration changed its guidance and now recommends that metformin can be used in patients with mild to moderate kidney disease, until a patient reaches an estimated glomerular filtration rate (eGFR) of 30 mL/min per 1.73 m2. 8 Because of the recommendations to reduce the use of metformin for those with diabetes mellitus and reduced kidney function, there is limited understanding of the impact of metformin on heart failure outcomes among this high‐risk group. Our aim was to test the hypothesis that among patients with T2D who developed reduced kidney function, the risk of heart failure hospitalizations would be lower among patients who persisted on metformin versus sulfonylureas.
Methods
Statement of Research Reproducibility
The protocol, statistical code, and deidentified and anonymized data sets are available from Dr Roumie with a written request per the Transparency and Openness Promotion Guidelines.
Study Design and Data Sources
We assembled a retrospective cohort of Veterans Health Administration (VHA) patients. 6 Pharmacy data included medication dispensed, date filled, days supplied, and number of pills dispensed. Demographic, diagnostic, and procedure information identified inpatient and outpatient VHA encounters. We collected laboratory results and vital signs data from clinical sources. For Medicare or Medicaid enrollees, we obtained enrollment, claims files, and prescription (Medicare Part D) data. 9 , 10 Dates and cause of death were obtained from vital status and the National Death Index files. 11 , 12 The institutional review board of VHA Tennessee Valley Healthcare System approved this study with a waiver of consent.
Study Population
The population comprised veterans aged 18 years and older who were regular users of VHA care, defined as an encounter or prescription filled at least once every 365 days for 2 or more years before cohort entry. We identified patients with new‐onset T2D by selecting those who were new users of metformin, glipizide, glyburide, or glimepiride. New users were patients who filled a first glucose‐lowering prescription without any diabetic drug filled in the 180 days before that first fill. We followed these patients with diabetes mellitus longitudinally and selected patients who experienced a decline in kidney function. Patients were required to persist with their initial monotherapy with no medication gaps for >180 days or medication switching before reaching the kidney threshold to be eligible for cohort entry.
The index date and start of follow‐up was the date of reaching a reduced kidney function threshold, defined as either an eGFR of <60 mL/min per 1.73 m2 or serum creatinine level of ≥1.5 mg/dL for men or ≥1.4 mg/dL for women (Figure S1). The index date and cohort entry were restricted to the period between January 1, 2002 and December 30, 2015 to allow sufficient collection of baseline data and to allow follow‐up through December 31, 2016. We excluded patients who added or switched glucose‐lowering medications at or before the kidney threshold or had a single episode of dialysis, organ transplantation, or enrollment in hospice care at or within the 2 years before reaching the reduced kidney function threshold.
Exposure
The study exposures were persistent use of metformin or a sulfonylurea (glyburide, glipizide, and glimepiride) after reaching the reduced kidney threshold. Follow‐up began on the date the kidney threshold (eGFR <60 mL/min per 1.73 m2 or serum creatinine level 1.4/1.5 mg/dL) was fulfilled and continued through an outcome (heart failure hospitalization), a competing risk (drug nonpersistence or death), or a censoring event (loss to follow‐up or end of the study). Competing risks were defined as informative events that may have been influenced by the study medications. Nonpersistence was defined as 90 days without an antidiabetic drug or the addition of or switch to a different glucose‐lowering drug. 13 Censoring events were defined as noninformative events not likely to be influenced by study medications; loss to follow‐up was defined as the 181st day of no VHA contact (inpatient, outpatient, or pharmacy use), or study end (December 31, 2016).
Outcomes
The primary outcome was hospitalization with a primary discharge diagnosis of heart failure, cardiomyopathy, or hypertensive heart disease with heart failure. Events were defined by primary discharge diagnosis codes of the International Classification of Diseases, Ninth or Tenth Revision (ICD‐9; ICD‐10) before or after 2015. 14 ICD‐9 , Clinical Modification codes included the following: 425.X, 428.X, 404.01, 404.03, 404.11, 404.13, 398.91, 402.01, 402.11, 402.91, 404.91, and 404.93. ICD‐10 codes included: I50.2*, I50.3*, I50.9, I42.9, I13.0, I13.2, I09.81, and I11.0. A heart failure hospitalization could also be captured if there was a diagnosis‐related group code for heart failure (diagnosis‐related group code 127 before fiscal year 2008 and 291–293 after fiscal year 2008). 15 , 16 The outcome date was the admission day.
To further understand the type of heart failure that was associated with the hospitalization, we used the Natural Language Processing echocardiogram algorithm developed and reported previously by Patterson et al. 17 Only echocardiogram reports conducted within the VHA were available to determine heart failure type on the basis of ejection fraction (EF) (reduced <40%, midrange 40%–49%, and preserved ≥50%). The echocardiogram used for heart failure classification was the study obtained closest to the day of admission and up to 7 days after admission. If no echocardiogram was obtained during that heart failure hospitalization, we evaluated echocardiograms for each patient up to 1 year before that admission and used the one closest to the date of admission. If no echocardiogram was available, including any obtained for a hospitalization outside of the VHA and within the Medicare claims files, then heart failure hospitalization type was considered unknown.
Covariates
Study covariates were included as the closest measured to the date of cohort entry and up to 720 days before the reduced kidney function threshold. Covariates included age, sex, race, fiscal year, number of months from initial antidiabetic medication to reaching the reduced kidney function threshold (diabetes mellitus duration), and Veterans Integrated Service Network of care. Each Veterans Integrated Service Network of care is a geographic designation for the VHA and allowed a more granular estimation of geographic variation of diabetes mellitus care. Physiologic variables were also collected for up to 720 days before the kidney threshold and defined as the most recent measure before kidney threshold. Physiologic variables included body mass index (calculated as weight in kilograms divided by height in meters squared), blood pressure, glycated hemoglobin, low‐density lipoprotein, hemoglobin, proteinuria, and creatinine values (both historical and the creatinine at cohort entry).
Creatinine was used to calculate eGFR using the Chronic Kidney Disease Epidemiology Collaboration equation. 18 , 19 Healthcare utilization (hospitalization, nursing home use, number of outpatient visits or medications, and Medicare or Medicaid insurance use) was measured in the year before the reduced kidney function threshold. We collected data on smoking and comorbidities as defined in Table S1. Selected medications filled within 180 days before the reduced kidney function threshold were also covariates. Because race is associated with heart failure outcomes, we collected patient self‐reported categorical race from VHA data and supplemented those data with patients with Medicare self‐reported categorical race data to minimize missing values. 20
Statistical Analysis
The primary analysis accounted for 2 competing risks, medication nonpersistence and all‐cause death, and compared the cause‐specific hazard of heart failure hospitalizations between metformin versus sulfonylurea users (referent) using a propensity score–weighted model. The propensity score modeled the probability of metformin or sulfonylurea at the time of the reduced kidney function threshold and included covariates, Veterans Integrated Service Network, and an indicator of missing covariates. Missing covariates were handled with multiple imputations using 20 iterations of chained imputations adjusted for canonical variates. 21 We used matching weights derived from the propensity score to balance both exposure groups on observed covariates (detailed methods in Table S2, Figures S2 and S3). 22 , 23 , 24 Standardized mean differences were calculated as the difference between groups in number of standard deviations, because this is the preferred measure of covariate balance when dealing with large sample sizes. 25 Smaller standardized mean difference values indicate less difference between groups, with 0 indicating perfect balance in mean or proportion.
Cox proportional hazards models estimated the cause‐specific hazard ratios (HRs) for metformin versus sulfonylurea (referent) in the weighted cohort while adjusting for the aforementioned covariates. Separate analyses were conducted for each type of heart failure hospitalization (reduced, midrange, preserved, or unknown EF). For each type of heart failure, all of the other types of heart failure were combined and called "other heart failure hospitalization," and considered a competing risk. For the cause‐specific hazard model of heart failure hospitalization, the concordance averaged 0.88 across the multiple imputed data sets indicating good predictive ability. The Akaike information criterion (AIC) for the unadjusted cause‐specific hazard model for heart failure hospitalization was 59 664.79, and the average AIC for the adjusted model was 54 352.34.
Statistical significance for the 2‐sided P value was set at 0.05. Fulfillment of the proportional hazards assumptions was verified through examination of Schoenfeld residuals over time. 26 Cumulative incidence plots for the weighted cohort were generated using the Aalen‐Johansen estimator. 27 When estimating potentially causal associations, it is preferable to report the cause‐specific HRs with competing risks treated as censored outcomes. 28 However, this approach will yield biased estimates of cumulative incidence. Thus, the Aalen‐Johansen estimator is used. The outcome, heart failure hospitalization, and the competing risks of nonpersistence and death were treated as terminal states.
Sensitivity and Subgroup Analysis
A planned sensitivity analysis excluded patients who were enrolled in Medicare Advantage during the baseline period and censored patients' follow‐up upon enrollment in Medicare Advantage programs. In this sensitivity analysis, Medicare Advantage (Part C) individuals were excluded because their claims tended to be missing or incomplete during the time frame of the study. 29 We also conducted subgroup analyses and tested for effect modification by stratifying by the following covariates: age (≥65, <65 years), race (Black and non‐Black), history of cardiovascular disease (yes, no) and history of heart failure (yes, no). Analyses were conducted using R. 30
Results
Study Cohort and Patient Characteristics
The study identified 67 762 new metformin users and 28 979 new sulfonylurea users who persisted on treatment, reached the reduced kidney function threshold, and satisfied cohort entry criteria (Figure 1). This cohort of persistent new users represented 55.3% of 174 882 new users of metformin or sulfonylurea who remained persistent on medication and reached the reduced kidney function threshold. We excluded 59 464 whose regimens changed before or on the day that the kidney threshold was reached. There were 12 505 who met the kidney threshold outside the prespecified study time frame, 5647 who had no supply of metformin or sulfonylurea in the 90 days before reaching the kidney threshold, and those with hospice care (n=219), organ transplant (n=206), data error (n=75), or dialysis use in the past 2 years (n=25). After propensity score calculation and weighting, the cohort included 24 685 metformin and 24 804 sulfonylurea users (54% glipizide, 45% glyburide, and 1% glimepiride).
Figure 1. Flow of eligible patients in the Veterans Health Administration diabetes mellitus kidney disease cohort.
Weighted number uses matching weights derived from the propensity score to balance both exposure groups on observed covariates.
The unweighted full cohort of patients were 96.5% men and 82.8% White. Metformin and sulfonylurea users had similar baseline characteristics. However, metformin users were younger than sulfonylurea users (median age 67 years versus 71 years, respectively). After weighting, patient characteristics were similar between metformin and sulfonylurea including age 70 years (interquartile range [IQR], 63–78) versus 70 years (IQR, 63–78), glycated hemoglobin 6.5% (IQR, 6.1–7.1) versus 6.6% (IQR, 6.1–7.2), and eGFR 55.8 (IQR, 51.6–58.2) versus 55.8 (IQR, 51.6–58.2), respectively, at the time of reduced kidney function threshold. The historical eGFR before cohort entry was also reduced (69.6 mL/min [IQR, 64.6–77.0]), and the difference between these 2 eGFRs was 14.6 mL/min (IQR, 9.6–23.5) for metformin and 14.6 mL/min (IQR, 9.6–23.2) for sulfonylurea. The median time between these 2 eGFR measures was 4.6 months (IQR, 2.4–7.0) for metformin users and 5.0 months (IQR, 2.6–7.5) for sulfonylurea users.
The prevalence of underlying congestive heart failure and cardiovascular disease was balanced between metformin and sulfonylurea users (heart failure 12.1% versus 12.1%, CVD 31.6% versus 31.7%, respectively) (Table 1). The median observed follow‐up in the weighted cohort was 1.03 years (IQR, 0.35–2.58) for patients taking metformin and 1.17 years (IQR, 0.46–2.66) for sulfonylurea.
Table 1.
Patient Characteristics at Cohort Entry of Reduced Kidney Function
| Full Unweighted Cohort | Weighted Cohort | SMD* | |||
|---|---|---|---|---|---|
| Metformin | Sulfonylurea | Metformin | Sulfonylurea | ||
| N=67 762 | N=28 979 | N=24 685 | N=24 804 | ||
| Age, y † | 67 [62–74] | 71 [63–79] | 70 [62–77] | 70 [62–77] | <0.001 |
| Men, n (%) | 64 933 (95.8) | 28 462 (98.2) | 24 195 (98.0) | 24 312 (98.0) | <0.001 |
| Race, n (%) | 0.001 | ||||
| Other § | 1473 (2.2) | 528 (1.8) | 457 (1.9) | 463 (1.9) | |
| Black | 9884 (14.6) | 4925 (17.0) | 4036 (16.4) | 4048 (16.3) | |
| White | 56 405 (83.2) | 23 526 (81.2) | 20 191 (81.8) | 20 293 (81.8) | |
| Medication start to kidney threshold, mo † | 16.2 [6.5–35.1] | 13.6 [5.9–29.0] | 14.0 [5.8–30.2] | 14.0 [6.0–30.3] | 0.01 |
| Years of cohort entry, n (%) | 0.03 | ||||
| 2002–2003 | 3167 (4.7) | 4904 (16.9) | 2925 (11.8) | 2919 (11.8) | |
| 2004–2005 | 5786 (8.6) | 5737 (19.8) | 4481 (18.2) | 4443 (17.9) | |
| 2006–2007 | 9075 (13.4) | 6101 (21.0) | 5208 (21.1) | 5439 (21.9) | |
| 2008–2009 | 9952 (14.7) | 4051 (14.0) | 3875 (15.7) | 3894 (15.7) | |
| 2010–2011 | 12 237 (18.0) | 3341 (11.6) | 3366 (13.6) | 5049 (13.3) | |
| 2012–2013 | 12 854 (19.0) | 2619 (9.0) | 2652 (10.8) | 2600 (10.4) | |
| 2014–2015 | 14 691 (21.6) | 2226 (7.7) | 2178 (8.8) | 2222 (9.0) | |
| Clinical variables | |||||
| Body mass index, kg/m2 † | 31.1 [27.7–35.2] | 30.1 [26.9–34.1] | 30.4 [27.1–34.4] | 30.3 [27.1–34.3] | 0.004 |
| Missing BMI measure, n (%) | 11 519 (17.0) | 5733 (19.8) | 4610 (18.7) | 4635 (18.7) | <0.001 |
| Systolic blood pressure, mm Hg † | 129 [118–140] | 131 [120–143] | 131 [119–142] | 131 [119–142] | 0.003 |
| Diastolic blood pressure, mm Hg † | 73 [65–80] | 71 [64–80] | 72 [64–80] | 72 [64–80] | <0.001 |
| Laboratory variables | |||||
| HbA1c, % † | 6.5 [6.1–7.0] | 6.6 [6.1–7.3] | 6.5 [6.1–7.1] | 6.6 [6.1–7.2] | 0.006 |
| Missing HbA1c, n (%) | 2768 (4.1) | 1138 (3.9) | 1010 (4.1) | 995 (4.0) | 0.004 |
| Hemoglobin, g/dL † | 14.0 [12.9–15.0] | 14.1 [13.0–15.2] | 14.1 [13.0–15.1] | 14.1 [13.0–15.2] | 0.003 |
| Missing hemoglobin, n (%) | 3630 (5.4) | 1712 (5.9) | 1513 (6.1) | 1508 (6.1) | 0.002 |
| Estimated glomerular filtration rate at cohort entry † | 55.9 [51.7–58.2] | 55.8 [51.5–58.2] | 55.8 [51.6–58.2] | 55.8 [51.6–58.2] | 0.002 |
| Estimated glomerular filtration rate before cohort entry † | 70.5 [65.1–78.6] | 69.2 [64.5–76.5] | 69.6 [64.7–77.0] | 69.6 [64.7–77.0] | <0.001 |
| Serum creatinine, mg/dL † | 1.33 [1.24–1.43] | 1.33 [1.24–1.43] | 1.33 [1.24–1.43] | 1.33 [1.24–1.43] | 0.002 |
| Low‐density lipoprotein, mg/dL † | 85 [67–106] | 89 [72–111] | 88 [70–110] | 88 [71–110] | 0.001 |
| Missing low‐density lipoprotein, n (%) | 1323 (2.0) | 1139 (3.9) | 797 (3.2) | 798 (3.2) | <0.001 |
| Urine protein on urinalysis, n (%) | 0.002 | ||||
| Negative | 32 970 (48.7) | 13 517 (46.6) | 11 651 (47.2) | 11 706 (47.2) | |
| Trace or 1+ | 10 072 (14.9) | 4185 (14.4) | 3574 (14.5) | 3606 (14.5) | |
| 2+ | 2187 (3.2) | 983 (3.4) | 803 (3.3) | 808 (3.3) | |
| 3+/4+/trace to 4+ | 632 (0.9) | 483 (1.7) | 347 (1.4) | 348 (1.4) | |
| Missing urine protein measure, n (%) | 21 901 (32.3) | 9811 (33.9) | 8309 (33.7) | 8335 (33.6) | |
| MACR stage, n (%) | 0.003 | ||||
| A1, <30 mg/g normal to mild increase albuminuria | 29 664 (43.8) | 10 626 (36.7) | 9472 (38.4) | 9532 (38.4) | |
| A2, 30–300 mg/g moderate increase albuminuria | 7400 (10.9) | 3076 (10.6) | 2675 (10.8) | 2676 (10.8) | |
| A3 and positive, >300 mg/g severely increased albuminuria | 1815 (2.7) | 931 (3.2) | 769 (3.1) | 763 (3.1) | |
| Missing MACR measure | 28 883 (42.6) | 14 346 (49.5) | 11 768 (47.7) | 11 833 (47.7) | |
| Baseline comorbidities, n (%) ‡ | |||||
| Malignancy | 7199 (10.6) | 3514 (12.1) | 2891 (11.7) | 2907 (11.7) | <0.001 |
| Liver disease | 1131 (1.7) | 820 (2.8) | 596 (2.4) | 593 (2.4) | 0.002 |
| HIV | 235 (0.3) | 118 (0.4) | 95 (0.4) | 97 (0.4) | 0.001 |
| Congestive heart failure | 5527 (8.2) | 4218 (14.6) | 2988 (12.1) | 3010 (12.1) | <0.001 |
| Cardiovascular disease | 17 701 (26.1) | 9811 (33.9) | 7798 (31.6) | 7869 (31.7) | 0.003 |
| Stroke | 1900 (2.8) | 1031 (3.6) | 833 (3.4) | 830 (3.3) | 0.002 |
| Transient ischemic attack | 710 (1.0) | 410 (1.4) | 321 (1.3) | 331 (1.3) | 0.003 |
| Serious mental illness | 16 591 (24.5) | 5827 (20.1) | 5048 (20.4) | 5122 (20.6) | 0.005 |
| Smoking | 8749 (12.9) | 3552 (12.3) | 3064 (12.4) | 3086 (12.4) | <0.001 |
| Chronic obstructive pulmonary disease | 10 304 (15.2) | 5266 (18.2) | 4196 (17.0) | 4234 (17.1) | 0.002 |
| History of respiratory failure | 1967 (2.9) | 963 (3.3) | 791 (3.2) | 791 (3.2) | <0.001 |
| History of kidney disease | 73 (0.1) | 52 (0.2) | 35 (0.1) | 38 (0.2) | 0.002 |
| History of sepsis | 961 (1.4) | 511 (1.8) | 397 (1.6) | 403 (1.6) | 0.001 |
| History of pneumonia | 2179 (3.2) | 1426 (4.9) | 1057 (4.3) | 1074 (4.3) | 0.002 |
| Arrhythmias | 9511 (14.0) | 5469 (18.9) | 4289 (17.4) | 4320 (17.4) | 0.001 |
| Cardiac valve disease | 1894 (2.8) | 1196 (4.1) | 898 (3.6) | 907 (3.7) | 0.001 |
| Parkinson | 496 (0.7) | 311 (1.1) | 228 (0.9) | 231 (0.9) | <0.001 |
| Urinary tract infection | 2267 (3.3) | 1375 (4.7) | 1035 (4.2) | 1046 (4.2) | 0.001 |
| Osteomyelitis | 309 (0.5) | 198 (0.7) | 155 (0.6) | 153 (0.6) | 0.002 |
| Osteoporosis | 475 (0.7) | 239 (0.8) | 196 (0.8) | 202 (0.8) | 0.002 |
| Falls | 147 (0.2) | 73 (0.3) | 55 (0.2) | 57 (0.2) | 0.001 |
| Fractures | 1258 (1.9) | 679 (2.3) | 549 (2.2) | 549 (2.2) | <0.001 |
| Amputation | 230 (0.3) | 170 (0.6) | 116 (0.5) | 120 (0.5) | 0.002 |
| Retinopathy | 508 (0.7) | 399 (1.4) | 290 (1.2) | 291 (1.2) | <0.001 |
| Use of medications, n (%) | |||||
| ACE inhibitors | 43 233 (63.8) | 18 811 (64.9) | 15 968 (64.7) | 16 091 (64.9) | 0.004 |
| Angiotensin receptor blockers | 8697 (12.8) | 3109 (10.7) | 2816 (11.4) | 2807 (11.3) | 0.003 |
| β‐blockers | 33 342 (49.2) | 14 798 (51.1) | 12 514 (50.7) | 12 587 (50.7) | 0.001 |
| Calcium channel blockers | 19 721 (29.1) | 8667 (29.9) | 7381 (29.9) | 7415 (29.9) | <0.001 |
| Thiazide/potassium‐sparing diuretics | 29 986 (44.3) | 11 573 (39.9) | 10 103 (40.9) | 10 195 (41.1) | 0.004 |
| Loop diuretics | 10 317 (15.2) | 6621 (22.8) | 4957 (20.1) | 4983 (20.1) | <0.001 |
| Other antihypertensives | 18 461 (27.2) | 7833 (27.0) | 6719 (27.2) | 6728 (27.1) | 0.002 |
| Lipid‐lowering statins | 49 915 (73.7) | 18 671 (64.4) | 16 548 (67.0) | 16 698 (67.3) | 0.006 |
| Nonstatin lipid‐lowering agents | 13 167 (19.4) | 4665 (16.1) | 4246 (17.2) | 4273 (17.2) | <0.001 |
| Antiarrhythmic digoxin and inotropes | 4395 (6.5) | 3143 (10.8) | 2260 (9.2) | 2272 (9.2) | <0.001 |
| Anticoagulants | 6029 (8.9) | 3099 (10.7) | 2488 (10.1) | 2496 (10.1) | <0.001 |
| Nitrates | 7812 (11.5) | 4715 (16.3) | 3628 (14.7) | 3664 (14.8) | 0.002 |
| Aspirin | 14 373 (21.2) | 6543 (22.6) | 5360 (21.7) | 5408 (21.8) | 0.002 |
| Platelet inhibitors | 6241 (9.2) | 3100 (10.7) | 2574 (10.4) | 2593 (10.5) | <0.001 |
| Antipsychotics | 5415 (8.0) | 1992 (6.9) | 1740 (7.0) | 1762 (7.1) | 0.002 |
| Oral glucocorticoids | 5050 (7.5) | 2139 (7.4) | 1795 (7.3) | 1813 (7.3) | 0.001 |
| Indicators of healthcare utilization | |||||
| Hospitalization within year (Veterans Health), n (%) | 9077 (13.4) | 4517 (15.6) | 3576 (14.5) | 3630 (14.6) | 0.004 |
| Hospitalizations within 30 d (Veterans Health), n (%) | 2510 (3.7) | 1197 (4.1) | 942 (3.8) | 961 (3.9) | 0.003 |
| Hospitalizations within y (Medicaid/Medicare), n (%) | 5634 (8.3) | 3597 (12.4) | 2771 (11.2) | 2788 (11.2) | <0.001 |
| Hospitalizations within 30 d (Medicaid/Medicare), n (%) | 987 (1.5) | 581 (2.0) | 439 (1.8) | 452 (1.8) | 0.003 |
| Medicaid insurance use in past year, n (%) | 663 (1.0) | 435 (1.5) | 323 (1.3) | 331 (1.3) | 0.002 |
| Medicare insurance use in past year, n (%) | 21 437 (31.6) | 10 540 (36.4) | 8810 (35.7) | 8815 (35.5) | 0.003 |
| Nursing home encounters, n (%) | 201 (0.3) | 137 (0.5) | 97 (0.4) | 101 (0.4) | 0.002 |
| No. of medications † | 7 [5–11] | 7 [4–10] | 7 [4–10] | 7 [4–10] | 0.003 |
| Outpatient visits in past year † | 6 [3–11] | 6 [4–11] | 6 [3–11] | 6 [3–11] | 0.002 |
| Medicare Advantage use, n (%) | 10 253 (15.1) | 4339 (15.0) | 3771 (15.3) | 3785 (15.3) | <0.001 |
ACE indicates angiotensin‐converting enzyme; HbA1c, glycated hemoglobin; MACR, microalbumin to creatinine ratio; and SMD, standardized mean difference.
SMDs are the absolute difference in means or percent divided by an evenly weighted pooled standard deviation, or the difference between groups in number of standard deviations. In the weighted cohort, all standardized differences were not statistically significant (see Figure S3 for the plot of the SMDs of the prematched and matched cohort).
Median [interquartile range].
Definitions of comorbidities can be found in Table S1.
Other races include: Asian, American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander
Primary Outcome: Hospitalization for Heart Failure
After propensity score weighting, there were 775 heart failure hospitalization events among metformin patients with reduced kidney function, and 992 events among patients taking sulfonylurea; this yielded 16.9 (95% CI, 15.8–18.1) versus 20.7 (95% CI, 19.5–22.0) events per 1000 person‐years of use, respectively. After covariate adjustment, the cause‐specific HR for heart failure hospitalizations was 0.85 (95% CI, 0.78–0.93) among metformin compared with sulfonylurea users (Table 2). The Aalen‐Johansen plot demonstrates the cumulative probability of heart failure hospitalizations over a 5‐year period (Figure 2 and Figure S4). Cumulative probability of a heart failure hospitalization between metformin and sulfonylurea was 1.6% versus 2.0% at 1 year and 3.0% versus 3.8% at 5 years. These estimates accounted for the competing risks of nonpersistence (82.8% metformin versus 80.8% sulfonylurea) and noncardiovascular death (3.1% versus 4.2%). Among nonpersistent metformin users, 58.4% stopped the drug, and 41.6% added another drug (7.3% added insulin, 82.9% added sulfonylurea, 6.6% added an alternative agent, and 3.3% added more than one medication). Among nonpersistent sulfonylurea users, 60.2% stopped the drug, and 39.8% added another drug (15.8% added insulin, 69% added metformin, 10.7% added an alternative agent, and 4.5% added more than one medication).
Table 2.
Rates, Adjusted Hazard Ratios, and Confidence Intervals for HF Hospitalization and HF Type for Metformin Versus Sulfonylurea Users With Reduced Kidney Function
| Metformin | Sulfonylurea | |
|---|---|---|
| No. at risk matched weighted | n=24 685 | n=24 804 |
| Primary outcome: HF hospitalization | 775 | 992 |
| Person‐years | 45 865 | 47 882 |
| Unadjusted rate/1000 person‐years (95% CI) | 16.9 (15.7–18.1) | 20.7 (19.5–22.0) |
| Adjusted HR (95% CI)* | 0.85 (0.78–0.93) | Reference |
| HF hospitalization type: reduced ejection fraction | 214 | 303 |
| Unadjusted rate/1000 person‐years (95% CI) | 4.67 (4.1–5.3) | 6.33 (5.7–7.1) |
| Adjusted HR (95% CI) † | 0.79 (0.67–0.93) | Reference |
| HF hospitalization type: midrange ejection fraction | 73 | 85 |
| Unadjusted rate/1000 person‐years (95% CI) | 1.6 (1.3–2.0) | 1.8 (1.4–2.2) |
| Adjusted HR (95% CI) † | 0.94 (0.71–1.26) | Reference |
| HF hospitalization type: preserved ejection fraction | 148 | 157 |
| Unadjusted rate/1000 person‐years (95% CI) | 3.2 (2.7–3.8) | 3.3 (2.8–3.8) |
| Adjusted HR (95% CI) † | 0.97 (0.79–1.20) | Reference |
| HF hospitalization type: unknown ejection fraction | 340 | 447 |
| Unadjusted rate/1000 person‐years (95% CI) | 7.4 (6.7–8.2) | 9.3 (8.5–10.2) |
| Adjusted HR (95% CI) † | 0.84 (0.74–0.96) | Reference |
| Sensitivity analysis: exclude Medicare Advantage | N=20 914 | N=21 019 |
| HF hospitalization | 676 | 862 |
| Person‐years | 36 939 | 38 766 |
| Unadjusted rate/1000 person‐years (95% CI) | 18.3 (17.0–19.7) | 22.2 (20.8–23.8) |
| Adjusted HR (95% CI)* | 0.85 (0.77–0.93) | Reference |
HF indicates heart failure; and HR, hazard ratio.
Primary analysis considers patients persistent on regimen until they do not have antidiabetic medications for 90 days. Model adjusted for full list of covariates: demographics, clinical information derived from the electronic health record, comorbidities, use of medications, and healthcare utilization. All continuous variables modeled as restricted cubic splines.
Primary analysis considers patients persistent on regimen until they do not have antidiabetic medications for 90 days. Reduced model to allow for convergence. All covariates in above model except: Veterans Integrated Service Networks of care regrouped into regions (North, South, Midwest, West) and excluded HIV, history of kidney disease, osteomyelitis, osteoporosis, falls, sepsis, Parkinson, amputation, and retinopathy.
Figure 2. Aalen‐Johansen cumulative incidence demonstrating heart failure event hospitalizations in weighted cohort.
Met indicates metformin; and Sul, sulfonylurea.
Heart Failure Hospitalization Type
Of the 775 metformin and 992 sulfonylurea users hospitalized for heart failure, 44% (340 out of 775) versus 45% (447 out of 992) did not have an echocardiogram, 28% (214 out of 775) versus 31% (303 out of 992) had reduced EF, 9% (73 out of 775) versus 9% (85 out of 992) had midrange EF, and 19% (148 out of 775) versus 16% (157 out of 992) had preserved EF respectively. Results were consistent in all types of heart failure hospitalizations and statistically significant for heart failure with reduced EF, with a cause‐specific HR of 0.79 (95% CI, 0.68–0.93), and in those with unknown EF, with a cause‐specific HR of 0.84 (95% CI, 0.74–0.96) (Table 2 and Figure 3).
Figure 3. Aalen‐Johansen cumulative incidence for the type of heart failure hospitalization.
A, Heart failure reduced ejection fraction. B, Heart failure midrange ejection fraction. C, Heart failure preserved ejection fraction. D, Heart failure unknown ejection fraction.
Sensitivity and Subgroup Analysis
Sensitivity analysis, which excluded patients with Medicare Advantage, were consistent with the results of the main findings (Table 2). Results were also consistent in all subgroups, although in small subgroups some confidence intervals crossed 1 (Figure 4 and Table S3).
Figure 4. Forest plot demonstrating the adjusted hazard ratio of heart failure hospitalization among patients in different subgroups.
Discussion
Among patients with T2D who developed worsening kidney function, persistent metformin use was associated with reduced heart failure hospitalization compared with sulfonylurea. We found that the risk difference between sulfonylurea and metformin was 0.8% (95% CI, 0.7–0.9) at 5 years. Our results were consistent when we evaluated the outcomes of heart failure hospitalizations with a reduced or unknown EF; the number of events with both midrange and preserved EF heart failure hospitalizations was limited. Although there is consensus that metformin is a first‐line diabetes mellitus treatment, metformin is often discontinued when kidney disease develops. The revised label for metformin use based on the Food and Drug Administration's 2016 safety advisory states that metformin can be safely used in patients with mild kidney function impairment (45–60 mL/min per 1.73 m2) and in some patients with moderate kidney function impairment (eGFR, 30–45 mL/min per 1.73 m2). 8 This study adds to the limited observational evidence for the beneficial association of metformin compared with sulfonylurea on heart failure outcomes among those who develop reduced kidney function. 6 , 31
Our findings are consistent with the results of a study by Masoudi et al., which was restricted to patients with underlying heart failure and evaluated heart failure rehospitalizations among a cohort of 16 417 Medicare beneficiaries with diabetes mellitus. 32 In multivariable models, treatment with metformin was associated with significantly lower risks of heart failure readmission (HR, 0.92; 95% CI, 0.86–0.99) when compared with those who did not use an insulin‐sensitizing agent (sulfonylurea or insulin). These results remained when confined to those with creatinine of >1.5 mg/dL (HR, 0.91; 95% CI, 0.84–0.99). Physiologically, these findings are thought to be related to the nonglycemic cardiac benefits of metformin on insulin sensitization, which include weight neutrality as well as modest improvement in lipoprotein and triglyceride levels. 33 Although the Masoudi et al. study evaluated the association of readmission stratified by EF at the index hospitalization, our study evaluated the association of metformin or sulfonylurea with the type of heart failure hospitalization. This study can potentially inform the association of metformin and sulfonylurea with the physiologic changes that may be associated with heart failure type in certain patient populations.
Although our study has strengths including its large sample size and day‐by‐day ascertainment of medication exposures to reduce misclassification and control for multiple covariates, there are limitations that should be noted. First, incident therapy persistence with either metformin or sulfonylureas at the kidney threshold was required and excluded many patients who discontinued, added, or switched to other medications at or before reaching the kidney threshold. By design, we also excluded those who began diabetes mellitus treatment after the onset of reduced kidney function. This study‐specified criteria allow us to make inferences and interpret study results among a population of patients who continue to use medications after reaching the kidney threshold. Medication changes can occur for multiple reasons, including comorbid status, provider preference, or rate of progression of kidney disease; therefore, these results should not necessarily be extrapolated to those who change or switch medications. Furthermore, newer agents such as DPP4, TZD, GLP1RA, and SGLT2s, although sometimes used as first‐line treatments, were outside of the scope of the study question. Second, because many veterans were not receiving all their care at VHA facilities, some data from hospitalizations, including echocardiography data, were not available. The high percentage of patients without echocardiography data (≈45% of all heart failure hospitalizations) suggests that cardiac care was received outside the VA using their Medicare benefits. Because of the high proportion of missing echocardiography data, the results evaluating association with the heart failure type should be interpreted with caution. Third, cohort entry and the start of follow‐up was either an elevated serum creatinine or reduced eGFR of <60 mL/min per 1.73 m2. It is possible that for some patients this kidney threshold may represent an acute kidney injury event rather than progression to chronic kidney disease; however, the historical eGFR was also reduced, indicating deteriorating kidney function rather than solely an acute event. Fourth, despite use of multiple analytic techniques, including propensity score weighting and covariate adjustment, residual confounding may exist, and furthermore, we did not adjust for the multiple comparisons made in the study. Finally, the study population was mostly elderly White men, and may not represent the larger population of patients with diabetes mellitus and reduced kidney function. Results may not be generalizable to women or populations with lower representation in the VHA system.
In conclusion, this study found that among the population of patients with diabetes mellitus and reduced kidney function, persistent metformin use was associated with reduced heart failure hospitalization compared with sulfonylurea.
Sources of Funding
This project was funded by the by Veterans Affairs (VA) Clinical Science Research and Development Investigator‐Initiated grant CX000570‐08 (Roumie). Drs Richardson, Elasy, Hackstadt, and Roumie were supported in part by the Center for Diabetes Translation Research grant P30DK092986. Support for Veterans Affairs/Centers for Medicare and Medicaid Services data was provided by the Department of Veterans Affairs, Veterans Affairs Health Services Research and Development Service, Veterans Affairs Information Resource Center (project numbers SDR 02‐237 and 98‐004). The VA Clinical Science Research and Development had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication. The contents do not represent the views of the US Department of Veterans Affairs or the United States government. Drs Roumie and Hackstadt 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.
Disclosures
Dr Roumie reports the VA Clinical Science Research and Development investigator‐initiated grant CX000570‐07 that funded this study. The remaining authors have no disclosures to report.
Supporting information
Tables S1–S3
Figures S1–S4
Acknowledgments
The protocol, statistical code, and deidentified and anonymized data set are available from Dr Roumie with a written request.
Author contributions: Design: Roumie, Greevy, Grijalva, Hung, Elasy, Griffin; conduct/data collection: Roumie, Hackstadt, Hung, Greevy, Griffin; analysis: Hackstadt, Greevy; drafting manuscript: Richardson; critical revision of manuscript: Richardson, Griffin, Hackstadt, Hung, Greevy, Grijalva, Elasy, Roumie; funding and administrative oversight: Roumie.
(J Am Heart Assoc. 2021;10:e019211. DOI: 10.1161/JAHA.120.019211.)
Supplementary Material for this article is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.120.019211
For Sources of Funding and Disclosures, see page 11.
REFERENCES
- 1. Dieleman JL, Baral R, Birger M, Bui AL, Bulchis A, Chapin A, Hamavid H, Horst C, Johnson EK, Joseph J, et al. US spending on personal health care and public health, 1996–2013. JAMA. 2016;316:2627–2646. DOI: 10.1001/jama.2016.16885. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. IDF Atlas 9th edition and other resources. n.d. Available at: https://diabetesatlas.org/en/resources/. Accessed April 16, 2020.
- 3. Xu G, Liu B, Sun Y, Du Y, Snetselaar LG, Hu FB, Bao W. Prevalence of diagnosed type 1 and type 2 diabetes among US adults in 2016 and 2017: population based study. BMJ. 2018;362:k1497. DOI: 10.1136/bmj.k1497. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Cavender MA, Steg PHG, Smith SC, Eagle K, Ohman EM, Goto S, Kuder J, Im K, Wilson PWF, Bhatt DL. Impact of diabetes mellitus on hospitalization for heart failure, cardiovascular events, and death. Circulation. 2015;132:923–931. DOI: 10.1161/CIRCULATIONAHA.114.014796. [DOI] [PubMed] [Google Scholar]
- 5. Jackson SL, Tong X, King RJ, Loustalot F, Hong Y, Ritchey MD. National burden of heart failure events in the United States, 2006 to 2014. Circ Heart Fail. 2018;11:e004873. DOI: 10.1161/CIRCHEARTFAILURE.117.004873. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6. Roumie CL, Chipman J, Min JY, Hackstadt AJ, Hung AM, Greevy RA, Grijalva CG, Elasy T, Griffin MR. Association of treatment with metformin vs sulfonylurea with major adverse cardiovascular events among patients with diabetes and reduced kidney function. JAMA. 2019;322:1167–1177. DOI: 10.1001/jama.2019.13206. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7. Messerli FH, Rimoldi SF, Bangalore S. The transition from hypertension to heart failure: contemporary update. JACC Heart Fail. 2017;5:543–551. DOI: 10.1016/j.jchf.2017.04.012. [DOI] [PubMed] [Google Scholar]
- 8. FDA Drug Safety Communication: FDA revises warnings regarding use of the diabetes medicine metformin in certain patients with reduced kidney function. FDA; 2019. [Google Scholar]
- 9. Humensky J, Carretta H, de Groot K, Brown MM, Tarlov E, Hynes DM. Service utilization of veterans dually eligible for VA and Medicare fee‐for‐service: 1999–2004. Medicare Medicaid Res Rev. 2012;2:1999–2004. DOI: 10.5600/mmrr.002.03.A06. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 10. Hynes D, Koelling K, Stroupe K, Arnold N, Mallin K, Sohn M‐W, Weaver F, Manheim L, Kok L. Veterans’ access to and use of Medicare and Veterans Affairs health care. Med Care. 2007;45:214–223. DOI: 10.1097/01.mlr.0000244657.90074.b7. [DOI] [PubMed] [Google Scholar]
- 11. McCarthy JF, Valenstein M, Kim HM, Ilgen M, Zivin K, Blow FC. Suicide mortality among patients receiving care in the Veterans Health Administration health system. Am J Epidemiol. 2009;169:1033–1038. DOI: 10.1093/aje/kwp010. [DOI] [PubMed] [Google Scholar]
- 12. Center of Excellence for Suicide Prevention . Joint Department of Veterans Affairs (VA) and Department of Defense (DoD) Suicide Data Repository—National Death Index (NDI). Available at: http://vaww.virec.research.va.gov/Mortality/Overview.htm. Accessed December 12, 2018.
- 13. Greevy RA, Huizinga MM, Roumie CL, Grijalva CG, Murff H, Liu X, Griffin MR. Comparisons of persistence and durability among three oral antidiabetic therapies using electronic prescription‐fill data: the impact of adherence requirements and stockpiling. Clin Pharmacol Ther. 2011;90:813–819. [DOI] [PubMed] [Google Scholar]
- 14. International Classification of Diseases, Ninth Revision, Clinical Modification. Washington, DC. Public Health Service US Department of Health and Human Services; 1988. [Google Scholar]
- 15. Presley CA, Min JY, Chipman J, Greevy RA, Grijalva CG, Griffin MR, Roumie CL. Validation of an algorithm to identify heart failure hospitalisations in patients with diabetes within the Veterans Health Administration. BMJ Open. 2018;8:e020455. DOI: 10.1136/bmjopen-2017-020455. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16. Saczynski JS, Andrade SE, Harrold LR, Tjia J, Cutrona SL, Dodd KS, Goldberg RJ, Gurwitz JH. A systematic review of validated methods for identifying heart failure using administrative data. Pharmacoepidemiol Drug Saf. 2012;21:129–140. DOI: 10.1002/pds.2313. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17. Patterson OV, Freiberg MS, Skanderson M, J Fodeh S, Brandt CA, DuVall SL. Unlocking echocardiogram measurements for heart disease research through natural language processing. BMC Cardiovasc Disord. 2017;17:151. DOI: 10.1186/s12872-017-0580-8. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18. Skali H, Uno H, Levey AS, Inker LA, Pfeffer MA, Solomon SD. Prognostic assessment of estimated glomerular filtration rate by the new Chronic Kidney Disease Epidemiology Collaboration equation in comparison with the Modification of Diet in Renal Disease Study equation. Am Heart J. 2011;162:548–554. DOI: 10.1016/j.ahj.2011.06.006. [DOI] [PubMed] [Google Scholar]
- 19. Levey AS, Stevens LA, Schmid CH, Zhang Y (Lucy), Castro AF, Feldman HI, Kusek JW, Eggers P, Van Lente F, Greene T, et al. A new equation to estimate glomerular filtration rate. Ann Intern Med. 2009;150:604–612. DOI: 10.7326/0003-4819-150-9-200905050-00006. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Lewis EF, Claggett B, Shah AM, Liu J, Shah SJ, Anand I, O’Meara E, Sweitzer NK, Rouleau JL, Fang JC, et al. Racial differences in characteristics and outcomes of patients with heart failure and preserved ejection fraction in the treatment of preserved cardiac function heart failure trial. Circ Heart Fail. 2018;11:e004457. DOI: 10.1161/CIRCHEARTFAILURE.117.004457. [DOI] [PubMed] [Google Scholar]
- 21. van Buuren S. Flexible Imputation of Missing Data. Boca Raton Florida: Chapman and Hall/CRC Press; 2012. [Google Scholar]
- 22. Parsons LS. 165‐29: performing a 1:N case‐control match on propensity score. n.d.:11.
- 23. D’Agostino RB, Rubin DB. Estimating and using propensity scores with partially missing data. J Am Stat Assoc. 2000;95:749–759. DOI: 10.1080/01621459.2000.10474263. [DOI] [Google Scholar]
- 24. Li L, Greene T. A weighting analogue to pair matching in propensity score analysis. Int J Biostat. 2013;9:215–234. DOI: 10.1515/ijb-2012-0030. [DOI] [PubMed] [Google Scholar]
- 25. Flury BK, Riedwyl H. Standard distance in univariate and multivariate analysis. Am Stat. 1986;40:249–251. [Google Scholar]
- 26. Grambsch PM, Therneau TM. Proportional hazards tests and diagnostics based on weighted residuals. Biometrika. 1994;81:515–526. DOI: 10.1093/biomet/81.3.515. [DOI] [Google Scholar]
- 27. Aalen OO, Johansen S. An empirical transition matrix for non‐homogeneous markov chains based on censored observations. Scand J Stat. 1978;5:141–150. [Google Scholar]
- 28. Austin PC, Lee DS, Fine JP. Introduction to the analysis of survival data in the presence of competing risks. Circulation. 2016;133:601–609. DOI: 10.1161/CIRCULATIONAHA.115.017719. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Virnig B. Strengths and limitations of CMS administrative data in research. Available at: https://www.resdac.org/articles/strengths‐and‐limitations‐cms‐administrative‐data‐research. Accessed June 21, 2020.
- 30. R: The R Project for Statistical Computing. n.d. Available at: https://www.r‐project.org/. Accessed April 16, 2020.
- 31. Crowley MJ, Diamantidis CJ, McDuffie JR, Cameron CB, Stanifer JW, Mock CK, Wang X, Tang S, Nagi A, Kosinski AS, et al. Clinical outcomes of metformin use in populations with chronic kidney disease, congestive heart failure, or chronic liver disease: a systematic review. Ann Intern Med. 2017;166:191. DOI: 10.7326/M16-1901. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Masoudi FA, Inzucchi SE, Wang Y, Havranek EP, Foody JM, Krumholz HM. Thiazolidinediones, metformin, and outcomes in older patients with diabetes and heart failure. Circulation. 2005;111:583–590. DOI: 10.1161/01.CIR.0000154542.13412.B1. [DOI] [PubMed] [Google Scholar]
- 33. DeFronzo RA, Goodman AM; Group the MMS . Efficacy of metformin in patients with non‐insulin‐dependent diabetes mellitus. N Engl J Med. 1995;333:541–549. DOI: 10.1056/NEJM199508313330902. [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Tables S1–S3
Figures S1–S4