Importance
Preferred second-line medication for diabetes treatment after metformin failure remains uncertain.
Objective
To compare time to acute myocardial infarction (AMI), stroke, or death in a cohort of metformin initiators who added insulin or a sulfonylurea.
Design, Setting, and Participants
Retrospective cohort constructed with national Veterans Health Administration, Medicare, and National Death Index databases. The study population comprised veterans initially treated with metformin from 2001 through 2008 who subsequently added either insulin or sulfonylurea. Propensity score matching on characteristics was performed, matching each participant who added insulin to 5 who added a sulfonylurea. Patients were followed through September 2011 for primary analyses or September 2009 for cause-of-death analyses.
Main Outcomes and Measures
Risk of a composite outcome of AMI, stroke hospitalization, or all-cause death was compared between therapies with marginal structural Cox proportional hazard models adjusting for baseline and time-varying demographics, medications, cholesterol level, hemoglobin A1c level, creatinine level, blood pressure, body mass index, and comorbidities.
Results
Among 178 341 metformin monotherapy patients, 2948 added insulin and 39 990 added a sulfonylurea. Propensity score matching yielded 2436 metformin + insulin and 12 180 metformin + sulfonylurea patients. At intensification, patients had received metformin for a median of 14 months (IQR, 5-30), and hemoglobin A1c level was 8.1% (IQR, 7.2%-9.9%). Median follow-up after intensification was 14 months (IQR, 6-29 months). There were 172 vs 634 events for the primary outcome among patients who added insulin vs sulfonylureas, respectively (42.7 vs 32.8 events per 1000 person-years; adjusted hazard ratio [aHR], 1.30; 95% CI, 1.07-1.58; P = .009). Acute myocardial infarction and stroke rates were statistically similar, 41 vs 229 events (10.2 and 11.9 events per 1000 person-years; aHR, 0.88; 95% CI, 0.59-1.30; P = .52), whereas all-cause death rates were 137 vs 444 events, respectively (33.7 and 22.7 events per 1000 person-years; aHR, 1.44; 95% CI, 1.15-1.79; P = .001). There were 54 vs 258 secondary outcomes: AMI, stroke hospitalizations, or cardiovascular deaths (22.8 vs 22.5 events per 1000 person-years; aHR, 0.98; 95% CI, 0.71-1.34; P = .87).
Conclusions and Relevance
Among patients with diabetes who were receiving metformin, the addition of insulin vs a sulfonylurea was associated with an increased risk of a composite of nonfatal cardiovascular outcomes and all-cause mortality. These findings require further investigation to understand risks associated with insulin use in these patients.
Diabetes mellitus and its complications represent an enormous health care burden and result in nearly 200 000 deaths annually. The American Diabetes Association and the European Association for the Study of Diabetes recommend that, for patients with preserved renal function, treatment begin with metformin and lifestyle changes to achieve a glycated hemoglobin (HbA1c) level of less than or equal to 7%. Often patients will require a second agent to reach this goal, but there is no consensus regarding which medication to choose: insulin, sulfonylureas, thiazolidinediones, glucagon-like peptide 1 receptor agonists, or dipeptidyl peptidase 4 inhibitors.1 Evidence to inform treatment choices after metformin monotherapy remains limited.
Clinicians begin administration of insulin to attain fast and flexible control of blood glucose levels. In addition, a few trials have suggested that early insulin initiation is effective in preserving beta cell function.2-4 Accordingly, there has been an increase in early initiation of insulin and its use as add-on therapy to metformin.5-7 However, patients often want to delay insulin initiation because of fears of difficulty with administration, weight gain, and hypoglycemia.
We sought to compare time to cardiovascular disease (CVD) or death among patients who intensified their diabetes treatment with addition of insulin vs a sulfonylurea. We hypothesized that intensification with insulin would be associated with a lower risk of CVD or death compared with sulfonylurea, according to the superiority of insulin in achieving glycemic control.8
Study Design and Data Sources
We assembled a retrospective cohort of Veterans Health Administration (VHA) patients.9 Veterans Health Administration data identified dispensed prescriptions, including medication, date filled, days supplied, pill number, and dosage10; VHA demographic data and International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) coded diagnostic and procedure information identified inpatient and outpatient encounters.11 We collected laboratory results from standard clinical sources. Vital signs data included all outpatient height, weight, and blood pressure measurements. For enrollees in Medicare or Medicaid, we obtained encounter, prescription (Part D), and self-reported race/ethnicity data (coded as white, black, Hispanic, American Indian, Asian/Pacific Islander, other) from the Centers for Medicare & Medicaid Services through VHA’s interagency exchange agreement.12,13 We obtained dates of death from VHA vital status files and cause of death from National Death Index data from VHA National Death Index agreements.14 The institutional review boards of Vanderbilt University and the VHA Tennessee Valley Healthcare System approved this study with waiver of informed consent.
The study population comprised veterans aged 18 years or older who received regular VHA care (encounter or a prescription fill at least once every 180 days) for at least 2 years. Incident users of metformin from October 2001 through September 2008 with at least 365 days of baseline data preceding their first prescription fill who had not filled any diabetic drug prescription within 180 days were identified. These metformin initiators were eligible for the treatment intensification cohort on the date that they subsequently filled either insulin or a sulfonylurea prescription. We selected patients who were adherent to metformin by excluding those with no metformin available on the date of their insulin or sulfonylurea prescription or the previous 180 days. Follow-up began 180 days after the intensified prescription to distinguish patients who continued intensified therapy from those who switched to either insulin or sulfonylurea monotherapy. We excluded patients receiving hospice care or dialysis at intensification.
The exposures of interest were insulin (long-acting, premixed, or short/fast-acting insulin) and sulfonylurea (glyburide, glipizide, or glimepiride) as metformin cotherapies. Follow-up continued through a study outcome or the study end. The study end was September 30, 2011, for all analyses except those that included cause of death as an outcome, for which the study end was September 30, 2009. Patients were censored for loss of follow-up, defined as the 181st day of no contact with any VHA facility (inpatient, outpatient, or pharmacy use); nonpersistence, defined as 90 days without metformin; or prescription for a third antidiabetic drug. In our population, allowing 90 days to refill medications approximates 80% adherence.15
Primary Outcome: CVD and All-Cause Death
The primary composite outcome was acute myocardial infarction, stroke hospitalization, or all-cause death. We defined acute myocardial infarction by a 410.x ICD-9-CM primary discharge diagnosis (positive predictive value 90% vs VHA medical record review). Stroke hospitalizations encompassed patients with a primary discharge diagnosis for ischemic stroke (433.x1, 434 [excluding 434.x0], or 436), intracerebral hemorrhage (431), and subarachnoid hemorrhage (430) and excluded traumatic brain injury (800-804 and 850-854) (positive predictive value 81%).16
We determined all-cause death by using the Vital Status file, which combines information from Medicare, VHA, Social Security, and VHA compensation and pension benefits to determine date of death (sensitivity 98.3%; specificity 99.8% relative to that of the National Death Index).17 When the date of death in the VHA vital status file conflicted with the National Death Index date of death (<3%), we used the latter.
Secondary outcomes included CVD events (acute myocardial infarction and stroke combined), all-cause deaths, and a composite of acute myocardial infarction, stroke, and cardiovascular death (through September 30, 2009). Cardiovascular deaths were identified from death certificates with an International Classification of Diseases, 10th Revision cause of death including I00-I78 (cardiovascular deaths) or R98, R99, R960, and R961 (unattended deaths), excluding I30.X (diseases of the pericardium). This definition included the Centers for Disease Control and Prevention’s broad definition of cardiac death and a validated strategy for identification of sudden cardiac deaths.18
Study covariates were collected in the 730 days before intensification and as time-varying covariates and included age, sex, race (white, black, other), fiscal year, indicators of health care use (hospitalization, months from hospitalization to intensification, nursing home use, number of outpatient visits, and Medicare or Medicaid use in past year), physiologic variables (blood pressure, creatinine level, HbA1c level, low-density lipoprotein levels, presence of proteinuria, and body mass index), duration of metformin monotherapy before intensifying diabetes regimen (diabetes duration), selected medications, smoking, and presence of comorbidities (eTable 1 in the Supplement). Because race can influence study outcome, it was included in all models.19
For patients missing covariates, we conducted multiple imputation with the Markov chain Monte Carlo method and a noninformative Jeffreys prior.20 All covariates from the primary analysis, survival time, and a censoring indicator were included in 20 imputation models and used to compute the final estimates.
The primary analysis was time to the composite: acute myocardial infarction, stroke, or all-cause death in a propensity score–matched cohort. The propensity score modeled the probability of metformin + insulin use, given covariates and Veterans Integrated Service Network of care. Because of size differences between the 2 groups, metformin + insulin observations were propensity score matched to metformin + sulfonylurea observations with a 1:5 optimal matching algorithm21,22 (eTables 2 and 3 and eFigure 1 in the Supplement).
Marginal structural Cox proportional hazards models were used to compare outcomes for metformin + insulin vs metformin + sulfonylurea (referent) while controlling for baseline and time-varying covariates in the matched cohort (eTables 2-4 in the Supplement). Because these model estimates can be unstable in the presence of disproportionately large inverse probability treatment weights,23,24 the primary analysis used stabilized inverse probability treatment weights and truncated weights at 5, the 99th percentile. Thus, the models included the main effects of metformin + insulin vs metformin + sulfonylurea weighting by inverse probability treatment weight. The proportional hazards assumptions were verified through examination of log-log plots. Statistical significance was considered a 2-sided P < .05.
Sensitivity and Subgroup Analyses
First, in an approach similar to that used in intention-to-treat analyses, we used the intensification regimen to define drug exposure and ignored subsequent changes (persistent exposure not required). Because patients were not censored for nonpersistence, this method increases follow-up and events. Second, we changed the stabilized inverse probability treatment weights threshold (untruncated, truncated at 100, and 10). Third, we conducted subgroup analyses, stratifying by CVD history and age (<65 and ≥65 years), to assess effect modification. Among the subgroup with death certificates, we analyzed specific causes of death to identify cardiovascular, cancer, and all other deaths. Finally, we estimated the absolute prevalence difference of a hypothetical unmeasured binary confounder that would be required to yield a statistically nonsignificant association between exposure and outcome.25 We assumed a confounder-outcome association similar to our observed covariates (hazard ratio = 1.25) and considered a broad range of confounder prevalences in both exposures.
Analyses were conducted with R (http://www.r-project.org; modules optmatch26 and RI tools27) and SAS version 9.2 (modules Proc MI, ProcPHREG for marginal structural Cox proportional hazards models, and Proc Lifetest).
Study Cohort and Patient Characteristics
There were 178 341 patients who initiated metformin during 2001-2008. Fifty-two percent (n = 92 045) never intensified therapy (median follow-up, 50 months [IQR, 19-67 months]), 6% (n = 9851) stopped filling metformin prescriptions, and 2% (n = 3577) had fewer than 6 months of follow-up. Among the remaining 41% (n = 72 868) of metformin initiators who started receiving another therapy, 40% (n = 29 523) were excluded because their regimen excluded metformin or included nonstudy medications.
Fifty-nine percent (43 345/72 868) of metformin patients intensified with 1 of the 2 study regimens. We excluded less than 1% (n = 407) of patients with data errors (n = 370), hospice (n = 0), or dialysis (n = 37). The cohort included 2948 patients (7%) who added insulin (47% long-acting, 22% both long- and short-acting, 17% premixed, and 11% short-acting) and 39 990 patients (92%) who added a sulfonylurea (55% glipizide, 43% glyburide, and 2% glimepiride). Seventy-six percent of matched patients who died had a death certificate available (Figure 1).
Patients were 95% men and 70% white. Compared with patients who added a sulfonylurea, patients who added insulin to metformin intensified therapy earlier (14 months vs 18 months), had higher median HbA1c levels (8.5% vs 7.5%), and had a higher prevalence of comorbidities. The proportion of patients prescribed metformin + insulin increased over time, with the odds increasing by an average of 17% (IQR, 14%-20%) per year (P < .001). After propensity score matching, we included 14 616 patients: 2436 metformin + insulin and 12 180 metformin + sulfonylurea. Baseline characteristics were not statistically different (Table 1).
The most common reasons for censoring were therapy change (58.7% metformin + insulin vs 61.7% metformin + sulfonylurea), leaving the VHA (1.3% vs 2.5%), or reaching the study end (32.9% vs 30.6%). The median number of years before censoring or the outcome was 1.15 (IQR, 0.5-2.4) among metformin + insulin patients and 1.15 (IQR, 0.5-2.2) among metformin + sulfonylurea patients. At 1 year, median HbA1c level declined to 7% (IQR, 6.3%-8.0%) among metformin + insulin users and 6.9% (IQR, 6.4%-7.7%) among metformin + sulfonylurea users. Patient characteristics at 1 year were not statistically different (eTable 5 in the Supplement).
Absolute and Relative Hazards of Cardiovascular Events and Deaths
There were 172 vs 634 events for the primary outcome among patients who added insulin vs sulfonylureas, respectively (42.7 vs 32.8 events per 1000 person-years; adjusted hazard ratio [aHR], 1.30; 95% CI, 1.07-1.58; P = .009) (Table 2, Figure 2A). Cardiovascular disease (acute myocardial infarction and stroke) events were 41 and 229 among patients who added insulin or sulfonylurea, respectively (10.2 and 11.9 per 1000 person-years; aHR, 0.88; 95% CI, 0.59-1.30; P = .52). All-cause deaths were 137 vs 444, respectively (33.7 and 22.7 per 1000 person-years; aHR, 1.44; 95% CI, 1.15-1.79; P = .001).
For the secondary outcome, fatal and nonfatal cardiovascular events, there were 54 vs 258 events (22.8 vs 22.5 per 1000 person-years; aHR, 0.98; 95% CI, 0.71-1.34; P = .87) (Table 2, Figure 2B).
Sensitivity and Subgroup Analyses
In sensitivity analysis in which persistent exposure was not required, there were 394 events for the primary outcome among those who added insulin (7456 person-years) and 1553 events among those who added a sulfonylurea (37 237 person-years), yielding 52.8 (IQR, 48.0-58.2) and 41.7 (IQR, 39.7-43.8) events per 1000 person-years, respectively (aHR, 1.29; 95% CI, 1.15-1.46; P < .001). Results of the stabilized nontruncated weights in the analysis of persistent exposure not required yielded comparable results (aHR, 1.30; 95% CI, 1.15-1.46; P < .001). Marginal structural Cox proportional hazards models analyses, which varied the threshold of the maximal stabilized weight, yielded consistent results (eFigure 2 in the Supplement). No interaction between exposure and CVD history was detected (P = .78). Subgroup analyses stratifying by CVD or age were consistent with the primary analysis, but CIs were wide (eFigure 3 in the Supplement). In separate analyses that evaluated cause of death, the aHRs for metformin + insulin vs metformin + sulfonylureas were increased for all groups, but statistically significant only for cancer death (Table 3). Assuming an association comparable to our measured covariates (ie, hazard ratio, 1.25), an unmeasured binary confounder would need to be 30% higher among metformin + insulin users compared with metformin + sulfonylurea users to yield nonsignificant results in the main findings and 70% higher to yield statistical nonsignificance in the outcome of all-cause mortality (eTables 6 and 7 in the Supplement).
Among patients with diabetes who were receiving metformin, the addition of insulin compared with a sulfonylurea was associated with an increased hazard of a composite of nonfatal cardiovascular outcomes and all-cause mortality. There is consensus that metformin is first-line diabetes treatment; however, uncertainty remains regarding additional therapy after inadequate control with metformin. Among the options, intensification with either insulin or a sulfonylurea is considered a high-efficacy strategy with reasonable costs.1
Although sulfonylurea use predominated as add-on therapy, we observed increasing use of insulin intensification during the study. Reasons may include increasing prevalence of obesity and insulin resistance, emphasis on metrics such as glycemic targets,28,29 increasing comfort with newer analog insulins, and benefit in microvascular outcome prevention.30
Two large randomized trials demonstrated that regimens including greater insulin use and tighter control did not reduce cardiovascular events compared with standard care. In the Action to Control Cardiovascular Risk in Diabetes (ACCORD) trial, patients were randomized to intensive control (target HbA1c level <6%) or standard care. Approximately 77% of patients in the intensive group received insulin compared with 55% in the standard group.31 ACCORD was stopped when interim analyses found more all-cause deaths in the intensive vs standard group (5.0% vs 4.0%; hazard ratio, 1.22; 95% CI, 1.01-1.46). Most excess mortality was due to cardiovascular deaths (2.63% vs 1.83% during a mean 3.5 observation years; P = .02). Whether insulin itself or other effects of intense treatment such as hypoglycemia32,33 contributed to the increased mortality remains unknown.
The Outcome Reduction With an Initial Glargine Intervention (ORIGIN) trial randomized 12 537 patients with CVD risk factors and prediabetes or diabetes to insulin glargine or standard care. Metformin (28%) and sulfonylurea (29%) use was similar in both groups, but insulin reached only 11% in the standard group by study end. After a median of 6 years, there was no difference in the incidence of cardiovascular death, myocardial infarction, or stroke between groups (2.94 vs 2.85 per 100 person-years, respectively; hazard ratio, 1.02; 95% CI, 0.94-1.11).34 There was also no difference in the incidence of cancer or cancer death. However, patients in the insulin group had more weight gain and hypoglycemic events.
Several observational studies have also reported no cardiovascular benefit of insulin relative to noninsulin comparators, and some have suggested worse outcomes. A Canadian study35 reported increased all-cause mortality among insulin users compared with nonusers in a dose-response manner. Similarly, a study of primary care patients in the UK General Practice Research database36 determined that metformin + insulin was associated with an elevated risk of all-cause mortality and cardiovascular- and cancer-related outcomes compared with metformin monotherapy. However, these studies did not address confounding by disease severity adequately. The first did not control for HbA1c, and the second compared more intensive therapies such as insulin (alone or in combination) with metformin monotherapy.
Our finding of a modestly increased risk of a composite of cardiovascular events and death in metformin users who add insulin compared with sulfonylurea is consistent with the available clinical trial and observational data. None of these studies found an advantage of insulin compared with oral agents for cardiovascular risk, and several reported increased cardiovascular risk or weight gain and hypoglycemic episodes, which could result in poorer outcomes. Although insulin remains a reasonable option for patients who have very high glucose levels or who desire flexible and fast glucose reduction, most patients prefer to delay insulin initiation.37 Our study suggests that intensification of metformin with insulin among patients who could add a sulfonylurea (HbA1c level less than ≈10%) offers no advantage in regard to risk of cardiovascular events and is associated with some risk.
Our findings must be interpreted in light of limitations. Although we applied an extensive set of strategies to address confounding by indication, including rigorous selection criteria, propensity score matching, and marginal structural models, residual confounding from difficult-to-measure factors such as patient frailty or diabetes severity remains possible. Nevertheless, our sensitivity analyses estimated that a large confounding effect would be needed for an unmeasured confounder to explain our observations. Using similar methods in a VHA diabetes cohort, we previously demonstrated drug effects on lipids, HbA1c, and body mass index that were concordant with that of clinical trials and meta-analyses.9,38-40 Our results are consistent with UK Prospective Diabetes Study results, which demonstrated a reduction of cardiovascular events with metformin but not insulin or sulfonylurea.
There are several other limitations. We used refill data as a proxy for receipt of medication. Nevertheless, prescription fills are a good proxy for medication use. Veterans may not receive all their care or medications in VHA facilities,12,13 resulting in missing events or medications, which we partially addressed through supplementation with Medicare and Medicaid information. Because we required patients to persist in receiving their medications, censoring was high. In addition, patients who added insulin composed only 7% of intensifiers, which resulted in a relatively small sample size and limited the precision of some estimates. The statistical significance of our primary outcome was driven by all-cause mortality, and a clinically significant cardiovascular benefit could not be excluded. Our primary analyses considered a matched population. Some patients who were prescribed metformin + insulin did not match metformin + sulfonylurea users. Excluded metformin + insulin users (N = 512) had a median HbA1c level of 11.8% and 67% of these patients were hospitalized in the 90 days before insulin initiation (eTable 8 in the Supplement). Results can be generalized only to metformin patients who were eligible to add either medication. Finally, our patients reflect a typical veteran population, with most patients being white men.
Among patients with diabetes who are receiving metformin, the addition of insulin compared with sulfonylurea was associated with an increased risk of a composite of nonfatal cardiovascular outcomes and all-cause mortality. These findings require further investigation to understand risks associated with insulin use in these patients and call into question recommendations that insulin is equivalent to sulfonylureas for patients who may be able to receive an oral agent.
Corresponding Author: Christianne L. Roumie, MD, MPH, Nashville VA Medical Center, 1310 24th Ave S, Geriatric Research Education Clinical Center, Nashville, TN 37212 (christianne.roumie@vanderbilt.edu).
Author Contributions: Drs Roumie and Greevy had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Roumie, Greevy, Grijalva, Hung, Elasy, Griffin.
Acquisition, analysis, or interpretation of data: Roumie, Greevy, Grijalva, Hung, Liu, Murff, Griffin.
Drafting of the manuscript: Roumie, Greevy.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Greevy, Liu.
Obtained funding: Roumie, Greevy, Griffin.
Administrative, technical, or material support: Murff, Elasy.
Study supervision: Greevy, Grijalva, Griffin.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: This project was funded by the Agency for Healthcare Research and Quality, US Department of Health and Human Services, contract HHSA2902010000161, as part of the Developing Evidence to Inform Decisions About Effectiveness (DEcIDE 2) program. This work was supported in part by VA Clinical Science Research and Development investigator-initiated grant I01CX000570-01 (Dr Roumie) and the Center for Diabetes Translation Research P30DK092986 (Drs Roumie and Elasy). Dr Hung (2-031-09S) was supported by a VA Career Development Award. Support for Veterans Affairs/Centers for Medicare & 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).
Role of the Sponsors: The funders had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; in the preparation, review, or approval of the manuscript; or in the decision to submit the manuscript for publication.
Disclaimer: The authors of this report are responsible for its content. Statements in the report should not be construed as endorsement by the Agency for Healthcare Research and Quality, the US Department of Health and Human Services, or the Department of Veterans Affairs.
1.Inzucchi
SE, Bergenstal
RM, Buse
JB,
et al; American Diabetes Association (ADA); European Association for the Study of Diabetes (EASD). Management of hyperglycemia in type 2 diabetes: a patient-centered approach: position statement of the American Diabetes Association (ADA) and the European Association for the Study of Diabetes (EASD).
Diabetes Care. 2012;35(6):1364-1379.
PubMedGoogle ScholarCrossref 2.Weng
J, Li
Y, Xu
W,
et al. Effect of intensive insulin therapy on beta-cell function and glycaemic control in patients with newly diagnosed type 2 diabetes: a multicentre randomised parallel-group trial.
Lancet. 2008;371(9626):1753-1760.
PubMedGoogle ScholarCrossref 3.Harrison
LB, Adams-Huet
B, Raskin
P, Lingvay
I. β-Cell function preservation after 3.5 years of intensive diabetes therapy.
Diabetes Care. 2012;35(7):1406-1412.
PubMedGoogle ScholarCrossref 4.Alvarsson
M, Sundkvist
G, Lager
I,
et al. Effects of insulin vs glibenclamide in recently diagnosed patients with type 2 diabetes: a 4-year follow-up.
Diabetes Obes Metab. 2008;10(5):421-429.
PubMedGoogle ScholarCrossref 6.Cohen
FJ, Neslusan
CA, Conklin
JE, Song
X. Recent antihyperglycemic prescribing trends for US privately insured patients with type 2 diabetes.
Diabetes Care. 2003;26(6):1847-1851.
PubMedGoogle ScholarCrossref 7.Holden
SE, Poole
CD, Morgan
CL, Currie
CJ. Evaluation of the incremental cost to the National Health Service of prescribing analogue insulin.
BMJ Open. 2011;1(2):e000258.
PubMedGoogle ScholarCrossref 8.Holman
RR, Paul
SK, Bethel
MA, Matthews
DR, Neil
HA. 10-Year follow-up of intensive glucose control in type 2 diabetes.
N Engl J Med. 2008;359(15):1577-1589.
PubMedGoogle ScholarCrossref 9.Hung
AM, Roumie
CL, Greevy
RA,
et al. Comparative effectiveness of incident oral antidiabetic drugs on kidney function.
Kidney Int. 2012;81(7):698-706.
PubMedGoogle ScholarCrossref 10.Arnold
N, Hines
D, Stroupe
K. Comparison of VA Outpatient Prescriptions in the DSS Datasets and the PBM Database. Hines, IL: Edward Hines, Jr VA Hospital; 2006.
11.National Center for Health Statistics. International Classification of Diseases, Ninth Revision, Clinical Modification. Washington, DC: Public Health Service, US Dept of Health & Human Services; 1988.
12.Humensky
J, Carretta
H, de Groot
K, Brown
MM, Tarlov
E, Hynes
D. Service utilization of veterans dually eligible for VA and Medicare fee-for-service: 1999–2004.
Medicare Medicaid Res Rev. 2012;2(3).
Google Scholar 13.Hynes
DM, Koelling
K, Stroupe
K,
et al. Veterans’ access to and use of Medicare and Veterans Affairs health care.
Med Care. 2007;45(3):214-223.
PubMedGoogle ScholarCrossref 14.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(8):1033-1038.
PubMedGoogle ScholarCrossref 15.Greevy
RA
Jr, Huizinga
MM, Roumie
CL,
et al. 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(6):813-819.
PubMedGoogle ScholarCrossref 16.Niesner
K, Murff
HJ, Griffin
MR,
et al. Validation of VA administrative data algorithms for identifying cardiovascular disease hospitalization.
Epidemiology. 2013;24(2):334-335.
PubMedGoogle ScholarCrossref 17.Sohn
MW, Arnold
N, Maynard
C, Hynes
DM. Accuracy and completeness of mortality data in the Department of Veterans Affairs.
Popul Health Metr. 2006;4:2.
PubMedGoogle ScholarCrossref 18.Ray
WA, Meredith
S, Thapa
PB, Meador
KG, Hall
K, Murray
KT. Antipsychotics and the risk of sudden cardiac death.
Arch Gen Psychiatry. 2001;58(12):1161-1167.
PubMedGoogle ScholarCrossref 20.Yuan
YC. Multiple Imputation for Missing Data: Concepts and New Development (Version 9.0). SAS; 2011.
22.D'Agostino
R, Rubin
D. Estimating and using propensity scores with partially missing data.
J Am Stat Assoc. 2000;95(451):749-759.
Google ScholarCrossref 23.Robins
JM, Hernán
MA, Brumback
B. Marginal structural models and causal inference in epidemiology.
Epidemiology. 2000;11(5):550-560.
PubMedGoogle ScholarCrossref 24.Hernán
MA, Brumback
B, Robins
JM. Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men.
Epidemiology. 2000;11(5):561-570.
PubMedGoogle ScholarCrossref 25.Schneeweiss
S, Glynn
RJ, Tsai
EH, Avorn
J, Solomon
DH. Adjusting for unmeasured confounders in pharmacoepidemiologic claims data using external information: the example of COX2 inhibitors and myocardial infarction.
Epidemiology. 2005;16(1):17-24.
PubMedGoogle ScholarCrossref 26. Optmatch: Functions for Optimal Matching [computer program]. Version R package 0.9-1. 2010.
27. RItools: Randomization Inference Tools [computer program]. Version: R package 0.1-11. 2010.
28.Selby
JV, Uratsu
CS, Fireman
B,
et al. Treatment intensification and risk factor control: toward more clinically relevant quality measures.
Med Care. 2009;47(4):395-402.
PubMedGoogle ScholarCrossref 29.Kerr
EA, Gerzoff
RB, Krein
SL,
et al. Diabetes care quality in the Veterans Affairs Health Care System and commercial managed care: the TRIAD study.
Ann Intern Med. 2004;141(4):272-281.
PubMedGoogle ScholarCrossref 30.UK Prospective Diabetes Study (UKPDS) Group. Intensive blood-glucose control with sulphonylureas or insulin compared with conventional treatment and risk of complications in patients with type 2 diabetes (UKPDS 33).
Lancet. 1998;352(9131):837-853.
PubMedGoogle ScholarCrossref 31.Gerstein
HC, Miller
ME, Byington
RP,
et al; Action to Control Cardiovascular Risk in Diabetes Study Group. Effects of intensive glucose lowering in type 2 diabetes.
N Engl J Med. 2008;358(24):2545-2559.
PubMedGoogle ScholarCrossref 32.Bonds
DE, Miller
ME, Bergenstal
RM,
et al. The association between symptomatic, severe hypoglycaemia and mortality in type 2 diabetes: retrospective epidemiological analysis of the ACCORD study.
BMJ. 2010;340:b4909.
PubMedGoogle ScholarCrossref 33.Bonds
DE, Miller
ME, Dudl
J,
et al. Severe hypoglycemia symptoms, antecedent behaviors, immediate consequences and association with glycemia medication usage: secondary analysis of the ACCORD clinical trial data.
BMC Endocr Disord. 2012;12:5.
PubMedGoogle ScholarCrossref 34.Gerstein
HC, Bosch
J, Dagenais
GR,
et al; ORIGIN Trial Investigators. Basal insulin and cardiovascular and other outcomes in dysglycemia.
N Engl J Med. 2012;367(4):319-328.
PubMedGoogle ScholarCrossref 35.Gamble
JM, Simpson
SH, Eurich
DT, Majumdar
SR, Johnson
JA. Insulin use and increased risk of mortality in type 2 diabetes: a cohort study.
Diabetes Obes Metab. 2010;12(1):47-53.
PubMedGoogle ScholarCrossref 36.Currie
CJ, Poole
CD, Evans
M, Peters
JR, Morgan
CL. Mortality and other important diabetes-related outcomes with insulin vs other antihyperglycemic therapies in type 2 diabetes.
J Clin Endocrinol Metab. 2013;98(2):668-677.
PubMedGoogle ScholarCrossref 37.Peyrot
M, Rubin
RR, Khunti
K. Addressing barriers to initiation of insulin in patients with type 2 diabetes.
Prim Care Diabetes. 2010;4(suppl 1):S11-S18.
PubMedGoogle ScholarCrossref 38.Roumie
CL, Huizinga
MM, Liu
X,
et al. The effect of incident antidiabetic regimens on lipid profiles in veterans with type 2 diabetes: a retrospective cohort.
Pharmacoepidemiol Drug Saf. 2011;20(1):36-44.
PubMedGoogle ScholarCrossref 39.Roumie
CL, Hung
AM, Greevy
RA,
et al. Comparative effectiveness of sulfonylurea and metformin monotherapy on cardiovascular events in type 2 diabetes mellitus: a cohort study.
Ann Intern Med. 2012;157(9):601-610.
PubMedGoogle ScholarCrossref 40.Roumie
CL, Liu
X, Choma
NN,
et al. Initiation of sulfonylureas versus metformin is associated with higher blood pressure at one year.
Pharmacoepidemiol Drug Saf. 2012;21(5):515-523.
PubMedGoogle ScholarCrossref