Comparative cardiovascular effects of GLP‐1 agonists using real‐world data

Abstract Aims There is limited research using real‐world data to evaluate protective cardiovascular effects of glucagon‐like peptide‐1 (GLP‐1) agonists among adults with type 2 diabetes (T2D) early in treatment. Materials and Methods We conducted a retrospective, active comparator cohort study using 2011–2015 administrative claims data to compare cardiovascular disease (CVD) event rates following initiation of exenatide extended‐release (E‐ER), exenatide immediate‐release (E‐IR) or liraglutide in T2D adults who previously received no other antidiabetic medication (ADM) except metformin. The primary outcome was time to first major adverse CVD event (ischaemic heart disease, stroke, congestive heart failure or peripheral arterial disease) after starting GLP‐1. Cox proportional hazards regression was used to model the association between index GLP‐1 and CVD events, adjusting for baseline patient, prescriber and plan characteristics. Primary analyses included all patients with ≥2 prescription fills for the index GLP‐1, regardless of subsequent refill adherence or initiation of other ADM after index date. Results Compared with liraglutide, neither E‐ER nor E‐IR was associated with risk of composite major CVD events (hazard ratios [HRs] for E‐ER and E‐IR: 1.33 [95% C.I. 0.73–2.39] and 1.30 [0.81–2.09]). No associations were observed between event rates for individual CVD components. The HR for an ischaemic event with E‐IR relative to liraglutide was 1.85 (95% C.I. 0.97–3.53). Adjusting for time‐varying exposure to other ADM and CVD medications after index date produced similar results. Conclusions Initiating either immediate or extended‐release exenatide rather than liraglutide was not associated with significant differences in CVD risk in this observational real‐world study.


| Real-world data sources
Data sources included patients' health plan enrolment files, pharmacy claims, and medical claims for both inpatient and ambulatory care provided from a large health payer. Diagnosis codes from medical claims were coded according to the International Classification of Diseases, 9th Revision (ICD-9). Pharmacy claims were used to identify different ADM fills, including new starts of GLP-1. Individual race/ethnicity data were imputed by the data vendor using individual-and area-level characteristics.

| Participants
The study population was comprised of adults with evidence of diabetes mellitus (≥1 ambulatory encounter with ICD-9-CM 250.XX) and each of the following: (1) a first pharmacy dispensing event for a GLP-1 medication [fill date defined as the index date]; (2) at least one refill for this GLP-1 to indicate a 'true start;' (3) no evidence of any ADM fill other than metformin before the index date; and (4) continuous enrolment to provide data for ≥365 days before and ≥90 days after the index date. We excluded patients with evidence of pregnancy, type 1 diabetes (any encounter with ICD-9-CM: 250. X1, 250.X3), and conditions or medications that might cause secondary diabetes. Detailed definitions of the eligibility criteria and all study variables are provided in the Appendix A.

| Exposure and outcomes
The main exposure of interest was a categorical variable identifying the specific GLP-1 receptor agonist that was initiated, as evidenced by a pharmacy claim, followed by a second claim (to demonstrate that the patient successfully initiated the drug

| Covariates
Demographic data included patients' age, sex and race/ethnicity. Though laboratory data are typically not available in administrative claims, we did have access to the result of a haemoglobin A1c test for 34.7% of patients, so used this additional information to categorize all patients by baseline A1c result status:: <8.0%, 8.0-10.0%, >10.0%, or result not available. An indicator for metformin use at baseline was also included. We used diagnostic codes from inpatient and/or ambulatory medical claims prior to the index date to estimate whether each patient had a past CVD event, a prior microvascular complication of diabetes (i.e. diabetic nephropathy, neuropathy and retinopathy), or evidence of another CVD risk factor (i.e. chronic kidney disease, dyslipidaemia, hypertension, overweight/obesity, tobacco use and family history of CVD). Similarly, we used pharmacy claims to identify patients' baseline use of the following medications known to impact CVD risk: angiotensin-converting enzyme (ACE) inhibitors, angiotensin receptor blockers (ARBs), aldosterone receptor antagonists, antiplatelet drugs, beta blockers, calcium channel blockers, diuretics, HMG CoA reductase inhibitors (statins) and other lipidlowering drugs, as well as their exposure to these same drugs or to other ADM beginning after their index date. In addition, time-varying covariates were constructed to capture changes in the use of these medications during follow-up, as well as exposure to other ADMs started after the index date, and CV medications. Finally, we used pharmacy data to define prescriber specialty of prescribers, patient's health plan and geographic region in which care was delivered.

| Statistical analysis
Summary statistics characterized the study population with respect to all pre-and post-index covariates. Chi-square tests were used to examine associations between baseline covariates and the index GLP-1 medication (Table 1). Because relatively few patients received albiglutide (n = 360) and dulaglutide (n = 99), evaluation of CVD outcomes was restricted to those starting liraglutide, exenatide ER or exenatide IR. Unadjusted Kaplan-Meier curves were plotted to display the primary outcome. After testing proportionality assumptions, Cox proportional hazards regression was used to model the association between the index GLP-1 receptor agonist and the composite cardiovascular events, adjusting for baseline patient, prescriber and health plan characteristics. As a subgroup analyses, we estimated relative risk for the primary outcome separately for patients on metformin at baseline and patients who were not on metformin at baseline. We also attempted to compare event rates for a high-risk subgroup of patients who had evidence of prior CVD events in the baseline period, but the subgroup was too small to generate stable hazard ratio estimates. In separate models, we examined individual component events as secondary outcomes. Because liraglutide was the most frequently prescribed, it serves as the comparison group in all models.
We conducted sensitivity analyses to assess the robustness of our findings and to explore the influence of GLP-1 medication adherence or of differing exposures to other medications that could influence CVD event rates after the index date. We included additional covariates to account for time-varying exposure to ADM and CVD drugs during follow-up. We also conducted subgroup analyses of only those patients in each group who had at least 6 months of continuous refill adherence to their index GLP-1 (defined as adherent) and who did not start another ADM after the index date.  (Table 1). Differences were also noted in the prevalence of obesity identified by ICD-9 diagnosis code between groups; patients starting albiglutide and dulaglutide 46.9% and 49.5% vs. 41.4% for E-ER and 28.3% E-IR and 37.3% for liraglutide, p-value <.0001. There were also small but appreciable between-group differences in CVD related medication use; for example, the percentages of patients in each group who were prescribed ACE inhibitors or ARBs before their index dates were as follows: 41.4% for E-ER; 38.6% for E-IR; 36.9% for liraglutide; and 35.0% for albiglutide, p = .004). Among the 3 larger comparison groups considered in the primary outcome analysis, evidence of prior CVD events was similar. Kaplan-Meier curves were plotted to display the primary outcome ( Figure 1). See Table S1 for number of events and unadjusted cumulative incidence rates.

| RE SULTS
In multivariable Cox models (Table 2)

| D ISCUSS I ON AND CON CLUS I ON
In pre-clinical and clinical studies, GLP-1 agonists have shown benefits including CVD risk factor reduction and glucose reduction. 9,10 Diabetes algorithms in clinical guidelines, which encourage use of medications such as GLP-1 in those with CVD diagnoses, have implications for our highest risk patients. 5,7 However, in pre-clinical studies, potential differences between drugs within the same class have been proposed based on time of action (half-life) and differences at the molecular level. 11 Data from large cardiovascular outcomes studies show mixed results in large high-risk patient populations. [1][2][3][4]  RTA had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
We would like to thank Susan Karam MD for her contributions to the manuscript.

CO N FLI C T O F I NTE R E S T
AW discloses that she has received research grant support from Eli Lilly and Novo Nordisk all of which are unrelated to this work.

DATA AVA I L A B I L I T Y S TAT E M E N T
Data sharing is not applicable to this article as no new data were created or analyzed in this study.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found in the online version of the article at the publisher's website.