Risk of heart failure hospitalization among users of dipeptidyl peptidase-4 inhibitors compared to glucagon-like peptide-1 receptor agonists

This was a retrospective cohort analysis using Truven Health Analytic MarketScan^® commercial and Medicare supplemental databases between January 2006 and December 2013. The databases include information on inpatient and outpatient medical claims, health expenditures, enrollment, and outpatient prescription claims of more than 150 million individuals. The commercial data includes privately insured employees and their dependents who are covered under plans including preferred provider organizations, health maintenance organizations, point-of-service plans, and indemnity plans. This study was approved by the University of Florida Institutional Review Board.

New-users of DPP-4 inhibitors (sitagliptin, saxagliptin, alogliptin, or linagliptin) or GLP-1 agonists (exenatide, liraglutide, albiglutide, and dulaglutide) were identified based on the absence of any prior use during the baseline period (12 months lookback period). The start date of the first eligible prescription was used as the index date. To be included in the cohort, patients were required to have a diagnosis of type 2 diabetes mellitus [International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM): 250.x0 or 250.x2] based on one inpatient or two outpatient encounters at two different service dates within 1 year prior to the index date, have 12 months of continuous enrollment in medical and pharmacy benefits prior to the index date, and be aged ≥ 18 years on cohort entry. Patients were excluded if they had a diagnosis of type 1 diabetes (ICD-9-CM: 250.x1 or 250.x3), end-stage renal disease (ESRD) (ICD-9-CM: 585.6), a prescription for DPP-4 inhibitors or GLP-1 agonists during the baseline period, or if they initiated both drugs on the same date. Further, to make our results comparable with those from the SAVOR-TIMI and other observational analyses, we excluded patients who had a hospitalization due to HF (ICD-9-CM: 402.×1, 404.×1, 404.×3, and 428.×) during the 60 days prior to the index date because a recent hospitalization due to HF is known to be highly predictive of subsequent HF exacerbations [2, 10].

Patients were considered exposed as long as they continued to refill their medication, allowing a gap no longer than 7 days between refills. Follow-up started from the index date until the first occurrence of one of the following censoring criteria: occurrence of the primary outcome (hospitalization due to HF), initiation of the study comparator (e.g. new users of DPP-4 inhibitors start or switch to GLP-1 agonists), discontinuation of therapy (i.e. no fill for 7 days), inpatient death (death was recorded if occurring during inpatient stay only), end of enrollment, or end of study period.

The study outcome was the first hospitalization due to HF defined as an inpatient hospital claim with a primary or secondary diagnosis of HF (ICD-9-CM: 402.×1, 404.×1, 404.×3, and 428.××). This definition has been validated previously with a positive predictive value [PPV] > 90% [14].

We performed three secondary analyses, first, stratifying by the presence of prior diagnosis of HF defined as the occurrence of HF between 1 year and 60 days prior to the index date. Secondly, we stratified by the presence of prior CVD [cerebrovascular diseases, myocardial infarction (MI), stroke, coronary atherosclerosis, ischemic heart diseases, angina pectoris]. Thirdly, we assessed HF hospitalization risk of individual DPP-4 inhibitors, saxagliptin and sitagliptin, compared to GLP-1 agonists.

To adjust for confounders and disease risk factors, propensity score (1:1) matching to the nearest neighbor was used. To estimate the propensity score, we used a logistic regression model to obtain the predicted probability of receiving DPP-4 inhibitors versus GLP-1 agonists as a function of pre-specified baseline covariates. Included covariates were demographics (age, sex), presence of comorbidities (asthma, chronic obstructive pulmonary disease [COPD], chronic kidney disease [CKD], ischemic heart disease, depression, stroke, hypoglycemia, hyperlipidemia, cancer, hypertension, MI, HF), use of other diabetes medications (insulin, metformin, thiazolidinediones [TZDs], α-Glucosidase inhibitors, sodium glucose co-transporter-2 [SGLT2] inhibitors, meglitinides, sulfonylureas), use of other medications (angiotensin-converting enzyme [ACE] inhibitors, aldosterone receptor blockers, β-blockers, calcium channel blockers, loop diuretics, potassium sparing diuretics, thiazide diuretics, and direct vasodilators), and measures of healthcare utilizations (total number of inpatient visits, total number of outpatient visits). The presence of baseline comorbidities were defined based on the presence of ICD-9 codes in either inpatient or outpatient claims. These variables were drawn from a prior protocol that was developed by the Mini-Sentinel for the assessment of cardiovascular outcomes among patients with type 2 diabetes [15].

Descriptive statistics were summarized using mean for continuous variables and proportion for categorical variables. Demographics and disease risk factors were compared between users of DPP-4 inhibitors and those of GLP-1 agonists using a Chi square test for categorical variables and independent t test for continuous variables. After propensity score matching, standardized differences were used to examine the balance in patient characteristics, where imbalance was defined as an absolute value higher than 0.2 [16, 17]. Proportional hazard models were used to obtain the HR and associated 95% CI after propensity score matching in the primary and secondary analyses. In secondary analyses, we examined the heterogeneity of treatment effect in patients with and without prior HF, with and without baseline CVD, and comparing saxagliptin and sitagliptin to GLP-1 agonists.

Sensitivity analyses were conducted to examine the robustness of the primary study results by changing the outcome definition to primary diagnosis only and, separately, extending the follow-up time to 45 days post-treatment discontinuation. Propensity score matching was performed for each of the secondary and sensitivity analyses. All analyses were performed using SAS 9.4 (SAS Institute Inc., Cary, NC).

A flow diagram of the study population is included in the Additional file 1: Figure S1. Prior to matching, a total of 358,632 new-users of DPP-4 inhibitors, and 174,711 new-users of GLP-1 agonists were identified. Table 1 summarizes baseline characteristics comparing the new-users of DPP-4 inhibitors to GLP-1 agonists users. Prior to matching, there were significant differences in patient demographics and clinical characteristics.

After propensity score matching, a total of 321,606 patients were included in the analysis (n = 160,803 each for DPP-4 inhibitors and GLP-1 agonists users). Table 2 summarizes patient demographics and disease characteristics post-matching. In the propensity score matched cohort, patient demographics, including age (mean, 52.7 years vs. 53.0 years), proportion of men (44.7% vs. 44.1%), presence of comorbidities such as asthma (6.4% vs. 6.4%), hyperlipidemia (57.4% vs. 56.8%), and hypertension (59.6% vs. 59.5%), and prior use of medications including metformin (62.2% vs. 62.0%), TZDs (24.4% vs. 24.5%), and β-blockers (22.5% vs. 23.0%) were comparable between groups.

The mean length of follow-up was 170 (± 290) days in the DPP-4 inhibitors group and 159 (± 285) days in the GLP-1 agonists group. Table 3 shows the unadjusted risk of hospitalization due to HF with DPP-4 inhibitors versus GLP-1 agonists users in the propensity score matched cohort. We identified 5714 of cases of HF hospitalization among DPP-4 inhibitors users (n = 160,803) and 6098 among GLP-1 agonists users (n = 160,803). The crude HF hospitalization incidence was 74 per 1000 person-years with DPP-4 inhibitors and 87 per 1000 person-years with GLP-1 agonists. In the proportional hazards model, use of DPP-4 inhibitors was associated with a 14% decreased risk of HF hospitalization compared to GLP-1 agonists use (HR, 0.86; 95% CI 0.83, 0.90), after adjusting for differences in baseline characteristics and disease risk factors (Fig. 1). Secondary analyses revealed similar results in patients without baseline HF (n = 158,543 for DPP-4 inhibitors; n = 158,543 for GLP-1 agonists; HR, 0.85; 95% CI 0.82, 0.89). However, among the comparatively small matched cohort of individuals with baseline HF (n = 1937 for DPP-4 inhibitors; n = 1937 for GLP-1 agonists), we observed no evidence of differential HF hospitalization risk (HR, 0.90; 95% CI 0.74, 1.07).

Results of the unadjusted and adjusted secondary analyses are summarized in Table 3 and Fig. 1. We observed a 18% reduction in the risk of hospitalization due to HF among patients with baseline CVD (n = 28,085 in each exposure group) comparing DPP-4 inhibitors to GLP-1 agonists users (HR, 0.82; 95% CI 0.77, 0.86). Similar results were observed in those without baseline CVD (n = 132,047 in each exposure group).

We identified a total of 1335 cases of HF hospitalization among users of saxagliptin (n = 49,214) compared to 2058 cases among matched GLP-1 agonists users (n = 49,214); and 6108 cases among sitagliptin users (n = 160,609) compared to 6239 among GLP-1 agonists matched users (n = 160,609). The crude incidences of HF hospitalization were 69 per 1000 person-years in saxagliptin users and 99 per 1000 person-years in matched GLP-1 agonists users. For sitagliptin users, the crude incidences were 75 per 1000 person-years compared to 89 per 1000 person-years in the matched GLP-1 agonists users. In the proportional hazards models, we observed differential within-class effect comparing individual DPP-4 inhibitors to GLP-1 agonists. Saxagliptin, compared with GLP-1 agonists, reduced HF hospitalization by 26% (HR, 0.74; 95% CI 0.69, 0.84), whereas sitagliptin reduced HF hospitalization by 8% (HR, 0.92; 95% CI 0.89, 0.95) compared to GLP-1 agonists. Study results remained consistent in the sensitivity analyses (Table 4).

The precise mechanism by which DPP-4 inhibitors may reduce HF exacerbation is not known. A recent single blinded randomized clinical trial found that DPP-4 inhibitors have no clinically meaningful effects on the B-type natriuretic peptide (BNP), a biomarker of HF. The study suggested that any changes in the level of HF-biomarkers over time is more likely to be due to the natural history of HF and not to be attributed to the effect of DPP-4 inhibitors [20]. However, there are several other possible mechanisms underlying the effect of DPP-4 inhibitors on HF. DPP-4 inhibitors are responsible for the degradation of several peptides that directly affect the heart and blood vessels [21]. One potential effect was observed in animal models where the continuous infusion of GIP led to the inhibition of the atherosclerotic lesion and the prevention of the penetration of macrophage in the aortic wall [21, 22]. The class also mediates control of the ratio of different receptor subtypes of neuropeptide Y (NPY) resulting in a regulation of blood pressure, blood flow, and inflammation [21]. Other studies suggest that DPP-4 inhibitors are associated with an improvement in left ventricular and endothelial functions leading to a delay in the development of HF, or worsening of HF, through its effect on blood pressure control, blood vessels, and cardiomyocytes [22–24].

Patients with type 2 diabetes are at higher risk of developing HF and the concomitant presence of both diseases can further complicate diabetes treatment. Thus, understanding the safety profile of diabetes medications—specifically the ones that were proven to be effective in reducing HbA1C level while maintaining body weight—is of particular interest to clinical practice. Our study found that despite documented safety warnings regarding the use of DPP-4 inhibitors in patients with HF, the class was not associated with an increased risk of HF compared with GLP-1 agonists. The findings complement the most recent findings from observational research and need to be considered within the context of other available alternative of diabetes therapy which have shown a lower risk of HF compared to DPP-4 inhibitors (i.e. SGLT2 inhibitors) [25]. Some may suggest that the observed protective effect of DPP-4 inhibitors on HF can be related to the higher risk of HF with the use of GLP-1 agonists. However, recent studies found no difference in the risk of HF when comparing GLP-1 agonists to placebo ruling out this alternative explanation [26–28].

The current study has several strengths. We used GLP-1 agonists as the comparator, and these agents represent reasonable second-line therapy for most patients according to existing guidelines. Secondly, we had a large sample size that allowed for the assessment of the heterogeneity of the treatment effect in selected subgroups of diabetes patients. Third, we performed several subgroup and sensitivity analyses providing confidence in the robustness of our primary findings. Nevertheless, the current study needs to be viewed in light of several limitations. First, the current study was conducted using observational data and, because of different route of administration (DPP-4 inhibitors are administered orally whereas GLP-1 agonists are injectable) as well as slightly different A1c lowering ability and adverse drug effects (e.g. weight loss) between DPP-4 inhibitors and GLP-1 agonists, the potential of selection bias may exist. However, we used a propensity score matching approach to minimize differences in baseline characteristics and disease risk factors between the two treatment groups. Second, deaths occurring outside of hospitals were not available in our data. Recent data suggest that GLP-1 agonists may have a mortality benefit compared with DPP-4 inhibitors, although the results of the analysis were limited by the small number of events (n = 20 in 7 trials) [13]. This, if true, could have led to survivor bias in our analysis, whereby GLP-1 agonists users may be less likely to die prior to a HF hospitalization, leading to more cases in this group. Nevertheless, we suspect any such bias to be negligible because the censoring rate due to end of enrollment (which include patients who died) was similar between the two groups (4% for DPP-4 inhibitors vs. 4% for GLP-1 agonists).

Third, the claims data used in this study were collected primarily for billing purposes and do not include clinical biomarkers, such as HbA1c. Thus, residual confounding resulting from missing data on important confounders such as demographics (e.g. race), or lab values (e.g. HbA1c) is still possible. Fourth, because outcome definition and the identification of confounders relies on ICD-9 codes only, there is the potential for bias resulting from errors in coding. However, the validity of ICD-9 codes used for the identification of the study outcome or the confounders has been assessed previously. Fifth, although informative censoring resulting from treatment discontinuation can impact the study results, our sensitivity analyses following the patients for a fixed period of time (45 days) provides some degree of confidence that this was not a major issue in our primary findings. Results from the current study are generalizable to patients with type 2 diabetes who are covered by commercial or Medicare supplemental insurance, limiting the generalizability to the broader Medicare population.

Not applicable.

The authors declare that they have no competing interests.

The data used for this study is not publicly available.

Each author has read and approved the manuscript and assumes responsibility of the content of the manuscript. The authors of the manuscript do not have any closely related papers or manuscripts that have been submitted or published elsewhere and declare that they do not have any competing interests.

The study was approved by the institutional review board of the University of Florida Health Science Centers.

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit-sector. Ghadeer Dawwas is a recipient of the American Association of Graduate Women (AAUW) fellowship for the 2017–2018 year.

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

The data used for this study is not publicly available.