Dipeptidyl peptidase-4 inhibitor compared with sulfonylurea in combination with metformin: cardiovascular and renal outcomes in a propensity-matched cohort study

We employed a retrospective matched cohort design using the Korean National Health Insurance Service-Health Screening Cohort (NHIS-HEALS), which included 514,866 individuals aged between 40 and 79 years from the Republic of Korea. This represents 10% of a random selection within all of the health screening participants in the index year 2002 or 2003 and followed up through 2013. The NHIS requires all insured employees and self-employed persons more than 40 years as well as their dependents to participate in a general health screen every 2 years in order to improve the health status of Koreans through the prevention and early detection of disease. This database contains longitudinal information including the subjects’ demographics as well as medical and pharmaceutical records including disease code records according to the International Classification of Disease, Tenth Revision (ICD-10), medical procedures, hospitalization, information of prescribed drugs, and death records. The detailed cohort protocol has been described previously [21]. From the original database, we selected patients with type 2 diabetes (ICD-10 codes E11–14), who had received at least one oral glucose-lowering agent from December 1, 2008 (the date DPP4i was first released in Korea) to September 30, 2013 (the date when the results of SAVOR-TIMI 53 were released). Metformin, SU, thiazolidinedione (TZD), and DPP4i were included in this study as oral glucose-lowering agents. Among the included patients, we identified all of those who had been prescribed DPP4i or SU in combination with metformin. We chose SU as a comparator drug because SUs were one of the most frequently used second-line oral hypoglycemic agents added on metformin all over the world, especially under the Korean insurance. Furthermore, only patients who were initially treated with the study drugs (DPP4i and SU) were included in the study, and any patients who had received these agents alone or in other combinations before the index date were excluded. We also excluded patients who had previously used insulin formulations. Patients who died in the first month in the index year were also excluded. The index year was defined as the date when DPP4i or SU in combination with metformin were first prescribed.

The study subjects were divided into the following two groups: DPP4i group (DPP4i plus metformin) and SU group (SU plus metformin). The disposition of patients for the study is shown in Additional file 2: Figure S1. Two separate cohorts were created based on their underlying histories of cardiocerebrovascular disease (CVD) and HF, respectively. In the first cohort (CVD cohort), baseline CVD was defined by any former diagnosis of ischemic heart disease (IHD) (ICD codes I20–25) and cerebrovascular disease (ICD codes I60–64), or HF (ICD codes I50, I42–43). On the basis of the presence or absence of a recorded history of CVD, two subgroups were classified, matched and analyzed. In the second cohort (HF cohort), baseline HF was also defined as stated above. The analysis was performed in the same way as that in the first cohort. In the analysis for the renal outcomes, we selected only patients with available laboratory data for creatinine levels at baseline and had no history of end-stage of renal disease (ESRD) before the index year.

This study was approved by the institutional review board of the Korea University Anam Hospital.

The cardiocerebrovascular outcomes of interest were incident IHD (ICD-10 codes I20-I25 plus a procedure of coronary artery angiography), ischemic stroke (IS) (ICD-10 codes I63–66 with an examination of brain imaging study), hospitalization for HF (HHF; ICD-10 codes I50, I42 and I43), and death from CVD (ICD codes I00-I99). Composite CVD events included any of the components of CVD events. HHF event was defined as a primary or secondary diagnosis with ICD-10 codes mentioned above.

The end-stage renal outcomes were defined as having any of the following: diagnosis of ESRD (ICD-10 codes N18.0 and N18.5), hospital visits involving renal dialysis (Z49.1 and Z49.2), kidney transplantation status (Z94.0), procedures for hemodialysis or peritoneal dialysis (O7020, O7030-7034, O7071, and O7072), or surgical procedures for kidney transplantation (R3280).

Each patient was followed up from the index date up to the earliest occurrence of any study outcomes, discontinuation of pre-specified regimens (the study drug was stopped, or the alternative study drug was added), death, or the end of the study period (September 30, 2013).

Confounder variables measured for this study included the patient age at index date, gender, duration of diabetes, fasting blood sugar (FBS), body mass index (BMI), systolic blood pressure (SBP), and previous prescribed use of TZD. Prescribed drugs including antihypertensive medication, statins, antiplatelet agents, and anticoagulation agents were regarded as those used more than 30 days before the index year. Due to the fact that adjustment of too many similar variables can cause conflict, we chose medications for hypertension and dyslipidemia instead of ICD codes. Diabetes duration (in years) was assessed from the date of the first medical treatment for a diagnosis of diabetes until the date of an event. The patients’ histories of smoking, alcohol consumption and physical activity were also adjusted. For statistical analyses, these data were classified further into three groups as follows: smoking (never, former, or current smokers), alcohol consumption (none, twice per week, or ≥ three times per week), and physical activity (none, ≤ twice per week, or ≥ three times per week).

Continuous variables were presented as mean ± standard deviation (SD), and the categorical variables were described as numbers with percentages. Cumulative incidence rates and 95% confidence intervals (CI) for the study outcome were calculated using the Kaplan–Meier method. The propensity score matching (PSM) and inverse probability of treatment weighting (IPTW) were used to reduce the effects of confounders between the DPP4i and SU groups. The propensity score defining each individual’s probability of receiving the DPP4i plus metformin treatment was developed by a multiple logistic regression model that included all of the variables in both cohorts mentioned in Additional file 1: Tables S1 and S2 in both cohorts: index year, age, sex, duration of diabetes, fasting blood sugar, ever-prescription for TZD, BMI, SBP, prescribed drugs, and histories of smoking, alcohol consumption, and physical activity. For the PSM, we performed a 2:1 matching (two cases per one control patient within strata based on the presence or absence of baseline CVD or HF. Baseline characteristics between the two groups were compared using generalized mixed models with appropriate link functions in each matched cohort. A stratified Cox’s proportional hazards regression analysis for matched data was performed to evaluate the relative hazard of events in the DPP4i group compared to that in the SU group after adjusting covariates used in the multiple logistic regressing model. We also used IPTW, which is a powerful tool for observational data [22]. The weights based on each individual’s propensity score were calculated by the inverse of the score in the DPP4i group and the inverse of 1 minus the score in the SU group [23]. We then estimated the hazard ratio (HR) and 95% CI from the weighted Cox’s proportional hazards regression analysis using IPTW. A P value < 0.05 was considered statistically significant. All statistical analyses were performed using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA).

The cohort included a total of 23,635 patients; 16,803 patients were treated with a DPP4i plus metformin, and 6832 were treated with a SU plus metformin (Additional file 2: Figure S1). The mean age of study subjects was 62 years, and 61% were women. 4.2% had been treated with TZD before the index date. The frequencies of statins and anti-thrombotics prescribed were 49% and 41%, respectively. Additional file 1: Table S1 describes the baseline characteristics of the DPP4i group (n =9368) and SU group (n =4684) according to the baseline CVD (1st cohort), which were well balanced after PSM. An additional table showing the baseline characteristics of the 2nd cohort according to the baseline HF also demonstrated well-matched profiles between the groups (Additional file 1: Table S2).

During a median follow-up of 19.6 months (interquartile range 7.2–36.4), 762 composite CVD events and 17 cases of ESRD occurred in the 1st cohort. In the 2nd cohort, there were 201 HHF events and 28 cases of ESRD during a median follow-up of 19.3 months (interquartile range 7.1–36.4).

The composite and individual CVD events were analyzed by a multiple logistic regression model (Table 1). Because the number of patients followed after 3 years were largely reduced due to changes of initial treatment regimens, we analyzed the risks separately for the 3rd and 5th years.

In the 1st cohort, DPP4i plus metformin therapy was not associated with an increased risk of composite CVD events (3rd year: HR, 1.02; 95% CI 0.88–1.19; 5th year: HR, 1.04; 95% CI 0.90–1.20) compared to SU plus metformin. The HRs of IHD, IS, and CVD deaths in the DPP4i group were 0.99 (95% CI 0.80–1.24), 0.89 (95% CI 0.68–1.17) and 0.68 (95% CI 0.41–1.14), respectively, in the 3rd year; and 1.00 (95% CI 0.81–1.23), 0.95 (95% CI 0.74–1.23) and 0.74 (95% CI 0.46–1.18), respectively, in the 5th year. However, the risk of HHF was significantly higher in the DPP4i group (3rd year: HR, 1.47; 95% CI 1.07–2.04; 5th year: HR, 1.34; 95% CI 1.01–1.81) than in the SU group (Table 1, Fig. 1). In the subgroup with no baseline CVD, the increased risk of HHF remained significant in the DPP4i group in the 3rd year (HR, 3.32; 95% CI 1.28–3.32). However, there were no significant differences in HHF between the groups in the subgroup with baseline CVD. From the 2nd cohort, DPP4i-related risk of HHF was also significantly higher in the 3rd year (HR, 1.39; 95% CI 1.02–1.90), but not in the 5th year (HR, 1.26; 95% CI 0.95–1.67) (Table 2; Additional file 2: Figure S2A). Further subgroup analyses for the risk of HHF was performed according to baseline CKD (Additional file 1: Table S3), history of TZD use (Additional file 1: Table S4 and Table S5) and stratified by individual DPP4 inhibitors (Additional file 1: Table S6), which resulted in similar findings of DPP4i-associated increased risk of HHF. However, those results are limited by small numbers of HHF events in each subgroups.

In the model of IPTW (Additional file 1: Table S7), the HRs of HHF were 1.59 (95% CI 1.16–2.17) in the 3rd year and 1.36 (95% CI 1.00–1.87) in the 5th year. These trends were maintained in subgroups both with and without baseline CVD.

For renal outcomes, we included patients with data on creatinine and without a previous diagnosis of ESRD. The median values of creatinine were 1.05 in DPP-4i group, and 1.01 in SU group (p = 0.01). The risk of ESRD was not higher in the DPP4i group than in the SU group. In the 1st cohort, the HRs were 1.23 (95% CI 0.42–3.62) in the 3rd year, and 1.02 (95% CI 0.40–2.63) in the 5th year (Table 1, Fig. 2). The results in the 2nd cohort were similar to those of the 1st cohort (3rd year: HR, 1.55; 95% CI 0.65–3.71; 5th year: HR, 1.10; 95% CI 0.54–2.28) (Table 2; Additional file 2: Figure S2B).