Dipeptidyl peptidase-4 inhibitor cardiovascular safety in patients with type 2 diabetes, with cardiovascular and renal disease: a retrospective cohort study

Clinical trials investigating cardiovascular safety of dipeptidyl peptidase-IV inhibitors (DPP-4i) among patients with cardiovascular and renal disease rarely recruit patients with renal impairment, despite associations with increased risk for major adverse cardiovascular events (MACE). We investigated the risk of MACE associated with the use of DPP-4i among these high-risk patients. Using a new-user, retrospective, cohort design, we analyzed 2010–2015 IBM MarketScan Commercial Claims and Encounters for patients with diabetes, comorbid with cardiovascular disease and/or renal impairment. We compared time to first MACE for DPP-4i versus sulfonylurea and versus metformin. Of 113,296 individuals, 9146 (8.07%) were new DPP-4i users, 17,481 (15.43%) were new sulfonylurea users, and 88,596 (78.20%) were new metformin users. Exposure groups were not mutually exclusive. DPP-4i was associated with lower risk for MACE than sulfonylurea (aHR 0.84; 95% CI 0.74, 0.93) and similar risk for MACE to metformin (aHR 1.07; 95% CI [1.04, 1.16]). DPP-4i use was associated with lower risk for MACE compared to sulfonylureas and similar risk for MACE compared to metformin. This association was most evident in the first year of therapy, suggesting that DPP-4i is a safer choice than sulfonylurea for diabetes treatment initiation in high-risk patients.


Results
We identified 12,166,812 individuals with diabetes from 2010 to 2015. After applying our inclusion and exclusion criteria, there were a total of 113,296 individuals in our cohort (Fig. 1). Of these, 9146 (8.07%) were new users of DPP-4i, 17,481 (15.43%) initiated sulfonylureas, and 88,596 (78.20%) started metformin. Three percent of included individuals were new users of both DPP-4i and metformin, and 1524 (1.72%) were new users of both sulfonylureas and metformin. Total patients exposed to DPP-4i, sulfonylurea, and/or metformin after application of exclusion criteria (N=772,540) Total excluded for no cardiovascular disease or renal impairment at baseline (N=659,244) Total patients exposed to other OHA 1 (N=9,425,241) Total patients exposed to DPP-4i, Sulfonylurea, or Metformin with cardiovascular disease and/or renal impairment at baseline (N=113,296) -DPP-4i (n=9,146) -Sulfonylurea (n=17,481) -Metformin (n=88,596) www.nature.com/scientificreports/ Approximately half of individuals in each exposure group had a cumulative exposure to the treatment of interest less than or equal to 6 months (DPP-4i: 54.05%; sulfonylureas: 56.28%; metformin: 53.65%). There were also several differences in baseline comorbidities, and corresponding medication use, between exposure groups. For example, rates of cardiovascular disease were higher among users of metformin than their counterparts, while www.nature.com/scientificreports/ kidney disease was more prevalent among users of sulfonylureas (17.22%) and DPP-4i (15.00%) than metformin (7.07%). Differences between users of DPP-4i, sulfonylurea, and metformin were not statistically significant after application of propensity score weighting. The median follow-up for the primary outcome was 160 days (interquartile range (IQR) 92-296 days) and there was no statistically significant difference in time-to-event distributions across the therapeutic classes examined. After propensity score weighting and adjusting for baseline demographics, clinical characteristics, and concomitant medications, the incidence rate for the primary outcome was statistically significantly less (30.20 per 1000 person-years) among new users of DPP-4i compared to sulfonylureas ( Table 2). There was a non-statistically significant increase in the incidence rate for the primary outcome among new users of DPP-4i compared to sulfonylureas (91.25 per 1000 person-years vs. 79.46 per 1000 person-years). Among the secondary outcomes, the incidence rate for heart failure was statistically significantly less (4.07 per 1000 person-years vs. 7.56 per 1000 person-years) for new users of DPP-4i compared to sulfonylurea. The incidence rates for the secondary outcomes were not statistically significantly different between DPP-4i and metformin.
After propensity score weighting and adjustment for baseline demographics, clinical characteristics, and concomitant medications, there was a statistically significant association between new use of DPP-4i and the primary outcome of MACE compared to sulfonylurea (adjusted hazard ratio (aHR) 0.84, 95% confidence interval (CI) 0.74-0.93). However, there was no statistically significant difference in risk of MACE among users of DPP-4i as compared with metformin (aHR 1.07, 95% CI 0.98-1.16) ( Table 3). There were no statistically significant differences between sexes in the association between new use of DPP-4i and the primary outcome compared to sulfonylurea or metformin (Additional File 1, Supplementary Appendix 3).
When comparing the association between DPP-4i and the primary outcome by cumulative exposure strata, there were differences from the overall effect. In the comparison between new users of DPP-4i and sulfonylurea, the adjusted hazard ratio for the primary outcome was statistically significant for a cumulative exposure of < 6 months (aHR 0.85, 95% CI 0.71-0.93) and 6-12 months (aHR 0.76, 95% CI 0.70-0.93). However, statistical significance was not evident for cumulative exposure 12-18 months and > 18 months. This suggests that the overall aHR might be driven by differences in MACE risk seen in the first year after initial exposure. There were no differences between the overall and cumulative exposure stratum specific adjusted hazard ratios for MACE between new users of DPP-4i and metformin (Table 4). Table 2. Incidence rates of primary composite outcome, acute myocardial infraction, stroke, and heart failure among new users of DPP-4 inhibitors, sulfonylureas, and metformin. IQR interquartile range. 1 Primary composite outcome includes myocardial infarction, cardiac arrest, coronary artery bypass, coronary angioplasty, heart failure, stroke, inpatient death. 2

Discussion
In this longitudinal study of commercial claims data for individuals with diabetes, having also cardiovascular disease and renal impairment, new use of DPP-4i was associated with a lower risk for MACE compared to sulfonylureas, and a comparable risk compared to metformin. This difference was most evident in the first year of use. New use of DPP-4i was not shown to be associated with the following individual components of MACE: heart failure, stroke, or acute myocardial infarction. These results contribute to the body of evidence examining the association of DPP-4i and MACE in individuals at higher risk for cardiovascular events.
Overall, our results corroborated those of previous clinical trials and observational studies of cardiovascular safety of DPP-4i. Our results showed no increased risk for MACE with the use of DPP-4i similar to the conclusions in the three completed clinical trials comparing DPP-4i and placebo 12,13,18 . However, unlike these trials our study population consisted of a high-risk group of individuals with established cardiovascular disease and renal impairment in a real-world setting with multiple comorbidities and concomitant medications. Additionally, our study allowed us to compare DPP-4i to other available therapies, namely sulfonylureas and metformin, showing a decreased risk for MACE when compared to sulfonylureas. Table 3. Hazard ratios for the association between DPP-4 inhibitor use and primary composite outcome, acute myocardial infraction, stroke, and heart failure compared to sulfonylureas and metformin. a Propensity score weighting only. b Propensity score weighting and demographics, comorbidities, and concomitant medications as regressors and stratifiers. c Primary composite outcome includes myocardial infarction, cardiac arrest, coronary artery bypass, coronary angioplasty, heart failure, stroke, inpatient death. Bold values indicate statistically significant results. www.nature.com/scientificreports/ Three notable observational studies provide additional context to our results. First, a 2015 administrative claims study using the Taiwan National Health Insurance Research Database of individuals with diabetes and chronic kidney disease admitted for acute myocardial infarction also showed no increased risk for MACE when comparing individuals on sitagliptin to those not on sitagliptin 19 . Expanding on this approach, our study design included individuals with a multitude of cardiovascular conditions at baseline, making our results more generalizable to high-risk individuals with diabetes. The second study of 127,555 individuals which used the Italian Nationwide OsMed Health-DB database showed decreased risk in hospitalization for heart failure risk associated with DPP-4i compared to sulfonylurea (HR 0.78; 95% CI 0.62-0.97) 20 . This suggests that the increased risk for MACE in our comparison might be driven by differences in risk for heart failure. This is partially evident in the statistically significant unadjusted hazard ratio for heart failure as a secondary outcome (Table 3). Finally, a study of Medicare patients with diabetes compared cardiovascular risk of DPP-4i to sulfonylurea and thiazolidinediones and found no increased risk in individuals over age 65 21 . While our study was limited to a patient population under age 65, this study shows that our results are similar to those found in a study of older patients.
A major strength of our study was the identification of a high-risk cohort through the use of a large administrative database. The results of the SAVOR TIMI-53 trial showed that renal impairment was independently associated with adverse cardiovascular outcomes, even after adjusting for cardiovascular risk factors 17 . As such, our decision to restrict the study cohort to individuals with diabetes, comorbid with cardiovascular disease and renal impairment allowed us to hone in on a patient population identified by FDA as more appropriate for the study of this drug-event association. In a 2008 Guidance for Industry 9 , regulators noted that such patients are often excluded from pre-approval clinical trials, and recommended studies of the association between oral antihyperglycemic agents and MACE should include high-risk patients.
Our study also had several limitations. First, we included individuals using multiple therapies of interest (e.g., both DPP-4i and sulfonylureas) at baseline. While this may have led to potential misclassification, their exclusion would have resulted in a much smaller and less generalizable sample 22 . Second, our new-user design did not account for time-varying hazards for MACE, although the cumulative exposure stratified analysis suggested that the differences in risk of MACE between DPP-4i and sulfonylureas were most evident during the first year of exposure. Third, our analyses are subject to potential informative censoring due to a change in exposure status. However, our results were insensitive to the lagged latency period after the last day of exposure, and since most events occurred within the first year of exposure, extending the latency period past 30-days would have been unlikely to have changed our results and could have potentially led to misclassification. Finally, due to limitations in the administrative claims data, we were unable to assess mortality outside of the inpatient setting as a component of the primary composite outcome or as a competing risk. However, it is unlikely that death outside of the inpatient setting would occur more frequently in one exposure group than another in this cohort, potentially leading to biased results.
Our results provide evidence of decreased risk for MACE when comparing DPP-4i versus sulfonylurea in a commercially insured patient population with diabetes, comorbid with cardiovascular disease and renal impairment in the United States. Additionally, we found that there was no difference in risk of MACE for these individuals when comparing DPP-4i and metformin. Further studies are needed to determine differences in risk for individual components of the composite outcome, particularly heart failure. Finally, the decreased risk seen with DPP-4i use compared to sulfonylurea is more likely due to cardiovascular risk associated with sulfonylurea rather than protective effects of DPP-4i, as there was no difference in effect between DPP-4i and metformin, which carries little cardiovascular risk.

Methods
Study design and data source. We conducted a population-based, retrospective cohort study with a new-user design using data from IBM MarketScan Commercial Claims and Encounters from January 2010 through December 2015. MarketScan houses linked paid claims and encounter data from approximately 350 payers, covering more than 25 million individuals annually. The data consists of de-identified individual-level healthcare utilization data including demographic characteristics and information on inpatient and outpatient medical services and pharmacy claims issued.
Cohort derivation. We identified individuals with type 2 diabetes as those with at least one prescription for an oral antihyperglycemic agent and either hemoglobin A1c greater than 6.5% twice, fasting glucose greater than 126 mg/dL twice on different days, random glucose > 200 mg/dL twice on different days, one inpatient diagnosis of diabetes (ICD-9(10) codes: 250x (E11.9), 357.2 (E11.42), 366.41 (E11.36), 362.01-362.07 (E11.3*)) or outpatient diagnosis for diabetes (ICD-9(10) codes: 250x (E11.9), 366.41 (E11.36), 362.01-362.07 (E11.3*)) twice on different days. We included individuals if they received at least one prescription for an FDA-approved DPP-4i, sulfonylurea, or metformin (Additional File 1, Supplementary Appendix 1). We assigned an index date as the date of first filled prescription for one of these products. The baseline period was defined as the six-month period preceding this. We used International Classification of Diseases, 9th and 10th Revisions (ICD-9/10) codes from inpatient and outpatient records to identify those with a history of cardiovascular disease and/or renal impairment (chronic kidney disease or acute renal failure) for inclusion into the study.
We excluded individuals based on: (1) no continuous medical or pharmacy enrollment in the six-month baseline period; (2) below the age of 35; (3) insulin use at baseline; (4) end-stage renal disease in the baseline period; (5) less than 12-weeks of exposure to index treatment; or (6) treatment with oral antihyperglycemic agents in the 6-month baseline period. We followed individuals until the first of the following occurrences: (1) 14-days after the last date of exposure; (2) switch to or addition of anti-hyperglycemic treatment that was not Definition of exposure. We assigned individuals to an exposure group based on their first prescription of DPP-4i, sulfonylurea, or metformin. In the event of multiple drug class prescriptions on the index date, individuals were counted in all relevant exposure groups. Due to the large number of individuals on multiple antihyperglycemic agents, particularly metformin in combination with other drug classes, dropping these individuals would have resulted in a considerably smaller sample size, reducing statistical power as well as the generalizability of our results.
Definition of outcome. We defined our primary composite outcome as the first of any of the following events: myocardial infarction, cardiac arrest, coronary artery bypass graft, coronary angioplasty, heart failure, and stroke. We did not examine all-cause mortality outside of the inpatient setting as an outcome, because information on these deaths were not available for our dataset. Our secondary outcomes were acute myocardial infarction, stroke, and heart failure. We used validated algorithms to identify myocardial infarction 23 , cardiac arrest 24 , coronary artery bypass graft 23 , coronary angioplasty 23 , heart failure 25 , and stroke 26 .

Definition of covariates.
We used the peer-reviewed literature 19,[27][28][29] , clinical guidelines 30-34 and expert opinion 9,35 in order to identify key covariates of interest. We assessed possible confounding due to patient age and sex; cardiovascular risk factors including hypertension and hypercholesterolemia, diabetic complications, diabetic severity at baseline as measured by the Adjusted Diabetes Comorbidities Severity Index (aDCSI) [36][37][38] , and other common comorbidities such as cerebrovascular disease, ischemic heart disease, asthma, and rheumatoid arthritis. Comorbidities coded under ICD-9 were identified using the Clinical Classification Software (CCS) developed by the Agency for Healthcare Research and Quality 39 . Additionally, we used National Drug Codes to assess for possible confounding due to concomitant medications including statins, angiotensin converting enzymes (ACE) inhibitors, platelet aggregation inhibitor, calcium channel blockers, angiotensin II receptor blockers, beta-blockers, diuretics, nitrates, and inhaled corticosteroids.
Propensity score. We used the Toolkit for Weighting and Analysis of Nonequivalent Groups (twang) package developed by the RAND Corporation 40 to compute the propensity scores and associated weights used in the analysis to balance the covariates between exposure groups. To do so, we identified all available covariates without an association to the exposure that were associated to the primary outcome in order to increase precision 41 (Additional File 1, Supplementary Appendix 2). Next, we used generalized boosted regression models to optimize the selection of covariates for the propensity score calculation, allowing for propensity scores estimation in the presence of multiple exposure groups. We used the standardized mean difference (SMD) to measure balance of covariates before and after weighting. Propensity score weighting reduced the SMD from a maximum of 0.23 to less than 0.01. We used the average treatment effect on the treated (ATT) propensity score weights to estimate the treatment effect of DPP-4i.

Statistical analysis.
We identified a baseline period of 6 months prior to the first filled prescription of the exposure group. Using chi-square statistics for categorical covariates and the Kruskal-Wallis test for continuous covariates, we compared differences at baseline between new users of DPP-4i, sulfonylurea, and metformin. We used exact Poisson tests to compare absolute differences in incidence rates for the primary and secondary outcomes for new users of DPP-4i compared to those of sulfonylurea and metformin. Additionally, we checked for differences in duration of exposure to treatment distributions using a Kolmogorov-Smirnov test. Using the propensity score weights described above, we calculated weighted crude and adjusted Cox proportional hazards for the association between new use of DPP-4i and the primary and secondary outcomes compared to new use of sulfonylureas and metformin. For the adjusted hazard ratios, we included indicators for age, sex, baseline comorbidities, and concomitant medications. Additionally, we checked for covariates that violated the proportional hazards assumption by plotting the scaled Schoenfeld residuals. These covariates were added to the Cox proportional hazards model as stratifying variables. We also used spline intervals every 6 months to address potential violations of the proportional hazard assumption. Finally, we included an indicator variable for individuals included in multiple exposure groups and an indicator variable for cumulative exposure, which was defined as the number of days exposed to the exposure group drug. We stratified the analysis by the cumulative exposure variable. All analyses were conducted in R, version 3.3.3.

Sensitivity analyses.
To check the robustness of our results, we first assessed possible sensitivity of our results to the latency of the period after drug discontinuation. We followed individuals for 14-, 7-, and 30-days after the last day of exposure to drug. Next, we recalculated the primary analysis without individuals with acute renal failure to determine whether our results were sensitive to the inclusion of those individuals. Finally, we compared the association of DPP-4i and the primary outcome between cumulative exposure strata and sex strata to assess possible effect modification.
The study was exempted from review by a Johns Hopkins Institutional Review Board, and all research was performed in accordance with relevant guidelines/regulations. Our study utilized commercial claims data through IBM Marketscan Research Databases. As such, we did not obtain informed consent. This was in accordance with local laws and regulations, and aligned with the terms of use from IBM Marketscan Research Databases.

Data availability
The data that support the findings of this study are available from IBM MarketScan Research Databases but restrictions apply to the availability of these data, which were used under license for the current study, and so are not publicly available. Data are however available from the authors upon reasonable request and with permission of IBM MarketScan Research Databases.