Glucagon‐like‐peptide‐1 receptor agonists versus dipeptidyl peptidase‐4 inhibitors and cardiovascular outcomes in diabetes in relation to achieved glycemic control. A Danish nationwide study

In Denmark, an identification number given at birth or immigration allows for cross‐linkage of health and administrative databases at the individual level and enables complete follow‐up.13 Every citizen, including primary and hospital care, is granted equal access to the health care system. Data for this study was cross‐linked using seven nationwide registries: (a) The Danish National Patient Registry contains information on all hospital admissions and outpatient contacts from 1977 onwards. According to the International Classification of Diseases, Ninth and Tenth Revision, each contact is coded with a primary diagnosis and one or more secondary diagnoses. Surgical procedures are coded using the Nordic Medico‐Statistical Committee Classification of Surgical Procedures.14, 15 (b) The Danish National Prescription Registry holds information (dosage, dates, and anatomical therapeutic chemical codes) on all prescriptions dispensed from a pharmacy since 1995.16 (c) The Danish Cause of Death Registry entails information on the date, cause, and place of death from 1970 onwards.17 (d) The Danish Population Registry contains information on the date of birth and sex.18 (e) Information on the living situation and educational level was extracted from the Civil Registration System and the Educational register.19 (f) An extensive collection of laboratory databases covers most blood tests from hospitals and general practitioners in four of five regions in Denmark.13 (g) The Danish Adult Diabetes Registry contains information on body mass index, smoking status, and HbA1c measurements on T2D patients managed by general practitioners or the hospitals.20

We included all incident GLP‐1 RA and DPP‐4i users between 2007 and 2021 >18 years who initiated GLP‐1 RA or DPP‐4i as a second‐line T2D treatment following metformin. Treatment with DPP‐4i was selected as an active comparator because it is a frequently used second‐line therapy for T2D, exerts its effect on the blood sugar through GLP‐1, but has a neutral effect on cardiovascular outcomes.

Patients were included 6 months after treatment initiation. We excluded patients who started other glucose‐lowering drugs (including insulin), who were not in metformin treatment at the time of inclusion, who emigrated during the study period, who immigrated to Denmark in the 10 years before inclusion, who were diagnosed with end‐stage renal disease or were affected with a condition that would make HbA1c invalid as a monitoring measure (see Tablesandin supplementary material). S1 S2

Available HbA1c measurements within 6 months after treatment initiation were included. We used the latest measures for each patient if several measurements were available. The primary exposure of interest was the achieved HbA1c level, categorized according to two groups: ≤53 mmol/mol (≤7.0%) or >53 mmol/mol (>7.0%). Patients with a missing post‐treatment HbA1c were categorized into one group to investigate the association among those patients with a missing HbA1c value.

The primary outcome was a composite outcome ofMACE: non‐fatal myocardial infarction, nonfatal stroke, and all‐cause death.

Patients were included 6 months after treatment initiation and followed until the outcome, emigration, end of study (31 December 2021), or 5 years in total, whichever came first.

Comorbidities were identified from before in‐ and out‐hospital contacts up to 10 years before the date of inclusion. Information on concomitant medical therapy was obtained from dispensed prescriptions listed in the Danish National Prescription Registry and defined by at least one redeemed prescription 6 months before the inclusion date. Hypertension was defined as treatment with at least two classes of antihypertensive drugs 6 months before inclusion, as done previously.21 Educational level was defined as the highest level of completed education and classified according to the International Standard Classification of Education. Missing information on the educational level was classified as basic level education.

Baseline characteristics were described using proportions for categorical variables and means and SDs or medians and interquartile ranges (IQR) for continuous variables. A reverse Kaplan–Meier estimator was used to estimate the median follow‐up time.22 We used the Kaplan–Meier estimator to assess the 5‐year risk of the primary outcome of MACE and the Aalen–Johansen estimator to calculate the risk of individual components of the outcome when death was present as a competing risk (ie, nonfatal myocardial infarction and nonfatal stroke). To investigate how different levels of achieved HbA1c affected the rate of the outcome among GLP1‐RA users compared to DPP‐4i users, a multivariate Cox‐proportional model was estimated with an interaction term between treatment and achieved HbA1c level. The model included the following covariates: age, gender, living alone (yes/no), T2D duration, pretreatment HbA1c category (≤53 mmol/mol [≤7.0%], >53 mmol/mol [>7.0%], or missing HbA1c), highest achieved educational level, body mass index category (underweight/normal weight, overweight, missing), smoking status (never smoker, ex‐smoker, occasional, daily smoker, missing), known comorbidities (atrial fibrillation, cancer, chronic obstructive pulmonary disease, hypertension, chronic kidney disease, cardiovascular disease, heart failure), and inclusion year. To investigate potential differences across subgroups, we reproduced these analyses in subgroups of specific patient characteristics (gender, age category, presence of cardiovascular disease at baseline (including heart failure), and inclusion year) to assess whether the interaction of achieved HbA1c category and treatment on the outcome persisted. A cause‐specific Cox regression model was used to investigate the individual components of the composite outcome where death was a competing risk. To estimate how different levels of achieved HbA1c were associated with the rate of the outcome, an interaction term between treatment and achieved HbA1c level was included. The model was adjusted for the same covariates mentioned. All estimates are reported with 95% confidence intervals (CIs).

To assess the robustness of the primary analyses, we performed the following analyses: The primary analysis was repeated with a finer categorization of HbA1c levels according to four groups: <48 mmol/mol (<6.5%), 48–53 mmol/mol (6.5%–6.9%), 54–58 mmol/mol (7.0%–7.5%), and <58 mmol/mol (>7.5%). Important to note that 47 and 48 mmol/mol are converted to 6.5% according to the National Glycohemoglobin Standardization Program HbA1c. To ensure the generalizability of our results, we repeated the primary analyses, including cardiovascular death instead of all‐cause death in the composite outcome of MACE. In the Cause of Death registry, registration of cardiovascular death is available only up to 31 December 2018; therefore, we repeated the primary analysis, including incident GLP‐1 RA or DPP‐4i users between 2007 and 2018. Patients were included 6 months following treatment initiation and followed until the outcome, emigration, end of study (31 December 2018), or 5 years in total, whichever came first. To ensure that the achieved HbA1c level was a consequence of the treatment, we restricted the study population to those with high adherence to GLP1‐RA and DPP‐4i before inclusion. This was done by calculating the treatment duration for each drug based on the available prescriptions. A patient was considered to be under treatment from the first prescription until the date of the subsequent prescription. If no additional prescription was obtained, a treatment duration of 3 months was assumed. Only persons with a proportion of days covered above 80% within the 6 months before inclusion were included. To investigate how the cumulative exposure of increased blood sugar levels could affect the risk of MACE among GLP‐1 RA users compared to DPP‐4i users, we used the mean HbA1c value to represent the cumulative exposure. Available HbA1c measurements were collected within 6 months before and after treatment initiation with GLP‐1 RA or DPP‐4i. If several measurements were present, the earliest value before treatment (but within the 6 months) and the latest value following treatment (within the 6 months) were collected. Each mean HbA1c value was categorized according to the same categories mentioned earlier. Patients with a missing pretreatment and/or post‐treatment HbA1c compromised their categories. GLP‐1 RA was introduced in Denmark in 2007–2012 and this could affect which patients would be treated with DPP‐4i, sulfonylurea, and thiazolidinedione following 2012. To investigate if our association was confounded by this, we created a cohort including only patients in treatment with DPP‐4i, sulfonylurea, and thiazolidinedione prior to 2012 and patients in treatment with GLP‐1 RA included after 2013. The primary analyses were repeated using DPP‐4i or sulfonylurea and thiazolidinedione as the reference group.

All analyses were performed in R version 4.2.1.23

By law, registered‐based research has been granted permission to be performed without ethical approval or informed consent. Therefore, the project has been approved by the Knowledge Centre on Data Protection Compliance‐The Capitol Region of Denmark (approval number: P‐2010‐191).

The final study population consisted of 53 473 T2D patients, of whom the following HbA1c distribution among GLP‐1 RA and DPP‐4i users was achieved: HbA1c ≤ 53 mmol/mol (≤7.0%): 3026 (22%) versus 4824 (12%); HbA1c > 53 mmol/mol (>7.0%): 6577 (48%) versus 17 508 (44%); missing HbA1c: 4031 (30%) versus 17 507 (44%), as shown in Figure 1. The frequency distribution of all HbA1c values according to treatment is shown in Figure S1 in the supplementary material. Patients in treatment were GLP‐1 RA were more likely to be treated with liraglutide (50.9%) or liraglutide (46.5%) than another type of GLP‐1 RAs (2.6%). Characteristics of the study population according to achieved HbA1c level are shown in Table 1. Patients treated with GLP‐1 RA were more likely to be female, younger, not living alone, better educated, overweight, and active/ex‐smokers, to have shorter diabetes duration and better kidney function compared to DPP‐4i users across all levels of achieved HbA1c category. Second, they were less likely to be diagnosed with comorbidities and receive concomitant medication than DPP‐4i users. No differences in the percentage of baseline cardiovascular disease were observed between GLP‐1 RA and DPP‐4i users within each HbA1c category.

Within a median follow‐up time of 5 years (IQR:2.6–5.0), the following number of events was observed in each achieved HbA1c category among GLP‐1 RA and DPP‐4i users: ≤53 mmol/mol (≤7.0%): 101 (3.3%) versus 796 (16.5%), >53 mmol/mol (>7.0%): 372 (5.7%) versus 2417 (13.8%), missing HbA1c: 481 (11.9%) versus 3880 (22.2%). The 5‐year risk (95% CI) of the composite outcome in GLP1‐RA and DPP4i users according to HbA1c categories was as follows: HbA1c ≤ 53 mmol/mol: 10.3% (8.2–12.3) versus 24.3% (22.7–25.8); HbA1c > 53 mmol/mol: 16.0% (14.3–17.6) versus 21.1% (20.3 –21.9); missing HbA1c: 17.1% (15.7 –18.5) versus 25.6% (24.9 –26.3), as shown in Figure 2. The hazard rate ratio from the multivariable‐adjusted analyses of the composite outcome among GLP‐1 RA users compared to DPP‐4i users across the achieved HbA1c categories was HbA1c ≤ 53 mmol/mol (≤7.0%): 0.65 (0.52–0.80), HbA1c > 53 mmol/mol (>7.0%): 0.92 (0.83–1.03), and missing HbA1c: 0.92 (0.84–1.02) (p value for interaction <.001) (Figure 3). The results for the individual components of the composite outcome are shown in Figure S2 in the supplementary material. The outcome of all‐cause death primarily drove the composite outcome. When investigating the association among subgroups, the results remained persistent as in the primary analyses but did not achieve statistical significance for all subgroups (Table S3 in supplementary material).

When using finer categories of achieved HbA1c levels, we achieved similar results as in the primary analysis, albeit the number of events in each HbA1c category were lower. Tablesandin the supplementary material depict the study population's characteristics and analysis results. S4 S5 Using a major adverse cardiovascular outcome that included cardiovascular death instead of all‐cause death did not change the results. Similar 5‐year risk and hazard ratios across different levels of achieved HbA1c categories were observed, as shown in (Tablein supplementary material). S6 Among patients with a high adherence to treatment (proportion of days covered >80%) in the 6 months before inclusion, the risk of MACE among patients in treatment with GLP‐1 RA compared to DPP‐4i yielded similar results as in our primary analysis. Lower HbA1c categories were associated with the lowest 5‐year risk and hazard rate of the event (see Tablein the supplementary material). S7 The cumulative HbA1c exposure yielded similar results to the primary analyses (Tablein the supplementary material). S8 Restricting the cohort to DPP‐4i users included prior to the introduction of GLP‐1 RA in Denmark (2012) yielded similar results to the primary analyses (Tableand Figurein the supplementary material). Using sulfonylurea or thiazolidinedione as the active comparator reference group showed similar association as in the primary analyses (Tableand Figure). S9 S3 S10 S4