All‐cause mortality and cardiovascular outcomes with sodium‐glucose Co‐transporter 2 inhibitors, glucagon‐like peptide‐1 receptor agonists and with combination therapy in people with type 2 diabetes

To assess the relationship of sodium‐glucose cotransporter‐2 inhibitors (SGLT2i), glucagon‐like peptide‐1 receptor analogues (GLP‐1RA) and their combination (SGLT2i + GLP‐1RA) with 5‐year risk of all‐cause mortality, hospitalization and cardiovascular/macrovascular disease in people with type 2 diabetes.

reduction in 5-year all-cause mortality when compared directly against either monotherapy.
K E Y W O R D S cardiovascular disease, chronic kidney disease, glucagon-like peptide-1 receptor agonists, hospitalization, mortality, sodium-glucose cotransporter 2 inhibitors, type 2 diabetes 1 | INTRODUCTION Cardiovascular disease is the most common cause of death in people living with type 2 diabetes mellitus. 1 In 2017, over 1 million deaths were directly attributed to type 2 diabetes, with cardiovascular death responsible for half of these. 2,3 In the same year it was estimated that 462 million live with type 2 diabetes, with the prevalence expected to rise significantly from 6059 to 7079 cases per 100 000 by 2030. 2 Historically, clinicians had a limited ability to reduce the risk of cardiovascular disease in people with type 2 diabetes through the modification of cardiovascular risk factors (dysglycaemia, hyperlipidaemia, blood pressure and smoking) and promotion of a healthy lifestyle. Metformin has been considered cardioprotective, primarily based on data from UKPDS. 4 In a recent meta-analysis in patients with type 2 diabetes with coronary artery disease, metformin reduced cardiovascular and all-cause mortality, and cardiovascular events. 5 However, this effect has not been true for all hypoglycaemic therapies as the thiazolidinedione, rosiglitazone, was withdrawn in the late 2000s because of excess cardiovascular risk. 6 Because of growing concerns relating to cardiovascular safety of novel glucose-lowering therapies, regulatory authorities stipulated that future therapies must show cardiovascular safety. Consequently, we saw the emergence of large, international, randomized controlled cardiovascular outcome trials (CVOT) assessing incident major adverse cardiovascular events (MACE) in patients taking novel glucose-lowering therapies. More recently, two classes of glucose-lowering therapies, namely sodium glucose cotransporter-2 inhibitors (SGLT2is) and glucagon-like peptide-1 receptor agonists (GLP-1RA) showed both cardiovascular and renal benefit. 7,8 SGLT2i have ubiquitously also shown a significant reduction in hospitalization from heart failure, with varying SGLT2i showing a neutral or positive effect on cardiovascular events, allcause mortality 8 and atrial fibrillation. 9 In addition, dapagliflozin became the first SGLT2i to be approved in Europe for the treatment of chronic kidney disease (CKD), irrespective of the presence or absence of type 2 diabetes, based on findings from the DAPA-CKD renal outcomes trial. 10 In addition, meta-analysis of randomized trials by Sattar et al. showed that GLP-1RA are associated with a reduced risk for all MACE events, all-cause mortality, hospital admission for heart failure, and worsening kidney function in people with type 2 diabetes. 11 In this study, we review the real-world impact of SGLT2i and GLP-1RA therapy, on all-cause mortality, hospitalization, cardiovascular outcomes and CKD when given either alone or together over 5 years following the initiation of treatment. Given the limited data regarding the impact of combination therapy on 'hard clinical endpoints' from clinical trials, we evaluated such combination therapy using a large real-world dataset and include a comparison of combination therapy versus monotherapy with either SGLT2i or GLP-1RA.

| Study population
This study was conducted with anonymized data from TriNetX, a global federated health research network that has access to both inpatient and outpatient electronic medical records (EMRs) from health care organizations (HCOs) from all over the world. This analysis was conducted on the Global Collaborative Network, which contains data from over 114 million patients with access to diagnoses, procedures, medications, laboratory values and genomic information worldwide. The global collaborative network collects data from across 14 different countries, from predominately US HCOs. In particular, for this retrospective cohort analysis, approximately 2.2 million patients with type 2 diabetes receiving insulin from 85 HCOs were included.
Data recorded between 1 January 2010 and 24 March 2023 were used in the subsequent analyses. As part of the data ingestion process when HCOs joins the network, data are mapped to a common data model to reflect individual institution, country and regional standards with regard to electronic health record data. All data collection, processing and transmission are performed in compliance with all Data Protection laws applicable to the contributing HCOs, including the EU Data Protection Law Regulation 2016/679, the General Data Protection Regulation on the protection of natural persons regarding the processing of personal data and the Health Insurance Portability and Accountability Act, the US federal law, which protects the privacy and security of health care data. The TriNetX Global Collaborative Network is a distributed network (with the majority of HCOs located in the United States), and analytics are performed at the HCO with only aggregate results being surfaced and returned to the platform. Data usage and publication agreements are in place with all HCOs.

| Building cohorts in TriNetX
Patients with type 2 diabetes were identified in each cohort based on the inclusion of the ICD-10-CM code E11 in their EMR. To avoid the potential of patients with type 1 diabetes being included and skewing the analysis, EMR code E10 was used as an exclusion criterium when building the cohorts. This analysis was conducted on four cohorts:  Tables S1-S5 for the complete list of pharmacological agents). This was done to ensure the people in each cohort were at a similar disease stage.
As this study was comparing specifically SGLT2i versus GLP-1RA effects on cardiovascular outcomes a control cohort excluding these medications was used as a standard reference point for comparison.
The analysis on the respective cohorts was based on the following outcomes over 5 years from the index event: all-cause mortality, any hospitalization, cardiovascular outcomes including acute-myocardial infarct, unstable angina, IHD, heart failure, atrial fibrillation, stroke, peripheral vascular disease (PVD), lower limb amputation and CKD.
Mortality data were collected directly from the electronic health records of HCOs, and were not collected from a government source, such as a registered persons database.
Using the above cohorts, a sub-analysis was performed comparing the combination cohort to both SGLT2i cohort and GLP-1RA cohort. The same PSM was performed between these cohorts as has previously been described, and the same outcomes by 5 years were recorded.

| Index event
The initiation of insulin was used as the index event for the control group and was incorporated into the index criteria for the treatment cohorts. The index events for the SGLT2i cohort were when both insulin and SGLT2i were prescribed, for the GLP-1RA cohort when both insulin and GLP-1RA were prescribed, and for the combination cohort when all three of insulin, SGLT2i and GLP-1RA were prescribed. The data collection started once all index criteria were met, which prevented the immortal time bias from affecting the cohorts.
The inclusion of insulin across the cohorts was undertaken to ensure the cohorts contained people with type 2 diabetes who were at a similar disease stage.
F I G U R E 1 Diagram of inclusion and exclusion criteria used to develop the different cohorts in the study. GLP-1RA, glucagon-like peptide 1 receptor agonist; HCO, health care organization; SGLT2is, sodium glucose cotransporter-2 inhibitors; T1DM, type 1 diabetes mellitus; T2DM, type 2 diabetes mellitus

| Statistical analysis
The control cohort was considered the reference cohort (HR = 1) when compared with the SGLT2i, GLP-1RA and SGLT2i + GLP-1RA cohorts. Using the TriNetX software a survival analysis was performed, which estimates the probability of an outcome at a respective time interval (daily time interval was used in these analyses) over 5 years from the index event and generates a hazard ratio (HR), log rank test and Kaplan-Meier survival curve. TriNetX uses the R Survival package v3.2-3 for its analysis. Patients were excluded from an outcome analysis if they had already experienced the outcome before the time window.
To account for patients who dropped out of the analysis, censoring is applied. A patient was removed (censored) from the analysis after the last event in their electronic record.

| Propensity score matching
After propensity matching for age, gender, presence of IHD, hypertension, heart failure, CKD and HbA1c value, each pair of cohorts (SGLT2i vs. control, GLP-1RA vs. control and combination vs. control) were deemed well matched with almost all differences between these characteristics being non-significant. PSM was also performed in the sub-analysis, using the same characteristics as in the primary analysis, each pair of cohorts (combination vs. SGLT2i and combination vs. GLP-1RA) were deemed to be well matched; however, some variables remained statistically significant between cohorts. Tables S1-S5 list the race and medication breakdowns of the different cohorts after propensity matching (Table 1).

| Sodium-glucose cotransporter 2 inhibitor versus control
In all the analysed events, treatment with SGLT2i reduced the hazard rate of that event occurring over 5 years when compared with the propensity matched control cohort's hazard rate, this is evi-

| Log rank test
For all of the survival analyses between these cohorts, SGLT2i versus control, GLP-1RA versus control and combination versus control, the log rank test evidenced a significant difference between the survival curves for each outcome (see Table 2 for each log rank test).

| Combination versus monotherapy treatment
3.4.1 | Combination sodium-glucose cotransporter 2 inhibitor and glucagon-like peptide-1 receptor agonist versus sodium-glucose cotransporter 2 inhibitor The combination cohort showed a modest reduction in risk for IHD ( Figure 2C.

| Log rank test
For each survival analysis performed between the combination versus SGLT2i cohorts the log rank test showed a significant difference between all survival curves except for the outcomes for unstable angina ( p = .327) and PVD ( p = 1). For the combination versus GLP-1RA cohorts the log rank test showed a significant difference between all survival curves except for the outcomes for unstable angina (p = 1), IHD ( p = 1) and PVD ( p = .090) ( Table 3).

| DISCUSSION AND CONCLUSION
Our study showed that treatment with SGLT2i or GLP-1RA, both as monotherapy and combination, confer prognostic cardiovascular benefit on people with type 2 diabetes when compared with people who are treatment naïve, over 5 years. Second, our data suggests that combination therapy confers a greater all-cause mortality benefit than either monotherapy. Finally, combination therapy appears to provide a greater benefit for most cardiovascular and CKD outcomes than monotherapy with GLP-1RA or SGLT2i. F I G U R E 2 (A) Hazard ratio (HR) of each outcome occurring, comparing the intervention cohorts to control cohort with 95% confidence intervals (CIs). (B) HR of each outcome comparing the combination cohort to the sodium glucose cotransporter-2 inhibitor (SGLT2i) cohort with 95% CIs. (C) HR of each outcome comparing the combination cohort to the glucagon-like peptide 1 receptor agonist (GLP-1RA) cohort with 95% CIs. CKD, chronic kidney disease; IHD, ischaemic heart disease; PVD, peripheral vascular disease T A B L E 3 Summary of outcome incidence, survival probability and Log rank test for each cohort and event in the sub-analysis  16 and DECLARE-TIMI 58 with dapagliflozin 17 ). Importantly, the EMPA-REG OUTCOME trial (empagliflozin) showed a significantly reduced all-cause mortality rate. 15   data. [37][38][39] This is a large real-world study undertaken to evaluate the benefit of monotherapy and combination therapy on mortality and cardiovascular outcomes. Our study was able to leverage data from 7 million patients with type 2 diabetes from 85 different HCOs in multiple countries and our final analysis included approximately 108 000 dual therapy patients compared with 186 000 in the GLP-1RA cohort and 143 000 in the SGLT2i cohort. The population size of our study and the long duration of the analysis adds credible evidence to the literature as to the beneficial effects of combination therapy on all-cause mortality.
Given the retrospective nature of the study, several limitations exist. First, these are real world data, which are not randomized comparisons nor are they controlled. Specifically, the effect of socioeconomic status on access to novel anti-glycaemic treatments cannot be accounted for in this study. Second, as data are extracted from the EHR from an administrative database, there is a potential for a lack of data completeness, data may not be recorded by the HCO or recorded in free text that we are unable to extrapolate. Specifically, mortality data were extracted only from the electronic health records of the participating HCOs and not a government source, such as a registered persons database. In addition, should participants move between HCO, it is possible that some of their data may not be avail-