Cardiovascular and mortality outcomes with GLP-1 receptor agonists vs other glucose-lowering drugs in individuals with NAFLD and type 2 diabetes: a large population-based matched cohort study

This large, population-based, retrospective cohort study was conducted using the TriNetX research network (Cambridge, MA, USA). TriNetX is a federated multicentre research network that provides real-time access to an anonymised dataset from participating healthcare organisations' electronic health records (EHR). Details of the data source, quality checks and diagnosis codes used for patient selection (according to predefined ICD-9 [http://www.icd9data.com/2007/Volume1/default.htm↗] and ICD-10 [https://icd.who.int/browse10/2019/en↗] codes) are described in the electronic supplementary material (ESM) Methods. Details of the TriNetX network are described in previous studies [15, 16]. We followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

We identified all adult (aged ≥20 years) patients with NAFLD and type 2 diabetes who had newly started treatment with non-insulin glucose-lowering drugs (metformin, GLP-1RAs, SGLT2 inhibitors, sulfonylureas, meglitinides, thiazolidinediones, acarbose or dipeptidyl peptidase-4 inhibitors) between 1 January 2013 and 31 December 2021. We limited the study cohort to patients with at least 1 year of follow-up before cohort entry (i.e. patients who received their first glucose-lowering prescription more than 1 year before cohort entry). Furthermore, to reduce reverse causality and detection bias, we only included individuals with more than 1 year of follow-up after the start of the study.

The identification of NAFLD at the study's outset was based on the presence of specific ICD‐9 and ICD-10 codes. Individuals diagnosed with other liver diseases with any ICD codes were excluded from the NAFLD group. Individuals were excluded if they met any of the following criteria: chronic liver disease other than NAFLD, including alcohol, viral, drug-induced, autoimmune and genetic diseases, liver cirrhosis, or clinical diagnosis of hepatic decompensation (such as oesophageal varices or ascites); history of excessive alcohol use, alcohol abuse, or alcohol use disorder or history of alcohol-related disorders; HIV infection; solid organ transplantation; or individuals with eGFR <30 ml/min per 1.73m² (within 6 months before the cohort entry) and history of undergoing dialysis treatment. Finally, individuals with a history of heart failure, ischaemic heart disease, unstable angina, myocardial infarction, aortic aneurysm or dissection, stroke (ischaemic or haemorrhagic stroke), cerebral infarction, transient ischaemic attack, carotid intervention or surgery, coronary stenting, percutaneous coronary intervention (PCI) or coronary artery bypass before inclusion to the cohort or prior to the index event were also excluded. We further excluded individuals with a history of treatment with insulin before their initial prescription for a non-insulin glucose-lowering drug (since insulin therapy at baseline is considered for patients with advanced disease), women with a history of polycystic ovary syndrome or gestational diabetes, and individuals with type 1 diabetes. Patients were followed from 1 year after cohort entry until a diagnosis of incident adverse CVEs, switch to a comparator drug, any cause of death or the end of the study (30 September 2022), whichever occurred first.

The study cohort consisted of new users of GLP-1RAs (dulaglutide, exenatide, liraglutide, lixisenatide or semaglutide), SGLT2 inhibitors (canagliflozin, dapagliflozin or empagliflozin), metformin and other second- or third-line glucose-lowering drugs (thiazolidinediones, dipeptidyl peptidase-4 inhibitors, sulfonylureas, meglitinides, α-glucosidase inhibitors, insulin or a combination of glucose-lowering drugs). Furthermore, the study cohort included individuals who either switched to or added on treatment with other glucose-lowering drugs. Cohort entry was defined as the date of the first-ever prescription for one of the drugs of interest (GLP-1RAs, SGLT2 inhibitors, metformin or other second- or third-line glucose-lowering drugs) during the study period, and their respective first exposure was defined as the index event. We used a lag of 6 months for all exposures to minimise protopathic bias and allow for a minimum and sufficient latency period after cohort entry [17, 18].

The primary outcome was to assess the incidence of new-onset adverse CVEs, which were categorised as: (1) heart failure; (2) major adverse cardiovascular events (MACE); and (3) composite cerebrovascular events. MACE, as a composite endpoint, was defined as the first occurrence of unstable angina, myocardial infarction or revascularisation (including PCI or coronary artery bypass graft) [19, 20]. A composite of cerebrovascular events was defined as the first occurrence of stroke (haemorrhagic or ischaemic stroke), transient ischaemic attack, cerebral infarction, carotid intervention or surgery [19, 20]. The secondary outcome was to evaluate the incidence of all-cause mortality.

All analyses were performed using the TriNetX real-time analytics platform. This approach involves dynamic and immediate data analysis, enabling continuous processing and interpretation of data as it is generated. Categorical variables were compared using the Pearson χ² test, and continuous variables were compared using an independent-sample t test. Continuous variables were expressed as mean ± SD and categorical variables were presented as frequency and percentage. Analyses were performed to examine the rate of adverse CVEs using Cox proportional hazards models. HRs and CIs, along with tests for proportionality, were calculated using R's Survival package v3.2-3. The results were validated by comparing them with the output from SAS version 9.4. Patients were censored when the time window ended or the day after the last fact in their record. We utilised a 1:1 propensity matching strategy to establish comparable groups of patients treated with different glucose-lowering drugs, including GLP-1RA, SGLT2 inhibitors and metformin. In addition, we used this matching approach to balance the covariates between the groups effectively. To account for clustering within the 1:1 propensity-matched sample and address the loss of independence among individuals resulting from the matching procedure, we incorporated a robust variance estimator in the Cox regression model [21]. The robust variance estimator is essential in enhancing the accuracy of our analytical approach and ensuring the validity of the study's findings. A priori-defined two-sided alpha of <0.05 was used for statistical significance.

We used metformin as a control in the secondary analysis. In this analysis, we matched new users of GLP-1RAs to new metformin users on propensity scores.

We conducted two sensitivity analyses to ensure the robustness of our findings due to the heterogeneous nature of adverse CVEs. The first sensitivity analysis evaluated the influence of different reference groups/disease stages. We compared the use of GLP-1RAs with second- or third-line glucose-lowering therapy to minimise confounding [22], thereby allowing us to explore the impact of including a reference group with the same disease stage, even though we adjusted for microvascular complications. In the second sensitivity analysis, we estimated the rates of new incidences of study outcomes by excluding patients with outcomes 2 years after the index event. Both analyses were performed using the same methods as the primary analysis.

The mean follow-up was 4.8±1.1 years for the GLP-1RAs group and 5.3±1.2 years for the SGLT2 inhibitors group. After PSM, the GLP-1RAs and SGLT2 inhibitors groups (n=38,804 in each) were well matched (ESM Fig. 1). Clinical characteristics of the GLP-1RAs and SGLT2 inhibitors groups are compared in ESM Table 1. After PSM, among the lipid profiles, TG were significantly higher among the SGLT2 inhibitors group (2.44 ± 2.84 mmol/l vs 2.51 ± 2.81 mmol/l, ESM Table 1).

Several studies have shown that metformin may correct components of the metabolic syndrome [23]. Moreover, in individuals with type 2 diabetes, metformin may provide cardiovascular protection that is not solely attributed to its glucose-lowering effects, as it also has antihypertensive effects [23]. These potential cardioprotective effects may be related to the favourable actions of metformin on lipid metabolism, vascular smooth muscle, hypercoagulation, platelet hyperactivity, cardiomyocyte intracellular calcium handling and endothelial function [24]. We therefore used metformin as a control group in an analysis including 288,675 patients with NAFLD and type 2 diabetes, of which 103,667 individuals were receiving metformin (metformin group; ESM Fig 2). A well-matched population was found in the GLP-1RAs vs metformin groups (n=47,097 each) after PSM (ESM Figs 2 and 3). The mean follow-up was 4.8±1.1 years for the GLP-1RAs group and 4.4±1.4 years for the metformin group. ESM Tables 2 and 3 compare the baseline, demographic and clinical characteristics of the GLP-1RAs and metformin groups.

The rate of new onset of heart failure (HR 0.92; 95% CI 0.88, 0.96) and composite cerebrovascular diseases (HR 0.91; 95% CI 0.87, 0.95) was significantly lower in the GLP-1RAs group. However, the two groups had no evidence of a difference in the composite incidence of MACE (HR 0.95; 95% CI 0.90, 1.08). Mortality rate was significantly lower in the GLP-1RAs group (HR 0.62; 95% CI 0.58, 0.66; ESM Fig.). 4

We conducted the first sensitivity analysis to compare GLP-1RAs with second- or third-line glucose-lowering medications. After PSM, the groups (n=52,166 each) were well matched (ESM Figs. 5 and 6). The mean follow-up was 4.8±1.1 years for the GLP-1RAs group and 4.9±1.9 years for the second- or third-line glucose-lowering medications group. A comparison of baseline demographics and clinical characteristics between GLP-1RAs and second- or third-line glucose-lowering medications is presented in ESM Tables 4 and 5.

The GLP-1RAs group had a lower rate of adverse CVEs, including new-onset heart failure (HR 0.88; 95% CI 0.85, 0.92), composite incidence MACE (HR 0.89; 95% CI 0.85, 0.94) and composite cerebrovascular events (HR 0.93; 95% CI 0.89, 0.96), compared with other second- or third-line glucose-lowering medications. Mortality rate was also significantly lower in the GLP-1RAs group (HR 0.70; 95% CI 0.66, 0.75; ESM Fig.). 7

In the second sensitivity analysis, we estimated the rates of new incidences of study outcomes by excluding patients with outcomes 2 years after the index event. The analysis led to generally consistent results, which were similar to the primary analysis between cohorts of new users of GLP-1RAs compared with new users of SGLT2 inhibitors, metformin or other second- or third-line medications. Both analyses were performed using the same methods as the primary analysis (ESM Tables–). 6 8

Our study had several strengths. First, we analysed large population-based, multicentre data collected nationwide from healthcare organisations. Second, baselines and potential confounders were adjusted to create a robust control group. Third, the large sample size in the propensity-matched analyses resulted in narrower CIs, demonstrating higher precision. Hence, using PSM provided additional strength to the study results. Furthermore, we used a new user cohort study design as a comparator, which reduced the potential for unmeasured confounding [42]. Lastly, we ensured that the sample comprised only individuals with NAFLD and type 2 diabetes using an exhaustive list of quality checks and diagnosis codes, including but not limited to the exclusion of alcohol-related disorders, alcohol abuse, alcohol dependence and alcohol-related liver diseases. We believe that this is the largest published study comparing adverse CVEs in patients with NAFLD and type 2 diabetes with and without GLP-1RA treatment.

Our study also had some limitations. First, the retrospective design and the reliance on an EHR-based database limited our results. Whenever patient information is translated into diagnosis codes, data from EHR-based databases are susceptible to errors in coding. Standardised measures were used to identify cases to minimise documentation errors. Second, we did not account for some residual confounding even after adjusting potential confounders in our analyses. However, we used new users as a cohort to reduce the potential for unmeasured confounding. Additionally, we did not grade comorbid conditions at baseline, which might have caused some selection bias in the cohorts. Third, we did not include imaging modalities to confirm the NAFLD diagnosis (e.g. ultrasonography, computed tomography, MRI or elastography), as it was not available in our dataset. In future studies, it would be ideal to link NASH and type 2 diabetes separately to adverse CVEs incidence and mortality rate. We acknowledge that there were some differences in the comparison between GLP-1RAs and other glucose-lowering agents, particularly concerning HbA_1c levels, the use of insulin and the use of SGLT2 inhibitors, which may have influenced the results. Excluding patients with prior CVD and cerebrovascular disease could have influenced the outcomes of our study and potentially impacted the comparison between GLP-1RAs and SGLT2 inhibitors. Finally, the inclusion criteria were not based on histological diagnosis, which could lead to some data contamination due to misdiagnosis.

The results of this large population-based cohort study indicate that in patients with NAFLD and type 2 diabetes, GLP-1RAs and SGLT2 inhibitors are associated with lower incidences of macrovascular diseases and all-cause mortality compared with other glucose-lowering drugs. Though our study provides novel information, randomised controlled trials and additional observational studies are needed to corroborate our findings, given the significant morbidity and mortality of CVEs in patients with NAFLD and type 2 diabetes. Compared with metformin and second- or third-line glucose-lowering medication groups, major adverse CVEs were significantly lower in patients with NAFLD and type 2 diabetes taking GLP-1RAs, but not different from those taking SGLT2 inhibitors. Routine cardiovascular risk assessments and timely implementation of GLP-1RAs and SGLT2 inhibitors are also essential for reducing the adverse CVEs in patients with NAFLD and type 2 diabetes.

Below is the link to the electronic supplementary material. Supplementary file1 (PDF 840 KB)