Sulfonylurea is associated with higher risks of ventricular arrhythmia or sudden cardiac death compared to metformin: a population-based cohort study

Background: Commonly prescribed diabetic medications such as metformin and sulfonylurea may be associated with different arrhythmogenic risks. This study compared the risk of ventricular arrhythmia or sudden cardiac death (VA/SCD) between metformin and sulfonylurea in patients with type 2 diabetes. Methods: Patients aged 40 years or older who were diagnosed with type-2 diabetes mellitus or prescribed anti-diabetic agents in Hong Kong between 1 st January 2009 and 31 st December 2009 were included and followed up till 31 st December 2019. Patients prescribed with both metformin and sulfonylurea or had prior myocardial infarction were excluded. The study outcome was a composite of VA/SCD. Metformin users and sulfonylurea users were matched at a 1:1 ratio by propensity score matching. Results: The study cohort consisted of 5756 metformin users (48.00% male, age 68±11 years, mean follow-up 5.04±2.57 years) and 5756 sulfonylurea users (50.30% male, age 69±11 years, mean follow-up 5.04±2.54 years). Sulfonylurea was associated with higher risk of VA/SCD than metformin (hazard ratio 1.84 [1.56 - 2.16], p<0.001). Such difference was consistently observed in subgroup analyses stratifying for insulin usage or known coronary heart disease. Conclusion: Sulfonylurea use is associated with higher risk of VA/SCD than metformin in patients with type 2 diabetes. Further to study the of in different of the


Introduction
Type 2 diabetes mellitus is a highly prevalent condition worldwide, with 1 in 11 adults suffering from it globally. Every year, type 2 diabetes mellitus alone causes over 1 million deaths, making it the ninth leading cause of mortality (1). Cardiovascular complications are signi cantly associated with death among type 2 diabetes patients (2), with ventricular arrhythmia and sudden cardiac death (VA/SCD) as the most common macrovascular complications (3). Siscovick and colleagues estimated the incidence rate of sudden cardiac arrest to be 3.15 per 1000 diabetic patients without prior clinically-recognized heart disease; and the incident rate was 13.80 per 1000 diabetic patients with clinically-recognized heart disease, 3.84 and 2.31 times higher when compared with non-diabetic patients respectively (4). The elevated risk can be attributed to a poor glycemic control and other risk factors such as dyslipidemia and nephropathy (4).
Metformin and sulfonylurea were frequently prescribed because of their effective glycemic control and low cost even though their cardiovascular risks continue to be debated (5). The choice of antidiabetic medications on atrial brillation has been well documented (6, 7), yet not many studies investigated the effect of such medications on the risk of VA/SCD, let alone the comparison between metformin and sulfonylurea use. There is a need for the above investigation as sudden cardiac death, the most devastating manifestation of VA, is the leading cause of death among type 2 diabetes patients with cardiovascular complications (8).
Therefore, this study aimed to compare the risk of developing VA/SCD between metformin and sulfonylurea users.

Data Source
This study has been approved by The Joint Chinese University of Hong Kong -New Territories East Cluster Clinical Research Ethics Committee and the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster. This was a population-based retrospective cohort study. Data was extracted from the Clinical Data Analysis and Reporting System (CDARS), a territory-wide electronic database storing patient healthcare records managed by the Hong Kong Hospital Authority, which manages 42 hospitals, 47 Specialist Out-patient Clinics, and 73 General Out-patient Clinics; serving a population of 7 million. This database has previously been used for cohort studies by other teams in Hong Kong (9,10). Informed consent was not required from patients as the data used in this study were anonymized.

Study Cohort
Patients who ful lled all of the following inclusion criteria were included: 1) aged 40 years or older; 2) had documented diagnosis of type 2 diabetes mellitus under the International Classi cation of Diseases, Ninth Revision (ICD-9) coding system, or prescribed antidiabetic agents between January 1st, 2009, to December 31st, 2009. Patients who were prescribed both metformin and sulfonylurea or had past medical history of myocardial infarction , as identi ed using ICD-9 codes, were excluded from the study.

Outcomes
The primary outcome of the study is a composite of ventricular arrhythmia and sudden cardiac death (VA/SCD), as identi ed using ICD-9 codes. A diagnosis of VA/SCD was made based on clinical judgment by the treating physician and was then coded according to the ICD-9 system into CDARS. VA/SCD episodes where a MI occurred within 1 week before or after the VA/SCD episode were considered AMI-related and thus excluded. Patients were followed up till December 31 st 2019.

Propensity Score Matching
To ensure comparability of the cohorts, propensity score matching was performed with the baseline demographics of gender, age, and duration since type 2 diabetes diagnosis; prior comorbidities including peripheral vascular disease, ischemic stroke, atrial brillation, heart failure, prior VA/SCD, intracranial hemorrhage, coronary heart disease, hypertension and chronic obstructive pulmonary disease; baseline medication use including ACE inhibitors, beta blockers, calcium channel blockers, diuretics, and use of diabetic medications including insulin, thiazolidinediones, dipeptidyl-peptidase 4 (DPP4) inhibitors and glucagon-like peptide (GLP-1) agonists; and laboratory tests including serum levels of high density lipoprotein, low density lipoprotein, triglycerides, and HbA1c.
The ICD-9 codes used for classi cation of the corresponding medical conditions are available in Supplementary Table 1.
Imputation was not performed as patients with missing data were excluded during the generation of the matched cohort. Patients with use of either metformin or sulfonylurea were matched on a 1:1 ratio on the propensity score which was derived from logistic regression using the nearest-neighbor matching algorithm. To assess the balance of covariates, the standardized mean difference (SMD), which is the difference in means or proportions over the pooled standard deviation (SD) was used for categorical covariates, while the variance ratio (VR), which is the ratio of variance between the treatment and control groups was used for both categorical and continuous covariates. A caliper width of 0.2 was chosen as it is considered as the optimal caliper width for estimating mean differences (11). Covariates were considered balanced in both groups when the standardized mean difference was less than 0.1, and when the VR lied between the range of 0.5 and 2.0 (12, 13).

Statistical Analysis
Continuous variables were presented as the mean (Standard deviation [SD]) and categorical variables were presented as frequency (%). Hazard ratios (HRs), along with 95% CIs and P-values were reported accordingly with the Cox proportional hazards regression model. Kaplan-Meier curves were plotted against the time-to-event for VA/SCD strati ed by either metformin or sulfonylurea use. The log-rank test was performed to investigate the statistical signi cance between metformin and sulfonylurea groups. Incidence rates for VA/SCD were presented as incidence per 1000 person-years along with 95% CIs. Subgroup analyses were conducted to investigate the risk of VA/SCD by users versus nonusers of insulin and patients with versus without prior coronary heart disease, and by different types of sulfonylurea. Sensitivity analysis was performed by excluding patients with prior cardiomyopathy or valvular disease. Computed E-values for HRs were computed to assess the effects of any unmeasured confounding on our study.
The E-value is de ned as the minimum strength of association that an unmeasured confounder would require to have with both the outcome and treatment to fully explain away a speci c association between the treatment and outcome, conditional on the measured covariates (14). In this study, a large E-value implies that an unmeasured confounder must be very strong to explain away the effect of sulfonylurea over metformin use in the risk of developing VA/SCD.
All signi cance tests were two-tailed and considered signi cant when P < 0.05. No imputation was performed for missing data.

Results
The steps for the selection of patients for this study cohort are shown in Figure 1. In total, 261,308 patients ful lled the inclusion criteria. After excluding patients with missing data, the study cohort consisted of 15,158 patients (46.7% male, mean baseline age: 68 ± 11 years old), where 8,542 (22.41%) patients received metformin and 6,616 (17.35%) patients received sulfonylurea. After 1:1 propensity score matching, the nal study cohort consisted of 5,756 metformin users and 5,756 matched sulfonylurea users. The baseline and clinical characteristics of the study population before and after propensity score matching were shown in Table 1. A Love plot summarizing covariate balances before and after propensity score matching is provided in Supplementary Figure 1. SMD values were less than 0.1 and VR values were within 0.5-2.0, indicating a good balance in baseline characteristics between the two cohorts; the only exception was age, which had a SMD of 1.1, but the VR value was 1.07 which was very close to 1.0, indicative of good balance (15).
The main and subgroup analysis of VA/SCD risk were presented in Table 2. The mean follow-up duration was 5.08±2.56 years for the metformin cohort, and 5.01±2.54 years for the sulfonylurea cohort. During the study period, 229 metformin users and 415 sulfonylurea users exhibited episodes of VA/SCD. Cox proportional hazards model analyses over the entire follow-up showed that sulfonylurea use was associated with an overall higher risk of VA/SCD than metformin use as visualized by the Kaplan-Meier curves in Figure 2.
Subgroup analyses were performed by sulfonylurea type, concurrent use of insulin, and the presence of coronary heart disease ( Table 2); sensitivity analysis was performed by excluding patients with either valvular heart disease or cardiomyopathy (

Discussion
The main nding of this study is that sulfonylurea users have a higher risk of ventricular arrhythmia and sudden cardiac death than metformin users. These results were consistent irrespective of severity of diabetes or history of coronary heart disease.

Mechanism underlying the observations
It is hypothesized that sulfonylurea may inhibit the delayed recti er potassium channel, leading to prolonged QT interval (16). A trial conducted in 30 type 2 diabetes patients found that patients randomized to glyburide were associated with an increase in corrected QT interval (QTc) compared to patients randomized to metformin (17). Sulfonylurea carries a high risk of hypoglycemia compared to other antidiabetic medications (18), in turn prolonging action potentials in myocardial tissue by blocking potassium channels at the cellular level (19). Multiple studies found that severe hypoglycemia, a major concern of sulfonylurea, increased the risk of VA/SCD (20)(21)(22). However, sulfonylurea may also have anti-arrhythmic effects by inhibiting reentrant arrhythmias by a mechanism known as ischemic preconditioning (IPC), and thus reducing the risk of developing cardiac arrest (16), but it has been suggested that the effects of IPC are abolished in type 2 diabetes (23). Metformin has pleotropic effects with many cardiovascular bene ts as shown in basic science studies (24,25); it was found to be associated with a decreased QTc in animal models (26, 27), but no decrease in ventricular arrhythmic outcomes were re ected in clinical trials (28).
Comparison with previous observational studies Two recent observational studies have investigated the association between metformin and sulfonylurea use and risk of VA/SCD (29,30). Ostropolets and colleagues found that diabetic patients on metformin monotherapy had a reduced risk of VA compared to sulfonylurea monotherapy (29). However, patients with a history of atrial brillation, ventricular tachycardia and ventricular brillation were excluded from their study, so their results may not be generalizable to patients with prior arrhythmic conditions. Moreover, younger type 2 diabetes patients were not captured as only patients above the age of 50 were included in their study, which may lead to biased results. A recent systematic review has shown that type 2 diabetes is increasingly diagnosed in patients under the age of 50 in many countries worldwide (31). Our study included type 2 diabetes patients aged 40 or above, which may be more generalizable to a larger population.
Conversely, Eroglu and colleagues found that sulfonylurea antidiabetics were associated with a lower risk of developing out-ofhospital cardiac arrest (30). However, their study did not match cases and controls by duration of diabetes, which is a risk factor for ventricular arrhythmias (4). The duration of diabetes was accounted for during propensity score matching in our study. Moreover, age was not evenly distributed in their study groups, as patients on sulfonylurea drugs alone were older (mean age: 75.2, SD: 9.7) than those on metformin alone (mean age: 69.6, SD: 10.1). Finally, their sample size was small for both patients on sulfonylurea alone (N=215) and patients on metformin alone (N=385), while our study included 5756 metformin users and 5756 sulfonylurea users after matching.

Implications of subgroup and sensitivity analyses
Sensitivity analysis was performed by excluding patients with a history of cardiomyopathy or valvular heart disease, as cardiomyopathies are a common cause of SCD (32, 33); a systematic review identi ed cardiomyopathies as one of the top causes of SCD in Chinese patients (34). Similarly, studies have found associations of aortic valve disease and mitral valve prolapse with VA/SCD (35-38). It was found that the risk for sulfonylurea compared to metformin to develop VA/SCD was similar in both sensitivity and overall analysis.
Common complications of type 2 diabetes include coronary artery disease and myopathy (39), which are risk factors for developing VA/SCD (40). The risk for these complications increases with the duration and severity of type 2 diabetes, which is consistent with the underlying low-grade in ammation and glycation (41,42). As such, it was possible that the risk of VA/SCD increases with the duration and severity of diabetes, and the in uence from medication-related effects may become less important accordingly. To better elucidate drug-related effects and minimize confounding by the above factors, we performed a subgroup analysis using insulin usage as a surrogate of diabetic duration and severity. We found that sulfonylurea was consistently associated with greater risk of developing VA/SCD than metformin regardless of diabetes severity, suggesting that the differences between these two drugs are clinically important in diabetic patients regardless of their condition's severity and duration.
Myocardial ischemia in coronary heart disease alters metabolic and electrical processes in the heart, altering the propagation and conduction of resting and acting membrane potentials, leading to cardiac arrhythmias (43). We performed subgroup analyses to explore if coronary heart disease would be a dominant risk factor for VA/SCD such that the differences in arrhythmogenicity between sulfonylurea and metformin would be considered relatively insigni cant. The results of the subgroup analyses found the risk of developing VA/SCD was consistently higher in metformin users compared to sulfonylurea users even in patients with coronary heart disease. This further suggested that the differences in arrhythmogenicity between the two drugs were signi cant and clinically important regardless of patients' inherent risks for VA/SCD.
Subgroup analyses were performed to investigate the risk of VA/SCD among individual types of sulfonylurea. Only glicazide and tolbutamide users were statistically signi cant for the risk of VA/SCD when compared to metformin. Tolbutamide users were at a higher risk of VA/SCD, but this may be attributed to a higher usage of pro-arrhythmic calcium channel blockers and beta blockers in the subgroup (44).

Clinical implications
Our study has important clinical implications. Although the use of sulfonylurea has decreased in recent years, it remains a commonly prescribed antidiabetic agent, second to metformin (45,46). Diabetic nephropathy is a common complication among type 2 diabetes patients with a prevalence of 31.6% in Hong Kong (47). When metformin is contraindicated, such as in patients with severe kidney impairment, sulfonylurea is a viable alternative (48, 49). The Kidney Disease: Improving Global Outcomes (KDIGO) 2020 guidelines recommends patients with an estimated glomerular ltration rate of ≤30 mL/min/1.73 m 2 to discontinue metformin therapy (48, 50) However, in practice, the risk of developing VA/SCD by antidiabetic choice is often neglected. Aside from increased mortality risk, VA/SCD necessitates further therapy such as the use of implantable cardioverterde brillators and antiarrhythmic agents and thereby imposes more healthcare burden (51, 52). Patient adherence to metformin may be di cult due to the common side effect of gastrointestinal disturbance (53), but this can be avoided with alternative formulations such as extended-release metformin (54). Given the ndings of our study, there exists a compelling case to move away from prescribing sulfonylurea for glycemic control.

Study limitations
This study has limitations. As this study was retrospective in nature, the effect of unmeasured confounders on the risk of developing VA/SCD cannot be ruled out. For instance, smoking status and alcohol consumption are not recorded by CDARS. Nonetheless, we have included multiple signi cant risk factors in the propensity score matching. Furthermore, the E-value suggested that the observed association of the higher VA/SCD risk in sulfonylurea over metformin users would only be insigni cant if an unmeasured risk factor exists with an HR of 2.69 to 3.12, which is realistically unlikely.
Recent studies have shown that the risk of VA/SCD differed by choice of sulfonylurea (55-58). Further research comparing risk of VA/SCD between different drugs of the sulfonylurea class is warranted, such that clinicians can avoid choosing sulfonylurea with high arrhythmogenic risk.
To conclude, this study found that among patients diagnosed with type 2 diabetes, use of sulfonylurea was associated with higher risk of developing VA/SCD compared to use of metformin. The increased risk was consistent in patients with severe diabetes and in those with a history of coronary heart disease. Hence, the use of sulfonylurea should be reconsidered in patients at risk of VA/SCD.
Further studies are warranted to study the risk of VA/SCD in different drugs of the sulfonylurea class.

Availability of data and materials
The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Competing interests
None.
Funding None. Tables Table 1 Baseline characteristics before and after 1:1 propensity score matching.