Atrial fibrillation (AF) is the most common arrhythmia in clinical practice and is considered an independent risk factor for death, with a twofold increased risk of all-cause mortality in women and a 1.5-fold increase in men, with an overall 3.5-fold mortality risk increase due to heart failure (HF), malignancy, infection and embolic cerebrovascular disease [1, 2, 3].

Type 2 diabetes is generally known to be an independent risk factor for AF. A meta-analysis of several cohort and case-control studies reported that patients with type 2 diabetes had a 34% greater risk of developing AF than the general population [4]. The pathophysiology of diabetes-related AF is not fully understood, but it is associated with structural and electrical remodelling and autonomic dysfunction in the atria as a result of glycemic fluctuation, oxidative stress and inflammation [5]. It is known that glucose-lowering drugs have differing effects on the incidents of AF. In general, glucose-lowering drugs that cause hypoglycemia are known to increase the risk of AF [5].

In observational studies, thiazolidinedione (TZD), an insulin sensitizer, has shown a protective effect against the development of new-onset AF. This may be because of its pleiotropic effects, such as anti-inflammatory and antioxidant properties [6].

A new class of oral hypoglycemic agents, sodium-glucose cotransporter-2 inhibitors (SGLT-2i), has been shown to potentially reduce the risk of cardiovascular outcomes in terms of HF and all-cause mortality [7, 8]. SGLT-2i has also shown benefit in reducing the relative risk of AF in patients with type 2 diabetes [9, 10]. The mechanism of this favorable outcome is incompletely understood, but it has been suggested that increased renal glucose excretion induces additional osmotic diuresis, leading to decreased arterial blood pressure, improved myocardial efficiency, delayed myocardial structural remodeling, and decreased atrial dilation [11].

Therefore, we designed this study to compare the effects that SGLT-2i and TZD has on the risk of AF in patients with type 2 diabetes.

This study used the National Health Insurance Service (NHIS) database, established by the Korean NHIS. In order to access the NHIS database, an application form for a research proposal must be submitted to the NHIS Korean Committee of Research Support. After a review and approval, data are made available in a deidentified format. This study was also approved by the Institutional Review Board of Dongguk University Ilsan Hospital (No.: 2019-02-002-002). Written informed consent was waived by the Institutional Review Board. All methods in this study were performed in accordance with the Declaration of Helsinki.

The NHIS in Korea is a single-payer healthcare system, mandatory for all residents of Korea. The NHIS established a national health information database which includes patient demographic information, medical claims, medications, health checkups and death information. Because the NHIS provides regular cost-free health checkups, which include a physical examination, blood tests, and urine tests to all applicable examinees, these results can be integrated with other medical information so that comprehensive and detailed analyses are possible.

2.1 Study Population

Because SGLT-2i has been available since 2014 in Korea, we included patients from January 2014 to December 2018 who were new users of either SGLT-2i or TZD. We defined type 2 diabetes as the presence of ICD-10 code (E11-14). We defined a new user as any patient who had received any SGLT-2i (dapagliflozin, empagliflozin, ipragliflozin, or ertugliflozin) or TZD (pioglitazone or lobeglitazone) between January 2014 and December 2018, with a one-year washout period. Among the new users, we only included those who had been prescribed study drugs for 90 or more days. Diagnostic codes of patients from January 2008 to December 2013 were collected and combined to determine the baseline characteristics of enrolled patients.

We identified a total of 481,515 new users of SGLT-2i or TZD. Index year was the year the drug of interest started. We excluded patients who has a history of AF or had been prescribed study drugs for fewer than 90 days or patients who had been prescribed SGLT-2i and TZD simultaneously. We also excluded patients younger than 18 or older than 84 years, patients who had malignancy, end-stage renal disease, or patients who had not had a health checkup within one year prior to the index year. Fifty-one individuals with extremely high low-density lipoprotein (LDL)-cholesterol levels ($\ge$300 mg/dL) were excluded due to possible familial hypercholesterolemia. We identified a total of 163,384 patients, of which 69,122 used SGLT-2i and 94,262 used TZD. After propensity matching, a total of 109,986 patients were identified, 54,993 in each group (Fig. 1). Participants were followed until the outcome event, death, or 31 December 2018 whichever came first.

Fig. 1.

Flow chart of study population. LDL, low-density lipoprotein; SGLT-2i, sodium-glucose cotransporter-2 inhibitor; TZD, thiazolidinedione.

In this retrospective cohort study using the NHIS database of Korea, we found no differences in the incidence of new-onset AF between type 2 diabetes patients treated with SGLT-2i and TZD. The SGLT-2i-treated patients tend to be associated with lower AF risk concurrent with hHF, although this did not reach statistical significance.

Recently published meta-analysis with 16 randomized control trials demonstrated that SGLT-2i (empagliflozin 4 studies, canagliflozin 6 studies, dapagliflozin 6 studies) had great benefits in reducing the risk of AF/AFL (atrial flutter) in type 2 diabetes populations [9]. These benefits were not affected by age, body weight, glycosylated haemoglobin A1c (HbA1c), or systolic blood pressure. In the study, dapagliflozin was associated with a significant reduction of AF/AFL, whereas canagliflozin and empagliflozin had no effect on reducing AF/AFL events. A subgroup analysis of the DECLARE-TIMI 58 trial of patients with type 2 diabetes also found that dapagliflozin reduced the risk of AF/AFL events during follow-up by 19%, as well as the number of AF/AFL events overall. These reductions were consistent across major subgroups, including sex, presence of atherosclerotic cardiovascular disease, history of AF/AFL, history of HF, history of ischemic stroke, HbA1c, body mass index, blood pressure, or eGFR, which are well-known to be associated with the risk of AF/AFL [10]. Although many studies have been conducted, it is still unclear how SGLT-2i affects AF. Considering that there is close relationship between cardiovascular impairment and renal dysfunction which worsen each other [15], the mechanism of SGLT-2i on AF may be that it mediates natriuresis and glucosuria, thus lowering cardiac preload and reducing pulmonary congestion and systemic edema, allowing for proportionally greater reductions of interstitial volume, as compared to intravascular volume [16]. This reduces sympathetic nervous system overdrive [17], oxidative stress, inflammation [18] and vascular stiffness [19]. The cardio-renal protective effect of SGLT-2i was confirmed in many recent clinical trials. Cardiovascular outcome trials with SGLT-2i in type 2 diabetes showed unprecedented outcomes in the prevention of worsening heart failure, renal disease progression and mortality, further proved by randomized controlled trials in patients with heart failure and chronic kidney disease, with or without diabetes [20].

TZDs, the insulin-sensitizing agents, are known to have a strong protective effect on atherosclerosis-driven events such as cardiac or cerebrovascular disease [21]. Zhang et al. [6] conducted a meta-analysis of 130,854 diabetic patients in seven studies to evaluate TZD’s effect on AF development. This elucidated pioglitazone’s beneficial effect on reducing the risk of new-onset or recurrent AF by 30% [6]. Although pioglitazone might worsen HF and peripheral edema by inducing sodium-water retention, it positively modulates numerous cardiovascular functions (reduced inflammation of plaque, improved left ventricular systolic-diastolic function, improved arterial stiffness) and risk factors (blood pressure, blood lipid, adipose tissue physiology), and it shows meaningful reduction of major adverse cardiovascular events [22] and AF incidences [6].

In the present study, we compared the effect of SGLT-2i and TZD on new-onset AF, and there was no difference between the two groups, possibly because of the beneficial effect of both drugs. These results are consistent in patients, regardless of the presence or absence of HF at baseline. Of note, SGLT-2i-treated patients tend to have a lower risk of new-onset AF concurrent with hHF, although this does not reach statistical significance (HR 0.647, 95% CI, 0.303–1.381). The presence of either AF or HF increases the risk of developing the other [14]. AF results in atrial dilatation and left atrial volume overload that can promote HF, and HF can facilitate atrial remodeling, which makes for the development of AF [14]. Considering that SGLT-2i reduces the risk of hHF, SGLT-2i may help in reducing the incidence of AF and HF in those with severe HF that requires hospitalization. We saw no statistically significant difference in this study, possibly because the incidence of hHF was too small. Further studies with longer-term follow-ups are needed to verify this observation.

The current study found that SGLT-2i-treated group tended to have a lesser risk of AF in those with lower fasting glucose level, although statistical significance was not achieved. These findings may be explained by previous studies suggesting that glucose-lowering drugs may increase the risk of AF at FBS less than 130 mg/dL [23]. Previously, Lee et al. [24], reported that patients taking TZD have a higher risk of hypoglycemia than those without, and the patients taking TZD were more likely to have been prescribed three or more classes of antidiabetic drugs simultaneously in Korea. Thus, it is possible that glucose-lowering drugs that induce hypoglycemia were concomitantly used in the TZD group.

There are several limitations that should be considered in this study. First, this retrospective observational study using the claims databases has limitations. There were unmeasured confounding factors, such as HbA1c levels or duration of diabetes. Considering a recent meta-analysis that showed a 10% increased risk of AF per 20 mg/dL increase in blood glucose [25], potential effect of differential hyperglycemia between groups could not be excluded. Additionally, we could not evaluate specific clinical information such as HFpEF (heart failure with preserved ejection fraction) or HFrEF (heart failure with reduced ejection fraction), which are not included in the NHIS database. Second, discontinuation of study drugs or switching to other classes of drug could not be considered due to lack of data, when performing as-treated analysis. This might dilute the benefit of study drugs. Third, SGLT-2i was introduced in Korea in 2014, and its prescription rate was low. Therefore, the NHIS database had included fewer at-risk patients at the end of the follow-up period. Therefore, this short follow-up period is also a limitation. Forth, in Korea, SGLT-2i has been mainly used for younger obese patients with normal renal function [26]. Because we used a propensity score matched analysis to balance baseline characteristics between groups, we were unable to thoroughly examine the effects of drugs in patients at high risk for AF with renal dysfunction. Also, unfortunately, we couldn’t include elderly patients over 85 years of age in the study since the Korean Ministry of Food and Drug Safety issued a label warning to limit the use of SGLT-2i in this age group due to insufficient data on the volume depletion of SGLT-2i when it was introduced in 2014. Fifth, detection rate of AF could be lower than real AF incidence. To overcome this fundamental issue of AF detection, we enrolled relatively large number of patients and evaluated in a country where standardized in treatment and evaluation guided by government insurance system. Lastly, we only compared two drugs, TZD and SGLT-2i. Notably, population-based studies showed protective effect of metformin [27] and dipeptidyl peptidase 4 inhibitors [28] and harmful effect of sulfonylurea [29] in terms of new-onset AF. Regarding the effect of metformin on incident AF, we could not analyze it since more than 80% of patients with diabetes were using it as the first-line treatment in Korea [30].

If we had performed an analysis including other antidiabetic drugs, the effects would have been more obvious. Future studies with prospective design, longer-term follow-ups, and multiple comparators are needed to confirm the results of this study. Despite these limitations, to our knowledge, this is the first study to directly compare the effect of SGLT-2i and TZD in terms of new-onset AF outcomes using a large nationwide database.

In this study, the risk of new-onset AF was comparable in patients with type 2 diabetes treated with SGLT-2i and TZD in the national cohort. SGLT-2i or TZD would be a reasonable choice for patients with type 2 diabetes at risk of developing AF compared to other glucose-lowering drugs with a risk of hypoglycemia.

YRK, KAK designed the research. HS, YRK, SEL, KAK performed the research and HS, SEL wrote manuscript. HN, HK, DSK analyzed the data. All authors contributed to editorial changes in the manuscript. All authors read and approved the final manuscript.

This study used the National Health Insurance Service (NHIS) database, established by the Korean NHIS. In order to access the NHIS database, an application form for a research proposal must be submitted to the NHIS Korean Committee of Research Support. After a review and approval, data are made available in a deidentified format. This study was also approved by the Institutional Review Board of Dongguk University Ilsan Hospital (No.: 2019-02-002-002). Written informed consent was waived by the Institutional Review Board. All methods in this study were performed in accordance with the Declaration of Helsinki.

Data used in this study was provided by the Korea NHIS. We are grateful to Korea NHIS for cooperation.

This research received no external funding.

The authors declare no conflict of interest. Hanmi Pharmaceutical Company provided support in the form of salaries for the other authors (H Nam, H Kim, DS Kyoung), but did not have any additional role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript.