Background
Inhibitors of sodium glucose cotransporter-2 (SGLT2i) lower the renal threshold for glucose resorption in the proximal renal tubule, thereby causing glycosuria. In patients with type 2 diabetes (T2D), SGLT2i are effective in controlling glycemia, blood pressure, and body weight [
1]. According to the results of large cardiovascular outcome trials (CVOTs), SGLT2i prevent hospitalization for heart failure (HHF) in patients with T2D with or without a prior history of HF or cardiovascular disease (CVD) at baseline [
2]. In two trials performed on patients with HF and reduced ejection fraction (HFrEF), 42–50% of whom had T2D, SGLT2i significantly improved HF outcomes [
3,
4].
Interestingly, atrial fibrillation (AF) has been shown to occur less frequently among patients who received dapagliflozin than among those who received placebo in the DECLARE trial (HR 0.81; 95% CI 0.68–0.95) [
5]. Other studies have reported similar lower rates of AF among patients randomized to SGLT2i, and two meta-analysis calculated a 21% relative risk reduction [
6,
7]. This finding is relevant because T2D is an established risk factor for AF [
8,
9], which can cause embolic stroke, precipitate HF [
10], or result in hospitalization for the need of rate control. Therefore, though mechanisms of this association are unclear [
11], preventing AF in people with diabetes can exert positive effects on global outcomes.
To date, whether such benefit observed in trials applies to clinical practice is unknown. The real-world setting differs from the experimental trial setting in many instances, including the way outcomes, like AF, are screened, ascertained, adjudicated, and reported. Thus, it is important that, in the absence of dedicated trials, potential benefits resulting from post-hoc trial analyses are confirmed in clinical practice. So far, limited real-world studies provided inconsistent results on the association between use of SGLT2i and rates of AF [
12‐
14]. Spontaneous reports of adverse events (AE) populate large databases with clinically-relevant information emerging from clinical practice. The analysis of pharmacovigilance databases can inform on the associations between drugs and health outcomes defined by AE. Though pharmacovigilance is traditionally used to detect signals of potential harm, more recently, spontaneous reporting systems have been exploited to uncover patterns indicative of reduced reporting (positive effects), e.g. for drug repurposing [
15,
16]. Furthermore, the consistency between trial results and pharmacovigilance analyses has been recently confirmed, thus raising the debate on the role of spontaneous reports as potential source of risk estimates under certain circumstances [
17]. Therefore, AE reporting databases can be re-used as a source of real-world evidence that complement available information from trials and traditional cohort studies. This approach has been successfully used to explore outcomes associated with SGLT2i and other diabetes therapies, not limited to the detection of rare AE [
18‐
21].
Herein, we examined a large pharmacovigilance database to evaluate the reporting frequency of SGLT2i with AF as compared to other glucose-lowering medications.
Discussion
We show that, in one of the world largest pharmacovigilance databases, AF was reported disproportionally less frequently among patients using SGLT2i than among patients using other glucose-lowering medications. This result was highly consistent in several sensitivity analyses performed to reduce the possibility of bias, including the use of negative and positive controls.
Along with the results of individual randomized controlled trials (RCTs) and meta-analyses [
6,
7], this finding supports the suggested protective effect of SGLT2i against AF. To date, observational studies have provided conflicting results. The CVD-Real Nordic, a retrospective observational study (n = 40,908 patients) performed in Denmark, Norway and Sweden, reported no difference in the incidence of AF among patients who received SGLT2i versus matched patients who received DPP-4 inhibitors [
13] or other glucose-lowering medications [
12]. On the other side, a study from Taiwan reported markedly lower rates of AF among 15,606 new users of SGLT2i versus 12,383 new users of DPP-4 inhibitors, after inverse probability of treatment weighting [
14]. In view of these conflicting results, more real-world data are needed to verify whether the protection exerted by SGLT2i against AF observed in trials could apply to clinical practice. Our new analysis adds further evidence because spontaneous AE reporting databases can be used to explore health outcome in the real-world. We contribute to extend the cumulative knowledge about the safety and effectiveness of SGLT2i in an unselected population using a global pharmacovigilance database, thus supporting generalizability of trial findings.
Though generally considered a relatively benign arrhythmia, new-onset AF is burdened by significant morbidity and mortality, driving markedly elevated risk for stroke, dementia, HF, and overall mortality [
29‐
31]. Due to its high healthcare and societal costs, preventing AF has become a major public health priority [
32]. Thus, prevention of AF by SGLT2i can yield substantial benefits on patients’ outcomes and quality of life. One could even argue that the effect against AF drives part of the extraordinary cardiorenal benefits observed during therapy with SGLT2i in patients with and without diabetes [
2‐
4,
33].
The mechanisms driving occurrence of AF are still debated, but include atrial fibrosis and electrical remodelling, with or without the coexistence of triggering ectopic foci [
34]. There are several potential mechanisms whereby SGLT2i may reduce the incidence of AF [
35]. SGLT2i can prevent glucotoxicity in the heart, switching cardiac metabolism to metabolically favourable substrates. Reduction of pre-load, decongestion and reduction of filling pressures, achieved by the peculiar diuretic effect of SGLT2i, would blunt the stress imposed to atrial cardiomyocytes. Also, thinning of epicardial adipose tissue, resulting from loss of body weight and fat mass, may spare atria from pro-inflammatory signals driving remodelling and fibrosis [
36]. Finally, SGLT2i can exert direct electrophysiological effects by modulating sodium handling and mitochondrial function [
37,
38], which may counteract AF triggers.
We wish to underline some important characteristics of pharmacovigilance studies and their intrinsic limitations. First, these studies differ from observational cohort research mainly because the population is composed of patients for whom at least one AE was filed for at least one drug, thereby missing information on the background population of individuals exposed to the drug(s) but not reporting any AE. If no true association exists between a given drug and an AE, such AE should have the same frequency in reports listing such drug as in other reports, yielding no disproportionality. Indeed, pharmacovigilance data cannot be used to derive absolute incidence of an event, but only to compare reporting of an event in association with different drugs. Another important limitation is that clinical-level data (such as HbA1c, BMI, kidney function) are not available in pharmacovigilance databases, preventing full assessment of comparability of patient characteristics within the various AE groups. For these reasons, after having shown a disproportionally lower rate of AF among reports for SGLT2i, we undertook a series of sensitivity analyses and internal controls to verify robustness of the finding. A positive control for AF (ibrutinib), as well as negative and positive control AE for SGLT2i confirmed internal validity of the method. Excluding reports for anti-arrhythmic drugs re-assured on competition with SGLT2i for AF, whereas excluding AE reporting renal disease as a drug indication reasonably ruled out that AF occurred less frequently in reports for SGLT2i because chronic kidney disease, a risk factor for AF, was long considered a contraindication to SGLT2i. Exclusion of reports for insulin allowed to rule out that the control group of drugs were being prescribed to sicker patients, driving a spurious association. Excluding patients with an indication for CVD ruled out that the finding was driven by patients who were already affected by cardiac problems, who are more likely to receive SGLT2i. Considering that only part of AE reports filed for glucose-lowering medications specify the diabetes indication and because some diabetes drugs could be used by non-diabetic individuals, all analyses were repeated excluding reports without the diabetes indication. In all cases, AF was reported disproportionally less frequently in association with SGLT2i than with control drugs. Pharmacovigilance traditionally aims to discover signals of potentially new drug-associated AEs, while its ability to reliably inform on inverse associations indicative of risk reduction is still debated. In the case of SGLT2i, their consolidated use in clinical practice and largely characterized safety profile make spontaneous reporting a likely indicator of risk in clinical practice, provided that major reporting biases are reasonably excluded with available methods. Dilution by competing AEs is of concern. In other term, the rate of AF among reports for SGLT2i might be diluted by common and well-recognized SGLT2i-associated AE, increasing denominator of the PRR. To address this issue, we repeated the analysis by excluding common AE associated with SGLT2i and AE that are less common but highly specific for SGLT2i. Even with such exclusion, on top of insulin exclusion and for both reports with and without the diabetes indication, disproportionality remained robust and significant. Remarkably, even after combining all the above-mentioned filters and exclusions together, AF rates remained 25% significantly lower among reports for SGLT2i than for other diabetes drugs.
Among other limitations, we wish to acknowledge that duration of treatment (time elapsed from drug initiation to AE occurrence) is described in a minority of reports in the FAERS. In addition, there is no mean of assessing how AF was diagnosed, e.g. based on symptoms or instrumental screening.
In face of these limitations, to strengthen the possible causal connection between use of SGLT2i and the lower rates of AF, we globally assess the evidence using Bradford Hill criteria (Table
1): with the exception of some items that we could not assess (namely biological gradient and reversibility), all criteria appear to be satisfied thereby supporting, though not proving, a cause-effect relationship.
Table 1
Bradford Hill criteria for causation
Strength | Moderate | The rate of AF among reports for SGLT2i was about halved compared to those for control drugs, or is about 2× for control drugs versus SGLT2i |
Consistency | High | Disproportionality changed little and remained significant in all sensitivity analyses. Therefore, the impact of unmeasurable confounders is likely to be negligible |
Specificity | Good | Positive and negative controls were satisfied |
Temporality and reversibility | Not assessed | Time from drug initiation to AE occurrence, as well as information on drug withdrawal and re-challenge was not available |
Biological gradient | Not assessed | We had no data on the dosage of SGLT2i. However, the gradient of SGLT2i dose is limited (max 2.5-fold) |
Plausibility | Satisfied | There are several potential mechanisms whereby SGLT2i may reduce the risk of AF |
Coherence | Good | Lower rates of AF among patients taking SGLT2i have been reported in randomized control trials, and in one of two observational studies |
Experimental support | Satisfied | Randomized controlled trials provide the most compelling experimental evidence for protection against AF by SGLT2i. Yet, it should be noted that none of such trials had AF as primary endpoint |
Analogy | Good | The FAERS has been used previously to demonstrate associations between AF rates and other specific drugs (e.g. ibrutinib) |
Publisher’s note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.