Heart rate variability and atrial fibrillation in the general population: a longitudinal and Mendelian randomization study

Atrial fibrillation (AF), the most common cardiac arrhythmia, is associated with substantial morbidity and mortality and represents a significant burden on healthcare [1‐4]. The exact AF pathogenesis remains to be identified. It has recently been suggested that cardiac autonomic imbalance could play a role in AF pathophysiology by promoting a decline in cardiac function [5‐10].

Heart rate variability (HRV) is considered a non-invasive, accessible measure that may reflect the complex interaction between the autonomic nervous system and the heart [11, 12]. A complex relationship between HRV and AF has been suggested [5‐10, 13]. Specifically, lower and higher levels of HRV may lead to decline in cardiac function and subsequently give rise to AF [5‐10, 13]. Recent evidence shows that sex differences with regard to AF burden, pathophysiology, and prognosis exist [14]. However, the previous observational studies have been limited to either a cross-sectional design or a single measurement of HRV and did not evaluate sex differences. In addition, observational studies are prone to residual confounding and reverse causality [15].

Genome-wide association studies (GWAS) have identified genetic variants/single nucleotide polymorphisms (SNPs) for multiple assessments of HRV [16] and AF [17, 18]. Pathway and tissue enrichment analyses suggest that HRV SNPs are preferentially expressed within the sinoatrial node [16]. Moreover, AF SNPs have been suggested to affect the cardiac ion channels, cardiac calcium signaling, and the heart and skeletal muscles [17, 18]. This suggests that there may be a genetic foundation underlying the association between HRV and AF.

We aimed to investigate the association between longitudinal measures of HRV and heart rate with the risk of new-onset AF in the general population. Additionally, we used a comprehensive Mendelian randomization (MR) analysis using summary-level data from GWAS on measures of HRV and AF to investigate the potential causal relationship between HRV and AF.

Our study shed light on the complex interaction between HRV and AF. Our joint model analyses showed that longitudinal measures of SDNN, and RMSSD were significantly associated with new-onset AF in the general population while SDNNc, RMSSDc, and heart rate were not significantly associated. Sex-stratified analyses showed that RMSSD among men, and SDNN, RMSSD, and heart rate among women were significantly associated with new-onset AF. MR analyses supported the causal association between SDNN, and RMSSD with AF. Our findings indicate that treatment to modulate HRV might prevent AF in the general population, in particular in women.

The exact mechanism that underlies the relationship between HRV and AF remains incompletely understood. Shared underlying risk factors, such as obesity, diabetes mellitus, and coronary heart disease, could influence HRV and are also implicated in AF pathophysiology [1, 5, 12, 38, 39]. In our study, however, the associations of HRV with incident AF slightly attenuated, but remained significant after extensive adjustment for shared cardiovascular risk factors. The increase in left atrial size that has been associated with HRV could suggest a role for HRV in AF pathogenesis that is mediated by the left atrium [13]. Moreover, autonomic imbalance could trigger an inflammatory response that can subsequently lead to AF [12]. Finally, the effect of the GWAS-identified HRV SNPs on the genes (especially, GNG11, RGS6) that are preferentially expressed within the sinoatrial node underlines the genetic basis that potentially underlies the association between HRV and AF. In short, these genes may affect acetylcholine release of the vagal nerves within the sinoatrial node and thereby influence HRV [16]. More specifically, GNG11 codes for the γ11 subunit of the heterotrimeric G-protein complex Gαβγ and may cause a decreased expression of this subunit [16]. This lower availability of this subunit may then reduce Gβγ induced GIRK activation. This potentially blunts heart rate changes caused by oscillatory changes in cardiac vagal activity, ultimately decreasing HRV [16]. Furthermore, RGS6 regulates the heterotrimeric G-protein complex signaling type 6 and may increase its availability. This leads to a decreased GIRK activation and potentially blunts the effects in cardiac vagal activation, and may thereby decrease HRV [16]. Subsequently, it has been suggested that sinus node disease (SND) may cause AF by promoting atrial extrasystoles, and re-entry [40, 41]. Atrial extrasystoles may occur during the slow atrial cycle in the presence of SND. Atrial extrasystoles are mostly followed by a compensatory pause. The pause may then be prolonged which allows other atrial ectopic activity to arise which possibly triggers AF [40]. Early premature beats that originate from areas other than the sinus node may result in conduction block and initiate re-entry, which may be a mechanism underlying AF [40]. Furthermore, stenosis in the sinus nodal artery is also common in patients with AF which implies that ischemic damage to the sinus node alone without atrial fibrosis, stretch or muscle loss may result in AF [40]. Overall, a combination of atrial extrasystoles, re-entry, and ischemia to the sinus node are mechanisms by which SND may cause and promote AF.

We investigated the longitudinal measures of HRV during a long follow-up time in relation to new-onset AF. Taking into account repeated measurements of HRV in relation to new-onset AF may provide more insight and prognostic information over a single baseline measurement that has been done by most of the previous studies [5‐10, 13]. Longitudinal measures of HRV during follow-up were associated with an increased risk of incident AF, especially among women. These findings extend previous evidence by simultaneously evaluating the repeated measurements of uncorrected and corrected HRV, heart rate, and sex differences while investigating the link between HRV and AF [5‐10, 13]. To some extent our findings support the association between heart rate and AF that has been previously reported in observational, [6, 7, 42] and Mendelian randomization studies [43]. However, we only found a significant association for heart rate in association with AF among women. One potential explanation could be differences in sex hormones. It has been demonstrated that an acute ovarian hormone withdrawal induced by oophorectomy leads to decline in different measures of HRV (SDNN, RMSSD), and an increase in heart rate in women [44]. The same study also showed that estrogen replacement therapy for three months within the oophorectomized women restored the HRV and heart rate to a pre-surgery level [44]. This might explain why uncorrected HRV and heart rate were only associated with incident AF in women, and not in men, in our study. We further hypothesize that competing risk of death is a possible explanation for the observed sex differences. AF is strongly associated with age, [1‐3] so it is likely that men die of other (cardiovascular) diseases before development of AF. This hypothesis was supported by our competing risk analyses which showed that SDNN, RMSSD, RMSSDc, and heart rate were significantly associated with mortality, especially among men. Nevertheless, we found a higher incidence of AF in men, than in women in our study.

Our MR approach sheds light on the causality of the association between HRV and AF. Our effect estimates were more or less in line with previous observational studies. However, we were unable to assess the association between SDNNc, RMSSDc and AF, since not enough instrumental variables for SDNNc and RMSSDc were available to be used for the MR analyses. Future GWAS with a larger sample size could identify new additional genetic variants that could be used to assess the association between heart rate corrected HRV and AF. This could be of importance, because of the strong inverse association that exists between HRV and heart rate [24, 25]. This relation is further underlined by Nolte et al. who showed attenuation in the HRV SNP associations when they corrected for heart rate [16]. This might imply that uncorrected measures of HRV may be, in part, confounded by heart rate. Although, we showed that heart rate itself was not significantly associated with new-onset AF (except in women), we cannot rule out the possibility that heart rate is the overall determining factor instead of HRV after all. Since our uncorrected measures of HRV were indeed not significantly associated with new-onset AF. Further, as heart rate is also associated with AF and cardiovascular mortality proper adjustment for heart rate is of importance [26, 43, 45, 46]. However, excluding a genetic variant that was also associated with heart rate, a potential confounder or horizontal mediator, did not substantially change our MR results. Future studies on HRV measures corrected for heart rate could further aid in elucidating the exact mechanisms underlying HRV and AF.

The major strengths of this study are its population-based nature, large sample size with detailed information on cardiovascular risk factors, meticulous adjudication of incident AF and long follow-up time, multiple sensitivity analyses including complete-case analyses, excluding prevalent and incident CHD prior to AF diagnosis, use of competing risk analyses to compute cause-specific hazards, and use of large-scale GWAS summary statistics. The availability of repeated measurements for different HRV measures during follow-up also enabled us to investigate longitudinal measures of HRV in association with new-onset AF in a joint modeling approach which may provide more insight and give more prognostic information over a single baseline measurement. Moreover, using a MR approach we were able to gain more insight in the complex interaction between HRV and AF and to avoid certain biases that are more common in traditional observational epidemiological studies, such as residual confounding and reverse causation [15]. However, our study also has some limitations that should be taken into consideration. Our HRV measures were based on 10-s ECGs, although HRV guidelines recommend that HRV measures are based on preferably 5-min or 24-h ECG recordings [22]. Nevertheless, 10-s ECGs are more commonly performed in healthcare, are cheaper, are faster, and thereby more patient friendly than longer ECG recordings. Additionally, HRV measures from 10-s ECGs have already been associated with left ventricular function, [47] heart failure, [47, 48] cardiac-[49] and all-cause mortality [50]. Additionally, other studies that investigated the reliability of 10-s ECGs in comparison to 5-min ECGs to assess HRV showed that 10-s ECGs are also a reliable tool for HRV risk assessment, in particular within population-based studies [51, 52]. We could not distinguish between paroxysmal, persistent, long-term persistent, and permanent AF as Holter monitoring has not been done in this large population-based cohort. In our MR analyses, we cannot rule out unobserved horizontal pleiotropy, although we tried to address horizontal pleiotropy using multiple MR sensitivity analyses, such as MR-Egger, WME, MR-PRESSO, and sensitivity plots, to identify and correct for horizontal pleiotropy. Additionally, not enough sex-stratified SNPs were available in the publically available genetic dataset to perform the MR for men and women separately. Furthermore, there was partial overlap in the samples that were used to obtain the genetic instruments which may cause bias toward observational findings [33]. However, the potential bias was probably negligible given that the maximum potential overlap was 2.1%. Finally, our findings may not be generalizable to younger populations and other ethnicities, as our analysis included mainly older participants from European descent.

In conclusion, longitudinal measures of SDNN, RMSSD, but not SDNNc, RMSSDc, and heart rate, were significantly associated with new-onset AF. In sex-stratified analyses, RMSSD among men and SDNN, RMSSD, and heart rate among women were significantly associated with new-onset AF. MR analysis confirmed the complex association between HRV and AF that has been indicated by our and previous observational studies. These findings indicate that measures to modulate HRV might prevent AF in the general population, especially among women, but future MR studies that investigate the causality between heart rate corrected measures of HRV and AF are warranted.