Insight on the relationship between heart rate variability parameters and the risk of stroke among non-valvular paroxysmal atrial fibrillation patients

Abstract

Background

Identifying the risk of stroke in patients with atrial fibrillation (AF) presents a challenge, as existing tools such as the CHA2DS2-VASc score primarily emphasize medical history, affording relatively less attention to the electrocardiogram (ECG). Exploring HRV (heart rate-related variability) parameters could help refine risk prediction for thromboembolic ischemic stroke and enhance personalized management. This paper investigates how HRV parameters may inform the risk of acute ischemic stroke in atrial fibrillation patients.

Methods

Data were retrospectively collected from Zhongnan Hospital’s electronic registry for patients with atrial fibrillation admitted in 2022. The sample was sorted into cases and controls based on specific inclusion and criteria. The outcome was the onset of acute ischemic stroke (AIS) in recurrent paroxysmal non-valvular atrial fibrillation. Cases and control data were analyzed using different statistical methods based on distribution, and the association between HRV parameters and the risk of stroke in paroxysmal atrial fibrillation was assessed. Binary logistic regression was utilized to assess a predictive model for stroke risk based on heart rate variability (HRV) parameters.

Results

A total of 3218 medical records with atrial fibrillation were retrieved. Out of 3218 medical records with atrial fibrillation, 192 were selected after screening, including 93 cases and 99 controls. The cohort comprised 93 females (50 cases) and 99 males (43 cases). Mean ages were 68.67 ± 1.95 for cases and 68.35 ± 1.19 for controls, with no significant difference (P > 0.05). Hypertension history was significantly more common in cases (P = 0.006), indicating a dependency between hypertension and acute ischemic stroke in non-valvular paroxysmal AF. SDNN mean rank was significantly lower in cases (P = 0.024), and LF was significantly reduced (P = 0.044). Other HRV parameters showed lower values in cases, but differences were not statistically significant. We found that AVTD (abnormal variation in the time domain) alone was significantly associated with acute ischemic stroke in non-valvular paroxysmal AF. While AVFD (abnormal variation in frequency domain) and AVA (abnormal variation in arrhythmogenicity) were not individually significant, a combined variable (CDV) from AVTD, AVFD, and AVA was also statistically significant (P = 0.029). In this model, SDNN, AVTD, history of hypertension, as well as the CHA2DS2-VASc score and the use of novel anticoagulant medicines (NOACS), were significant predictors in both univariate and multivariate analyses. At the same time, CDV was significant only in univariate analysis. These findings may suggest that HRV parameters may provide clues to the risk of acute ischemic stroke in non-valvular paroxysmal AF.

Conclusion

This study found that abnormal variation in HRV parameters was more observed in patients with acute ischemic stroke. Then, unlike previous research, this study uniquely integrated multiple HRV parameters to develop novel indices such as AVTD, AVFD, AVA, and CDV, which somehow demonstrated a significant association with the risk of AIS. These new metrics, combined with conventional risk factors and CHA2DS2-VASc score, brought great improvement to the model prediction for AIS risk.

Introduction

Atrial fibrillation (AF) is the most common sustained arrhythmia encountered in clinical practice and is a significant risk factor for stroke, particularly ischemic stroke [1, 2]. The prevalence of AF increases with age and is associated with a substantial burden on public health due to its complications and associated risk of stroke [3]. Defining high-risk stroke patients from individuals with AF remains a critical challenge, as conventional risk stratification tools such as the CHA2DS2-VASc score do not always provide sufficient granularity in predicting individual risk. Heart rate variability (HRV), a measure of the variation in time intervals between consecutive heartbeats, has emerged as a potential biomarker for assessing autonomic dysfunction and overall cardiovascular health [8]. Reduced HRV has been associated with an increased risk of cardiovascular events, including stroke, in various populations [2, 4]. In previous research, the link between HRV and cardiovascular outcomes has been demonstrated to be higher in patients with type 2 diabetes. Diabetes is a well-known CVD (cardiovascular disease) risk factor associated with increased odds of onset of stroke. Another factor associated with a higher risk for stroke is the presence of atrial fibrillation. Current management strategies in atrial fibrillation include anticoagulation, which aims to reduce the onset risk for events, amidst which acute ischemic stroke. Since ischemic stroke remains a more significant burden in China, according to a recent 2020 report, stroke's estimated prevalence, incidence, and mortality rate were 2.6%, 505.2 per 100,000 person-years, and 343.4 per 100,000 person-years. This indicated the need to improve the related prevention strategy for the Chinese population. Atrial fibrillation (AF) raises a person's risk of stroke by 4–6 times on average. The risk increases with age, and AF is the direct cause of one in every four strokes in adults over the age of 80 years.

Many studies have shown that specific HRV metrics, such as the standard deviation of NN intervals (SDNN), the root mean square of successive differences (RMSSD), and low-frequency/high-frequency ratio (LF/HF), exhibit significant associations with the risk of stroke in diverse settings [4,5,6]. These parameters may provide insights into the underlying pathophysiological processes predisposing AF patients to stroke, potentially offering a complementary approach to existing predictive models. Understanding the relationship between HRV parameters and the risk of stroke in AF patients could refine risk prediction strategies and enable more personalized management approaches. By integrating HRV data with clinical assessments, healthcare providers may improve their ability to identify high-risk individuals and tailor interventions to reduce stroke incidence among this vulnerable population [7, 8].

This paper assesses how HRV parameters could inform of the risk of acute ischemic stroke in people with atrial fibrillation.

Materials and methods

Design

This investigation followed the retrospective case controls clinical research design principles. The data came from the Neurology department of Zhongnan Hospital’s electronic registry. Data collection was from medical records of patients admitted to the hospital from January 2022 to December 2022.

Target population

This study's sample population included patients admitted to the hospital whose admission ECG records showed an atrial fibrillation pattern and who met the specific inclusion criteria to be considered cases or controls. The grouping into case or control groups was based on evidence of acute ischemic stroke diagnosis, based on criteria established by the latest guidelines provided by the Chinese Stroke Association.

The development of acute ischemic stroke in a recurrent paroxysmal atrial fibrillation setting as evidenced by the presence of focal neurological symptoms such as difficulty with speech and weakness in one half of the body, following rapid assessment of neurological deficits using scales like the National Institutes of Health Stroke Scale (NIHSS), added to suggestive ischemic brain lesion on non-contrast CT scans and the acute nature confirmed by diffusion-weighted magnetic resonance imaging (DWI).

To be included in the study, all the following conditions were mandatory:

• Admission ECG with obvious atrial fibrillation pattern.

• Availability of data from Holter ECG following admission.

• Patients with a history of recurrent non-valvular paroxysmal atrial fibrillation (NVP AF). Atrial paroxysmal AF is an irregular heartbeat occurring intermittently, typically resolving within 7 days or less. In contrast to persistent atrial fibrillation, which remains constant for durations extending beyond one week and may last up to less than a year, episodes of paroxysmal atrial fibrillation can initiate abruptly and resolve spontaneously.

• Patients with a confirmed acute ischemic stroke diagnosis on non-contrast brain CT and DWI.

• Patients who met the first three criteria and had a normal brain CT were recruited as controls: these patients were admitted for other reasons, among which dizziness, headaches, trans myelitis, cervical spine compression, and cognitive function disorders.

Paroxysmal AF was identified through documented medical history, prior ECGs/Holter reports indicating self-terminating episodes (< 7 days), and physician notes confirming recurrent, intermittent AF. Patients with persistent AF (episodes > 7 days) or long-standing AF (> 1 year) were excluded using these criteria.

The exclusion criteria were as follows:

• Incomplete reports.

• Presence of other severe arrhythmia (VT, SVT).

• Recent diagnosis of atrial fibrillation less than 4 years.

• Patient with less than 1-year history of systemic embolism (excluded cerebral embolism), DVT, or other coagulopathies.

Outcome

To evaluate abnormal variation in HRV-derived parameters as risk factors and predictors of acute ischemic stroke in patients with non-valvular paroxysmal atrial fibrillation (AF).

Data collection and statistical analyses

Data were extracted to Excel, filtered, and sorted into different cases and controls based on the abovementioned criteria. Data collected were age, gender, hypertension history (coded as positive if indicated in the patient admission medical and treatment history taking), diabetes mellitus history (coded as positive if indicated in the patient admission medical and treatment history taking), BP measurements, BNP, LDL heart rate measurements, CHA2DS2-VASc score, the presence of novel oral anticoagulation therapy (NOAC) such as and heart rate variability parameters grouped in three categories: time domain SDNN (the standard deviation of NN intervals), SDANN (standard deviation of the average NN intervals), RMSSD (the root mean square of successive differences), SDNN index; the frequency domain high frequency (HF), low frequency (LF), LF/HF. Data were reported as mean ± standard deviation (X̄ ± SD) or median and quartiles for continuous variables and as frequency or count for categorical variables. Abnormal variation in the time domain (AVTD) was the categorical variable that corresponded to any two of SDNN < 102–180 ms, SDANN < 92–162 ms, SDNN index < 39–69 ms, and rMSSD < 15–39 ms. Abnormal variation in the frequency domain (AVFD) was the categorical variable defined by LF < (1 170 ± 416) ms² or HF < (975 ± 203) ms². Abnormal variation in arrhythmogenicity (AVA) was the variable that indicated higher arrhythmogenic activity based on 24-h PVC count > 100 or 24-h PAC count > 30. The simultaneous presence of AVTD, AVFD, and AVA was coded as the compound degree of variation.

The analyses conducted include case versus control comparison using independent samples t-Student parametric test and independent two samples Mann–Whitney non-parametric analyses depending on whether the variable was normally distributed. The Chi-square statistic was used for categorical variables to determine the association of the various categorical variables derived from HRV parameters with the risk of acute ischemic stroke in a paroxysmal atrial fibrillation setting. Finally, binary logistic regression analyses were performed to determine whether HRV parameters or its derived categorical variables could also be predictors of the risk of acute ischemic stroke among recurrent non-valvular paroxysmal atrial fibrillation patients. The variables in the model were tested for collinearity via the variance inflation estimation approach.

All the analyses were performed with SPSS 27 IBM Corp software. The P value statistics obtained were two-tailed, and the level of significance was set to 0.05.

Results

Cases selection and cases vs controls comparison

A total of 3218 medical records with atrial fibrillation were retrieved. After a thorough screening based on the inclusion and exclusion criteria, 192 reports were selected, including 93 that fulfilled the criteria for cases and 99 that were eligible as controls. The detailed screening process is illustrated below (Fig. 1).

Fig. 1

Case selection process

The selected cohort consisted of 93 females, 50 of whom were cases, and 99 males, 42 of whom were cases. The mean ages were 68.67 ± 1.95 in cases and 68.35 ± 1.19 for the control group, but the difference did not reach statistical significance (P > 0.05).

The positive history of hypertension was more pronounced in cases than controls; Chi-square statistics yielded a significant result (P = 0.006), which implied that history of hypertension and the onset of acute ischemic stroke are dependent variables in the setting of non-valvular paroxysmal AF. The difference of mean in CHA2DS2-VASc score was also statistically significant, with it being higher among cases than controls. Moreover, the frequency of use of NOAC was lower among cases than controls. These findings altogether demonstrate that our data did not deviate that much from reality facts.

The difference in SDNN mean rank between cases and controls was also statistically significant P = 0.024, denoting a decreased trend of this HRV parameter time domain in the non-valvular paroxysmal AF setting (see Table 1).

Table 1 Cases versus control comparisons

The mean of LF was also lower in cases than in controls, and the difference reached statistical significance (P = 0.044). Other HRV parameter values also seemed lower in cases than controls, but the non-parametric analyses did not reach a clear significance.

Test of dependence between heart rate variability parameters and risk of stroke

We used the categorical variables derived from the HRV parameters to evaluate and assess the nature of the association for each category, namely the time domain, frequency domain, and arrhythmogenic domain. This analysis was noticeable for a significant association between AVTD alone and the onset of acute ischemic stroke in non-valvular paroxysmal AF. Although AVFD and AVA depicted an independent relationship with the onset of acute ischemic stroke (AIS), it could be observed that a compound degree of variation, which was a variable defined by the simultaneous presence of AVTD, AVFD, and AVA, also yielded a statistically significant association with the onset of AIS in the non-valvular paroxysmal AF setting (Table 2).

Table 2 Statistical relationship between types of HRV variation and risk of acute ischemic stroke

Regression analyses (binary logistics)

Multivariate univariate regression analyses were used to estimate the impact of several variables (such as age, gender, DM history, HTN history, SDNN, SDANN, LF, HF, AVTD, AVFD, AVA, CDV, LDL, BNP), CHA2DS2-VASc and intake of NOAC score on the dichotomous dependent outcome, which was the development or onset of acute ischemic stroke as a dichotomous variable, the values of which were coded as case or control. The multivariate binary logistic regression carried out on SPSS demonstrated the fitness of our model. This was supported by the significance of the Omnibus tests of model coefficients (P value < 0.001). Moreover, there was a slight similarity between the observed and predicted variable values. However, the Nagelkerke R-square revealed that the predictor variables in the model were responsible for only a 54% change in the risk of AIS. The model's sensitivity was estimated to be 80.6%, and its specificity was estimated to be 79.8%. Overall, the model correctly classified 80.2% of the patients, an improvement from the null model that correctly classified 80.2%. The table below summarizes the contribution and significance of each variable included in the model (Table 3). This table shows that SDNN and hypertension history, both on uni and multivariate, were significant predictors in this model. At the same time, AVTD and CDV were only significant in univariate analysis. All these results point to the fact that the contribution of the HRV parameter may carry or give some clues to the risk of onset of acute ischemic stroke in NVP AF.

Table 3 Model prediction analyses

Discussion

This study investigated the association between heart rate variability (HRV) parameters and the risk of stroke in patients with paroxysmal non-valvular atrial fibrillation (AF). Our analysis revealed significant differences between the case and control groups in SDNN (standard deviation of normal-to-normal intervals) and LF (low-frequency component), indicating a potential link between these HRV metrics and the risk of stroke in non-valvular AF patients. The difference of mean in CHA2DS2-VASc score was also statistically significant, with it being higher among cases than controls. Moreover, the frequency of use of NOAC was lower among cases than controls. These findings altogether demonstrate that our data did not deviate that much from the ideal scenario. Additionally, the integration of HRV into categorical derived variables highlighted significant associations of AVTD (abnormal variation of time domain) and CDV (compound degree of variability) with stroke risk. In predictive models, SDNN and hypertension history, CHA2DS2-VASc score and AVTD were significant predictors in univariate and multivariate analyses, while CDV showed significance only in univariate analysis. These findings suggest that HRV parameters could serve as partial predictors of the risk of stroke in non-valvular AF patients.

Atrial fibrillation is a prevalent arrhythmia characterized by disorganized atrial electrical activity, which increases the risk of thromboembolic events, including stroke. This condition manifests when the heart's upper chambers, known as the atria, exhibit quivering or irregular contractions, which consequently impact the blood flow throughout the body. This arrhythmia significantly impairs quality of life and is associated with increased morbidity and mortality. Globally, the incidence and prevalence of AF are rising, with projections estimating that by 2050, Asia will have at least 72 million AF patients, and approximately 3 million of these will experience AF-related strokes [9]. The burden of AF, coupled with the high risk of stroke, underscores the critical need for practical risk assessment tools. Previous research has demonstrated that AF often coexists with complications such as heart failure and acute stroke, and AF is a significant risk factor for stroke, substantially increasing its likelihood. The prevalence of asymptomatic AF further complicates risk management, placing a considerable burden on healthcare systems.

Anticoagulation remains the primary treatment strategy for AF, aimed at reducing the risk of thromboembolic events, including stroke. Therefore, identifying tools that can predict the risk of stroke in AF patients is of significant value. Current predictive tools include the CHADS2 VASc score and various biomarkers, such as the neutrophil percentage to albumin ratio and indicators of cardiovascular disease and diabetes [10, 11]. Hypertension, in particular, is a crucial predictor that significantly increases the risk of stroke in AF patients [12], aligning with our study's findings. Despite extensive research on AF mechanisms, the risk of stroke varies among AF patients, highlighting the need for additional tools and indicators for accurate risk assessment and management. Untreated paroxysmal AF may progress to persistent AF, further increasing stroke risk. Evidence suggests that HRV may have predictive value for early-stage paroxysmal AF [13], although its application in identifying high-risk acute ischemic stroke populations still needs to be explored. This study addressed this gap by evaluating the association and predictive value of HRV abnormal variation for acute stroke in paroxysmal AF. These insights could enhance clinical management strategies and improve risk stratification for AF patients.

In this study, the parameters included in the prediction model were selected based on the existing evidence of a link between the parameter and the risk of acute ischemic stroke. As already established, in literature; diabetes mellitus, female gender, abnormal increase in BNP, high serum LDL cholesterol, advanced age, the intake of oral anticoagulation therapy and CHA2DS2-VASc score can inform on the risk of the onset of acute ischemic stroke among the susceptible population. We also added HRV-derived indices such as AVTD, AVFD, AVA, and CDV, designed to assess how the various categories, namely the time and frequency domains of HRV markers, affect or associate with the risk of acute ischemic stroke. Multicollinearity in regression analysis occurs when several predictor variables are closely related, resulting in a lack of unique or independent information in the regression model. When the correlation between variables is significantly high, it can create issues during the fitting and interpretation of the regression model. The variance inflation factor (VIF) metric was used to avoid multicollinearity in the model. When the VIF score was close to 1 and lesser than two, collinearity was unlikely, which was the case for most variables except for SDANN, SDNN, SDNN index, rMSSD, LF, HF, LF/HF, and heart rate, which yielded a VIF greater than 10. Once multicollinearity was detected in the model, some correlated variables were removed to address it. The model was then analyzed using univariate and multivariate binary logistic regression, a plausible statistical method to predict the probability that an observation falls into one of the two responses (acute ischemic stroke or not) based on one or more independent variables. This analysis has helped to clarify whether variations in HRV parameters would have a predictive value, whether alone or in combination with other known predictors whatsoever, on the risk of acute ischemic stroke.

The initial model's sensitivity was estimated to be 67.7%, and its specificity was estimated to be 68.7%. Overall, the model correctly classified 69% of the patients, an improvement from the null model that correctly classified only 50%. However, after introducing the CHA2DS2-VASc score and the NOAC, there was a significant improvement, and the model reached 80.6 in sensitivity, 79.8%, and this overall model classified up to 80.2% of cases. When taken in multivariate modality, the SDNN index performed better in predicting acute ischemic stroke than the other conventional ischemic stroke predictor. In the univariate modality where only one variable is selected while controlling for confounding, the proposed new HRV-derived parameter CDV and AVTD added to SDNN performed better than conventional risk factors or predictors. While cases had higher CHA2DS2-VASc scores and lower NOAC use, multivariate analysis demonstrated that SDNN and hypertension history remained significant predictors even after adjustment. This suggests HRV parameters may provide incremental prognostic value beyond traditional risk factors.

These findings and the descriptive data and association analysis results suggest that considering variations in HRV indices in conjunction with previously established predictors may significantly improve the overall acute ischemic stroke risk prediction efficiency.

HRV, reflecting the variability between successive sinus heartbeats, is a well-established predictor of cardiovascular events. Using a 24-h Holter monitor allows for the capture of diurnal heart rate variations and the detection of transient arrhythmias, making it a valuable tool for HRV analysis [14]. HRV parameters can be categorized into time-domain measures (e.g., SDNN, SDANN, SDNN Index, rMSSD, PNN50) and frequency-domain measures (e.g., LF, HF, LF/HF). The research on HRV mainly focuses on psychological analysis and neuropsychological tools can be used for specific analysis [15]. Research has shown that HRV can be modulated through biofeedback, potentially improving autonomic balance and impacting cardiovascular health [16]. HRV changes are also observed in various neurological disorders [17], suggesting that HRV may serve as a bridge between cardiovascular and neurological health. SDNN reflects the overall standard deviation of RR intervals, indicating overall autonomic regulation, while LF represents low-frequency oscillations linked to sympathetic and parasympathetic regulation [18]. Composite HRV indices may enhance predictive capabilities. Traditional predictors, such as CHADS2 scores, correlate with stroke outcomes [19], but HRV parameters provide a more comprehensive view of heart rate variability, potentially offering greater predictive power. Advances in wearable technology, including smart patches and optical volumetric plethysmography-based devices [20], have made HRV monitoring more accessible and convenient. Therefore, monitoring for acute ischemic stroke susceptibility or risk may become even more feasible, comprehensive and more accurate with history-based score (CHA2DS2-VASc score) in association with the HRV parameters derived measures suggested in this paper. Whenever monitoring for AIS in that subset of peoples it will contribute to better risk stratification and appropriate therapy assignment both for the non-valvular atrial fibrillation control and the AIS prevention.

Limitations

This paper presents significant findings, but several limitations should be noted. First, its retrospective design may introduce selection bias and hinder the establishment of causality, as existing medical records may be incomplete. Second, although 192 patients were analyzed, this sample size may not be adequate to generalize findings across diverse populations with paroxysmal non-valvular atrial fibrillation; a larger cohort would strengthen the conclusions. Third, as a single-center study conducted at Zhongnan Hospital, the results may not reflect broader populations due to potential differences in clinical characteristics based on geographic or sociocultural factors. Additionally, while several heart rate variability (HRV) parameters were examined, other relevant markers of autonomic function were not included, limiting a holistic understanding of patient risk. Measurement conditions could have also influenced HRV readings. Although some confounders were controlled for, lifestyle factors like body mass index and smoking were inadequately recorded and not addressed, which may impact both HRV and stroke risk. Lastly, the determination of acute ischemic stroke was based on clinical diagnosis, possibly overlooking subclinical strokes that could affect HRV parameters.

In summary, while this study offers initial insights into HRV and stroke risk in atrial fibrillation patients, the limitations emphasize the need for further research, particularly through prospective studies with larger, more diverse samples to validate and enhance predictive models for stroke risk.

Conclusion

Atrial fibrillation remains a significant risk factor for acute ischemic stroke, underscoring the need for enhanced focus on this patient group. This study conclusively demonstrated a robust link between heart rate variability (HRV) parameters and ischemic risk of stroke in patients with non-valvular paroxysmal atrial fibrillation. This study found that abnormal variation in HRV parameters was more observed in patients with acute ischemic stroke. Then unlike previous research, this study uniquely integrated multiple HRV parameters to develop novel indices such as AVTD, AVFD, AVA, and CDV that demonstrated somehow a significant association with the risk of AIS. These new metrics combined to conventional risk factors and CHA2DS2-VASc score brought great improvement to the model prediction for AIS risk. Therefore, these new metrics may offer the potential for refining the risk of stroke assessment. Future research should explore the incorporation of these indices into existing risk assessment tools to evaluate whether they provide improved predictive or prognostic value compared to traditional methods.