A clinical validation of AViTA home blood pressure monitor for atrial fibrillation detection in adults

Authors: Fu-Chun Chiu, I-Chih Huang

Abstract

Background

Atrial fibrillation (AF) is a stroke and heart complication risk but is often overlooked due to subtle symptoms. The accessibility of sphygmomanometers that can detect AF, may play crucial roles in identifying asymptomatic patients. AViTA BPM63B is an automatic blood pressure (BP) monitor for atrial fibrillation detection. This study validated the performance of BPM63B for AF detection using two novel algorithms of time-domain analysis and frequency-domain analysis which evaluate multiple statistics.

Methods

The study included 100 participants, with 47 participants being male (47%) and ages ranging from 24 to 89 years (mean ± SD: 59 ± 17 years). Each participant received three consecutive readings from the subject device and a 12-lead ECG assessment. The pulse wave data from subject device was analyzed using the two algorithms. The atrial fibrillation status thus determined was compared to the ECG results interpreted by a physician.

Results

Among the 100 participants, 52 patients had chronic atrial fibrillation, along with 48 outpatients exhibiting sinus rhythm or non-atrial fibrillation arrhythmias. Using the time-domain analysis method for atrial fibrillation detection, we achieved a sensitivity of 94.8% [95% confidence interval (CI), 90.08–97.75%] and a specificity of 98.6% (95% CI, 95.07–99.83%). Similarly, utilizing the frequency-domain analysis method resulted in a sensitivity of 91.6% (95% CI, 86.08–95.46%) and specificity of 94.4% (95% CI, 89.35–97.57%).

Conclusions

These findings suggest that AViTA BPM63B, which integrates two novel algorithms for atrial fibrillation diagnosis, demonstrates high sensitivity and specificity. Utilizing the AViTA BPM63B for BP monitoring could aid in the early detection of atrial fibrillation among outpatients in cardiology clinics.

Introduction

Hypertension is the most prevalent cardiovascular condition in patients with atrial fibrillation. Uncontrolled high blood pressure (BP) significantly influences the size and function of the left atrium, making it a crucial contributor to the development of a substrate susceptible to atrial fibrillation [1,2]. Atrial fibrillation is the most common cardiac arrhythmia and its prevalence increases annually. It is currently estimated to affect between 0.4 and 1% (approximately 2.3 million) of people in the USA. Atrial fibrillation, which is associated with other cardiac conditions, can have dangerous implications, including an increased risk of stroke, heart disease, and mortality. Because of these complications, there has been an increased need for hospitalization and medical care, thus increasing healthcare costs associated with atrial fibrillation [3].

The diagnosis of atrial fibrillation is usually based on long-term ECG monitoring and a variety of diagnostic tests. While current methods are costly in terms of both time and money, a quicker approach that can be implemented in home care is desirable. In 2004, an algorithm that evaluates pulse irregularity during BP measurement was integrated with home BP monitor [4]. Stergiou and Wiesel used an automated oscillometric device for self-home BP monitoring (Microlife BPA100 Plus; Microlife, Heerbrugg, Switzerland and BP3MQ1-2D; Microlife USA, Dunedin, Florida, USA) with an atrial fibrillation detector that using statistics based on time-domain analysis to carry out triplicate BP measurements in subjects with sinus rhythm, atrial fibrillation, and non-atrial fibrillation arrhythmias. During each BP measurement, the ECG was recorded simultaneously [5,6]. Similarly, frequency-domain analysis using fast Fourier transform of pulsewave signal of blood pressure monitor showed favorable accuracy [7]. These results demonstrate that the methods utilized have high diagnostic accuracy in detecting atrial fibrillation.

Following the development of the home blood pressure monitoring market, AViTA developed two algorithms to extract and analyze additional health-related insights from regular blood pressure measurements. The first method uses a pulse sequence and the time-domain analysis method [root mean squares of successive differences (RMSSD)/mean, Shannon entropy (ShEn), turning points ratio (TPR)] [8] which has been integrated into the blood pressure monitor. The second method, the discrete Fourier transform (DFT) frequency-domain analysis method, was the primary algorithm featured in the AViTA Toolbox, an application developed by AViTA. This toolbox was designed to be accessible to both tablets and mobile phones. The time-domain pressure wave is converted to the frequency-domain via DFT [9,10]. These algorithms could be extended to home blood pressure monitoring with simultaneous atrial fibrillation detection, serving as an initial screening tool for atrial fibrillation. The aim of this article is to validate the performance of AViTA BPM63B in atrial fibrillation detection when using time-domain analysis and frequency-domain analysis.

Methods

Participants

The study was conducted from May 28, 2020 to January 6, 2021. This trial was approved by the Ethics Committee of National Taiwan University Hospital, Taiwan (no. 201911111DSB). In this study, 100 participants were recruited from the outpatient population at the Division of Cardiology, Department of Internal Medicine of the National Taiwan University Hospital Yun-Lin Branch. Fifty-two patients with chronic atrial fibrillation, which were previously diagnosed using ECG, were enrolled, along with 48 outpatients exhibiting sinus rhythm or other non-atrial fibrillation arrhythmias. The age range of the subjects was 24–89 years, and 47 males and 53 females were included in the study.

Device

AViTA BPM63B (AViTA Co., Ltd., New Taipei, Taiwan) is an oscillometric upper-arm BP monitor intended for self-BP measurements at home. The device has a semi-conductive pressure sensor to measure static pressure values ranging from 0 to 300 mmHg and pulse rate values ranging from 40 to 199 beats per minute. The declared specific accuracy was ±3 mmHg for the pressure and ±4% for the pulse rate. The device uses automatic inflation and deflation to measure blood pressure and incorporates an atrial fibrillation algorithm to estimate atrial fibrillation. The device has one memory zone and a memory capacity of 60. The device requires four 1.5 V AA (LR6) alkaline batteries as energy sources. At least 300 blood pressure readings were obtained for each battery set. Universal adult cuffs (Easy Cuff, AViTA Corporation) are suitable for arm circumferences of 22–42 cm (9–17 in.). Throughout the measurement, the forearm and device were positioned on a table, with the palm facing upward and the device aligned with the heart level.

Procedure

Following informed consent, pertinent details, such as sex, age, and arm circumference, were recorded for each participant. After a 5-min resting period, a 12-lead ECG (FUKUDA FX-8322, K173226) was performed on the subject’s chest and limbs. Simultaneously, the blood pressure monitor cuff were placed on the left arm, and synchronization was maintained for approximately 1–2 min during the measurement. Following this setup, triplicate blood pressure measurements were taken with the participants seated, ensuring a minimum 1-min interval between each measurement. The entire procedure lasted for approximately 15–20 min (Fig. 1). The examining physician interpreted the ECG results and the pulse wave form detected from BP monitor was analyzed (Supplementary Figure 1, Supplemental digital content 1, http://links.lww.com/BPMJ/A229).

Analysis

Atrial fibrillation analysis

This study primarily utilizes a blood pressure monitor to conduct pulse wave sampling during the pressurization and depressurization processes of blood pressure measurements. The digital signal was sampled at regular intervals using an analog-to-digital converter. Subsequently, the signal was subjected to digital bandpass filtering to eliminate noise, such as hand muscle tremors or shaking, while amplifying the desired signal. The pulse wave data obtained can directly measure the presence of atrial fibrillation through an algorithm (time-domain analysis) in the blood pressure monitor that includes calculation of three statistics (i.e. RMSSD/mean, ShEn, and TPR). Atrial fibrillation is only found by the time-domain analysis when all three statistics reach their threshold (Supplementary Figure 2, Supplemental digital content 2, http://links.lww.com/BPMJ/A230). In addition, the pulse wave data were transmitted via Bluetooth to a mobile phone (AViTA Toolbox), where algorithms (frequency-domain analysis) were employed to perform atrial fibrillation detection. These methods were implemented to improve the sensitivity and specificity of atrial fibrillation detection (Fig. 2).

Root mean squares of successive differences/mean of the pulse sequence

RMSSD is a parametric statistic used to quantify the variability within a dataset, particularly in pulse-to-pulse intervals. It is sensitive to outliers and is calculated by summing the squares of the differences between each interval [9]. For each N pulse segment xi, the mean pulse-to-pulse intervals (mean RMSSD) was calculated according to the following

The atrial fibrillation algorithm selected RMSSD/mean over the threshold 1 for atrial fibrillation detection.

Shannon entropy of the pulse sequence

ShEn is a parametric statistic that measures uncertainty in a dataset related to its complexity and predictability. It ranges from zero to one, with zero indicating complete predictability and one indicating maximum randomness. ShEn is sensitive to outliers and is commonly used to differentiate between normal and abnormal rhythms such as atrial fibrillation. The number of bins used in the calculation affects the resolution, with 16 bins empirically found to offer an optimal balance between the resolution and distortion. The number of pulse-to-pulse intervals in each bin is then computed. The probability for each bin is computed as

where Ni is the number of beats in a particular bin, l the segment length, and Noutliers is the number of outliers. ShEn is then calculated as

In the atrial fibrillation algorithm selected ShEn over the threshold 2 for atrial fibrillation detection.

Turning points ratio of pulse sequence over threshold 3

TPR is a nonparametric statistic employed in our atrial fibrillation algorithm to gauge the randomness of fluctuations within a dataset. Unlike ShEn and RMSSD, the TPR is unaffected by assumptions regarding the distribution of the dataset. It assesses the number of turning points, defined as points either greater than both the preceding and succeeding terms or less than both, relative to the maximum possible turning points. The TPR assumes stationarity of the data, implying random fluctuations without discernible trends [10]. The statistical test within the algorithm posits a null hypothesis of stationarity (H0) versus an alternative hypothesis (H1) of non-stationarity. Specifically, H0 suggests randomness in pulse-to-pulse intervals, indicative of atrial fibrillation, while H1 suggests nonrandomness, which is characteristic of a normal sinus rhythm. For instance, white noise typically exhibits a turning point approximately every 1.5 data points.

Statistical analysis

The authenticity of the diagnostic test was assessed using various indicators including sensitivity, specificity, positive predictive value, and negative predictive value [10]. To enhance statistical analysis, the positive likelihood ratio and negative likelihood ratio were incorporated [11]. These measures provided additional insights into diagnostic test performance.

Results

One hundred subjects were recruited and all were included in the analysis. Forty-seven participants (47%) were male. The age range was 24–89 years (average, 59 years; SD, 17 years) and the arm size range was 22–34.1 cm (average, 27 cm; SD, 2 cm). Fifty-two patients (52%) had chronic atrial fibrillation, and 48 (48%) had a non-atrial fibrillation rhythm or sinus rhythm (Table 1).

A total of 299 simultaneous blood pressure measurements and ECG recordings were obtained from 100 subjects, and three readings were collected for each individual, except for one subject who had two readings instead of three. Half of the subjects (155/299, 51.8%) had atrial fibrillation, and half (143/299, 47.8%) had sinus rhythm or non-atrial fibrillation arrhythmias.

Using the time-domain analysis method to detect atrial fibrillation had 94.84% sensitivity [95% confidence interval (CI), 90.08–97.75%], 98.61% specificity (95% CI, 95.07–99.83%), and 95.03% accuracy. Additionally, while the average values of sensitivity of using only RMSSD/mean (99.35%), ShEn (96.13%), or TPR (98.71%) were better than that using all statistics, the average values of specificity of using only RMSSD/mean (71.53%), ShEn (52.78%), or TPR (63.19%) were much lower than that using all statistics. Using the frequency-domain analysis method to detect atrial fibrillation had 91.61% sensitivity (95% CI, 86.08–95.46%), 94.44% specificity (95% CI, 89.35–97.57%), and 91.75% accuracy. These data suggest that the AViTA blood pressure monitor, which implements two algorithms for atrial fibrillation diagnosis, has high performance (Tables 2 and 3).

Discussion

This study evaluated the diagnostic precision of an automated device designed for self-home blood pressure monitoring for the detection of atrial fibrillation. From the 299 modified BP monitor readings, 155 records from 52 patients with atrial fibrillation were measured three times, and only one record had no measured atrial fibrillation; the other 144 records had sinus rhythm or non-atrial fibrillation arrhythmias, including 51 pairs of irregular heartbeats in non-atrial fibrillation arrhythmias. It is worth noting that atrial fibrillation arrhythmia may be caused by other arrhythmias or movements at the time of measurement.

The time-domain analysis method demonstrated a sensitivity of 94.84%, specificity of 98.61%, and accuracy of 95.03%. The high specificity may be attributed to that atrial fibrillation is only found when all three statistics reach their threshold, thus avoiding false positive results when using only one statistics.

Time-domain analysis may also show limitations in detecting atrial fibrillation in specific populations. For instance, with participant #23 (atrial fibrillation-positive by ECG) that showed small irregularities in the R-R interval during atrial fibrillation, only the ShEn reached the threshold in time-domain analysis, while the RMSSD/mean and TPR did not. In contrast, the results from frequency-domain analysis found this small irregularity and classified this participant as atrial fibrillation-positive (see Supplementary Figure 3, Supplemental digital content 3, http://links.lww.com/BPMJ/A231). On the other hand, none of the participants in this study exhibited characteristics of large irregularities in sinus rhythm (sinus arrhythmia), which could also lead to difficulties in atrial fibrillation detection. Future studies are warranted that focus on specific atrial fibrillation patient populations and employ an algorithm that combines results from both time-domain and frequency-domain analyses to further improve atrial fibrillation detection.

In this study, the sensitivity (94.8%; 95% CI, 90.08–97.75%) and specificity (98.6%; 95% CI, 95.07–99.83%) using time-domain analysis are not significantly superior to the sensitivity (91.6%; 95% CI, 86.08–95.46%) and specificity (94.4%; 95% CI, 89.35–97.57%) using frequency-domain analysis. The results of both methods are comparable to previous studies using similar methods (Table 4) [4–7,12–17], and this concurrently verifies the utility of the algorithm employed in this study. The average values using time-domain analysis are slightly higher, which may be attributed to the usage of different statistics in respective methods. Additionally, the DFT used in frequency-domain analysis may lose some useful characteristics, which could be a chance for future optimization.

Although the validation protocols for blood pressure monitors in atrial fibrillation have been working for over 20 years, there are no guidelines to follow. The American National Standards Institute/Association for the Advancement of Medical Instrumentation/International Organization for Standardization (ANSI/AAMI/ISO) 81060-2:2013 standard for the evaluation of the accuracy of BP measuring devices states that the accuracy of the auscultation method for determination of BP measurement in subjects with atrial fibrillation is not known [18]. It also means there are no generally accepted guidelines for determining atrial fibrillation in BP monitors.

Conclusion

The use of the time-domain analysis method to detect atrial fibrillation had 94.84% sensitivity and 98.61% specificity. The use of the frequency-domain analysis method to detect atrial fibrillation had 91.61% sensitivity and 94.44% specificity. These data suggest that the AViTA blood pressure monitor BPM63B, which implements two algorithms for atrial fibrillation diagnosis, has favorable assessment accuracy.

Acknowledgements

This study was funded by the Conventional Industry Technology Development (CITD), Taiwan (no. E10800026012-187).

Conflicts of interest

I-C.H. is employed by and receives a salary from the AViTA Corporation, the manufacturer of AViTA BPM63B. The authors declare no conflicts of interest.

Supplementary Material