Results from a nationwide atrial fibrillation screening effort in Belgium

Introduction

Atrial Fibrillation (AF) is associated with an increased risk of stroke that can be mitigated with anticoagulation therapy. Opportunistic screening for AF for primary stroke prevention is recommended in subjects above 65. However, the paroxysmal and asymptomatic nature of AF hampers early detection with a single time point screening. Multiple time point measurements are superior to single time point measurements for the detection of AF. New technologies such as photoplethysmography (PPG) enable large scale AF screening with repetitive measurements at low-cost using only a smartphone.

Purpose

To explore an entirely online AF screening program in subjects with an elevated stroke risk.

Methods

The Belgian Heart Rhythm Association launched a digital marketing campaign, to promote AF screening during “The Belgian Week of the Heart Rhythm”. Candidates were referred to an online questionnaire to calculate their CHADS-VASC score. Subjects older than 18 with a CHADS-VASC score of 2 or more were allowed to enter the screening program. AF screening was performed with a PPG-based smartphone application. A 60-second PPG trace is captured by placing a fingertip on the smartphone's camera. The smartphone application analyses the PPG trace with an artificial intelligence software. Subjects were instructed to perform measurement twice daily and while experiencing symptoms over the course of 7 days. Measurements were classified as AF or non-AF by the algorithm and were reviewed by medical technicians.

Results

Of the 12.602 candidates who completed the questionnaire, 6.020 subjects met the inclusion criteria and were offered screening. However, only 2.111 (35%) participated in the screening program. The mean age of participants was 63±11 years, 37.3% was male, median CHADS-VASC was 2 (2–3). 257 participants (12.2%) were previously known with AF. In total 25.362 PPG recordings of 60 seconds were performed of which 258 demonstrated AF. AF was detected in 56 participants (2.7%). This was a new finding in 36 participants (1.7%) meaning that 64.3% of participants demonstrating AF were not previously known with AF. The number needed to screen was 58.6 to detect AF in a population without a history of AF and the number needed to invite was 167.2. Only 20 participants (7.8%) with a history of AF demonstrated AF during the screening program.

Conclusions

AF screening in subjects with an elevated stroke risk is feasible with an entirely online screening program without the need for medical hardware or medical personnel with an acceptable number needed to screen. However, this approach failed to target subjects in the highest age groups and since almost two thirds of the subjects interested in the screening program failed to commence screening, approaches to increase this response (specifically in high-risk groups) needs to be explored.

Funding Acknowledgement

Type of funding sources: None.

Performance of an artificial intelligence algorithm to detect atrial fibrillation on a 24-hour continuous photoplethysmography recording using a smartwatch: ACURATE study

Background

In the awakening era of mobile health, wearable devices capable of detecting atrial fibrillation (AF) are on the rise. Smartwatches and wristbands are equipped with photoplethysmography (PPG) technology that enables (semi)continuous rhythm monitoring. These devices have been pioneered already in a few screening trials. However, such devices are being spread among consumers at a pace that is not paralleled by the evidence supporting their clinical performance. This imbalance reflects the urgent need for validation studies.

Purpose

To determine the diagnostic performance of an artificial intelligence algorithm to detect AF using photoplethysmography acquired by a smartwatch.

Methods

One hundred patients (≥18 years) without a pacemaker-dependent heart rhythm who were referred to a university hospital or a large tertiary hospital for elective 24-hour ECG Holter monitoring were asked to wear a continuous PPG monitoring smartwatch (i.e. Samsung GWA2 or Empatica E4) simultaneously with the Holter. All activities of daily life were allowed. The ECG trace and PPG waveform were synchronised and fragmented in one-minute fragements. The one-minute ECG fragments were labelled as AF, non-AF, or insufficient quality based on the routine clinical interpretation of the 24-hour Holter (i.e. software + physician overreading). The one-minute PPG fragments were analysed by an artificial intelligence (AI) algorithm (i.e. FibriCheck) and were given the same labels. Diagnostic metrics of the PPG AI algorithm were calculated with respect to the ECG interpretation, for all fragments with sufficient quality for both PPG and ECG.

Results

Four patients had to be excluded due to technical error (3 Holter errors, 1 smartwatch error). The mean age in the remaining study population (n=96) was 59±16 years, 51 (53%) were men and 15 (15.6%) were known with permanent AF. In this population, simultaneous ECG and PPG monitoring was recorded for 115,245 one-minute fragments. Fragments of insufficient quality for ECG (n=1,454; 1.3%), PPG (n=25,704; 22.3%) or both (n=15,362; 13.3%) were excluded. PPG fragments were more frequently of insufficient quality (p<0.001). AF was present in 10,255 (14.1%) of the resulting 72,725 high-quality one-minute fragments. The sensitivity of PPG to detect AF was 93.4% (CI 92.9% - 93.8%). The specificity of PPG to exclude AF was 98.4% (CI 98.3% - 98.5%). As a result, the overall accuracy of the PPG algorithm on one-minute fragment level was 97.7% (CI 97.6%- 97.8%).

Conclusion

Continuous out-of-hospital PPG monitoring using a smartwatch in combination with an AI algorithm can accurately discriminate between AF and non-AF rhythms in a heterogenous patient population. PPG quality is more often affected than ECG quality during daily life activities.

Funding Acknowledgement

Type of funding sources: Foundation. Main funding source(s): Research Foundation-Flanders, Strategic Basic Research Fund

Assessment of heart rate agreement on continuous photoplethysmography monitoring using a smartwatch versus beat-to-beat synchronized ECG monitoring

Background

In the awakening era of mobile health, wearables equipped with photoplethysmography (PPG) technology to monitor the heart rate (HR) and rhythm are on the rise. Smartwatches and wristbands enable HR monitoring for consumers at massive scale. Unfortunately, once consumers become patients, physicians are limited by insufficient evidence to support the clinical use of PPG based wearables. Accurate identification of heartbeats is the first step in the interpretation of PPG traces and should be validated.

Purpose

To assess the agreement between continuous PPG monitoring using a smartwatch and continuous ECG Holter monitoring in the identification of heartbeats and calculation of the HR.

Methods

One hundred patients (≥18 years) without a pacemaker-dependent heart rhythm who were referred to a university hospital and a large tertiary hospital for elective 24-hour ECG Holter monitoring were asked to wear a continuous PPG monitoring smartwatch (i.e. Samsung GWA2 or Empatica E4) simultaneously with the 24-hour Holter monitor. All activities of daily life were allowed. The ECG trace and PPG waveform were synchronised and fragmented in one-minute fragments. The one-minute ECG fragments were labelled as AF, non-AF, or insufficient quality based on the routine clinical interpretation (i.e. software + physician overreading), and the average HR during each fragment was calculated by Holter algorithm. The PPG fragments were analysed by an artificial intelligence (AI) algorithm (i.e. FibriCheck) that labelled fragments as sufficient or insufficient quality, identified the number of heartbeats and calculated the HR. The agreement between the HR on ECG and PPG in sufficient quality tracings was analysed with linear regression, Pearson's product-moment correlation and Bland-Altman analysis. A subanalysis was performed for AF rhythm and non-AF rhythms.

Results

A total of 72,725 simultaneous ECG and PPG one-minute fragments were recorded in 96 patients, after excluding 4 patients (due to 3 Holter and 1 smartwatch technical error) and 42,520 minutes (36.9%) of insufficient quality (ECG 1,454 (1.3%); PPG 25,704 (22.3%), ECG and PPG 15,362 (13.3%)). The correlation (r=0.935) between ECG and PPG HR was statistically significant (CI 0.934–0.936; P<0.001), with a mean difference between ECG and PPG of 0.8bpm. The lower and upper limit boundary (LLB and ULB; defined as ±1.96 SD) were −8.0bpm and 9.7bpm, respectively, i.e. 95% of PPG measurements identified the HR within 8bpm below or 10bpm above the ECG reference. The mean difference between ECG and PPG HR in the AF subgroup (n=10,255 (14.1%)) was 0.9bpm (LLB −8.4bpm; ULB 10.2bpm) and 0.8bpm in the non-AF subgroup (LLB −0.8bpm; ULB 9.6bpm).

Conclusion

The AI algorithm analysing continuous out-of-hospital PPG tracings can annotate heartbeats and assess HR without a clinically significant bias compared to continuous ECG monitoring, both during AF and non-AF rhythms in a heterogenous patient population.

Funding Acknowledgement

Type of funding sources: Public grant(s) – National budget only. Main funding source(s): Research Foundation-Flanders, Strategic Basic Research Fund Correlation plot & Bland-Altman plot

Evaluation of the device independent nature of a photoplethysmography-deriving smartphone app

Funding Acknowledgements

Type of funding sources: None.

Background

Smartphone apps using photoplethysmography (PPG) technology enable digital heart rhythm monitoring through their built-in camera, without the need for additional, specific, or costly hardware. This may positively impact the availability and scalability of remote monitoring. However, the diversity of smartphone specifications on the consumer market may raise concerns regarding the robustness of AF detection algorithms between various devices.

Purpose

To study the device independency of AF detection performance by a PPG-based smartphone application.

Methods

Patients from the cardiology department were consecutively enrolled. Patients were handed 7 iOS models and 1 Android model and were asked to consecutively perform one PPG measurement per device. A 12-lead electrocardiogram (ECG) was collected during the same consultation and interpreted by a cardiologist as reference diagnosis. To allow an objective comparison across the devices, patients who failed to perform one successful measurement on each device were excluded. Additional exclusions were atrial flutter rhythms and insufficient quality results. Sensitivity, specificity and accuracy were calculated with respect to the reference diagnosis. McNemar’s analysis was used for the head-to-head comparison of the sensitivity and specificity of the proprietary algorithm on the different smartphone devices.

Results

A total of 150 patients participated in the study with a median CHA2DS2-VASc score of 3 (interquartile range: 1-5). The median age of the study population was 70 (interquartile range: 56-79) years. In total, 54.7% of the population was male and the AF-prevalence was 35.3%. After the exclusion of patients with atrial flutter (n = 14) and patients who did not successfully perform a PPG measurement on each device (n = 5), diagnostic-grade results of 131 patients were used to calculate the performance of the proprietary algorithm. The sensitivity and specificity of the AF detection algorithm ranged from 90.9% (95% CI 75.7-98.1) to 100.0% (95% CI 91.0-100) and 94.5% (95% CI 86.6-98.5) to 100.0% (95% CI 94.6-100), respectively. The overall accuracy across the devices ranged from 94.4% (95% CI 88.3-97.9) to 99.0% (95% CI 94.6-100). Head-to-head comparisons of the results did not reveal significant differences in sensitivity (P = 0.125-1.000) or specificity (P = 0.375-1.000) of the proprietary AF detection algorithm among the different devices.

Conclusion

This study demonstrated the device-independent nature of the PPG-deriving smartphone application with respect to 12-lead ECG diagnosis.

The accuracy of physician interpretation of PPG vs single-lead ECG vs 12-lead ECG for the detection of atrial fibrillation

Funding Acknowledgements

Type of funding sources: None.

Introduction

The increasing availability of smartphones has enabled rhythm monitoring in large populations using standalone photoplethysmography (PPG) apps or singe-lead electrocardiography (ECG) with add-on devices. Current guidelines note that when atrial fibrillation (AF) is suspected by an automated algorithm, confirmation on an ECG tracing is required. The use of PPG alone to establish the diagnosis is not generally accepted, even when overread. The performance of physicians to discriminate between sinus rhythm (SR) and AF based on PPG alone is unknown.

Purpose

To study the performance of physicians to detect AF based on PPG vs single-lead ECG vs 12-lead ECG, and to explore the incremental value of a tachogram, Poincaré plot, and algorithm output to the interpretation of the PPG waveform by physicians.

Methods

PPG, single-lead ECG and 12-lead ECG data were simultaneously recorded in 30 patients. Diagnostic reference was the 12-lead ECG, read by two cardiologists. Cardiologists, electrophysiologists and cardiology fellows were invited to analyse the data of 30 patients (10 in SR, 10 in SR with extrasystoles and 10 in AF) through online surveys and classify the readings as ‘SR’, ‘ectopic/missed beats’, ‘AF’, ‘flutter’ or ‘unreadable’. For dichotomous analysis, ‘unreadable’ was reclassified as incorrect, the other options were reclassified as AF ‘present’ or ‘absent’. In the first survey, PPG data were presented subsequently as a waveform, stepwise adding the tachogram and Poincaré plot, and algorithm information. In the next two surveys, the single-lead and 12-lead ECG traces were presented. Sensitivity and specificity for all presentations were calculated with respect to the reference diagnosis. Diagnostic performances were compared with the Obuchowski-Rockette’s ANOVA approach with Jackknife covariance estimation and Benjamini-Hochberg correction.

Results

Sixty-five physicians completed the PPG survey and analysed the PPG waveforms with 88.8% sensitivity and 86.3% specificity for AF. The diagnostic metrics significantly increased to 95.5% sensitivity (P < 0.001) and 92.5% specificity (P < 0.001) after providing the tachogram and Poincaré plot. Fifty-seven physicians completed both ECG surveys and analysed the single-lead ECG outputs with 91.2% sensitivity and 93.9% specificity, while 12-lead ECG outputs were analysed with 93.9% sensitivity and 98.6% specificity. Hence, qualitative analysis of a PPG waveform with tachogram and Poincaré plot had a similar diagnostic performance to detect AF compared to single-lead ECG analysis and a similar sensitivity (P = 0.792) but lower specificity (P = 0.035) compared to 12-lead ECG.

Conclusions

PPG rhythm recordings, analysed by physicians as a waveform in combination with the corresponding tachogram and Poincaré plot, achieve similar diagnostic accuracy as single-lead ECG to detect AF. Abstract Figure.