This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in the Journal of Medical Internet Research, is properly cited. The complete bibliographic information, a link to the original publication on https://www.jmir.org/, as well as this copyright and license information must be included.
Significant morbidity, mortality, and financial burden are associated with cardiac rhythm abnormalities. Conventional investigative tools are often unsuccessful in detecting cardiac arrhythmias because of their episodic nature. Smartwatches have gained popularity in recent years as a health tool for the detection of cardiac rhythms.
This study aims to systematically review and meta-analyze the diagnostic accuracy of smartwatches in the detection of cardiac arrhythmias.
A systematic literature search of the Embase, MEDLINE, and Cochrane Library databases was performed in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines to identify studies reporting the use of a smartwatch for the detection of cardiac arrhythmia. Summary estimates of sensitivity, specificity, and area under the curve were attempted using a bivariate model for the diagnostic meta-analysis. Studies were examined for quality using the Quality Assessment of Diagnostic Accuracy Studies 2 tool.
A total of 18 studies examining atrial fibrillation detection, bradyarrhythmias and tachyarrhythmias, and premature contractions were analyzed, measuring diagnostic accuracy in 424,371 subjects in total. The signals analyzed by smartwatches were based on photoplethysmography. The overall sensitivity, specificity, and accuracy of smartwatches for detecting cardiac arrhythmias were 100% (95% CI 0.99-1.00), 95% (95% CI 0.93-0.97), and 97% (95% CI 0.96-0.99), respectively. The pooled positive predictive value and negative predictive value for detecting cardiac arrhythmias were 85% (95% CI 0.79-0.90) and 100% (95% CI 1.0-1.0), respectively.
This review demonstrates the evolving field of digital disease detection. The current diagnostic accuracy of smartwatch technology for the detection of cardiac arrhythmias is high. Although the innovative drive of digital devices in health care will continue to gain momentum toward screening, the process of accurate evidence accrual and regulatory standards ready to accept their introduction is strongly needed.
PROSPERO International Prospective Register of Systematic Reviews CRD42020213237; https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=213237.
Cardiac arrhythmia encompasses a group of conditions in which the heart beats too quickly, too slowly, or in an irregular pattern. Significant morbidity, mortality, and financial burden are associated with cardiac rhythm abnormalities [
Although AF is the most common type of cardiac arrhythmia, other arrhythmias, such as premature cardiac contractions, are responsible for significant symptomatic burden. Premature atrial contractions have been shown to be an independent risk factor for all strokes in a longitudinal study [
Conventional screening tools, in the form of 12-lead electrocardiograms (ECGs) and ambulatory electrocardiography monitors, are often unsuccessful in detecting AF or other cardiac arrhythmias, such as bradyarrhythmias or tachyarrhythmias, because of the transient nature of episodes. The episodic and infrequent nature of cardiac arrhythmias means that they are not captured within the investigation period, making diagnosis very difficult.
Recent advances in mobile health technology and wearable electronic devices allow heart rhythm monitoring to be undertaken in real time with greater comfort, ease, and engagement [
Smartwatches have gained popularity in recent years, especially as a health tool for the detection of heart rhythms. Patients with a smartwatch can self-diagnose their heart rhythm within 30 seconds using one finger [
This study aims to systematically review and meta-analyze the diagnostic accuracy of smartwatches in the detection of cardiac arrhythmias.
This review was carried out and reported in accordance with the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) statement [
A thorough literature search was performed using the Embase, MEDLINE, and Cochrane Library databases. All articles published until February 2021 were included in the study. The appropriate MeSH (Medical Subject Headings) terms and free text all field searches were performed and combined with appropriate Boolean operator terms for
Inclusion criteria were as follows:
studies reporting detection of cardiac arrhythmias using smartwatches;
studies reporting sensitivity, specificity and diagnostic accuracy; or studies with adequate information to calculate these data; and
studies published or translated into English.
Exclusion criteria were as follows:
studies with no original data present (eg, review article, letter);
studies with no full text available;
studies >20 years; and
studies without adequate data to calculate sensitivity, specificity and diagnostic accuracy data.
Studies obtained from the literature search were analyzed, and duplicates were removed. Title, abstract, and full-text review were performed by 2 reviewers independently, and irrelevant studies were excluded. Disagreements were settled by consensus among the reviewers.
Data were extracted onto a standard spreadsheet template. Information regarding the journal, author, study design, type of smartwatch, number of subjects, and diagnostic accuracy data (sensitivity, specificity, accuracy, positive predictive value [PPV], and negative predictive value [NPV]) was selected from each paper.
The Quality Assessment of Diagnostic Accuracy Studies 2 tool was used to assess the risk of bias of the included studies [
Summary estimates of sensitivity, specificity, and area under the curve data were attempted using a bivariate model for diagnostic meta-analysis. Independent proportions and their differences were calculated and pooled using DerSimonian and Laird random effects modeling [
The database searches identified 292 studies that matched the criteria. Duplicates were removed, and 215 studies were eligible for title and abstract screening. Following this, a full-text review was undertaken, and a total of 18 studies were included in this review. Studies that failed to satisfy the inclusion criteria were excluded, and the reasons for exclusion of these articles included wrong intervention (such as the lack of use of a smartwatch) or wrong outcomes (such as studies that did not involve the detection of cardiac arrhythmias or reports on diagnostic accuracy). The study screening and selection process is shown in
The studies included in this systematic review were all published between 2017 and 2021. The outcome measure in the studies was mainly AF detection but also included bradyarrhythmias, tachyarrhythmias, and premature contractions. The studies measured diagnostic accuracy using smartwatches in 424,371 subjects in total. The Apple watch was used in 7 studies, Samsung smartwatches were used in 5 studies, and the remaining studies used a Huawei, Huami, or Empatica smartwatch. One study used the Wavelet wristband. Three different types of Huawei smartwatches were used in 2 studies to assess the diagnostic accuracy [
The reference standard was an ECG in most studies in the form of a 12-lead ECG, a Holter monitor, an ECG patch, telemetry, or an internet-enabled mobile ECG. In one study, an implantable cardiac monitor was used as the standard [
PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram for study selection.
Characteristics of included studies on detection of cardiac arrhythmias.
Authors (year) | Primary outcome | Study design | Type of sensor | Reference standard | Research or real-life setting | Type of smartwatch | Number of subjects |
Corino et al (2017) [ |
AFa detection | Prospective | PPGb | —c | Research | Empatica E4 | 70 |
Bumgarner et al (2018) [ |
AF detection | Prospective, nonrandomized, adjudicator blinded | — | 12-lead ECGd (physician reviewed) | Research | Apple watch | 100 |
Tison et al (2018) [ |
AF detection | Multinational, cohort | PPG | 12-lead ECG | Research | Apple watch | 1617 |
Wasserlauf et al (2019) [ |
AF detection | Prospective | PPG | Insertable cardiac monitor | Research | Apple watch | 24 |
Perez et al (2019) [ |
AF detection | Prospective, single group, open label, site less, pragmatic | PPG | ECG patch | Real life | Apple watch | 419,297 |
Zhang et al (2019) [ |
AF detection | Pilot, cohort | PPG | 12-lead ECG and physical examination | Real life | Huawei Watch GT | 263 |
|
|
|
|
|
|
The Honor Watch (Huawei) | 263 |
|
|
|
|
|
|
The Honor Band4 (Huawei) | 209 |
Ding et al (2019) [ |
AF detection | Observational | PPG | Holter monitor ECG | Research | Samsung Simband 2 | 40 |
Dorr et al (2019) [ |
AF detection | Prospective, two center, case-control | PPG | Internet-enabled mobile ECG | Research | Samsung GearFit 2 | 508 |
Bashar et al (2019) [ |
AF detection | Prospective | PPG | Holter monitor ECG | Research | Samsung Simband | 20 |
Bashar et al (2019) [ |
AF detection | Prospective | PPG | Holter monitor ECG | Research | Samsung Simband | 37 |
Valiaho et al (2019) [ |
AF detection | Multicenter prospective case-control | PPG | Three-lead ECG | Research | Empatica E4 | 213 |
Guo et al (2019) [ |
AF detection | Prospective | PPG | Clinical evaluation, ECG, or 24-hour Holter monitoring | Real life | Huawei Watch GT | 212 |
|
|
|
|
|
|
The Honor Watch (Huawei) | 265 |
|
|
|
|
|
|
The Honor Band4 (Huawei) | 264 |
Chen et al (2020) [ |
AF detection | Prospective | PPG | 12-lead ECG (physician reviewed) | Research | Amazfit Health Band 1S (Huami) | 401 |
Rajakariar et al (2020) [ |
AF detection | Prospective, multicenter validation | PPG | 12-lead ECG | Research | Apple watch | 200 |
Seshadri et al (2020) [ |
AF detection | Prospective | — | Telemetry | Research | Apple watch | 50 |
Selder et al (2020) [ |
AF detection | Observational, prospective cohort | PPG | One-lead ECG | Research | Wavelet wristband | 60 |
Han et al (2020) [ |
Premature atrial contraction or premature ventricular contraction | Prospective | PPG | ECG patch | Research | Samsung Gear S3 | 2 |
Caillol et al (2021) [ |
AF, atrial flutter, brady arrhythmias, and tachyarrhythmias | Prospective | PPG | 12-lead ECG | Research | Apple watch | 256 |
aAF: atrial fibrillation.
bPPG: photoplethysmography.
cNot available.
dECG: electrocardiogram.
The diagnostic accuracy of the smartwatch in detecting cardiac arrhythmias was analyzed, reporting a pooled sensitivity of 100% (95% CI 0.99-1.00;
Of the 18 studies, 7 (39%) reported data on accuracy. Among the 1769 subjects, the pooled accuracy for arrhythmia detection was 97% (95% CI 0.96-0.99;
Pooled analysis for sensitivity of cardiac arrhythmia detection by smartwatches. Effect sizes are shown with 95% CIs. A random effects model was used. ES: effect sizes.
Pooled analysis for specificity of cardiac arrhythmia detection by smartwatches. Effect sizes are shown with 95% CIs. A random effects model was used. ES: effect sizes.
Pooled analysis for accuracy of cardiac arrhythmia detection by smartwatches. Effect sizes are shown with 95% CIs. A random effects model was used. ES: effect sizes.
The PPV for cardiac arrhythmia detection was assessed in 9 studies using a smartwatch. These included a total of 421,267 subjects and reported a PPV of 85% (95% CI 0.79-0.90;
Pooled analysis for PPV of cardiac arrhythmia detection by smartwatches. Effect sizes are shown with 95% CIs. A random effects model was used. ES: effect sizes; PPV: positive predictive value.
Pooled analysis for NPV of cardiac arrhythmia detection by smartwatches. Effect sizes are shown with 95% CIs. A random effects model was used. ES: effect sizes; NPV: negative predictive value.
There was a high degree of variation between studies assessing cardiac arrhythmia detection using a smartwatch. The heterogeneity was statistically significant when all the studies were compared (
The assessment of bias using the Quality Assessment of Diagnostic Accuracy Studies 2 tool for the included studies is highlighted in
To the best of our knowledge, this systematic review and meta-analysis is the first to investigate the diagnostic accuracy of smartwatches for all cardiac arrhythmias. We have shown that the detection of cardiac arrhythmias using commercially available smartwatches is possible, with very high diagnostic accuracy. The overall sensitivity, specificity, and accuracy of these digital systems were 100%, 95%, and 97%, respectively. The pooled PPV and NPV for detecting cardiac arrhythmias were 85% and 100%, respectively. These values may offer clinicians a quantifiable appreciation for the use of smartwatches in a health care setting.
Although the aim of this study is to review the diagnostic accuracy of smartwatches in detecting cardiac arrhythmias, it is clear from the results that there are currently very few studies that assess the ability of PPG technology on smartwatches to detect non-AF arrhythmias.
A wide variety of smartwatches are commercially available, and this is reflected in the diverse range of smartwatches used in these studies (
Characteristics of smartwatches used in included studies.
Smartwatch | Company | Country | Approximate pricea, £ (US $) | Type | Photoplethysmography | Single-lead ECGb | Food and Drug Administration clearance ECG tracking | Electrodermal activity sensor |
Apple Watch | Apple | USA | 388 (531) | Watch | ✓c | ✓ | ✓ |
|
Honor Watch | Honor | China | 86 (117) | Watch | ✓ |
|
|
|
Huawei GT | Huawei | China | 89 (122) | Watch | ✓ |
|
|
|
Gear S3 | Samsung | South Korea | 160 (219) | Watch | ✓ |
|
|
|
Simband | Samsung | South Korea | N/Ad | Watch | ✓ | ✓ |
|
✓ |
Honor Band | Honor | China | 45 (61) | Fitness Band | ✓ |
|
|
|
Amazfit Healthband | Huami | China | 33 (45) | Fitness Band | ✓ |
|
|
|
GearFit2 | Samsung | South Korea | 49 (67) | Fitness Band | ✓ |
|
|
|
Wavelet wristband | Biostrap or Wavelet Health | USA | 180e (246) | Wristband | ✓ |
|
|
|
Empatica E4 | Empatica | USA | 1227f (1682) | Wristband | ✓ | ✓ |
|
✓ |
aPricing as per Amazon UK website on 22/04/2021.
bECG: electrocardiogram.
cIncluded with smartwatch.
dN/A: not applicable or data not available.
ePricing as per Biostrap shop on 22/04/2021.
fPricing as per Empatica store on 22/04/2021.
Overview of photoplethysmography sensor detection of arrhythmia. PPG: photoplethysmography.
The incidence of AF increases annually with an increase in the prevalence of risk factors, such as advancing age, obesity, hypertension, and type 2 diabetes. The challenge with detection is the ability of AF to remain asymptomatic or intermittent before eventually revealing itself. This poses a huge economic burden, accounting for 1%-2% of health care expenditure [
The detection of cardiac arrhythmias using smartwatches has multiple functionalities. It can be used to diagnose an abnormal rhythm, for monitoring of an arrhythmia, for example, in those with known paroxysmal AF, or for screening. Current methods of AF detection are criticized for their periodic investigative approach, during which an irregular pulse may be absent [
Wearable devices for
Furthermore, there is insufficient evidence on the burden of smartwatch-detected AF, which would prompt further evaluation and treatment. Guidance on what the clinician is expected to do with an episode of AF detected by a smartwatch is lacking. We suggest that this should be a critical prerequisite before introducing a digital detection tool into the general population; otherwise, overdiagnosis and an expectant role of clinicians from the public to assess their device-detected condition will become an even bigger burden on the health care system. A recent study evaluating the clinical outcome of the Apple smartwatch concluded that false-positive screening results may lead to overutilization of the health care system [
With evolving technology in the field of health care applications, there is a move to a more personalized and
There are many limitations to the studies in our review. At present, most studies have assessed the use of a PPG sensor and an accompanying algorithm to detect cardiac arrhythmias. However, they have not gone further to assess the use of such systems in health care. The largest study within our systematic review did not go beyond the participants’ self-reporting of an irregular pulse [
Many studies had a large proportion of data excluded because of insufficient PPG signal quality [
Finally, for smartwatch devices to be used as a screening tool for cardiac arrhythmias, such as AF detection, the value is highly dependent on disease prevalence. The estimated prevalence of AF in adults is between 2% and 4%. The prevalence increases with age, especially for those aged >65 years [
Regardless of the current studies, the future of health technology is undeniably advancing. Thus, measures should be taken early to ensure that such smartwatch technology supports ongoing national public health programs rather than having it run in parallel. Given the lack of recent success with the national NHS test and trace program in the United Kingdom, in which it fell short of its uptake aims when reaching contacts of people who tested positive for SARS-CoV-2 [
This systematic review and meta-analysis demonstrates the evolving field of digital disease detection and the increased role of machine learning in health care. The current diagnostic accuracy of smartwatch technology for the detection of cardiac arrhythmias is high. This shift signals a new direction in the field, allowing patients to play a greater role in disease diagnosis. However, before the use of these devices as a screening tool in health care is widely adopted, more studies are needed to clearly define the ideal population for the use of these systems, as well as to help form specific guidance on the conduct of device-detected disease. Consideration should also be placed for the wider use of smartwatch technology and similar digital tools in policy making decisions by health care departments in the future. Although the innovative drive of digital devices in health care will continue to gain momentum toward screening, the process of accurate evidence accrual and regulatory standards ready to accept their introduction is strongly needed.
Search strategy.
Quality assessment.
atrial fibrillation
electrocardiogram
Food and Drug Administration
Medical Subject Headings
negative predictive value
photoplethysmography
positive predictive value
Preferred Reporting Items for Systematic Reviews and Meta-Analyses
Infrastructure support for this research was provided by the National Institute for Health Research Imperial Biomedical Research Centre.
AD is Chair of the Health Security initiative at Flagship Pioneering UK Ltd. The remaining authors declare no conflicts of interest.