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 JMIR Cardio, is properly cited. The complete bibliographic information, a link to the original publication on http://cardio.jmir.org, as well as this copyright and license information must be included.
Atrial fibrillation (AF) is often paroxysmal and minimally symptomatic, hindering its diagnosis. Smartwatches may enhance AF care by facilitating long-term, noninvasive monitoring.
This study aimed to examine the accuracy and usability of arrhythmia discrimination using a smartwatch.
A total of 40 adults presenting to a cardiology clinic wore a smartwatch and Holter monitor and performed scripted movements to simulate activities of daily living (ADLs). Participants’ clinical and sociodemographic characteristics were abstracted from medical records. Participants completed a questionnaire assessing different domains of the device’s usability. Pulse recordings were analyzed blindly using a real-time realizable algorithm and compared with gold-standard Holter monitoring.
The average age of participants was 71 (SD 8) years; most participants had AF risk factors and 23% (9/39) were in AF. About half of the participants owned smartphones, but none owned smartwatches. Participants wore the smartwatch for 42 (SD 14) min while generating motion noise to simulate ADLs. The algorithm determined 53 of the 314 30-second noise-free pulse segments as consistent with AF. Compared with the gold standard, the algorithm demonstrated excellent sensitivity (98.2%), specificity (98.1%), and accuracy (98.1%) for identifying irregular pulse. Two-thirds of participants considered the smartwatch highly usable. Younger age and prior cardioversion were associated with greater overall comfort and comfort with data privacy with using a smartwatch for rhythm monitoring, respectively.
A real-time realizable algorithm analyzing smartwatch pulse recordings demonstrated high accuracy for identifying pulse irregularities among older participants. Despite advanced age, lack of smartwatch familiarity, and high burden of comorbidities, participants found the smartwatch to be highly acceptable.
Atrial fibrillation (AF) is the most common heart rhythm problem in the world and the number of patients living with AF is increasing rapidly [
Systematic and opportunistic screening of older populations for AF using mobile and digital health is feasible and can identify asymptomatic, community-dwelling individuals with undiagnosed AF [
Although appealing, the use of a smartwatch for AF monitoring introduces unique and significant technical challenges, such as motion and noise artifacts generated during activities of daily living (ADLs), as well as usability concerns, as individuals at risk for AF tend to be older, less familiar with mHealth devices, and frequently affected by physical and cognitive impairments that can impede operation of, and comfort with, mHealth technologies [
In this investigation, we sought to test the performance of a novel, real-time realizable, automated algorithm for AF discrimination using pulse data obtained from a smartwatch among older individuals with, or at risk for, AF while they executed simulated ADLs. Furthermore, we assessed study participants’ impressions of the smartwatch, generally and across specific usability domains. Finally, we identified characteristics associated with comfort using a smartwatch for rhythm analysis.
This observational study was designed to evaluate the performance and usability of a smartwatch for heart rhythm analysis among older individuals with, or at high risk for, AF. Participants were enrolled between June 2016 and November 2017 from the ambulatory clinics at the University of Massachusetts Medical Center. All participants provided written informed consent before the study participation. This study was approved by the University of Massachusetts Medical School (UMMS) Institutional Review Board (UMMS IRB number H00009953).
Study staff reviewed the electronic health records (EHR) of all patients presenting for an ambulatory visit. To be considered eligible for enrollment, participants were required to be 21 years of age or older, capable of and willing to provide informed consent, and able to speak and read English. Individuals were excluded from study participation if they were pregnant, incarcerated, had reported an adverse reaction to ECG electrodes or a Holter monitor, or refused to adhere to any aspect of the proposed study protocol, including a brief walk test. The staff telephoned potentially eligible study participants (both those with and without AF) 1 to 2 weeks before their clinic visit to assess interest in study participation. A total of 78 patients were telephoned and 48 patients expressed interest in the study. These patients were then approached for consent after their clinic visit. However, 7 patients declined participation at this time and 1 patient was excluded due to wrist size being too large for the smartwatch, resulting in a final sample of 40 participants.
Trained staff abstracted clinical, electrocardiographic, and laboratory data from the EHR on all participants, including data obtained during the ambulatory visit immediately preceding the study examination. Resting heart rate and rhythm status, as well as vital signs, including respiratory rate, systolic and diastolic blood pressure, and body mass index, were obtained on all participants. The use of cardiovascular medications was also abstracted from the EHR. The wrist circumference and skin tone of the participants were determined by the research staff at the time of their study examination as these factors may potentially influence pulse recordings obtained by the smartwatch.
After providing a brief overview of the smartwatch (Samsung Simband 2,
The Samsung Simband 2 (
Samsung Simband 2 smartwatch showing simultaneous single-lead ECG (electrocardiogram) and PPG (photoplethysmogram) recordings.
Representative electrocardiographic, pulse, and pulse interval waveforms as recorded by the Samsung Simband 2.0 smartwatch from study participants in various rhythm states. Panels 1 to 4 (top to bottom) represent patients in various rhythm states: (1) shows a patient in normal sinus rhythm, (2) shows a patient with premature atrial contractions, (3) shows a patient with premature ventricular contractions, and (4) shows a patient in atrial fibrillation. For each patient, the single-lead electrocardiogram (a) and pulse plethysmography waveforms (b) are collected by the smartwatch. The pulse interval (c) and Poincare plots (graphed on right) for each patient are also calculated and represented. BPM: beats per min; HR: heart rate; ECG: electrocardiogram; PPG: photoplethysmogram.
Once downloaded onto secure UConn study servers, each participant’s pulse data were divided into 30-second segments and then analyzed using a novel motion noise artifact (MNA) detection algorithm. The details of the MNA detection algorithm have been described elsewhere [
AF is characterized by disorganized atrial electrical activity that stimulates the ventricles in a random fashion that increases beat-to-beat variability. Our approach to pulse analysis from PPG data has been described in detail and uses 2 validated statistical techniques. The first of these methods, sample entropy (SampEn), measures the complexity of pulse variability. The second, root mean square of successive difference of RR intervals (RMSSD), quantifies peak-to-peak pulse waveform variability; the pulse peak determination methodology has also been previously described [
A 7-lead Holter monitor was used to record an ECG as the gold standard for comparison. ECG data from each participant were labeled with the participant’s study ID and recorded on an Secure Digital card. Staff downloaded the ECG data and transferred it securely to UConn investigators (KC, DH, SKB), where it was stored on firewall protected servers and analyzed separately from the pulse data from watches. Holter data were obtained for 41 participants, but were unavailable for 1 participant, likely because of inadequate contact of leads.
The ECG data were divided into 30-second segments and linked to PPG data using time-stamps and study IDs. Each segment of ECG data was analyzed by a highly accurate and well validated AF detection algorithm using a combination of time-varying coherence functions and Shannon Entropy (ShE) [
We characterized system usability, as well as the psychosocial, cognitive, and sociodemographic characteristics of study participants using validated instruments. We also assessed other factors known to influence perceptions of mHealth devices, including education level, yearly income, employment status, and prior experience with smart devices, as measured by smartphone or smartwatch ownership and social media use.
Usability in this study was measured globally and across several usability domains. The Brooke System Usability Scale (SUS) is a widely used and validated 10-item questionnaire that assesses multiple dimensions of usability and has been used to assess prior mHealth devices [
A separate investigator-generated assessment was also administered to measure participants’ perceptions of the smartwatch’s ease of use, its overall importance, privacy concerns related to smartwatch use, perceived fit of the device into daily activities, and comfort with use, as measured by stress associated with using the device. Participants responded to all questions using the same 5-point Likert scale used in Brooke SUS.
We employed 3 validated questionnaires to assess cognitive impairment and mood. The Montreal Cognitive Assessment is a validated tool for screening of mild cognitive impairment and is widely used in clinical settings [
We first calculated the proportion of noise-free data segments out of the total number of data segments. We then calculated the appropriate test characteristics, including sensitivity, specificity, and accuracy, to examine the performance of our automated pulse analysis algorithm from 314 noise-free pulse segments for the detection of an irregular pulse consistent with AF compared with the results of a validated algorithm examining contemporaneous 7-lead Holter ECG (criterion standard), using established threshold values of RMSSD, SampEn, and Poincare plot [
The overall usability of the smartwatch for arrhythmia monitoring was examined using validated Brooke SUS. Unadjusted linear regression was then used to identify patient-level characteristics associated with overall Brooke SUS score, as well as Likert-type scores across several usability domains (system ease of use, system importance to user, system privacy concerns, perceived fit of system into daily activities, and comfort with the system, as measured by stress induced by use). Age, sex, history of coronary artery disease, coronary bypass graft procedure, cardioversion, stroke, education level, smartphone ownership, social media use, cognitive impairment, depression, and anxiety were examined as predictors of these usability outcomes. All analyses were performed in MATLAB 9.1 (MathWorks) and Stata 13 (StataCorp).
The characteristics of the 40 participants enrolled in the study are shown in
A total of 40 participants wore the smartwatch for an average of 42 (SD 14) min, generating a total of 2538 30-second data segments of pulse waveform recordings, of which 314 were noise-free. All data from 1 participant were corrupted by motion/noise artifact, likely from a poor wristband fit. Furthermore, 63 out of the 314 clean 30-second pulse segments were deemed to be
All 40 participants were included in the usability analysis. The smartwatch demonstrated high usability for rhythm analysis, as determined using the validated Brooke SUS, with over two-thirds of participants (67.7%) considering the watch to be highly usable. The average Brooke SUS score was 72.9 (SD 17.5). Individual usability domains, including ease of use, importance to user, fit into daily activity, comfort with privacy, and stress associated with use were also assessed using a Likert scale (1 to 5). These usability domains were generated by the investigators for the specific purpose of evaluating this technology and are presented as univariate dot plots to best represent distribution of responses, which may be more meaningful than summary statistics. Overall usability was high across usability domains (
Baseline characteristics of study participants (N=40).
Characteristics | Statistics | ||
Age (years), mean (SD) | 70.6 (8) | ||
Sex, male, n (%) | 32 (80) | ||
40 (100) | |||
Skin tone: tan | 27 (68) | ||
Skin tone: pale | 10 (25) | ||
Skin tone: unspecified | 3 (8) | ||
Wrist circumference, inches, mean (SD), (n=32) | 6.9 (0.7) | ||
CHA₂DS₂-VASc scorea, mean (SD) | 2.6 (1.3) | ||
Body mass index, kg/m2, mean (SD) | 29.3 (5.1) | ||
Respiratory rate, bpmb, mean (SD) | 16.7 (1.2) | ||
Resting heart rate (per electrocardiogram), bpm, mean (SD) | 68.8 (14.3) | ||
Systolic blood pressure, mm Hg, mean (SD) | 126.3 (17.7) | ||
Diastolic blood pressure, mm Hg, mean (SD) | 72.0 (10.4) | ||
Hypertension, n (%) | 25 (63) | ||
Hyperlipidemia, n (%) | 23 (58) | ||
Current smoking, n (%) | 0 (0) | ||
Diabetes mellitus, Type 2, n (%) | 9 (23) | ||
Coronary artery disease, n (%) | 13 (33) | ||
Prior coronary artery bypass graft, n (%) | 7 (18) | ||
Congestive heart failure, n (%) | 2 (5) | ||
Sleep apnea, n (%) | 10 (25) | ||
Prior cardioversion (%) | 14 (35) | ||
Stroke, n (%) | 2 (5) | ||
History of atrial fibrillation | 28 (70) | ||
Paroxysmal | 17 (60) | ||
Permanent | 4 (14) | ||
Persistent | 5 (18) | ||
Unspecified | 2 (7) | ||
Sinus rhythm | 30 (77) | ||
Atrial fibrillation | 9 (23) | ||
Beta-blocker | 27 (68) | ||
Calcium channel blocker | 12 (30) | ||
Statin | 29 (73) | ||
Antiarrhythmic drug | 12 (30) | ||
Digoxin | 1 (3) | ||
Anticoagulant | 21 (53) | ||
Cognitive impairment | 13 (59) | ||
Anxiety | 4 (18) | ||
Depression | 4 (18) | ||
Completed some high school | 2 (5) | ||
Graduated high school | 9 (23) | ||
Graduated high school, some college | 5 (13) | ||
Graduated college | 8 (20) | ||
Graduated college, some graduate school | 2 (5) | ||
Completed a graduate degree | 14 (35) | ||
Employed full-time | 7 (17) | ||
Employed part-time | 4 (10) | ||
Full time home-maker or caretaker | 1 (3) | ||
Retired | 28 (70) | ||
Less than $10,000 | 1 (4) | ||
$10,000 to $29,999 | 1 (4) | ||
$30,000 to $49,999 | 3 (11) | ||
$50,000 to $69,000 | 3 (11) | ||
$70,000 to $89,999 | 8 (30) | ||
$90,000 to $149,999 | 9 (33) | ||
$150,000 or more | 2 (7) | ||
Unreported | 13 (33) | ||
Own smartphone | 22 (55) | ||
Own smart watch | 0 (0) |
aCHA₂DS₂-VASc score: clinically used tool for stroke risk assessment.
bbpm: beats per minute.
cCognitive impairment is based on Montreal Cognitive Assessment score <26, anxiety is based on Generalized Anxiety Disorder–7 score >4, and depression is based on Patient Health Questionnaire–9 score >4. Data are available for 22 participants for all measures.
Responses to smartwatch for atrial fibrillation usability questions. Each circle represents an individual participant’s coded response.
Factors associated with individual usability domains and system usability score.
Variables | Individual usability domains | Overall Brooke system usability scoreb | ||||
Ease of useb | Importance to userb | Fit into daily activityb | Comfort with privacyb | Comfort with useb | ||
Age | 0.02 | −0.01 | −0.02 | 0.00 | −0.039a | −0.082 |
Sex, male | −0.38 | −0.01 | 0.14 | 0.52 | −0.063 | −8.905 |
History of coronary artery diseasec | 0.86 | 0.13 | −0.19 | −0.23 | 0.046 | −2.680 |
Prior coronary artery bypass graftc | −0.54 | −0.04 | −0.26 | 0.21 | −0.935a | −2.857 |
Prior cardioversionc | −0.22 | 0.15 | 0.11 | 0.72a | −0.088 | 0.238 |
Prior strokec | 1.37 | 0.27 | 0.68 | 0.49 | 0.842 | 11.042 |
Education leveld | 0.04 | 0.02 | 0.06 | −0.04 | 0.152 | −1.622 |
Smartphone ownershipc | −0.07 | 0.09 | 0.21 | −0.45 | 0.468 | −8.295 |
Social media usee | 0.37 | −0.02 | 0.03 | −0.06 | 0.030 | −0.814 |
Cognitive impairmentf | −0.05 | −0.19 | −0.17 | 0.16 | −0.188 | 6.656 |
Depressionf | 0.37 | 0.14 | 0.47 | −0.61 | −0.158 | −6.542 |
Anxietyf | 0.21 | 0.44 | 0.17 | −0.61 | −0.158 | −8.125 |
aIndicates statistical significance at
bUnadjusted beta coefficients from univariate regression analysis.
cCoronary artery disease, coronary artery bypass graft, cardioversion, stroke, smartphone ownership are all coded as yes (1) or no (0).
dEducation coded from 1 to 6, from completed some high school to some graduate degree.
eSocial media use coded from 1 to 5, from no use to >6 hours per week.
fCognitive impairment coded as Montreal Cognitive Assessment <26, depression coded as Patient Health Questionnaire–9 >4, anxiety coded as Generalized Anxiety Disorder−7 >4. Cognitive impairment, depression, and anxiety available for 22 participants.
We enrolled 40 older patients presenting to an ambulatory clinic in an observational study of mHealth devices for arrhythmia monitoring. Participants wore a smartwatch (the Samsung Simband 2.0,
Early AF detection results in treatment with oral anticoagulants that reduce stroke risk by up to 70% [
The most commonly prescribed noninvasive ECG monitors (24-hour Holter monitors) demonstrate low yield for AF detection across a wide array of high-risk subgroups, likely as the monitoring coverage of a 24- or 48-hour monitor is simply too brief [
Owing to the suspected prevalence and adverse health impact of undiagnosed AF, there is interest in developing new mHealth tools capable of enabling long-term, noninvasive monitoring. Recently, several companies have developed new software and hardware needed to harness the ubiquity and usability of commercial wrist-based wearable devices for AF screening and monitoring [
Like the Apple Watch–AliveCor KardiaBand dyad and similar to the recently FDA-cleared Apple Watch 4, the Samsung Simband 2.0 records PPG and single-lead ECG signals using an electrode embedded in the watch (
The pulse analysis approach tested in our study demonstrated high accuracy for the detection of AF using smartwatch pulse recordings despite the fact that participants were asked to perform activities intentionally designed to create MNAs [
Our approach to AF detection using a smartwatch involves passive pulse acquisition and real-time analysis using methods that are accurate but not computationally demanding [
Consistent with this hypothesis, most participants deemed the smartwatch system highly usable overall and expressed comfort with using the system for home heart rhythm monitoring. Our finding that the majority of participants owned a smartphone and were willing to use a smartwatch to monitor themselves debunks the commonly held misconceptions about older Americans at risk for AF and their facility with mobile devices, but is entirely consistent with an emerging literature showing that older Americans are increasing using smart devices and are open to using wearables for disease prevention and treatment [
Although we did not find significant associations between any participant characteristics and overall smartwatch usability, we did observe that older age and history of coronary artery bypass graft surgery were associated with lesser comfort using the smartwatch for heart rhythm monitoring, consistent with prior studies [
In contrast to prior investigations, our study focused on establishing both the accuracy and usability of a scalable smartwatch-based approach to AF monitoring and screening among a population of older potential users during active periods intended to simulate ADLs. Not only will future studies need to be conducted to examine long-term adherence to smartwatches for AF monitoring among at-risk populations, further work to tune the AF detection algorithm for ideal performance using large, diverse study cohorts will be required. Our findings suggest, however, that older users, when provided support in learning to use smartwatches, can use them well, and that data derived from these devices are of sufficient quality so as to enable high quality rhythm analysis.
Our study has several strengths. First, in contrast to most mHealth studies involving younger participants, our study was able to enroll older participants at high risk for incident or recurrent AF. Second, in contrast to other mHealth studies that examine performance of signal processing algorithms from
Several limitations of our study warrant presentation. Our sample was enrolled from an ambulatory clinic at a single tertiary care medical center in central Massachusetts and was relatively homogenous with regard to race and gender. This racial and gender homogeneity significantly limits our ability to generalize these findings to members of other racial groups, community-dwelling individuals not under the care of a physician, or among individuals from other geographic areas. We plan to rectify this and enrich our study population by targeting a more diverse sample in future data collection. Furthermore, the sample was highly educated and reported a relatively high annual income, potentially limiting generalizability to less well-educated or poorer individuals. In addition, our sample was older, affected by a moderate to significant burden of physical and cognitive impairments, and the penetrance of smart device use was lower than the national average [
A novel, real-time realizable software algorithm analyzing pulse data from a smartwatch exhibits excellent performance for the detection of an irregular pulse consistent with AF among older individuals creating motion noise to simulate ADLs. Furthermore, the smartwatch system was deemed highly usable by older participants enrolled in our study, suggesting that long-term monitoring for AF using wrist-based mHealth devices holds promise. Future work is needed to assess provider impressions of the system, to validate findings from our study in much larger and more diverse cohorts, and examine long-term adherence to daily home use, as well as in-field accuracy of AF diagnosis among older individuals at risk for incident or recurrent AF.
Descriptive statistics of participant responses to smartwatch usability assessment.
activities of daily living
atrial fibrillation
American Heart Association
electrocardiogram
electronic health record
Food and Drug Administration
identification number
mobile health
motion noise artifact
National Institute of Health
normal sinus rhythm
premature atrial contraction
photoplethysmogram
premature ventricular contraction
root mean square of successive difference of RR intervals
sample entropy
Shannon Entropy
System Usability Scale
University of Connecticut
University of Massachusetts Medical School
The authors would like to thank all the patients who participated in the study, as well as the UMMS ambulatory cardiology clinic and its providers. ED is supported by National Institute of Health (NIH) grants 1T32GM107000-01 and 5T32HL120823-05. AS is supported by the National Center for Advancing Translational Sciences (TL1-TR001454) and the Eunice Kennedy Shriver National Institute of Child Health and Human Development (1F30HD091975-01A1). SAL is supported by NIH grant 1R01HL139731 and American Heart Association 18SFRN34250007. DDM is supported by NIH grants 5R01HL126911, 1R01HL137734, 1R01HL137794, 5R01HL135219-02, and 5UH3TR000921-04, as well National Science Foundation grant NSF-12-512. KC has a patent on the algorithm described in the paper. This work was funded in part by NIH grant 1R15HL121761, 1R01HL137734, as well as the office of Naval Research work unit N00014-12-1-0171. Smartwatches used in this study were provided by Samsung Electronics for investigator-initiated research. Samsung Electronics engineers were provided the paper presubmission, but the Samsung team had no input into the design or analysis plan, and the content in the paper was solely generated by the research team. This project was directly supported by the NIH grant 1R01HL137734.
ED was responsible for data analysis/interpretation, drafting majority of the paper, statistics, and data collection. DH, CW, SKB, and OA were responsible for data analysis/interpretation, drafting sections of the paper, critical revision of the paper, statistics, and data collection. AS, JS, TF, MM, and SL were responsible for data interpretation and critical revision of the paper. DL and BB were responsible for data analysis and statistics. HTH was responsible for critical revision of article and approval of the paper. KC and DDM were responsible for concept/design, drafting, critical revision and approval of the paper, and securing of grant funding.
DDM receives sponsored research support from Bristol Myers Squibb, Pfizer, Biotronik, and Boehringer Ingelheim. He has consulted for Bristol Myers Squibb, Pfizer, Samsung Electronics, and FlexCon and has inventor equity in Mobile Sense Technologies, LLC. KC is Chief Technology Officer and cofounder of Mobile Sense Technologies, LLC. SAL receives sponsored research support from Bristol Myers Squibb/Pfizer, Bayer HealthCare, and Boehringer Ingelheim, and has consulted for Abbott, Quest Diagnostics, Bristol Myers Squibb/Pfizer.