Association of Wearable Device Use With Pulse Rate and Health Care Use in Adults With Atrial Fibrillation | Atrial Fibrillation | JAMA Network Open | JAMA Network
[Skip to Navigation]
Sign In
Figure 1.  Study Population Selection Flowchart
Study Population Selection Flowchart

Study population selected from all patients with at least 1 International Classification of Diseases (ICD) code for atrial fibrillation (AF) from 2017 through 2019.

Figure 2.  Mean Pulse Rate and Composite Health Care Use During Follow-up Period
Mean Pulse Rate and Composite Health Care Use During Follow-up Period

A, For each patient, mean value of all pulse rates in the 90-day period following the index date was calculated. B, Composite use score included evaluation and management encounters, cardioversions, telephone encounters, and rate and rhythm control medications. bpm indicates beats per minute.

Table 1.  Patient Characteristics Prior to Exclusions and Matching
Patient Characteristics Prior to Exclusions and Matching
Table 2.  Patient Characteristics After Exclusions and Propensity Matching
Patient Characteristics After Exclusions and Propensity Matching
Table 3.  Outcomes of Matched Patients
Outcomes of Matched Patients
1.
Lu  T-C, Fu  C-M, Ma  MH, Fang  C-C, Turner  AM.  Healthcare applications of smart watches. a systematic review.   Appl Clin Inform. 2016;7(3):850-869. doi:10.4338/ACI-2016-03-R-0042PubMedGoogle ScholarCrossref
2.
Wu  M, Luo  J; Online Journal of Nursing Informatics Contributors. Wearable technology applications in healthcare: a literature review. Healthcare Information and Management Systems Society. Accessed August 15, 2020. https://www.himss.org/resources/wearable-technology-applications-healthcare-literature-review
3.
Gordon  WJ, Landman  A, Zhang  H, Bates  DW.  Beyond validation: getting health apps into clinical practice.   NPJ Digit Med. 2020;3:14. doi:10.1038/s41746-019-0212-zPubMedGoogle ScholarCrossref
4.
Device classification under section 513(f)(2)(de novo). FDA.gov. Accessed September 29, 2020. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN180044
5.
510(k) premarket notification. FDA.gov. Accessed September 29, 2020. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?id=K142743
6.
Giebel  GD, Gissel  C.  Accuracy of mHealth devices for atrial fibrillation screening: systematic review.   JMIR Mhealth Uhealth. 2019;7(6):e13641. doi:10.2196/13641PubMedGoogle Scholar
7.
Seshadri  DR, Bittel  B, Browsky  D,  et al.  Accuracy of the Apple Watch 4 to measure heart rate in patients with atrial fibrillation.   IEEE J Transl Eng Health Med. 2019;8:2700204. doi:10.1109/JTEHM.2019.2950397PubMedGoogle Scholar
8.
Al-Kaisey  AM, Koshy  AN, Ha  FJ,  et al.  Accuracy of wrist-worn heart rate monitors for rate control assessment in atrial fibrillation.   Int J Cardiol. 2020;300:161-164. doi:10.1016/j.ijcard.2019.11.120PubMedGoogle ScholarCrossref
9.
Digital Health Center of Excellence. FDA.gov. Accessed September 24, 2020. https://www.fda.gov/medical-devices/digital-health-center-excellence
10.
Carroll  AE. That new Apple Watch EKG feature: there are more downs than ups. The New York Times. October 8, 2018. Accessed March 8, 2021. https://www.nytimes.com/2018/10/08/upshot/apple-watch-heart-monitor-ekg.html
11.
Gonzalez  R. The new ECG Apple Watch could do more harm than good. Wired. September 13, 2018. Accessed March 8, 2021. https://www.wired.com/story/ecg-apple-watch/
12.
Husten  L, Darzi  A, Afeyan  N,  et al Apple Watch and atrial fibrillation detection: more harm than good? Stat. March 15, 2019. Accessed May 12, 2020. https://www.statnews.com/2019/03/15/apple-watch-atrial-fibrillation/
13.
Rosenbaum  PR, Rubin  DB. The central role of the propensity score in observational studies for causal effects.  Biometrika.1983;70(1):41-55. doi:10.1093/biomet/70.1.41Crossref
14.
Kind  AJH, Buckingham  WR.  Making Neighborhood-Disadvantage Metrics Accessible - The Neighborhood Atlas.   N Engl J Med. 2018;378(26):2456-2458. doi:10.1056/NEJMp1802313PubMedGoogle ScholarCrossref
15.
Quan  H, Li  B, Couris  CM,  et al.  Updating and validating the Charlson Comorbidity Index and score for risk adjustment in hospital discharge abstracts using data from 6 countries.   Am J Epidemiol. 2011;173(6):676-682. doi:10.1093/aje/kwq433PubMedGoogle ScholarCrossref
16.
Austin  PC.  The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments.   Stat Med. 2014;33(7):1242-1258. doi:10.1002/sim.5984PubMedGoogle ScholarCrossref
17.
Steinberg  BA, Kim  S, Thomas  L,  et al; Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT‐AF) Investigators and Patients.  Increased heart rate is associated with higher mortality in patients with atrial fibrillation (AF): results from the Outcomes Registry for Better Informed Treatment of AF (ORBIT-AF).   J Am Heart Assoc. 2015;4(9):e002031. doi:10.1161/JAHA.115.002031PubMedGoogle Scholar
18.
Bengtson  LGS, Lutsey  PL, Loehr  LR,  et al.  Impact of atrial fibrillation on healthcare utilization in the community: the Atherosclerosis Risk in Communities study.   J Am Heart Assoc. 2014;3(6):e001006. doi:10.1161/JAHA.114.001006PubMedGoogle Scholar
19.
Mulcahy  AW, Sorbero  ME, Mahmud  A,  et al Measuring health care utilization in Medicare Advantage encounter data: methods, estimates, and considerations for research. Rand.org. Accessed March 8, 2021. https://www.rand.org/content/dam/rand/pubs/research_reports/RR2600/RR2681/RAND_RR2681.pdf
20.
Eadicicco  L. The Apple Watch is still growing in popularity and dominating the competition, even while people are stuck at home. Business Insider. May 7, 2020. Accessed October 2, 2020. https://www.businessinsider.com/apple-watch-sales-numbers-strong-despite-coronavirus-2020-5
21.
Statt  N. Apple now sells more watches than the entire Swiss watch industry. The Verge. February 5, 2020. Accessed July 13, 2020. https://www.theverge.com/2020/2/5/21125565/apple-watch-sales-2019-swiss-watch-market-estimates-outsold
22.
Guo  Y, Lane  DA, Wang  L,  et al; mAF-App II Trial Investigators.  Mobile health technology to improve care for patients with atrial fibrillation.   J Am Coll Cardiol. 2020;75(13):1523-1534. doi:10.1016/j.jacc.2020.01.052PubMedGoogle ScholarCrossref
23.
Uittenbogaart  SB, Verbiest-van Gurp  N, Lucassen  WAM,  et al.  Opportunistic screening versus usual care for detection of atrial fibrillation in primary care: cluster randomised controlled trial.   BMJ. 2020;370:m3208. doi:10.1136/bmj.m3208PubMedGoogle ScholarCrossref
24.
Wyatt  KD, Poole  LR, Mullan  AF, Kopecky  SL, Heaton  HA.  Clinical evaluation and diagnostic yield following evaluation of abnormal pulse detected using Apple Watch.   J Am Med Inform Assoc. 2020;27(9):1359-1363. doi:10.1093/jamia/ocaa137PubMedGoogle ScholarCrossref
25.
Aggarwal  N, Ahmed  M, Basu  S,  et al Advancing artificial intelligence in health settings outside the hospital and clinic. NAM Perspectives. November 30, 2020. Accessed March 8, 2021. doi:10.31478/202011f
26.
Varma  N, Marrouche  NF, Aguinaga  L,  et al.  HRS/EHRA/APHRS/LAHRS/ACC/AHA worldwide practice update for telehealth and arrhythmia monitoring during and after a pandemic.   J Arrhythm. 2020. doi:10.1002/joa3.12389PubMedGoogle Scholar
Limit 200 characters
Limit 25 characters
Conflicts of Interest Disclosure

Identify all potential conflicts of interest that might be relevant to your comment.

Conflicts of interest comprise financial interests, activities, and relationships within the past 3 years including but not limited to employment, affiliation, grants or funding, consultancies, honoraria or payment, speaker's bureaus, stock ownership or options, expert testimony, royalties, donation of medical equipment, or patents planned, pending, or issued.

Err on the side of full disclosure.

If you have no conflicts of interest, check "No potential conflicts of interest" in the box below. The information will be posted with your response.

Not all submitted comments are published. Please see our commenting policy for details.

Limit 140 characters
Limit 3600 characters or approximately 600 words
    Original Investigation
    Cardiology
    May 27, 2021

    Association of Wearable Device Use With Pulse Rate and Health Care Use in Adults With Atrial Fibrillation

    Author Affiliations
    • 1Division of Cardiovascular Medicine, University of Utah School of Medicine, Salt Lake City
    • 2Department of Internal Medicine, University of Utah, Salt Lake City
    • 3Department of Epidemiology, Geisel School of Medicine at Dartmouth, Hanover, New Hampshire
    • 4Department of Biomedical Data Science, Geisel School of Medicine at Dartmouth, Hanover, New Hampshire
    • 5University of Colorado School of Medicine, Aurora
    • 6Study Design and Biostatistics Center, Center for Clinical and Translational Science, University of Utah, Salt Lake City
    • 7Department of Biomedical Informatics, Vanderbilt University Medical Center, Nashville, Tennessee
    • 8Department of Biostatistics, Vanderbilt University Medical Center, Nashville, Tennessee
    • 9Department of Medicine, Vanderbilt University Medical Center, Nashville, Tennessee
    • 10Geriatric Research Education and Clinical Center, Tennessee Valley Healthcare System, Nashville VA Medical Center, Nashville
    JAMA Netw Open. 2021;4(5):e215821. doi:10.1001/jamanetworkopen.2021.5821
    Key Points

    Question  Do individuals with known atrial fibrillation (AF) who use smartwatches or wearable devices use more health care resources and achieve better AF control?

    Findings  In this propensity-matched cohort study with 90-day follow-up among 16 320 patients with AF, individuals who used wearable devices had similar pulse rates compared with individuals who did not use these devices. Individuals using wearable devices had significantly increased rates of health care use; among use measures, the difference was significantly increased for ablation procedures.

    Meaning  These findings suggest that individuals with AF using wearable devices may have increased health care use but similar pulse rates on follow-up.

    Abstract

    Importance  Increasingly, individuals with atrial fibrillation (AF) use wearable devices (hereafter wearables) that measure pulse rate and detect arrhythmia. The associations of wearables with health outcomes and health care use are unknown.

    Objective  To characterize patients with AF who use wearables and compare pulse rate and health care use between individuals who use wearables and those who do not.

    Design, Setting, and Participants  This retrospective, propensity-matched cohort study included 90 days of follow-up of patients in a tertiary care, academic health system. Included patients were adults with at least 1 AF-specific International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code from 2017 through 2019. Electronic medical records were reviewed to identify 125 individuals who used wearables and had adequate pulse-rate follow-up who were then matched using propensity scores 4 to 1 with 500 individuals who did not use wearables. Data were analyzed from June 2020 through February 2021.

    Exposure  Using commercially available wearables with pulse rate or rhythm evaluation capabilities.

    Main Outcomes and Measures  Mean pulse rates from measures taken in the clinic or hospital and a composite health care use score were recorded. The composite outcome included evaluation and management, ablation, cardioversion, telephone encounters, and number of rate or rhythm control medication orders.

    Results  Among 16 320 patients with AF included in the analysis, 348 patients used wearables and 15 972 individuals did not use wearables. Prior to matching, patients using wearables were younger (mean [SD] age, 64.0 [13.0] years vs 70.0 [13.8] years; P < .001) and healthier (mean [SD] CHA2DS2-VASc [congestive heart failure, hypertension, age ≥ 65 years or 65-74 years, diabetes, prior stroke/transient ischemic attack, vascular disease, sex] score, 3.6 [2.0] vs 4.4 [2.0]; P < .001) compared with individuals not using wearables, with similar gender distribution (148 [42.5%] women vs 6722 women [42.1%]; P = .91). After matching, mean pulse rate was similar between 125 patients using wearables and 500 patients not using wearables (75.01 [95% CI, 72.74-77.27] vs 75.79 [95% CI, 74.68-76.90] beats per minute [bpm]; P = .54), whereas mean composite use score was higher among individuals using wearables (3.55 [95% CI, 3.31-3.80] vs 3.27 [95% CI, 3.14-3.40]; P = .04). Among measures in the composite outcome, there was a significant difference in use of ablation, occurring in 22 individuals who used wearables (17.6%) vs 37 individuals who did not use wearables (7.4%) (P = .001). In the regression analyses, mean composite use score was 0.28 points (95% CI, 0.01 to 0.56 points) higher among individuals using wearables compared with those not using wearables and mean pulse was similar, with a −0.79 bpm (95% CI −3.28 to 1.71 bpm) difference between the groups.

    Conclusions and Relevance  This study found that follow-up health care use among individuals with AF was increased among those who used wearables compared with those with similar pulse rates who did not use wearables. Given the increasing use of wearables by patients with AF, prospective, randomized, long-term evaluation of the associations of wearable technology with health outcomes and health care use is needed.

    Introduction

    Commercially available wearable devices (hereafter wearables) have flooded the market in recent years. Wearables are often marketed as health and wellness devices, with smartwatches and wristbands leading the wearable technology space. Hardware and software improvements have enabled additional features, such as heart rate monitoring, sleep monitoring, pulse oximetry, and respiration tracking. These advances open up new possibilities for disease-specific screening and management.1-3 Several devices cleared by the Food and Drug Administration (FDA) (eg, Apple Watch4 and AliveCor Kardia Band5) allow individuals to generate a single lead electrocardiogram (ECG) to screen for arrhythmias, such as atrial fibrillation (AF). Studies from 20196,7 and 20208 found that such devices were reasonably sensitive and specific in diagnosing new-onset AF, albeit among populations with high disease prevalence. The FDA recently reorganized the Center for Device and Radiological Health to reduce burdensome premarket approval processes. This is likely to increase the number of wearables available on the market.9

    As a result of this market evolution, wearables are commercially available and commonly used by patients. While wearables measure pulse rate and, in some cases, alert individuals for incident AF events,9 the associations of these devices with health care use and clinical outcomes among patients with known AF are unknown. A looming clinical concern is that wearables may lead to an excessive number of false positives or true positive but nonactionable alerts, increasing health care use—or worse, increasing treatment-associated complications—without improvement in morbidity or mortality.10-12 And despite a lack of FDA clearance for the use of these devices among individuals with known arrhythmias, individuals with AF are using wearables with pulse rate and ECG capabilities. The goal of this exploratory project was to characterize individuals with AF who used and sought care based on a wearable’s output and to compare subsequent pulse rate and health care use between individuals who used wearables and those who did not. We hypothesized that wearable use among patients with AF would be associated with increased health care use.

    Methods

    This cohort study was approved with waiver of informed consent granted by the institutional review board at the University of Utah because this was for retrospective research. This study is reported following the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline. The main objectives of our study were to compare the characteristics of individuals using wearables with those not using wearables and compare subsequent pulse rate and health care use patterns of individuals using wearables with those not using wearables.

    Patient Population and Exposures

    Data for this study were extracted from the enterprise data warehouse (EDW) at University of Utah Health. The EDW aggregates data from sources across the health system for use in operations and research efforts. Adult patients with at least 1 AF-specific International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) code (ie, I48.0, I48.1, I48.11, I48.19, I48.2, I48.20, or I48.21) from January 1, 2017, through December 31, 2019, comprised the candidate population for these analyses. Figure 1 depicts the cohort assembly process. For these candidate patients, clinical notes were extracted as text files, and search terms specific to wearables were identified using regular expression matching (eg, “kardia”; eTable 1 in the Supplement). This approach yielded 786 notes for 452 unique patients with at least 1 term specific to a wearable. If a relevant term was present in a note, the text from that note was populated into a Research Electronic Data Capture form and manually reviewed to identify individuals using wearables based on the description and context. Patients with AF who did not have a note that included a term specific to a wearable were included in the group of individuals who did not use wearables. This population was used to compare general characteristics of individuals who did and did not use wearables (ie, the study’s first objective).

    Initial Exclusions Prior to Propensity Matching

    We used a 2-stage process (ie, initial inclusion criteria followed by propensity matching) to control for confounding as rigorously as possible and to create the final cohort for comparison between individuals who did and did not use wearables (Figure 1). In the first step, initial criteria were applied to the nonusing candidate pool. Candidates who did not use wearables were included if they matched at least 1 individual who used wearables based on sex, year of birth (within 1 year), and index date, defined as follows. Candidates for the nonusing group had to have been seen for AF (defined by a documented relevant ICD code) within 60 days of at least 1 of the encounters by a member of the using group that referred to a wearable. Each candidate from the 2 groups was assigned an index date that served as day 0 in the analyses. For individuals using wearables, the index date was defined as the earliest date of the note that included a reference to a wearable within the study time frame. For individuals not using wearables, the index date was defined as the date of an encounter that included an AF-specific billing code and occurred within 60 days of an index date for an individual using wearables. If multiple dates were available, the earliest date during the 60-day interval was used.

    For both groups, we excluded patients whose index date occurred after September 30, 2019, and those with insufficient follow-up data during the 90-day outcome time period (ie, those with fewer than 3 unique dates with a recorded pulse rate). This approach allowed for a more rigorous comparison in this observational analysis because individuals who did and did not use wearables were seen at about the same time and all included patients had sufficient follow-up data.

    Propensity Matching After Initial Exclusions

    Individuals who did not use wearables were then matched to those who used wearables using propensity score matching at a ratio of 4 to 1.13 Specifically, we used a logistic regression model to estimate the propensity score, defined as the probability of being an individual who used wearables, as a function of variables that were potentially associated with choice of using a wearable for each individual in the cohort. Socioeconomic status is an important confounder in studies examining consumer technologies. We accounted for socioeconomic status by using the area deprivation index of the patient’s zip code within 90 days of the index date, with a higher area deprivation index indicating increased deprivation.14 The variables in the propensity score model included patient characteristics present at the time of and prior to the index date: age, sex, area deprivation index, index year, first visit for AF, number of days between index date and first AF visit in our system, baseline pulse, baseline pulse ≥ 110 beats per minute (bpm), history of ablation, history of cardioversion, CHA2DS2-VASc (congestive heart failure, hypertension, age ≥ 75 or 65-74 years, diabetes, prior stroke/transient ischemic attack, vascular disease, sex) score ≥ 2, Charlson Comorbidity Index score,15 mean pulse value from prior visits, oral anticoagulant use, rate control drug use, rhythm control drug use, number of prior evaluation and management (EM) codes, and number of prior telephone notes. The matching was performed without replacement based on the distance between each observation of an individual who used wearables and the observations in its neighborhood of individuals who did not use wearables, using the propensity score as a metric of distance. Standardized mean differences were used to assess covariate balance after matching (eTable 2 in the Supplement).16

    Outcomes

    The co-primary outcomes of interest were pulse rate and a composite health care use score, assessed over a 90-day period following the index date through December 31, 2019, to maintain a consistent follow-up time for all patients. Pulse rate was chosen as a surrogate for clinical outcome given that it is measured at almost all clinic visits and has prognostic value in AF.17 If multiple pulse rates were recorded on the same day for the same patient, the mean value for that day was used to avoid overcounting clinically active days (ie, procedures). Individuals who used wearables and those who did not use wearables were compared using the mean and median pulse rates from all outpatient-recorded pulse rates during the outcome period for each patient. Mean and median pulse rates were further divided into 5-unit intervals to approximate clinically relevant changes.

    The composite use score was designed to approximate health care use specific to AF management. Prior methods to estimate use have focused on inpatient hospital days or, broadly, outpatient claims for AF.18 We chose a more specific approach, incorporating health care actions that could be triggered by information from a wearable (eg, cardioversions and prescription orders). Similar to prior approaches, we accounted for professional services (ie, EM codes), but we did not account for inpatient hospitalizations.19 An ordinal variable was created from components of clinical encounters and AF-specific treatments, ranging from 0 to 8, with higher values indicating more use. We found that 38 out of 575 patients (6.6%) had scores of 6, 7, or 8, so these values were aggregated into a single value of 6. The components of the use outcome were: (1) Number of encounters with an EM or ablation Common Procedural Terminology (CPT) code. This number was categorized into 3 groups: 0 if no codes were present, 1 if the number of codes present was below the median of the group, and 2 if the number of codes present was above the median of the group. (2) Number of encounters with a cardioversion CPT code. This number was categorized as 0 if no codes were present and 1 if at least 1 code was present. (3) Number of telephone encounter notes (ie, 0 if no notes, 1 if the number of notes was below the median, and 2 if the number of notes was above the median). (4) Number of rate control medication orders (ie, 0 if no orders and 1 if at least 1 order). (5) Number of rhythm control medication orders (ie, 0 if no orders and 1 if at least 1 order).

    Patient Characteristics

    Prior conditions were identified from the presence of ICD billing codes prior to the index date. The EDW maintains an ICD code mapping to generate the Charlson Comorbidity Index score for each patient, using all diagnostic codes available in each patient's electronic medical record. We used CPT codes to identify EM visits (ie, codes 99201, 99202, 99203, 99204, 99205, 99212, 99213, 99214, 99215, 99281, 99282, 99283, 99284, 99285, 99304, 99305, 99306, 99307, 99308, 99309, 99310, 99324, 99325, 99326, 99327, 99328, 99334, 99335, 99336, 99337, 99341, 99342, 99343, 99344, 99345, 99347, 99348, 99349, and 99350), cardioversions (ie, code, 92960), and ablations (ie, codes, 93650, 93651, 93653, 93656, and 93657).

    Similarly, the EDW maintains an autocalculated CHA2DS2-VASc score for AF patients using demographic characteristics and historical ICD diagnoses billing codes. Each time an AF-specific ICD code appears in the electronic medical record, the system generates a CHA2DS2-VASc score for the corresponding date. We used the score closest in time to the index date as the baseline score for these analyses.

    Statistical Analysis

    Data cleaning, matching, and statistical analysis were performed using Python programming language version 3.8 (Python Software Foundation) and R statistical software version 3.6.0 (R Project for Statistical Computing). Characteristics of individuals who used and did not use wearables were compared using univariate logistic regression to generate odds ratios for binary variables and univariate linear regression to generate mean differences for continuous variables. Mixed-effects linear regression models were used to estimate the association between wearable use status and mean pulse rate, median pulse rate, and composite use score. The random component in the model was used to account for clustering structure associated with the matching procedure. P values were 2-sided, and statistical significance was set at .05. Data were analyzed from June 2020 through February 2021.

    Results
    Baseline Characteristics

    Among 16 320 patients identified with at least 1 AF-specific ICD diagnosis billing code from January 1, 2017, through December 31, 2019, 452 patients had at least 1 note with a reference to a wearable, with 851 device mentions among 786 notes; the study population consisted of 348 patients who were determined to have used wearables among patients with a wearable mention and 15 972 patients with at least 1 AF-specific ICD code and no device mentions. Apple watch was the most frequently mentioned device, with 240 mentions, followed by Fitbit, with 200 mentions, and Kardia, with 153 mentions (eFigure in the Supplement). Baseline characteristics for individuals using wearables compared with individuals not using wearables, prior to any exclusions or matching, are found in Table 1. The individuals who used wearables, compared with individuals who did not use wearables, were younger (mean [SD] age, 64.0 [13.0] years vs 70.0 [13.8] years; P < .001), were healthier (mean [SD] CHA2DS2-VASc score, 3.6 [2.0] vs 4.4 [2.0]; P < .001), had similar gender proportion (148 [42.5%] women vs 6722 women [42.1%]; P = .91), were less socioeconomically deprived (mean [SD] area deprivation index, 28.1 [18.0] vs 37.7 [19.2]; P < .001), and had a lower mean (SD) Charlson Comorbidity Index score (3.3 [3.1] vs 4.0 [3.2]; P < .001). Individuals who used wearables had greater mean (SD) EM encounters (7.2 [7.6] vs 4.4 [6.2]; P < .001) and median (interquartile range [IQR]) number of telephone notes (13.0 [3.0-39.0] vs 1.0 [0-7.0]; P < .001) prior to the index date compared with individuals who did not use wearables.

    After initial exclusions and propensity matching, 125 individuals who used wearables and 500 individuals who did not use wearables were compared using the index date for baseline characteristics (Table 2). Age, sex, rhythm control strategies, comorbidities, CHA2DS2-VASc score, number of ECGs, and health care use were statistically similar in the matched group. Individuals who used wearables had more median (IQR) electrocardiograms prior to the index date compared with those who did not use wearables (5 [2-11] vs 4 [1-9]; P = .02). Mean (SD) pulse prior to the index date was similar among those who did and did not use wearables (75.3 [11.5] bpm vs 76.2 [11.0] bpm; P = .42), as was mean (SD) pulse at the time of the index (75.3 [18.0] bpm vs 76.5 [15.2] bpm, P = .51).

    Follow-up Outcomes

    During the follow-up period, the mean pulse rate was similar between the 2 groups (75.01 bpm [95% CI, 72.74-77.27] vs 75.79 [95% CI, 74.68-76.90] bpm; P = .54) (Table 3, Figure 2A), as was the mean of all median pulse rates (73.68 [95% CI, 71.44-75.92] bpm vs 75.03 [95% CI, 73.86-76.20] bpm, P = .30) (Table 3; eTable 3 in the Supplement). Overall use was higher during the follow-up period for individuals who used wearables (Figure 2B), with increased mean (3.55 [95% CI, 3.31-3.80] vs 3.27 [95% CI, 3.14-3.40]; P = .04) composite use scores, while the median (IQR) composite use score was similar for the 2 groups (3.0 [3.0-4.0] vs 3.0 [2.0-4.0]; P = .08). Among the measures in the composite outcome, there was a significant difference in use of ablation, with at least 1 ablation CPT code found in 22 individuals who used wearables (17.6%) vs 37 individuals who did not use wearables (7.4%) (P < .001). The number of telephone notes, rate control and anti-arrhythmic drug prescriptions, and cardioversions were similar between the 2 groups (Table 3).

    In the regression analyses, which clusters based on propensity matching, the mean composite use score was 0.28 points (95% CI, 0.01 to 0.56 points) higher among individuals using wearables compared with those not using wearables. The mean pulse was similar, with a −0.79 bpm (95% CI, −3.28 to 1.71 bpm) difference, as was the mean of all median pulse rates, with a −1.35 bpm (95% CI, −3.95 to 1.24 bpm) difference. Because 1-unit differences in pulse may not be clinically meaningful, we also evaluated for 5-unit differences in pulse between the 2 groups and found no significant differences (eTable 3, eTable 4, and eTable 5 in the Supplement).

    Discussion

    Wearable sales are booming.20,21 Large randomized clinical trials published in 202022,23 focus on using such devices as screening tools among patients without arrhythmias. Our cohort study is among the first studies to systematically characterize the use of wearable use among individuals with AF as a part of routine health care delivery. Individuals who used wearables were younger, healthier, and socioeconomically better off than those who did not use wearables. We found no difference in clinic-measured pulse rates between the 2 groups but higher health care use rates among individuals who used wearables.

    Like our results, those of prior studies raised questions about effective ways to apply wearable technology to improve health outcomes. A 2020 retrospective analysis24 of Apple Watch alerts found that 11% of alerts were associated with a clinically actionable event. Another study of inpatients on telemetry7 found poor sensitivity of smartwatch AF detection among patients after cardiac surgical treatment, a group known to have a high arrhythmia burden. These findings, along with those reported in our study, support the need for more disease-specific data to guide the clinical use of wearables.

    Our observational analyses provide important information, given the absence of randomized trials in this population thus far. First, clinicians may be concerned about an excessive number of phone calls based on readings from a wearable; we found no evidence to suggest that pattern. Second, our data suggest that individuals who use wearables may be healthier than those who do not use wearables. This finding is unsurprising considering adoption patterns of new technology by younger patients, and it raises the possibility that subsequent clinical interventions based on wearables may not make patients feel better if they are already well. Third, most individuals in our study who used wearables had previously diagnosed AF. The wearable was not used for screening or diagnosis of AF, but rather for self-management; however, most of the published evidence focuses on these wearables as screening tools. These findings may have important implications for ongoing implementation of wearables into clinical care and their potential (or lack thereof) to improve personal health and health care delivery. Prospective, randomized clinical trials focused on wearable use among patients with AF are needed to fully understand the association of wearables with health outcomes.

    Wearables have potential use in longitudinal care of the arrhythmic population and will likely play increasing roles in clinical management.25 Indeed, the coronavirus disease 2019 pandemic has accelerated the shift in health care delivery away from traditional settings, with increasing use of telehealth clinical encounters.26 In addition to randomized studies, integration of wearables with electronic medical records could provide comprehensive capture of health outcomes and new ways to monitor treatment outcomes. For example, wearables could capture patient-reported symptoms and associate them with the electronic medical record, which could facilitate a better understanding of the association of wearables with health outcomes.

    Limitations

    This study has several limitations. The study included a single center, and patterns at other centers could differ. Wearable use rates, for example, could be lower in centers that serve poor or more ethnically diverse populations. The characteristics and AF severity among individuals using wearables would be different, in which would be associated with different outcome patterns. Multicenter studies, including randomized studies, are needed to ensure external validity. One cannot infer cause and effect from these exploratory, observational results; using electronic medical record data in this study, we cannot make inferences about the appropriateness of health care use, particularly because we could not reliably ascertain symptoms or why patients sought care.

    Exclusions and propensity score matching were used to limit confounding, but residual confounding is a limitation. Because wearables are commercially available, we could not classify patients using orders and instead relied on references to wearable-specific terms in record text. While the group classified as individuals who did not use wearables referred to patients without mention of a wearable, patients may have been misclassified as not using wearables if their clinical notes failed to mention devices. Assuming that use of wearables was associated with increased health care use, misclassification of individuals who used wearables as those who did not might lead to a reduction in the difference in pulse rate and use rate between the 2 groups. In other words, the misclassification could bias the results toward the null. Ultimately, a systematic method to capture individuals who use wearables will be needed. We were unable to ascertain if the specific wearables included rhythm classification software or were limited to pulse measurements. Additionally, composite health care use was defined a priori based on clinical relevance to AF; different definitions could be associated with different results.

    Conclusions

    Wearables are an increasing part of health care delivery, particularly in the fields of electrophysiology and cardiology. Many patients with or without AF use wearables for self-directed management or in conjunction with the health care delivery system, so it is important to understand the association of these devices with health outcomes and health care use. This study’s finding suggests that wearable use among patients with AF is associated with increased health care use and support the need for randomized clinical trials to measure the impact of wearables on health outcomes and use among patients with AF.

    Back to top
    Article Information

    Accepted for Publication: February 24, 2021.

    Published: May 27, 2021. doi:10.1001/jamanetworkopen.2021.5821

    Open Access: This is an open access article distributed under the terms of the CC-BY License. © 2021 Wang L et al. JAMA Network Open.

    Corresponding Author: Rashmee U. Shah, MD, MS, Division of Cardiovascular Medicine, University of Utah School of Medicine, 30 N 1900 E, Room 4A100, Salt Lake City, UT 84132 (rashmee.shah@utah.edu).

    Author Contributions: Drs Shah and Nielsen had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.

    Concept and design: Wang, Brown, Rumsfeld, Steinberg, Zhang, Matheny, Shah.

    Acquisition, analysis, or interpretation of data: Wang, Nielsen, Goldberg, Brown, Steinberg, Zhang, Shah.

    Drafting of the manuscript: Wang, Nielsen, Goldberg, Brown, Steinberg, Zhang, Shah.

    Critical revision of the manuscript for important intellectual content: Wang, Nielsen, Brown, Rumsfeld, Steinberg, Matheny, Shah.

    Statistical analysis: Wang, Nielsen, Brown, Zhang, Matheny, Shah.

    Obtained funding: Shah.

    Administrative, technical, or material support: Wang, Goldberg, Brown, Steinberg.

    Supervision: Wang, Brown, Steinberg, Shah.

    Conflict of Interest Disclosures: Dr Shah reported receiving a grant from the National Institutes of Health (NIH) National Heart Lung and Blood Institute, donations from Women as One and the Council for Educational Development and Research, and honoraria from the American College of Cardiology. Dr Steinberg reported receiving grants from Abbott, Boston Scientific, and Janssen; consulting fees from Janssen, AltaThera, Merit Medical, Bayer, and Crowley Fleck; and speaking fees from the North American Center for Continuing Medical Education (funded by Sanofi). Dr Rumsfeld reported serving as a member of the Apple Heart Study steering committee at Stanford University. No other disclosures were reported.

    Funding/Support: This work was supported by grant No. R03HL148372 from the National Heart Lung and Blood Institute and a grant from the University of Utah Study Design and Biostatistics Center, with funding in part from grant No. 8UL1TR000105 from the NIH National Center for Advancing Translational Sciences (formerly grant No. UL1RR025764 from the NIH National Center for Research Resources). Dr Steinberg was supported by grant No. K23HL143156 from the National Heart, Lung, and Blood Institute.

    Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

    Disclaimer: The content is solely the responsibility of the authors.

    References
    1.
    Lu  T-C, Fu  C-M, Ma  MH, Fang  C-C, Turner  AM.  Healthcare applications of smart watches. a systematic review.   Appl Clin Inform. 2016;7(3):850-869. doi:10.4338/ACI-2016-03-R-0042PubMedGoogle ScholarCrossref
    2.
    Wu  M, Luo  J; Online Journal of Nursing Informatics Contributors. Wearable technology applications in healthcare: a literature review. Healthcare Information and Management Systems Society. Accessed August 15, 2020. https://www.himss.org/resources/wearable-technology-applications-healthcare-literature-review
    3.
    Gordon  WJ, Landman  A, Zhang  H, Bates  DW.  Beyond validation: getting health apps into clinical practice.   NPJ Digit Med. 2020;3:14. doi:10.1038/s41746-019-0212-zPubMedGoogle ScholarCrossref
    4.
    Device classification under section 513(f)(2)(de novo). FDA.gov. Accessed September 29, 2020. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/denovo.cfm?id=DEN180044
    5.
    510(k) premarket notification. FDA.gov. Accessed September 29, 2020. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfpmn/pmn.cfm?id=K142743
    6.
    Giebel  GD, Gissel  C.  Accuracy of mHealth devices for atrial fibrillation screening: systematic review.   JMIR Mhealth Uhealth. 2019;7(6):e13641. doi:10.2196/13641PubMedGoogle Scholar
    7.
    Seshadri  DR, Bittel  B, Browsky  D,  et al.  Accuracy of the Apple Watch 4 to measure heart rate in patients with atrial fibrillation.   IEEE J Transl Eng Health Med. 2019;8:2700204. doi:10.1109/JTEHM.2019.2950397PubMedGoogle Scholar
    8.
    Al-Kaisey  AM, Koshy  AN, Ha  FJ,  et al.  Accuracy of wrist-worn heart rate monitors for rate control assessment in atrial fibrillation.   Int J Cardiol. 2020;300:161-164. doi:10.1016/j.ijcard.2019.11.120PubMedGoogle ScholarCrossref
    9.
    Digital Health Center of Excellence. FDA.gov. Accessed September 24, 2020. https://www.fda.gov/medical-devices/digital-health-center-excellence
    10.
    Carroll  AE. That new Apple Watch EKG feature: there are more downs than ups. The New York Times. October 8, 2018. Accessed March 8, 2021. https://www.nytimes.com/2018/10/08/upshot/apple-watch-heart-monitor-ekg.html
    11.
    Gonzalez  R. The new ECG Apple Watch could do more harm than good. Wired. September 13, 2018. Accessed March 8, 2021. https://www.wired.com/story/ecg-apple-watch/
    12.
    Husten  L, Darzi  A, Afeyan  N,  et al Apple Watch and atrial fibrillation detection: more harm than good? Stat. March 15, 2019. Accessed May 12, 2020. https://www.statnews.com/2019/03/15/apple-watch-atrial-fibrillation/
    13.
    Rosenbaum  PR, Rubin  DB. The central role of the propensity score in observational studies for causal effects.  Biometrika.1983;70(1):41-55. doi:10.1093/biomet/70.1.41Crossref
    14.
    Kind  AJH, Buckingham  WR.  Making Neighborhood-Disadvantage Metrics Accessible - The Neighborhood Atlas.   N Engl J Med. 2018;378(26):2456-2458. doi:10.1056/NEJMp1802313PubMedGoogle ScholarCrossref
    15.
    Quan  H, Li  B, Couris  CM,  et al.  Updating and validating the Charlson Comorbidity Index and score for risk adjustment in hospital discharge abstracts using data from 6 countries.   Am J Epidemiol. 2011;173(6):676-682. doi:10.1093/aje/kwq433PubMedGoogle ScholarCrossref
    16.
    Austin  PC.  The use of propensity score methods with survival or time-to-event outcomes: reporting measures of effect similar to those used in randomized experiments.   Stat Med. 2014;33(7):1242-1258. doi:10.1002/sim.5984PubMedGoogle ScholarCrossref
    17.
    Steinberg  BA, Kim  S, Thomas  L,  et al; Outcomes Registry for Better Informed Treatment of Atrial Fibrillation (ORBIT‐AF) Investigators and Patients.  Increased heart rate is associated with higher mortality in patients with atrial fibrillation (AF): results from the Outcomes Registry for Better Informed Treatment of AF (ORBIT-AF).   J Am Heart Assoc. 2015;4(9):e002031. doi:10.1161/JAHA.115.002031PubMedGoogle Scholar
    18.
    Bengtson  LGS, Lutsey  PL, Loehr  LR,  et al.  Impact of atrial fibrillation on healthcare utilization in the community: the Atherosclerosis Risk in Communities study.   J Am Heart Assoc. 2014;3(6):e001006. doi:10.1161/JAHA.114.001006PubMedGoogle Scholar
    19.
    Mulcahy  AW, Sorbero  ME, Mahmud  A,  et al Measuring health care utilization in Medicare Advantage encounter data: methods, estimates, and considerations for research. Rand.org. Accessed March 8, 2021. https://www.rand.org/content/dam/rand/pubs/research_reports/RR2600/RR2681/RAND_RR2681.pdf
    20.
    Eadicicco  L. The Apple Watch is still growing in popularity and dominating the competition, even while people are stuck at home. Business Insider. May 7, 2020. Accessed October 2, 2020. https://www.businessinsider.com/apple-watch-sales-numbers-strong-despite-coronavirus-2020-5
    21.
    Statt  N. Apple now sells more watches than the entire Swiss watch industry. The Verge. February 5, 2020. Accessed July 13, 2020. https://www.theverge.com/2020/2/5/21125565/apple-watch-sales-2019-swiss-watch-market-estimates-outsold
    22.
    Guo  Y, Lane  DA, Wang  L,  et al; mAF-App II Trial Investigators.  Mobile health technology to improve care for patients with atrial fibrillation.   J Am Coll Cardiol. 2020;75(13):1523-1534. doi:10.1016/j.jacc.2020.01.052PubMedGoogle ScholarCrossref
    23.
    Uittenbogaart  SB, Verbiest-van Gurp  N, Lucassen  WAM,  et al.  Opportunistic screening versus usual care for detection of atrial fibrillation in primary care: cluster randomised controlled trial.   BMJ. 2020;370:m3208. doi:10.1136/bmj.m3208PubMedGoogle ScholarCrossref
    24.
    Wyatt  KD, Poole  LR, Mullan  AF, Kopecky  SL, Heaton  HA.  Clinical evaluation and diagnostic yield following evaluation of abnormal pulse detected using Apple Watch.   J Am Med Inform Assoc. 2020;27(9):1359-1363. doi:10.1093/jamia/ocaa137PubMedGoogle ScholarCrossref
    25.
    Aggarwal  N, Ahmed  M, Basu  S,  et al Advancing artificial intelligence in health settings outside the hospital and clinic. NAM Perspectives. November 30, 2020. Accessed March 8, 2021. doi:10.31478/202011f
    26.
    Varma  N, Marrouche  NF, Aguinaga  L,  et al.  HRS/EHRA/APHRS/LAHRS/ACC/AHA worldwide practice update for telehealth and arrhythmia monitoring during and after a pandemic.   J Arrhythm. 2020. doi:10.1002/joa3.12389PubMedGoogle Scholar
    ×