http://orcid.org/0000-0001-7858-1923
Abstract
Background: The smartphone apps provide a user-friendly option for measurement of heart rate (HR) by detecting pulsatile photoplethysmographic signals with built-in cameras from the fingertips, however, the validation study is limited.
Methods: We compared HR detected by the smartphone apps (App1 = Instant HR, App2 = Cardiio: HR Monitor and App3 = Runtastic HR Monitor) with simultaneous standard ECG monitoring in the adult patients at the critical care unit.
Results: HR measurements were obtained from 140 patients with mean age 67.6 ± 15.3 years. Mean baseline HR was 89.1 ± 19.1 bpm (range, 32–136 bpm). Sinus rhythm was presented in 111 patients (79.3%), atrial fibrillation in 25 patients (17.9%), pacemaker rhythm in 3 patients (2.1%), and high-grade AV block in 1 patient (0.7%). The ECG-derived HR correlated well with App1 (r = 0.98), App2 (r = 0.97), and App3 (r = 0.92). In patients with regular rhythm, mean absolute deviation was 0.8 ± 1, 0.7 ± 0.9, 1.0 ± 1.3 bpm on App1, App2 and App3, respectively. In the patients with irregular rhythm, median absolute deviation (IQR) was 3 (2–5.5), 4 (1.5–11.5), and 6 (2–13) bpm. Skin colour did not affect with the HR measurement.
Conclusions: HR measurements from all applications were correlated well with ECG monitoring. However, it was less accurate in case of irregular rhythm such as atrial fibrillation.
Several reports on inaccuracy of mobile health apps have been published. We conducted the validation study in the real patients by using popular mobile apps.
Heart rate measurements from mobile apps were correlated well with standard ECG. The accuracy of HR from apps was worse at irregular rate and tachycardia.
Key messages
1. Introduction
Heart rhythm disorders or arrhythmias are the conditions that the heartbeat is irregular, too fast, or too slow. A heart rate (HR) above 100 beats per minute (bpm) is called tachycardia and a HR less than 60 bpm is called bradycardia. Symptoms may include palpitation, light-headedness, syncope, dyspnoea on exertion or chest pain. However, patients with these serious arrhythmias may have no symptoms at all. Some cardiac arrhythmias are hazardous to the patient and require prompt diagnosis and treatment. It can predispose a patient to serious complications such as stroke, heart failure or cardiac arrest. Moreover, episode of arrhythmias can be transient and terminated spontaneously without treatment, particularly in its early stage [Citation1]. The 12-lead electrocardiogram (ECG) is essential to identify a patient with a suspected arrhythmia. Ambulatory ECG monitoring is more sensitive for detecting occult cardiac arrhythmias and may also be used to evaluate the efficacy of therapy or for prognostic purposes. However, these monitors cannot be used without a healthcare professional. Palpating pulses or listening to the heart sounds may be challenging for patients who are not trained in these medical skills. There is a need for a convenient method which could allow patients or caregivers to accurately measure HR outside of the hospital setting.
The use of technology is increasing day by day, smartphones become essential devices in every aspect of daily life. Not only used for communication, but there are also many applications, including entertainment, education, daily convenience as well as health assessment. Mobile health (mHealth) is generally referred to the mobile communication technologies to health and medical care. The mHealth has been created to help people manage their medical conditions at the home or ambulatory settings. A percentage of mHealth smartphone apps have rapidly grown. Unfortunately, the reports on inaccuracy of mHealth apps have been published [Citation2–4]. In addition, there is a lack of scientific evidence of mHealth apps in the literature [Citation5]. Only a few applications have undergone rigorous studies [Citation6]. Until such information is available, clinicians may be hesitant to recommend any application to their patients. Although this mHealth may be seemingly innocuous, research is needed to provide data to support the health impacts of applications and how we can best adapt them in our clinical practice to promote better patient health [Citation7]. The mHealth technologies need to be safe, accurate and reliable.
The mHealth monitoring technologies, including applications using the built-in camera for HR measurement, are widely available, inexpensive and user-friendly. These technologies used to measured HR, primarily for exercise and fitness. As HR monitoring is an important tool for screening HR disorders and guiding to adjust medications. The validation of HR measurements detected from smartphone apps is limited. The pre-existing validation studies of HR measurement showed a good correlation between mobile application and standard methods [Citation8]. However, the majority of previous studies conducted in a small number (less than 30) of healthy volunteers that had sinus rhythm [Citation8]. In addition, pigments in human skin could affect the performance of pulse detection [Citation9]. Testing of the application in larger and ethnically diverse cohorts is necessary. Therefore, this study aims to evaluate the accuracy of HR measurement from smartphone applications in the real-patient situations.
2. Materials and methods
2.1. Study population
This study was a diagnostic test study with cross-sectional design and prospective data collection. During 2016–2017, we randomly recruited adult patients who were at least 15 years old and required HR monitoring at the critical care unit of Ramathibodi Hospital. The inclusion criteria were patients who underwent the standard continuous ECG monitoring. The exclusion criteria were patients whose fingertip pulsation cannot be obtained such as amputation, limb ischaemia, infection, contact precaution and patients who had hemodynamic instability. All patients provided their written informed consent to participate in the study. The study was approved by the committee on human rights related to research involving human subjects, Faculty of Medicine Ramathibodi Hospital, Mahidol University.
2.2. Devices and measurement
2.2.1. Applications
The top 3 lists of free applications for HR measurement by used search term “heart rate monitor” were downloaded from the App Store (https://itunes.apple.com/store/). Instant Heart Rate (App1) (Azumio Inc., USA), Cardiio: Heart Rate Monitor (App2) (Cardiio, Inc., USA), and Runtastic Heart Rate Monitor (App3) (Runtastic, Inc., Austria) were used in the present study.
All three applications for HR measurement work via the technology of contact photoplethysmography (PPG). PPG is a simple and low-cost optical biomonitoring technique used to noninvasively measure the pulsatile changes of blood volume that appear in the microvascular bed under the skin [Citation10]. PPG is based on the principle that blood absorbs light more than the surrounding tissues. The PPG technique has become a popular non-invasive method for extracting physiological measurement such as HR and blood oxygen saturation. The contact PPG technique works by using the built-in white light-emitting diode (LED) flashlight and rear camera of the smartphone to illuminating the skin and detecting the light absorption by arterial pulsation at the fingertip pulp, respectively (). By the detection of pulsatile signal, pulse peak can be interpreted as an R wave with high accuracy [Citation11].
Figure 1. For use the contact photoplethysmography application, (A), the patient’s index finger placed to cover on the flashlight and rear camera of the smartphone simultaneously. When the pulse rate showed on the smartphone screen consistently, we recorded the HR displayed on the smartphone and electrocardiography monitor at the same time. (B) is the example of a report produced by the Instant Heart Rate (Azumio Inc., USA). In this study, we used alcohol cotton ball for disinfecting the patient’s finger and smartphone before and after each measurement.
2.2.2. Mobile devices
We tested the three different applications on iPhone 5s (Apple Inc., USA). All the HR measurements were performed by the same researcher (W.P.) throughout this study.
2.2.3. Process of HR measurement
The patient’s palmar side of the index finger was placed over the flashlight and the camera lens. Patients were instructed to place a fingertip on the camera lens without pressing down and to hold the finger still to diminish any motion artefacts. The HR measured by the mobile application () was a simultaneous standard continuous ECG monitoring which connected on the patient’s chest wall with leads and showed the results on the bedside monitors. This validated approach is based on the previous study [Citation12]. If the application fails the measurement of HR, we will try to repeat the measurement. If it persistently fails to measure 3 times, then it will be recorded as the application failure.
The baseline characteristics (such as sex, age, body temperature, blood pressure, respiratory rate, skin colour, oxygen saturation, haemoglobin, clinical status, underlying heart disease, illness conditions, medications) were collected. Skin colour at the pulp of the index finger was measured as melanin index by using portable dermaspectrometer (DSM II Colorimeter, Cortex Technology, Denmark). The device was calibrated prior to use against a manufacturer-provided white calibration plate. The colourimeter’s narrow-band reflectance system utilises two high-intensity white LEDs to reflect light onto a 7-mm2 target area of the patients’ skin. An RGB (Red–Green–Blue) sensor determines the intensities of the red, green and blue lights being reflected back, which were used to compute specific skin indices including the melanin index. The measurement of skin colour was performed by the same researcher (S.H.).
2.3. Statistical analysis
Continuous data were described as mean ± standard deviation (SD) or median and interquartile range (IQR) as appropriate. Categorical variables were expressed as percentages. The chi-squared test was used to compare the accuracy of the HR measurement between different groups. We used the Pearson’s correlation analysis and Bland–Altman analysis to compare the HR from the three applications and ECG monitoring. To assess the level of agreement between PPG-estimated HR and standard ECG, we calculated the mean absolute deviation. According to the American National Standard Institute/Association for the Advancement of Medical Instrumentation EC-13 standard, the HR accuracy defined by the difference of HR is within 5 beats per minute (bpm). Linear and multiple regression analysis were used to define the factors that influenced the accuracy of the application. A p-value of less than .05 was considered statistically significant. Statistical analyses were performed using SPSS Statistics for Windows, version 17.0 (SPSS Inc., Chicago, IL).
3. Results
We randomly recruited 140 patients who were admitted to the medical intensive care unit. A total of 20 HR measurements (App1, n = 4; App2, n = 6; App3, n = 10) were excluded from the study because applications failed to perform HR detection for more than 3 times. Finally, 400 measurements from PPG-based applications were included in the analysis.
3.1. Baseline characteristics
The details of baseline characteristics are shown for the 140 participants in .
Table 1. Baseline characteristics of the patients (n = 140).
3.2. The accuracy of application-based HR measurement
The accuracy of HR measured by applications compared to standard ECG, reported as mean absolute deviation (in bpm ± SD) or median absolute deviation (IQR) as appropriate. Overall, the median differences of HR measurement from applications and standard continuous ECG monitoring were 1 (0–1 bpm) in App1; 1 (0–1.25) bpm in App2, and 1(0–5) bpm in App3.
In patients with regular rhythm (n = 110), mean absolute deviation (in bpm) was 0.8 ± 1 in App1; 0.7 ± 0.9 in App2 and 1.0 ± 1.3 in App 3. In the patients with irregular rhythm (n = 30), median absolute deviation (IQR) was 3 (2–5.5) in App1; 4 (1.5–11.5) in App2, and 6 (2–13) bpm in App3.
Standard ECG-derived HR correlated well with mobile applications, App1 (r = 0.98, p-value <.001), App2 (r = 0.97, p-value<.001), and App3 (r = 0.92, p-value <.001). All applications measurement tend to correlate well with standard ECG when patients were within normal HR. HR from mobile applications were not correlated well in patients with tachycardia (App1, r = 0.28, p-value = .001; App2, r = 0.29, p-value = .001, and App3, r = 0.20, p-value = .02). Furthermore, App3 was not correlated well in patients with tachycardia as well as bradycardia (). HR measurement from mobile applications had wider 95% level of agreement (1.96 standard deviation of the difference) when patients had irregular rhythm, as illustrated in the Bland–Altman plots ().
Figure 2. Bland-Altman plots of limits of agreement in heart rate (HR) measurement by tested applications (App1, App2, and App3) compared to standard electrocardiogram (ECG). In patients with regular heart rate (A–C), the accuracy of all applications was excellent. However, the accuracy was less when measured in the patients with irregular and higher heart rates (D–F). The 95% level of agreement is equal to 1.96 standard deviation of the difference.
Table 2. This table showed the percentage of the accuracy of heart rate (HR) from mobile applications in the settings of normal rate (60-100 bpm), bradycardia (<60 bpm), and tachycardia (>100 bpm). The accuracy of HR was defined by the percentage of the mobile application performance that can accurately measure HR within ±5 bpm in comparison with the standard electrocardiogram.
The percentage of accuracy based on the mean absolute deviation of HR derived from 3 applications compared with standard ECG less than 5 bpm was 89.3%. Fifteen patients with inaccuracy had median absolute deviation of 10.2 (5.75–16) bpm and all were diagnosed with atrial fibrillation.
Due to many zero responses (the difference of HR is equal to zero), the Pearson correlation was performed instead of the regression analysis. The factors that can significantly influence the accuracy of HR measurement were irregular heart rhythm (rApp1=0.53, p-value<.001; rApp2=0.61, p-value<.001; rApp3=0.54, p-value<.001) as well as tachy- and bradycardia (rApp1=0.28, p-value<.001; rApp2=.29, p-value<.001; rApp3=.20, p-value<.001). Skin colour, haemoglobin and other baseline characteristics (age, body temperature, respiratory rate, systolic blood pressure, oxygen saturation, cooperative, did not associate with accuracy of HR measurement (p = NS).
4. Discussion
Since the modern lifestyle is a mobile lifestyle, it has the great potential to introduce mobile healthcare for promoting a healthier lifestyle in the near future. Furthermore, smartphone-based healthcare technology could easily extend beyond self-monitoring and could also be used to convey information to health care professionals. In the present study, we showed the data in the real patients that HR measurements from all three mobile applications were correlated very well with standard ECG monitoring. Our results were similar with prior meta-analysis of 14 validated studies, but this study conducted in the larger sample size of cardiac patients included patients with implanted pacemaker [Citation8]. The meta-analysis concluded that HR-derived from mobile applications via PPG signal was correlated well with a validated method in resting sinus rhythm with HR between 60 and 100 bpm [Citation8]. Our data also indicated that no difference in accuracy in the Southeast Asian patients with dark skin pigmentation.
However, the performance of different mobile applications for HR measurements was very heterogeneous. The accuracy of HR from applications compared with standard ECG was significantly worse at irregular rate and tachycardia, in particular with patients with atrial fibrillation with rapid ventricular rate. It is not surprising because the measure PPG waveform comprises a pulsatile physiological waveform that reflects cardiac synchronous changes in the blood volume with each heartbeat [Citation13]. The pulse deficit, the difference between the HR counted by auscultation and the pulse rate at the radial artery, is commonly found in patients with atrial fibrillation. The irregularities of the peripheral pulse rate and pulse deficit explain why the measurement of HR by applications is not accurate [Citation14]. Accuracy of HR detection was decreased in the patients with HR ≥120 bpm or supraventricular tachycardia as previously reported in the paediatric population [Citation15,Citation16]. During tachycardia, short RR intervals do not allow sufficient time for left ventricular diastolic filling, causing in a low stroke volume and may affect the inaccuracy of mobile applications to detection of heart rate. Fortunately, many developers produced technologies to easily detection of atrial fibrillation and other tachyarrhythmias [Citation17–19]. Another drawback of these applications is the very short periods (less than 10 seconds) of HR detection. During atrial fibrillation, the 60 s count is the most accurate technique in measurement of HR [Citation14]. The newly developed application needs longer time about 2 min to analyze the heart rhythm and rate [Citation17]. Thus, these 3 mobile applications may not be the suitable option for detection of atrial fibrillation. Novel technological advances in mobile applications, smartphones and connected wearable devices could be a solution. Recent study showed the Kardia Band (AliveCor, Mountain View, California) coupled with an automated atrial fibrillation detection algorithm using smartwatch technology, e.g. Apple Watch (Apple, Cupertino, California) can help screen patients with atrial fibrillation by recording a rhythm strip and transmit this result to a patient’s treating physician [Citation20]. The generalizability of these applications warrants further investigation to define the clinical utility for arrhythmia detection.
The possibility of dark skin phototype interfering HR measurement has been raised concern. The accuracy of contact PPG requires a good signal quality which may be influenced by the amount of melanin pigment within the human epidermis [Citation9]. It is well known that the ability of light absorption of melanin pigment reduces the light penetration through the skin and may cause inaccurate measurement results [Citation9,Citation21]. The majority of subjects in the present study were dark skin type (phototype IV–V) which provided an average melanin index within a range of Asians (40.2 ± 4.4) [Citation22,Citation23]. This was different from the previous contact PPG validated studies that were conducted in Caucasians, light skin-colour population (phototype I–III). However, our study revealed that high melanin pigment-containing skin did not affect the accuracy of HR measurement. Therefore, the novel technology detecting PPG signals by a multi-wavelength sensor reveals its effectiveness in various skin phototypes [Citation24]. Moreover, measurement of HR by these contact PPG applications mostly affected by patients’ co-operation, finger movement, moist or dry crackle skin, and cold temperature of fingertips [Citation13]. Another drawback, free applications have occasionally an advertisement pop-up while we are measuring the HR. This can prevent convenient efficiency measurement.
The strengths of our study are the measurement occurred in the real-patient situation. However, this can be transformed to the limitations because all the subjects were in hospital setting, not in the ambulatory outpatient settings. The study population that had irregular HR or tachy- and bradycardia was in the small numbers. We used only iPhone5s, which may not apply to other smartphones or application platforms (e.g. Android). Newer smartphones and applications may have better features to measure HR. In addition, it is still unclear whether the same accuracy of these applications could be attained in an unsupervised condition.
The mHealth is a promising technology due to its inexpensive, non-invasiveness, portable and simplicity. It has potential for early screening and self-monitoring for various diseases in the primary care settings. Mobile technologies are profoundly transforming the clinical practice and the way medical decisions are made. However, there is a need for more research into a full understanding of the diagnostic value of the different features. Importantly, it is caveat emptor when decided on applications for widely performing in general population.
5. Conclusions
HR measurements from mobile applications were correlated well with standard ECG monitoring. These smartphone applications could be applied for diagnosis and monitoring in the patients. However, it was less accurate in patients who had atrial fibrillation with rapid rate. Novel technologies could be the solution to help for screening patients with arrhythmia.
Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.
Acknowledgements
We thank Umaporn Udomtrupayakul, PhD for her assistance in statistical analysis.
Disclosure statement
No potential conflict of interest was reported by the authors.
References
- Humphries KH, Kerr CR, Connolly SJ, et al. New-onset atrial fibrillation: sex differences in presentation, treatment, and outcome. Circulation. 2001;103:2365–2370.
- Wolf JA, Moreau JF, Akilov O, et al. Diagnostic inaccuracy of smartphone applications for melanoma detection. JAMA Dermatol. 2013;149:422–426.
- Plante TB, Urrea B, MacFarlane ZT, et al. Validation of the instant blood pressure smartphone app. JAMA Intern Med. 2016;176:700–702.
- Murakami H, Kawakami R, Nakae S, et al. Accuracy of wearable devices for estimating total energy expenditure: comparison with metabolic chamber and doubly labeled water method. JAMA Intern Med. 2016;176:702–703.
- Buechi R, Faes L, Bachmann LM, et al. Evidence assessing the diagnostic performance of medical smartphone apps: a systematic review and exploratory meta-analysis. BMJ Open. 2017;7:e018280.
- Burke LE, Ma J, Azar KM, et al. Current science on consumer use of mobile health for cardiovascular disease prevention: a scientific statement from the American Heart Association. Circulation. 2015;132:1157–1213.
- Eapen ZJ, Turakhia MP, McConnell MV, et al. Defining a mobile health roadmap for cardiovascular health and disease. J Am Heart Assoc. 2016;5:e003119.
- De Ridder B, Van Rompaey B, Kampen JK, et al. Smartphone apps using photoplethysmography for heart rate monitoring: meta-analysis [review]. JMIR Cardio. 2018;2:e4.
- Fallow BA, Tarumi T, Tanaka H. Influence of skin type and wavelength on light wave reflectance. J Clin Monit Comput. 2013;27:313–317.
- Allen J. Photoplethysmography and its application in clinical physiological measurement. Physiol Meas. 2007;28:R1–39.
- Scully CG, Lee J, Meyer J, et al. Physiological parameter monitoring from optical recordings with a mobile phone. IEEE Trans Bio-Med Eng. 2012;59:303–306.
- Vandenberk T, Stans J, Van Schelvergem G, et al. Clinical validation of heart rate apps: mixed-methods evaluation study. JMIR mHealth Uhealth. 2017;5:e129.
- Sun Y, Thakor N. Photoplethysmography revisited: from contact to noncontact, from point to imaging. IEEE Trans Biomed Eng. 2016;63:463–477.
- Sneed NV, Hollerbach AD. Accuracy of heart rate assessment in atrial fibrillation. Heart Lung. 1992;21:427–433.
- Ho CL, Fu YC, Lin MC, et al. Smartphone applications (apps) for heart rate measurement in children: comparison with electrocardiography monitor. Pediatr Cardiol. 2014;35:726–731.
- Wackel P, Beerman L, West L, et al. Tachycardia detection using smartphone applications in pediatric patients. Jour Pediatr. 2014;164:1133–1135.
- Mc MD, Chong JW, Soni A, et al. PULSE-SMART: pulse-based arrhythmia discrimination using a novel smartphone application. J Cardiovasc Electrophysiol. 2016;27:51–57.
- Tarakji KG, Wazni OM, Callahan T, et al. Using a novel wireless system for monitoring patients after the atrial fibrillation ablation procedure: the iTransmit study. Heart Rhythm. 2015;12:554–559.
- Chan PH, Wong CK, Poh YC, et al. Diagnostic performance of a smartphone-based photoplethysmographic application for atrial fibrillation screening in a primary care setting. J Am Heart Assoc. 2016;5:e003428.
- Bumgarner JM, Lambert CT, Hussein AA, et al. Smartwatch algorithm for automated detection of atrial fibrillation. J Am Coll Cardiol. 2018;71:2381–2388.
- Zonios G, Bykowski J, Kollias N. Skin melanin, hemoglobin, and light scattering properties can be quantitatively assessed in vivo using diffuse reflectance spectroscopy. J Invest Dermatol. 2001;117:1452–1457.
- Fitzpatrick TB. The validity and practicality of sun-reactive skin types I through VI. Arch Dermatol. 1988;124:869–871.
- Ash C, Town G, Bjerring P, et al. Evaluation of a novel skin tone meter and the correlation between Fitzpatrick skin type and skin color. Photonics Lasers Med. 2015;4:177–186.
- Yan L, Hu S, Alzahrani A, et al. A multi-wavelength opto-electronic patch sensor to effectively detect physiological changes against human skin types. Biosensors. 2017;7:22.