Introduction
Monitoring of heart rate (HR) and Heart Rate Variability (HRV) has been extensively investigated for many years in healthy subjects, in general population as well as in various diseases \cite{24110126,Malik_1993,8598068}. HRV describes the fluctuation of instantaneous heart rate over time providing information on two quantities, the HR and its variability. As HRV is highly correlated to the temporal autonomic nervous system (ANS) function, it has been shown that increased HRV is an indication of high level of fitness, wellbeing and good health whereas decreased have shown predictive role of HRV for mortality, morbidity and adverse outcomes \cite{22851700}.
Additionally, decreased HRV is associated with fatigue, stress, and even burnout state\cite{von_Rosenberg_2017,Jim_nez_Morgan_2017}.
Photoplethysmography (PPG) is a simple and low cost method used to detect volumetric changes in blood in peripheral circulation at the skin surface \cite{17322588,2649304}.
In recent years, several wearable pulse rate monitors using PPG technology have been developed and are widely available. These small, robust and user friendly devices have sensors that reliably monitor minor changes in the intensity of light transmitted through or reflected from the human tissues which is emitted from high intensity light emitting diodes (LEDs). Although they have obvious advantages over the classical ambulatory ECG recording, the fact that they use PPG, i.e. a totally different detecting approach raises the question of the accuracy of their results when compared to the gold standard of ECG method. Therefore, the question rises if wearable PPG devices provide a reliable and precise measurement of HRV parameters in rest as well as during exercise.
Aim
The purpose of this systematic review is to collectively examine studies that compare ECG derived RR and HRV with that of the wearable commercially available devices.
Search Strategy
This systematic review was conducted by searching medical literature in MEDLINE and SCOPUS, guided by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement \cite{19631508}in conjunction with the PRISMA explanation and elaboration document \cite{19631507}. The last search was conducted in April 2017. All the retrieved article titles and abstracts were screened for relevant manuscripts. A full text review of the selected relevant articles was made in order to detect the studies included in this systematic review. Relevant full text review manuscripts or systematic review manuscripts were used to retrieve articles of any publishing date from their reference list and manually include them to this systematic review.
Additionally we performed a google search for wearable devices available on the market during the last 5 years claiming that they can measure via plethysmography HR. The results were reviewed for devices able to measure HRV. The devices list (Table 1) was used as an additional retrograde search tool for any relevant studies through manufacturers’ commercial websites.
Research papers found were filtered according to the inclusion and exclusion criteria of this systematic review.
Medical Subject Heading (MeSH) terms and text words were used based on the following search strategy:
Group A terms: (hrv) OR heart rate variability
Group B terms: ((((smart) OR smartphone$) OR wearable$) OR phone$ OR plethysmography OR photoplethysmography OR impedance)
Group C terms: ((((((holter) OR continuous ecg) OR continuous electrocardiogram) OR continuous electrocardiography) OR ambulatory ecg) OR ambulatory electrocardiogram OR ambulatory electrocardiography.
Group A, group B, group C, were combined and humans’ studies and english language limits were applied.
Inclusion - exclusion criteria
Of the articles retrieved through the above described search strategy only those that met the following criteria were included to this systematic review:
1. Studies related to human subjects were included. Studies not related to human subjects were excluded.
2. Studies reported in full text English language were included. Studies having only their title / abstract reported in English language were excluded.
3. Studies on HRV were included. Studies on heart rate and/or blood pressure and/or any other variable that did not include HRV were excluded.
4. Studies on HRV detection and/or analysis and/or interpretation and/or filtering were included.
5. Studies that compare ecg/holter recordings of HRV with any other wearable HRV detection and capture method(s) were included.
6. Original papers were only included. Review papers that met the rest of the inclusion / exclusion criteria were not included; however, their reference lists were used to retrieve any relevant studies.
7. Studies that did not use commercially available hardware were excluded.
Two of the authors (AUT1 and AUT3) independently applied the above described search strategy to retrieve articles. Also, two of the authors (AUT1 and AUT2) independently screened the retrieved articles. Any disagreements were resolved by a third author (AUT3) and a final decision was made accordingly.
Results
The search strategy with the limits applied retrieved 57 articles from MEDLINE and 269 from SCOPUS. After duplicates were removed it yielded 308 articles. Of these articles 272 were excluded according to the predefined criteria through title and abstract screening. There were 36 articles selected for full text review. Thirty of them were also excluded as not relevant (n=28) or no comparison papers (n=2). The remaining 6 articles were included to the study. Additionally, 15 relevant articles were identified from the reference list of the reviews or from manufacturers’ commercial websites and added to the study. Finally, 21 articles were included to this systematic review. The flow diagram of the selection process is shown in Figure 1.
The characteristics of the included studies are presented in Table 2.
\cite{27708585}
\cite{23366214}
\cite{24511344}
\cite{16672842}
\cite{17614016}
\cite{28269383}
\cite{26708360}
\cite{23751411}
\cite{27749728}
\cite{19292626}
\cite{15643533}
\cite{28289645}
\cite{18427831}
\cite{19092682}
\cite{25685174}
\cite{28290720}
\cite{24756693}
\cite{18853042}
\cite{Vasconcellos_2015}
\cite{21766225}
\cite{20225081}
The total number of the subjects involved was 725, with one study having 339 subjects examined (Wallen et al.). All 21 articles have examined healthy subjects except two; Lee et al \cite{Lee_2017} studied myocardial infarction (MI) adults and Vasconcellos et al \cite{Vasconcellos_2015} studied obese adolescents. Also, one study \cite{17614016} examined children. In the vast majority of the articles (n=12) a wrist device was used, chest device was used in 4 studies and 6 articles used finger sensors. Four of the studies integrated a similar to ECG - HRV technology and the rest 17 of them integrated PPG - PRV technology.
In all articles, the primary end point was the comparison between R-R interval measurements, while several HRV indices were calculated as R-R derivatives.
Rest
All 21 studies examined as baseline the HRV at rest, having 725 participants in total. Overall, correlation was very good to excellent ranging from 0.85 to 0.99. Primarily, the RR interval correlation was ranging from 0.91 to 0.999. Additive to these, in 2 studies \cite{16672842,17614016} the error rate in detection of R waves was evaluated and found to range between 0.28 and 0.4% which was estimated as an accepted ratio. Regarding time domain indices of HRV, correlations were ranging from 0.98 to 0.99 while in frequency domain, the correlation was found to range from 0.85 to 0.94. Finally, in the non-linear dynamics analysis, the correlation was found to be > 0.9.
Exercise
Ten out of 21 studies used an exercise protocol, in a total of 150 subjects. All of them presented that although there was an excellent correlation at rest, this seemed to decrease up to 0,85 as the level of exercise and/ or motion increased.
Overall, RR interval correlation was moderate to excellent ranging from 0,786 to 1. Regarding time domain HRV parameters correlation was found to range from 0,786 (especially in the peak of the exercise) to 1(resting and 1st phase of the exercise).Similar pattern occurred in the frequency domain HRV parameters where the correlation was found to range from 0,8 to 1. Also, RR to PP wave correlation was ranging from 0.8 to 0.998.
Discussion
In recent years tech advances are used in order to capture HRV through wearable devices in daily activities. The accuracy of these devices versus classical methods like ECG is still under evaluation.
This systematic review aims to present the burden of relative literature and reveal their controversies and limitations as well as to provide possible explanations for these findings.
It is important to realize that the basic difference among photoplethysmography and electrocardiogram is the captured signal per se. Heart’s electric activity is depicted by ECG, whereas the PPG is a mechanical signal measuring the propagation of the peripheral pulse wave. Therefore, the time needed to propagate the pulse pressure (PP) wave to the distal arterioles is called Pulse Transit Time (PTT) and measures the time elapsed between the R-wave of QRS complex in the ECG and the arrival point to PPG device \cite{27278521,22809539}. Several studies showed that PTT seems to be a surrogate marker of ANS \cite{11556663} and that it dependents on the properties of the pulse wave velocity, the vascular path from the heart to the location of the detector and it is negatively correlated with blood pressure, arterial stiffness, and age \cite{17322588}.
HRV fundamentals
HRV is a widely available, accurate and reproducible dynamic noninvasive technique widely used as a quantitative assessment tool of the autonomic nervous system (ANS) function \cite{24804906,24602150,24215747}. Studies have shown that reduced HRV assessed by the RR interval analysis has been associated with increased cardiovascular morbidity and mortality in patients with various diseases and in general population \cite{Maheshwari_2016,Fyfe_Johnson_2016,Au_yeung_2015}. Upon this predictive value, heart rate and HRV analysis have been used for estimation of mental stress and athlete’s fitness levels, fatigue and overload.
The quantification of ANS function is being able due to calculation of several HRV parameters according to time-domain, frequency-domain, and nonlinear analysis of consecutive sinus origin R-R intervals (RRI) \cite{Ben_Amor_2012,24110126}. These indices represent different components of the sympathetic and/or parasympathetic system of the ANS. For instance the high frequency (HF) component derived by the frequency domain analysis, represents the parasympathetic activity, and the low frequency/HF ratio represents the balance of sympathetic to parasympathetic activity \cite{Georgiou_2017}.
These HRV indices are dependent to the quality of the recording, the condition of the patient during the recording, the exclusion of artifacts, the detection of arrhythmic beats and the duration of recording ranging from seconds (Short Term HRV) to even days (Long Term). There are indices like Root Mean Square of Standard Deviation of R-R interval that can be calculated from 10 sec recording time and others that need more than one hour. Also, recording during rest favors parasympathetic indices augmentation while diseased state, motion and exercise increases sympathetic tone thus decreases heart rate variability.
Although the classical ECG or ambulatory Holter represents the gold standard method for RR interval monitoring and analysis, several drawbacks regarding the proper and accurate capture of RR interval exist: Patients with tremor and elderly with fragile skin have bad quality of recordings with a lot of noise and artifacts due to the inconvenience caused from the presence of too many electrodes attached to their body. Also, other factors such as surface electromyography, increased impedance of the electrode body, respiration induced baseline drift, and electrode contact movement can cause noise and motion artifacts. Another problem is the use of a gel electrolyte which can stimulate the skin causing a reaction and/or rash. Additionally, morphological variations in the ECG waveform and heterogeneity in the QRS complex often make difficult the identification of RR interval. Another limitation can be the presence of specialized technician/doctor which increases the cost and decreases the wide applicability \cite{19848857}.
PPG fundamentals
Photoplethysmography (PPG), a cheap, simple and widespread technology has been used as an alternative approach to obtain HRV indices \cite{17322588}. The PPG based devices have a sensor that uses infrared emitter and a detector integrated to a comfortable to wear probe in stable places of the body with rich microcirculation. Thus the synchronous to the heartbeat blood volume changes in the microvascular bed can be traced without the inconvenience of installation of neither any electrode nor the examinee to be undressed \cite{24110126}. The simplicity of the technique, cost-effectiveness, easy signal acquisition and remote monitoring are the main and obvious advantages of the PPG versus the gold standard ECG. Therefore PPG is often used in conditions for measurements where mobility, simplicity, time efficiency, flexibility, and low cost are of paramount importance, e.g., in exercise monitoring, every day motion, monitoring of the elderly, or at disabled patients etc \cite{17322588,22809539}.
Recently, a lot of latest technology devices, such as smartphones \cite{23751411} or wearable devices \cite{17614016}, based on PPG are used in clinical research for assessing HRV as an alternative to ECG monitoring \cite{15643533,16672842,19140912,20225081}.
In the relevant literature of the wearable PPG devices the terms ‘‘heart rate’’ and ‘‘pulse rate’’ are frequently used interchangeably, along with the derived from PPG term “pulse rate variability” (PRV), has been introduced as a potential analog of HRV \cite{26511754,19864707,19728090,18663635,17987395,16259639}.
As PRV is further processed identically as HRV, the same derived parameters can be extracted from both methods such as SDNN which denotes the standard deviation of normal to normal R–R intervals (NN), where R is the peak of a QRS complex in an ECG recording, root mean square of successive difference between adjacent NN intervals (RMSSD), proportion of NN50 in total NN intervals (pNN50), low frequency (LF) power, high frequency (HF) power and LF to HF ratio (LF/HF).
A major drawback in PPG derived HRV is the inability to distinguish arrhythmias and ectopic beats which doesn’t occur in the gold standard method. Thus in a study using the Polar RS800 with appropriate software (PPT 5) system to recognize, correct and eliminate non-valid cases, the PPT5 failed to recognize 18 of 21 recordings with ectopic beats \cite{21766225}.
Wavelength used
PPG uses low intensity infrared (IR) or green light which are more strongly absorbed by blood than the surrounding tissues \cite{8564151}. It has been shown that 530 nm light (green) PPG showed higher accuracy of pulse rate detection than the 645 (red) and 470 (blue) nm light for monitoring HR \cite{24110039,19163152}. However there is only one study with 12 participants measuring the accuracy of pulse rate (PR) during motion \cite{24110039}.
Measurement site
A flat skin surface with rich microvasculature is required to attach firmly the PPG sensor in order to obtain an accurate measurement .Thus usual measurement sites for wearable PPGs are the finger, forehead, ear, wrist and chest. For obvious reasons most of the wearables are placed in the wrist or the chest. The ear is chosen because it is a natural anchoring point, and it is discrete since the device can be partially hidden by hair \cite{23366488}. As mentioned in the results section, most of the reviewed articles used the wrist as a measuring site.
Parameters involved in PPG measurements
Several parameters have to be considered when PPG measurements have to be interpreted:
1. Motion artifact
Since remote measuring sites are more susceptible to motion artifacts \cite{24608066,20389499}and the contact force between the site and the sensor as PPG is vulnerable to motion-induced artifacts \cite{18301578} special attention must be addressed during data PPG acquisition to eliminate them \cite{19728090,17322588}.
2. Respiration
Since respiration alters the intrathoracic pressure and causes blood flow variations in both the veins \cite{8665771} as well as in the arteries \cite{9336733}, the DC component of the PPG waveform shows minor changes with respiration \cite{26390439}.
Thus it has been shown that the short-term variability (rMSSD, SD1 and HF) and LF/HF agreement between PRV and HRV deteriorates as it is vulnerable to respiratory changes \cite{26668634}.
3. Age, gender
Normal HRV values for various age and gender subsets are not still available in the literature. It is well known that the elderly have increased arterial stiffness, which leads to faster pulse transmission in the periphery \cite{12420195}. Thus, pulse transit time (PTT) differences observed between HRV and PRV, could be dependent on aging \cite{20702919}. In the reviewed articles age was not mentioned in 2 studies. The vast majority of the rest (except 2 studies, one with 16 participants of mean age 55.94±-9,05 and the second of mean age 60,8±-5,76 with 12 subjects) involved young population (mean range from 20,9 to 39,2 years).
Regarding gender influence, one study showed that the validity of the Polar RS800–PPT 5 system to measure HRV at rest was age and gender dependent:
Women showed lower correlation with ECG than men which was further deteriorating in older women \cite{21766225}. In our review the gender of 696/725 participants was available, 272 of them being females and 424 males.
Therefore as it is not clear so far if different and/or gender can play a role, further studies should investigate different samples with large prospective population studies with longitudinal follow-up \cite{27278521,8319331}.
Additionally, it has been suggested that mental stress can increase atrial stiffness which in turn alters pulse wave velocity \cite{18273040}.
It is also suggested that ambient temperature could induce a difference in the short-term variables that reflect the parasympathetic activity between HRV and PRV \cite{26511754}.
4. Software analysis
There is a couple of proprietary available software systems for collection and analysis of PPG data like the PPT5 used with Polar devices or the ithleteTM software application which in tandem with a miniature HR receiver plugged into the headset jack of an iPhone or iPod touch collects the HR signal from a chest strap wirelessly. Afterwards data filtration and analysis can be performed by, e.g., the PPT5 software. Another option is to use freely available software eg.Kubios \cite{26708360,24054542,23751411}.
However, automated filtering which exclude some intervals from the original R–R sequence, should be avoided and instead manual editing should be preferred according to the ‘‘Task Force’’ 1996 \cite{8598068}, which states that automatic filters are known to be unreliable and may potentially introduce errors \cite{2606116}.
In our review only 6 studies used automated analysis whether the rest used both manual and automated analysis
5. Statistical analysis
The Bland–Altman plot should be used to compare the agreement among a new measurement technique with a gold standard, as even a gold standard does not imply to be without error. It allows the identification of any systematic difference between the measurements \cite{2868172}.
In our review only 17 studies used this technique, while 4 studies did not applied the Bland–Altman analysis and therefore only the correlation, but not agreement among the two methods, could be determined from these publications.
6. Sampling rate
The sampling rate is a matter of difference between the two approaches. Sampling rate of PPG is usually 20 Hz much less than that of ECG which is 125 to 250 Hz. This obviously implies smaller events detection ability of the PPG devices.
Comparison of PPG vs. ECG for HRV measurement (συζήτηση για εμας? )
There are several studies examining the correlation among HRV and PRV with inconclusive results \cite{22350367,20702919}. This may be due to different experimental settings or to the absence of standardization of the methods of analysis used \cite{26668634}. It is also worth to note that the disagreement between the two methods does not apply to the same extent to all HRV parameters. Additionally, since PPG is susceptible to motion artifacts, this lack of maintaining consistent methodology in most studies, suggests that the accuracy of PRV obtained from PPG should be interpreted with caution \cite{22809539}.
Rest
Our search revealed that most of the comparison studies have been performed in stationary conditions generally revealed that PRV is a good surrogate of HRV. This is in line with other studies that at sitting, resting position the agreement between HRV and PRV was mostly acceptable \cite{17987395,17282607,12415532,7554427,8021000}.
However, when the position of a subject changes from supine to upright even in resting conditions a PRV divergence from HRV is appearing due to the fact that PRV reflects the mechanical coupling between respiration and thoracic vasculature which is stronger at a standing than at a supine position \cite{1928405}.
It must also be noted that HRV indices derived through PPG data are very sensitive to different conditions including noise, artifacts, stature, atherosclerosis, location of sensor and sampling rate. It is probably due to these reasons that some studies comparing the two methods found differences in both healthy normal subjects \cite{25570719,24111246,22809539,22350367,20702919,17987395,17322588,12212675} as well as in patients \cite{22350367,20980188,19864707,10491338}.
Exercise (Εδώ θέλει δουλειά)
There are many non-stationary situations where autonomic balance significantly changes like in stress, during motion or exercise the use of PRV as a surrogate measurement of HRV is questionable \cite{20702919}. It is obvious that this is an area where wearable devices could be more useful than at rest.
A moderate agreement is observed when exercising or having mental stress, due to increased noise production, contraction of muscles that are in contact with the sensors, sweating, increased intrathoracic pressure which alters the venous return, increased peripheral vasoconstriction and the respiratory effort during exercise \cite{7503271,8301336,866577}.
Depending of the wearable device, different problems arise concerning exercise and HRV. For instance, Lee et al showed that HR detected by camera based wearable at intense level of exercise could not be measured accurately during the exercise, due to difficulty of keeping the fingertip accurately on the camera, as the hand shaking. Also, Ghamari et al \cite{28269383} showed that during walking, serious motion artifacts caused by the participant’s hand movement which would influence the peak detection algorithm ineffective. A disadvantage of chest band wearable device during intense exercise was the feeling of discomfort that a subject sensed as the chest expands with deep breathing.
Hong et al mentioned that the correlation coefficient of R-R interval variables between the ECG using Bioshirt and the conventional ECG was very good indicating that R-peak detecting capabilities of these two devices were largely similar however as the level of exercise was increasing, the correlation was decreasing due to artifact production.
Hernando et al although they had an agreement between the detected R-peaks, and the RR intervals from the ECG, it was observed that as the intensity level was increased, the discrepancy from Bland-Altman plot increases from 1.67% of the RR pairs out of the limits of agreement (at rest) to 4.80% in the last interval (I100). They conclude that they have good correlation in some of the indices but not all of the HRV. This occurs because there is a disagreement of the relative error of the Polar derived High Frequency with that of the ECG as the level of the exercise is increasing. This might be partly due to the augmentation of the noise level as the exercise increases and partly due to the influence of the respiratory effort and depth that becomes more intense. Both of them influence enormous the High Frequency. Akintola et al in his work showed enormous amount of artifacts during daily activities which might limit the use of the wearable device.
In agreement with all the above, Kingsley et al showed that the limits of agreement were increased as the exercise was intensified. Also, the HRV indices in absolute power were reducing as the level of exercise was augmenting implying influence of adrenergic input, respiratory effort and unreliable algorithm detection and recording RR ability.
Regretfully most of the findings of our review showed that the correlation was fading out as the level of exercise and/ or motion increased. Indeed, Lee et al have noticed that HR measurements using a smartphone were lower than control measurements as the level of exercise was increased, because the application algorithm was cancelled in the process of filtering inaccurate signals or that deficits occurred in the PPG input, because fingers were not well attached to the camera because of movement during exercise. In addition, we assumed that HR measurements using a smartphone would be higher than the control HR measurements, because vibrations from exercise were transmitted to the camera via a finger and perceived by the PPG as HR changes, resulting in the addition of signal input. They conclude that it is essential to develop advanced devices with higher accuracy, that allow the cameras can keep contact with the skin even during higher exercise intensity. Algorithms also need to be improved to better reflect errors from movement during exercise.
In contrary to the negative results that all the above studies have shown during intense exercise, there are other groups that reported an overall stronger agreement \cite{25570719,24111246}.
Also, some other research groups reported that PPG yielded higher HRV values \cite{22809539,10491338}. However, all these studies involved only a few subjects.
Our common belief is that technology needs to be developed to better detect and identify errors that may occur during exercises of higher intensity.
Some research groups reported that PPG yielded higher HRV values \cite{22809539,10491338}. However, these studies involved only a few subjects.
In summary, PRV may be used as an acceptable surrogate for HRV signal, provided that subjects exhibiting either very high or very low HRV are avoided \cite{10491338}.
Future directions
As wearable healthcare technology as well as the research of light propagation in human tissues is progressing, it is expected that applications of PPG should expand. For instance, there is a growing interest to remotely depict PPG through imaging such as contactless video-photoplethysmography (vPPG) \cite{Lewandowska_2012,Zaunseder_2014,Blackford_2015,25700104,25150821,24681430,23744659,23111602,20952328,19104573} or Imaging PPG (IPPG) \cite{Tuchin_2015,27681456,26390439}.
Additionally, as the availability of wearable devices is expanding, more research is obviously warranted to establish age- and sex- dependent normal PRV values as well as to standardize both acquisition protocols and analytical methods in order to get reliable and accurate results, thus permitting these methods to become a valid surrogate for HRV parameters.
Conclusion
Our systematic review revealed that wearable devices using PPG may provide a promising alternative solution for measuring HRV. However it is evident that more robust studies in non stationary conditions are needed in terms of number of subjects involved, acquisition and analysis techniques implied. Therefore so far wearable devices using PPG can only be recommended as a surrogate for HRV at resting position as their accuracy fades out as exercise load increases. Thus they need to be used with caution in such conditions.
ACKNOWLEDGMENT None
LEGENDS
AUTHOR BIOGRAPHY:
Konstantinos Georgiou MD, is a surgery resident and a PhD candidate at the Medical School, National and Kapodistrian University of Athens, Greece. He is focusing in stress estimation in surgery as well as in medical simulators. Email:
[email protected]