ArticlePDF Available

Validation of the Apple Watch for Heart Rate Variability Measurements during Relax and Mental Stress in Healthy Subjects

Authors:

Abstract and Figures

Heart rate variability (HRV) analysis is a noninvasive tool widely used to assess autonomic nervous system state. The market for wearable devices that measure the heart rate has grown exponentially, as well as their potential use for healthcare and wellbeing applications. Still, there is a lack of validation of these devices. In particular, this work aims to validate the Apple Watch in terms of HRV derived from the RR interval series provided by the device, both in temporal (HRM (mean heart rate), SDNN, RMSSD and pNN50) and frequency (low and high frequency powers, LF and HF) domain. For this purpose, a database of 20 healthy volunteers subjected to relax and a mild cognitive stress was used. First, RR interval series provided by Apple Watch were validated using as reference the RR interval series provided by a Polar H7 using Bland-Altman plots and reliability and agreement coefficients. Then, HRV parameters derived from both RR interval series were compared and their ability to identify autonomic nervous system (ANS) response to mild cognitive stress was studied. Apple Watch measurements presented very good reliability and agreement (>0.9). RR interval series provided by Apple Watch contain gaps due to missing RR interval values (on average, 5 gaps per recording, lasting 6.5 s per gap). Temporal HRV indices were not significantly affected by the gaps. However, they produced a significant decrease in the LF and HF power. Despite these differences, HRV indices derived from the Apple Watch RR interval series were able to reflect changes induced by a mild mental stress, showing a significant decrease of HF power as well as RMSSD in stress with respect to relax, suggesting the potential use of HRV measurements derived from Apple Watch for stress monitoring.
Content may be subject to copyright.
sensors
Article
Validation of the Apple Watch for Heart Rate
Variability Measurements during Relax and Mental
Stress in Healthy Subjects
David Hernando 1, 2, *ID , Surya Roca 3, Jorge Sancho 3ID ,Álvaro Alesanco 3ID
and Raquel Bailón1,2
1Biomedical Signal Interpretation & Computational Simulation (BSICoS) Group, Aragón Institute of
Engineering Research (I3A), IIS Aragón, University of Zaragoza, 50018 Zaragoza, Spain; rbailon@unizar.es
2Centro de Investigación Biomédica en Red en Bioingeniería, Biomateriales y Nanomedicina (CIBER-BBN),
28029 Madrid, Spain
3
Communications Networks and Information Technologies (CeNIT) Group, Aragón Institute of Engineering
Research (I3A), University of Zaragoza, 50018 Zaragoza, Spain; surya@unizar.es (S.R.);
jslarraz@unizar.es (J.S.); alesanco@unizar.es (Á.A.)
*Correspondence: dhernand@unizar.es; Tel.: +34-876-555462
Received: 4 July 2018; Accepted: 8 August 2018; Published: 10 August 2018


Abstract:
Heart rate variability (HRV) analysis is a noninvasive tool widely used to assess autonomic
nervous system state. The market for wearable devices that measure the heart rate has grown
exponentially, as well as their potential use for healthcare and wellbeing applications. Still, there is a
lack of validation of these devices. In particular, this work aims to validate the Apple Watch in terms
of HRV derived from the RR interval series provided by the device, both in temporal (HRM (mean
heart rate), SDNN, RMSSD and pNN50) and frequency (low and high frequency powers, LF and HF)
domain. For this purpose, a database of 20 healthy volunteers subjected to relax and a mild cognitive
stress was used. First, RR interval series provided by Apple Watch were validated using as reference
the RR interval series provided by a Polar H7 using Bland-Altman plots and reliability and agreement
coefficients. Then, HRV parameters derived from both RR interval series were compared and their
ability to identify autonomic nervous system (ANS) response to mild cognitive stress was studied.
Apple Watch measurements presented very good reliability and agreement (>0.9). RR interval series
provided by Apple Watch contain gaps due to missing RR interval values (on average, 5 gaps per
recording, lasting 6.5 s per gap). Temporal HRV indices were not significantly affected by the gaps.
However, they produced a significant decrease in the LF and HF power. Despite these differences,
HRV indices derived from the Apple Watch RR interval series were able to reflect changes induced
by a mild mental stress, showing a significant decrease of HF power as well as RMSSD in stress with
respect to relax, suggesting the potential use of HRV measurements derived from Apple Watch for
stress monitoring.
Keywords:
heart rate variability; wearable device; Apple Watch; RR series; validation; ANS assessment
1. Introduction
Heart rate variability (HRV) analysis is a widely accepted tool for the noninvasive assessment of
autonomic nervous system (ANS) [
1
]. Its analysis and use are becoming increasingly common, since it
is sensitive to both physiological and psychological changes [
2
]. Altered HRV measurements have
been reported in several diseases related to ANS dysregulation, including cardiovascular diseases,
such as ischemia, myocardial infarction and heart failure [
1
,
3
], metabolic diseases, such as diabetes and
Sensors 2018,18, 2619; doi:10.3390/s18082619 www.mdpi.com/journal/sensors
Sensors 2018,18, 2619 2 of 11
obesity [
3
], and mental disorders, such as anxiety and depression [
4
,
5
]. Moreover, HRV measurements
have also been used to monitor sleep [6], stress [7,8], drowsiness [9] and exercise training [1012].
Over the past 10 years, the market for health related wearable devices has grown exponentially,
as well as their potential use for healthcare and wellbeing applications. The most popular and
accepted ones, especially among the non-patient population, are the watch-like devices, which
usually provide average heart rate (HR) estimates, derived from the pulse signal recorded at the
wrist. They are based on photophethysmography (PPG), including a light source for illuminating
the skin and a photodetector, which measures the intensity of the reflected light. HR estimates are
based on the pulsatile changes in reflected light induced by fluctuations in blood flow every heartbeat.
Although technically feasible, only a small subset of these devices allows HRV measures. As pointed
out in [
13
], these devices usually lack from proper validation, limiting its applicability for enhancing
clinical care.
Different wearable devices have been evaluated and compared for HR estimation. In [
13
] the
Apple Watch achieved the best performance estimating HR at a one minute granularity, with median
error below 3% during sitting, walking, running, and cycling. However, neither of these devices has
been validated for HRV estimation. Although pulse rate variability (PRV) derived from PPG has been
validated as a surrogate of HRV derived from the electrocardiogram (ECG) [
14
], it is not the case for
PRV derived from the wrist. Only a few studies have addressed this issue, such as [
15
], where wrist
PPG-based HRV was evaluated for Emphatica E4 (Empatica Srl, Milano, Italy) and PulseOn (PulseOn
Oy, Espoo, Finland) devices, yielding reliable results only during sitting condition.
The objective of this study is to validate the Apple Watch for HRV measurement. Prior studies have
only validated aggregate HR estimates [
13
,
16
19
], but its outperformance over other devices [
13
,
20
]
makes it a potential candidate among wrist-worn devices for HRV measurement. Moreover, its spread
and acceptance are beating the wearable market [
21
]. The present study sets out the comparison of RR
intervals derived from Apple Watch using as reference the RR intervals derived from the Polar H7 chest
belt heart rate monitor (Polar Electro Oy, Kempele, Finland), which has been already validated [
22
,
23
].
Additionally, this study includes the validation of HRV measurements, derived from the RR intervals
from both devices, in two scenarios with different ANS balance: relax and metal stress.
2. Materials and Methods
2.1. Experimental Setup
A total of 20 volunteers agreed to participate in the study. All of them were apparently healthy
subjects. Written informed consent was obtained from each subject. The study protocol was approved
by the institutional ethics committee and was in accordance with the Declaration of Helsinki for
Human Research of 1974 (last modified in 2008).
Participants were asked to abstain from caffeine and smoking prior to the test and to only consume
a light meal 2 h prior to testing. The Polar H7 chest belt was placed tightly just below the chest muscles
and the Apple Watch was placed tightly as well in the left wrist. Both devices were cleaned up between
uses to avoid possible misreading effects. Measures were taken while participants were sitting in front
of a 21” computer screen, completing relax and mental stress test protocols. Instructions were given to
minimize left arm movements, where the Apple Watch was placed.
The experimental protocol consisted of two consecutive 5-min stages, namely relax and stress
stages, separated by a 1-min break. During the relax stage volunteers watched a relaxing video,
consisting in relaxing music and pleasant images, which was previously used in a study on emotion
recognition [
24
]. During the stress stage, participants performed an online version of the Stroop test
(www.onlinestrooptest.com), which is an attentional test in which participants see words (names of
different colors) written in colored ink and they are asked to name the color of the ink while ignoring
the meaning of the word.
Sensors 2018,18, 2619 3 of 11
Raw RR data was obtained from Polar H7 (
RRH7
) using Elite HRV app installed in an Android
device. This app is able to export a text file containing the raw RR interval series with a precision
of milliseconds. On the other hand, the Breath app was used to extract RR interval series from
Apple Watch (
RRAW
). Breath app, developed by Apple, is the only way at this moment (watchOS
4.2 operating system) to obtain RR raw values since Apple does not include any programing method
for developers to directly access the values. This app stores the raw RR values, with a precision of
centiseconds, in the user’s Personal Health Record, accessible to be exported in XML format using
Apple’s Health App.
2.2. Synchronization and RR Matching
Figure 1shows an example of both RR series,
RRH7
and
RRAW
. It can be seen that the RR series
derived from the Apple Watch has some gaps, probably corresponding to time instants where Apple
Watch proprietary algorithms have not been able to reliably detect the PPG pulses. These gaps were
present in almost all Apple Watch recordings, with no apparent relation with the subject or the test
stage (relax or stress). Synchronization between
RRH7
and
RRAW
was performed using the delay
that maximized their cross correlation using the first 20 RR intervals, where no gaps appeared in
Apple Watch recordings. All RR intervals need to be matched from both devices: we create
RRH7g
by
intentionally creating the same gaps in
RRH7
than those in
RRAW
. This way, both
RRAW
and
RRH7g
have the same number of samples and all RR intervals are matched. For each recording, the percentage
of missing RR intervals missed with respect to the total number of RR intervals were noted.
Sensors 2018, 18, x 3 of 11
Raw RR data was obtained from Polar H7 () using Elite HRV app installed in an Android
device. This app is able to export a text file containing the raw RR interval series with a precision of
milliseconds. On the other hand, the Breath app was used to extract RR interval series from Apple
Watch ( ). Breath app, developed by Apple, is the only way at this moment (watchOS 4.2
operating system) to obtain RR raw values since Apple does not include any programing method for
developers to directly access the values. This app stores the raw RR values, with a precision of
centiseconds, in the user’s Personal Health Record, accessible to be exported in XML format using
Apple’s Health App.
2.2. Synchronization and RR Matching
Figure 1 shows an example of both RR series,  and . It can be seen that the RR series
derived from the Apple Watch has some gaps, probably corresponding to time instants where
Apple Watch proprietary algorithms have not been able to reliably detect the PPG pulses. These
gaps were present in almost all Apple Watch recordings, with no apparent relation with the subject
or the test stage (relax or stress). Synchronization between  and  was performed using
the delay that maximized their cross correlation using the first 20 RR intervals, where no gaps
appeared in Apple Watch recordings. All RR intervals need to be matched from both devices: we
create  by intentionally creating the same gaps in  than those in . This way, both
 and  have the same number of samples and all RR intervals are matched. For each
recording, the percentage of missing RR intervals missed with respect to the total number of RR
intervals were noted.
Figure 1. Example of the RR series for a subject in both stages: Polar H7 (blue) and Apple Watch (red).
In the stress stage, there are gaps in the Apple Watch recording where no beats are detected.
2.3. Validation of RR Series
A Bland–Altman plot [25] was used to study the validity of the Apple Watch RR series, by
representing both  and  series. The bias, the limits of agreement (LOA, ±2*std values)
and the percentage of paired RR measurements out of the LOA were also obtained.
To measure the interchangeability between both RR series, two reliability indices were used. First,
Lin’s concordance correlation coefficient (CCC) determines how far the observed data deviate from the
line of perfect concordance line at 45° on a square axis scatter plot [26]. Second, intraclass correlation
coefficient (ICC) represents the ratio of between-sample variance and the total variance (between- and
within-sample) to measure precision under the model of equal marginal distributions [27]. Lastly, the
Figure 1.
Example of the RR series for a subject in both stages: Polar H7 (blue) and Apple Watch (red).
In the stress stage, there are gaps in the Apple Watch recording where no beats are detected.
2.3. Validation of RR Series
A Bland–Altman plot [
25
] was used to study the validity of the Apple Watch RR series, by
representing both
RRH7g
and
RRAW
series. The bias, the limits of agreement (LOA,
±
2*std values)
and the percentage of paired RR measurements out of the LOA were also obtained.
To measure the interchangeability between both RR series, two reliability indices were
used. First, Lin’s concordance correlation coefficient (CCC) determines how far the observed
data deviate from the line of perfect concordance line at 45
on a square axis scatter plot [
26
].
Second, intraclass correlation coefficient (ICC) represents the ratio of between-sample variance and
the total variance (between- and within-sample) to measure precision under the model of equal
Sensors 2018,18, 2619 4 of 11
marginal distributions [
27
]. Lastly, the agreement was measured by an information-based measure
of disagreement (IBMD), which is based on Shannon’s entropy [
28
]. This index equals 0 when the
observers agree (no disagreement); i.e., there is no information in the differences between both methods.
The agreement (A) can be quantified as A = 1IBMD.
2.4. Heart Rate Variability Parameters
Heart rate variability parameters were divided in two classes: time and frequency domain indices.
Since both are calculated in different ways, the gaps also are addressed differently. For the temporal
domain analysis, we will derive HRV parameters from
RRAW
,
RRH7
and
RRH7g
, the latter created
with simulated gaps as defined in Section 2.2. For the frequency domain analysis, most of the methods
used for power spectral density estimation require evenly sampled series, and interpolation is usually
employed to overcome the intrinsic unevenly sampled RR interval series. However, since gaps in
RRAW
and
RRH7g
are too large (from 3.3 to 10.4 s), we first filled the gaps with artificial RR intervals,
created by linear interpolation. Note that, for simplicity, we use the same nomenclature,
RRAW
and
RRH7g
, for temporal and frequency domain analysis, although they are slightly different: for temporal
HRV parameters, the gaps are omitted and the RR intervals are concatenated (
RRAW
and
RRH7g
are
shorter than
RRH7
), while for frequency HRV parameters the gaps are filled with artificially created
RR intervals (RRAW and RRH7gare the same length than RRH7).
For the temporal parameters, four parameters were proposed [
1
]: HRM, SDNN, RMSSD and
pNN50. HRM is the mean heart rate, which is obtained as the inverse of the mean heart period (mean
of the RR intervals). SDNN is the standard deviation of the NN intervals (or normal RR intervals),
which is a measure of the total power in the analyzed segment. A dispersion measure can be obtained
by calculating the root mean-square of successive differences of adjacent intervals (RMSSD). The last
parameter, pNN50, represents the percentage of pairs of adjacent intervals differing by more than
50 ms. RMSSD and pNN50 describe short-time variability. Note that whenever a gap appears in
RRAW
, the corresponding intervals in
RRH7
are also removed, and the remaining RR intervals are
concatenated. The difference of the intervals just before and after the gaps are not included in the
calculation of RMSSD and pNN50.
For the frequency domain analysis, the heart rate variability signal was obtained following a
method based on the integral pulse frequency modulation (IPFM) model described in [
29
]: from the
RR interval series, we derived the instantaneous heart rate signal,
dHR (n)
, which is a continuous
and evenly sampled signal (sampled at 4 Hz). Then, we obtained the mean heart rate,
dHR M(n)
,
using a low pass filter up to 0.03 Hz. The variability signal,
dHRV (n)
, was obtained as
dHRV (n)=
dHR (n)dHRM (n)
. Finally, the modulating signal, which is assumed to represent the ANS modulation
on the sinoatrial node, was obtained as ˆ
m(n)=dHRV (n)/dHR M(n)[30].
The Welch periodogram was applied to estimate the spectral properties of the HRV signals
ˆ
mAW(n)
,
ˆ
mH7(n)
and
ˆ
mH7g(n)
(derived from
RRAW
,
RRH7
and
RRH7g
, respectively), using a
Hamming window with a length of 150 s with an overlap of 120 s. Low frequency (LF) and high
frequency (HF) powers,
PLF
and
PHF
respectively, were obtained integrating the power in their bands
(LF: 0.04–0.15 Hz, HF: 0.15–0.4 Hz) [1].
2.5. Statistical Analysis
None of all HRV parameters followed a normal distribution (tested with a Kolmogorov test).
To compare HRV parameters, a paired Wilcoxon test was applied. Then we performed 3 different
comparisons: (1) we analyzed the influence of the gaps by comparing HRV parameters derived from
RRH7
and
RRH7g
; (2) we compared HRV parameters derived from
RRAW
and
RRH7g
; and (3) HRV
parameters derived from
RRAW
were compared in relax vs. stress stage to evaluate the ability of those
parameters to reflect ANS response to stress, using changes in HRV parameters derived from
RRH7
as the reference ANS response. The difference was considered to be significantly different from zero
when p< 0.05.
Sensors 2018,18, 2619 5 of 11
Distribution of HRV parameters are shown using boxplots: on each box, the central mark indicates
the median, and the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively.
The whiskers extend to the most extreme data points not considered outliers, and the outliers are
plotted individually using the ‘+’ symbol. An outlier is defined as a point that falls more than 1.5
times the interquartile range (difference between 75th and 25th percentiles) above the 75th percentile
or below the 25th percentile.
3. Results
3.1. Validity of RR Series
A total of 206 gaps were found in the Apple Watch recordings, equivalent to 1321 missing RR
intervals (around 10% of total intervals). On average, each recording presents 5 gaps, with a length of
6 s per gap. The minimum length found was 3.3 s and the maximum length was 10.4 s. No differences
in the number or length of these gaps were found between relax and stress recordings.
Figure 2shows the Bland–Altman plot which evaluates the RR interval discrepancies between
Polar H7 and Apple Watch measurements and the stability across the different values of the intervals.
A total of 12,109 paired RR intervals were used from both relax and stress stages. The bias (central
horizontal line) was 0.06 ms. The LOA (upper and lower lines) contained 96.21% of the intervals.
No visual differences were found in the discrepancies between lower and higher RR intervals.
Sensors 2018, 18, x 5 of 11
respectively. The whiskers extend to the most extreme data points not considered outliers, and the
outliers are plotted individually using the ‘+’ symbol. An outlier is defined as a point that falls more
than 1.5 times the interquartile range (difference between 75th and 25th percentiles) above the 75th
percentile or below the 25th percentile.
3. Results
3.1. Validity of RR Series
A total of 206 gaps were found in the Apple Watch recordings, equivalent to 1321 missing RR
intervals (around 10% of total intervals). On average, each recording presents 5 gaps, with a length
of 6 s per gap. The minimum length found was 3.3 s and the maximum length was 10.4 s. No
differences in the number or length of these gaps were found between relax and stress recordings.
Figure 2 shows the Bland–Altman plot which evaluates the RR interval discrepancies between
Polar H7 and Apple Watch measurements and the stability across the different values of the intervals.
A total of 12,109 paired RR intervals were used from both relax and stress stages. The bias (central
horizontal line) was 0.06 ms. The LOA (upper and lower lines) contained 96.21% of the intervals. No
visual differences were found in the discrepancies between lower and higher RR intervals.
Figure 2. Bland-Altman plot:  vs. . Mean of the difference of the RR series ±2*std values
(limits of agreement, LOA).
Table 1 shows the mean values of the RR intervals for each stage, as well as the reliability and
agreement indices. While slightly lower in the stress stage, these indices show excellent reliability
and agreement between the measurements in both stages. Bias and LOA are also shown for relax and
stress stages, which are very similar to those obtained with the combined data. The last row shows
the mean of the percentage of missing RR intervals in .
Table 1. Validity of RR series from Apple Watch. CI = Confidence interval.
RELAX STRESS
Mean (SD) H7 RR intervals (ms) 869.28 (114.01) 834.78 (97.43)
Mean (SD) AW RR intervals (ms) 869.23 (114.39) 834.70 (97.84)
CCC (90% CI) 0.989 (0.981, 0.998) 0.977 (0.970, 0.985)
ICC (90% CI) 0.989 (0.984, 0.996) 0.982 (0.977, 0.987)
A (90% CI) 0.993 (0.987, 0.999) 0.983 (0.975, 0.991)
Bias (Out LOA) 0.049 (3.29%) 0.078 (3.93%)
Mean (SD)% missed RR intervals 10.98 (9.78) 9.45 (7.30)
Figure 2.
Bland-Altman plot:
RRH7gvs
.
RRAW
. Mean of the difference of the RR series
±
2*std values
(limits of agreement, LOA).
Table 1shows the mean values of the RR intervals for each stage, as well as the reliability and
agreement indices. While slightly lower in the stress stage, these indices show excellent reliability and
agreement between the measurements in both stages. Bias and LOA are also shown for relax and stress
stages, which are very similar to those obtained with the combined data. The last row shows the mean
of the percentage of missing RR intervals in RRAW.
Table 1. Validity of RR series from Apple Watch. CI = Confidence interval.
RELAX STRESS
Mean (SD) H7 RR intervals (ms) 869.28 (114.01) 834.78 (97.43)
Mean (SD) AW RR intervals (ms) 869.23 (114.39) 834.70 (97.84)
CCC (90% CI) 0.989 (0.981, 0.998) 0.977 (0.970, 0.985)
ICC (90% CI) 0.989 (0.984, 0.996) 0.982 (0.977, 0.987)
A (90% CI) 0.993 (0.987, 0.999) 0.983 (0.975, 0.991)
Bias (Out LOA) 0.049 (3.29%) 0.078 (3.93%)
Mean (SD)% missed RR intervals 10.98 (9.78) 9.45 (7.30)
Sensors 2018,18, 2619 6 of 11
3.2. HRV Parameters: Temporal Domain
Figure 3shows the temporal parameters for both relax and stress stages. No significant differences
were found in any parameter neither between RRH7and RRH7g, nor between RRH7gand RRAW.
Sensors 2018, 18, x 6 of 11
3.2. HRV Parameters: Temporal Domain
Figure 3 shows the temporal parameters for both relax and stress stages. No significant
differences were found in any parameter neither between  and , nor between  and
.
Figure 3. Heart rate variability (HRV) parameters: time domain.
3.3. HRV Parameters: Frequency Domain
Figure 4 shows the frequency HRV parameters derived from ,  and  in relax
and stress stages. Both  and  show significant differences (marked with an asterisk between
the boxplots) in  when comparing to  , being lower in the presence of the gaps. When
comparing HRV parameters derived from  and  , neither  nor  presented
significant differences.
Figure 4. HRV parameters: frequency domain, derived from  ,  and  . * denotes
significant differences (p < 0.05) between adjacent boxplots. Adim refers to adimensional units.
Figure 3. Heart rate variability (HRV) parameters: time domain.
3.3. HRV Parameters: Frequency Domain
Figure 4shows the frequency HRV parameters derived from
RRH7
,
RRH7g
and
RRAW
in
relax and stress stages. Both
PLF
and
PHF
show significant differences (marked with an asterisk
between the boxplots) in
RRH7g
when comparing to
RRH7
, being lower in the presence of the gaps.
When comparing HRV parameters derived from
RRAW
and
RRH7g
, neither
PLF
nor
PHF
presented
significant differences.
Figure 4.
HRV parameters: frequency domain, derived from
RRH7
,
RRH7g
and
RRAW
. * denotes
significant differences (p< 0.05) between adjacent boxplots. Adim refers to adimensional units.
Sensors 2018,18, 2619 7 of 11
3.4. HRV Parameters: Relax vs. Stress
Figures 5and 6display HRV parameters derived from
RRAW
and
RRH7
in relax vs. stress stages.
All temporal HRV parameters (HRM, SDNN, RMSSD and pNN50) presented changes from relax to
stress in both Apple Watch and Polar H7 recordings: an increase in HRM and a decrease in the rest of
parameters. We observed a decrease in
PLF
in stress with respect to relax stage, being only statistically
significant in Polar H7 recordings (p= 0.030 for Polar H7, p= 0.065 for Apple Watch). A significant
decrease was found in both devices when comparing PHF.
Sensors 2018, 18, x 7 of 11
3.4. HRV Parameters: Relax vs. Stress
Figures 5 and 6 display HRV parameters derived from  and  in relax vs. stress
stages. All temporal HRV parameters (HRM, SDNN, RMSSD and pNN50) presented changes from
relax to stress in both Apple Watch and Polar H7 recordings: an increase in HRM and a decrease in
the rest of parameters. We observed a decrease in  in stress with respect to relax stage, being only
statistically significant in Polar H7 recordings (p = 0.030 for Polar H7, p = 0.065 for Apple Watch). A
significant decrease was found in both devices when comparing .
Figure 5. HRV parameters: temporal domain (Relax vs. Stress). * denotes significant differences
(p < 0.05) between adjacent boxplots.
Figure 6. HRV parameters: frequency domain (Relax vs. Stress). * denotes significant differences (p <
0.05) between adjacent boxplots. Adim refers to adimensional units.
Figure 5.
HRV parameters: temporal domain (Relax vs. Stress). * denotes significant differences (p<
0.05) between adjacent boxplots.
Sensors 2018, 18, x 7 of 11
3.4. HRV Parameters: Relax vs. Stress
Figures 5 and 6 display HRV parameters derived from  and  in relax vs. stress
stages. All temporal HRV parameters (HRM, SDNN, RMSSD and pNN50) presented changes from
relax to stress in both Apple Watch and Polar H7 recordings: an increase in HRM and a decrease in
the rest of parameters. We observed a decrease in  in stress with respect to relax stage, being only
statistically significant in Polar H7 recordings (p = 0.030 for Polar H7, p = 0.065 for Apple Watch). A
significant decrease was found in both devices when comparing .
Figure 5. HRV parameters: temporal domain (Relax vs. Stress). * denotes significant differences
(p < 0.05) between adjacent boxplots.
Figure 6. HRV parameters: frequency domain (Relax vs. Stress). * denotes significant differences (p <
0.05) between adjacent boxplots. Adim refers to adimensional units.
Figure 6.
HRV parameters: frequency domain (Relax vs. Stress). * denotes significant differences
(p< 0.05) between adjacent boxplots. Adim refers to adimensional units.
Sensors 2018,18, 2619 8 of 11
4. Discussion
This is the first study to validate HRV measurements using an Apple Watch to the authors’
knowledge. Many of the wrist-worn devices aimed at assessing well-being, stress, and fitness provide
average HR values, derived from a pulse signal, instead of HRV. Fitbit, Jawbone, TomTom, Garming,
Withing, Samsumg, among others, commercialize such devices. However, although changes in
average HR are mainly induced by ANS and can be significantly different in some physio-pathological
situations, it cannot be considered a measure of autonomic function [
3
,
31
]. Moreover, there are
situations where altered ANS function is manifested in HRV but not in HR changes, such as in
depressed patients with respect to controls [32] or in exercise contexts [33,34].
In order to validate HRV measurements derived from an Apple Watch, in this work an
experimental study was conducted on young healthy volunteers subjected to mild relax and stress.
HRV measurements derived from Apple Watch were compared to those obtained from a validated
heart rate monitor, in particular Polar H7 chest band. Several works in the literature have already
validated, using as reference the RR interval series derived from a synchronous ECG, the RR interval
series given by different models of Polar devices: S810 [
35
,
36
], RS800 [
37
,
38
], and V800 with a H7
chest belt [22,23].
Regarding the RR intervals series, which are the basis for any HRV analysis, the bias was almost 0
and most pairs of RR intervals (around 96%) were within the limits of agreement in the Bland Altman
plot. These results are consistent when studying both stages separately. Moreover, the discrepancy
between Apple Watch and Polar H7 is similar with no dependency of the heart rate. The RR series
present excellent reliability and agreement when analyzing matched pairs of RR intervals. These results
agree with other studies for Apple Watch [17,18] and other heart rate watches [13,16,19].
However, it is important to note the existence in Apple Watch measurements of missing beats.
Around 10% of the beats could not be detected by the Apple Watch in both the relax and stress stages.
This could be not problematic if they were isolated beats, but they usually were consecutive beats,
which lead to gaps in the RR series. The shorter gaps were longer than 3 s, being easily identifiable as
abnormally large RR intervals. The origin of these gaps could be varied, from bad skin contact to fast
arm movement, but ultimately, we cannot conclude anything about these gaps because we do not have
direct access to the algorithms used by the Apple Watch. In [
18
], they also reported missing values
in the Apple Watch, with the proportion of heart rate values actually measured by the Apple Watch
decreasing with increasing exercise intensity. They also reported that these missing heart rate values
were higher in the first minute of exercise (between 20 and 40% of RR intervals), particularly at higher
intensities of exercise. In our data, however, we did not find more missing beats in the stress stage
compared to the relax stage, and the total missing RR intervals was about 10% of total intervals.
These missing beats are important when extracting HRV parameters in the frequency domain,
since the RR interval series needs to be interpolated in the gaps. This might influence HRV frequency
parameters. To study the influence of these gaps in frequency parameters
PLF
and
PHF
, we created
RRH7g
by intentionally creating gaps in the
RRH7
series, removing the missing RR intervals found
in
RRAW
. This analysis showed that both
PLF
and
PHF
were significantly lower in the presence of
simulated gaps. However, when comparing HRV frequency parameters derived from
RRAW
and from
RRH7g
, no significant difference was found, suggesting that differences between HRV parameters
derived from RR series provided by Apple Watch and Polar H7 are indeed due to the presence of gaps
in the Apple Watch series, rather than to the different signal from which they are derived or to the
different time resolution. The effect of gaps on HRV temporal parameters was also investigated, but no
significant differences were found, suggesting that these parameters are less sensitive to the presence
of gaps.
Despite these differences, HRV measurements derived from Apple Watch were still able to
reflect the ANS response induced by the stress stage, replicating most of the trends observed in HRV
measurements derived from the Polar H7. Induced mental stress resulted in a significant decrease
in
PHF
with respect to relax (in both the Apple Watch and Polar H7 measures, note that there were
Sensors 2018,18, 2619 9 of 11
no gaps in the Polar H7 recordings), suggesting the inhibition of parasympathetic stimulation, as
already reported in numerous studies [
8
,
39
]. A similar conclusion can be extracted with the decrease in
RMSSD parameter in the stress phase, being this parameter associated to the parasympathetic activity.
Although not shown in the results, the ratio between LF and HF powers (R), as well as the normalized
LF power (
PLFn
), were also obtained. Comparing relax and stress, these parameters have shown in
other studies a significant increase during stress [
8
], which means a predominance of the sympathetic
activity. In this work, however, the increase found in Rand
PLFn
was not significant even in the Polar
H7 measures, possibly due to the fact that the level of stress was moderate: see the moderately low
maximum HR achieved in the stress stage (around 100 bpm). Still, the significant decrease in
PHF
was
evident using both the Polar H7 and the Apple Watch, supporting the potential use of the Apple Watch
for stress monitoring.
However, one of the main limitations of using the Apple Watch to obtain and use HRV measures
is that, at this moment, the RR series is only available through the use of the Apple Breath App
in an indirect way (exporting the measured values using the Health App, where all Health-related
data from the user is stored). Apple does not provide within the watch OS 4.2 SDK any function to
access raw RR measures. Thus, developers are not able to develop an App that takes advantage of
the potential shown by Apple Watch in HRV monitoring. Besides, Breath App stops when repeated
motion is detected. This does not allow us to validate HRV measurements in low or moderate exercise.
Nevertheless, this SDK limitation can be removed by Apple in a further SDK release. This inclusion
would leverage the development of Apps able to monitoring stress conditions, among other things, by
the developer’s community.
5. Conclusions
Heart rate variability parameters extracted from an Apple Watch device have been validated using
a Polar H7 band as a reference during relax and mental stress in 20 healthy volunteers. Reliability and
agreement coefficients were computed for the RR interval series provided by both devices in relax and
stress stages, achieving very good results (reliability and agreement > 0.9). No significant differences
were found when comparing temporal HRV parameters (SDNN, RMSSD, pNN50 and HRM) derived
from the RR interval series provided by both devices. However, frequency HRV parameters LF and
HF powers were significantly different when derived from the Apple Watch, due to the appearance of
gaps in the RR series, both in relax and stress stages. Nonetheless, a decrease in the HF power (and in
the RMSSD parameter) was observed in stress with respect to relax stage when using both devices,
which supports the potential use of the Apple Watch for stress monitoring.
Author Contributions:
Conceptualization, D.H., Á.A. and R.B.; Methodology, D.H. and R.B.; Formal Analysis,
D.H.; Data Curation, S.R., J.S. and Á.A.; Writing-Original Draft Preparation, D.H., S.R., J.S., Á.A. and R.B.;
Writing-Review & Editing, D.H., Á.A. and R.B.; Supervision, Á.A. and R.B.
Funding:
This work was supported by Centro de Investigación Biomédica en Red–Bioingeniería, Biomateriales y
Nanomedicina (CIBER-BBN) through Instituto de Salud Carlos III, by the Ministerio de Economía, Industria y
Competitividad, Gobierno de España, European Regional Development Fund (TIN2016-76770-R) and FEDER
“Construyendo Europa desde Aragón” (Grupo de investigación T31_17R), by Aragón Government through Grupo
de Referencia BSICoS (T39_17R), by Aragón Institute of Engineering Research (I3A), IIS Aragón and European
Social Fund (EU).
Acknowledgments:
The computation was performed by the ICTS “NANBIOSIS”, more specifically by the
High Performance Computing Unit of the CIBER-BBN at the University of Zaragoza. The authors also want to
acknowledge the support from Ogilvy Paris.
Conflicts of Interest:
The authors declare no conflict of interest. The funders had no role in the design of the
study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, and in the decision to
publish the results.
Sensors 2018,18, 2619 10 of 11
References
1.
Camm, A.J.; Malik, M.; Bigger, J.T.; Brethardt, G.; Cerutti, S.; Cohen, R.J.; Coumel, P.; Fallen, E.L.;
Kennedy, H.L.; Kleiger, R.E.; et al. Heart rate variability: Standards of measurement, physiological
interpretation, and clinical use. Circulation 1996,93, 1043–1065.
2.
Thayer, J.; Ahs, F.; Fredrikson, M.; Sollers, J.; Wager, T. A meta-analysis of heart rate variability
and neuroimaging studies: Implications for heart rate variability as a marker of stress and health.
Neurosci. Biobehav. Rev. 2012,36, 747–756. [CrossRef] [PubMed]
3.
Billman, G. Heart rate variability—A historical perspective. Front. Physiol.
2011
,2, 86. [CrossRef] [PubMed]
4.
Beauchaine, T.; Thayer, J. Heart rate variability as a transdiagnostic biomarker of psychopathology.
Int. J. Psychophysiol. 2015,98, 338–350. [CrossRef] [PubMed]
5.
Koenig, J.; Kemp, A.; Beauchaine, T.; Thayer, J.; Kaess, M. Depression and resting state heart rate variability
in children and adolescents—A systematic review and meta-analysis. Clin. Psychol. Rev.
2016
,46, 136–150.
[CrossRef] [PubMed]
6.
Tobaldini, E.; Nobili, L.; Strada, S.; Casali, K.; Braghiroli, A.; Montano, N. Heart rate variability in normal
and pathological sleep. Front. Physiol. 2013,4, 294. [CrossRef] [PubMed]
7.
Kim, H.; Cheon, E.; Bai, D.; Lee, Y.; Koo, B. Stress and heart rate variability: A meta-analysis and review of
the literature. Psychiatry Investig. 2018,15, 235–245. [CrossRef] [PubMed]
8.
Hernando, A.; Lázaro, J.; Gil, E.; Arza, A.; Garzón, J.; López-Antón, R.; de la Cámara, C.; Laguna, P.;
Aguiló, J.; Bailón, R. Inclusion of respiratory frequency information in heart rate variability analysis for
stress assessment. IEEE J. Biomed. Health Inform. 2016,20, 1016–1025. [CrossRef] [PubMed]
9.
Vicente, J.; Laguna, P.; Bartra, A.; Bailón, R. Drowsiness detection using heart rate variability. Med. Biol.
Eng. Comput. 2016,54, 927–937. [CrossRef] [PubMed]
10.
Kiviniemi, A.; Hautala, A.; Kinnunen, H.; Tulppo, M. Endurance training guided individually by daily heart
rate variability measurements. Eur. J. Appl. Physiol. 2007,101, 743–751. [CrossRef] [PubMed]
11.
Plews, D.; Laursen, P.; Stanley, J.; Kilding, A.; Buchheit, M. Training adaptation and heart rate variability in
elite endurance athletes: Opening the door to effective monitoring. Sports Med.
2013
,43, 773–781. [CrossRef]
[PubMed]
12.
Tian, Y.; He, Z.; Zhao, J.; Tao, D.; Xu, K.; Earnest, C.; McNaughton, L. Heart rate variability threshold values
for early-warning nonfunctional overreaching in elite female wrestlers. J. Strength Cond. Res.
2013
,27,
1511–1519. [CrossRef] [PubMed]
13.
Shcherbina, A.; Mattsson, C.; Waggott, D.; Salisbury, H.; Christle, J.; Hastie, T.; Wheeler, M.; Ashley, E.
Accuracy in wrist-worn, sensor-based measurements of heart rate and energy expenditure in a diverse
cohort. J. Pers. Med. 2017,7, 3. [CrossRef] [PubMed]
14.
Gil, E.; Orini, M.; Bailón, R.; Vergara, J.; Mainardi, L.; Laguna, P. Photoplethysmography pulse rate variability
as a surrogate measurement of heart rate variability during non-stationary conditions. Physiol. Meas.
2010
,
31, 1271–1290. [CrossRef] [PubMed]
15.
Eskola, H.; Väisänen, O.; Viik, J.; Hyttinen, J. Evaluation of the accuracy and reliability for
photoplethysmography based heart rate and beat-to-beat detection during daily activities. In Proceedings
of the Joint Conference of the European Medical and Biological Engineering Conference (EMBEC) and the
Nordic-Baltic Conference on Biomedical Engineering and Medical Physics (NBC) 2017, Tampere, Finland,
11–15 June 2017.
16.
Wallen, M.; Gomersall, S.; Keating, S.; Wisløff, U.; Coombes, J. Accuracy of heart rate watches: Implications
for weight management. PLoS ONE 2016,11, e0154420. [CrossRef] [PubMed]
17.
Abt, G.; Bray, J.; Benson, A. The validity and inter-device variability of the Apple Watch for measuring
maximal heart rate. J. Sports Sci. 2017,36, 1447–1452. [CrossRef] [PubMed]
18.
Khushhal, A.; Nichols, S.; Evans, W.; Gleadall-Siddall, D.; Page, R.; O’Doherty, A.; Carroll, S.; Ingle, L.;
Abt, G. Validity and reliability of the AppleWatch for measuring heart rate during exercise. Sports Med.
Int. Open 2017,1, 206–211.
19.
Wang, R.; Blackburn, G.; Desai, M.; Phelan, D.; Gillinov, L.; Houghtaling, P.; Gillinov, M. Accuracy of
wrist-worn heart rate monitors. JAMA Cardiol. 2017,2, 104–106. [CrossRef] [PubMed]
20.
Bai, Y.; Hibbing, P.; Mantis, C.; Welk, G. Comparative evaluation of heart rate-based monitors: Apple Watch
vs Fitbit Charge HR. J. Sports Sci. 2017,36, 1734–1741. [CrossRef] [PubMed]
Sensors 2018,18, 2619 11 of 11
21.
Canalys. 18 Million Apple Watches Ship in 2017, up 54% on 2016. 2018. Available online: https://www.
canalys.com/newsroom/18-million-apple-watches-ship-2017-54-2016 (accessed on 4 July 2018).
22.
Giles, D.; Draper, N.; Neil, W. Validity of the Polar V800 heart rate monitor to measure RR intervals at rest.
Eur. J. Appl. Physiol. 2016,116, 563–571. [CrossRef] [PubMed]
23.
Caminal, P.; Sola, F.; Gomis, P.; Guasch, E.; Perera, A.; Soriano, N.; Mont, L. Validity of the Polar V800
monitor for measuring heart rate variability in mountain running route conditions. Eur. J. Appl. Physiol.
2018,118, 669–677. [CrossRef] [PubMed]
24.
Valderas, M.T.; Bolea, J.; Laguna, P.; Vallverdú, M.; Bailón, R. Human Emotion Recognition Using Heart
Rate Variability Analysis with Spectral Bands Based on Respiration. In Proceedings of the 37nd Annual
International Conference of the IEEE EMBS, Milan, Italy, 25–29 August 2015; pp. 6134–6137.
25.
Bland, J.; Altman, D. Statistical methods for assessing agreement between two methods of clinical
measurement. Lancet 1986,8, 307–310. [CrossRef]
26.
Lin, L. A concordance correlation coefficient to evaluate reproducibility. Biometrics
1989
,45, 255–268.
[CrossRef] [PubMed]
27. Fisher, R. Statistical Methods for Research Workers; Oliver & Boyd: Edinburgh, UK, 1925.
28.
Costa-Santos, C.; Antunes, L.; Souto, A.; Bernardes, J. Assessment of disagreement: A new information-based
approach. Ann. Epidemiol. 2010,20, 555–561. [CrossRef] [PubMed]
29.
Mateo, J.; Laguna, P. Improved heart rate variability signal analysis from the beat occurrence times according
to the IPFM model. IEEE Trans. Biomed. Eng. 2000,47, 985–996. [CrossRef] [PubMed]
30.
Bailón, R.; Laouini, G.; Grao, C.; Orini, M.; Laguna, P.; Meste, O. The integral pulse frequency modulation
with time-varying threshold: Application to heart rate variability analysis during exercise stress testing.
IEEE Trans. Biomed. Eng. 2011,58, 642–652. [CrossRef] [PubMed]
31.
Sacha, J. Interplay between heart rate and its variability: A. prognostic game. Front. Physiol.
2014
,5, 347.
[CrossRef] [PubMed]
32.
Valenza, G.; Garcia, R.; Citi, L.; Scilingo, E.; Tomaz, C.; Barbieri, R. Nonlinear digital signal processing in
mental health: Characterization of major depression using instantaneous entropy measures of heartbeat
dynamics. Front. Physiol. 2015,6, 74. [CrossRef] [PubMed]
33.
Weippert, M.; Behrens, K.; Rieger, A.; Stoll, R.; Kreuzfeld, S. Heart rate variability and blood pressure during
dynamic and static exercise at similar heart rate levels. PLoS ONE 2013,8, e83690. [CrossRef] [PubMed]
34.
Buchheit, M. Monitoring training status with HR measures: Do all roads lead to Rome? Front. Physiol.
2014
,
5, 73. [CrossRef] [PubMed]
35.
Gamelin, F.X.; Baquet, G.; Berthoin, S.; Bosquet, L. Validity of the polar S810 to measure R-R intervals in
children. Int. J. Sports Med. 2008,29, 134–138. [CrossRef] [PubMed]
36.
Nunan, D.; Jakovljevic, D.G.; Donovan, G.; Hodges, L.D.; Sandercock, G.R.; Brodie, D.A. Levels of agreement
for RR intervals and short-term heart rate variability obtained from the Polar S810 and an alternative system.
Eur. J. Appl. Physiol. 2008,103, 529–537. [CrossRef] [PubMed]
37.
Barbosa, M.P.; da Silva, N.T.; de Azevedo, F.M.; Pastre, C.M.; Vanderlei, L.C. Comparison of Polar
®
RS800G3
heart rate monitor with Polar
®
S810i
and electrocardiogram to obtain the series of RR intervals and analysis
of heart rate variability at rest. Clin. Physiol. Funct. Imaging 2016,36, 112–117. [CrossRef] [PubMed]
38.
Hernando, D.; Garatachea, N.; Almeida, R.; Casajús, J.A.; Bailón, R. Validation of Heart Rate Monitor Polar
RS800 for Heart Rate Variability Analysis During Exercise. J. Strength Cond. Res.
2018
,32, 716–725. [CrossRef]
[PubMed]
39.
Porges, S. Cardiac vagal tone: A physiological index of stress. Neurosci. Biobehav. Rev.
1995
,19, 225–233.
[CrossRef]
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
... It has been considered the reference device for HR measurement; in fact, chest-worn wearable devices are generally more accurate with respect to wrist-worn ones [31], mainly for the different sensing principle, which is based on electrodes (hence, an electrical signal is acquired) instead of PPG (optical signal) and also due to the fact that they are placed on the thoracic area, in correspondence to the cardiac muscle [29]. Moreover, it is commonly used as a gold standard in the literature [32,33]; ...
... Few studies in the literature report data related to the metrological characterisation of the employed devices. For example, Gilgen-Ammann et al. found a signal quality (i.e., relative number of correctly detected RR intervals) of 99.6% for the Polar H10 cardiac belt [36], which is often considered a reference instrument for HR measurement [33,37]). Polar manufacturer published a white paper in 2019, reporting an error of HR < 4% in all the tested activities (i.e., running, cycling, weight training, and a combination of them), proving that its performance in measuring RR intervals is excellent [38]. ...
Article
Full-text available
Nowadays, the use of wearable devices is spreading in different fields of application, such as healthcare, digital health, and sports monitoring. In sport applications, the present trend is to continuously monitor the athletes’ physiological parameters during training or competitions to maximize performance and support coaches. This paper aims to evaluate the performances in heart rate assessment, in terms of accuracy and precision, of both wrist-worn and chest-strap commercial devices used during swimming activity, considering a test population of 10 expert swimmers. Three devices were employed: Polar H10 cardiac belt, Polar Vantage V2, and Garmin Venu Sq smartwatches. The former was used as a reference device to validate the data measured by the two smartwatches. Tests were performed both in dry and wet conditions, considering walking/running on a treadmill and different swimming styles in water, respectively. The measurement accuracy and precision were evaluated through standard methods, i.e., Bland–Altman plot, analysis of deviations, and Pearson’s correlation coefficient. Results show that both precision and accuracy worsen during swimming activity (with an absolute increase of the measurement deviation in the range of 13–56 bpm for mean value and 49–52 bpm for standard deviation), proving how water and arms movement act as relevant interference inputs. Moreover, it was found that wearable performance decreases when activity intensity increases, highlighting the need for specific research for wearable applications in water, with a particular focus on swimming-related sports activities.
... Such interest arises because researchers have shown that HRV is one of the most promising markers to assess autonomic nervous system (ANS) activity, which can be directly correlated to cardiovascular disease [36]. At the same time, with massive improvements and advances in consumer and wearable technology, HRV data can be calculated using smart watches or small portable chest straps, which makes it more accessible for diagnosis [37], [38]. Existing works on HRV analysis and their limitations will be addressed in Chapter 3. ...
Preprint
Full-text available
Many types of ventricular and atrial cardiac arrhythmias have been discovered in clinical practice in the past 100 years, and these arrhythmias are a major contributor to sudden cardiac death. Ventricular tachycardia, ventricular fibrillation, and paroxysmal atrial fibrillation are the most commonly-occurring and dangerous arrhythmias, therefore early detection is crucial to prevent any further complications and reduce fatalities. Implantable devices such as pacemakers are commonly used in patients at high risk of sudden cardiac death. While great advances have been made in medical technology, there remain significant challenges in effective management of common arrhythmias. This thesis proposes novel arrhythmia detection and prediction methods to differentiate cardiac arrhythmias from non-life-threatening cardiac events, to increase the likelihood of detecting events that may lead to mortality, as well as reduce the incidence of unnecessary therapeutic intervention. The methods are based on detailed analysis of Heart Rate Variability (HRV) information. The results of the work show good performance of the proposed methods and support the potential for their deployment in resource-constrained devices for ventricular and atrial arrhythmia prediction, such as implantable pacemakers and defibrillators.
... Finally, continuous and daily HR monitoring helps to improve users well-being and their awareness of their lifestyle. In fact, HR and HRV provide useful information on stress level [15] and sleep quality [16]. ...
Article
Full-text available
This work presents an overview of the main strategies that have been proposed for non-invasive monitoring of heart rate (HR) in extramural and home settings. We discuss three categories of sensing according to what physiological effect is used to measure the pulsatile activity of the heart, and we focus on an illustrative sensing modality for each of them. Therefore, electrocardiography, photoplethysmography, and mechanocardiography are presented as illustrative modalities to sense electrical activity, mechanical activity, and the peripheral effect of heart activity. In this paper, we describe the physical principles underlying the three categories and the characteristics of the different types of sensors that belong to each class, and we touch upon the most used software strategies that are currently adopted to effectively and reliably extract HR. In addition, we investigate the strengths and weaknesses of each category linked to the different applications in order to provide the reader with guidelines for selecting the most suitable solution according to the requirements and constraints of the application.
... Due to the progression of the chest strap HRV monitoring technology, research have slowly moved away from using the smartwatches in combination with the chest straps (Hernández-Vicente et al., 2021;Hernando, Roca, Sancho, Alesanco, & Bailón, 2018;Plews et al., 2017). The chest straps, rather, are connected to a smart phone application developed by Polar® which monitors HRV in real time. ...
Chapter
Firefighters are required to maintain all aspects of their health and wellness in order to sustain their fitness for duty. Heart rate variability (HRV) has been used as a reliable tool when assessing the stressors placed on firefighters, be it physical, emotional, or psychological. This review determined the usefulness of using HRV as a tool to determine the physical, physiological, and psychological health of firefighters at a more regular and frequent scale. HRV is a versatile technology with a plethora of uses, particularly in monitoring the cardiovascular strain as a result of firefighting and recovery post-fire suppression. In addition, the literature showed that HRV could be used to successfully monitor physical fitness, physiological stress, psychological stress, decision making, risk taking behavior and recovery in firefighters. The use of mobile technology measuring HRV may be used to successfully assess firefighter occupational performance. In future research, longitudinal studies investigating HRV use in firefighters are warranted.
... MoBI devices (such as C-grids, Debener et al., 2015) will be unobtrusive and will probably be incorporated into the daily activities of participants/users/patients as wearable/hearable devices. These devices already provide continuous monitoring of people's activity and physiological dynamics (e.g., heartrate variability in consumer wearable technology, Hernando et al., 2018). For example, in the future we will have people who have profiled their heart physiology for most of their adult life. ...
Article
Full-text available
Cognitive dynamics are multimodal, and they need to integrate real-time feedback to be adaptive and appropriate. However, cognition research still relies on mostly unimodal paradigms using simple motor tasks in laboratory-based static situations. This paper addresses this limitation by presenting the Mobile Brain/Body Imaging approach based on the Embodied, Embedded, Extended, and Enactive perspective, which complements traditional laboratory work while also facilitating ecologically valid applications. First, we briefly review Mobile Brain/Body Imaging technologies used to obtain functional and structural images of the Brain/Body System during natural cognition. Specifically: mobile cognitive electrophysiology, mobile functional neurovascular dynamics, and mobile behavioral measurements. Second, we review the development of Mobile Brain/Body Imaging/4E in Chile. Finally, we discuss challenges and opportunities. We conclude that although this new epistemic/methodological approach is promising, there is a need for greater portability, robust equipment, and data-analysis tools that can integrate signals from the brain/body-in-the-world system. Future experimental designs need to re-consider their underlying logic and increase their ecological validity by perhaps-modifying the physical spaces in which experiments are conducted.
... Such interest arises because researchers have shown that HRV is one of the most promising markers to assess autonomic nervous system (ANS) activity, which can be directly correlated to cardiovascular disease [36]. At the same time, with massive improvements and advances in consumer and wearable technology, HRV data can be calculated using smart watches or small portable chest straps, which makes it more accessible for diagnosis [37], [38]. Existing works on HRV analysis and their limitations will be addressed in Chapter 3. ...
Thesis
Full-text available
Many types of ventricular and atrial cardiac arrhythmias have been discovered in clinical practice in the past 100 years, and these arrhythmias are a major contributor to sudden cardiac death. Ventricular tachycardia, ventricular fibrillation, and paroxysmal atrial fibrillation are the most commonly-occurring and dangerous arrhythmias, therefore early detection is crucial to prevent any further complications and reduce fatalities. Implantable devices such as pacemakers are commonly used in patients at high risk of sudden cardiac death. While great advances have been made in medical technology, there remain significant challenges in effective management of common arrhythmias. This thesis proposes novel arrhythmia detection and prediction methods to differentiate cardiac arrhythmias from non-life-threatening cardiac events, to increase the likelihood of detecting events that may lead to mortality, as well as reduce the incidence of unnecessary therapeutic intervention. The methods are based on detailed analysis of Heart Rate Variability (HRV) information. A range of features based on HRV analysis are investigated, including features from time, frequency, bispectrum and nonlinear analysis, and a range of classification techniques is used for prediction based on these features. Firstly, a thorough review of existing methods for ventricular arrhythmia prediction is conducted, including a detailed comparative experimental study. Following this, feature extraction for ventricular arrhythmia prediction is investigated using feature ranking methods based on mutual information. Using classification techniques such as support vector machine, k-nearest neighbour and random forests, the proposed approaches compare well with related work in the literature using different signal analysis durations. Furthermore, for paroxysmal atrial fibrillation prediction, seven novel features are proposed and investigated using a number of standard classification techniques. Using only the seven newly-proposed features, classification performance outperforms those of the classical state-of-the-art feature set, and the results further improve when the features are combined with several of the classical features. The results of the work show good performance of the proposed methods and support the potential for their deployment in resource-constrained devices for ventricular and atrial arrhythmia prediction, such as implantable pacemakers and defibrillators.
... Recent advances in the manufacturing process have enabled devices to be designed with a high degree of form freedom. Figure 1 shows a change in the number of publications for the recent research topics "Wearable devices and three-dimensional (3D) electronics/In-mold electronics" and their related research areas. Wearable devices as sensor-integrated systems are primarily used in healthcare management to enhance a healthy life expectancy and are also highlighted by real-time patient monitoring for medical care [2][3][4][5][6][7]. Since 2000, there has been an increase in the number of publications on this topic, which has increased further since 2014. ...
Article
Full-text available
Recent technological innovations, such as material printing techniques and surface functionalization, have significantly accelerated the development of new free-form sensors for next-generation flexible, wearable, and three-dimensional electronic devices. Ceramic film sensors, in particular, are in high demand for the production of reliable flexible devices. Various ceramic films can now be formed on plastic substrates through the development of low temperature fabrication processes for ceramic films, such as photocrystallization and transferring methods. Among flexible sensors, strain sensors for precise motion detection and photodetectors for biomonitoring have seen the most research development, but other fundamental sensors for temperature and humidity have also begun to grow. Recently, flexible gas and electrochemical sensors have attracted a lot of attention from a new real-time monitoring application that uses human breath and perspiration to accurately diagnose presymptomatic states. The development of a low-temperature fabrication process of ceramic film sensors and related components will complete the chemically stable and reliable free-form sensing devices by satisfying the demands that can only be addressed by flexible metal and organic components.
Thesis
Full-text available
Alleviating the burden of breast cancer has become in one of the biggest challenges of our times. The advances in surgery, radiotherapy, and systemic therapy have improved the survival rates of patients with breast cancer, but have also produced a higher number of patients suffering short- and long-term side effects, with high the risk of recurrence, developing comorbidities, and death. Therapeutic exercise poses a means to address this issues; however, exercise interventions in patients with cancer are often adhered to the same therapeutic exercise guidelines. This results in one-size-fits-all exercise prescriptions for all adults, regardless their individual exercise capabilities and needs, which may lead to inadequate training adaptation. The mobile health (mHealth) paradigm has enabled the remote and individual monitoring of health through wearable sensors and smartphones. Personalizing training adaptation with an mHealth approach has already been successfully conducted in sports settings, and the literature suggests that similar strategies may translated to patients with chronic conditions such as breast cancer. However, recent works do not target the adjustment of training doses to the individual needs of the patients. This thesis presents three contributions to support the personalization of therapeutic exercise intervention in patients with breast cancer. First, ATOPE+, an mHealth system to support the remote monitoring of patients’ training load through heart rate variability (HRV), self-reported wellness, and Fitbit physical activity and sleep data. ATOPE+ also integrates a decision-support system with expert rules that automatically trigger daily exercise recommendations for patients. Second, the ATOPE+Breast dataset, an open dataset describing the continuous evolution of training load during therapeutic exercise intervention for 23 patients with breast cancer. Third, a clustering approach to assess training needs in patients with breast cancer. Data science and artificial intelligence (AI) are leveraged in this approach to better understand the different states of the patient throughout an exercise intervention, and eventually serve as a tool to make more informed decisions when prescribing an exercise dose. The potential of these contributions may lead to new research directions in the personalization of therapeutic exercise interventions in real-life scenarios, specially regarding the application of mHealth and AI to improve chronic conditions.
Thesis
Full-text available
El tecnoestrés se asume como cualquier impacto negativo provocado por el uso repetido de las tecnologías de la información y la comunicación. Si bien ha sido estudiado en diversos países, su impacto apenas comienza a ser comprendido en México. El cambio en los procesos de enseñanza provocados por la pandemia SARS-CoV-2 obligó a muchos profesores universitarios de la Ciudad de México y municipios urbanos aledaños del Estado de México a cambiar su esquema de instrucción y depender de las TIC para llevar a cabo su trabajo, lo que incrementó su nivel de tecnoestrés. Mediante una metodología mixta con herramientas tradicionales y una capa agregada de recolección de datos biométricos, esta investigación encontró que el tecnoestrés tiene diversas manifestaciones fisiológicas y psicológicas en los maestros estudiados. También halló que los profesores con mayores niveles de tecnoestrés son más propensos encontrarse insatisfechos laboralmente, aunque existen diferencias categóricas entre ellos. Al final de la investigación se propone un modelo de recolección, análisis y visualización de data para disminuir el tecnoestrés a nivel institucional mediante una mejor toma de decisiones en tiempo real.
Article
The current protocol aims to showcase the technology integration, providing a detailed description of adopting the HealthCloud app, developed by the Healthy Landscape and Healthy People Lab, National Taiwan University (HLHP-NTU), on smartphones and smartwatches to collect data on users' real-time psychological and physiological responses and environmental information. A flexible and integrated research method was proposed because it can be difficult to measure multi-dimensional aspects of personal data in on-site studies in landscape and outdoor recreation research. An on-site study conducted in 2020 at the National Taiwan University campus was used as an application example. A dataset of 385 participants was used after excluding invalid samples. During the experiment, participants were asked to walk around campus for 30 min when their heart rate and psychological-scale items were measured, together with several environmental metrics. This work aimed to provide a possible solution to help on-site studies track real-time human responses that match ambient factors. Due to the app's flexibility, its use on wearable devices shows excellent potential for multidisciplinary research studies.
Article
Full-text available
Objective: Physical or mental imbalance caused by harmful stimuli can induce stress to maintain homeostasis. During chronic stress, the sympathetic nervous system is hyperactivated, causing physical, psychological, and behavioral abnormalities. At present, there is no accepted standard for stress evaluation. This review aimed to survey studies providing a rationale for selecting heart rate variability (HRV) as a psychological stress indicator. Methods: Term searches in the Web of Science®, National Library of Medicine (PubMed), and Google Scholar databases yielded 37 publications meeting our criteria. The inclusion criteria were involvement of human participants, HRV as an objective psychological stress measure, and measured HRV reactivity. Results: In most studies, HRV variables changed in response to stress induced by various methods. The most frequently reported factor associated with variation in HRV variables was low parasympathetic activity, which is characterized by a decrease in the high-frequency band and an increase in the low-frequency band. Neuroimaging studies suggested that HRV may be linked to cortical regions (e.g., the ventromedial prefrontal cortex) that are involved in stressful situation appraisal. Conclusion: In conclusion, the current neurobiological evidence suggests that HRV is impacted by stress and supports its use for the objective assessment of psychological health and stress.
Article
Full-text available
Purpose: This study was conducted to test, in mountain running route conditions, the accuracy of the Polar V800™ monitor as a suitable device for monitoring the heart rate variability (HRV) of runners. Method: Eighteen healthy subjects ran a route that included a range of running slopes such as those encountered in trail and ultra-trail races. The comparative study of a V800 and a Holter SEER 12 ECG Recorder™ included the analysis of RR time series and short-term HRV analysis. A correction algorithm was designed to obtain the corrected Polar RR intervals. Six 5-min segments related to different running slopes were considered for each subject. Results: The correlation between corrected V800 RR intervals and Holter RR intervals was very high (r = 0.99, p < 0.001), and the bias was less than 1 ms. The limits of agreement (LoA) obtained for SDNN and RMSSD were (- 0.25 to 0.32 ms) and (- 0.90 to 1.08 ms), respectively. The effect size (ES) obtained in the time domain HRV parameters was considered small (ES < 0.2). Frequency domain HRV parameters did not differ (p > 0.05) and were well correlated (r ≥ 0.96, p < 0.001). Conclusion: Narrow limits of agreement, high correlations and small effect size suggest that the Polar V800 is a valid tool for the analysis of heart rate variability in athletes while running high endurance events such as marathon, trail, and ultra-trail races.
Article
Full-text available
We examined the validity and reliability of the Apple Watch heart rate sensor during and in recovery from exercise. Twenty-one males completed treadmill exercise while wearing two Apple Watches (left and right wrists) and a Polar S810i monitor (criterion). Exercise involved 5-min bouts of walking, jogging, and running at speeds of 4 km.h−1, 7 km.h−1, and 10 km.h−1, followed by 11 min of rest between bouts. At all exercise intensities the mean bias was trivial. There were very good correlations with the criterion during walking (L: r=0.97; R: r=0.97), but good (L: r=0.93; R: r=0.92) and poor/good (L: r=0.81; R: r=0.86) correlations during jogging and running. Standardised typical error of the estimate was small, moderate, and moderate to large. There were good correlations following walking, but poor correlations following jogging and running. The percentage of heart rates recorded reduced with increasing intensity but increased over time. Intra-device standardised typical errors decreased with intensity. Inter-device standardised typical errors were small to moderate with very good to nearly perfect intraclass correlations. The Apple Watch heart rate sensor has very good validity during walking but validity decreases with increasing intensity.
Article
Full-text available
The ability to measure physical activity through wrist-worn devices provides an opportunity for cardiovascular medicine. However, the accuracy of commercial devices is largely unknown. The aim of this work is to assess the accuracy of seven commercially available wrist-worn devices in estimating heart rate (HR) and energy expenditure (EE) and to propose a wearable sensor evaluation framework. We evaluated the Apple Watch, Basis Peak, Fitbit Surge, Microsoft Band, Mio Alpha 2, PulseOn, and Samsung Gear S2. Participants wore devices while being simultaneously assessed with continuous telemetry and indirect calorimetry while sitting, walking, running, and cycling. Sixty volunteers (29 male, 31 female, age 38 ± 11 years) of diverse age, height, weight, skin tone, and fitness level were selected. Error in HR and EE was computed for each subject/device/activity combination. Devices reported the lowest error for cycling and the highest for walking. Device error was higher for males, greater body mass index, darker skin tone, and walking. Six of the devices achieved a median error for HR below 5% during cycling. No device achieved an error in EE below 20 percent. The Apple Watch achieved the lowest overall error in both HR and EE, while the Samsung Gear S2 reported the highest. In conclusion, most wrist-worn devices adequately measure HR in laboratory-based activities, but poorly estimate EE, suggesting caution in the use of EE measurements as part of health improvement programs. We propose reference standards for the validation of consumer health devices (http://precision.stanford.edu/).
Article
Full-text available
Wrist-worn fitness and heart rate (HR) monitors are popular.¹,2 While the accuracy of chest strap, electrode-based HR monitors has been confirmed,³,4 the accuracy of wrist-worn, optically based HR monitors is uncertain.⁵,6 Assessment of the monitors’ accuracy is important for individuals who use them to guide their physical activity and for physicians to whom these individuals report HR readings. The objective of this study was to assess the accuracy of 4 popular wrist-worn HR monitors under conditions of varying physical exertion.
Article
Full-text available
Background: Wrist-worn monitors claim to provide accurate measures of heart rate and energy expenditure. People wishing to lose weight use these devices to monitor energy balance, however the accuracy of these devices to measure such parameters has not been established. Aim: To determine the accuracy of four wrist-worn devices (Apple Watch, Fitbit Charge HR, Samsung Gear S and Mio Alpha) to measure heart rate and energy expenditure at rest and during exercise. Methods: Twenty-two healthy volunteers (50% female; aged 24 ± 5.6 years) completed ~1-hr protocols involving supine and seated rest, walking and running on a treadmill and cycling on an ergometer. Data from the devices collected during the protocol were compared with reference methods: electrocardiography (heart rate) and indirect calorimetry (energy expenditure). Results: None of the devices performed significantly better overall, however heart rate was consistently more accurate than energy expenditure across all four devices. Correlations between the devices and reference methods were moderate to strong for heart rate (0.67-0.95 [0.35 to 0.98]) and weak to strong for energy expenditure (0.16-0.86 [-0.25 to 0.95]). All devices underestimated both outcomes compared to reference methods. The percentage error for heart rate was small across the devices (range: 1-9%) but greater for energy expenditure (9-43%). Similarly, limits of agreement were considerably narrower for heart rate (ranging from -27.3 to 13.1 bpm) than energy expenditure (ranging from -266.7 to 65.7 kcals) across devices. Conclusion: These devices accurately measure heart rate. However, estimates of energy expenditure are poor and would have implications for people using these devices for weight loss.
Article
The purpose of this investigation was to examine the validity of energy expenditure (EE), steps, and heart rate measured with the Apple Watch 1 and Fitbit Charge HR. Thirty-nine healthy adults wore the two monitors while completing a semi-structured activity protocol consisting of 20 minutes of sedentary activity, 25 minutes of aerobic exercise, and 25 minutes of light intensity physical activity. Criterion measures were obtained from an Oxycon Mobile for EE, a pedometer for steps, and a Polar heart rate strap worn on the chest for heart rate. For estimating whole-trial EE, the mean absolute percent error (MAPE) from Fitbit Charge HR (32.9%) was more than twice that of Apple Watch 1 (15.2%). This trend was consistent for the individual conditions. Both monitors accurately assessed steps during aerobic activity (MAPEApple: 6.2%; MAPEFitbit: 9.4%) but overestimated steps in light physical activity. For heart rate, Fitbit Charge HR produced its smallest MAPE in sedentary behaviors (7.2%), followed by aerobic exercise (8.4%), and light activity (10.1%). The Apple Watch 1 had stronger validity than the Fitbit Charge HR for assessing overall EE and steps during aerobic exercise. The Fitbit Charge HR provided heart rate estimates that were statistically equivalent to Polar monitor.
Article
Maximal heart rate (HRmax) is a fundamental measure used in exercise prescription. The Apple Watch™ measures heart rate yet the validity and inter-device variability of the device for measuring HRmax are unknown. Fifteen participants completed a maximal oxygen uptake test while wearing an Apple Watch™ on each wrist. Criterion HRmax was measured using a Polar T31™ chest strap. There were good to very good correlations between the watches and criterion (left: r = 0.87 [90%CI: 0.67 to 0.95]; right: r = 0.98 [90%CI: 0.94 to 0.99]). Standardised mean bias for the left and right watches compared to the criterion were 0.14 (90%CI: -0.12 to 0.39; trivial) and 0.04 (90%CI: -0.07 to 0.15; trivial). Standardised typical error of the estimate for the left and right watches compared to the criterion were 0.51 (90%CI: 0.38 to 0.80; moderate) and 0.22 (90%CI: 0.16 to 0.34; small). Inter-device standardised typical error was 0.46 (90%CI: 0.36 to 0.68; moderate), ICC = 0.84 (90%CI: 0.65 to 0.93). The Apple Watch™ has good to very good criterion validity for measuring HRmax, with no substantial under- or over-estimation. There were moderate and small prediction errors for the left and right watches. Inter-device variability in HRmax is moderate.
Article
Heart rate variability (HRV) analysis during exercise is an interesting non-invasive tool to measure the cardiovascular response to the stress of exercise. Wearable heart rate monitors are a comfortable option to measure RR intervals while doing physical activities. It is necessary to evaluate the agreement between HRV parameters derived from the RR series recorded by wearable devices and those derived from an ECG during dynamic exercise of low to high intensity.23 male volunteers performed an exercise stress test on a cycle ergometer. Subjects wore a Polar RS800 device while ECG was also recorded simultaneously to extract the reference RR intervals. A time-frequency spectral analysis was performed to extract the instantaneous mean heart rate (HRM), and the power of low frequency (PLF) and high frequency (PHF) components, the latter centred on the respiratory frequency. Analysis was done in intervals of different exercise intensity based on oxygen consumption. Linear correlation, reliability and agreement were computed in each interval.The agreement between the RR series obtained from the Polar device and from the ECG is high throughout the whole test, although the shorter the RR is, the more differences there are. Both methods are interchangeable when analysing HRV at rest. At high exercise intensity, HRM and PLF still presented a high correlation (ρ>0.8) and excellent reliability and agreement indices (above 0.9). However, the PHF measurements from the Polar showed reliability and agreement coefficients around 0.5 or lower when the level of the exercise increases (for levels of O2 above 60%).
Conference Paper
The work presented in this paper aims at assessing human emotion recognition by means of the analysis of the heart rate variability (HRV) with varying spectral bands based on respiratory frequency (RF). Three specific emotional states are compared corresponding to calm-neutral state (Relax), positive elicitation (Joy) and negative elicitation (Fear). Standard HRV analysis in time and frequency domain is performed. In order to better characterize the HRV component related to respiratory sinus arrhythmia, the high frequency (HF) band is centered on RF. Results reveal that the power content in low band (PLF), the normalized power content in HF band (PHFn) and the sympathovagal ratio (LF/HF) can be suitable indices to distinguish Relax and Joy. Mean heart rate and RF are significantly different between Relax and Fear. Different HRV indices show significant differences between Joy and Fear, such as pNN50, PLF, PHFn and LF/HF. Statistical analysis of HRV indices with HF centered in the RF results in a lower p-value than the ones with a HF standard band.