Conference PaperPDF Available

Ultra-Short-Term Analysis of Heart Rate Variability for Real-time Acute Pain Monitoring with Wearable Electronics

2017 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)
978-1-5090-3050-7/17/$31.00 ©2017 IEEE 1025
Ultra-Short-Term Analysis of Heart Rate Variability for Real-time
Acute Pain Monitoring with Wearable Electronics
Mingzhe Jiang1, Riitta Mieronkoski2, Amir M. Rahmani3,4, Nora Hagelberg5,
Sanna Salanter¨
a2and Pasi Liljeberg1
1Department of Future Technologies, University of Turku, Finland
2Department of Nursing Science, University of Turku, Finland
3Department of Computer Science, University of California Irvine, USA
4Institute of Computer Technology, TU Wien, Austria
5Pain Clinic, Turku University Hospital, Finland
Email: {mizhji, ritemi},,, {sansala, pasi.liljeberg}
Abstract—In medical care, it is essential to assess and manage
acute painful conditions adequately. Heart rate variability (HRV)
analysis is based on the acquisition of electrocardiogram (ECG),
which is available from both patient monitor and wearable device.
As HRV analysis can reflect autonomic nervous system activity
which is unconsciously regulated, HRV analysis in ultra-short-
term is getting attention in indicating the reaction due to acute
pain. Different HRV features in different window lengths are
involved in pain monitoring studies as a signal index or part
of a multi-parameter model. In this work, seven HRV features
and median heart rate (HR) in ultra-short-term are evaluated for
their competence in indicating experimental acute pain. Also, the
choice of time window length in HRV analysis and its relation
with pain detection are discussed. The results of the normalized
HRV analysis from healthy volunteers show that the changes of
lnRMSSD, pNN20 and median HR associated with the intensity
of experimental electrical pain; and in the tests with experimental
thermal pain, lnLF and ln(LF/HF) changed along with pain
intensity. The fusion of the HRV features could tell pain from
no pain. With either experimental pain stimulation, optimal time
window length was observed around or larger than 40 seconds
with better correlation analysis result and HRV feature fusion
Keywords— Heart rate variability, pain assessment, wearable
device, experimental pain.
Pain assessment, as one of the critical points in pain
management, relates closely to the quality of care in hospital
and quality of life. Pain is common with a large population
of people suffering from it. One in every four Americans was
reported suffering from pain that lasts longer than 24 hours and
a lot more suffered from acute pain in the year of 2006 [1].
Similarly, by the end of May in 2015, between one-third and
one-half of the population in the UK were affected by chronic
pain where pain lasts longer than three months. In addition,
the proportion is likely to increase because of population aging
[2]. Undertreatment of pain can lead to adverse consequences
physiologically, psychologically and financially. For example,
poorly management of acute pain may cause serious complica-
tions and progress to chronic pain; also, under-treated chronic
pain influences daily activities and lowers the quality of life
In pain assessment, pain intensity assessment is essential
for pain management intervention and its evaluation. Patients
are guided by caregivers to describe their pain intensities on
a scale from 0 to 10, where 0 represents no pain, and 10
denotes the worst pain imaginable. Such self-report method
is considered the ” gold standard ” as pain is believed to
be a highly subjective experience, whereas it is impossible
to obtain self-report in uncommunicatable situations from
sedated or delirious patients for instance. In these situations,
behavioral indicators of pain are observed by the nurse as
an alternative way. No matter use which assessment tool,
timely reassessment of pain by inquiry or observation is
needed to assess treatment efficacy, but it barely can be real-
time especially for acute pain. Furthermore, these tools rely
heavily on nurses, their knowledge, interviewing techniques
and physical assessment skill [4]. Therefore, an automatic
pain intensity assessment tool is needed for real-time and
continuous pain monitoring.
One potential approach towards automatic pain assessment
is the fusion of physiological parameters, as multi-parameter
analysis is superior to any individual physiological parameter
[5]–[8]. Automatic pain assessment with physiological param-
eters has the advantage of leveraging wearable devices. With
wearable devices, human physiological parameters are measur-
able and recordable even in everyday life. Wearable devices
provide the foundation of health monitoring especially when
integrated into an Internet-of-things system which enables
wearable devices network with mobile devices, fog [9], as
well as cloud servers. Such system may support various real-
time health monitoring services by continuously processing
and analyzing data from wearable devices.
Among the physiological signals available in wearable de-
vices, signals that influenced by sympathetic tone are consid-
ered useful to indicate the perception of pain, and electro-
cardiogram (ECG) is one of them. From ECG record, heart
rate (HR) and heart rate variability (HRV) can be derived to
index sympathetic and parasympathetic nervous system activ-
ity when internal body functions are involuntarily regulated.
HR, as one of the vital signs, is the most frequently observed
physiological signal in pain studies. While HRV features
measuring the variation time interval between heartbeats with
different methods are observed individually and selectively in
some pain studies.
Therefore, HRV features in several categories are examined
in this paper with their intercorrelation and correlation with
pain for future physiological data fusion in pain detection
or pain intensity assessment, serving as a reference in HRV
features selection. In addition, HRV analysis in ultra-short-
term is performed in this study instead of classical short-term
in 5-minute windows and long-term in 24-hour time windows
to address timely assessment and reassessment of acute pain.
In the rest of paper, HRV analysis methods and their
relations with pain are introduced in Section II. Section III
presents the study protocol designed for this study, where 1-
lead ECG was recorded with a wearable strap around chest
from thirty healthy volunteers under two experimental pain
stimuli. Signal processing and data analysis methods are also
presented in Section III. The following Section IV reveals
the observations in HRV features as well as classification
performance with selected HRV features. In the end, Section
V concludes and discusses the study.
Extracting HRV features from ECG starts with QRS de-
tection. Only inter-beat intervals between adjacent normal R
peaks are kept for HRV analysis. Those normal inter-beat
intervals are usually referred as NN intervals in the literature.
HRV features are measured over a period, typically in short-
term over 5 minutes or in long-term over 24 hours. However,
some HRV features in less than 1 minute as ultra-short-term
analysis are also adopted, but in tracing acute psychologi-
cal and physiological changes [10]–[12]. Within each time
window, HRV features are calculated with numerous methods
falling in three categories and they are i) time domain analysis,
ii) frequency domain analysis and iii) nonlinear analysis [13].
HRV features in the time domain are statistically calculated
from the interval lengths within each time window. Some
common short-term time domain HRV features are:
AVNN: average of NN intervals
SDNN: standard deviation of NN intervals
RMSSD: root mean square of differences between adja-
cent NN intervals
pNNx: percentage of differences between adjacent NN
intervals that are greater than xmilliseconds;
HRV features in the frequency domain analysis are extracted
from the spectrum of NN intervals in the corresponding time
axis. To get the power spectrum with fast Fourier transform,
the interval sequence can be evenly re-sampled. Alternatively,
Lomb-Scargle periodogram [14] can be used directly to un-
evenly sampled data. Some common short-term frequency
domain HRV features are:
LF: low-frequency component, total spectral power be-
tween 0.01 and 0.15 Hz
HF: high-frequency component, total spectral power be-
tween 0.15 and 0.04 Hz
LF/HF: ratio of low to high-frequency power
where LF and HF are considered to index sympathetic and
parasympathetic activity, respectively [10], [15]. Besides, cor-
relations between SDNN and LF as well as RMSSD and
HF were observed in healthy subjects, and the former was
assumed as a surrogate for the latter [16]. The third category
of HRV features uses nonlinear methods, which are considered
more effective to describe the process generated by nonlinear
biological systems. Among these methods, entropy in NN
intervals is looked into in this study, where ApEn (approximate
entropy) measures the disorder in NN intervals and SampEn
(sample entropy) is similar to ApEn but better in studying the
dynamics of human cardiovascular physiology [13].
HRV analysis in ultra-short-term has been applied as a
path to measure pain in different types. For example, it was
observed that standard deviation of Poincare plot, HR, and
RMSSD in 1-minute windows were significantly different
between eight chronic low back pain patients and eight healthy
controls during movements [12]. Regarding acute pain, a
finding was presented in Sesay et al. ’s work [10], where LF
and LF/HF in 30-second epochs significant increased among
120 patients when pain intensity was larger than 3 (numeric
rating scale) after minor spinal surgery, but no significant
change was found in HF. HF in every 1 minute was involved
in [5] to monitor the clinical pain level with multi-parameter
regression method.
Relatively, more studies were carried out with experimen-
tally induced acute pain such as thermal pain or electrical pain.
By reviewing the studies on HRV features in experimentally
induced pain, LF was concluded to be valuable in response
to pain induction, whose increase indicated an increase in
sympathetic-baroreflex activity [15]. Also, the decrease of HF
was highlighted to index a decrease in vagal-parasympathetic
activity. In addition to extracting HRV features, a signal
processing approach was developed to recognize experimental
pain [17]. However, current studies also show that individual
HRV feature is not capable to differ pain intensities or the
subgroups of pain intensities although HRV analysis has
potential in indicating the presence of pain [10], [18].
In this study, tests with experimental pain were carried out
on healthy volunteers. From the gathered ECG records, ultra-
short-term HRV features in different lengths were extracted
to check the correlations between each of them and the
presence of pain and compare the result with the studies
mentioned above. HRV features were then compared and
selected for the fusion in pain detection. The performance
of HRV features fusion was presented as receiver operating
characteristic (ROC) curve of classification.
Fifteen male and fifteen female volunteers in healthy con-
dition with an average age of 35 (SD = 8) and 33 (SD = 11.9)
were included in data analysis. The exclusion criteria include
chronic, acute somatic and mental illness. Besides, cases with
regular medication taken two weeks before or during the study
were avoided as well. This study was approved by the Ethics
Committee of the Hospital District of Southwest Finland. The
study was designed as follows:
A. Two experimental pain stimuli
Two types of experimental pain were employed in the tests
to stimulate two types of sensory receptor. Electrical pain
stimulus was delivered to the fourth finger of one hand with
a digital transcutaneous electrical nerve stimulation (TENS)
device [19]. The device generated biphasic rectangular pulses
with a width of 250 microseconds and a repeat frequency of
100 Hz to cause pain. The voltage of the pulse can be incre-
mented and decremented between 0 and 50, where the peak
to peak output voltage in full scale was 100 volts. The second
pain stimulus was thermal pain delivered to inner forearm skin
from a heating element with a diameter of 3 centimeters. The
temperature of the element surface was controlled to increase
approximately linearly from room temperature and up to 52
degrees centigrade to avoid skin burn injury.
Each experimental pain stimulus was applied to the right
and left side respectively, and thereby four tests were carried
out on each volunteer. The four tests were conducted succes-
sively in a random order. The following test proceeded only
when the HR of the subject was observed returning to his or
her baseline if there was any change.
B. Test process
The subject was guided to first settle down in an armchair
for around 10 minutes. After that, one pain stimulus was
applied to him or her as the start of one test. The process of
each test and data definition are presented in Figure 1. From
the start time at t0, the intensity of pain stimulus increased
with a speed of 1 TENS level every 3 seconds or 1C every
3 to 5 seconds until the volunteer reported reaching his or her
pain tolerance at t2and the pain stimulus was then removed.
During this process, the volunteer could indicate his or her
pain threshold at t1by pressing a button which activated a
Data in the time period without pain stimulus was defined
as No pain; data between t0and t1was defined as Mild pain,
which was equivalent to pain intensity below 3 and 4 in visual
analog scale [20]; data between t1and t2was defined as
Moderate/Severe pain, where the corresponding pain intensity
in visual analog scale was between 5 and 10.
C. ECG measurement
During the whole experiment, the volunteer was wearing
the Zephyr R
Bioharness 3 wearable sensor and sitting in
an armchair. The device can capture 1-lead ECG from chest
area at a sample rate of 250 Hz with 12-bit analog-to-digital
resolution. Its Bluetooth connection with a laptop and PC-
based software ensured real-time wireless data transmission
and real-time waveform visualization. Raw ECG records were
saved as files for off-line processing.
Fig. 1. Experimental study design
D. ECG signal processing
The aim of signal processing with ECG records in this study
was to detect QRS wave in ECG waveform to extract NN
intervals. Four steps were conducted in ECG processing:
1) 10-point moving average filter was applied to the whole
record to remove the baseline wander in low frequency.
2) R peaks were detected with a threshold-based method,
where the threshold was adaptively defined based on
the maximum amplitude of the waveform. Moreover, for
further precise R peak detection especially from P wave
and T wave, minimum peak distance was added as a
constraint according to normal heart rate range.
3) After the one round of detection, the detected R peaks
were marked in the plotted ECG waveform, and they
were manually validated in the third step. The distorted
parts of the ECG record were abandoned, which are
mostly at the beginning of the experiment due to device
adjustment. Also in this step, several omitted R peaks
were manually added.
4) Validated NN intervals were calculated from the po-
sitions of all detected R peaks. Meanwhile, all NN
intervals were labeled as 1-No pain,2-Mild pain and
3-Moderate/Severe pain respectively according to time
stamps and the data definition in Figure 1.
E. HRV features extraction
Seven HRV features were evaluated including SDNN,
RMSSD, pNNx, LF, HF, LF/HF and ApEn, which were from
three categories as introduced in Section II. In the process
of feature extraction, NN intervals were HR normalized by
dividing by their average (AVNN) due to mathematical bias
in NN intervals when referring different HRs [21]. This
normalization is emphasized especially among people with
different average HRs or during interventions that change HR
[22]. The HR normalization narrows the gap between the big
NN interval change at low heart rate and small NN interval
change at high heart rate with the same amount of heart rate
change. When extracting frequency domain features, Lomb-
Scargle periodogram was applied as the spectral estimation
method. Besides these HRV features, median HR was also
As the intensity of each pain stimulus increased pro-
gressively, HRV features in ultra-short time windows were
extracted to track their dynamic change. HRV features in
different time window lengths were first observed separately
regarding their correlation with pain intensity. The window
length was defined from 10 seconds to 60 seconds increasing
in a step of 10 seconds. The HRV features were labeled as
the most frequently occurring label among NN intervals in
the time window. The sample sizes of each HRV feature in
different time lengths are listed in Table I, where the sample
size of 1-No pain was the largest and the sample size of 2-
Mild pain in electrical pain tests were small due to the short
transition from no pain to pain threshold.
Stimuli Label HRV features window length
type 10 s 20 s 30 s 40 s 50 s 60 s
1 1903 964 646 481 389 321
2 89 44 21 16 11 11
3 220 109 77 60 48 40
1 2251 1125 747 564 452 375
2 411 209 142 102 80 74
3 244 113 77 58 47 31
To achieve normal distributions among the HRV features,
SDNN, RMSSD, LF, HF and LF/HF were logarithmically
transformed with natural logarithm [18]. Therefore, they are
denoted as lnSDNN, lnRMSSD, lnLF, lnHF and ln(LF/HF)
in Section IV. The signal processing and HRV analysis were
implemented in Matlab.
The root mean squares (RMSs) of HRV features in different
time windows were first checked for its change with pain
intensities. Several observations are obtained from Figure 2.
Firstly, HRV features in electrical tests reacted more dramati-
cally than those in thermal tests in general. Take median HR as
an example. It changed from 70 bpm during No Pain to around
85 bpm during Moderate/Severe pain in electrical tests, while
increased to around 75 bpm in thermal tests. This difference
may validate that different pain-conducting nerve fibers were
excited by these two different pain stimuli. Secondly, the
relative relations of HRV features in different pain intensities
may change with the length of the time window. In electrical
tests, RMS of 10 s lnSDNN increased as pain intensity
increase. However, it first went down and then went back to
a close value when the time window length was 50 s. Similar
pattern change can also be observed in ln(LF/HF) and median
HR in both types of tests. Thirdly, pNN20 was responsive
from No Pain to pain, but it tended to recover despite the
ongoing pain stimulus. Last but not least, lnRMSSD, lnHF and
median HR were found changing along with pain intensity in
10 s 20 s 30 s 40 s 50 s 60 s
Electrical tests
10 s 20 s 30 s 40 s 50 s 60 s
Thermal tests
No pain Mild pain Moderate/Severe pain
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
10 s 20 s 30 s 40 s 50 s 60 s
Fig. 2. Comparison of HRV features in different window lengths and different pain intensities. RMSs are compared in tests with the same pain stimulus.
Electrical pain, window length Thermal pain, window length
HRV feature 10 s 20 s 30 s 40 s 50 s 60 s 10 s 20 s 30 s 40 s 50 s 60 s
lnSDNN -0.1623*-0.1504 -0.1021 NR** NR NR -0.1115 -0.1333 -0.1458 -0.1491 -0.1280 -0.1086
lnRMSSD -0.2338*** -0.2545 -0.2400 -0.2617 -0.2766 -0.2334 -0.0897 -0.1044 -0.1266 -0.1123 -0.1175 -0.0953
pNN20 -0.1894 -0.2437 -0.2894 -0.3313 -0.3117 -0.3471 NR -0.0533 -0.0740 NR NR NR
lnLF -0.1094 -0.1309 -0.1309 -0.1268 -0.1528 -0.1036 -0.1354 -0.1785 -0.2080 -0.2192 -0.2119 -0.2280
lnHF -0.1569 -0.1696 -0.1596 -0.1667 -0.1546 -0.1403 NR NR NR NR NR NR
ln(LF/HF) NR NR NR NR NR NR -0.1406 -0.1919 -0.2060 -0.2369 -0.2064 -0.2551
ApEn 0.1355 0.1186 0.1522 0.1452 NR NR NR NR NR NR NR NR
median HR 0.3140 0.3363 0.3293 0.3226 0.3108 0.3495 0.1152 0.1198 0.1311 0.1327 0.1586 0.1106
Absolute val-
ues sum**** 1.0035 1.1350 1.1493 1.2090 1.2064 1.1738 0.5923 0.7278 0.8175 0.8502 0.8224 0.7976
*Underlined coefficient: statistically highly significant (p<0.001)
** NR: not related. Hypothesis of no correlation: p>0.05
*** Coefficient in bold: HRV features that were commonly correlated with pain intensities in all window lengths
**** lnRMSSD, pNN20, lnLF, lnHF and median HR in electrical tests; lnSDNN, lnRMSSD, lnLF, ln(LF/HF) and median HR in thermal tests.
electrical tests, while in thermal tests those were lnSDNN,lnLF
and median HR.
The correlations between each HRV feature and the three
pain categories in increasing pain intensities were then further
looked into with Pearson correlation coefficient r, which are
presented in Table II. The correlation whose p-value larger
than 0.05 was considered as not related, and was marked as
NR. The correlations that were statistically highly significant
(p<0.001) were highlighted with underlines. Additionally,
for HRV features that were commonly correlated with pain
intensities with all window lengths, their coefficients were
highlighted with bold font. In electrical pain tests, HRV
features including lnRMSSD, pNN20, lnLF, lnHF and median
HR each showed small (rwas between 0.1 and 0.29) to
medium (rwas between 0.3 and 0.49) correlation with the
three pain categories. While there was a difference in the
thermal tests, where lnSDNN, lnRMSSD, lnLF, lnLF/HF and
median HR showed small correlation. The coefficients of these
correlated HRV features were marked in bold font, and their
absolute values in the same window length were summed.
The highlighted HRV features were roughly consistent with
the finding in Figure 2. pNN20 and lnLF in electrical tests
as well as lnRMSSD and ln(LF/HF) in thermal tests were
also found potentially useful from Table II in addition to
the finding in RMS change from Figure 2. The sum of the
highlighted HRV features showed that the overall correlation
between HRV features and pain detection increased along with
the increase of window length from 10 s to 40 s. After 40 s,
the absolute values sum dropped slightly, but still better than
that in shorter window length.
The relative relations among HRV features were next
analyzed with Pearson correlation and principal component
analysis (PCA). HRV features in 40 s window length were
analyzed as a representative, and the PCA biplots in separate
pain stimuli tests are presented in Figure 3. The similarity
between lnSDNN and lnLF can be observed in Figure 3
(r=0.87 in electrical tests and 0.89 in thermal tests). A similar
similarity was found between lnRMSSD and lnHF (r=0.90 in
electrical tests and 0.94 in thermal tests). However, normalized
lnSDNN and lnLF or lnRMSSD and lnHF in ultra-short-term
performed inconsistently in the tests, as summarized from
Figure 2 and Table II. Therefore, although there is strong linear
correlation between lnSDNN and lnLF as well as between
lnRMSSD and lnHF, the HRV analysis in time domain and in
Component 1
-10 0 10
Component 2
median HR
Electrical pain tests
Component 1
-20 -10 0 10 20
Component 2
median HR
Thermal pain tests
Fig. 3. Principal component analysis biplots of HRV features in 40 s time window
frequency domain are not equivalent.
Then, the HRV features highlighted in Table II were fused
with support vector machine (SVM) classifier. Multi-class
SVM classifier was first applied to the labeled HRV features
in three categories, but the result was unsatisfactory where
the overall classification accuracy in 10-fold cross validation
was around 50%. Relatively, the binary classification results
were more valuable in pain detection when rearranging the
data labels as No pain and Pain. The binary classification
results are shown in Figure 4 and Figure 5 with ROC curve
and area under ROC curve (AUC). In accordance with prior
observations, electrical pain stimulus was more predictable by
HRV analysis than thermal pain stimulus because AUC with
the former tests was higher than that with the latter tests in
the same length of the time window. Moreover, either pain
stimulus was better predicted with HRV features as the time
window length increased.
0 0.2 0.4 0.6 0.8 1
1Electrical pain tests
10 s; AUC=0.72
20 s; AUC=0.74
30 s; AUC=0.72
40 s; AUC=0.82
50 s; AUC=0.82
60 s; AUC=0.78
Fig. 4. Results of No pain and Pain classification in electrical pain tests
0 0.2 0.4 0.6 0.8 1
1Thermal pain tests
10 s; AUC=0.68
20 s; AUC=0.7
30 s; AUC=0.7
40 s; AUC=0.74
50 s; AUC=0.72
60 s; AUC=0.75
Fig. 5. Results of No pain and Pain classification in thermal pain tests
To observe how the combination of HRV features influences
the classification performance, each possible combination of
features in each window length was trained and tested by a
binary SVM classifier. AUC was calculated as the performance
measure. The results are presented as two heat maps shown
in Figure 6. The total number of all combinations is 255. It
includes the 8 cases of choosing 1 feature from the total 8
features (C1
8), 28 cases of choosing 2 features (C2
8), etc. The
ROC curves in Figure 4 and 5 are from one case in the C6
combinations. To highlight the combinations having the best
performance, AUCs higher than 0.8 in electrical pain tests and
AUCs higher than 0.75 in thermal pain tests are presented
in red color. Figure 6 shows that 40 s and 50 s windows
fit electrical pain tests better, while 60 s window fit thermal
pain tests better. In terms of different feature combinations,
the more features involved, the better chance to have good
performance. The combination of lnSDNN, lnRMSSD and
ln(LF/HF) reached similar good performance with the least
number of features in both pain tests, where AUC(40, 40)
= 0.8001 and AUC(40, 50) = 0.8066 in electrical pain tests,
AUC(40, 60) = 0.7539 in thermal pain tests.
Two experimental pain models in the skin are employed
in this study. Thermal stimulation by slow heating that
less than 1C/s gives a preferential activation of C fibers,
while electrical stimulation from TENS device activates non-
nociceptive nerve fibers and also nociceptive nerve fibers
directly bypassing nociceptors [23]. Human experimental pain
models offer the possibility to explore the pain system under
controlled settings and help understand pain mechanism [23],
[24]. Usually, experimental pain tests on healthy volunteers
are restricted in time with short pain stimulus in seconds
or several minutes and the choice of time window length in
HRV analysis is limited. For example, in BioVid database [6],
thermal pain stimulus was maintained for 4 s, and it stopped
for 8-12 s. In this case, 10 s time window was chosen for
HRV analysis in time domain [25]. In this study, the time
length of each test is determined by subject, so it varies among
subjects and between pain stimuli. In electrical pain tests,
Mild pain and Moderate/Severe pain took an average of 16
s and 37 s. In thermal pain tests, they took an average of
68 s and 50 s separately. With the same processing method,
electrical pain is better identified from no pain than thermal
pain when comparing Figure 4 and 5. In both types of tests,
classification performance improves slightly with the increase
of time window length in terms of AUC.
Independent of time window length in ultra-short-term
analysis, there are some HRV features showing potentials
in pain detection. Table II shows that lnRMSSD, pNN20,
and median HR are correlated with electrical pain stimulus
with high significance independent of time window length
and lnLF and ln(LF/HF) are correlated with thermal pain
stimulus with high significance. The decrease of RMSSD
and/or HF is in line with the previous conclusion that it indexes
a decrease in parasympathetic activity [15]. LF and LF/HF are
observed decrease with the increase of thermal pain, which is
contrary to previous studies where they increased [10], [15].
However, the consensus is reached that LF and LF/HF are
Fig. 6. AUCs of classifications with all possible feature combinations
potentially valuable HRV features in pain detection despite
the disagreement in trend.
In healthy volunteers, several HRV features in ultra-short-
term within 1 minute changed accordingly with the intensity
of experimental acute pain stimulation. In the tests with elec-
trical stimulation activating nociceptive nerve fibers directly,
lnRMSSD, pNN20 and median HR showed correlation higher
than the others with the three pain categories. While in the
test with thermal stimulation where C fibers were activated,
lnLF and ln(LF/HF) showed a higher correlation with pain
intensity than the other HRV features. These HRV features
were small to medium linearly correlated to pain intensity.
However, the multiple feature classification results did not
show strong evidence in indicating pain intensity. Instead, they
can potentially indicate the change from No pain to Pain.
Furthermore, from a real-time point of view in acute pain
monitoring, a better HRV feature fusion performance were
achieved when time window length was around or larger than
40 s. The findings in this paper support ultra-short-term HRV
features in acute pain monitoring and also provide a reference
in HRV analysis method selection and time window length
selection when less than 1 minute.
[1] National Institutes of Health, “NIH Fact Sheets-Pain Management.”
[2] A. Fayaz, P. Croft, R. Langford, L. Donaldson, and G. Jones, “Prevalence
of chronic pain in the UK: a systematic review and meta-analysis of
population studies,” BMJ open, vol. 6, no. 6, p. e010364, 2016.
[3] P. H. Berry, C. Chapman, E. Covington, J. Dahl, J. Katz, C. Miaskowski,
and M. McLean, “Pain: current understanding of assessment, manage-
ment, and treatments,” National Pharmaceutical Council and the Joint
Commission for the Accreditation of Healthcare Organizations, VA,
USA, p. b44, 2001.
[4] E. Wuhrman and M. F. Cooney, “Acute pain: Assessment and treatment,”
[5] N. Ben-Israel, M. Kliger, G. Zuckerman, Y. Katz, and R. Edry, “Mon-
itoring the nociception level: a multi-parameter approach,Journal of
clinical monitoring and computing, vol. 27, no. 6, pp. 659–668, 2013.
[6] S. Gruss, R. Treister, P. Werner, H. C. Traue, S. Crawcour, A. Andrade,
and S. Walter, “Pain intensity recognition rates via biopotential feature
patterns with support vector machines,” PloS one, vol. 10, no. 10,
p. e0140330, 2015.
[7] R. Treister, M. Kliger, G. Zuckerman, I. G. Aryeh, and E. Eisenberg,
“Differentiating between heat pain intensities: the combined effect of
multiple autonomic parameters,” PAIN, vol. 153, no. 9, pp. 1807–1814,
[8] M. Huiku, K. Uutela, M. Van Gils, I. Korhonen, M. Kym¨
P. Meril ¨
ainen, M. Paloheimo, M. Rantanen, P. Takala, H. Vierti¨
et al., “Assessment of surgical stress during general anaesthesia,” British
journal of anaesthesia, vol. 98, no. 4, pp. 447–455, 2007.
[9] Y. Shi, G. Ding, H. Wang, H. E. Roman, and S. Lu, “The fog computing
service for healthcare,” in International Symposium on Future Infor-
mation and Communication Technologies for Ubiquitous HealthCare,
pp. 1–5, IEEE, 2015.
[10] M. Sesay, G. Robin, P. Tauzin-Fin, O. Sacko, E. Gimbert, J.-R. Vi-
gnes, D. Liguoro, and K. Nouette-Gaulain, “Responses of heart rate
variability to acute pain after minor spinal surgery: optimal thresholds
and correlation with the numeric rating scale,” Journal of neurosurgical
anesthesiology, vol. 27, no. 2, pp. 148–154, 2015.
[11] L. Salahuddin, J. Cho, M. G. Jeong, and D. Kim, “Ultra short term
analysis of heart rate variability for monitoring mental stress in mobile
settings,” in 29th Annual International Conference of the IEEE Engi-
neering in Medicine and Biology Society, pp. 4656–4659, IEEE, 2007.
[12] A. J. Hautala, J. Karppinen, and T. Sepp¨
anen, “Short-term assessment
of autonomic nervous system as a potential tool to quantify pain
experience,” in 38th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society, pp. 2684–2687, IEEE,
[13] U. R. Acharya, K. P. Joseph, N. Kannathal, C. M. Lim, and J. S. Suri,
“Heart rate variability: a review,” Medical and biological engineering
and computing, vol. 44, no. 12, pp. 1031–1051, 2006.
[14] P. Laguna, G. B. Moody, and R. G. Mark, “Power spectral density
of unevenly sampled data by least-square analysis: performance and
application to heart rate signals,” IEEE Transactions on Biomedical
Engineering, vol. 45, no. 6, pp. 698–715, 1998.
[15] J. Koenig, M. N. Jarczok, R. J. Ellis, T. Hillecke, and J. F. Thayer, “Heart
rate variability and experimentally induced pain in healthy adults: A
systematic review,” European Journal of Pain, vol. 18, no. 3, pp. 301–
314, 2014.
[16] H.-M. Wang and S.-C. Huang, “SDNN/RMSSD as a surrogate for
LF/HF: a revised investigation,Modelling and Simulation in Engineer-
ing, vol. 2012, p. 16, 2012.
[17] S. Tejman-Yarden, O. Levi, A. Beizerov, Y. Parmet, T. Nguyen, M. Saun-
ders, Z. Rudich, J. C. Perry, D. G. Baker, and T. Moeller-Bertram, “Heart
rate analysis by sparse representation for acute pain detection,” Medical
& biological engineering & computing, vol. 54, no. 4, pp. 595–606,
[18] J. J. Meeuse, M. S. L¨
owik, S. A. L¨
owik, E. Aarden, A. M. Roon,
R. O. Gans, M. Wijhe, J. D. Lefrandt, and A. K. Reyners, “Heart rate
variability parameters do not correlate with pain intensity in healthy
volunteers,” Pain Medicine, vol. 14, no. 8, pp. 1192–1201, 2013.
[19] “Sanitas SEM43 Digital EMS/TENS.” http://www.sanitas- dokumente/GAs/therapie/752.907-0212 SEM43.pdf.
[20] G. A. Hawker, S. Mian, T. Kendzerska, and M. French, “Measures
of adult pain: Visual analog scale for pain (vas pain), numeric rating
scale for pain (nrs pain), mcgill pain questionnaire (mpq), short-form
mcgill pain questionnaire (sf-mpq), chronic pain grade scale (cpgs), short
form-36 bodily pain scale (sf-36 bps), and measure of intermittent and
constant osteoarthritis pain (icoap),” Arthritis care & research, vol. 63,
no. S11, pp. S240–S252, 2011.
[21] J. Sacha, “Why should one normalize heart rate variability with respect
to average heart rate,Frontiers in physiology, vol. 4, 2013.
[22] G. E. Billman, H. V. Huikuri, J. Sacha, and K. Trimmel, “An introduction
to heart rate variability: methodological considerations and clinical ap-
plications,” Heart Rate Variability: Clinical Applications and Interaction
between HRV and Heart Rate, p. 6, 2015.
[23] A. E. Olesen, T. Andresen, C. Staahl, and A. M. Drewes, “Human
experimental pain models for assessing the therapeutic efficacy of
analgesic drugs,” Pharmacological reviews, vol. 64, no. 3, pp. 722–779,
[24] C. Staahl and A. M. Drewes, “Experimental human pain models: a
review of standardised methods for preclinical testing of analgesics,
Basic & clinical pharmacology & toxicology, vol. 95, no. 3, pp. 97–
111, 2004.
[25] F. M. Lima, S. Walter, S. Schach, H. C. Traue, and A. O. Andrade, “Pain-
related heart rate variability responses in the context of experimental pain
stimulation,” in IX Simpsio de Engenharia Biomdica, pp. 42–45, 2016.
... One study (18) did not even report patients' mean age or age range. All the studies reporting the patients' age range involved younger adults while only 7 (8,9,13,(19)(20)(21)(22) included middle-aged adults and only one (22) involved patients younger than 18, even though studies with children and adolescents were not systematically excluded. As shown in Fig. 4a, half of the studies reported sample sizes ranging from 20 and 50 patients, while only one document (9) reported a sample size greater than 100. ...
... The vast majority of studies (24/26 = 92.3%) reported the use of one single nociceptive stimulation modality (i.e., thermal, mechanical, electrical, or chemical), and only 2 articles (7,20) reported the use of multiple modalities. Thermal stimuli, either heat, cold, or both, were used in more than half of the included studies (18/26 = 69.2%; ...
... Thermal stimuli, either heat, cold, or both, were used in more than half of the included studies (18/26 = 69.2%; Fig. 5), and while only cold water immersion (i.e., cold pressor test) was used for inducing pain by cold, different techniques were used to deliver noxious heat stimuli (contact heating element: [8,9,11,12,15,16,[19][20][21]; hot water immersion: [17,30]; laser-induced heat: [7]). Mechanical, electrical, and even chemical stimulation were used in only 6 (6,7,10,14,18,27), 4 (7,13,20,23), and one (25) study, respectively. ...
Background: Pain is essential for survival, but it is also a major clinical, social, and economic problem that demands adequate management. The latter involves timely and accurate assessment, so several efforts have been made to develop accurate and reliable pain assessment tools. Advances in objective pain assessment include a large body of work focused on determining whether autonomic-mediated peripheral responses can be used to predict pain intensity. However, there is still no clinically validated autonomic marker for objective pain assessment. Objectives: In order to identify possible causes of this situation, the present study reviews the most recent advances examining peripheral autonomic markers' ability to describe pain intensity. Study design: Systematic literature review. Methods: We conducted an online search on PubMed using terms such as "pain assessment," "experimental pain," "autonomic arousal," "heart rate," "heart rate variability," "electrodermal activity," "pupillary diameter," and "blood pressure." Articles published from 2010 through 2020 examining the abilities of peripheral autonomic markers to describe experimental pain intensity were collected and reviewed. From each of the included studies, we extracted information regarding autonomic parameters and stimulation modalities used by experimenters, as well as the sample size, gender, and health condition of the patients. Results: Twenty-six articles were included for analysis, from which only 2 studies reported the use of multiple modalities. Half of the documents reported sample sizes ranging from 20 to 50 patients, and only 3 studies used formal power calculation to determine the sample size. Most of the articles included only healthy patients, so the influence of age, gender, and pre-existing health conditions on the autonomic peripheral parameters' capabilities to reflect the experience of pain remains unexplored. Limitations: It is possible that several documents were not retrieved due to a potential search engine bias or the use of very specific terms. Furthermore, only studies reporting pain intensity as a unique measure of its severity were included. Conclusion: The measurement of autonomic responses elicited by experimentally induced pain is one crucial step toward the development of reliable pain assessment tools. Still, several issues need to be addressed before continuing to explore the use of autonomic parameters for the assessment of pain. It is also recommended that future research endeavors in capturing the singularity of the pain experience involve the measurement of both peripheral (end organs) and central (brain) autonomic responses to pain.
... Heart rate (HR) and heart rate variability (HRV) are the essential parameters that can be derived from ECG, as they are both coupled to autonomic nervous system activity, when internal body functions are involuntarily regulated, and they can provide a suitable proxy for examining pain intensity [16]. The most frequently used vital sign in pain studies is HR as the number of heartbeats, while HRV features with the extent to which the heart rate changes over a time interval or the extent to which it is spread over different frequencies are observed individually and selectively in some pain studies [17][18][19]. There are several approaches for classifying pain intensity of healthy subjects using machine learning techniques. ...
... To monitor the nociception level of patients with multiple physiological parameters, HF in a 1-minute window was calculated in Ben-Israel et al [11]. Jiang et al [17] experimented with the correlation of HRV features in the ultrashort term with acute pain. They suggested that multiple HRV features can indicate the change from no pain to pain. ...
Full-text available
Background There is a strong demand for an accurate and objective means of assessing acute pain among hospitalized patients to help clinicians provide pain medications at a proper dosage and in a timely manner. Heart rate variability (HRV) comprises changes in the time intervals between consecutive heartbeats, which can be measured through acquisition and interpretation of electrocardiography (ECG) captured from bedside monitors or wearable devices. As increased sympathetic activity affects the HRV, an index of autonomic regulation of heart rate, ultra–short-term HRV analysis can provide a reliable source of information for acute pain monitoring. In this study, widely used HRV time and frequency domain measurements are used in acute pain assessments among postoperative patients. The existing approaches have only focused on stimulated pain in healthy subjects, whereas, to the best of our knowledge, there is no work in the literature building models using real pain data and on postoperative patients. Objective The objective of our study was to develop and evaluate an automatic and adaptable pain assessment algorithm based on ECG features for assessing acute pain in postoperative patients likely experiencing mild to moderate pain. Methods The study used a prospective observational design. The sample consisted of 25 patient participants aged 18 to 65 years. In part 1 of the study, a transcutaneous electrical nerve stimulation unit was employed to obtain baseline discomfort thresholds for the patients. In part 2, a multichannel biosignal acquisition device was used as patients were engaging in non-noxious activities. At all times, pain intensity was measured using patient self-reports based on the Numerical Rating Scale. A weak supervision framework was inherited for rapid training data creation. The collected labels were then transformed from 11 intensity levels to 5 intensity levels. Prediction models were developed using 5 different machine learning methods. Mean prediction accuracy was calculated using leave-one-out cross-validation. We compared the performance of these models with the results from a previously published research study. Results Five different machine learning algorithms were applied to perform a binary classification of baseline (BL) versus 4 distinct pain levels (PL1 through PL4). The highest validation accuracy using 3 time domain HRV features from a BioVid research paper for baseline versus any other pain level was achieved by support vector machine (SVM) with 62.72% (BL vs PL4) to 84.14% (BL vs PL2). Similar results were achieved for the top 8 features based on the Gini index using the SVM method, with an accuracy ranging from 63.86% (BL vs PL4) to 84.79% (BL vs PL2). Conclusions We propose a novel pain assessment method for postoperative patients using ECG signal. Weak supervision applied for labeling and feature extraction improves the robustness of the approach. Our results show the viability of using a machine learning algorithm to accurately and objectively assess acute pain among hospitalized patients. International Registered Report Identifier (IRRID) RR2-10.2196/17783
... Widely used contact sensors in eHealth applications record electroencephalogram (EEG, electrical activity of the brain), electrocardiogram (ECG, heart activity (heart rate (HR) and heart rate variability (HRV))), electrodermal activity (EDA often measured using skin conductance level (SCL), and sometimes the old term "galvanic skin response" (GSR)), surface electromyogram (sEMG, muscle activity), photoplethysmogram (PPG, blood perfusion of the skin for pulse and other measures, also called blood volume pulse or BVP), respiration (RSP), or acceleration (ACC, movement). For pain monitoring, the uni-dimensional assessment tools have been questioned and debated for their oversimplification and limited applicability in non-communicative patients, since they require interactive communication between patient and caregiver [26]. As a result, physiological sources of data comprising of heart rate (HR), heart rate variability (HRV), SpO2, skin temperature, electrodermal activity (EDA), and facial expression and frontal muscle activity using computer vision or facial electromyography (EMG) and electroencephalogram (EEG) are prioritized for pain assessment. ...
Full-text available
Smart eHealth applications deliver personalized and preventive digital healthcare services to clients through remote sensing, continuous monitoring, and data analytics. Smart eHealth applications sense input data from multiple modalities, transmit the data to edge and/or cloud nodes, and process the data with compute intensive machine learning (ML) algorithms. Run-time variations with continuous stream of noisy input data, unreliable network connection, computational requirements of ML algorithms, and choice of compute placement among sensor-edge-cloud layers affect the efficiency of ML-driven eHealth applications. In this chapter, we present edge-centric techniques for optimized compute placement, exploration of accuracy-performance trade-offs, and cross-layered sense-compute co-optimization for ML-driven eHealth applications. We demonstrate the practical use cases of smart eHealth applications in everyday settings, through a sensor-edge-cloud framework for an objective pain assessment case study.
... Moreover, HRV analysis has also been widely used to explain players' behavioral responses when interacting with games and VR applications, complementing the conventional methods to capture perceived user experiences [14]. In non-VR settings, the real-time analysis of HRV is a critical approach to determine psychological aspects of a person, such as pain [23], mental workload [18], or stress [15,36]. However, measurement of heart activity in VR systems often poses challenges (e.g., due to wired connections and data synchronization) that interfere with the natural interaction of the user; an example can be found in a project that estimated stress for VR pain management using HRV biofeedback [7]. ...
... In the future, we consider that collecting ECG and EDA in real-life applications tends to require the use of UST analysis. We found that UST analysis gives good results in monitoring emotional states [55] and acute pain [56]. Still, UST requires researchers to test for its validity. ...
Full-text available
This review investigated research works on affective computing by using electrocardiogram (ECG) and electrodermal activity (EDA). The 27 related research papers, including 23 from IEEE Journals and 4 from other Q1 Journals in the last five years, were studied. The main goals have been to summarize common trends in this field in recent years as well as provide discussions and a big picture of how future research should be done. The content of this review covers the fundamental pipeline of affective computing research consisting of stimuli, intelligent affective computing, and sensor design and data processing. Additionally, we discuss future research directions and potential applications, including stress monitoring, music cognition, and robotics, where ECG and EDA will play a significant role with wearable devices. Therefore, this review serves as an information summary for everyone who is interested in affective computing and improving AI’s understanding of human emotion.
... Moreover, HRV analysis has also been widely used to explain players' behavioral responses when interacting with games and VR applications, complementing the conventional methods to capture perceived user experiences [14]. In non-VR settings, the real-time analysis of HRV is a critical approach to determine psychological aspects of a person, such as pain [23], mental workload [18], or stress [15,36]. However, measurement of heart activity in VR systems often poses challenges (e.g., due to wired connections and data synchronization) that interfere with the natural interaction of the user; an example can be found in a project that estimated stress for VR pain management using HRV biofeedback [7]. ...
... With regards to PPG, we used only the time domain-based indicators (HR, SDNN and PNN50) because the long epoch time was required to extract frequency domain-based PPG-indicators (Smith et al., 2013;Jiang et al., 2017;Castaldo et al., 2019). We also made our decision based on previous PPG studies related to addiction (including IGD). ...
Full-text available
The proliferating and excessive use of internet games has caused various comorbid diseases, such as game addiction, which is now a major social problem. Recently, the American Psychiatry Association classified “Internet gaming disorder (IGD)” as an addiction/mental disorder. Although many studies have been conducted on the diagnosis, treatment, and prevention of IGD, screening studies for IGD are still scarce. In this study, we classified gamers using multiple physiological signals to contribute to the treatment and prevention of IGD. Participating gamers were divided into three groups based on Young’s Internet Addiction Test score and average game time as follows: Group A, those who rarely play games; Group B, those who enjoy and play games regularly; and Group C, those classified as having IGD. In our game-related cue-based experiment, we obtained self-reported craving scores and multiple physiological data such as electrooculogram (EOG), photoplethysmogram (PPG), and electroencephalogram (EEG) from the users while they watched neutral (natural scenery) or stimulating (gameplay) videos. By analysis of covariance (ANCOVA), 13 physiological features (vertical saccadic movement from EOG, standard deviation of N-N intervals, and PNN50 from PPG, and many EEG spectral power indicators) were determined to be significant to classify the three groups. The classification was performed using a 2-layers feedforward neural network. The fusion of three physiological signals showed the best result compared to other cases (combination of EOG and PPG or EEG only). The accuracy was 0.90 and F-1 scores were 0.93 (Group A), 0.89 (Group B), and 0.88 (Group C). However, the subjective self-reported scores did not show a significant difference among the three groups by ANCOVA analysis. The results indicate that the fusion of physiological signals can be an effective method to objectively classify gamers.
This secondary data analysis study aimed to (1) investigate the use of two sense-based parameters (movement and sleep hours) as predictors of chronic pain when controlling for patient demographics and depression, and (2) identify a classification model with accuracy in predicting chronic pain. Data collected by Oregon Health & Science University between March 2018 and December 2019 under the Collaborative Aging Research Using Technology Initiative were analyzed in two stages. Data were collected by sensor technologies and questionnaires from older adults living independently or with a partner in the community. In Stage 1, regression models were employed to determine unique sensor based behavioral predictors of pain. These sensor-based parameters were used to create a classification model to predict the weekly recalled pain intensity and interference level using a deep neural network model, a machine learning approach, in Stage 2. Daily step count was a unique predictor for both pain intensity (75% Accuracy, F1=0.58) and pain interference (82% Accuracy, F1=.59). The developed classification model performed well in this dataset with acceptable accuracy scores. This study demonstrated that machine learning technique can be used to identify the relationship between patients’ pain and the risk factors.
This paper covers the feasibility of electrogastrogram (EGG) in multi-modal mental stress assessment in conjunction with electrocardiogram (ECG) and respiratory signal (RESP). In this study, twenty-one healthy participants were repeatedly relaxed, stressed, and highly stressed according to our experimental protocol, which was based on combined arithmetic and Stroop tasks, and their EGG, ECG, and RESP were simultaneously captured. Subsequently, various features were extracted from the signals, and correlation analysis was performed between mental stress levels and the features, especially the EGG features. Furthermore, conventional machine learning models were optimized and validated to verify the feasibility of EGG in mental stress detection. Some EGG features exhibited significant correlation to mental stress levels (ρbi,menaDP = -0.187 and ρmulti,meanDP = -0.177, p < 0.001). The correlation degree was comparable to that of the RESP features. The EGG features largely reflected individual differences regarding mental stress response compared to the ECG and RESP features. Most importantly, the utilization of the EGG features along with the ECG and RESP features significantly improved the accuracy of several models by up to 8% regarding mental stress detection. Especially, logistic regression exhibited moderate accuracy in detecting mental stress (70.15% accuracy and 0.741 area under the receiver operating characteristic curve). We believe that EGG monitoring could significantly contribute to in-depth mental stress evaluation, and potentially be used for the development of real-time mental stress monitoring system and personalized mental stress assessment modality.
Conference Paper
Full-text available
Various pain conditions may result in altered autonomic nervous system regulation. We evaluated whether autonomic regulation, analyzed with short-term heart rate variability (HRV), differs between low back pain (LBP) patients and healthy controls. Additionally, we tested if short-term HRV recordings are feasible and informative in planning home monitoring of LBP patients. The study population consisted of 16 volunteers (8 LBP patients and 8 healthy subjects) (age 42±10 years, body mass index 26±4 kg/m2, 7 men and 9 women). Usually 3- to 5-minute R-R interval recordings have been used as short-term recordings of HRV, but recent evidence supports even shorter R-R interval recording procedure for short-term HRV assessment. We collected R-R interval data for 1 minute in sitting, standing and bending down tasks. Mean heart rate (HR) and vagally mediated beat-to-beat variability (SD1 and rMSSD) were analyzed. The results showed that autonomic nervous system function assessed with the short-term measurement HRV method differentiates LBP patients from healthy controls in sitting and standing. Vagally mediated SD1 and rMSSD were significantly lower and the HR was higher among the patients compared to the controls. A novel finding was also the feasibility of 1-minute measurement of HRV, which may open entirely new opportunities to assess accurately concomitant changes in autonomic nervous system function and self-reported individual pain experience. This could lead to a more personalized pain treatment and more efficient health care resource allocation as the new measurement methods is more suitable for home monitoring than the previously used ones.
Full-text available
Objectives There is little consensus regarding the burden of pain in the UK. The purpose of this review was to synthesise existing data on the prevalence of various chronic pain phenotypes in order to produce accurate and contemporary national estimates. Design Major electronic databases were searched for articles published after 1990, reporting population-based prevalence estimates of chronic pain (pain lasting >3 months), chronic widespread pain, fibromyalgia and chronic neuropathic pain. Pooled prevalence estimates were calculated for chronic pain and chronic widespread pain. Results Of the 1737 articles generated through our searches, 19 studies matched our inclusion criteria, presenting data from 139 933 adult residents of the UK. The prevalence of chronic pain, derived from 7 studies, ranged from 35.0% to 51.3% (pooled estimate 43.5%, 95% CIs 38.4% to 48.6%). The prevalence of moderate-severely disabling chronic pain (Von Korff grades III/IV), based on 4 studies, ranged from 10.4% to 14.3%. 12 studies stratified chronic pain prevalence by age group, demonstrating a trend towards increasing prevalence with increasing age from 14.3% in 18–25 years old, to 62% in the over 75 age group, although the prevalence of chronic pain in young people (18–39 years old) may be as high as 30%. Reported prevalence estimates were summarised for chronic widespread pain (pooled estimate 14.2%, 95% CI 12.3% to 16.1%; 5 studies), chronic neuropathic pain (8.2% to 8.9%; 2 studies) and fibromyalgia (5.4%; 1 study). Chronic pain was more common in female than male participants, across all measured phenotypes. Conclusions Chronic pain affects between one-third and one-half of the population of the UK, corresponding to just under 28 million adults, based on data from the best available published studies. This figure is likely to increase further in line with an ageing population.
Full-text available
Background: The clinically used methods of pain diagnosis do not allow for objective and robust measurement, and physicians must rely on the patient's report on the pain sensation. Verbal scales, visual analog scales (VAS) or numeric rating scales (NRS) count among the most common tools, which are restricted to patients with normal mental abilities. There also exist instruments for pain assessment in people with verbal and / or cognitive impairments and instruments for pain assessment in people who are sedated and automated ventilated. However, all these diagnostic methods either have limited reliability and validity or are very time-consuming. In contrast, biopotentials can be automatically analyzed with machine learning algorithms to provide a surrogate measure of pain intensity. Methods: In this context, we created a database of biopotentials to advance an automated pain recognition system, determine its theoretical testing quality, and optimize its performance. Eighty-five participants were subjected to painful heat stimuli (baseline, pain threshold, two intermediate thresholds, and pain tolerance threshold) under controlled conditions and the signals of electromyography, skin conductance level, and electrocardiography were collected. A total of 159 features were extracted from the mathematical groupings of amplitude, frequency, stationarity, entropy, linearity, variability, and similarity. Results: We achieved classification rates of 90.94% for baseline vs. pain tolerance threshold and 79.29% for baseline vs. pain threshold. The most selected pain features stemmed from the amplitude and similarity group and were derived from facial electromyography. Conclusion: The machine learning measurement of pain in patients could provide valuable information for a clinical team and thus support the treatment assessment.
Full-text available
Heart rate variability (HRV), the beat-to-beat variation in either heart rate or the duration of the R-R interval, has become a popular clinical and investigational tool (Task Force of the European Society of Cardiology and the North American Society of Pacing and Electrophysiology, 1996; Billman, 2011). Indeed, the term “heart rate variability” yields nearly 18,000 “hits” when placed in the pubmed search engine. These temporal fluctuations in heart rate exhibit a marked synchrony with respiration (increasing during inspiration and decreasing during expiration—the so called respiratory sinus arrhythmia) and are widely believed to reflect changes in cardiac autonomic regulation (Billman, 2011). Although the exact contributions of the parasympathetic and the sympathetic divisions of the autonomic nervous system to this variability are controversial and remain the subject of active investigation and debate, a number of time and frequency domain techniques have been developed to provide insight into cardiac autonomic regulation in both health and disease (Billman, 2011). It is the purpose of this book to provide a comprehensive assessment of the strengths and limitations of HRV techniques. Particular emphasis will be placed on the application of HRV techniques in the clinic and on the interaction between prevailing heart rate and HRV. This book contains both state-of-the art review and original research articles that have been grouped into two main sections: Methodological Considerations and Clinical Application. A brief summary of the chapters contained in each section follows below.
Full-text available
Thousands of papers involved in heart rate variability (HRV). However, little was known about one important measure of HRV, the root mean square of successive heartbeat interval differences (RMSSDs). Another fundamental measure SDNN indicates standard deviation of normal to normal R-R intervals, where R is the peak of a QRS complex (heartbeat). Compared with SDNN, RMSSD is a short-term variation of heart rate. Through a time-frequency transformation, the ratio of low- and high-frequency power LF/HF represents the sympatho-vagal balance of the autonomic nervous system (ANS). Some research claimed that SDNN/RMSSD was a good surrogate for LF/HF. However, only two special cases supported this hypothesis in the literature survey. The first happened in resting supine state and the other was a group of prefrontal cortex patients. Both of their Pearson correlation coefficients reached 0.90, a reasonable criterion. In our study, a 6-week experiment was performed with 32 healthy young Asian males. The Pearson correlation coefficients had a normal distribution with average values smaller than 0.6 for 3 and 5-minute epochs, respectively. Our findings suggest this surrogate aspect could remain as a hypothesis.
Conference Paper
Fog Computing is a new architecture to migrate some data center’s tasks to the edge of the server. The fog computing, built on the edge servers, is viewed as a novel architecture that provides the limited computing, storing, and networking services in the distributed way between end devices and the traditional cloud computing Data Centers. It provides the logical intelligence to the end devices and filters the data for Data Centers. The primary objective of fog computing is to ensure the low and predictable latency in the latency-sensitive of Internet of Things (IoT) applications such as the healthcare services. This paper discusses the characteristics of fog computing and services that fog computing can provide in the healthcare system and its prospect.
Objective pain assessment methods pose an advantage over the currently used subjective pain rating tools. Advanced signal processing methodologies, including the wavelet transform (WT) and the orthogonal matching pursuit algorithm (OMP), were developed in the past two decades. The aim of this study was to apply and compare these time-specific methods to heart rate samples of healthy subjects for acute pain detection. Fifteen adult volunteers participated in a study conducted in the pain clinic at a single center. Each subject's heart rate was sampled for 5-min baseline, followed by a cold pressor test (CPT). Analysis was done by the WT and the OMP algorithm with a Fourier/Wavelet dictionary separately. Data from 11 subjects were analyzed. Compared to baseline, The WT analysis showed a significant coefficients' density increase during the pain incline period (p < 0.01) and the entire CPT (p < 0.01), with significantly higher coefficient amplitudes. The OMP analysis showed a significant wavelet coefficients' density increase during pain incline and decline periods (p < 0.01, p < 0.05) and the entire CPT (p < 0.001), with suggestive higher amplitudes. Comparison of both methods showed that during the baseline there was a significant reduction in wavelet coefficient density using the OMP algorithm (p < 0.001). Analysis by the two-way ANOVA with repeated measures showed a significant proportional increase in wavelet coefficients during the incline period and the entire CPT using the OMP algorithm (p < 0.01). Both methods provided accurate and non-delayed detection of pain events. Statistical analysis proved the OMP to be by far more specific allowing the Fourier coefficients to represent the signal's basic harmonics and the wavelet coefficients to focus on the time-specific painful event. This is an initial study using OMP for pain detection; further studies need to prove the efficiency of this system in different settings.
Background: The autonomic nervous system is influenced by many stimuli including pain. Heart rate variability (HRV) is an indirect marker of the autonomic nervous system. Because of paucity of data, this study sought to determine the optimal thresholds of HRV above which the patients are in pain after minor spinal surgery (MSS). Secondly, we evaluated the correlation between HRV and the numeric rating scale (NRS). Methods: Following institutional review board approval, patients who underwent MSS were assessed in the postanesthesia care unit after extubation. A laptop containing the HRV software was connected to the ECG monitor. The low-frequency band (LF: 0.04 to 0.5 Hz) denoted both sympathetic and parasympathetic activities, whereas the high-frequency band (HF: 0.15 to 0.4 Hz) represented parasympathetic activity. LF/HF was the sympathovagal balance. Pain was quantified by the NRS ranging from 0 (no pain) to 10 (worst imaginable pain). Simultaneously, HRV parameters were noted. Optimal thresholds were calculated using receiver operating characteristic curves with NRS>3 as cutoff. The correlation between HRV and NRS was assessed using the Spearman rank test. Results: We included 120 patients (64 men and 56 women), mean age 51±14 years. The optimal pain threshold values were 298 ms for LF and 3.12 for LF/HF, with no significant change in HF. NRS was correlated with LF (r=0.29, P<0.005) and LF/HF (r=0.31, P<0.001) but not with HF (r=0.09, NS). Conclusions: This study suggests that, after MSS, values of LF>298 m and LF/HF>3.1 denote acute pain (NRS>3). These HRV parameters are significantly correlated with NRS.