ArticlePDF Available

Abstract

The circulatory system is oscillatory in its nature. Oscillatory components linked to physiological processes and underlying regulatory mechanisms are identifiable in circulatory signals. Autonomic regulation is essential for the system's ability to deal with external exposure, and the integrity of oscillations may be considered a hallmark of a healthy system. Loss of complexity is seen as a consequence of several diseases and aging. Heart rate variability is known to decrease after cardiac surgery and remain reduced for up to 6 months. Oscillatory components of circulatory signals are linked to the system's overall complexity. We therefore hypothesize that the frequency distributions of circulatory signals show loss of oscillatory components after cardiac surgery and that the observed changes persist. We investigated the development of the circulatory frequency distributions of eight patients undergoing cardiac surgery by extracting three time series from conventional blood pressure and electrocardiography recordings: systolic blood pressure, heart rate, and amplitude of the electrocardiogram's R-wave. Four 30-min selections, representing key events of the perioperative course, were analyzed with the continuous wavelet transform, and average wavelet power spectra illustrated the circulatory frequency distributions. We identified oscillatory components in all patients and variables. Contrary to our hypothesis, they were randomly distributed through frequencies, patients, and situations, thus, not representing any reduction in the overall complexity. One patient showed loss of a 25-s oscillation after surgery. We present a case where noise is misclassified as an oscillation, raising questions about the robustness of such analyses.
Physiological Reports. 2020;8:e14423.
|
1 of 9
https://doi.org/10.14814/phy2.14423
wileyonlinelibrary.com/journal/phy2
1
|
INTRODUCTION
The circulatory system is oscillatory in its nature. Oscillations
can be traced back to physiological processes, such as heart
contraction, respiration, and rhythmic contraction of the vas-
culature, and these oscillatory processes are controlled by
regulatory mechanisms. The result is a highly irregular and
complex oscillatory profile. Autonomic regulation is essential
DOI: 10.14814/phy2.14423
ORIGINAL RESEARCH
Cardiac surgery does not lead to loss of oscillatory components in
circulatory signals
KathrineKnai1
|
PetterAadahl1,2
|
Nils K.Skjaervold1,2
This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original
work is properly cited.
© 2020 The Authors. Physiological Reports published by Wiley Periodicals LLC on behalf of The Physiological Society and the American Physiological Society.
1Department of Circulation and Medical
Imaging, Faculty of Medicine and Health
Sciences, Norwegian University of Science
and Technology, Trondheim, Norway
2Department of Cardiothoracic Anaesthesia
and Intensive Care, Clinic of Anaesthesia
and Intensive Care, Trondheim University
Hospital, Trondheim, Norway
Correspondence
Kathrine Knai, NTNU, Faculty of Medicine
and Health Sciences, Department of
Circulation and Medical Imaging, Postboks
8905, 7491 Trondheim, Norway.
Email: kathrine.knai@ntnu.no
Funding information
Norwegian University of Science and
Technology
Abstract
The circulatory system is oscillatory in its nature. Oscillatory components linked
to physiological processes and underlying regulatory mechanisms are identifiable
in circulatory signals. Autonomic regulation is essential for the system's ability to
deal with external exposure, and the integrity of oscillations may be considered a
hallmark of a healthy system. Loss of complexity is seen as a consequence of several
diseases and aging. Heart rate variability is known to decrease after cardiac surgery
and remain reduced for up to 6months. Oscillatory components of circulatory sig-
nals are linked to the system's overall complexity. We therefore hypothesize that the
frequency distributions of circulatory signals show loss of oscillatory components
after cardiac surgery and that the observed changes persist. We investigated the de-
velopment of the circulatory frequency distributions of eight patients undergoing
cardiac surgery by extracting three time series from conventional blood pressure and
electrocardiography recordings: systolic blood pressure, heart rate, and amplitude
of the electrocardiogram's R-wave. Four 30-min selections, representing key events
of the perioperative course, were analyzed with the continuous wavelet transform,
and average wavelet power spectra illustrated the circulatory frequency distributions.
We identified oscillatory components in all patients and variables. Contrary to our
hypothesis, they were randomly distributed through frequencies, patients, and situ-
ations, thus, not representing any reduction in the overall complexity. One patient
showed loss of a 25-s oscillation after surgery. We present a case where noise is mis-
classified as an oscillation, raising questions about the robustness of such analyses.
KEYWORDS
cardiac surgery patients, circulatory oscillations, continuous blood pressure, electrocardiogram, loss of
complexity
2 of 9
|
KNAI et Al.
for the system's ability to deal with external exposure, and
the integrity of circulatory oscillations can be considered a
hallmark of a healthy system. Circulatory signals, such as
continuous blood pressure (BP) and electrocardiography
(ECG) signals, show traces of these underlying oscillations.
Traditional heart rate variability and frequency analyses have
identified specific oscillations and they have been attributed
to different parts of autonomic regulation (Akselrod et al.,
1981; Bracic & Stefanovska, 1998; Pomeranz et al., 1985).
Loss of complexity has been reported in several cardiac
and non-cardiac diseases (Claydon & Krassioukov, 2008;
Goldstein et al., 1998; Kleiger, Miller, Bigger, & Moss, 1987;
Riordan, Norris, Jenkins, & Morris, 2009; Wolf, Varigos,
Hunt, & Sloman, 1978) and simply as a feature of aging
(Kaplan et al., 1991; Lipsitz & Goldberger, 1992; Umetani,
Singer, McCraty, & Atkinson, 1998; Takahashi et al., 2012).
Heart rate variability is known to decrease after cardiac sur-
gery, and remain reduced for up to 6months (Hogue, Stein,
Apostolidou, Lappas, & Kleiger, 1994; Kuo et al., 1999).
There is no clear definition of complexity. However, com-
plex systems are built up by components that interact in mul-
tiple ways and with the external environment, resulting in
organized and disorganized behavior that cannot be predicted
from the components alone (Johnson, 2009). Linking this to
biological signals, complexity is related to the degree of infor-
mation in the signal, the predictability of the signal, and the
ability to describe the signal in a simple manner (Goldberger,
Moody, & Costa, 2012). The definition is too diffuse to pro-
vide a quantitative measure of complexity that applies univer-
sally. Oscillatory components of biological systems represent
underlying components that interact and produce the behav-
ior of the system as a whole. Altogether, they both reflect
the system's overall complexity (Goldberger, 1996) and are
linked to underlying regulation. On this basis, the exploration
of oscillatory distributions of biological signals provides in-
formation about the overall state of biological systems, which
could be altered by disease or invasive procedures. If the ob-
served changes are generalizable between patients, such in-
formation can be implemented to future monitoring tools.
This is beneficial as changes in patients’ clinical state could
be identified and clinicians notified before overall variables
such as heart rate (HR) or BP are changed.
We explore frequency and amplitude modulations of BP
and ECG signals by extracting three time series: systolicBP
(SBP), HR, and amplitude of the ECG’s R-wave. The Brody
effect states that variations in R-wave amplitude are related
to ventricular preload (Brody, 1956). R-wave amplitude can
thus be seen in relation with SBP and HR. By combining
the frequency distributions of the three mentioned variables,
we illustrate unique circulatory frequency distributions. In
this work, we investigated the development of the circulatory
frequency distributions of eight patients undergoing cardiac
surgery. Four 30-min selections, representing key events of
the perioperative course, were analyzed with the continuous
wavelet transform (CWT), and average wavelet power spec-
tra illustrated the circulatory frequency distributions. We hy-
pothesize that the circulatory frequency distributions show
loss of oscillatory components with surgery and that the
observed changes persist, measured until the morning after
surgery.
2
|
MATERIAL AND METHODS
2.1
|
Study population, ethics, and
confidentiality
From March to May 2016, patients scheduled for coronary
artery bypass grafting were invited to participate in the study,
recruiting a total of 10 patients. Two patients were excluded
due to non-sinus rhythm at one or several time points of the
recording. Other exclusion criteria are left ventricular ejec-
tion fraction below 0.5, severe valve disease, right ventricular
failure, pulmonary hypertension, and severe postoperative
hemorrhage. Finally, we had a study group consisting of six
men and two women, age ranging from 47 to 88. The patients
were enumerated 1–10, with patient 6 and 9 excluded.
The surgery was performed at Trondheim University
Hospital, Norway. Written consent was collected prior to data
collection. The study protocol was approved by the Regional
Committee for Medical and Health Research Ethics (refer-
ence: 2015/2019/REK midt). Confidentiality was strictly
maintained throughout the study.
2.2
|
Equipment and study protocol
Data collection was performed in two sessions: before and
after surgery. The patients were lying in bed during both peri-
ods. The study equipment includes a 3-electrode ECG, a laser
Doppler flowmeter (LDF) attached to the calf, and an arterial
cannula inserted to the left radial artery. Additionally, pa-
tients 1–3 had a photoplethysmograph (PPG) finger sensor at-
tached. The study equipment was provided by ADInstruments
(Oxford, UK), as well as hardware and software (PowerLab
16/35 and LabChart 8.1.3). The signals were recorded with a
sampling rate of 400Hz.
The preoperative recordings were collected with the patients
resting in bed in a quiet room without disturbances at the tho-
racic surgery ward. The duration of the recordings ranged from
47 to 86min. The patients did not receive premedication prior to
surgery, and surgery was performed under general balanced an-
esthesia (thiopental, fentanyl, isoflurane, and propofol). During
surgery, the study equipment was removed. After surgery, the
study equipment was reattached using new ECG patches and
a new arterial cannula inserted to the right radial artery. The
|
3 of 9
KNAI et Al.
postoperative recording was collected from the patients arrived
at the thoracic intensive care unit, until the next morning. The
duration of the recordings ranged from 14 to 18.5hr.
2.3
|
Data handling and preprocessing
The BP and ECG recordings were exported from LabChart as
mat.files and analyzed in R, version 3.5.1, with the packages
R.matlab, signal, robustHD, and WaveletComp (Alfons, 2016;
Bengtsson, 2016; R Foundation for Statistical Computing,
2018; Roesch & Schmidbauer, 2018; Signal developers, 2013).
We subdivided the data into four situations: preoperatively (A);
postoperatively, on respirator (B); postoperatively, after extu-
bation (C); and postoperatively, the next morning (D).
We extracted 30-min selections representing each situ-
ation and preprocessed the BP and ECG signals into three
time series: SBP, HR, and R-wave amplitude (Figure 1).
Baseline wander was removed from the ECG signals by ap-
plying a Savitzky–Golay smoothing filter before further anal-
yses (Nahiyan & Amin, 2017). We defined the SBP and the
R-wave amplitude as the maxima of the BP and ECG, respec-
tively. The heart ratewas defined as HR=60/RRi, where RRi
is the time interval in seconds between R-peak i and i+1 of
the ECG. Some episodes of noise were misclassified as heart-
beats; thus, we removed outliers from SBP, RR-intervals, and
R-amplitude before further calculation. To provide evenly
sampled time series, we performed a cubic spline interpo-
lation to a sampling frequency of 10Hz. The final variables
were called interpolated SBP (iSBP), interpolated HR (iHR),
and interpolated R-wave amplitude (iAmp).
To examine one specific identified oscillation, the PPG
and LDF signal of patient 1 were included in a subanalysis.
To provide comparable results, the PPG was preprocessed
with the same algorithm as iSBP and iAmp, creating a new
time series of the amplitude of the signal, interpolated to a
sampling rate of 10Hz. The final variable was called PPG-
iAmp. The LDF was downsampled to 10Hz.
2.4
|
Analysis
We performed the CWT to identify frequency components
present in iSBP, iHR, and iAmp. The CWT is a convolu-
tion of the signal with a function generated from the mother
wavelet (Fugal, 2009). We used the Morlet wavelet, which
by mathematical definition is a Gaussian enveloped cosine
wave, and has been widely used for investigation of bio-
logical signals, especially the ECG (Addison, 2005). In the
convolution process, it is shifted in time and stretched and
shrunk, quantifying different frequency components’ pres-
ence in the signal at different time points. We presented the
results in average wavelet power spectra, illustrating the av-
eraged frequency distributions of the signals. Furthermore,
we performed the CWT for bivariate time series identifying
frequency components that are present in two time series
with a significance level of 0.05. The results are presented
in cross-wavelet spectra, with significant frequencies marked
by white lines and phasedifferences by arrows. The CWT
for bivariate time series was performed on the variable pairs,
iAmp-iSBP and iSBP-iHR.
In order to visually examine the individual time-series’
oscillations, we decomposed the time series with locally
weighted estimated scatterplot smoothing (Loess) (Cleveland
& Devlin, 1988). We applied the regression three times, each
time subtracting the smoothed curve from the signal, provid-
ing a set of oscillating components of increasing frequency.
The extracted components were called Loess #1, Loess #2,
and Loess #3. By plotting the components of all variables
together, we visually inspected their oscillating behavior and
phasedifferences. From the CWT and Loess, we identified
the components that are highly present in all variables and
performed a cross-correlation analysis, which calculates the
correlation of two time series as a function of the displace-
ment of one relative to the other—the cross-correlation func-
tion (CCF). By defining CCFmax, we identified at which time
lag the correlation is highest, and thus at which relative dis-
placement the studied variables oscillate.
FIGURE 1 Preprocessing of the BP and ECG signals generating the variables iSBP, iHR, and iAmp. The SBP and the R-wave
amplitudewere defined as the maxima of the BP and ECG, respectively. The HRwas calculated from the time interval between two subsequent
R-peaks of the ECG (RRi [s]). The time serieswere interpolated to a sampling rate of 10Hz
4 of 9
|
KNAI et Al.
3
|
RESULTS
By performing the CWT on iSBP, iHR, and iAmp, we identi-
fied each patient's circulatory frequency distribution through-
out the perioperative course, illustrated by situation A–D. We
hypothesize that we will see loss of oscillatory components
from situation A to B, and that the changes persist through
situations C and D. Figure2 shows the average wavelet power
spectra of iSBP, iHR, and iAmp of all patients and situations.
The spectra include periods between 10 and 1,000s. Lower
periods are not included, as the respiration is a powerful os-
cillatory component, overshadowing the presence of slower,
less dominating oscillations.
We identified oscillatory components in all patients and
situations. Patient 1 shows a distinct peak around 25s in
situation A. This oscillation is present in all three variables,
and not visible in any of the situations B, C, or D. A less
prominent oscillation around 800s in 3A is observed, which
arguably disappears after surgery. Patient 8 showed loss of
oscillations in the range between 20 and 50s in iHR, and
in situation D, an oscillation just above 100s is observed in
all variables. We see other examples of loss of power in the
mid-range after surgery, but no cases with loss of distinct
oscillatory components. Altogether, we illustrate frequency
distributions that change through the perioperative course,
but the observed changes do not display any trend or sys-
tem. Overall, the number of oscillatory components and their
power are more or less randomly distributed through patients
and situations. Linking this to the signals’ overall complexity,
we did not identify any clear reduction of such after surgery.
iAmp does not show any distinct oscillatory peaks in any pa-
tients in situation B. This is due to a domination of the respi-
ration during mechanical ventilation.
The 25-s oscillation in 1A stands out as the only oscilla-
tion that is clearly present in all variables before surgery and
gone after. To examine the specific variables’ oscillatory be-
havior, we performed a Loess regression, illustrated by Loess
#2 in Figure3.
Figure 3 shows that iSBP and iHR oscillate in phase,
iHR leading. iAmp oscillates off phase with respect to the
other two. By performing a cross-correlation analysis on
the variable pairs, iAmp-iSBP and iSBP-iHR, we found that
CCFmax of iAmp and iSBP is 0.75, with a lag of 8.2s. The
corresponding values for iSBP and iHR are 0.80 and 3.5s.
Altogether, this tells us that iHR is leading, with a time lag of
3.5s to iSBP and 11.7s to iAmp. Maximum correlation val-
ues of 0.75 and 0.80 are high when it comes to biological time
series. The preoperative recording of patient 1 included both
a PPG and a LDF signal. The signals were preprocessed as
described in Methods and analyzed with the CWT. Figure4
shows the average wavelet spectrum of all variables, and we
see that the 25-s oscillation is present in the amplitude of the
PPG signal but not in LDF.
Figure 2 shows an 800-s oscillation in patient 3, situa-
tion A.It is present in all three variables and is partly gone
postoperatively. Looking at the raw signals and performing
a Loess regression, we find that the extracted component is
caused by short events of noise, thus not representing a true
physiological oscillator.
Patient 8 showed loss of oscillations in the range between
20 and 50s in iHR, and an oscillation just above 100s in
situation D that is present in all variables. Figure5 shows the
cross-wavelet spectra of iAmp-iSBP and iSBP-iHR of situa-
tion D.
Both variable pairs show high power just above 128s,
confirming that the oscillation is present in all three time se-
ries. Both variable pairs show variations in phasedifferences
through the time course, but mostly iAmp-iSBP show arrows
pointing to the lower right, meaning that they oscillate in
phase, iSBP leading. iSBP-iHR show arrows pointing to the
lower left, meaning that they oscillate off phase, iSBP lead-
ing. Thus, iHR and iAmp oscillate in phase, and iSBP out of
phase with respect to the other two.
4
|
DISCUSSION
We illustrate the frequency distributions of variables ex-
tracted from BP and ECG signals of eight patients under-
going coronary artery bypass grafting. The circulatory
frequency distributions illustrate the presence of oscillatory
components in all variables, patients and situations, and the
oscillations are randomly distributed over the examined fre-
quency range. The high variety seems to represent interindi-
vidual variations, more than factors of the performed surgery.
Considering the heterogeneity of our findings, we have not
presented information that is suitable for use in any monitor-
ing device or other clinical decision tools.
The identified oscillations do not correspond to the dis-
tinct pattern of frequency bands that are described in the lit-
erature (Akselrod et al., 1981; Bracic & Stefanovska, 1998;
Pomeranz et al., 1985). Linking the circulatory frequency
distributions to the overall complexity of circulatory sig-
nals, no reduction of such is identified. Either the complex-
ity is not reduced with cardiac surgery, or our method is not
able to identify it. One case showed a 25-s oscillation that
is present preoperatively (1A) and not postoperatively (1B,
1C, 1D). The oscillation is found in all three variables, and
additionally in the amplitude of the PPG (PPG-iAmp). It has
a frequency of 0.04Hz, corresponding to the limit between
low frequencies and very low frequencies (Task Force of
the European Society of Cardiology & the North American
Society of Pacing & Electrophysiology, 1996). According to
the literature, low frequencies reflect baroreceptor activity,
but findings are inconsistent regarding whether this activ-
ity is mediated by the sympathetic or the parasympathetic
|
5 of 9
KNAI et Al.
FIGURE 2 The average wavelet power spectra of iSBP, iHR, and iAmp of all patients through situation A to D. Each spectrum represents
one patient in one situation, named with a number and a letter. Patients are separated by rows, and situations by columns. The situations represent
key events of the perioperative course: preoperatively (A); postoperatively, on respirator (B); postoperatively, after extubation (C); postoperatively,
the next morning (D). Average wavelet power is shown on the x-axis and period (in seconds) on a logarithmic scale on the y-axis. The variables are
distinguished by color. Patient 1 shows loss of a 25-s oscillation between situation A and B. Patient 3 shows loss of an 800-s oscillation, and patient
8 shows loss of oscillations in the range between 50 and 100s in iHR
012345
10 50 500
A1
Average wavelet power
Period (saa
iSBP
iHR
iAmp
012345
10 50 500
B1
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C1
Average wavelet power
Period (s)
iSBP
iHR
iAmp
0
10 50 500
D1
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A2
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B2
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C2
Average wavelet power
Period (s)
iSBP
iHR
iAmp
0
10 50 500
D2
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A3
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B3
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C3
Average wavelet power
Period (s)
iSBP
iHR
iAmp
0
10 50 500
D3
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A4
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B4
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C4
Average wavelet power
Period (s)
iSBP
iHR
iAmp
0
12345
12345
12345
12345
10 50 500
D4
tverage wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A5
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B5
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C5
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
D5
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A7
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B7
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C7
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
D7
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A8
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B8
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C8
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
D8
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
A10
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
B10
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
C10
Average wavelet power
Period (s)
iSBP
iHR
iAmp
012345
10 50 500
D10
Average wavelet power
Period (s)
iSBP
iHR
iAmp
6 of 9
|
KNAI et Al.
nervous system (Shaffer, McCraty, & Zerr, 2014; Task
Force of the European Society of Cardiology & the North
American Society of Pacing & Electrophysiology, 1996).
Patient 8 showed loss of oscillations in iHR with surgery and
return of an oscillation in situation D that is present in all
variables. The oscillation in 8D has a frequency of 0.007Hz,
corresponding to the ultra-low frequency band. The physi-
ological correlate to ultra-low frequencies is insecure. This
study was not designed to explore underlying physiological
mechanisms of identified oscillations.
We are developing methods for analyzing biological sig-
nals, both focusing on preprocessing and choice of analyses.
We find that there are several challenges related to analyz-
ing biological signals, mainly related to noise-handling. We
have identified a case where noise is misinterpreted as an
oscillation. This leaves us wondering which of the identified
oscillations are true physiological oscillators, and which rep-
resent methodological errors. As we are not able to provide
completely noise-free recordings in controlled settings of
bedbound patients, we believe that tools meant for clinical
use must be robust for noise. Thus, the algorithms must either
remove all noise, or the results not being affected by the pres-
ence of it. We did not identify the distinct frequency bands that
are reported in the literature (Shaffer et al., 2014; Task Force
of the European Society of Cardiology & the North American
Society of Pacing & Electrophysiology, 1996). Noise could
be the problem here as well. However, the signals are mostly
noise-free, so we would expect to identify the oscillations if
they were present. This raises the question if the use of strict
frequency intervals is a simplification of a highly complex
and variable field. Features that are incorporated in biologi-
cal signals, such as nonlinearity and nonstationarity, are chal-
lenging when analyzing them. We have earlier addressed the
challenge by applying Fourier-based analyses to such signals,
suggesting the data-driven Hilbert–Huang Transform as a
better approach (Knai, Kulia, Molinas, & Skjaervold, 2017).
However, the Hilbert–Huang transform is hampered by being
computationally challenging and requiring thorough valida-
tion. If the developed methods at some time point are meant
to be used real-time, for instance in intelligent alarm systems,
the chosen analyses should be quick and easily adaptive to
different biological signals. In this work, visual inspection
was required to secure that the algorithms applied correctly
and to validate the results, identifying the case where noise
was misinterpreted as an oscillation.
4.1
|
Methodological considerations
Our study population is small. We recruited 10 patients
scheduled for coronary artery bypass grafting over a period
FIGURE 3 Loess regression of iSBP, iHR, and iAmp of 1A, illustrated by Loess #2 which is the second extracted component. We see that
iSBP and iHR oscillates in phase, and iAmp off phase
0 200 400 600 800
A1
Time (s)
Loess #2
iSBP
iHR
iAmp
1000 1200 1400 1600 1800
A1
Time (s)
Loess #2
iSBP
iHR
iAmp
FIGURE 4 Average wavelet power spectra of 1A including PPG
and LDF signals. The PPG is preprocessed with the same algorithm as
iSBP and iAmp, giving a time series of the maxima of the PPG signal,
called PPG-iAmp. The LDF is downsampled to a sampling frequency
of 10Hz, as the other time series. The25-s oscillation is presentin
PPG-iAmp, iSBP, iHR, and ECG-iAmp, but not in the LDF
012345
10 20 50 100 200 500
Average wavelet power
Period (s)
PPG−iAmp
iSBP
iHR
ECG−iAmp
LDF
a
|
7 of 9
KNAI et Al.
of 3months in 2016, whereof two patients were excluded due
to non-sinus rhythm. The final data material includes three
extracted variables from four situations of eight patients—96
analyzed time series. Being derived from only eight patients,
we cannot generalize our findings to the total population of
cardiac surgery patients.
The recruited patients are heterogeneous individuals
featuring different medical backgrounds, pharmacological
profiles, and general health. However, the group altogether
holds common features such as high age and coronary heart
disease. One could raise the question of selection bias as we
recruited patients over a short time period and excluded pa-
tients with serious illness such as heart failure, valve disease,
and perioperative complications. However, we investigate
universal physiological features without performing statisti-
cal hypothesis testing or other comparisons on group level.
The comparisons we do are only between situations of the
perioperative course, and in such cases, the patients serve as
their own controls. Interpretation of the results must be done
with these aspects in thought, and the results’ generalizability
should be investigated in bigger study groups.
To minimize autonomic activation and artifacts caused
by postural changes, the patients were kept lying during
data collection. The data were collected with research
hardware and software to secure complete control of fil-
tering and preprocessing algorithms applied to the data.
To avoid putting the patients through unnecessary stress
by inserting two arterial cannulas prior to surgery, we used
different cannulas pre- and postoperatively. A consequence
of this could be different absolute values of the BP record-
ings before and after surgery. However, we believe that
the frequency distributions of the signals are unchanged.
Vasoactive and analgesic medications, and fluids were
administered postoperatively according to the individual
FIGURE 5 Cross-wavelet spectra of
the variable pairs iAmp-iSBP and iSBP-
iHR of patient 8, situation D. Time (in
seconds) is shown on the x-axes, and period
(in seconds) on a logarithmic scale on the
y-axes. Power is given by color, according
to the scale next to the plots. Significant
frequencies are marked by white lines and
phase differences by arrows. Both spectra
show high power just above 128s
0.0
0.1
0.2
0.3
0.6
2.7
Cross−wavelet power levels
8D, iAmp−iSBP
64
128
256
Period [s]
0 200 400 600 800 1000 1300 1600
Time [s]
0.0
0.2
0.3
0.6
1.0
3.8
Cross−wavelet power levels
8D, iSBP−iHR
64
128
256
Period (s)
0 200 400 600 800 1000 1300 1600
Time (s)
a
b
8 of 9
|
KNAI et Al.
patient's clinical state. Thus, the patients may have received
different amounts of medications, with varying contribu-
tion to their oscillatory profile.
5
|
CONCLUSION
In this study, we decomposed BP and ECG recordings from
eight cardiac surgery patients to time series of SBP, HR,
and R-wave amplitude. Four 30-min selections, represent-
ing key events of the perioperative course, were analyzed
with the CWT and average wavelet power spectra were used
to illustrate the patients’ circulatory frequency distributions.
We identified oscillatory components in all variables, pa-
tients, and situations, and they were more or less randomly
distributed through the examined frequency range. The high
variety in circulatory oscillations seems to represent inter-
individual variations, more than factors of the performed
surgery. Linking the circulatory frequency distributions to
the overall complexity of circulatory signals, no reduction of
such is identified. Considering the heterogeneity of our find-
ings, we have not presented information that is suitable for
use in any monitoring device or other clinical decision tools.
The study is limited by challenges regarding noise-handling,
and generalizability due to small sample size.
ACKNOWLEDGMENTS
We thank Bjørn Gardsjord Lio and Fredrik Einar Tobias
Axelsson for assistance in collecting data, and Tord Åsnes
for designing Figure1.
CONFLICT OF INTEREST
All authors declare that they have no competing interests.
ORCID
Kathrine Knai https://orcid.org/0000-0002-7586-8029
REFERENCES
Addison, P. S. (2005). Wavelet transforms and the ECG: A review.
Physiological Measurement, 26(5), R155–R199. https://doi.
org/10.1088/0967-3334/26/5/R01
Akselrod, S., Gordon, D., Ubel, F. A., Shannon, D. C., Berger, A. C., &
Cohen, R. J. (1981). Power spectrum analysis of heart rate fluctuation:
A quantitative probe of beat-to-beat cardiovascular control. Science,
213(4504), 220–222. https://doi.org/10.1126/scien ce.6166045
Alfons, A.. (2016). robustHD: Robust Methods for High-Dimensional
Data [Internet]. Available from: https://CRAN.R-proje ct.org/packa
ge=robustHD
Bengtsson, H. (2016). R.matlab: Read and Write MAT Files and Call
MATLAB from Within R [Internet]. Available from: https://github.
com/Henri kBeng tsson /R.matlab
Bracic, M., & Stefanovska, A. (1998). Wavelet-based analysis of human
blood-flow dynamics. Bulletin of Mathematical Biology, 60(5),
919–935. https://doi.org/10.1006/bulm.1998.0047
Brody, D. A. (1956). A theoretical analysis of intracavitary blood mass
influence on the heart-lead relationship. Circulation Research, 4(6),
731–738. https://doi.org/10.1161/01.RES.4.6.731
Claydon, V. E., & Krassioukov, A. V. (2008). Clinical correlates of
frequency analyses of cardiovascular control after spinal cord
injury. American Journal of Physiology-Heart and Circulatory
Physiology, 294(2), H668–H678. https://doi.org/10.1152/ajphe
art.00869.2007
Cleveland, W. S., & Devlin, S. J. (1988). Locally weighted regression:
An approach to regression analysis by local fitting. Journal of the
American Statistical Association, 83(403), 596–610. https://doi.
org/10.1080/01621 459.1988.10478639
Fugal, D. L. (2009). Conceptual wavelets in digital signal processing:
An in-depth, practical approach for the non-mathematician (p. 382).
Space & Signals Technical Pub.
Goldberger, A. L. (1996). Non-linear dynamics for clinicians:
Chaos theory, fractals, and complexity at the bedside. The
Lancet, 347(9011), 1312–1314. https://doi.org/10.1016/S0140
-6736(96)90948 -4
Goldberger, A. L., Moody, G. B., & Costa, M. D. (2012). Variability vs.
Complexity [Internet]. Physionet.org. Available from: https://archi
ve.physi onet.org/tutor ials/cv/
Goldstein, B., Fiser, D. H., Kelly, M. M., Mickelsen, D., Ruttimann, U.,
& Pollack, M. M. (1998). Decomplexification in critical illness and
injury: Relationship between heart rate variability, severity of ill-
ness, and outcome. Critical Care Medicine, 26(2), 352–357. https://
doi.org/10.1097/00003 246-19980 2000-00040
Hogue, C. W., Stein, P. K., Apostolidou, I., Lappas, D. G., & Kleiger,
R. E. (1994). Alterations in temporal patterns of heart rate vari-
ability after coronary artery bypass graft surgery. Anesthesiology,
81(6), 1356–1364. https://doi.org/10.1097/00000 542-19941
2000-00009
Johnson, N. (2009). Simply complexity: A clear guide to complexity the-
ory (p. 202). Oxford, UK: Oneworld Publications.
Kaplan, D. T., Furman, M. I., Pincus, S. M., Ryan, S. M., Lipsitz, L.
A., & Goldberger, A. L. (1991). Aging and the complexity of car-
diovascular dynamics. Biophysical Journal, 59(4), 945–949. https://
doi.org/10.1016/S0006 -3495(91)82309 -8
Kleiger, R. E., Miller, J. P., Bigger, J. T., & Moss, A. J. (1987).
Decreased heart rate variability and its association with increased
mortality after acute myocardial infarction. The American Journal
of Cardiology, 59(4), 256–262. https://doi.org/10.1016/0002-
9149(87)90795 -8
Knai, K., Kulia, G., Molinas, M., & Skjaervold, N. K. (2017).
Instantaneous frequencies of continuous blood pressure a compari-
son of the power spectrum, the continuous wavelet transform and the
Hilbert-Huang transform. Advances in Data Science and Adaptive
Analysis, 09(04), 1750009.
Kuo, C.-D., Chen, G.-Y., Lai, S.-T., Wang, Y.-Y., Shih, C.-C., &
Wang, J.-H. (1999). Sequential changes in heart rate variability
after coronary artery bypass grafting. The American Journal
of Cardiology, 83(5), 776–779. https://doi.org/10.1016/S0002
-9149(98)00989 -8
Lipsitz, L. A., & Goldberger, A. L. (1992). Loss of “complexity”
and aging: Potential applications of fractals and chaos theory
to senescence. JAMA, 267(13), 1806. https://doi.org/10.1001/
jama.1992.03480 13012 2036
Nahiyan, K. M. T., & Amin, A. A. (2017). Removal of ECG baseline
wander using Savitzky-Golay filter based method. Bangladesh
|
9 of 9
KNAI et Al.
Journal of Medical Physics, 8(1), 32–45. https://doi.org/10.3329/
bjmp.v8i1.33932
Pomeranz, B., Macaulay, R. J., Caudill, M. A., Kutz, I., Adam, D.,
Gordon, D., … Cohen, R. J.. (1985). Assessment of autonomic func-
tion in humans by heart rate spectral analysis. American Journal of
Physiology-Heart and Circulatory Physiology, 248(1), H151–H153.
https://doi.org/10.1152/ajphe art.1985.248.1.H151
R Foundation for Statistical Computing. (2018). R: A Language and
Environment for Statistical Computing [Internet]. Vienna, Austria:
R Foundation for Statistical Computing. Available from: https://ww-
w.R-proje ct.org
Riordan, W. P., Norris, P. R., Jenkins, J. M., & Morris, J. A. (2009).
Early loss of heart rate complexity predicts mortality regardless of
mechanism, anatomic location, or severity of injury in 2178 trauma
patients. Journal of Surgical Research, 156(2), 283–289. https://doi.
org/10.1016/j.jss.2009.03.086
Roesch, A., & Schmidbauer, H. (2018). WaveletComp: Computational
Wavelet Analysis [Internet]. Available from: https://CRAN.R-proje
ct.org/packa ge=Wavel etComp
Shaffer, F., McCraty, R., & Zerr, C. L. (2014). A healthy heart is
not a metronome: An integrative review of the heart's anatomy and
heart rate variability. Frontiers in Psychology, 5, 1040. https://doi.
org/10.3389/fpsyg.2014.01040
Signal developers. (2013). signal: Signal processing [Internet].
Available from: http://r-forge.r-proje ct.org/proje cts/signa l/
Takahashi, A. C. M., Porta, A., Melo, R. C., Quitério, R. J., da Silva,
E., Borghi-Silva, A., … Catai, A. M.. (2012). Aging reduces com-
plexity of heart rate variability assessed by conditional entropy and
symbolic analysis. Internal and Emergency Medicine, 7(3), 229–
235. https://doi.org/10.1007/s1173 9-011-0512-z
Task Force of the European Society of Cardiology and the North
American Society of Pacing and Electrophysiology. (1996). Heart
rate variability standards of measurement, physiological interpreta-
tion, and clinical use. Circulation, 93(5), 1043–1065.
Umetani, K., Singer, D. H., McCraty, R., & Atkinson, M. (1998).
Twenty-four hour time domain heart rate variability and heart
rate: Relations to age and gender over nine decades. Journal of
the American College of Cardiology, 31(3), 593–601. https://doi.
org/10.1016/S0735 -1097(97)00554 -8
Wolf, M. M., Varigos, G. A., Hunt, D., & Sloman, J. G. (1978). Sinus
arrhythmia in acute myocardial infarction. Medical Journal of
Australia, 2(2), 52–53.
SUPPORTING INFORMATION
Additional supporting information may be found online in
the Supporting Information section.
How to cite this article: Knai K, Aadahl P,
Skjaervold NK. Cardiac surgery does not lead to loss
of oscillatory components in circulatory signals.
Physiol Rep. 2020;8:e14423. https://doi.org/10.14814/
phy2.14423
... Complexity analysis of RR and SAP variability series was less frequently performed after cardiac surgery. Complexity analysis of RR and SAP variability series suggested a more variable response to cardiac surgery across individuals and groups [26,28]. Information about the impact of cardiac surgery on cerebrovascular control is even more limited [29]. ...
Article
Full-text available
We assessed the effect of surgical aortic valve replacement (SAVR) on cardiovascular and cerebrovascular controls via spontaneous variability analyses of heart period, approximated as the temporal distance between two consecutive R-wave peaks on the electrocardiogram (RR), systolic, diastolic and mean arterial pressure (SAP, DAP and MAP) and mean cerebral blood flow (MCBF). Powers in specific frequency bands, complexity, presence of nonlinear dynamics and markers of cardiac baroreflex and cerebral autoregulation were calculated. Variability series were acquired before (PRE) and after (POST) SAVR in 11 patients (age: 76±5 yrs, 7 males) at supine resting and during active standing. Parametric spectral analysis was performed based on the autoregressive model. Complexity was assessed via a local nonlinear prediction approach exploiting the k-nearest-neighbor strategy. The presence of nonlinear dynamics was checked by comparing the complexity marker computed over the original series with the distribution of the same index assessed over a set of surrogates preserving distribution and power spectral density of the original series. Cardiac baroreflex and cerebral autoregulation were estimated by assessing the transfer function from SAP to RR and from MAP to MCBF and squared coherence function via the bivariate autoregressive approach. We found that: i) orthostatic challenge had no effect on cardiovascular and cerebrovascular control markers in PRE; ii) RR variance was significantly reduced in POST; iii) complexity of SAP, DAP and MAP variabilities increased in POST with a greater likelihood of observing nonlinear dynamics over SAP compared to PRE at supine resting; iv) the amplitude of MCBF variations and MCBF complexity in POST remained similar to PRE; v) cardiac baroreflex sensitivity decreased in POST, while cerebrovascular autoregulation was preserved. SAVR induces important changes of cardiac and vascular autonomic controls and baroreflex regulation in patients exhibiting poor reactivity of cardiovascular regulatory mechanisms, while cerebrovascular autoregulation seems to be less affected.
Article
Full-text available
Continuous biological signals, like blood pressure recordings, exhibit non-linear and non-stationary properties which must be considered when analyzing them. Heart rate variability analyses have identified several frequency components and their autonomic origin. There is need for more knowledge on the time-changing properties of these frequencies. The power spectrum, continuous wavelet transform and Hilbert-Huang transform are applied on a continuous blood pressure signal to investigate how the different methods compare to each other. The Hilbert-Huang transform shows high ability to analyzing such data, and can, by identifying instantaneous frequency shifts, provide new insights into the nature of these kinds of data.
Article
Full-text available
p>ECG (Electrocardiogram) is a measure of heart’s electrical activity. As the body is a volume conductor, ECG signal can be recorded from the body surface. The signal while being recorded from the body surface gets corrupted by various types of noise or artifact. Among these, baseline wander is a type of noise that severely hampers the ECG signal. Baseline wander is particularly of very low frequency; it causes the ECG signal to deviate from its isoelectric line and causes the signal to ride on the lower frequency artifact. The proposed method is based on Savitzky-Golay filter, which is a moving average filter that takes into consideration the polynomial order along with moving averaging when approximating the signal. It enables to approximate the baseline wander quite efficiently. Though in some cases it distorts the ECG signal to some extent, when compared with usual polynomial fitting method, it demonstrates superiority in terms of accuracy, simplicity and generalization. Bangladesh Journal of Medical Physics Vol.8 No.1 2015 32-45</p
Article
Full-text available
Heart rate variability (HRV), the change in the time intervals between adjacent heartbeats, is an emergent property of interdependent regulatory systems that operate on different time scales to adapt to challenges and achieve optimal performance. This article briefly reviews neural regulation of the heart, and its basic anatomy, the cardiac cycle, and the sinoatrial and atrioventricular pacemakers. The cardiovascular regulation center in the medulla integrates sensory information and input from higher brain centers, and afferent cardiovascular system inputs to adjust heart rate and blood pressure via sympathetic and parasympathetic efferent pathways. This article reviews sympathetic and parasympathetic influences on the heart, and examines the interpretation of HRV and the association between reduced HRV, risk of disease and mortality, and the loss of regulatory capacity. This article also discusses the intrinsic cardiac nervous system and the heart-brain connection, through which afferent information can influence activity in the subcortical and frontocortical areas, and motor cortex. It also considers new perspectives on the putative underlying physiological mechanisms and properties of the ultra-low-frequency (ULF), very-low-frequency (VLF), low-frequency (LF), and high-frequency (HF) bands. Additionally, it reviews the most common time and frequency domain measurements as well as standardized data collection protocols. In its final section, this article integrates Porges’ polyvagal theory, Thayer and colleagues’ neurovisceral integration model, Lehrer, Vaschillo, and Vaschillo’s resonance frequency model, and the Institute of HeartMath’s coherence model. The authors conclude that a coherent heart is not a metronome because its rhythms are characterized by both complexity and stability over longer time scales. Future research should expand understanding of how the heart and its intrinsic nervous system influence the brain.
Article
Full-text available
The concept of "complexity," derived from the field of nonlinear dynamics, can be adapted to measure the output of physiologic processes that generate highly variable fluctuations resembling "chaos." We review data suggesting that physiologic aging is associated with a generalized loss of such complexity in the dynamics of healthy organ system function and hypothesize that such loss of complexity leads to an impaired ability to adapt to physiologic stress. This hypothesis is supported by observations showing an age-related loss of complex variability in multiple physiologic processes including cardiovascular control, pulsatile hormone release, and electroencephalographic potentials. If further research supports this hypothesis, measures of complexity based on chaos theory and the related geometric concept of fractals may provide new ways to monitor senescence and test the efficacy of specific interventions to modify the age-related decline in adaptive capacity. (JAMA. 1992;267:1806-1809)
Article
Full-text available
This study sought to define the effects of age and gender effects on the normal range of time domain heart rate variability (HRV) over nine decades in healthy subjects. Low HRV is considered an independent marker of mortality risk. However, the age-related decline in HRV may limit its predictive value, particularly in the elderly. Delineation of the range of HRV in healthy subjects over the life span is needed. Gender-related differences in HRV also need clarification. We determined, according to decade, 24-h heart rate (HR) and HRV of 260 healthy subjects (10 to 99 years old; 112 male, 148 female) by means of five standard time domain measures: standard deviation of all normal sinus RR intervals over 24 h (SDNN), standard deviation of the averaged normal sinus RR intervals for all 5-mm segments (SDANN), mean of the standard deviations of all normal sinus RR intervals for all 5-min segments (SDNN index), root-mean-square of successive normal sinus RR interval difference (rMSSD) and the percentage of successive normal sinus RR intervals >50 ms (pNN50). 1) HRV decreased with aging, the pattern of change being measure dependent. HRV (SDNN and SDANN) decreased only very gradually, reaching 60% of baseline (second-decade values) by the tenth decade. With the SDNN index, HRV decreased linearly with aging, reaching 46% of baseline by the tenth decade. Using pNN50 and rMSSD, HRV decreased most rapidly, reaching 24% and 47% of baseline, respectively, by the sixth decade and then stabilized. 2) Using the SDNN index, rMSSD and pNN50, HRV of subjects >65 years old fell below published cutpoints for increased risk of mortality in 25%, 12% and 4%, respectively. 3) At age <30 years, HRV for all measures was lower in female than male subjects. Gender differences decreased at age >30 years and disappeared at age >50 years. 4) HR also declined with aging but much more slowly. HR at age <50 years was faster in female than in male subjects. Gender differences disappeared thereafter. 1) Using all measures, HRV of healthy subjects declines with aging, with measure-dependent patterns. 2) Using the SDNN index, rMSSD and pNN50, HRV of healthy subjects, particularly those >65 years old, may decrease to below levels associated with increased risk of mortality. 3) Gender influences HRV. Gender differences in HRV are age and measure dependent. 4) Age and gender also affect heart rate.
Article
Full-text available
Increasing age is associated with a reduction in overall heart rate variability as well as changes in complexity of physiologic dynamics. The aim of this study was to verify if the alterations in autonomic modulation of heart rate caused by the aging process could be detected by Shannon entropy (SE), conditional entropy (CE) and symbolic analysis (SA). Complexity analysis was carried out in 44 healthy subjects divided into two groups: old (n = 23, 63 ± 3 years) and young group (n = 21, 23 ± 2). It was analyzed SE, CE [complexity index (CI) and normalized CI (NCI)] and SA (0V, 1V, 2LV and 2ULV patterns) during short heart period series (200 cardiac beats) derived from ECG recordings during 15 min of rest in a supine position. The sequences characterized by three heart periods with no significant variations (0V), and that with two significant unlike variations (2ULV) reflect changes in sympathetic and vagal modulation, respectively. The unpaired t test (or Mann-Whitney rank sum test when appropriate) was used in the statistical analysis. In the aging process, the distributions of patterns (SE) remain similar to young subjects. However, the regularity is significantly different; the patterns are more repetitive in the old group (a decrease of CI and NCI). The amounts of pattern types are different: 0V is increased and 2LV and 2ULV are reduced in the old group. These differences indicate marked change of autonomic regulation. The CE and SA are feasible techniques to detect alteration in autonomic control of heart rate in the old group.
Article
Objectives: To determine if decomplexification of heart rate dynamics occurs in critically ill and injured pediatric patients. We hypothesized that heart rate power spectra, a measure of heart rate dynamics, would inversely correlate with measures of severity of illness and outcome. Design: A prospective clinical study. Setting: A 12-bed pediatric intensive care unit (ICU) in a tertiary care children's hospital. Patients: One hundred thirty-five consecutive pediatric ICU admissions. Interventions: None. Measurements and Main Results: We compared heart rate power spectra with the Pediatric Risk of Mortality (PRISM) score, the Pediatric Cerebral Performance Category (PCPC), and the Pediatric Overall Performance Category (POPC). We found significant negative correlations between minimum low-frequency and high-frequency heart rate power spectral values recorded during ICU stay and the maximum PRISM score (log low-frequency heart rate power vs. PRISM, r2 = .293, p < .001; and log high-frequency heart rate power vs. PRISM, r2 = .243, p < .001) and outcome at ICU discharge (log low-frequency heart rate power vs. POPC or PCPC, r2 = .429, p < .001; and log high-frequency heart rate power vs. POPC or PCPC, r2 = .271, p < .001). Conclusions: Our data support the hypothesis that measures of heart rate power spectra are inversely related and negatively correlated to severity of illness and outcome in critically ill and injured children. The phenomenon of decomplexification of physiologic dynamics may have important clinical implications in critical illness and injury. (Crit Care Med 1998; 26:352-357)
Article
Locally weighted regression, or loess, is a way of estimating a regression surface through a multivariate smoothing procedure, fitting a function of the independent variables locally and in a moving fashion analogous to how a moving average is computed for a time series. With local fitting we can estimate a much wider class of regression surfaces than with the usual classes of parametric functions, such as polynomials. The goal of this article is to show, through applications, how loess can be used for three purposes: data exploration, diagnostic checking of parametric models, and providing a nonparametric regression surface. Along the way, the following methodology is introduced: (a) a multivariate smoothing procedure that is an extension of univariate locally weighted regression; (b) statistical procedures that are analogous to those used in the least-squares fitting of parametric functions; (c) several graphical methods that are useful tools for understanding loess estimates and checking the assumptions on which the estimation procedure is based; and (d) the M plot, an adaptation of Mallow's Csubp/sub procedure, which provides a graphical portrayal of the trade-off between variance and bias, and which can be used to choose the amount of smoothing.
Article
Reduced heart rate (HR) complexity (e.g., a lack of randomness or unpatterned variability) is an established predictor of trauma patient mortality. However, this finding has not been validated across the diverse spectrum of traumatic injury, and underlying mechanisms of this relationship are poorly understood. Two thousand one hundred seventy-eight trauma patients were admitted directly to the intensive care unit (ICU), and had sufficient (>6h) continuous integer heart rate data within the first d. Patients were stratified by location of isolated severe injury (head, torso, both, or neither), primary mechanism (blunt or penetrating), and probability of survival, an accepted scoring system based on age, admission vital signs, and injury type and severity. HR multiscale entropy (MSE) was calculated (sum of scales, Costa's algorithm, physionet.org, m=2, r=0.15) to estimate complexity. Univariate analysis was performed by comparing MSE between survivors and nonsurvivors in each subgroup. Multivariate analysis incorporated logistic regression to characterize the relationship between MSE and risk of death, controlling for probability of survival. The MSE odds ratios (OR) and area under the receiver operator curve (AUC) were calculated. Reduced MSE was significantly associated with increasing mortality, and was independent of probability of survival in all multivariate analyses (OR 0.87-0.94). This range of odds ratios implies that a patient with an MSE of 15 has roughly a 2- to 6-fold increase in odds of death versus a patient with an MSE of 25. The relationship between MSE and death was moderately stronger in patients with isolated severe head injury versus torso injury, and significantly stronger in patients with penetrating versus blunt mechanism of injury. MSE measured early in the hospital stay remained a robust predictor of mortality in all subgroups, even stratified by narrow ranges of probability of survival. Early reduction of heart rate complexity is an important risk factor across diverse injury etiology. This suggests common underlying physiologic mechanisms linking the loss of biologic complexity to death.