Content uploaded by Kristian Solem
Author content
All content in this area was uploaded by Kristian Solem
Content may be subject to copyright.
IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012 45
Noninvasive Techniques for Prevention of
Intradialytic Hypotension
Leif Sörnmo, Senior Member, IEEE, Frida Sandberg, Eduardo Gil, and Kristian Solem
Methodological Review
Abstract—Episodes of hypotension during hemodialysis treat-
ment constitutes an important clinical problem which has received
considerable attention in recent years. Despite the fact that
numerous approaches to reducing the frequency of intradia-
lytic hypotension (IDH) have been proposed and evaluated, the
problem has not yet found a definitive solution—an observation
which, in particular, applies to episodes of acute, symptomatic hy-
potension. This overview covers recent advances in methodology
for predicting and preventing IDH. Following a brief overview
of well-established hypotension-related variables, including
blood pressure, blood temperature, relative blood volume, and
bioimpedance, special attention is given to electrocardiographic
and photoplethysmographic (PPG) variables and their significance
for IDH prediction. It is concluded that cardiovascular variables
which reflect heart rate variability, heart rate turbulence, and
baroreflex sensitivity are important to explore in feedback control
hemodialysis systems so as to improve their performance. The
analysis of hemodialysis-related changes in PPG pulse wave
properties hold considerable promise for improving prediction.
Index Terms—Cardiac information, ECG, feedback control,
hemodialysis, intradialytic hypotension (IDH), photoplethysmog-
raphy (PPG), prediction.
I. INTRODUCTION
HEMODIALYSIS is since long a well-established treat-
ment of patients with serious kidney problems. The treat-
ment improves dramatically the living conditions for this group
of patients, but it is also associated with episodes of intradia-
lytic hypotension (IDH) which occur in approximately 25% of
all sessions, thereby making IDH the most common complica-
tion during hemodialysis [1]. This percentage increases even
Manuscript received February 03, 2012; revised May 13, 2012; accepted July
06, 2012. Date of publication July 24, 2012; date of current version December
06, 2012. This work was supported by the Swedish Research Council under
Grant 2010-4772. The work of E. Gil in Lund was supported by the Ministerio
de Ciencia y Tecnología (FEDER, TEC2010-21703-C03-02), Centro de Inves-
tigacíon Biomédica en Red de Bioingeniería, Biomateriales y Nanomedicina
(CIBER-BBN), ARAID and Ibercaja (Programa de Apoyo A LA I+D+i), Grupo
Consolidado GTC from DGA.
L. Sörnmo and F. Sandberg are with the Department of Electrical and Infor-
mation Technology and Center for Integrative Electrocardiology, Lund Univer-
sity, Lund, Sweden (e-mail: leif.sornmo@eit.lth.se; frida.sandberg@eit.lth.se).
E. Gil is with CIBER-BBN and with the Communications Technology
Group, Aragón Institute of Engineering Research, Zaragoza University,
Zaragoza, Spain (e-mail: edugilh@unizar.es).
K. Solem is with the Research Department, Gambro Lundia AB, Lund,
Sweden (e-mail: kristian.solem@gambro.com).
Digital Object Identifier 10.1109/RBME.2012.2210036
further as the population grows older. The short-term conse-
quence of such episodes is obviously decreased wellbeing for
the patient, manifested by symptoms of fatigue, cramps, and
vomiting. Since IDH can lead to premature termination of the
session, the patient may end up with insufficient clearance of the
blood. In the long-term perspective, IDH can cause permanent
damage to the heart and intestines as well as occlusion of the
arteriovenous fistula [2]. Repeated episodes of IDH have been
established as a significant and independent risk factor for in-
creased morbidity and mortality in hemodialysis patients [3],
[4]. Hypotension which occurs at the onset of the hemodialysis
session has been established as particularly serious [5]. Further-
more, it is well known that IDH can induce cardiac arrhythmias
and predispose to myocardial ischemia, which in turn increases
the risk for sudden cardiac death, being a common cause of
death in dialysis patients [6].
The causes of hypotension are multifactorial. The primary
factor is the decrease in blood volume that occurs during
hemodialysis that results from fluid withdrawal of the vascular
compartment during ultrafiltration and insufficient refilling of
fluid from the interstitial compartment to the vascular com-
partment. The picture is further complicated by several, rather
diverse factors, including impaired peripheral vasoconstric-
tion, autonomic dysfunction, arteriosclerosis, cardiovascular
pathologies such as left ventricular hypertrophy and dilated
cardiomyopathy, hydration, and medication [7]. During a
hemodialysis session, eating and changes in body position may
also serve as triggers of a hypotensive episode.
The occurrence of IDH is found to be more frequent in
the standard thrice-weekly hemodialysis treatment than in
short daily or nocturnal treatment because the former type of
treatment requires a higher dose of ultrafiltration. While short
everyday treatment is not practically feasible when delivered
in the hospital, the increasing dissemination of home-based
hemodialysis may change this as future treatment paradigms
are evolving.
In many hospitals, the clinical management of IDH remains
synonymous to the placement of the patient in Trendelenburg
position, i.e., supine body position with the feet held higher
than the head [8]. This placement is accompanied with sub-
stantial slowing of the ultrafiltration rate (UFR) so that the re-
duction in blood volume due to fluid removal is slowed down.
Another means is to infuse a hypertonic solution (low-volume
saline or glucose) which increases the blood volume and accord-
ingly the blood pressure; the solution facilitates osmotic shift of
1937-3333/$31.00 © 2012 IEEE
46 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
fluid from the extravascular to the intravascular compartment
[9]. These types of actions are invoked when the patient al-
ready exhibits symptoms, and therefore it is highly desirable
to prevent episodes of IDH well in advance so that appropriate
measures can be taken. A complicating factor is that the occur-
rence of IDH cannot always be observed through external signs.
While most patients complain about dizziness or nausea when
hypotension is about to occur, some patients do not display any
precursory symptoms whatsoever [8].
The prevention of IDH has been the subject of intense re-
search in recent years and has resulted in a range of techniques
which aim at solving this issue. These techniques include
improved assessment of the patient’s “dry weight”, cooler
dialysate temperature, dialysate sodium concentration (DSC)
profiling, and online blood volume monitoring. Interesting
reviews have been published on this topic, all of them authored
from a clinically oriented perspective [7], [10]–[12].
Considering that the problem of predicting IDH entails sig-
nificant engineering challenges, it is quite remarkable that the
problem has received so little attention among researchers in
the biomedical engineering community; the vast majority of ar-
ticles on this topic has been published in clinical journals. In
fact, the significance of technology in hemodialysis is far from
uncontroversial, and opinions diverge quite considerably. While
many researchers are optimistic about its significance [10], [11],
[13], others are much more skeptical since the major technolog-
ical advances in dialysis have not yet been translated into longer
patient survival [7], [12]. Irrespective of the opinion held on this
issue, continuous monitoring of patient status during hemodial-
ysis, based on lightweight, nonintrusive sensor technology, ap-
pears as inevitable if the goal is to ensure that the nephrologist
has immediate access to relevant clinical information.
Hence, there is room for much more engineering research
whose purpose is to develop methods which can dramatically
reduce the frequency of IDH. This opinion is further under-
lined by a recent review which concluded that [13]: “What
is badly needed in this area of clinical research are improved
methods to reduce the frequency of intradialytic hypotension,
thereby avoiding its untoward effects. Until progress is made
in mitigating the incidence of intradialytic hypotension, stan-
dard thrice-weekly 3- to 4-h hemodialysis will continue to be
episodically unpleasant”.
The present paper takes more of an engineering perspective
when reviewing and discussing noninvasive techniques de-
veloped for the prediction and prevention of IDH. Following
a brief overview of well-studied hypotension-related vari-
ables in Section III, special attention is given to more recent
techniques with which variables from electrocardiographic
or photoplethysmographic signals are subject to analysis; see
Sections IV and V, respectively. Feedback control in hemodial-
ysisisdiscussedinSectionVI,withreferencetobothfuzzyand
model-based control. The paper is concluded with a discussion
on future perspectives.
DEFINITION OF INTRADIALYTIC HYPOTENSION
The clinical definition of IDH has not been universally agreed
upon, but varies quite considerably among studies in the liter-
ature. Since blood pressure values can indicate hypotension in
one patient, while they are judged as normal in another patient
who suffers from chronic low blood pressure, a definition which
only involves absolute blood pressure values is not meaningful.
Rather, it is warranted to define criteria which account for rela-
tive reduction in blood pressure during a session as well as for
episodes of acute symptomatic hypotension.
One recently proposed definition of IDH embraces the fol-
lowing three criteria [14] (see also [15]):
1) if predialysis systolic arterial pressure (SAP) is greater
than 100 mmHg, then any episode with SAP less than 90
mmHg, even without complaints;
2) if predialysis SAP is less than 100 mmHg, then any SAP
reduction by at least 10% associated with complaints;
3) any SAP reduction of 25% or more of the predialysis value
with the typical symptoms requiring specific intervention.
Simpler definitions of IDH have also been suggested. For
example, only the last of the three criteria has been employed,
replacing the 25% reduction in SAP with 20% as threshold
value [16].
The time course of blood pressure can be involved in the def-
inition of IDH as reflected by the duration of the reduction in
blood pressure [17]. The reduction is then characterized by its
rate in terms of mmHg/min, with a high rate categorized as acute
IDH, whereas a low rate as nonacute IDH. An important reason
for including rate in the definition is that the etiology of IDH
may differ depending on duration and/or rate. Since changes and
counteractions of the body relate to etiology, methods should
likely differ in design depending on whether the target is to pre-
dict acute IDH or not.
The predominant problem treated in the literature is to predict
prior to the dialysis session whether the patient will suffer from
IDH or not (“offline prediction”). Since no temporal aspect is in-
volved, the related design problem must be considered as being
simpler than that associated with prediction of acute IDH since
the latter type will have to involve online data processing.
II. WELL-STUDIED HYPOTENSION-RELATED VARIABLES
Over the years, numerous clinical studies have investigated
the significance of variables that characterize the circulatory
system in relation to IDH, notably blood temperature and blood
volume. Bioimpedance is another well-established variable
which may be continuously monitored for the purpose of
assessing a patient’s dry weight, i.e., the weight at the end of a
session at which most excess body fluid has been removed, and
below which the patient is likely to develop IDH [8]. Changes
in “sensor” variables, such as blood temperature, blood volume,
and bioimpedance, are analyzed and used to modify “actuator”
variables, such as UFR, DSC, and dialysate temperature, so
that the frequency of IDH may be reduced. The relationship be-
tween an actuator variable, in this case UFR, and the frequency
of IDH, expressed as a percentage, is exemplified in Fig. 1. For
many years, the settings of the dialysis machine were mod-
ified manually by the operator; however, such modifications
have more recently been subjected to automation within the
framework of feedback control; see Section VI. In this section
follows a brief description of the variables blood pressure,
blood temperature, relative blood volume, and bioimpedance.
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 47
Fig. 1. Percentage of intradialytic hypotension as a function of ultrafiltration
rate (adapted from [18]). Hypotension was defined as a drop in arterial blood
pressure of more than 30 mmHg. It was concluded that the percentage of hy-
potension increases exponentially with ultrafiltration rate.
A. Blood Pressure
The arterial blood pressure may seem the natural starting
point for developing an algorithm for online prediction and pre-
vention of hypotension. However, such an approach is compli-
cated by the fact that sensors for the measurement of continuous
blood pressure, e.g., inflatable cuff-based devices, are highly in-
convenient to wear throughout the entire hemodialysis session.
Moreover, frequent cuff-based measurements have a direct in-
fluence on the blood pressure itself. On the other hand, indirect
measurement of changes in SAP can be easily accomplished
with the photoplethysmographic technique, involving a pulse
oximeter attached to the finger and combined with temporal in-
formation derived from an ECG lead; see Section V.
B. Blood Temperature
It is well known that the patient’s temperature tends to in-
crease during hemodialysis for the conventional dialysate tem-
perature of 37 C. Hemodialysis increases the core body temper-
ature since more blood, normally flowing to the body surface for
heat dissipation, remains in the central circulation to preserve
the core blood plasma volume which is reduced in response to
ultrafiltration. Sympathetic activity is known to increase in re-
sponse to ultrafiltration and leads to peripheral vasoconstriction
which reduces heat dissipation. Although a certain amount of
thermal energy is removed during hemodialysis through the ex-
tracorporeal blood circulation system, the net effect is still heat
accumulation which tends to increase the core body tempera-
ture. When this increase in temperature overcomes peripheral
vasoconstriction, the risk of IDH increases, especially in pa-
tients with low core body temperatures [11].
Cooling of the dialysate was already in the 1980s found to
have a beneficial effect on cardiovascular stability and to re-
duce the frequency of IDH [19]; the reduction in temperature
being from the standard 37 C to about 35 C. Several studies
have since then reported a similar finding, many of which were
compiled in a recent meta-analysis of 22 studies showing that
a reduction in dialysate temperature is an effective intervention
for reducing the frequency of IDH. Cool hemodialysis led to
that IDH occurred 7.1 times less frequently than when standard
hemodialysis was prescribed [20]. The results applied to dif-
ferent cooling profiles that spanned in complexity from a fixed,
empirical reduction of the dialysate temperature to online moni-
toring of blood temperature and cooling based on feedback con-
trol, cf. Section VI. A disadvantage with blood cooling is the
increased risk of shivering and cold sensation [21].
C. Relative Blood Volume
The loss in blood volume that occurs during hemodialysis
constitutes an important cause of IDH [22], [23]. This loss
is related to the ultrafiltration process during which fluid is
withdrawn from the circulation. Since UFR is almost always
higher than the plasma refilling rate, i.e., the rate at which fluid
moves from the interstitial tissue into the circulation, the blood
volume will decrease as fluid is withdrawn, and hypovolemia
may develop. When the cardiovascular compensatory mecha-
nisms that counteract hypovolemia are insufficient or impaired,
hypotension may develop. Thus, the preservation of blood
volume during hemodialysis represents an important target for
preventing IDH.
Historically, blood volume has always been expressed as a
relative measure that reflects changes in blood volume relative
the time for onset of hemodialysis. Although it is preferable to
measure absolute blood volume, there are many confounding
factors which renders such a measurement difficult. A number
of techniques are available for continuous, noninvasive mea-
surement of relative blood volume which explore different types
of blood constituent such as hemoglobin, hematocrit, or the con-
centration of total plasma proteins. Hemoglobin and hematocrit
are measured by quantifying the absorption of monochromatic
light in blood, whereas protein concentration is estimated from
the velocity of sound waves in blood. Comparing the measure-
ments produced by the different techniques, all of them could
detect changes in relative blood volume during hemodialysis,
although the resulting measurements differed appreciably be-
tween the techniques [24].
Several clinical studies have investigated the relationship be-
tween the reduction in relative blood volume and the develop-
ment of ultrafiltration-induced IDH. However, as pointed out in,
e.g., [7] and [25], most studies have been unable to establish a
close relationship between these two factors: the reduction in
relative blood volume at the time for symptomatic IDH did not
differ significantly from what was observed in hypotension-free
sessions. This finding suggests that IDH cannot be predicted by
simply requiring the relative blood volume to drop below a cer-
tain fixed threshold level. Rather, the level is highly patient-de-
pendent and, accordingly, much more difficult to determine. Im-
proved prediction performance was achieved when the shape of
the trend of relative blood volume was analyzed with respect to
features such as decreased long-term variability [26], increased
irregularity [27], or time for switch from an exponential to a
linear decay [28].
48 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
Relative blood volume can be monitored for the purpose of
controlling UFR and DSC in response to changes in relative
blood volume, cf. Section VI. Sodium concentration is an im-
portant determinant of plasma conductivity [29] and has an in-
direct influence on relative blood volume as it modifies plasma
osmolarity which controls the plasma refilling process. Thus,
DSC serves as an actuator variable on relative blood volume,
and is increased during hemodialysis in order to reduce the fre-
quency of IDH. The role of sodium profiling for reducing this
frequency is not fully concordant. One study concluded that the
frequency can be reduced with about 15% [30], whereas an-
other study found no difference between hemodialysis with ei-
ther fixed or profiled sodium concentration [31].
D. Bioimpedance
The measurement of bioelectrical impedance is a well estab-
lished, noninvasive technique for assessing the hydration status
of the body [32], [33]. Impedance is decomposed into resis-
tance, being the opposition to the flow of a current passing
through both intra- and extra-cellular fluid, and reactance, being
the capacitative component of cellular membranes. Changes in
whole-body fluid volume can be assessed by measuring changes
in impedance, with increased impedance corresponding to de-
creased fluid volume. The contribution of the trunk and limbs
to whole-body impedance is about 10% and 90%, respectively
[34]. The impedance is measured by injecting an alternating cur-
rent, traditionally with a single frequency of 50 kHz, using elec-
trodes placed on the hand and foot. Before being analyzed, the
impedance measurement is usually normalized with the sub-
ject’s height as it approximates the length of the human con-
ductor. The impedance measurements are highly site-depen-
dent, and consequently one cannot compare measurements from
different sites. In general, measurements made from the legs
should be more profoundly influenced by ultrafiltration than are
central measurements.
In hemodialysis, the impedance measurement has been ex-
plored for the purpose of estimating dry weight [35], [36]. While
the use of impedance-based techniques has been claimed to
offer “the benefits of avoiding deliberate search of hypotension”
[35], very few studies have actually considered such techniques
for the prediction of IDH. In an early study, it was found that
impedance is associated with both a very low positive predictive
value (42%) and poor sensitivity (66%) when employed for the
prediction of IDH [37]. It was concluded that the investigated
impedance-based technique was unsuitable for predicting hy-
potensive episodes.
More recently, the prediction of IDH was suggested as a
possible application of a method for bioelectrical impedance
vector analysis, however, the method’s potential was not fur-
ther explored for this purpose [38], [39]. The method analyzes
impedance values in the complex plane, spanned by height-nor-
malized resistance and reactance, by comparing the values
to a reference population characterized by the equidensity
contours of a bivariate Gaussian probability density function.
A value is judged to reflect tissue dehydration when it falls
outside an ellipsoid-shaped tolerance zone so that both resis-
tance and reactance exceed their respective mean values. The
trajectory resulting from repeated measurements thus indicates
how hydration changes during hemodialysis treatment. From
their results, the authors concluded that a trajectory pattern of
dehydration is consistent with hypovolemia as a cause of hy-
potension, whereas a pattern of hyperhydration would indicate
other causes [38].
Multifrequency bioelectrical impedance offers the advantage
of providing measurements on both intra- and extra-cellular vol-
umes. Such an approach has been investigated as a means to im-
prove the poor performance of single frequency measurements
[40]. The study showed though that the intradialytic time course
of multifrequency measurements could not serve as a predictor
of hypotension in the individual patient. In another study, it
was concluded that monitoring of relative segmental changes in
extracellular volumes using multifrequency bioimpedance may
help to prevent IDH [41]; no quantitative results were, however,
presented to further support this conclusion.
Impedance cardiography is, just like bioelectrical impedance,
a special form of impedance plethysmography, but designed to
characterize variations in blood flow during the cardiac cycle,
in particular the estimation of stroke volume (cardiac output)
[42]. The measurement setup typically involves four electrodes,
two symmetrically positioned on both sides of the neck and
two on both sides of the chest. The resulting impedance cardio-
graphic signal can be used to derive hemodynamic information
which characterize, e.g., thoracic fluid content, blood pressure
(systolic, diastolic, and mean value), systemic vascular resis-
tance, and heart rate. This type of information was investigated
in a study with the aim to monitor impedance-related parame-
ters so that significant hemodynamic instability could be iden-
tified. Out of 35 patients, five experienced instability but none
of them could be correctly identified [43]. Impedance cardio-
graphic monitoring was found to be more promising in a study
involving 48 patients prone to IDH [16]. Out of the 18 patients
with a fall in SAP of 20% or more relative to the baseline value
at the onset of hemodialysis, 11 exhibited a decrease in cardiac
output, and seven exhibited a significant fall in peripheral re-
sistance. No information was presented on whether changes in
cardiac output or peripheral resistance were predictive of IDH
episodes.
Whether impedance measurements are whole body, seg-
mental, or cardiographic in nature, they have, to date, only
found an indirect importance for the prediction of IDH as
reliableinformationondryweightmayhelptoreducethe
frequency of IDH. This type of measurement has a number of
advantages such as being noninvasive, low-cost, and safe. Its
rather poor reproducibility has been improved over the years
[44], [45]. However, modern systems for impedance measure-
ments require the use of a multiple electrode configuration,
making them less suitable for adoption in clinical routine.
Thoracic admittance, which is the reciprocal to thoracic
impedance, has also been studied as a predictor of IDH since it
is considered to reflect thoracic fluid content (and thus central
blood volume) [46]. It was found that the mean absolute value
of the admittance differed significantly in patients which were
prone and resistant to IDH, suggesting that patients with a large
central blood volume are unlikely to experience an IDH, and
vice versa.
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 49
III. ELECTROCARDIOGRAPHIC VARIABLES
Cardiac function has been studied extensively in hemodial-
ysis patients owing to the fact that cardiovascular diseases cause
almost 50% of all deaths [47]—a figure which is dramatically
higher than in the general population. Sudden cardiac death, is-
chemic heart disease, and heart failure are the most common
causes of mortality. Some of this disease burden is unfortunately
caused by the prevailing clinical opinion on what the prescribed
dose of hemodialysis should be [48]. It has been established that
ventricular premature beats (VPBs) and complex ventricular
arrhythmias are both more common during hemodialysis than
after, especially during the last hour of the session [49]. More-
over, myocardial ischemia may be induced during hemodialysis
in its silent manifestation, usually reflected by ST-segment de-
pression, and has been found to occur in a substantial number of
patients. Better understanding of these types of cardiac events
are important for improved patient management as it may lead
to modifications of the hemodialysis prescription and reduced
disease burden.
Although the ECG signal has been recorded during hemodial-
ysis for analysis purposes in numerous cardionephrological
studies, it has not yet become part of clinical routine, most
likely due to the inconvenience of having to attach and wear
electrodes. As a consequence, information on, e.g., the oc-
currence of paroxysmal arrhythmias or ischemic episodes, is
not readily available to the clinical staff. Even if such cardiac
information would be available, it remains largely unclear how
it should be integrated in the dialysis procedure so that more
appropriate treatment parameters can be selected for prevention
of IDH.
A. Heart Rate
The maintenance of blood pressure is a result of compen-
satory actions mediated by the baroreflex, of which one action
is a modest increase in heart rate which is usually observed
during hemodialysis. However, no study has so far been able
to translate information on heart rate into a useful predictor of
IDH. Several studies have concluded that the change in heart
rate during hemodialysis does not differ significantly between
patients which are prone and resistant to hypotension; see, e.g.,
[17] and [50]–[53].
B. Heart Rate Variability
The analysis of heart rate variability (HRV) has proven to be
a powerful noninvasive tool for quantifying the neural regula-
tory mechanisms that control the cardiovascular system. Heart
rate variability has been investigated with reference to countless
pathologies, including myocardial infarction, congestive heart
failure, and diabetic neuropathy [54], [55]. The dynamics of
HRV are usually characterized by time domain analysis, e.g.,
the standard deviation of the RR intervals between normal sinus
beats, or frequency domain analysis where the resulting power
spectrum, by convention, is divided into a low frequency (LF)
band (0.04–0.15 Hz) and a high frequency (HF) band (0.15–0.40
Hz). These two bands are predominantly related to the activities
of the sympathetic and parasympathetic branches of the auto-
nomic nervous system, respectively, and therefore the LF/HF
TAB L E I
HRV LF/HF POWER RATIO I N PAT I E N T S BEING RESISTANT AND PRONE
TO INTRADIALYTIC HYPOTENSION.ASTERISK INDICATES A STATIS TI CAL LY
SIGNIFICANT DIFFERENCE.SEE TEXT FOR FURTHER COMMENTS ON RESULTS
power ratio is a commonly studied spectral parameter reflecting
autonomic balance.
Heart rate variability has been extensively studied in dialysis
patients with respect to hemodynamic instability; see, e.g., [17]
and [50]–[53]. At an early stage, it was pointed out that the LF
component plays a dominant role since it is representative of
an autonomically mediated compensatory response [50]. With
insufficient cardiovascular compensatory mechanisms to coun-
teract a reduction in blood volume during hemodialysis, the car-
diopulmonary and arterial baroreceptor reflex leads to excita-
tion of the sympathetic activity and inhibition of the parasym-
pathetic activity. As a consequence of this reflex, the LF com-
ponent tends to dominate the oscillations that constitute HRV
during sessions without IDH.
Studies on HRV have focused on uremic patients which
are prone or resistant to hypotension, almost unanimously
concluding that the spectral power parameters can discriminate
between these two patient groups, using either the LF power
separately or the LF/HF power ratio. It has been reported on
elevated values of the LF/HF power ratio in sessions without
hypotension, whereas the ratio dropped markedly at the time of
crisis in hemodialysis sessions with hypotension [52]. Table I
presents the results from five HRV studies of which four found
statistical significant differences between patients being resis-
tant and prone to IDH. It should be noted that the comparison
of LF/HF power ratio between studies is rendered somewhat
difficult since the computation of this variable differs slightly
from study to study.
Some of the above-mentioned studies have performed spec-
tral analysis in short successive segments during dialysis, thus
making it possible to monitor the evolution of computed param-
eters in real time. The obvious idea to let changes in the spectral
parameters predict an approaching IDH has so far not been ex-
plored at greater length. While spectral HRV parameters may
not be powerful enough to alone predict IDH with a clinically
acceptable accuracy, they are likely a part of the set of physio-
logical parameters which would be embraced in an online pre-
dictor.
Although frequency domain analysis appears to be the pre-
dominant approach to characterizing HRV, nonlinear time do-
main analysis represents another powerful approach which typ-
ically relies on entropy-based measures [56]–[58]. One of the
few dialysis-related studies pursuing a nonlinear approach was
the one which investigated Shannon entropy of RR intervals oc-
curring just prior to performing dialysis in patients with chronic
50 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
renal failure [59]. This measure of complexity is computed by
integrating the logarithm of the HRV power spectrum over all
frequency bands. The results showed that the Shannon entropy
is strongly correlated to a change in SAP during a hemodialysis
session, making the authors conclude that the entropy measure
may serve as a predictor of a patient’s proneness to IDH.
C. Ventricular Premature Beats and Heart Rate Turbulence
It is well known that VPBs are frequent in dialysis patients
[60], [61] and that they increase in number during hemodial-
ysis when excess potassium is removed [62]. Ventricular ar-
rhythmias in hemodialysis patients has recently been studied
in long-term, ambulatory ECG recordings. The results showed
that VPBs occurred much more frequently during hemodialysis
than they did during the postdialysis period [49]. Patients with
regional wall motion abnormalities, ischemic heart disease, and
left ventricular hypertrophy all had a higher frequency of VPBs
during hemodialysis than those without.
Given these findings, it is somewhat surprising that so few
studies on HRV in dialysis patients disclose information on the
handling of VPBs as well as other ectopic beats. One study
which actually provided such information showed that HRV
analysis could only be performed in 18 out of 30 sessions when
the exclusion criterion required the number of VPBs to be less
than 4% [17]. Since the presence of VPBs perturbs the impulse
pattern initiated by the sinoatrial node, RR intervals neighboring
to a VPB cannot be analyzed. The presence of occasional VPBs
can be adequately handled prior to HRV analysis using correc-
tion techniques [63], [64]. If correction is not performed, the
resulting power spectrum will exhibit fictitious frequency com-
ponents, manifested as a “white noise” level [65]. Signal seg-
ments with too many ectopic beats should be completely ex-
cluded from further analysis.
The frequent occurrence of VPBs in dialysis patients makes
it possible to compute parameters which characterize heart
rate turbulence (HRT). This phenomenon refers to a short-term
fluctuation in heart rate, triggeredbyasingleVPB[66],[67].
Such turbulence is considered to be a blood-pressure-regu-
lating mechanism which, in normal subjects, compensates the
VPB-induced hypotension by an accelerated sinus rate. The
heart rate then decelerates to its baseline level and the blood
pressure returns to its pre-extrasystolic level. The magnitude of
the HRT depends on the preceding heart rate such that higher
heart rates are coupled to lower magnitudes, and vice versa.
Blunted or missing turbulence reflects autonomic dysfunction
and is associated with various conditions. Several studies have
established HRT as one of the most powerful risk predictors of
mortality and sudden cardiac death following acute myocardial
infarction [68].
For hemodialysis patients, there are good reasons to believe
that HRT conveys clinically significant information since auto-
nomic neuropathy is known to be associated with a marked fall
in blood pressure during hemodialysis [69]. To date, only one
study has addressed the issue whether higher propensity to IDH
is reflected in HRT parameters [17]. The results showed that
the acceleration in heart rate that follows a VPB is significantly
lower in patients which are prone to IDH than in patients which
are resistant; both groups exhibited blunted HRT according to
the standard criterion [67]. The significance of the “local” hy-
potension which immediately follows a VPB was explored in
[70]. This study concluded that HRT is physiologically modu-
lated by the duration of the local hypotension. It remains to be
established if a relationship exists between the degree of local
hypotension and the prevalence of IDH.
The parameters conventionally used to characterize HRT,
i.e., turbulence onset and turbulence slope [67], were devel-
oped for use in long-term, ambulatory recordings, based on the
assumption that the poor signal-to-noise ratio (SNR) of single
HRTs could be improved by averaging of the VPB series.
For hemodialysis sessions with a duration of 3–4 h, a much
lower number of VPBs is expected and therefore novel HRT
parameters are needed which perform better at lower SNRs
than do the conventional parameters. The ultimate goal would,
of course, be to analyze single HRTs, without having to resort
to averaging, so that the possible existence of HRT dynamics
can be monitored over time [71].
Model-based detection and characterization of HRT was
recently proposed as a step toward this goal [72]–[74]. The em-
ployed signal model is an extended version of the well-known
integral pulse frequency modulation model which also ac-
counts for the presence of VPBs and HRT. Based on a set
of Karhunen–Loève basis functions which characterize tur-
bulence shape, the generalized likelihood ratio test statistic
was employed for HRT detection. Using a small dataset from
hemodialysis patients, the model-based parameters achieved
better separation between patients being prone and resistent
to IDH than what could be achieved by the conventional pa-
rameters [72]. The model-based HRT parameters, computed
from an average of 10 VPBs, performed similarly to the con-
ventional parameters but computed from an average of 50
VPBs [74]. Interestingly, it was shown in the same study that
the model-based HRT parameters, but not the conventional
parameters, remained predictive of cardiac death in a popula-
tion of patients with ischemic cardiomyopathy and congestive
heart failure when computed from 4-h instead of 24-h ECG
recordings.
Based on 48-h ambulatory recordings or longer, changes in
the conventional HRT parameters were studied before and after,
but not during, hemodialysis [75]. The dataset consisted of 71
patients with end-stage renal disease, but only 31 of these had
VPBs which qualified for HRT analysis. The parameters were
found to be significantly blunted in all patients, but were not
altered by hemodialysis. This latter finding may suggest that
HRT is not a phenomenon which exhibit considerable variation
over time.
D. Baroreflex Sensitivity
Arterial–cardiac baroreceptor reflex sensitivity (BRS) is
usually assessed with the sequences technique which relies on
noninvasive measurements on cardiac activity as well as SAP
[76]–[78]. With this technique, the slope of the regression line
between SAP measurements and RR intervals is computed
in each of the baroreflex sequences, after which the resulting
slopes of all sequences are averaged in order to produce an esti-
mate of BRS. A sequence is delimited in the beat-to-beat series
of SAP measurements and RR intervals whenever both types
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 51
of samples increase or decrease. Impaired BRS is characterized
by smaller slope values.
The assessment of spontaneous BRS in patients prone to IDH
is of particular interest since the baroreflex arc is under auto-
nomic control and regulates the short-term dynamics of blood
pressure. Two recent studies have shed light on the role of BRS
in this group of patients. Using a variant of the sequences tech-
nique, the first study showed that cool dialysate reduces asymp-
tomatic IDH (cf. Section III-B); however, absolute BRS values
did not change significantly by lowering the dialysate tempera-
ture [79]. Increased variability in BRS during cool hemodialysis
was observed which may suggest improved hemodynamic sta-
bility. In that study, it was concluded that early identification
of patients with reduced BRS variability may reduce the preva-
lence of IDH by individualizing the therapy. The other study
investigated the contribution of impaired BRS to the pathophys-
iology of IDH [80]. The major finding was that BRS, measured
at rest immediately prior to the onset of hemodialysis, is ex-
tremely heterogenous and, therefore, no individual patterns of
hemodynamic response could be identified, not even in patients
prone to IDH.
It is well known that the above-mentioned sequences tech-
nique sometimes fails to produce an estimate in patients with
impaired BRS, and dialysis patients are obviously among those
where the technique will fail. To address this serious short-
coming, a novel method was recently developed which employs
adifferentdefinition of baroreflex sequences and which intro-
duces global/total slope estimators, as replacements of the local
slope estimator, in order to ensure more robust estimation [81].
The results showed that the method could always produce a BRS
estimate, also in those cases where baroreflex sequences could
not be identified.
Baroreflex sensitivity can also be assessed at different fre-
quencies, relying on a measure of spectral coherence between
the variability which is present in heart rate and systolic blood
pressure [82]. Using the LF and HF bands, definedinthesame
way as for the HRV analysis mentioned earlier, the BRS was
found to differ significantly in the HF band in patients prone
and resistant to IDH [83]. In contrast to the results reported in
[80], this result suggests that failure of the baroreflex function
is likely to be one of the factors which is responsible for IDH.
IV. PHOTOPLETHYSMOGRAPHIC VARIABLES
The pulse oximeter is an optical technique, based on photo-
plethysmography (PPG), for measuring blood volume changes
in the microvascular bed of the tissue. This device is clinically
attractive since it offers noninvasive, continuous monitoring
relying on existing, low-cost technology. Photoplethysmog-
raphy has gained widespread clinical use because it can provide
information on diverse physiological variables such as arterial
oxygen saturation, respiration, heart rate, vasoconstriction,
blood pressure, and autonomic function [84]. The interaction
of light with arterial blood generates a pulsatile response due to
changes in blood volume with each heartbeat, while the inter-
action with skin, bone, and venous blood is more constant. As
a result, the PPG signal comprises a pulsatile component, syn-
chronized to each heartbeat, which is superimposed on a slowly
Fig. 2. Pulse wave transit time estimated from ECG and PPG signals.
varying baseline related to the average blood volume and tissue
properties. The baseline varies slowly due to the influence of
respiration, sympathetic activity, and thermoregulation.
Different characteristics of the finger PPG signal have been
explored for continuous monitoring of hemodynamic stability
during hemodialysis [85]–[87], though none of these studies
have devised an algorithm for prediction of IDH. Depending on
the underlying hypothesis on hemodynamic information in the
PPG signal, basic pulse wave characteristics such as amplitude,
occurrence time, and area are subjected to analysis and can be
trended for display.
A. IDH and Pulse Transit Time
The pulse wave transit time (PTT) is related to the time re-
quired for transit of the pulse wave to the periphery and can be
used to monitor changes in systolic blood pressure [85], [88].
The pulse transit time can be estimated by the time interval be-
tween the peak location of the R wave, determined from the
ECG signal, and the PPG pulse onset; see Fig. 2. The PTT pro-
vides an indirect estimate of blood pressure changes since arte-
rial compliance is reduced when blood pressure increases which
makes the pulse wave travel faster, and vice versa. The manifes-
tation of hypotension in PTT measurements was simulated by
placing healthy subjects in an airtight box in which a lower body
negative pressure was induced [85].
The measurement of PTT was also central to a study which
evaluated the “harmonized alert sensing technology” (HASTE)
device, developed especially for dialysis monitoring [87],
[89]. Besides PPG and ECG sensors, this device includes a cuff
placed on the arm for arterial pressure whose purpose is to make
intermittent control measurements and to exclude outliers. The
SAP was estimated as being linearly proportional to PTT, with
coefficients adjusted during hemodialysis. Using data from 18
patients, the performance of the HASTE device was evaluated
by comparing the PTT-based estimates of systolic arterial
pressure to invasive SAP taken as [87].
Interestingly, the results reported in the two above-mentioned
PTT-based studies were almost identical, despite the fact that
the investigated datasets were completely different. The cor-
relation between PTT and a cuff-based reference measure-
ment on SAP was 0.66 in [85], whereas it was 0.65 between
52 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
Fig. 3. Arterial and diastolic blood pressure (upper panel) and preprocessed PPG signal (lower panel) during a hemodialysis session with two hypotension episodes
occurring at time instants indicated by vertical, dotted lines.
PTT-based estimates of SAP and invasive SAP [87]. No quan-
titative results were presented on the agreement between mea-
surements in terms of temporal dynamics, and thus no conclu-
sions can be drawn on the methods’ suitability for prediction of
IDH.
While recent results suggest that PTT is unsuitable as a sur-
rogate marker of systolic blood pressure [90], PTT can still be
useful for assessing variability in blood pressure as observed,
e.g., during hemodialysis.
B. IDH and Pulse Wave Amplitude/Morphology
As mentioned in Section III-B, sympathetic activity increases
in response to ultrafiltration and leads to peripheral vasocon-
striction. As a result of vasoconstriction, the core body temper-
ature tends to increase since heat dissipation is impaired. When
combined with the increase in central heat production that ac-
companies dialysis, increased core body temperature can over-
come peripheral vasoconstriction and accelerate the occurrence
of acute hypotension [11]. In the PPG signal, peripheral vaso-
constriction is manifested as a decrease in pulse wave amplitude
when measured in the finger [91].
This pulse wave amplitude was further explored through
the development of a PPG-based method for the prediction of
IDH [92]. The method takes the envelope of the PPG signal as
its starting point since the envelope was considered to reflect
changes in relative blood volume of the capillaries. Using a
sliding window approach, statistical hypothesis testing is per-
formed in each window in order to determine if the amplitude
of the envelope has dropped or not, using a window length of
5 min. Thus, the method does not explore the amplitude of
individual pulse waves but long-term changes in amplitude.
With good accuracy, the noise was found to be characterized by
a Laplacian probability density function. Using leave-one-out
cross validation, the results showed that the method could
predict six out of seven hypotensive events, whereas it only
produced one false prediction out of 17 possible. The mean
time of prediction was found to be 38 min. The method is illus-
trated by the example presented in Fig. 3. While these results
are promising, they were obtained on a rather small dataset
and therefore need to be established on a more comprehensive
dataset before more far-reaching conclusions can be drawn.
The variability of the pulse wave amplitude series has been
investigated with reference to the reduction in blood volume
during hemodialysis [93]. The proposed analysis closely re-
sembles that performed in the frequency domain of HRV, but
with the crucial difference that pulse wave amplitude instead
of rhythm is investigated. As a consequence, the physiological
interpretation of PPG variability differs from that of HRV: the
PPG-LF band mostly reflects sympathetic-related vascular ac-
tivity with minimal direct vagal influence, whereas the PPG-HF
band is governed by the mechanical effect of respiration on
venous return. For fluid overloaded patients, both PPG-LF
power and PPG-HF power were found to increase significantly
due to a progressive reduction in relative blood volume during
hemodialysis, but not so for nonfluid overloaded patients. Early
findings published in the literature showed from the analysis of
intraneural recordings that IDH is related to acute withdrawal
of sympathetic vasoconstrictor drive [22]. With the PPG-based
method described in [93], changes in peripheral autonomic
control can be monitored and, possibly, detect this type of
sympathetic withdrawal.
Monitoring of pulse wave morphology in the PPG signal has
been proposed as a means to quantifying intermittent hemody-
namic instability [86]. In that study, the authors introduced a
“reflectiveindex,”defined as an average of the samples that
make up the dicrotic notch of the diastolic component of each
pulse wave. An increase in this index is associated with in-
creased peripheral pulse wave reflection due to, e.g., vasocon-
striction. In 15 out of 20 patients with end-stage renal failure,
the index was found to increase during hemodialysis, suggesting
increased systemic vasoconstriction. While the reflective index
was continuously trended during the session, its temporal dy-
namics was not explored for various purposes such as the pre-
diction of IDH; only the change in index value from onset to
end of the hemodialysis session was studied in statistical terms.
When the dicrotic notch is absent, which is often the case in
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 53
patients prone to hypotension, the reflective index is no longer
defined, and thus the resulting trend will contain gaps.
C. IDH and Oxygen Saturation
Short-term variability of oxygen saturation has been pro-
posed as a “warning parameter” of IDH [94]. Although
measurements of oxygen saturation were acquired from blood
entering the dialyzer, similar information is provided by a PPG
finger sensor since both types of measurements are made on
arterial blood. Short-term variability was quantified by the
standard deviation of the samples in a sliding window of 4-min
length. It was hypothesized that increased variability precedes
hypotension and is a consequence of changes in cardiac output
and tissue perfusion. The results were promising since 17 out
of 20 treatments with hypotension were correctly predicted,
and no prediction was made in 18 out of 20 treatments without
hypotension. The mean time of prediction was 14 min. Un-
fortunately, the significance of these results cannot be easily
assessed since information on the annotated occurrence time of
hypotension was made use of when determining the estimated
occurrence time. Hence, it is difficult to compare these results
with those presented in [92], although this would be highly
interesting since the two methods are the only ones published
to date which address the problem of real-time IDH prediction.
D. Hypovolemia in Other Applications
The withdrawal of blood volume can be monitored using
the PPG signal for the purpose of detecting progressive hypo-
volemia. The significance of such monitoring has been thor-
oughly investigated within the contexts of intensive care and
anesthesia, but not so for hemodialysis despite the fact that with-
drawal of blood volume is an important cause of IDH. In the
following, recent advances in methodology for detecting hypo-
volemia are brieflyreviewed.
Several studies have established that respiratory-induced am-
plitude changes in the PPG signal can serve as a parameter for
detecting progressive hypovolemia; see, e.g., [95] and [96] for
some early results. The changes are commonly quantified by
first determining the pulse wave amplitude, i.e., the difference
in amplitude between the peak and the preceding trough, and the
respiratory amplitude, i.e., the maximal and minimal pulse wave
amplitudes within a respiratory cycle. A “hypovolemic” param-
eter can be defined as the difference between these two ampli-
tudes and normalized by their mean value. Since this parameter
requires peak-picking, which is known to be vulnerable to noise,
recent efforts have aimed at robustifying the peak-picking pro-
cedure [97], [98]. The significance of respiration-induced vari-
ability in the PPG signal was evaluated in 33 anesthesized pa-
tients by withdrawing blood in steps of 2% of estimated circu-
lating blood volume, up to 20% [99]. Using the hypovolemic
parameter, it was found that the PPG signal is useful for de-
tecting hypovolemia for reductions in blood volume of 6% or
more.
Similar results were obtained from 18 healthy subjects who
underwent a progressive reduction in central blood volume in-
duced by lower body negative pressure [100]. In that study, the
amplitude, width, and area of each individual pulse wave were
analyzed and averaged over minute-long segments. Changes in
these features were found to be strongly correlated with pro-
gressive reduction in stroke volume during lower body negative
pressure; subsequent restoration of the central blood volume
was reflected by a return of the parameter values to their respec-
tive baseline levels. It was concluded that PPG analysis is suit-
able for detection of clinically significant hypovolemia before
the onset of cardiovascular decompensation in spontaneously
breathing patients.
Rather than relying on time domain features of individual
pulse waves for detecting hypovolemia, more detailed informa-
tion can be obtained by spectrally decomposing the PPG signal
into different components which, e.g., reflect respiratory ac-
tivity. A recent method explores modulation of the PPG signal
by heart rate and respiratory rate [101]. Such modulation is con-
sidered to be related to changes in arterial and venous pulsations.
An advanced demodulation technique was developed for multi-
component narrowband signals whose center frequencies can be
time varying [102]. At every time instant, the largest value of the
respiration spectral band is determined, forming a sample of the
signal which represents the time-varying amplitude modulation
due to respiration; the heart rate related modulation signal is de-
termined in an analogous way. From the PPG signals recorded
from 11 healthy subjects exposed to lower body negative pres-
sure, it was found that a reduction in amplitude modulation of
the respiration rate, as well as in amplitude modulation of the
heart rate, serve as markers for early detection of blood volume
reduction.
E. Cardiac Rhythm
Although it is well known that heart rate can be accurately es-
timated from the PPG signal [103], it is only until recently that
pulse rate variability, determined from the PPG signal, has been
investigated as a possible surrogate for HRV [104], [105]. When
analyzing data from healthy subjects, it has been shown that nei-
ther time domain nor frequency domain parameters differ sig-
nificantly when determined from the ECG or the PPG. However,
this finding remains to be established for different groups of pa-
tients before the PPG can be employed as a definitive surrogate
signal. An important difference between ECG- and PPG-based
signal analysis is that the latter is confounded by the influence
of PTT which fluctuates on a beat-to-beat basis. Although the
fluctuations introduce an error component to the analysis, its in-
fluence appears to be quite negligible, both for HRV determined
at rest and during nonstationary conditions. Since PTT is related
to blood pressure, its influence on the HRV spectrum is largely
confinedtotheLFregion.
The analysis of HRT has also been performed on rhythm in-
formation derived from the PPG signal, the outcome suggesting
that this may be a feasible approach [106]. To make such anal-
ysis meaningful, it is crucial that VPBs can be determined from
the pulse wave characteristics. This requirement applies to the
analysis of pulse rate variabilityaswellastothesituationwhere
the occurrence of VPBs must be dealt with.
F. Signal Quality
To ensure that PPG-based methods are robust enough for clin-
ical use, the influence of motion artifacts caused by the patient
moving his/her arm to communicate, drink, or eat is essential to
54 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
minimize. For example, patient movement can compromise the
accuracy of the analysis based on the PPG sensor in the non-fis-
tula arm [87]. Over the years, considerable research has been
directed toward reducing the influence of motion artifacts on
PPG measurements: see, e.g., [107] and [108]. With more so-
phisticated PPG applications being developed, it has become
increasingly important to devise algorithms which improve the
SNR. At the same time, it is essential that such algorithms do
not remove those frequency components which are constituent,
e.g., to PPG variability, and, therefore, their design should be
evaluated in terms which are relevant for the target analysis.
A broad range of signal processing techniques have recently
been proposed for improving the SNR. While a review of those
techniques is outside the scope of this paper, some interesting
studies are still listed here. Accelerometers were used to detect
the presence of motion artifacts so that an adaptive filter could
be employed to reduce the influence of such artifacts [109].
Based on a model derived from photon diffusion analysis, a
method was proposed for motion-resistant measurement of
blood oxygen saturation [110]. Statistical model-based ap-
proaches to signal estimation in noise have been pursued which
rely on basis functions including sinusoids [111], [112], see also
[113]. Such an approach was, for example, found to provide
accurate monitoring even in signals with abrupt changes in
heart rate or respiration rate [112].
V. FEEDBACK CONTROL IN HEMODIALYSIS
Traditionally, hemodialysis has been performed with pa-
rameter settings of the dialysis machine which remain fixed
throughout the session. As more information on the patient’s
clinical history unfolds, suitable changes in the parameter
settings can be introduced at the onset of subsequent sessions
in order to adjust the treatment to the patient’s need. For ex-
ample, the patient’s history of hypotension is a factor which
should exert influence on how to choose a suitable UFR profile.
Nonetheless, the use of session-fixed parameters stands out as
rather crude when contrasted with the body’s continuous regu-
lation of the normal kidney which ensures that a stable internal
environment can be maintained (homeostasis). Consequently, it
is not surprising that recent research interest has been directed
toward the design and implementation of feedback control1in
the hemodialysis machine in order to prevent hemodynamic in-
stability [115], [116]. Feedback control represents an important
step toward individualized treatment and should, ideally, lead
to improved efficacy of hemodialysis and reduced frequency
of IDH, especially when combined with information on patient
history.
Feedback control is today implemented in clinical systems
whose aim is to control one target variable, reflecting either
blood temperature or relative blood volume.2When blood tem-
1In the literature, feedback control in hemodialysis has, somewhat mis-
leading, been referred to as “biofeedback” though it is widely accepted that
this terms refers to a process that enables an individual to learn how to change
physiological activity for the purposes of improving health and performance;
see, e.g., [114].
2Arterial blood pressure has also been treated as a target variable for control,
together with measurements on the arterial blood pressure itself as sensor infor-
mation [14]. The overall purpose of the system was to prevent IDH. However,
no information was disclosed on the employed sensor principle.
perature is subject to control, the dialysate temperature is con-
tinuously adjusted to ensure that the required amount of thermal
energy is removed by cooling of the extracorporeal circulation
system [117]. The controller performs this adjustment based on
the information provided by different sensors which measure ar-
terial and venous line blood temperatures as well as blood flow.
When, instead, relative blood volume is subject to control,
UFR and DSC serve as actuators for stabilizing the hemody-
namic response during hemodialysis. These two actuators are
known to have a major influence on the volume of blood circu-
lating in the body. Information on the status of relative blood
volume is provided by sensors which measure different types of
blood constituent such as hemoglobin, hematocrit, or the con-
centration of total plasma proteins; see Section III-C.
Javed et al. [118] have just recently published an excellent
review of feedback control of hemodynamic variables during
hemodialysis. The reader is referred to that review for a more
detailed perspective on feedback techniques than offered as
follows.
A. Fuzzy Control
Since the relationship between process (output) and actuator
(input) variables is nonlinear, time varying, and patient de-
pendent, the design of a controller is far from straightforward.
Some early attempts explored the use of the proportional
integral derivative (PID) controller in the hemodialysis context
[29]; however, the use of fuzzy logic has since long become the
predominant principle for controller design since it can have
nonlinear characteristics [119], [120]. The knowledge from
designing a PID controller can nonetheless serve as a starting
point since a linear fuzzy controller can be designed to have
performance identical to a PID controller. The resulting linear
fuzzy controller can then be made nonlinear through a trial
and error procedure so that more complicated input–output
relationships can be handled.
Under stable hemodynamic conditions, as reflected by normal
values of the process variables, UFR and DSC are adjusted to its
maximum and minimum value, respectively. When the relative
blood volume is decreasing, possibly reflecting an approaching
IDH, the fuzzy rules should ensure that UFR is decreased and
DSC increased in relation to the deviation of the process vari-
ables from their normal values [119], [120]. In a fuzzy con-
troller, these relations are translated into a set of if–then state-
ments, which thus may include many input and output variables
and logical operators, in order to mimic heuristic reasoning.
An alternative approach is to formulate the if–then statements
of the fuzzy controller based on the analysis of different sce-
narios which can be simulated with a physiological model of
IDH. For example, a model was developed for simulation of sys-
temic arterial pressure, heart rate, total systemic resistance, and
cardiac output during hemodialysis [121]; see also [122]. A set
of differential equations defines the dynamics of the cardiovas-
cular system, the solute and water kinetics, and the exchange of
water between body compartments. This simulation model has
served as the basis for the design of a fuzzy controller which
adjusts the DSC and UFR profiles based on the prescribed de-
crease in body weight and sodium content [15]; the controller
performance regarding reduced frequency of IDH is discussed
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 55
Fig. 4. (a) Ultrafiltration rate and (b) dialysate sodium concentration subjected
to feedback control for an asymptomatic hypotensive patient. Time instant
at which systolic blood pressure dropped to the lower constraint is circled.
(Reprinted from [123] with permission.)
as follows. Unfortunately, the technical details of this controller,
as well as other fuzzy controllers employed in clinical studies,
are poorly disclosed since their design is proprietary and imple-
mented in commercial products, thereby limiting their signifi-
cance for the engineering research community.
B. Model-Based Control
Model-based control was recently investigated for the pur-
pose of maintaining hemodynamic stability during hemodial-
ysis and represents a novel, promising approach to feedback
control in hemodialysis [123]; see also the precursor study
in [124]. The model, assumed to be linear and time varying,
embraces variables similar to the ones of fuzzy control, i.e.,
UFR and DSC as actuator variables, and relative blood volume,
systolic blood pressure, and change in heart rate as process
variables. The matrices of the system state equation are defined
by several time-varying parameters, identified in individual
patients using a multistage estimation technique. Hemodial-
ysis sessions profiled with respect to UFR and DSC, either
being constant, linearly decreasing, step decreasing, or square
changing, are crucial to the identification procedure as the
model response of an individual patient can then be fine-tuned.
The average goodness of fit between the measured and the es-
timated process variables, expressed in terms of , was found
to be approximately 0.7 for all variables. Changes in heart rate
were more difficult to model which, according to the authors,
could be explained by the presence of HRV. However, it can
be questioned whether heart rate should be modeled at all con-
sidering that a change in heart rate does not differ significantly
between IDH prone and resistant patients, cf. Section IV-A.
The controller design aims at regulating relative blood
volume and heart rate by varying UFR and DSC while, at
the same time, maintaining a stable systolic blood pressure.
Model-based predictive controller design was explored for
this purpose as it can, among other things, easily account
for constraints which must be fulfilled so that the controlled
variables remain within safety margins. The behavior of this
type of controller is illustrated in Fig. 4.
Preliminary results were obtained from a dataset with 12
patients who underwent profiled hemodialysis treatment. The
results suggest that the proposed feedback control system can
maintain the systolic blood pressure within certain bounds
and avoid the occurrence of sudden changes in relative blood
volume and heart rate. In contrast to fuzzy controllers, the
model-based controller has the scientific advantage of being
described in considerable detail. It remains to be established
whether this approach to feedback control can reduce the
frequency of IDH.
C. Clinical Results
The use of feedback control based on relative blood volume
has in several studies been found to reduce the frequency of
IDH—a result which may be attributed to the fact that the feed-
back control helps to avoid rapid fluctuations in relative blood
volume [25], [29], [125]–[127]. In an early study, 12 hypoten-
sion-prone patients were treated with both standard (manual)
UFR control and feedback control of UFR and DSC [128]. The
latter type of control was associated with considerably fewer
hypotensive episodes: the number of sessions with hypotensive
episodes dropped from 59 to 24 out of a total of 72 sessions. In a
more recent study, involving 26 hypotension-prone patients, the
percentage of sessions with symptomatic IDH was found to de-
crease from 32% in standard hemodialysis to 24% when using
feedback control of UFR [129]. A similar reduction ratio was
reported in [14], where the percentage of sessions with severe
IDH decreased from 13.8% to 8.3%.
Among the most impressive results presented to date are
those of a multicenter study which involved 55 patients, all
having had IDH during at least one session per week in the
6 months preceding hemodialysis treatment [15]. Employing
the above-mentioned model-based fuzzy controller, labelled
“automatic adaptive system dialysis (AASD),” more stable
intradialytic blood pressure and lower heart rate were found
when compared to standard treatment. The number of sessions
complicated by hypotension decreased from 58.7% to as low
as 0.9%. While differences in results can be explained to
some extent by different datasets, the methodology adopted
for feedback control of UFR and DSC is likely to be the main
explanation.
The outcome of clinical studies may be partially biased by
a patient’s knowledge on whether hypotension-preventive mea-
sure are taken or not during the hemodialysis session. It is well
known that such knowledge can, by itself, act to reduce the fre-
quency of IDH.
Despite the fact that feedback control in hemodialysis has
produced promising results in terms of reducing the frequency
of IDH, not all nephrologists are equally convinced about the
overall value of this technology. The use of feedback control
may be time consuming for the clinical staff and may therefore
represent a barrier to widespread deployment. Furthermore, it
is felt that the technological advances in hemodialysis treatment
have yet to be translated into longer patientsurvival [12]; a sim-
ilar view has been expressed in [7] and [130]. It is, however,
essential to contrast such pessimistic views with the fact that
patients who undergo hemodialysis treatment today are much
older and have more serious medical problems than had patients
who were treated some 20 years ago.
56 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
VI. FUTURE PERSPECTIVES
Since cardiovascular problems remain the most common in-
tradialytic complication [131], it is remarkable that so scarce
information on cardiovascular status is provided to the clinical
staff during hemodialysis. This lack can partially be explained
by the need to use additional sensors for recording such activity,
which may cause discomfort to the patient. Moreover, it is clin-
ically undesirable to be forced to spend time on attaching ECG
electrodes and cables for every session. The finger-based pulse
oximeter represents a more comfortable sensor alternative and
is resistant to electrical interference and loosening of electrodes.
Ultimately, however, the clinical goal would be to develop tech-
niques which make it possible to extract cardiac information
without having to introduce additional sensors.
An interesting concept, recently proposed, is the “patient de-
teriorationindex”whichisdesignedtoreflect the risk ofapa-
tient to develop IDH [132]. This index is inspired by previous
work on a patient monitoring system which involved data fu-
sion [133], [134]. Based on a probabilistic model of normality,
the main idea is to identify abnormalities in hypotension-related
variables of a patient, where the statistics characterizing nor-
mality are first learned from data collected from a representative
group of IDH-resistant patients. Prediction of IDH is then ac-
complished in real-time by searching for substantial deviations
from the model. While this approach does not replace feedback
control, it can provide the clinical staff with valuable informa-
tion on patient status as well as to enrich the information which
serves as input to the feedback controller.
Feedback systems represent a great leap forward in hemodial-
ysis technology and offer the potential to improve the treatment
of patients with end stage renal disease. To date, their imple-
mentation has largely been confined to the analysis of “wet”
variables such as relative blood volume, UFR, and DSC. Be-
sides requiring additional sensors, a technical reason to why in-
formation on cardiac activity has not been much explored for
feedback control may be that variables reflecting HRV and HRT
are irregularly sampled in time. Indeed, certain cardiac variables
are represented by just a single value for the entire hemodialysis
session, thus serving as an indicator of the patient’s proneness to
IDH for the entire session. Consequently, research is needed on
how to design a feedback controller which not only processes
dynamic, regularly sampled data, but also sparsely and irregu-
larly sampled data so as to improve the overall interpretation of
the data.
REFERENCES
[1] J. T. Daugirdas, “Pathophysiology of dialysis hypotension: An update,”
Amer.J.KidneyDis., vol. 38, pp. S11–17, 2001.
[2] M. J. Schreiber Jr., “Clinical dilemmas in dialysis: Managing the hy-
potensive patient. Setting the stage,” Amer.J.KidneyDis., vol. 38, pp.
S1–S10, 2001.
[3] A. Tislér, K. Akócsi, B. Borbás, L. Fazakas, S. Ferenczi, S. Görögh, I.
Kulcsár,L.Nagy,J.Sámik,J.Szegedi,E.Tóth,G.Wágner,andI.Kiss,
“The effect of frequent or occasional dialysis-associated hypotension
on survival of patients on maintenance haemodialysis,” Nephrol. Dial.
Transplant., vol. 18, pp. 2601–2605, 2003.
[4] T. Shoji, Y. Tsubakihara, M. Fujii, and E. Imai, “Hemodialysis-associ-
ated hypotension as an independent risk factor for two-year mortality
in hemodialysis patients,” Kidney Int., vol. 66, pp. 1212–1220, 2004.
[5] B. Charra, “Hypertension/hypotension in dialysis,” Kidney Int.,vol.55,
pp. 1128–1130, 1999.
[6] N.M.SelbyandC.W.McIntyre,“The acute cardiac effects of dial-
ysis,” Semin. Dial., vol. 20, pp. 220–228, 2007.
[7] A. Davenport, “Can advances i n hemodialysis machine technology pre-
vent intradialytic hypotension?,” Semin. Dial., vol. 22, pp. 231–236,
2009.
[8] R. A. Sherman, J. T. Daugirdas, and T. S. Ing, “Complications during
hemodialysis,” in Handbook of Dialysis,J.T.Daugirdas,P.G.Blake,
and T. S. Ing, Eds., 4th ed. Philadelphia, PA: Lippincott, Williams &
Wilkins, 2007, ch. 10, pp. 170–191.
[9] K. Shimizu, T. Kurosawa, and T. Sanjo, “Effect of hyperosmolality
on vasopressin secretion in intradialytic hypotension: A mechanistic
study,” Amer. J. Kidney Dis., vol. 52, pp. 294–304, 2008.
[10] F. Locatelli, U. Buoncristiani, B. Canaud, H. Köhler, T. Petitclerc, and
P. Zucchelli, “Special feature haemodialysis with on-line monitoring
equipment: Tools or toys?,” Nephrol. Dial. Transplant., vol. 20, pp.
22–33, 2005.
[11] B. F. Palmer and W. L. Henrich, “Recent advances in the prevention
and management of intradialytic hypotension,” J. Amer. Soc. Nephrol.,
vol. 19, pp. 8–11, 2008.
[12] N. Lameire, W. Van Biesen, and R. Vanholder, “Did 20 years of tech-
nological innovations in hemodialysis contribute to better patient out-
comes?,” Clin. J. Amer. Soc. Nephrol., vol. 4, pp. S30–40, 2009.
[13] W. L. Henrich, “Intradialytic hypotension: A new insight to an old
problem,” Amer.J.KidneyDis., vol. 52, pp. 209–210, 2008.
[14] E. Mancini, E. Mambelli, M. Irpinia, D. Gabrielli, C. Cascone, F.
Conte, G. Meneghel, F. Cavatorta, A. Antonelli, G. Villa, A. Dal
Canton, L. Cagnoli, F. Aucella, F. Fiorini, E. Gaggiotti, G. Triolo, V.
Nuzzo, and A. Santoro, “Prevention of dialysis hypotension episodes
using fuzzy logic control system,” Nephrol. Dial. Transplant., vol. 22,
pp. 1420–1427, 2007.
[15] L. Colì et al., “Automatic adaptive system dialysis for hemodialysis-
associated hypotension and intolerance: A noncontrolled multicenter
trial,” Amer. J. Kidney Dis., vol. 58, pp. 93–100, 2011.
[16] A. Bayya, D. Rubinger, D. M. Linton, and S. Sviri, “Evaluation of
intradialytic hypotension using impedance cardiography,” Int. Urol.
Nephrol., vol. 43, pp. 855–864, 2010.
[17] K. Solem, A. Nilsson, and L. Sörnmo, “An ECG-based method for
early detection of abrupt changes in blood pressure during hemodial-
ysis,” ASAIO J., vol. 52, pp. 282–290, 2006.
[18] C. Ronco, A. Fabris, S. Chiaramonte, E. De Dominicis, M. Feriani, A.
Brendolan, L. Bragantini, M. Milan, R. Dell’Aquila, and G. La Greca,
“Comparison of four different short dialysis techniques,” Int. J. Artif.
Organs., vol. 11, pp. 169–174, 1988.
[19] Q. Maggiore, F. Pizzarelli, S. Sisca, C. Zocalli, S. Parlongo, F. Ni-
colo, and G. Creazzo, “Blood temperature and vascular stability during
hemodialysis and hemofiltration,” ASAIO Trans., vol. 28, pp. 523–527,
1982.
[20] N. M. Selby and C. W. McIntyre, “A systematic review of the clinical
effects of reducing dialysate fluid temperature,” Nephrol. Dial. Trans-
plant., vol. 21, pp. 1883–1898, 2006.
[21] F. M. van der Sande, G. Wystrychowski, J. P. Kooman, L. Rosales, J.
Raimann, P. Kotanko, M. Carter, C. T. Chan, K. M. Leunissen, and N.
W. Levin, “Control of core temperature and blood pressure stability
during hemodialysis,” Clin. J. Amer. Soc. Nephrol., vol. 4, pp. 93–98,
2009.
[22] R. L. Converse Jr., T. N. Jacobsen, C. M. Jost, R. D. Toto,
P. A. Grayburn, T. M. Obregon, F. Fouad-Tarazi, and R. G. Victor,
“Paradoxical withdrawal of reflex vasoconstriction as a cause of
hemodialysis-induced hypotension,” J. Clin. Invest., vol. 90, pp.
1657–1665, 1992.
[23] F. M. van der Sande, J. P. Kooman, and K. M. Leunissen, “Intradia-
lytic hypotension—New concepts on an old problem,” Nephrol. Dial.
Transplant., vol. 15, pp. 1746–1748, 2000.
[24] J. J. Dasselaar, R. M. Huisman, P. E. de Jong, and C. F. Franssen,
“Relative blood volume measurements during hemodialysis: Compar-
isons between three noninvasive devices,” Hemodial. Int., vol. 11, pp.
448–455, 2007.
[25] J. J. Dasselaar, R. M. Huisman, P. E. de Jong, and C. F. Franssen,
“Measurement of relative blood volume changes during haemodial-
ysis: Merits and limitations,” Nephrol. Dial. Transplant., vol. 20, pp.
2043–2049, 2005.
[26] J. Beige, J. Sone, A. M. Sharma, M. Rudwaleit, G. Offermann, A. Dis-
tler, and L. Preuschof, “Computational analysis of bloodvolume curves
and risk of intradialytic morbid events in hemodialysis,” Kidney Int.,
vol. 58, pp. 1805–1809, 2000.
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 57
[27] S. Andrulli, S. Colzani, F. Mascia, L. Lucchi, L. Stipo, M. C. Bigi,
M. Crepaldi, B. Redaelli, A. Albertazzi, and F. Locatelli, “The role of
blood volume reduction in the genesis of intradialytic hypotension,”
Amer.J.KidneyDis., vol. 40, pp. 1244–1254, 2002.
[28] S. Mitra, P. Chamney, R. Greenwood, and K. Farrington, “Linear decay
of relative blood volume during ultrafiltration predicts hemodynamic
instability,” Amer.J.KidneyDis., vol. 40, pp. 556–565, 2002.
[29] A. Santoro, E. Mancini, F. Paolini, G. Cavicchioli, A. Bosetto, and P.
Zucchelli, “Blood volume regulation during hemodialysis,” Amer. J.
Kidney Dis., vol. 32, pp. 739–748, 1998.
[30] K. Moret, J. Aalten, W. van den Wall Bake, P. Gerlag, C. Beeren-
hout, F. van der Sande, K. Leunissen, and J. Kooman, “The effect of
sodium profiling and feedback technologies on plasma conductivity
and ionic mass balance: A study in hypotension-prone dialysis pa-
tients,” Nephrol. Dial. Transplant., vol. 21, pp. 138–144, 2006.
[31] N. M. Selby, M. W. Taal, and C. W. McIntyre, “Comparison of pro-
gressive conductivity reduction with diacontrol and standard dialysis,”
ASAIO J., vol. 53, pp. 194–200, 2007.
[32] S.GrimnesandØ.G.Martinsen, Bioimpedance and Bioelectricity Ba-
sics. London, U.K.: Academic, 2000.
[33] U. G. Kyle, I. Bosaeus, A. D. De Lorenzo, P. Deurenberg, M. Elia, J.
Manuel Gómez, B. Lilienthal Heitmann, L. Kent-Smith, J. C. M elchior,
M. Pirlich, H. Scharfetter, A. Schols, and C. Pichard, “Bioelectrical
impedance analysis part I: Review of principles and methods,” Clin.
Nutr., vol. 23, pp. 1226–1243, 2004.
[34] K. F. Foster and H. C. Lukaski, “Whole-body impedance—What does
it measure?,” Amer.J.Clin.Nutr., vol. 64, pp. 388S–396S, 1996.
[35] P. Kotanko, N. W. Levin, and F. Zhu, “Current state of bioimpedance
technologies in dialysis,” Nephrol. Dial. Transplant., vol. 23, pp.
808–812, 2008.
[36] J. Raimann, L. Liu, S. Tyagi, N. W. Levin, and P. Kotanko, “A fresh
look at dry weight,” Hemodial. Int., vol. 12, pp. 395–405, 2008.
[37] S. Mandolfo, M. Farina, and E. Imbasciati, “Bioelectrical impedance
and hemodialysis,” Int. J. Artif. Organs, vol. 18, pp. 700–704, 1995.
[38] A. Piccoli, “Bioelectric impedance vector distribution in peritoneal
dialysis patients with different hydration status,” Kidney Int., vol. 65,
pp. 1050–1063, 2004.
[39] A. Piccoli, “Bioelectric impedance measurement for fluid status as-
sessment,” in Fluid Overload: Diagnosis and Management (Contrib.
Nephrol.), C. Ronco, M. R. Costanzo, R. Bellomo, and A. S. Maisel,
Eds. Basel: Karger, 2010, vol. 164, pp. 143–152.
[40] M. J. Koziolek, S. G auczinski, E. Kahler, C. P. Bramlage, A. K. Scheel,
G. A. Mueller, and F. Strutz, “Bioimpedance analysis and intradialytic
hypotension in intermittent hemodialysis,” Clin. Nephrol., vol. 66, pp.
39–50, 2006.
[41] J. Booth, J. Pinney, and A. Davenport, “The effect of vascular access
modality on changes in fluid content in the arms as determined by mul-
tifrequency bioimpedance,” Nephrol. Dial. Transplant., vol. 26, pp.
227–231, 2011.
[42] J. Malmivuo and R. Plonsey, Bioelectromagnetism. Oxford, U.K.:
Oxford Univ. Press, 1995.
[43] J. L. Wynne, L. O. Ovadje, C. M. Akridge, S. W. Sheppard, R. L. Vogel,
and J. M. Van De Water,“Impedance cardiography: A pote ntial monitor
for hemodialysis,” J. Surg. Res., vol. 133, pp. 55–60, 2006.
[44]B.H.Greenberg,D.D.Hermann,M.F.Pranulis,L.Lazio,andD.
Cloutier, “Reproducibility of impedance cardiography hemodynamic
measures in clinically stable heart failure patients,” Congest. Heart
Fail., vol. 6, pp. 19–26, 2000.
[45] J. M. Van De Water, T. W. Miller, R. L. Vogel, B. E. Mount, and M. L.
Dalton, “Impedance cardiography. The next vital sign technology?,”
Chest, vol. 123, pp. 2028–2033, 2003.
[46] J.Cordtz,B.Olde,K.Solem,and S. D. Ladefoged, “Central venous
oxygen saturation and thoracic admittance during dialysis: New ap-
proaches to hemodynamic monitoring,” Hemodial. Int., vol. 12, pp.
369–377, 2008.
[47] “Patient mortality and survival,” Amer. J. Kidney Dis., vol. 28, pp.
79–92, 1996.
[48] C. W. McIntyre, “Effects of hemodialysis on cardiac function,” Kidney
Int., vol. 76, pp. 371–375, 2009.
[49] J. O. Burton, S. Korsheed, B. J. Grundy, and C. W. McIntyre,
“Hemodialysis-induced left ventricular dysfunction is associated with
an increase in ventricular arrhythmia,” Renal Failure, vol. 30, pp.
701–709, 2008.
[50] S. Cavalcanti, S. Severi, and G. Enzmann, “Analysis of oscillatory
components of short-term heart rate variability in hemodynamically
stable and unstable patients during hemodialysis,” Artif. Organs, vol.
22, pp. 98–106, 1998.
[51] M.G.W.Barnas,W.H.Boer,andH.A.Koomans,“Hemodynamic
patterns and spectral analysis of heart rate variability during dialysis
hypotension,” J. Amer. Soc. Nephrol., vol. 10, pp. 2577–2584, 1999.
[52] G. Pelosi, M. Emdin, C. Carpeggiani, M. A. Morales, M. Piacenti, P.
Dattolo, T. Cerrai, A. Macerata, A. L’Abbate, and Q. Maggiore, “Im-
paired sympathetic response before intradialytic hypotension: A study
based on spectral analysis of heart rate and pressure variability,” Clin.
Sci. (Lond), vol. 96, pp. 23–31, 1999.
[53] D. Rubinger, N. Revis, A. Pollak, M. H. Luria, and D. Sapoznikov,
“Predictors of haemodynamic instability and heart rate variability
during haemodialysis,” Nephrol. Dial. Transplant., vol. 19, pp.
2053–2060, 2004.
[54] M. Malik and A. J. Camm, Eds., Dynamic Electrocardiography New
York, Wiley-Blackwell, 2004.
[55] M. Pagani, V. Somers, R. Furlan, S. Dell’Orto, J. Conway, G. Baselli,
S. Cerutti, P. Sleight, and A. Malliani, “Changes in autonomic regula-
tion induced by physical training in mild hypertension,” Hypertension,
vol. 12, pp. 600–610, 1988.
[56] D. H oyer, K. Schmidt, R. Bauer, U. Zwiener, M. Köhler, B. Lüthke, and
M. Eiselt, “Nonlinear analysis of heart rate and respiratory dynamics,”
IEEE Eng. Med. Biol. Mag., vol. 16, pp. 31–39, 1997.
[57] S. J. Richman and J. R. Moorman, “Physiological time-series analysis
using approximate entropy and sample entropy,” Amer.J.Physiol., vol.
278, pp. H2039–H2049, 2000.
[58] M. Costa, A. L. Goldberger, and C. K. Peng, “Multiscale entropy anal-
ysis of complex physiologic time series,” Phys. Rev. Lett., vol. 89, p.
068102, 2002.
[59] N. Yamanaka, T. Aoyama, N. Ikeda,M.Higashihara,andK.Kamata,
“Characteristics of heart rate variability entropy and blood pres-
sure during hemodialysis in patients with end-stage renal disease,”
Hemodial. Int., vol. 9, pp. 303–308, 2005.
[60] O. M. Shapira and Y. Bar-Khayim, “ECG changes and cardiac arrhyth-
mias in chronic renal failure patients on hemodialysis,” J. Electrocar-
diol., vol. 25, pp. 273–279, 1992.
[61] S. Abe, M. Yoshizawa, N. Nakanishi, T. Yazawa, K. Yokota, M. Honda,
and G. Sloman, “Electrocardiographic abnormalities in patients re-
ceiving hemodialysis,” Amer. Heart J., vol. 131, pp. 1137–1144, 1996.
[62] A. Santoro, E. Mancini, R. Gaggi, S. Cavalcanti, S. Severi, L.
Cagnoli, F. Badiali, B. Perrone, G. London, H. Fessy, L. Mercadal,
and F. Grandi, “Electrophysiological response to dialysis: The role
of dialysate potassium content and profiling,” Contrib. Nephrol., vol.
149, pp. 295–305, 2005.
[63] M. V. Kamath and E. L. Fallen, “Correction of the heart rate variability
signal for ectopics and missing beats,” in Heart Rate Variability,M.
Malik and A. J. Camm, Eds. New York: Futura Publ., 1995, ch. 6,
pp. 75–86.
[64] K. Solem, P. Laguna, and L. Sörnmo, “An efficient method for handling
ectopic beats using the heart timing signal,” IEEE Trans. Biomed. Eng.,
vol. 53, no. 1, pp. 13–20, Jan. 2006.
[65] L. Sörnmo and P. Laguna, Bioelectrical Signal Processing in Cardiac
and Neurological Applications. Amsterdam, The Netherlands: Else-
vier , 2005.
[66] G. Schmidt, M. Malik, P. Barthel, R. Schneider, K. Ulm, L. Rolnitzky,
A. J. Camm, J. T. Bigger, Jr, and A. Schomig, “Heart-rate turbulence
after ventricular premature beats as a predictor of mortality after acute
myocardial infarction,” Lancet, vol. 353, pp. 1390–1396, 1999.
[67] A.Bauer, M.Malik,G.Schmidt,P.Barthel,H.Bonnemeier,I.Cy-
gankiewicz, P. Guzik, F. Lombardi, A. Müller, A. Oto, R. Schneider,
M. Watanabe, D. Wichterle, and W. Zareba, “Heart rate turbulence:
Standards of measurement, physiological interpretation, and clinical
use: International Society for Holter and Noninvasive Electrophysi-
ology Consensus,” J. Amer. Coll. Cardiol., vol. 52, pp. 1353–1365,
2008.
[68] M. Watanabe and G. Schmidt, “Heart rate turbulence: A 5-year re-
view,” Heart Rhythm, vol. 1, pp. 732–738, 2004.
[69] C. Calvo, S. Maule, F. Mecca, R. Quadri, G. Martina, and P. Cavallo
Perin, “The influence of autonomic neuropathy on hypotension during
hemodialysis,” Clin.Auton.Res., vol. 12, pp. 84–87, 2002.
[70] S. Raj, R. Sheldon, M. Koshman, and D. Roach, “Role of hypotension
in turbulence physiology,” Heart Rhythm, vol. 2, pp. 710–717, 2005.
[71] J. L. RojoÁlvarez, Ó. Barquero Pérez, I. M. Jiménez, E. Everss, A. B.
Rodríguez González, and A. García Alberola, “Heart rate turbulence
denoising using support vector machines,” IEEE Trans. Biomed. Eng.,
vol. 56, no. 2, pp. 310–319, Feb. 2009.
[72] K. Solem, P. Laguna, J. P. Martínez, and L. Sörnmo, “Model-based
detection of heart rate turbulence,” IEEE Trans. Biomed. Eng., vol. 55,
no. 10, pp. 2712–2722, Oct. 2008.
58 IEEE REVIEWS IN BIOMEDICAL ENGINEERING, VOL. 5, 2012
[73] D. Smith, K. Solem, P. Laguna, J. P. Martínez, and L. Sörnmo, “Model-
based detection of heart rate turbulence using mean shape informa-
tion,” IEEE Trans. Biomed. Eng., vol. 57, no. 3, pp. 334–342, Mar.
2010.
[74] J. P. Martínez, I. Cygankiewicz, D. Smith, A. Bayés de Luna, P. La-
guna, and L. Sörnmo, “Detection performance and risk stratification
using model-based heart rate turbulence analysis,” Ann. Biomed. Eng.,
vol. 38, pp. 3173–3184, 2010.
[75] A.Celik,M.Melek,S.Yuksel,E.Onrat,andA.Avsar,“Cardiacau-
tonomic dysfunction in hemodialysis patients: The value of heart rate
turbulence,” Hemodial. Int., vol. 15, pp. 193–199, 2011.
[76] M. Di Rien zo, G. Bertinieri, G. Mancia, and A. Pedotti, “A new method
for evaluating the baroreflex role by a joint pattern analysis of pulse
interval and systolic blood pressure series,” Med. Biol. Eng. Comput.,
vol. 23, pp. 313–314, 1985.
[77] H. W. Robbe, L. J. Mulder, H. Rüddel, W. A. Langewitz, J. B.
Veldman, and G. Mulder, “Assessment of baroreceptor reflex sen-
sitivity by means of spectral analysis,” Hypertension, vol. 10, pp.
538–543, 1987.
[78] M. Di Rienzo, P. Castiglioni, G. Mancia, A. Pedotti, and G. Parati,
“Advancements in estimating baroreflex function: Exploring different
aspects of autonomic control of the heart through the sequence tech-
nique,” IEEE Eng. Med. Biol. Mag., vol. 20, pp. 25–32, 2001.
[79] L. J. Chesterton, M. N. Selby, J. O. Burton, and C. W. McIntyre,
“Cool dialysate reduces asymptomatic intradialytic hypotension and
increases baroreflex variability,” Hemodial. Int., vol. 13, pp. 189–196,
2009.
[80] L. J. Chesterton, N. M. Selby, J. O. Burton, J. Fialova, C. Chan, and
C. W. McIntyre, “Categorization of the hemodynamic response to
hemodialysis: The importance of baroreflex sensitivity,” Hemodial.
Int., vol. 14, pp. 18–28, 2010.
[81] S. Gouveia, A. P. Rocha, P. Laguna, and P. Lago, “Time domain
baroreflex sensitivity assessment by joint analysis of spontaneous SBP
and RR series,” Biomed. Signal Proc. Control, vol. 4, pp. 254–261,
2009.
[82] D. Laude et al., “Comparison of various techniques used to estimate
spontaneous baroreflex sensitivity (the EuroBaVar study),” Amer. J.
Physiol. Regul. Integr. Comp. Physiol., vol. 286, pp. R226–R231,
2004.
[83] D. Hernando, R. Bailón, P. Laguna, and L. Sörnmo, “Heart rate vari-
ability during hemodialysis and its relation to hypotension,” in Proc.
Comput. Cardiol., 2011, vol. 38, pp. 189–192.
[84] J. Allen, “Photoplethysmography and its application in clinical physi-
ological measurement,” Physiol. Meas., vol. 28, pp. R1–R39, 2007.
[85] C. Ahlström, A. Johansson, F. Uhlin, T. Länne, and P. Ask, “Nonin-
vasive investigation of blood pressure changes using the pulse wave
transit time: A novel approach in the monitoring of hemodialysis pa-
tients,” J. Artif. Organs, vol. 8, pp. 192–197, 2005.
[86] A. Burkert, A. Scholze, and M. Tepel, “Noninvasive continuous moni-
toring of digital pulse waves during hemodialysis,” ASAIO J.,vol.52,
pp. 174–179, 2006.
[87] E. Mambelli, E. Mancini, and A. Santoro, “A continuous and non-inva-
sive arterial pressure monitoring system in dialysis patients,” Nephron.
Clin. Pract., vol. 107, pp. c170–c176, 2007.
[88] W. Chen, T. Kobayashi, S. Ichikawa, Y. Takeuchi, and T. Togawa,
“Continuous estimation of systolic blood pressure using the pulse ar-
rival time and intermittent calibration,” Med. Biol. Eng. Comput., vol.
38, pp. 569–574, 2000.
[89] A. Sifil,D.Bednarz,B.Larive,M.Garcia,S.M.Lau,andE.P.Pa-
ganini, “Evaluation of the harmonized alert sensing technology de-
vice for hemodynamic monitoring in chronic hemodialysis patients,”
ASAIO J., vol. 49, pp. 667–672, 2003.
[90] R.A.Payne,C.N.Symeonides,D.J.Webb,andS.R.J.Maxwell,
“Pulse transit time measured from the ECG: An unreliable marker of
beat-to-beat blood pressure,” J. Appl. Physiol., vol. 100, pp. 136–141,
2006.
[91] M. Nitzan, A. Babchenko, B. Khanokh, and D. Landau, “The vari-
ability of the photoplethysmographic signal. A potential method for
the evaluation of the autonomic nervous system,” Physiol. Meas., vol.
19, pp. 93–102, 1998.
[92] K. Solem, B. Olde, and L. Sörnmo, “Prediction of intradialytic hy-
potension using photoplethysmography,” IEEE Trans. Biomed. Eng.,
vol. 57, no. 6, pp. 1611–1619, Jun. 2010.
[93] F. Javed, P. M. Middleton, P. Malouf, G. S. Chan, A. V. Savkin, N. H.
Lovell, E. Steel, and J. Mackie, “Frequency spectrum analysis of finger
photoplethysmographic waveform variability during haemodialysis,”
Physiol. Meas., vol. 31, pp. 1203–1216, 2010.
[94] E. Mancini, L. Corazza, D. C. Cannarile, M. L. Soverini, S. Caval-
canti, S. Cavani, A. Fiorenzi, and A. Santoro, “Short term variability
of oxygen saturation during hemodialysis is a warning parameter for
hypotension appearance,” in Proc. Comput. Cardiol., 2008, vol. 35,
pp. 881–883.
[95] M.Shamir,L.A.Eidelman,Y.Floman,L.Kaplan,andR.Pizov,
“Pulse oximetry plethysmographic waveform during changes in blood
volume,” Brit. J. Anaesth., vol. 82, pp. 178–181, 1999.
[96] L. B. Cook, “Extracting arterial flow waveforms from pulse oximeter
waveforms apparatus,” Anaesthesia, vol. 56, pp. 551–555, 2001.
[97] S. P. Linder and S. Wendelken, “Using the morphology of the pho-
toplethysmogram envelope to automatically detect hypovolemia,” in
IEEE Int. Symp. Comput.-Based Med. Sys., 2006, vol. 19, p. 371.
[98]L.Chen,A.T.Reisner,A.Gribok,andJ.Reifman,“Isrespiration-
induced variation in the photoplethysmogram associated with major
hypovolemia in patients with acute traumatic injuries?,” Shock, vol.
34, pp. 455–460, 2010.
[99] R. Pizov, A. Eden, D. Bystritski, E. Kalina, A. Tamir, and S. Gelman,
“Arterial and plethysmographic waveform analysis in anesthetized pa-
tients with hypovolemia,” Anesthesiol., vol. 113, pp. 83–91, 2010.
[100] S.P.McGrath,K.L.Ryan,S.M.Wendelken,C.A.Rickards,andV.
A. Convertino, “Pulse oximeter plethysmographic waveform changes
in awake, spontaneously breathing, hypovolemic volunteers,” Anesth.
Analg., vol. 112, pp. 368–374, 2011.
[101] N.Selvaraj,K.H.Shelley,D.G.Silverman,N. Stachenfeld, N. Ga lante,
J. P. Florian, Y. Mendelson, and K. H. Chon, “A novel approach using
time-frequencyanalysis of pulse-oximeter data todetect progressive hy-
povolemia in spontaneously breathing healthy subjects,” IEEE Trans.
Biomed. Eng., vol. 58, no. 11,pp. 2272–2279, Nov. 2011.
[102] K. H. Chon, S. Dash, and K. Ju, “Estimation of respiratory rate from
photoplethysmogram data using time-frequency spectral estimation,”
IEEE Trans. Biomed. Eng., vol. 56, no. 9, pp. 2054–2063, Sep. 2009.
[103] K. Nakajima, T. Tamura, and H. Miike, “Monitoring of heart and res-
piratory rates by photoplethysmography using a digital filtering tech-
nique,” Med. Eng. Phys., vol. 18, pp. 365–372, 1996.
[104] N. Selvaraj, A. Jaryal, J. Santhosh, K. K. Deepak, and S. Anand, “As-
sessment of heart rate variability derived from finger-tip photoplethys-
mography as compared to electrocardiography,” J. Med. Eng. Technol.,
vol. 32, pp. 479–484, 2008.
[105] E. Gil, M. Orini, R. Bailón, J. M. Vergara, L. Mainardi, and P. La-
guna, “Photoplethysmography pulse rate variability as a sur rogate mea-
surement of heart rate variability during non-stationary conditions,”
Physiol. Meas., vol. 31, pp. 1271–1290, 2010.
[106] E. Gil, L. Sörnmo, and P. Laguna, “Detection of heart rate turbulence
in photoplethysmographic signals,” in Proc. Comput. Cardiol.,2011,
vol. 38, pp. 665–668.
[107] T. L. Rusch, R. Sankar, and J. E. Scharf, “Signal processing methods
for pulse oximetry,” Comput. Biol. Med., vol. 26, pp. 143–159, 1996.
[108] S. Rhee, B.-H. Yang, and H. H. Asada, “Artifact-resistant power-ef-
ficient design of finger-ring plethysmographic sensors,” IEEE Trans.
Biomed. Eng., vol. 48, no. 7, pp. 795–805, Jul. 2001.
[109] J. Y. A. Foo and S. J. Wilson, “A computational system to optimise
noise rejection in photoplethysmography signals during motion or poor
perfusion states,” Med. Biol. Eng. Comput., vol. 44, pp. 140–145, 2006.
[110] Y.-S. Yan and Y.-T. Zhang, “An efficient motion-resistant method for
wearable pulse oximeter,” IEEE Trans. Info. Tech. Biomed., vol. 12,
no. 2, pp. 399–405, Apr. 2008.
[111] J. McNames and M. Aboy, “Statistical modeling of cardiovascular sig-
nals and parameter estimation based on the extended Kalman filter,”
IEEE Trans. Biomed. Eng., vol. 55, no. 1, pp. 119–129, Jan. 2008.
[112] B. Yang and K. H. Chon, “A novel approach to monitor nonstationary
dynamics in physiological signals: Application to blood pressure,
pulse oximeter, and respiratory data,” Ann. Biomed. Eng., vol. 38, pp.
3478–3488, 2010.
[113] R. Krishnan, B. Natarajan, and S. Warren, “Two-stage approach for
detection and reduction of motion artifacts in photoplethysmographic
data,” IEEE Trans. Biomed. Eng., vol. 57, no. 8, pp. 1867–1876, Aug.
2010.
[114] M. S. Schwartz and F. Andrasik, Eds., Biofeedback: A Practitioner’s
Guide3rd ed. Guilford Press, 2005.
[115] A. Santoro, E. Mancini, F. Paolini, M. Spongano, and P. Zucchelli,
“Automatic control of blood volume trends during hemodialysis,”
ASAIO J., vol. 40, pp. M419–M422, 1994.
[116] A. Santoro, E. Ferramosca, and E. Mancini, “Biofeedback-driven dial-
ysis: Where are we?,” in Hemodialysis—From Basic Research to Clin-
ical Trials, C. Ronco and D. N. Cruz, Eds. Basel: Karger, 2008, vol.
161, pp. 199–209.
SÖRNMO et al.: NONINVASIVE TECHNIQUES FOR PREVENTION OF INTRADIALYTIC HYPOTENSION 59
[117] D. Schneditz, C. Ronco, and N. Levin, “Temperature control by the
blood temperature monitor,” Semin. Dial., vol. 16, pp. 477–482, 2003.
[118] F. Javed, A. V. Savkin, G. S. Chan, J. D. Mackie, and N. H. Lovell,
“Recent advances in the monitoring and controlof haemodynamic vari-
ables during haemodialysis: A review,” Physiol. Meas., vol. 33, pp.
R1–R31, 2012.
[119] M. Nordio, S. Giove, S. Lorenzi, P. Marchini, and E. Saporiti, “A
new approach to blood pressure and blood volume modulation during
hemodialysis: An adaptive fuzzy control module,” Int. J. Artif. Organs,
vol. 18, pp. 513–517, 1995.
[120] R. Schmidt, O. Roeher, H. Hickstein, and S. Korth, “Prevention of
haemodialysis-induced hypotension by biofeedback control of ultrafil-
tration and infusion,” Nephrol. Dial. Transplant., vol. 16, pp. 595–603,
2001.
[121] M. Ursino and M. Innocenti, “Mathematical investigation of some
physiological factors involved in hemodialysis hypotension,” Artif.
Organs, vol. 21, pp. 891–902, 1997.
[122] M. Ursino, L. Colì, E. Magosso, P. Capriotti, A. Fiorenzi, P. Baroni,
and S. Stefoni, “A mathematical model for the prediction of solute ki-
netics, osmolarity and fluid volume changes during hemodiafiltration
with on-line regeneration of ultrafiltrate (hfr),” Int. J. Artif. Organs,
vol. 29, pp. 1031–1041, 2006.
[123] F. Javed, A. V. Savkin, G. S. Chan, J. D. Mackie, and N. H. Lovell,
“Identification and control for automated regulation of hemodynamic
variables during hemodialysis,” IEEE Trans. Biomed. Eng., vol. 58, no.
7, pp. 1686–1697, Jul. 2011.
[124] F. Javed, A. V. Savkin, G. S. H. Chan, P. M. Middleton, P. Malouf,
E. Steel, J. Mackie, and T. M. Cheng, “Model predictive control of
relative blood volume and heart rate during hemodialysis,” Med. Biol.
Eng. Comput., vol. 48, pp. 389–397, 2010.
[125] A. Santoro, E. Mancini, C. Basile, L. Amoroso, S. Di Giulio, M. Us-
berti, G. Colasanti, G. Verzetti, A. Rocco, E. Imbasciati, G. Panzetta,
R. Bolzani, F. Grandi, and M. Polacchini, “Blood volume controlled
hemodialysis in hypotension-prone patients: A randomized, multi-
center controlled trial,” Kidney Int., vol. 62, pp. 1034–1045, 2002.
[126] V. Bégin, C. Déziel, and F. Madore, “Biofeedback regulation of ultra-
filtration and dialysate conductivity for the prevention of hypotension
during hemodialysis,” ASAIO J., vol. 48, pp. 312–315, 2002.
[127] C. F. Franssen, J. J. Dasselaar, P. Sytsma, J. G. Burgerhof, P. E. de Jong,
andR.M.Huisman,“Automaticfeedback control of relative blood
volume changes during hemodialysis im proves blood pressure stability
during and after dialysis,” Hemodial. Int., vol. 9, pp. 383–392, 2005.
[128] C. Ronco, A. Brendolan, M. Milan, M. P. Rodeghiero, M. Zanella,
and G. La Greca, “Impact of biofeedback-induced cardiovascular sta-
bility on hemodialysis tolerance and efficiency,” Kidney Int., vol. 58,
pp. 800–808, 2000.
[129] D. Gabrielli, B. Krystal, K. Katzarski, M. Youssef, T. Hachache, F.
Lopot, C. Lasseur, T. Gunne, B. Draganov, R. Wojke, and A. Gauly,
“Improved intradialytic stability during haemodialysis with blood
volume-controlled ultrafiltration,” J. Nephrol., vol. 22, pp. 232–240,
2009.
[130] M. J. Damasiewicz and K. R. Polkinghorne, “Intra-dialytic hy-
potension and blood volume and blood temperature monitoring,”
Nephrology, vol. 16, pp. 13–18, 2011.
[131] A. Davenport, “Intradialytic complications during hemodialysis,”
Hemodial. Int., vol. 10, pp. 162–167, 2006.
[132] Y. Borhani, S. Fleming, D. Clifton, S. Sutherland, L. Hills, D.
Meredith, C. Pugh, and L. Tarassenko, “Towards a data fusion model
for predicting deterioration in haemodialysis patients,” in Proc.
Comput. Cardiol., 2010, vol. 37, pp. 967–970.
[133] N. Kannathal, U. R. Acharya, E. Y. Ng, S. M. Krishnan, L. C. Min,
and S. Laxminarayan, “Cardiac health diagnosis using data fusion of
cardiovascular and haemodynamic signals,” Comput. Meth. Progr.
Biomed., vol. 82, pp. 87–96, 2006.
[134] L. Tarassenko, A. Hann, and D. Young, “Integrated monitoring and
analysis for early warning of patient deterioration,” Brit. J. Anaesth.,
vol. 97, pp. 64–68, 2006.
Leif Sörnmo (S’80–M’85–SM’02) received the
M.Sc. and Ph.D. degrees in electrical engineering
from Lund University, Lund, Sweden, in 1978 and
1984, respectively.
From 1983 to 1995, he was a Research Fellow at
the Department of Clinical Physiology, Lund Uni-
versity, where he was engaged in research on ECG
signal processing. Since 1990, he has been with the
Signal Processing Group, Department of Electrical
and Information Technology, Lund University, where
he is currently a Professor of biomedical signal pro-
cessing. He is the author of Bioelectrical Signal Processing in Cardiac and Neu-
rological Applications (Elsevier, 2005). His research interests include statistical
signal processing, modeling of biomedical signals, methods foranalysis of atrial
fibrillation, multimodal signal processing in hemodialysis, and power-efficient
signal processing in implantable devices.
Dr. Sörnmo is an Associate Editor of the IEEE TRANSACTIONS ON
BIOMEDICAL ENGINEERING and the Journal of Electrocardiology, a member
of the Editorial Board of the Medical and Biological Engineering and Com-
puting, and was an Associate Editor of the Computers in Biomedical Research
(1997–2000).
Frida Sandberg received the M.Sc. degree in
electrical engineering and the Ph.D. degree in signal
processing from Lund University, Lund, Sweden, in
2003 and 2010, respectively.
She is currently a Postdoctoral Fellow in the
Signal Processing Group, Department of Electrical
and Information Technology, Lund University. Her
research interests include statistical signal pro-
cessing, time–frequency analysis, and modeling of
biomedical signals, with particular interest in signals
of cardiovascular origin.
Eduardo Gil received the M.Sc. degree in telecom-
munication engineering and Ph.D. degree in
biomedical engineering from Zaragoza University,
Zaragoza, Spain, in 2002 and 2009, respectively.
He is currently Assistant Professor in automatic
control at the Department of Computer Science and
Engineering Systems, University of Zaragoza, and
a Researcher at the Aragón Institute for Engineering
Research. He is also a member of the Spanish
Center for Biomedical Engineering, Biomaterial,
and Nanomedicine Research. His research interests
include biomedical signal processing and time–frequency analysis, with
particular interest in photoplethysmographic signals.
Kristian Solem received the M.Sc. degree in elec-
trical engineering and the Ph.D. degree in biomed-
ical signal processing from Lund University, Lund,
Sweden, in 2003 and 2008, respectively.
He is currently with the Research Department,
Gambro Lundia AB, Lund, Sweden. His research
interests include ECG, pressure, and photoplethys-
mographic signals and their significance in heart rate
variability analysis, heart rate turbulence analysis,
and hemodialysis.
