Computation

of aortic flow from pressure

in humans using a nonlinear, three-element
K. H. WESSELING,

J. R. C. JANSEN,

J. J. SETTELS,

AND

model

J. J. SCHREUDER

Netherlands
Organization for Applied Scientific Research, TN0 Biomedical Instrumentation,
Academic
Medical Centre, 1105 AZ Amsterdam; Department of Pulmonary Diseases, Erasmus University, 3000 DR
Rotterdam; and Department of Anesthesiology, Academic Hospital, University of Limburg, 6202 AZ
Maastricht, The Netherlands

WESSELING, K.H.,J.R.C.
JANSEN, J.J. SETTELS,AND J.J.
SCHREUDER. Computation of aortic flow from pressurein humans using a nonlinear, three-element
model. J. Appl. Physiol.
74(5): 2566-2573, 1993.-We
computed aortic flow pulsations
from arterial pressure by simulating
a nonlinear,
time-varying
three-element
model of aortic input impedance. The model elements represent aortic characteristic
impedance, arterial compliance, and systemic vascular resistance. Parameter values for
the first two elements were computed from a published, age-dependent, aortic pressure-area
relationship
(G. J. Langewouters
et al. J. Biomech. 17: 425-435, 1984). Peripheral
resistance was
predicted
from mean pressure and model mean flow. Model
flow pulsations
from aortic pressure showed the visual aspects
of an aortic flow curve. For evaluation
we compared model
mean flow from radial arterial pressure with thermodilution
cardiac output estimations,
76 times, in eight open heart surgical patients. The pooled mean difference was +7%, the SD
22%. After using one comparison
per patient to calibrate the
model, however, we followed quantitative
changes in cardiac
output that occurred either during changes in the state of the
patient or subsequent to vasoactive drugs. The mean deviation
from thermodilution
cardiac output was +2%, the SD 8%.
Given these small errors the method could monitor cardiac output continuously.
aortic input impedance; nonlinear
model; time-varying
model;
aortic arctangent
pressure-area
relationship;
age dependency;
simulation;
stroke volume; cardiac output

OUTPUT
is an important
hemodynamic
variable. It is studied in physiological
experiments often in
relation with changes in pressure. It is monitored in surgical and critical care patients. There is a trend in physiological research, as well as in the operating room and
intensive care unit, to require essentially continuous cardiac output monitoring.
Pressure pulse analysis h .as been U sed for beat-to -beat
ca rdiac output monitorin g (9, 15) Traditionally,
such
methods corn .puted cardiac stroke output from certai n
characteristic s of an arterial pressure pulse, based on a
CARDIAC

2566

0161-7567193

$2.00 Copyright

variety of simple models of the arterial system. Characteristics used included systolic area, heart rate, and mean
and diastolic pressure. In these methods, aortic properties were assumed constant under varying distending
pressures. They were not constant, however, and models
were not sufficiently
detailed. Consequently,
pressure
pulse analysis often proved to be of insufficient precision (6).
More recently, several studies have shown the suitability of a three-element
model of arterial input impedance
to describe the relationship between aortic pressure and
flow (2, 16). Once model parameters are found, flow can
be computed from measured pressure by simulating the
model (2). This flow then provides a continuous measure
of cardiac output. Integrated over one heartbeat it provides stroke volume; integrated over one minute it provides cardiac minute volume.
We investigated whether this model could monitor cardiac output continuously.
We were fortunate in that a
precise thermodilution
technique is now available (4),
and we anticipated that this improved technique would
permit a more reliable evaluation.
METHODS

Computing Flow From Pressure

During the development we made three decisions. 1)

We would compute flow by simulating the response of a
three-element
model of arterial input impedance to arterial pressure (Fig. 1). To do this, the parameter values of
the model elements must be known. Two of the model
parameters, characteristic impedance and arterial compliance, can be derived from an aortic pressure-area relationship. 2) We would use the arctangent model of aortic
mechanics of Langewouters et al. (7) (Fig. 2). The use of
this arctangent relationship
causes the model elements
to be nonlinearly
dependent on pressure. The third element, total systemic peripheral resistance, is a variable.
Its value is not known but is an outcome of the model

0 1993 the American

Physiological

Society

AORTIC

FLOW

COMPUTED

FROM

stole this inflow is dissipated in the periphery. Given input pressure, aortic systolic inflow is principally
determined by the time constant Z,C,. If this time constant is
too short, computed stroke volume will be too small. The
peripheral resistance element value is not a major determinant of systolic inflow. Diastolic outflow from the
windkessel into the peripheral resistance and windkessel
pressure decay are principally
determined by the time
constant R,C,. If input pressure decay equals model
windkessel pressure decay, the diastolic inflow is zero.
Windkessel compliance is a common factor in both time
constants.
Model parameter ualues. To carry out the computations numerically,
proper values for the model parameters Z,, C,, and R, must be found. We do have precise
and detailed results on the viscoelasticity of the human
aorta in the form of pressure-area relations (7), and Z,
and C, can be computed from it.
The aortic characteristic
impedance Z, can be expressed as
Z, = \JpI(AC)
(1)
where p is the density of blood, A is the cross-sectional
area of the aorta, and C is the aortic compliance per unit
length. C is the derivative of the pressure-area relationship with respect to pressure (P)
C = dA/dP
(2)
Windkessel compliance, C,, represents the lumped compliance of the entire arterial system. We assume its value
to be equal to the compliance of one unit length of thoracic aorta times an aortic effective length, 1
C - LCI
(3)
For an adult patient we assumed for L a value of 80 cm
(19). More strictly, L could depend on patient height and
weight.

Mt)L
/

P(1)

-f

RP

PIA)

U)

FIG.
1. Diagram
of &element
model used in this study to compute
flow. ZO, characteristic
impedance
of proximal
aorta; C,, windkessel
compliance
of arterial
system; I$,, total systemic
peripheral
resistance.
2, and C, have nonlinear,
pressure-dependent
properties,
indicated
by
stylized
S symbol.
R, varies with time, as symbolized
by arrow. Q(t),
blood flow as function
of time; P(t), arterial
pressure
waveform;
P,(t),
windkessel
pressure.

simulation. 3) We would use the current peripheral resistance value as a best estimate of that parameter in the
simulation
of the next beat.
The model. To compute aortic flow from pressure we
base the pressure-flow relationship on the traditional
(1,
20) but adequate (16) model shown in Fig. 1. This model
has three elements representing the three major properties of the aorta and arterial system: aortic characteristic
impedance (Z,), a dynamic property of the aorta that
impedes pulsatile outflow from the ventricle; windkessel
or buffer compliance (C,), the ability of the aorta and
arterial system to elastically store the cardiac stroke output from the left ventricle; and peripheral
resistance
(R,), the Poiseuille
resistance of all vascular beds together. The value of total peripheral resistance in the
model is the sum of Z, and R,. The special symbols in Fig.
1 indicate the nonlinear,
pressure-dependent
properties
of Z, and C, and the time-varying
property of I$.
During systole, flow is into the model; during dia-

1
4

l-

1 1 .

.-SO
PO
70

,,

5-

--

m-c-

4-

100

1 ,

150

1 1 1 .

-8

.o0.0.
&

0O*
uo-

0: l

0
0

l-

P [ mmHgJ
1 1 I

W-

I a - *s-c

32-

I
I
I
I

I
I
I

6-

200

A [Yrl

1111111111

0
0

50

100

PO + PI

60
50
40 r
x

2567

PRESSURE

0
0

FIG. 2. A: arctangent
pressure-area
relationship.
It
fits measured
pressure-area
curves of human
aorta
well (7) and is characterized
by 3 parameters:
maximal
area (A,,,),
pressure
at inflection
point
(p,>, and
width parameter
(Pi). Measurements
of the 3 parameters are plotted
in R-D, together
with their regressions on age (A), as listed in Table 1. Open circles,
female (F) aortas; filled circles, male (M) aortas. Parameters
PO and P, show only small scatter
around
regression,
but
scatter
for A,,,
is substantial.
[Adapted
from data in Langewouters
et al. (7).]

2568

AORTIC

FLOW

1. Summay
of regression equations
for arctangent aortic model parameters
vs. patient
and gender (7)

COMPUTED

TABLE

Parameter

A max?

4.12
72 - 0.89A
57 - 0.44A

cm2

PO, mm&

PI, mmHg

age

M
5.62
76 - 0.89A
57 - 0.44A

F, female; M, male; A,,,, maxi ma1 area; PO, inflection

pressure;
P, ,
width parameter;
A, patient
age ( in yr). We used these data to obtain
proper pressure-area
relation
for a patient,
given gender and age.

We define R, as the ratio of average pressure to average flow. Its value changes only slowly (15) compared
with a heartbeat interval. We therefore use its current
computed value to simulate the flow of the next beat. For
the first beat, at the start of the simulation, a reasonable
initial value is assumed. We take the ratio of 100 mmHg
mean pressure and 3 l/min cardiac output. From true
mean pressure and computed mean flow the next approximation
is computed, and so on. R, converges from
the initial value to the correct value in a few heartbeats.
Pressure-area relationship. Z, and C, have been presented in terms of the aortic pressure-area relation and
its derivative. We must now define that relation. According to Langewouters et al. (7), the thoracic aortic crosssectional area can be described as a function of pressure
by an arctangent with three parameters (see Fig. 2A)
1

A(P) = A,,, 0.5 + -7r arctan

[

P - PO
p

i 11
P1

(4)

A max is the maximal

cross-sectional
area at very high
pressure. Parameter POdefines the position of the inflection point on the pressure axis (40 mmHg in Fig. 2); P,
defines the width between the points at one-half and
three-quarter
amplitude
(50 mmHg in Fig. 2). Parameters POand P, compare with the mean and the SD of a
probability
function.
The compliance per unit length is the derivative of
A(P) with respect to P (Eq. 2) and depends on P
C(P) =

(5)

Given correct values for the three arctangent parameters

Z, and C, can be
computed for any pressure P.
Arctangent parameters. The three arctangent parameters have been determined for 45 human thoracic aortas
in vitro (7) and are presented in graphic form in Fig. 2
(B-D). It appears that PO and P, regress tightly on patient age (A), given patient gender. A,,, does not regress
on age, at least not over the observed age range. Its average value is indicated separately for females and males.
However, an individuals
value can deviate as much as
40% from the group average. The statistical information
is in Table 1. Thus given patient gender and age, we can
compute A,,, , PO, and P, using the data in Table 1 to find
a proper pressure-area relationship
for a subject, but
with uncertainty in the maximal area. To improve accuA max, PO, and P, , the model parameters

FROM

PRESSURE

racy in monitoring
an individual patient, a proper value
for maximal area must be obtained in another way. If no
other information
is available, the group average value is
used as a best estimate. The error involved from this is as
follows.
Importance of maximal area. Calculations
show that
computed model flow using the group average maximal
area deviates by the same percentage amount from true
flow as the group average maximal area deviates from a
patients true area. Given the 20% scatter SD in the maximal area, model flow cannot be computed with an accuracy better than this percentage. An accurate measurement of aortic diameter can yield a patient individual
value for maximal area, improving accuracy. An indicator dilution estimate of cardiac output can also be used,
as was proposed by Warner (17). The initial population
average value of A,,, is then multiplied
by K, an individual patients ratio of thermodilution
to model flow cardiac output.
Model flow computation. Model flow, finally, is computed by simulating the behavior of the model under the
applied arterial pressure pulsation. The values for windkessel compliance and characteristic
impedance, as a
function of instantaneous
pressure, are inserted in the
model. Because these values depend on pressure as was
shown above, the model behavior is nonlinear. Simulation is done digitally, and model computations
are repeated for each new pressure sample taken. Left ventricular stroke volume is computed by integrating model flow
during systole; cardiac output is computed by multiplying stroke volume with instantaneous
heart rate. The
systolic duration and the heartbeat interval are derived
from the pressure waveform. We call this approach the
model flow method.
Patients

We evaluated the model flow method using arterial

pressure and thermodilution
cardiac output recorded in
the operating room. After obtaining informed consent,
we studied eight male patients who suffered from multiple coronary vessel disease but had intact aortic valves.
All patients were monitored in 1989, and their signals
were digitized. Except for two patients whose data were
accidentally destroyed, all patients studied in that year
were analyzed. The patients underwent elective coronary
artery bypass graft surgery. Lorazepam (5 mg) was administered as premeditation
2 h before surgery. A peripheral venous catheter, a 20-gauge radial artery cannula
with a Medisize continuous flush device, and a 7-Fr
Swan-Ganz pulmonary
artery catheter were inserted
under local anesthesia. Before induction of anesthesia a
series of baseline hemodynamic
measurements was performed under spontaneous breathing conditions. Anesthesia was induced with an initial dose of 7.5 pg/kg of
sufentanil and was maintained
with a continuous infusion of sufentanil at a rate of 3.75 pg. kg- h-l. Pancuronium bromide (0.1 mg/kg) was given for muscle relaxation. The patients were ventilated at 10 cycles/min with
an oxygen-air mixture having an inspired oxygen fraction of 50%.
l

AORTIC

Experimental

FLOW

COMPUTED

Techniques

In all patients, we recorded pressure in the radial artery using a standard

Hewlett-Packard
model 78534A
arterial pressure channel with a 15-Hz built-in cutoff frequency. In two patients, we could additionally
record ascending aortic pressure using a Millar microtip pressure
transducer,
but only during series e (see Measurement
Protocol), just before going on bypass. A Baxter COM-2
device computed thermodilution
cardiac output. A Siemens model 900B servoventilator
was used for mechanical ventilation.
It also provided the time pulses to trigger
injections of 5 ml of room temperature
5% glucose with a
computer-controlled
power injector. An interface box
passed all monitor and control lines, providing electrical
isolation.
This interface
comprises
three buttons
to
flush, start, and, if necessary, break the automatic thermodilution injections. The interface was connected to an
IBM-compatible
personal computer of AT level controlling the hardware
system. Because aortic pressure was
only incidentally
available in these patients but radial
pressure was available throughout
the operation, the latter was used as input to the model. An aortic flow pulsation signal for comparison with the computed model flow
pulsation
was also not available in these patients. Instead, average model flow was compared with thermodilution estimates of average cardiac output.
To improve the accuracy of the thermodilution
estimates, we used the technique of phase-controlled
injections, equally spread over the ventilatory
cycle (4). Using
bolus injections, we estimated cardiac output four times,
each in a different phase of the ventilatory
cycle, and
averaged the results. Each injection was delayed from the
start of a ventilatory
period over either 0, 25, 50, or 75%
of the duration of the period. We waited -30 s between
injections. We averaged model flow cardiac output over
two ventilatory
cycles beginning at each thermodilution
injection. Each series of four provided one comparison
between model flow and thermodilution
cardiac output.
Measurement

Protocol

At various moments during the operation,

measurements were done after major changes in patient state.
Surgery was suspended, and hemodynamics
were usually
stable during the measurement
periods. We did series s
during spontaneous
breathing, before induction of anesthesia. Series a-k were taken with the patient maintained on the ventilator. The code letters a-k correspond
to the following
11 instants: a, 3 min after induction; b,
immediately
after sternotomy;
c-e, at 15, 10, and 5 min
before bypass, respectively;
f-i, at 3, 8, 13, and 18 min
after bypass, respectively; j, 3 min after sternal fixation;
and k, after end of surgery. Because of surgical and anesthetic events and depending on the duration of the operation, not every series was always performed
in all patients. Seven to 12 series of four thermodilution
estimations were thus carried out. In one patient an extra series
was done. The series a thermodilution
cardiac output was
always obtained and was used to calibrate model flow
cardiac output. During series b, c, and e before bypass,
and during series h, j, and k after bypass, sodium nitroprusside
was infused in doses of between 0.5 and 1.5

FROM

2569

PRESSURE

pg kg-. min.
This drug has a strong vasodilating
effect. Other vasoactive drugs were administered
as needed
during the operation.
We performed
an off-line sensitivity
analysis of the
model flow method by repeating the model flow computations on the digitized data base. Patient age and pressure zero level were artificially varied. The first estimate
was computed with unmodified
parameters.
Computations were then repeated with patient age entered incorrectly at +lO and -10 yr off true age. Age appears in the
values for the arctangent
parameters
PO and P,. The
computations
were repeated two more times with the
correct age but with a pressure offset of +lO and -10
mmHg. Insensitivity
to the absolute level of pressure is
important
because the pressure transducer
membrane is
not always precisely at heart level. The mean difference
of model flow with simultaneous
thermodilution
cardiac
output and its SD and range were used to judge sensitivity.
Our study could have produced good results accidentally because circumstances
were right or patients were
stable. We decided, therefore, to use the earlier, so-called
corrected
characteristic
impedance (cZ) method (19)
on the same data. This method has been evaluated before
(5), and its behavior is known. It computes beat-to-beat
stroke volume, V,, by integrating the area under the systolic portion of the arterial pressure pulse, dividing the
area by the CZ methods characteristic
impedance calibration factor, Z,,
l

V Z=

1
Z ao s T ,

[P(t) - P,ldt

(6)

A correction is then applied that accounts for changes

in mean pressure (P,) and heart rate (f). This results in
corrected stroke volume estimates, V,,
V cz = [0.66 + 0.005f - O.OlA(O.O14P,

- 0.8)]Vz

(7)

where P, is arterial end-diastolic

pressure, T, is the ejection period, and A is the age of the patient. The method is
calibrated
for each individual patient similarly
to the
model flow method, but adjusting Z,,.
RESULTS

Figure 3 presents an example of measured aortic and

radial arterial pressure and computed model flow pulsations from each pressure signal. Compared with aortic
pressure, radial pressure appears delayed and distorted.
Model flow computed from aortic pressure shows the typical characteristics
of an aortic waveform:
a steep upstroke, a slow downstroke,
a steep end-systolic
downstroke terminated
in a sharp dicrotic wave, and a flat
near-zero diastolic flow. The flow wave computed from
the distorted radial pressure is also distorted. In particular, peak flow is exaggerated.
Table 2 lists patient age, height, and model flow calibration factor K, pressure, pulse rate, and thermodilution
cardiac output ranges. K indicates by which factor the
patient aortic cross-sectional
area is different from the
population
average. For our eight patients, K averages
107 t 22%. Th is is not significantly
different from 100%
(t test, Z-sided, at P = 0.05). The 22% SD is not signifi-

2570

AORTIC

FLOW

COMPUTED

FROM

PRESSURE

TABLE 3. Model flow and CZ method errors

after calibration
Model

0.5

1.5

2.5

I
4

3.5

I
0

0.5

1.5

25

3.5

Patient

Avg ad,
l/min

8
9
10
11
12
15
16
17

5.12
4.89
4.67
5.42
4.59
4.10
4.69
4.29

10
7
8
12
10
7
9
13

Mean
+SD

4.72
kO.43

Mean,
l/min

Flow

Error

SD,
l/min

CZ Method

Mean,
l/min

Error

SD,
l/min

-0.14
0.34
0.04
0.20
0.04
-0.14
0.45
-0.05

0.41
0.29
0.39
0.33
0.34
0.40
0.27
0.46

8.2
5.7
8.4
6.0
7.3
9.5
5.8
10.5

0.30
0.30
0.63
T 0.58
0.79
0.45
0.33
0.04

0.49
0.65
0.40
0.35
0.40
0.69
0.43
0.92

9.6
13.3
8.6
6.5
8.7
16.8
9.2
21.4

0.09
kO.22

0.36
20.07

7.7
k1.8

0.28
20.42

0.54
-to.20

11.8
k5.1

t M

3. A: simultaneous
aortic
(solid
line) and radial
pressures
(dashed line). B: simulated
model flow from aortic (solid line) and radial pressures
(dashed line). Model flow computed
from aortic pressure
shows typical characteristics
of an aortic flow pulsation:
steep upslope
at beginning
of systole, a gradual down slope terminated
in a steep final
phase, and sharp dicrotic
notch at end systole, followed
by a period of
almost zero flow in diastole.
Model flow computed
from radial pressure
is delayed and distorted
and shows typical characteristics
less convincingly.
FIG.

9.5

cZ, corrected
characteristic
impedance.
%, Error
SD given as percentage of corresponding
average cardiac output.
Error figures exclude
the 8 pairs of cardiac output used for calibration.
The N - 1 comparisons in each patient
are averaged.
Bottom
lines give errors pooled for
group.
8

TABLE

2. Summary of patient information

P,, mm&

Patient

8
9
10
11
12
15
16
17

A, yr

62
67
60

1.53
1.02
1.32
1.01
0.81
1.02

51

62
56

Min

59
54
68
66
57
55

51

1.00

71

56

0.88

55

Max

110
94
111
98

93
98
92
110

f, beats/min

Qtd, l/min

Min

Max

Min

Max

59
61
62
71
61
50
82
49

73
96
80
109
81
85
103
94

4.2
3.9
3.8
4.3
3.6
3.2
3.3
3.1

6.4
5.9
5.9
6.9
5.7
4.6
5.8
5.8

K, model flow calibration

factor (dimensionless);
P,, mean pressure;
f, heart rate; Qtd, quadruple
thermodilution
cardiac output.
Deviations
of K values from 1 are indication
of accuracy
of model flow method
before calibration.
Averaged
over group of 8 patients,
K is 1.07 k 0.24
(SD). All patients
were male.

cantly different (F test, at P = 0.05) from the 17% SD

found earlier in the A,,, value for male patients (7).
Pressure, heart rate, and cardiac output varied approximately over a 58 range during the operations.
The results of the 76 comparisons with thermodilution
cardiac output are summarized in Table 3. We used determination a in each patient to compute the calibration
factor K. The remaining 68 comparisons were taken into
the postcalibration
error statistics. Mean error ranged
from -0.1 to 0.4 Vmin. Its average of 0.1 l/min (2%) is
not significantly
different from zero. The SD ranged
from 6 to 11% with an average of 8%. By comparison, the
offset of the CZ method ranged from -0.6 to 0.8 l/min
with a pooled average of 0.3 l/min. Scatter for the CZ
method ranged from 7 to 21% with a pooled average
of 12%.
A scatter diagram of model flow vs. thermodilution
cardiac output is shown in Fig. 4. It presents the individual
data taken in the eight patients summarized in Table 3.
We obtained from 7 to 13 quadruple measurements in a
patient. After using measurement
a for calibration,
at
least 6 and at most 12 measurements
per patient re-

quad &
in
I/min

. FIG. 4. Scattergram
of 68 individual
model flow [quadruple
(quad)
Qmf] vs. phase-controlled
thermodilution
(quad Qti) cardiac output estimates.
Values are obtained
after calibration
of model parameters.
Statistics
are summarized
in Table 3.

mained,

representing

major changes in patient

state (see
of the
model flow method showing the difference between the
model flow and thermodilution
estimates vs. their paired
means. All differences except one are within 21 l/min.
Except for five outliers all differences stay within -0.50
and +0.65 Vmin. The distribution
of the 68 errors is also
shown.
Pooled results of the sensitivity analysis for age and
pressure are given in Table 4. Overall, errors do not differ
greatly under the applied changes in the age parameter
and in the offset pressure. Judging from the error minima and maxima per condition, neither do they differ
greatly in individual patients.

Measurement Protocol). Figure 5 is the errorgram

DISCUSSION

The model flow method uses a three-element

model of
aortic input impedance to compute a flow pulsation from
an arterial pressure pulsation (2). A new feature in this
model is that two parameter values are computed (pre-

AORTIC

FLOW

Error
0

1.5

-1.5

COMPUTED

histogram
5

10

15

I,llll,,,,l,,,,J

1
3

,
4

1
t&n,

,
5
+

,
6

{
7

&d/2

in
I/min

FIG. 5. Scattergram
of difference
between
individual
model flow
(Q,,)
and thermodilution
(Q,)
cardiac
output
estimates
plotted
vs.
their mean. Differences
are independent
of level of cardiac output and
remain,
excepting
5 outliers,
within
-0.50 and +0.65 l/min.
Right: histogram
of differences.
Excepting
the 5 outliers,
errors are uniformly
distributed
(x2 test). We did not, therefore,
plot +2 SD lines.

dieted) from published arctangent aortic pressure-area

relationships.
This computation
is carried out for every
pressure sample taken on the waveform, resulting in a
nonlinear, pressure-dependent
model. Another new feature is that the third model parameter is obtained by
iteration, is linear, but varies per heartbeat.
Aortic pressure-area relations were obtained from
unrelated, in vitro measurements on segments of human
thoracic aorta. The arctangent model for the aortic pressure-area relation describes the measurements with only
three parameters, which depend on age and sex (7). Patient age and sex were thus used as inputs. No data fitting
was done, nor was any prior knowledge of a patients
aortic properties available. Still, our results verify the
validity of this approach.
The evaluation of the model flow method was performed in open heart surgical patients. This group probably presents the most taxing environment
for any hemodynamic monitoring
method. Our comparisons
were
done with surgery suspended for the duration of each
quadruple measurement. The small error of the comparison of two cardiac output techniques may be due in part
to this cooperation.
Arteriosclerosis

Because the method depends critically on pressurearea relationships, a logical question is, How does arteriosclerosis affect it? Langewouters et al. (8) analyzed their
aortic measurements
(7) for effects of sclerosis. If an
aorta showed only fatty streaks and spots and fibrous
plaques, it was classified as only intima affected, mild or
atherosclerotic.
If, in addition, atheromas, hemorrhage,
ulcerations, and calcium deposits were found, it was classified as also media affected, severe or arteriosclerotic.
There was about an equal number of aortas in each
group. It appeared that the severely sclerotic group had
larger A,,, values by 9% compared with the pooled average, whereas the mildly affected aortas had smaller areas
by 7%. The position of the inflection point of the pressure-area relation, PO, did not depend on the degree of
sclerosis. Width parameter, P,, was 4 mmHg lower for
the severely than for the mildly affected aortas. As a re-

FROM

2571

PRESSURE

sult, the aortic compliance did not depend on degree of

sclerosis for pressures in the physiological
range.
Apparently, the increased aortic wall stiffness with increasing degree of sclerosis is effectively compensated by
the larger diameter. For the patient group studied we
may expect arteriosclerosis to be present and thus may
find aortic areas larger than average. This we did find
(Table 3, K = 1.07) but could not prove statistically
(SD = 0.24).
Adequacy

of Three-Element Model

The three-element
model was proposed in 1930 by
Broemser and Ranke (1,20). Burkhoff et al. (2) state that
the three-element
model provides a reasonable representation of left ventricular afterload for predicting stroke
volume but significantly underestimates
peak aortic flow
in dogs. McDonald
and Nichols (lo), using a pressure
gradient method, similarly found that peak flow is more
difficult to compute. We could not do a peak flow comparison because pulsatile flow was not measured in these
patients. For most applications, however, possibly incorrect peak flows would have little consequence as long as
beat-to-beat
stroke volume and cardiac output are
correct.
Sensitivity Analysis

The arctangent parameters PO and P, regress tightly

on age (see Fig. 2) but not without scatter. This scatter
can be interpreted as some patients having apparently
older aortas for their age and some having younger aortas for their age. The possible effect this has on model
flow-computed cardiac output was studied with our sensitivity analysis. We found that varying the patient age
t10 yr hardly affected the error figures. We conclude
that any effect of difference between calendar age and
physiological
age on aortic mechanical
properties is
small and does not affect significantly
the precision of
the model flow method.
During surgery it is often difficult to monitor the
height of the arterial pressure transducer with respect to
heart level. Awake subjects may change their position,
anesthetized subjects may be tilted for surgical reasons,
or the table may be changed in height. This may cause
errors in the pressure level. We assumed that this effect
would be less than t10 mmHg and computed the subsequent model flow error. Table 4 shows that there is little
effect. Relative position of the pressure transducer, thereTABLE

4. Summary of sensitivity analysis

Error

Correct
A - 10
A + 10
P - 10
P + 10

yr
yr
mmHg
mmHg

Offset, l/min

Error

Mean

SD

Min

Max

0.09
0.17
0.07

0.22
0.16
0.29

-0.14
-0.06
-0.36

0.45
0.35
0.47

0.09
0.10

0.15
0.23

-0.12
-0.21

0.34
0.44

Mean

Scatter,

l/min

SD

Min

Max

0.36
0.42
0.38

0.07
0.17
0.06

0.27
0.27
0.31

0.46
0.68
0.50

0.37
0.36

0.11
0.04

0.26
0.33

0.57
0.42

Situations:
correct,
pooled errors of Table 3 given for comparison;
A + 10 yr, all patient
ages were entered
10 yr too young or too old; P +
10 mmHg, pressure
transducer
simulated
10 mmHg (13 cm) too high or
too low with respect to heart level.

2572

AORTIC

FLOW

COMPUTED

fore, is not critical for a correct flow computation

in these
subjects. We conclude that the flow computation
is robust with respect to age and mean pressure level.
Use of Radial For Aortic Pressure
Because aortic pressure was not available routinely in
our patients, radial pressure had to be used instead. This
pressure wave is distorted with respect to aortic pressure
(11). The effect is that the computed aortic flow wave is
also distorted (see Fig. 3). The error analysis shows that
cardiac output estimates from radial pressure, nevertheless, are rather precise.
Comparison With cZ Method
Pressure pulse analysis has been used to obtain beatto-beat cardiac output monitoring (9). Under stable hemodynamic circumstances these techniques performed
quite well, but when blood pressure, heart rate, peripheral resistance, or vascular tone changed, deviations occurred. An earlier method, the CZ method, tried to correct
for this aberrant behavior (Eqs. 6 and 7), achieving improved precision under varying hemodynamic conditions
and drug regimens (5, 13, 14, 19). Recently published
errors for the CZ method are near the ones reported here
in Table 3; older results indicate larger errors. This
may be due to a less precise reference cardiac output
method (18).
Precision of Thermodilution Estimates
The thermodilution
technique is usually associated
with large errors. Jansen et al. (4) have shown that,
under conditions of mechanical ventilation, these errors
are due in part to flow modulation and follow a systematic pattern. Averaging several randomly injected estimates reduces this error. Averaging a series of phasecontrolled estimations, however, reduces the error much
more. In a previous study (4), averaging four phase-controlled injections reduced the errors from 14 to 3%. Thus
our thermodilution
errors are probably limited to no
more than 5% SD.
Error Analysis
The postcalibration error mean and SD (Table 3) thus
represent the differences between two imprecise techniques. Averaging over a large number of estimations,
thermodilution probably has a mean error near zero (3).
This has not been demonstrated for the phase-controlled
injection technique in patients. We assume, however,
that any mean error is small enough to be ignored. The
model flow mean error, after calibration with thermodilution, is also near zero (Table 3). Thus only the scatter
errors of the thermodilution and model flow methods remain. They are statistically independent errors because
the methods are based on different physical principles.
With certainty we can state, therefore, that the upper

FROM

PRESSURE

limit of scatter error for each technique must be below

the 8% observed in this study. For the phase-controlled
thermodilution technique we estimated an error SD of
maximally 5%. Thus part of the 8% error in the comparison is due to errors in the reference method. We subtracted this error as explained elsewhere (18). The resulting error of the model flow method alone becomes
6%; that of the CZ method, 11%.
This 6% error includes five outlier errors. The outliers
all occurred immediately after cardiopulmonary bypass.
We checked that outliers were not correlated with sodium nitroprusside infusion or with changes in pressure
or pulse rate. It has been reported (12) that radial artery
pressure immediately postbypass can become unreliable
with both mean and pulse pressure in error. We tend to
consider this a probable cause for the outliers. These
outlier errors should not be ignored; after all they are
part of routine clinical procedure. The important fact is
that even the outlier errors are not enormous. The model
flow method apparently uses a model with the nonlinearity control needed for a precise stroke volume computation when hemodynamics change.
Conclusion
The model flow method extracts precise cardiac output
information from arterial pressure, a signal that is often
already recorded for other reasons. The nonlinear threeelement model, representing the three major characteristics of arterial input impedance, allows a precise computation of stroke volume and cardiac output. The
arctangent aortic pressure-area relation is a useful approximation of aortic nonlinearity. Radial artery pressure can be used to compute beat-to-beat values, although aortic pressure provides a visually better model
flow waveform. For highest accuracy and precision a calibration of the model parameters is required.
Address for reprint
requests:
K. H. Wesseling,
strumentation,
Academic
Medical
Centre,
Suite
15, 1105 AZ Amsterdam,
The Netherlands.
Received

27 January

1992; accepted

in final

form

TN0 Biomedical
InLO-002
Meibergdreef
2 October

1992.

REFERENCES
1. BROEMSER,
I?, AND 0. F. RANKE. Uber die Messung
des Schlagvolumens des Herzens
auf unblutigem
Weg. 2. Biol. 90: 467~507,193O.
2. RURKHOFF,
D., J. ALEXANDER,
JR., AND J. SCHIPKE. Assessment
of
windkessel
as a model of aortic input impedance.
Am. J. Physiol.
255 (Heart Circ. Physiol. 24): H742H753,
1988.
3. CONWAY J., AND P. LUND-JOHANSEN.
Thermodilution
method for
measuring
cardiac output.
Eur. Heart J. 11, Suppl. I: 26-32, 1990.
4. JANSEN, J. R. C., J. J. SCHREUDER,
J. J. SETTELS, J. J. KLOEK, AND
A. VERSPRILLE.
An adequate
strategy
for the thermodilution
technique in patients
during
mechanical
ventilation.
Intensive Cure
Med. 16: 422-425, 1990.
5. JANSEN, J. R. C., K. H. WESSELING,
J. J. SETTELS, AND J. J.
SCHREUDER.
Continuous
cardiac output
monitoring
by pulse contour during cardiac surgery.
Eur. Heart J. 11, Suppl. I: 26-32,199O.
6. KOUCHOUKOS,
N. T., L. C. SHEPPARD, AND D. A. MCDONALD.
Estimation of stroke volume in the dog by a pulse contour
method. Circ.
Rex 26: 61 l-623, 1970.
7. LANGEWOUTERS,
G. J., K. H. WESSELING,
AND W. J. A. GOEDHARD. The static elastic properties
of 45 human thoracic
and 20
abdominal
aortas in vitro and the parameters
of a new model. J.
Biomech. 17: 425-435, 1984.
8. LANGEWOUTERS,
G. J., K. H. WESSELING,
AND W. J. A. GOED-

AORTIC

9.
10.

11.
12.

FLOW COMPUTED

HARD. Age-related changes in visoelasticity of normal and arteriosclerotic human aortas. In: Biomechanics:
Current
Interdisciplinary
Research, edited by S. M. Perren and E. Schneider. Dordrecht: Nijhoff, 1985.
MCDONALD,
D. A. Blood Flow in Arteries
(2nd ed.). London: Arnold, 1974.
MCDONALD,
D. A., AND W. W. NICHOLS. Cardiac output derived
from the time derivative and phase velocities of the aortic pressure
wave. Med. Biol. Eng. 11: 678-690, 1973.
NICHOLS,
W. W., AND M. F. OROURKE.
McDonalds
Blood Flow in
Arteries
(3rd ed.). London: Arnold, 1990.
PAUCA, A. L., A. S. HUDSPETH,
S. L. WALLENHAUPT,
W. Y.
TUCKER,
N. D. KON, S. A. MILLS, AND A. R. CORDELL. Radial
artery to aorta pressure difference after cardiopulmonary bypass.

Anesthesiology
13. PURSCHKE,

70: 935-941,

1989.

R., P. BRUCKE, AND H. D. SCHULTE. Untersuchung zur

Zuverlassigkeit der Schlagvolumenbestimmung
aus der Aortendruckkurve. II. Langzeitbeobachtung
bei Patienten. Anaesthesist

23: 525-534,
14. SMITH, N.

1974.

T., K. H. WESSELING, J. A. P. WEBER, AND B. DE WIT.

Preliminary evaluation of a pulse contour cardiac output computer
in man. Feasibility of brachial or radial arterial pressures. Proc.

San Diego

Biomed.

Symp.

13: 113-119,

1974.

FROM

2573

PRESSURE

15. SPRANGERS,
R. L. H., K.
IMHOLZ,
AND W. WIELING.

H. WESSELING, A. L. T. IMHOLZ, B. P. M.
Initial blood pressure fall on stand up
and exercise explained by changes in total peripheral resistance. J.

Appl. Physiol.
16. TOOROP, G.

70: 523-530,

1991.

P., N. WESTERHOF, AND G. ELZINGA. Beat-to-beat estimation of peripheral resistance and arterial compliance during
pressure transients. Am. J. Physiol. 252 (Heart Circ. Physiol. 21):

H1275-H1283,
17. WARNER,
H.,
AND E. WOOD.

1987.

H. J. C. SWAN, D. C. CONNOLLY,
R. G. TOMPKINS,
Quantitation of beat-to-beat changes in stroke volume from the aortic pulse contour in man. J. Appl. Physiol. 5: 495-

507, 1953.
18. WESSELING,
K.
J. J. SCHREUDER.

H., B. DE WIT, J. J. SELLS,

J. R. C. JANSEN, AND
Monitoring cardiac output continuously in the
OR and ICU and a note on precision. In: Monitoring:
Konzepte
und
Klinische
Realisation,
edited by A. Versprille. Berlin: Springer,
1982, p. 18-32.
19. WESSELING,
K. H., B. DE WIT, J. A. P. WEBER, AND N. T. SMITH. A
simple device for the continuous measurement of cardiac output.
Its model basis and experimental verification. Adu. Cardiouasc.
Phys. 5: 16-52, 1983.
20. WEITERER,
E, AND
terienpulses.
Berlin:

T. KENNER. Grundlagen
Springer, 1968.

der Dynamik

des Ar-