Non-invasive Estimation of Cardiac Output in Mechanically Ventilated

Patients: A Prolonged Expiration Method

STEFANO CECCHINI,1 EMILIANO SCHENA,1 MARIA NOTARO,2 MASSIMILIANO CARASSITI,2

and SERGIO SILVESTRI1

1Unit of Measurements and Biomedical Instrumentation, Center for Integrated Research, Universita Campus Bio-Medico diRoma, Via Alvaro Del Portillo, 21, 00128 Rome, Italy; and 2Unit of Anaesthesia and Intensive Care, Center for Integrated

Research, Universita Campus Bio-Medico di Roma, Via Alvaro Del Portillo, 21, 00128 Rome, Italy

(Received 5 September 2011; accepted 14 February 2012; published online 24 February 2012)

Associate Editor James Tunnell oversaw the review of this article.

Abstract—A non-invasive method for the estimation ofcardiac output in mechanically ventilated patients isdescribed. The method is based on prolonged expiration, andrelies on measurement of gas concentrations and flow rate. Apneumatic system, with an ad hoc designed orifice resistance,has been made and experimentally characterized to adapt thebreathing circuit to this application. Cardiac output iscalculated using two algorithms and the results are comparedwith the ones obtained by thermodilution. To this purpose,we prospectively enrolled twenty mechanically ventilatedpatients, who had undergone cardiac surgery, and bothalgorithms show good correlation with thermodilution(R> 0.8). The application of the first algorithm gave meancardiac output values slightly lower than those obtained bythermodilution (26%), while the application of the secondalgorithm gave higher values (+30%). Difference standarddeviations between paired measurements is 0.72 L min21 forthe first algorithm and 1.07 L min21 for the second one.Standard deviation obtained by the application of the firstalgorithm is slightly lower than those relative to otherminimally invasive techniques. Through prolonged expira-tion, and standardization and automation of the procedureon mechanically ventilated patients, the proposed systemallows to obtain a non-invasive estimation of cardiac output.

Keywords—Cardiacoutput,Fickmethod,Mechanicalventilation.

GLOSSARY OF TERMS

BE base excess, the amount of acid or alkalineeded to titrate 1 L of fully oxygenatedblood to a pH of 7.40, mEq L21

CaCO2 arterial concentration of carbon dioxide,carbon dioxide volume per volume unit ofarterial blood, mL L21

CaO2 arterial oxygen concentration, oxygen vol-ume per volume unit of arterial blood,mL L21

CbCO2 carbon dioxide concentration in blood,carbon dioxide volume per 100 mL ofblood, mL (100 mL)21

CO cardiac output, volume of blood pumpedby one ventricle in unit of time, L min21

CpCO2 carbon dioxide concentration in plasma,carbon dioxide volume per 100 mL ofplasma, mL (100 mL)21

CvCO2 venous concentration of carbon dioxide,carbon dioxide volume per volume unit ofvenous blood, mL L21

CvO2 venous oxygen concentration, oxygen vol-ume per volume unit of venous blood,mL L21

F shunting fraction, ratio between shuntedblood volume (Qs) and stroke volume (Qt)

FECO2 expiratory fraction of carbon dioxide,fraction of carbon dioxide in the expiratorygas, %

FEO2 expiratory fraction of oxygen, fraction ofoxygen in the expiratory gas, %

FICO2 inspiratory fraction of carbon dioxide,fraction of carbon dioxide in the inspira-tory gas, %

FIO2 inspiratory fraction of oxygen, fraction ofoxygen in the inspiratory gas, %

k1, k2, empirical parameters, parameters definedand k3 by empirical relationships

Address correspondence to Sergio Silvestri, Unit of Measure-

ments and Biomedical Instrumentation, Center for Integrated

Research, Universita Campus Bio-Medico di Roma, Via Alvaro Del

Portillo, 21, 00128 Rome, Italy. Electronic mail: s.silvestri@

unicampus.it

Annals of Biomedical Engineering, Vol. 40, No. 8, August 2012 (� 2012) pp. 1777–1789

DOI: 10.1007/s10439-012-0534-3

0090-6964/12/0800-1777/0 � 2012 Biomedical Engineering Society

1777

Hb hemoglobin concentration, mass of hemo-globin per 100 mL of blood, g (100 mL)21

PaO2 arterial partial pressure of oxygen, partialpressure of oxygen in the arterial blood,mmHg

PaCO2 arterial partial pressure of carbon dioxide,partial pressure of carbon dioxide in thearterial blood, mmHg

PACO2 alveolar partial pressure of carbon dioxide,partial pressure of carbon dioxide in thealveolar gas, mmHg

PAO2 alveolar partial pressure of oxygen, partialpressure of oxygen in the alveolar gas,mmHg

PBF pulmonary blood flow, volume of bloodthat actively participates to the gasexchange in unit of time, L min21

pH measure of the acidity or basicity of asolution defined as the negative logarithm(base 10) of the molar concentration ofhydronium ions

pK negative logarithm (base 10) of the bicar-bonate dissociation constant

PvCO2 venous partial pressure of carbon dioxide,partial pressure of carbon dioxide in thevenous blood, mmHg

Qs shunted blood volume, amount of blood,which does not participate to the alveolargas exchange, L

Qt stroke volume, amount of blood pumpedby the left ventricle during heart contrac-tion, L

R instantaneous exchange ratio, instanta-neous ratio between carbon dioxide pro-duction and oxygen consumption

RQ exchange ratio, ratio between carbondioxide production ( _VCO2

) and oxygenuptake ( _VO2

) during the whole respiratory actS solubility of carbon dioxide in blood, slope

of the carbon dioxide dissociation curve inthe blood, mL L21 mmHg21

S* solubility of carbon dioxide in blood, calcu-lated frommeasuredphysiological parameters

s slope, slope of the parabolic curve derivedfrom the quadratic regression of PACO2 asa function of PAO2 registered during aprolonged expiration

SaO2 oxygen saturation, percentage of hemo-globin saturated with oxygen, %

[HCO3] bicarbonate ion concentration, amount ofbicarbonate ion dissolved into plasma,mEq L21

_VCO2carbon dioxide production, amount ofcarbon dioxide produced by the subject inunit of time, mL min21

_VO2oxygen uptake, amount of oxygen con-sumed by the subject in unit of time,mL min21

INTRODUCTION

Cardiac output (CO) is the volume of bloodpumped by one ventricle per unit of time and can bedefined as the product of stroke volume and heart rate.Its value provides an indication of ventricular function,making its monitoring an important component in thehaemodynamic management of both critically illpatients and patients with suspected cardiovasculardisease.7 CO also helps guide therapy, in order tomaintain adequate tissue perfusion in the high-risksurgical patient.29 Although the topic is still debated,27

the Fick method11 and pulmonary artery thermodilu-tion24 have been considered as reference methods or‘‘gold standard’’ for CO measurement. Their maindrawback is that they are both invasive, since theyrequire the use of central venous catheter, and Swan-Ganz catheter, respectively, thus exposing the patientto risk of sepsis, pneumothorax, thrombosis, or pul-monary artery rupture.9

In the last decade, several minimally invasive ornon-invasive techniques for CO monitoring in ICUhave been developed and tested. They include, amongothers: transthoracic bioimpedance, dye-dilution,Doppler ultrasound, arterial pulse contour and acety-lene–helium rebreathing. These methods have notachieved widespread use in clinical practice mainly dueto: high cost of both devices and disposable compo-nents, outcome dependence on the operator experi-ence, non-continuous assessment of CO, and concernsabout accuracy, precision, and reproducibility.25

CO value is highly correlated with pulmonary bloodflow (PBF), which is the volume of blood that activelyparticipates in the gas exchanges per unit of time.Some techniques, derived from the differential appli-cation of the Fick method to the carbon dioxide, are,by nature, suitable to be used in the estimation of PBFon mechanically ventilated patients. They require thefollowing phases: a first phase involves measurementsperformed during the steady state, and a second phasestarts when a sudden perturbation is introduced intothe CO2 elimination process. Several methods havebeen successfully used to induce such perturbation,e.g., changing the minute ventilation12 or, morerecently, the respiratory rate set on the ventilator,26 andeven adding pure CO2 to the inspiratory gas.21 Finally,one of these methods, based on the partial CO2

rebreathing, has led to the development of a com-mercial device (NICO, Novametrix Medical Systems,

CECCHINI et al.1778

Inc.).18 All these techniques require a perturbation ofthe steady state lasting a minimum of 15–20 s.13 Inorder to reduce the duration of the procedure, Gedeonet al.13 used a different approach on mechanicallyventilated patients: after one breath-holding, theyobtained an estimation of the PBF from the analysis ofthe steady state recovery.

Several studies have investigated a prolonged expira-tionmethod,20which allows a non-invasive estimation ofthe artero-venous content of CO2 and consequentlyPBF, according to the Fick method. None of thesestudies have used thermodilution as referencemethod, asthey were applied only on collaborative subjects at restand during exercise: in fact, the invasiveness of thermo-dilution makes it unsuitable when healthy subjects areenrolled. Only one survey, performed on animals, com-pared theprolonged expirationmethodagainst amethodsimilar to thermodilution, the dye-dilution method, andno significant difference was found.23

To our knowledge, this is the first work using theprolonged expiration method to estimate CO onmechanically ventilated patients. To this purpose, anew system has been designed, realized and validatedto modify the standard breathing circuit, making itsuitable for the proposed application. Two differentalgorithms are used for the data processing: the tra-ditional one proposed by Kim et al.20 and another oneproposed by Godfrey.14

For the first time, we compare the prolonged expi-ration technique against thermodilution, which is themost commonly used method in the monitoring ofcritically ill patients in ICU.

Moreover, taking advantage of the comparison withthermodilution, we try to address some issues that havebeen raised over the years about the data-reduction forthe calculation of PBF using prolonged expiration.7,16

These include: the choice of the expiratory waveformspoints effectively referring to alveolar gas fractions, theduration of the expiration, and the choice of the data-reduction procedure. All these aspects affect the accu-racy of the method. The reduction of intra-subjectvariability16,19 is highly influenced by type and value ofpneumatic resistance, used to induce the prolongedexpiration, thus our focus on designing and character-izing an orifice resistance with the aim to improve therepeatability of measurements. Other issues, related topreservation of the patient’s clinical conditions andpotential loss of accuracy, are also taken into account.

In the following section, we report the theoreticalbackground, describing the prolonged expirationmethod and its assumptions. Afterwards, clinical trials,patient population, measurement setup, and protocolare introduced. Finally, a correlation is showed bycomparing the results obtained with the proposedtechnique with the ones obtained with thermodilution.

THEORETICAL BACKGROUND

Two algorithms, based on prolonged expiration, areused in this work to estimate CO value. These derivefrom a modified version of the Fick equation,11 whichstates that the PBF can be calculated as follows:

PBF ¼_VO2

CaO2 � CvO2ð Þ ð1Þ

However, Eq. (1) may also be applied to CO2 massconservation in the following form:

PBF ¼_VCO2

CvCO2 � CaCO2ð Þ ð2Þ

By supposing a linear CO2 dissociation curve,Eq. (2) becomes:

PBF ¼_VCO2

S PvCO2 � PaCO2ð Þ ð3Þ

In this work only Eqs. (2) and (3) are utilized, aschoosing CO2 as the reference gas introduces someadvantages10: (1) when a patient receives high con-centrations of supplemental O2, even a small error inthe measurement of O2 concentration yields pooraccuracy on _VO2

estimation; (2) some oxygen analyzersshow poor accuracy at high O2 concentrations; (3) CO2

is virtually absent in the inspired air, it is more solublethan O2 in blood and its dissociation curve is morelinear than the oxygen one, allowing the use of aconstant value for S28; (4) CO2 is so diffusible that itsend-capillary partial pressure can be considered almostequal to PACO2: this is particularly true during theplateau of the PACO2 curve, when all dead space hasbeen exhaled and only the alveolar gas is sampled.20

Both algorithms investigated in this work estimatethe values of PaCO2 and PvCO2, in the denominatorof Eq. (2), using a prolonged expiration. This approach,described by Kim et al.,20 requires the analysis of theexpired gas content during both normal breathingand prolonged expiration. The estimation of PaCO2

and PvCO2 is possible through the calculation ofinstantaneous exchange ratio (R) with the followingformula:

R ¼ s� FIO2 � s� FICO2

1� FIO2 � s� FICO2ð4Þ

where s is the slope of the parabolic curve obtained fromthe quadratic regression of PACO2 as a function of PAO2

data, registered during a prolonged expiration.20

In Fig. 1 an example of PACO2 vs. PAO2 graph isshown. It has been obtained during a prolonged expira-tion performed by a patient ventilated with FIO2 =

40%: bothmeasured data and parabolic curve fitting arereported.

Non-invasive Estimation of Cardiac Output in ICU 1779

The data reduction procedure consists of two steps:(1) a preliminary parabolic regression, (2) the pointshaving a PACO2 value within the range 0–0.7 mmHgfrom the related points on the first fitted curve areconsidered for the calculation of a further parabolicregression6: this last regression is successively used tocalculate s, and therefore R using Eq. (4). The leastsquare method is applied to calculate all the fittingcurves. The adopted data-reduction procedure rejectsall the points with PACO2 values lower than two-thirdsof the peak value, since these measurements are per-formed on gas coming predominantly from dead space.This criterion, based on experimental observations,allows the best agreement between the prolonged

expiration method and the chosen reference methodfor almost every patient.

During a prolonged expiration, PAO2 diminishes ata fairly constant rate, while PACO2 rises at a decreas-ing rate. According to Kim et al., R linearly diminisheswith the increase of PACO2 (Fig. 2).

From the linear regression between R and PACO2

data, and considering PACO2 and PAO2 instanta-neously equal to PaCO2 and PaO2, respectively, thetwo parameters of interest can be obtained in the fol-lowing two steps: (1) PaCO2 is obtained as the level ofPACO2 corresponding to a value of R equal to themean of exchange ratio (RQ ¼ _VCO2

= _VO2) within the

minute preceding the prolonged expiration; (2) PvCO2

FIGURE 1. PACO2 as a function of PAO2 during a prolonged expiration obtained in a patient ventilated with FIO2 5 40%. Measureddata (d) in blue and parabolic curve fit (—) in red.

FIGURE 2. R as a function of PACO2 during a prolonged expiration obtained in a patient ventilated with FIO2 5 40%. Measureddata (d) in blue and linear fit (—) in red. PaCO2 and PvCO2 are obtained in correspondence of R 5 0.32 and R 5 RQ, respectively.

CECCHINI et al.1780

corresponds to the PACO2 value at which R equals0.32, that is assumed to be the magnitude of theHaldane effect.20 Moreover, if S is assumed constantand equal to 4.7 mL L21 mmHg21, the denominatorof Eq. (3) is determined.

By measurement of _VCO2during normal breathing,

and by knowing the values of PaCO2 and PvCO2

determined as above described, Eq. (3) allows an esti-mation of PBF.

The second algorithm disregards the simplifyinghypothesis of CO2 dissociation curve linearity. There-fore, with reference to Eq. (2), the findings byMcHardy22 and Godfrey14 about CO2 solubility inblood are considered. McHardy rearranged Visser’sequation for the calculation of CO2 concentration inblood (CbCO2):

CbCO2 ¼ CpCO2 � ½1� ðk1 þ k2 þ k3Þ� ð5Þ

where k1, k2, and k3 are expressed by the followingempirical relationships:

k1 ¼ 0:0288Hb ð6Þ

k2 ¼1

2:244� 0:422 SaO2ð7Þ

k3 ¼1

8:74� pHð8Þ

CpCO2 can be calculated using the Henderson–Hasselbach (H–H) equation:

CpCO2 ¼ 2:226 � 0:0307 PaCO2½1þ 10ðpH� pKÞ�ð9Þ

where, using the H–H equation once again, but forplasma bicarbonate14:

pK ¼ pH� log½HCO3�

0:0307 PaCO2ð10Þ

with 2.226 being the conversion factor from mEq L21

to mLCO2 (100 mL)21 and 0.0307 the solubility coef-ficient of CO2 in plasma [mEq (L mmHg)21].

By posing CbCO2 = CaCO2, and the values ofPaCO2 and PvCO2 being determined as previouslydescribed, it is possible to obtain CvCO2 as follows

14:

CvCO2 ¼ CaCO2 � 10S� log

PvCO2PaCO2 � 1

� �ð11Þ

where S* is defined by the following equation:

S� ¼ 1

2:5þ ðBE � 0:0469Þ � 0:0141 � ð15�HbÞ ð12Þ

PBF can, therefore, be expressed by introducingEq. (11) in Eq. (2):

PBF ¼_VCO2

CaCO2 � 10S� log

PvCO2PaCO2 � 2

� � ð13Þ

The values of the parameters SaO2, pH, Hb, [HCO3]and BE are measured by arterial blood-gas analysis.

The two above described methods per se estimatethe non-shunted portion of the PBF, i.e., the fractionof blood taking part to gas exchange. The shuntingfraction, F, is here calculated using FIO2 and PaO2,obtained from arterial blood-gas analysis. Iso-shuntdiagrams are used to obtain F: these diagrams are aseries of continuous curves showing the relation-ship between PaO2 and FIO2 for different shuntingfractions18:

F ¼ Qs

Qtð14Þ

Finally, CO is calculated using the followingequation:

CO ¼ PBF

1� Fð15Þ

MATERIALS AND METHODS

Patient Population

The twenty patients recruited in this prospectivestudy underwent cardiac surgery under general anes-thesia, and had to be mechanically ventilated aftersurgery. We excluded patients who were hemody-namically unstable requiring high doses of vasoactivemedications (>5 lg kg21 min21 of dopamine ordobutamine), fluids or colloidal solutions to maintaintheir pressure, or inspired oxygen concentration higherthan 60%.

Positioning of a Swan-Ganz catheter was considerednecessary to monitor the clinical and therapeuticcourses of the recruited patients. A continuousinfusion of morphine at 1 mg h21, Propofol at1–3 mg kg21 h21, and vecuronium bromide at 10 mgin bolus sedated and paralyzed all the patients whotook part in the study. They were all in reasonablystable conditions; neither spontaneous respiratoryefforts nor spontaneous muscular activity were regis-tered. Patients were all ventilated by Servo-i ventilators(Maquet Gmbh & Co. KG) in Synchronized Inter-mittent Mandatory Volume (SIMV) mode with tidalvolume equal to 6–8 mL kg21; FIO2 (‡40%) andPEEP, remained unchanged during the trial.

This prospective study was approved by the localInstitute Ethical Committee of the University CampusBio-Medico of Rome, and the recruited patients or the

Non-invasive Estimation of Cardiac Output in ICU 1781

closest relatives expressed their informed consent forthe clinical protocol-based treatment and data collec-tion (Prot. N. 19/2011 ComEt CBM).

Protocol and Measurements

Patients were lying supine during the whole periodof study. The beginning of the trials was at least 90 minafter any intervention or perturbation that wouldchange the circulatory or ventilatory state of thepatient. Furthermore, any intervention or perturbationof the system (patient-ventilator) was avoided during thestudy, and none of the patients had any alterations intheir therapy. A Swan-Ganz catheter (131 HVF,Edwards Lifesciences, Inc.) and a radial artery catheter,previously inserted for intra-operative management,were used to monitor all the patients. The Swan-Ganzcatheter was interfaced to the CO module of thepatient’s monitor (MP70 IntelliVue, Philips Healthcare,Inc.). Before the study, an inspection of the pulmonaryartery pressure waveform checked the correct position-ing of the Swan-Ganz catheter and the code for theinjection bolus temperature and the patient’s anthro-pometric characteristics were manually set in thepatient’s monitor. All the patients received infusion ofsaline solution during the study.

The determination of cardiac output by thermodi-lution (COT) was performed ten times: 4 times atsteady condition, 3 times after a sequence of 10prolonged expirations and 3 times at the end of thesession. Injections of 10 mL of a room-temperature(21–24 �C) solution of 0.9% NaCl were given. Injec-tion time was always shorter than 4 s to reduce anyeffect of varying injection rates on calculations. Thesame operator (M.N.) executed all the injections. The

morphology of thermodilution curves was alwaysmonitored to detect artifacts.

At the end of the session, arterial and mixed venousblood samples were taken using syringes (BD PresetCase Becton, Dickinson and Company) for blood-gasanalysis (ABL 700, Radiometer Medical ApS).

In steady state conditions, measurements of _VCO2,

exhaled air-flow, RQ, FEO2, and FECO2 were contin-uously recorded using a metabolic monitor (QuarkRMR, Cosmed srl), which operated by sampling gasfrom the Y-piece of the mechanical ventilator’sbreathing circuit (Fig. 3).

The inspired and expired O2 and CO2 fractions weremeasured by a paramagnetic differential oxygen sensor(range 0–60%, accuracy ±0.02%) and an infraredsensor (range 0–10%, accuracy ±0.02%), respectively.A continuous gas flow was withdrawn from thebreathing circuit in correspondence of the Y-piece(Fig. 3A) by means of a suction pump. In order toequalize the humidity of sample gas to the level ofambient air, a special Nafion tubing (Fig. 3C) was usedfor the gas sampling line. All gas values were correctedto standard temperature and pressure dry conditions(STPD). The air-flow was measured by a 18 mm-tur-bine flowmeter (range 4.8–480 L min21, accuracy±2%) placed at the ventilator outlet (Fig. 3E). Theoxygen and carbon dioxide sensors of the QuarkRMR, and the turbine flowmeter measured the gasfractions and the expiratory flow with a frequency of25 samples per second.

All data were elaborated by a custom madeapplication implemented in LabView� environment(National Instruments Corporation), able to detect therespiratory act and recognize the inspiration and theexpiration phases on the basis of flow measurements

FIGURE 3. Schematic representation of the measurement setup.

CECCHINI et al.1782

and gas concentration trends. It also calculated themean values of _VCO2

and RQ within a time interval ofabout 1 min preceding the prolonged expiration toassess the steady state conditions of the patient’s gasexchange.

Before each individual study period, the oxygen andcarbon dioxide analyzers of the metabolic monitorwere calibrated with a precise mixture containing 5%of CO2 and 15% of O2, according to the manufac-turer’s instruction. The turbine flowmeter was cali-brated once a day using a calibration syringe of 3 L.

The metabolic monitor was previously validatedand tested in vitro and in vivo (on mechanically venti-lated patients) to assess accuracy and reproducibil-ity.4,5 Under a variety of simulated ventilatoryconditions, the average uncertainties of _VCO2

and RQwere about 7 and 4%, respectively,5 and clinical trialsshowed a good accuracy in the measurement of theexpiratory volume and FIO2.

4

Both Kim and Godfrey algorithms for the estima-tion of CO require the assessment of PvCO2 andPaCO2 through the induction of a prolonged expira-tion. This maneuver was executed 20 times for eachpatient: a pause of about 2 min between consecutivemaneuvers allowed the recovering of steady conditionsby the subject and the assessment of the subject’s gasexchange.

In summary, the sequence of trials, as scheduled bythe clinical protocol, was the following: 4 measure-ments by thermodilution, 10 by prolonged expiration,3 by thermodilution, 10 by prolonged expiration, and 3measurements by thermodilution at the end of thesession.

In order to obtain a prolonged expiration, the set-upshowed in Fig. 3 was slightly changed by adding acustom developed element (Fig. 4) in the expiratorybranch.

The element shown in Fig. 4 has 2 branches with apipe diameter of about 12 mm. In one of these, anorifice pneumatic resistance with a diameter of 1.2 mm(Fig. 4B–a) is inserted to increase the expiratory time.When the balloon valve (Fig. 4B–b; 9300 inflatableballoon-type valve, Hans Rudolph, Inc.) is open, themajority of the expired flow goes through it. When thevalve is closed, the whole expiratory flow goes throughthe orifice resistance: this makes the patient expire witha longer time constant.

During the prolonged expiration, the ‘expiratorypause hold’ key of the ventilator was pushed to avoidthe delivery of an inspiratory act and the risk ofvolutrauma and barotrauma for the patient. At the endof the maneuver the valve’s balloon was deflated andthe ventilator unlocked: then an inspiratory act wasdelivered.

Experimental data of CO2 and O2 concentrationswere recorded and processed after the measurementsession thanks to an ad hoc developed LabView�

application, which implements the above describedprocedures for the estimation of the CO. In particular:(1) it converts the gas fractions into partial pressures(PACO2 and PAO2); (2) it compensates for STPDconditions (T = 0 �C, P = 760 mmHg, and no watervapor); (3) it segments the trends; (4) it executes thedata-reduction; and (5) after obtaining the values ofPvCO2 and PaCO2 as described, it calculates theCO using the above mentioned algorithms. Other

FIGURE 4. Custom made 2-branch element for realization of prolonged expiration: in vivo application and positioning (A), detail(B). In B the orifice pneumatic resistance (a) and the balloon-type valve (b) are indicated.

Non-invasive Estimation of Cardiac Output in ICU 1783

parameters were manually set before the execution ofthe software application: PaO2, FIO2, SaO2, pH, Hb,[HCO3], and BE.

Orifice Resistance Characterization

The orifice pneumatic resistance, used in the devel-oped adaptor for prolonged expiration, has beenin vitro characterized by delivering various volumetricflow rate values and recording the pressure drop. Aflow rate controller (Bronkhorst El-Flow, range 0.05–10.00 L min21, accuracy 0.2% of the set-point value)was utilized, and a pressure sensor (163PC01D48,Honeywell, range 21.96 to +11.8 kPa, accuracy±0.15% full scale) measured the pressure drop acrossthe orifice.

The pressure-flow rate relation is reported in Fig. 5.As the mean measured expiratory flow, during the

clinical trials, was about 1.6 L min21, we can considerthat the orifice introduces a mean pneumatic resistanceof about 5 cmH2O L21 min. This value appears toimprove the repeatability of the measurementsaccordingly to the findings of Hlastala et al.,16 whoobtained a higher repeatability than unconstrainedprolonged expiration by placing a resistance of0.33 cmH2O L21 min at the subject’s mouth. A highervalue of resistance could improve this effect.

Statistical Analysis

CO measurements were repeated ten times usingthermodilution method and twenty times with the twonon-invasive approaches (Kim and Godfrey algorithms)

for each patient. In the following, all experimental dataare reported as mean ± the expanded uncertainty,which is calculated by multiplying the standarduncertainty by a coverage factor (CF) of 2.26 forthermodilution and of 2.09 for non-invasiveapproaches. The two CFs are obtained by consideringa Student’s reference distribution with 9 and 19 degreesof freedom for thermodilution and non-invasiveapproaches, respectively, and a level of confidence of95%.

In order to assess the agreement between the twonon-invasive methods and thermodilution, each non-invasive measurement is paired to the set of thermo-dilution closer in time: then, the single CO valueobtained by the prolonged expiration is compared tothe mean value calculated from the set of thermodi-lution, to which it is paired.

Both the comparison between the Kim and Godfreyalgorithms and the thermodilution, and the analysis oftheir agreement are performed using several methods:(1) the analysis of the correlation between pairedmeasurements, evaluated using the mean square erroralgorithm; (2) the Bland–Altman approach2; (3) thepercentage differences (DCO) for each patient. Thesedifferences were calculated as the relative percentageerror between the CO mean values assessed by thenon-invasive methods and those assessed bythermodilution:

DCO ¼ CONI � COT

COT� 100 ð16Þ

where CONI is the CO value estimated with a non-invasive approach (Kim or Godfrey method) and COT

FIGURE 5. Pressure (DP) vs. flow rate (Q) for the orifice resistance used to realize the prolonged expiration.

CECCHINI et al.1784

is the CO value obtained by thermodilution; and (4)Student’s t test used to compare the average of the 10values of COT and the average of the 20 values ofCONI for every patient.

The bias of the non-invasive method is expressed asthe mean difference between paired measurements. Inthe same way, the precision of the non-invasivemethods is expressed as one standard deviation (1 SD)of the differences between paired measurements.

Limits of agreement (LoA) introduced into theBland–Altman plot are the extremes of the interval,which contains 95% of the differences between pairedmeasurements: lower and upper limit of agreement.

Percentage error (E) is calculated from the SD ofagreement and the mean CO value assessed by ther-modilution (COT):

E ð%Þ ¼ 2 � SDCOT

� 100 ð17Þ

All the statistics have been developed in MATLAB�

(MathWorks, Inc.) environment.

RESULTS

The CO values, obtained using the three differentmethods, are reported in Table 1: for each patient,throughout a period of about 45 min, thermodilution(COT) was executed 10 times, and prolonged expira-tion 20 times. Elaborations according to the Kimalgorithm (COK) and the Godfrey algorithm (COG)were executed after the measurement session.

Experimental data show that CO values estimatedusing the Godfrey algorithm are, on average, greaterthan the values obtained using the Kim algorithm(COG >COK for all patients). The Kim algorithmappears to slightly underestimate COT: a difference ofabout 26% occurs and an underestimation lower than215% is reported in 90% of the cases. On the otherhand, the Godfrey algorithm overestimates COT: COG

is greater than COT of about +30%, the percentagedifference is greater than +30% in 11 patients (55% ofthe cases), and+67% in theworst case. Figure 6 reportsthe percentage differences for each patient between COaverage values obtained using Kim and Godfrey algo-rithms, and those measured by thermodilution.

A further analysis to compare the CO values mea-sured using the abovementioned two algorithms hasbeen carried out by implementing the Student’s t test:

TABLE 1. CO measurements for every patient (mean 6expanded uncertainty).

Patient COT (L min21) COK (L min21) COG (L min21)

P1 3.8 ± 0.1 3.7 ± 0.2 5.4 ± 0.3

P2 2.2 ± 0.2 1.7 ± 0.3 2.3 ± 0.3

P3 4.6 ± 1.0 4.5 ± 0.5 6.0 ± 0.6

P4 4.3 ± 1.0 4.0 ± 0.3 5.1 ± 0.3

P5 2.6 ± 0.1 2.4 ± 0.2 3.3 ± 0.3

P6 2.7 ± 0.1 2.8 ± 0.6 3.8 ± 0.8

P7 4.8 ± 1.0 4.2 ± 0.4 6.0 ± 0.6

P8 2.4 ± 0.4 2.5 ± 0.3 3.3 ± 0.4

P9 4.0 ± 0.3 3.5 ± 0.9 4.9 ± 1.0

P10 6.8 ± 3.0 6.6 ± 2.0 8.7 ± 2.0

P11 4.6 ± 0.3 4.3 ± 0.3 6.6 ± 0.5

P12 4.6 ± 1.0 4.6 ± 1.0 6.6 ± 1.0

P13 3.3 ± 1.0 3.1 ± 0.6 4.2 ± 0.6

P14 3.4 ± 0.2 2.8 ± 0.4 3.8 ± 0.4

P15 3.6 ± 0.2 3.6 ± 0.7 5.1 ± 0.9

P16 4.1 ± 0.3 3.5 ± 0.6 5.3 ± 0.8

P17 4.0 ± 0.2 3.8 ± 0.3 4.3 ± 0.4

P18 2.6 ± 0.1 2.6 ± 0.7 3.6 ± 0.8

P19 3.6 ± 0.2 3.8 ± 0.3 4.3 ± 0.4

P20 2.6 ± 0.2 2.6 ± 0.3 4.4 ± 0.4

FIGURE 6. Percentage difference between CO values obtained with different methods, for each patient: COK and COT (red bars),COG and COT (green bars).

Non-invasive Estimation of Cardiac Output in ICU 1785

the values are significantly different (p< 0.05) for onlyone patient.

All values of COK and COG vs. COT are shown inFig. 7: as reported in the ‘‘Statistical Analysis’’ section,each non-invasive estimation is paired with the meanCO value belonging to the set of thermodilutionassessments closer in time.

The trends confirm the slight underestimation by theKim algorithm and the more marked overestimationby the Godfrey algorithm. The linear regression

implemented between COK or COG (dependent vari-able) and COT (independent variable) shows: in thefirst case (COK vs. COT), a slope of the best fitting lineequal to 0.95 (R = 0.82), and in the second one (COG

vs. COT) a slope equal to 1.30 (R = 0.81).Bland–Altman plots are reported to show the rate of

agreement between the methods (Fig. 8).The following table shows the main endpoints of the

comparison between the two non-invasive techniquesand the thermodilution (Table 2).

FIGURE 7. Data correlation: COK vs. COT (left) and COG vs. COT (right).

FIGURE 8. Bland–Altman plots: COK vs. COT (left) and COG vs. COT (right).

CECCHINI et al.1786

DISCUSSION

Several comparative evaluations can be found inliterature between minimally invasive techniques andthermodilution. Peyton and Chong25 analyzed theresults from 47 studies about the comparison betweenpulse contour technique, partial CO2 rebreathing,esophageal Doppler, transthoracic electrical bioim-pedance, and thermodilution. They reported that noneof the four methods showed a percentage differencelower than 30% if compared with thermodilutionvalues (a suggested criterion of acceptability8). Themethod of prolonged expiration has also been com-pared with other methods, although never before withthermodilution: for example, Chen et al.6 and Inmanet al.17 asserted that the Kim algorithm was not sta-tistically different either from the Fick method or frompartial CO2 rebreathing.

In the present study, theKim algorithm shows a goodagreement with the thermodilution (20.21 L min21,26%), which is in line with the outcomes reported byPeyton and Chong.25

On the contrary, the Godfrey algorithm shows asystematic and substantial overestimation of the ther-modilution (+30%).

The Kim method seems to have a better precision,and a percentage error (0.72 L min21, 39%) slightlylower than other minimally invasive techniques (about1.1 L min21, 42%),25 while the precision and percent-age error estimated for the Godfrey method are com-parable to those reported by Peyton et al.

In a previous study, Hlastala et al.16 applied pro-longed expiration to collaborative subjects, and theyreported a mean SD, over 13 patients, of about1.7 L min21. This higher variability, if compared to ourstudy, might be related to the data-reduction proce-dure, as discussed by Christensen and Grønlund,7 to anexpiratory phase variability which is higher in collab-orative subjects than in sedated ones, and to the use of alower pneumatic resistance than the one used in thepresent survey (0.33 vs. 5 cmH2O L21 min), as alsosuggested by Hlastala et al. themselves.

Moreover, cardiac surgery patients generally showmarked haemodynamic and respiratory changes,which might increase the variability of the CO

measurements. This can add to the limitations intrinsicto the method.

CO underestimation obtained with the Kim methodmay be due to two classes of factors. The first are di-rectly related to the approach: (1) systematic overesti-mation of the slope value of the PACO2–R linearrelationship; (2) underestimation of CO2 values mea-sured at the mouth, probably caused by contaminationof gas during expiration through dead space, andstorage of some expired CO2 in the lungs; (3) decreasein venous return to the heart that accompanies theprolonged expiration, as in the Valsalva maneuver; and(4) underestimation of the shunt fraction. The secondgroup of factors is related to the overestimation of COby thermodilution: Botero et al.3 found an overesti-mation greater than 41%, and Bajorat et al.1 of about48% if compared to a method using an aortic flowprobe. Moreover, thermodilution can be affected byinjection time and bolus temperature.

The method introduced by Godfrey, as shown, leadsto an overestimation of the CO value. Since theGodfrey algorithm takes into account exclusivelyhealthy subjects, some parameter values measured inthis study may exceed the normal ranges, and specifi-cally: all the post-surgery ventilated patients enrolledin the trials have high value of SaO2 (approximately100%) and low value of Hb, under 10 g (100 mL)21.This might account for an underestimation of theartero-venous difference of CO2 concentration.

Some limitations and assumptions regarding bothutilized methods must also be addressed.

An underlying assumption, shared with the Fickmethod, and with all those methods not allowing beat-to-beat estimates, including thermodilution, is that theobtained value constitutes the mean value of CO dur-ing the whole measurement period. Moreover, it isassumed that CO is not altered by the prolongedexpiration. Generally speaking, the underlying theoryconsiders that gas exchange in pulmonary capillaries isequal to the one taking place in the alveolar gas. Thisimplies that all CO2 participating in the alveolarexchange is released, without any retention in the lungs,while expiration proceeds and CO2 partial pressureraises.

Another hypothesis is that the alveolar gas equili-brates with the gas in the mixed arterial blood and thatall the dead space is soon exhaled, corresponding tothe linear increase in the partial pressure curves ofCO2.

The method also assumes that the concentrationsand fractions of the expired gas in the curve plateausare equal to those of the alveolar gas, considering thehypothesis of perfect mixing.

A further potential limitation of the Kim algorithmis the difficulty in the assessment of PvCO2 changes

TABLE 2. Agreement between both methods andthermodilution.

Method n

Bias

(L min21)

Precision

(L min21)

LoA

(L min21)

Percentage

error (%)

Kim 400 20.21 0.72 Lower: 21.63

Upper: +1.21

39

Godfrey 400 +1.11 1.07 Lower: 21.00

Upper: +3.22

58

Non-invasive Estimation of Cardiac Output in ICU 1787

during prolonged expiration: these can be caused by aCO2 buildup in the blood, related to re-circulation.

Another potential source of error can be associatedto pulsations in gas flow due to pulsatile blood flow:fast changes in the alveolar gas concentration inducefurther oscillations in O2 and CO2 flow.16 Such oscil-lations cause slight irregularities in alveolar gas pres-sures as a function of time.15

Finally, the prolonged expiration method may notbe suitable for subjects with obstructive pulmonarydisease, since a representative mixed alveolar samplecannot be obtained with any degree of certainty inthese subjects, and in subjects with uneven ventilation–perfusion.

The described method, as executed in this study,results safe thanks to three precautionary measures: (1)the ‘expiratory pause hold’ key of the ventilator waspushed during the maneuver, which resulted com-pletely passive: this avoided any risk of volutrauma andbarotrauma; (2) patients were continuously monitoredby both ventilator and cardiac monitor; (3) the physi-cian never left the patients and was ready to interruptthe maneuver in case of potential hazard.

Potential risks of hypoxia and hypercapnia could berelated to interposing an obstruction in the expiratorybranch of the breathing circuit: a slight hypoxia can becaused by a longer time, without ‘‘fresh gases’’, and aslight hypercapnia by a progressive reduction of CO2

expulsion rate during the prolonged expiration. How-ever, no potentially negative effects have been regis-tered in any patient enrolled in the study: (1)infrequent transitory desaturations or end-tidal CO2

increases went back to normal, on average, within afew breathing acts, as also confirmed by literature,13

without increasing FIO2 or breathing rate, (2) thelowest SaO2 values, after the prolonged expiration,were comprised between 95 and 100%, which is anacceptable range (at FIO2 ‡ 40%), and (3) PaO2 andPaCO2 values, measured at the end of the sessionthrough blood-gas analysis, were always within phys-iological ranges. This confirms that no cumulativeeffects were induced and that the transitory ones fullyrecovered within the pause period of 2 min.

The encouraging results reported in this studyshould be evaluated also considering the advantagesthat the technique could bring, together with theabsence of cannulation and blood sampling: it is simpleto perform, cumbersome equipment is not needed, themeasurements can be repeated at short time intervals,and the time required to perform the maneuver is rel-atively brief. This allows a serial execution of themaneuver on a subject, without requiring an excessiverecovery time, and it makes the method suitable forrelatively unsteady states, as in patients undergoingintensive care.26

The present study is based on a limited set ofobservations and, therefore, caution should be usedwhen considering the general clinical applicability ofthe described method. The results obtained motivatefurther clinical studies and also analytical work tobetter determine the potential of this application inICU for the monitoring of critically ill patients.

The recent developments in the field of mechanicalventilation could allow for the integration of the pro-posed approach into modern artificial ventilators,making non-invasive CO measurements a routine partof post-surgery monitoring for ventilated patients.

In summary, this study describes a single breathmethod for the non-invasive assessment of CO onmechanically ventilated patients. We designed, real-ized, and characterized a system to induce passiveprolonged expirations when connected to the patientcircuit. An in vivo validation of the method has beenperformed on post-surgery mechanically ventilatedpatients receiving an FIO2 (‡40%) higher than air. Themethod showed good agreement with the thermodilu-tion and a precision comparable to those of otherminimally invasive methods. All the issues related tothe connection to mechanical ventilator, the preserva-tion of the patient’s clinical conditions, and thepotential causes of error have been addressed. Thestandardization and automation of the single breathprocedure have also proven feasible in the applicationof the prolonged expiration on mechanically ventilatedpatients, Even though further validation is needed, thisapproach could foster the development of a newmonitoring device for the non-invasive estimation ofCO, which would reduce the risks derived from the useof catheters.

ACKNOWLEDGMENTS

This work has been carried out under the financialsupport of Filas-Regione Lazio in the framework of theITINERIS2 project (CUP code F87G10000060009).The authors gratefully acknowledge Cosmed s.r.l.(Rome, Italy) for the precious support provided, Dr.Antonio Menichetti and Dr. Lucio Di Pirro for theirsupport in the CO assessment by thermodilution.

REFERENCES

1Bajorat, J., R. Hofmockel, D. A. Vagts, M. Janda, B. Pohl,C. Beck, and G. Noeldge-Schomburg. Comparison ofinvasive and less-invasive techniques of cardiac outputmeasurement under different haemodynamic conditions ina pig model. Eur. J. Anaesthesiol. 23:23–30, 2006.

CECCHINI et al.1788

2Bland, J. M., and D. G. Altman. Statistical methods forassessing agreement between two methods of clinicalmeasurement. Lancet 1(8476):307–310, 1986.3Botero, M., D. Kirby, E. B. Lobato, E. D. Staples, andN. Gravenstein. Measurement of cardiac output beforeand after cardio-pulmonary bypass: comparison amongaortic transit-time ultrasound, thermodilution, and non-invasive partial CO2 rebreathing. J. Cardiothorac. Vasc.Anesth. 18:563–572, 2004.4Cecchini, S., E. Schena, R. Cuttone, M. Carassiti, andS. Silvestri. Influence of ventilatory settings on indirect cal-orimetry inmechanically ventilated patients. In: Proceedingsof the 33rd Annual International Conference of the IEEEEngineering in Medicine and Biology Society, EMBS, Art.No. 6090293, 30 Aug–3 Sept 2011, pp. 1245–1248.5Cecchini, S., E. Schena, M. N. Di Sabatino, andS. Silvestri. Uncertainty evaluation of a calibration methodfor metabolic analyzer in mechanical ventilation. In:MeMeA 2011—IEEE International Symposium on Medi-cal Measurements and Applications, Proceedings, Art. No.5966647, 30–31 May 2011, pp. 143–147.6Chen, H., N. P. Silverton, and R. Hainsworth. Evaluationof a method for estimating cardiac output from a singlebreath in humans. J. Appl. Physiol. 53(4):1034–1038, 1982.7Christensen, P., and J. Grønlund. Repeatability of thesingle-breath method for estimation of pulmonary blood-flow: comparison among four data-reduction procedures.Clin. Physiol. 6:221–234, 1986.8Critchley, L. A., and J. A. Critchley. A meta-analysis ofstudies using bias and precision statistics to compare car-diac output measurement techniques. J. Clin. Monit.Comput. 15:85–91, 1999.9Ehlers,K.C.,K.C.Mylrea,C.K.Waterson, andJ.M.Calkins.Cardiac output measurement. A review of current techniquesand research. Ann. Biomed. Eng. 14(3):219–239, 1986.

10Faddy, S. C. Cardiac output, Fick technique for. In:Encyclopedia for Medical Devices and Instrumentation,edited by J. G. Webster. New York: Wiley–Interscience,1988, pp. 12—21.

11Fick, A. Ueber die messung des blutquantums in denhertzventrikeln. Sitzungsberichte der Physiologisch- Mediz-inosche Gesellschaft zu Wuerzburg 2:16, 1870.

12Gedeon, A., L. Forslund, G. Hedenstierna, and E.Romano. A new method for noninvasive bedside determi-nation of pulmonary blood flow. Med. Biol. Eng. Comput.18:411–418, 1980.

13Gedeon, A., P. Krill, and B. Osterlund. Pulmonary bloodflow (cardiac output) and the effective lung volume deter-mined from a short breath hold using the differential Fickmethod. J. Clin. Monit. Comput. 17(5):313–321, 2002.

14Godfrey, S. Manipulation of the indirect Fick principle bya digital computer program for the calculation of exercisephysiology results. Respiration 27(6):513–532, 1970.

15Hlastala, M. P. Model of fluctuating of alveolar gas ex-change during the respiratory cycle. Respir. Physiol.15(2):214–232, 1972.

16Hlastala, M. P., B. Wranne, and C. J. Lenfant. Singlebreath method of measuring cardiac output—a reevalua-tion. J. Appl. Physiol. 33(6):846–848, 1972.

17Inman, M. D., R. L. Hughson, and N. L. Jones. Com-parison of cardiac output during exercise by single-breathand CO2-rebreathing methods. J. Appl. Physiol. 58(4):1372–1377, 1985.

18Jaffe, M. B. Partial CO2 rebreathing cardiac output—operating principles of the NICOTM system. J. Clin. Monit.Comput. 15(6):387–401, 1999.

19Kilty, B. G., C. H. G. Wright, J. M. Calkins, and S. F.Barrett. Automation of a noninvasive cardiac outputcomputation from respiratory gases and flow. Biomed. Sci.Instrum. 44:386–391, 2008.

20Kim, T. S., H. Rahn, and L. E. Fahri. Estimation of truevenous and arterial PCO2

by gas analysis of a single breath.J. Appl. Physiol. 21(4):1338–1344, 1966.

21Linnarson, D., and H. Larsson. Pulmonary blood flowdetermination with selective rebreathing of CO2. Clin.Physiol. 5:39–48, 1985.

22McHardy, G. J. The relationship between the differences inpressure and content of carbon dioxide in arterial andvenous blood. Clin. Sci. 32:299–309, 1967.

23Mohammed, M. M. J., and R. Hainsworth. Evaluationusing dogs of a method for estimating cardiac output froma single breath. J. Appl. Physiol. 50(1):200–202, 1981.

24Norman, R. A., R. W. Johnson, J. E. Messinger, and B.Sohrab. A continuous cardiac output computer based onthermodilution principle. Ann. Biomed. Eng. 17(1):61–73,1989.

25Peyton, P. J., and S. W. Chong. Minimally invasive mea-surement of cardiac output during surgery and critical care:a meta-analysis of accuracy and precision. Anesthesiology113(5):1220–1235, 2010.

26Peyton, P. J., D. Thompson, and P. Junor. Non-invasiveautomated measurement of cardiac output during stablecardiac surgery using a fully integrated differential CO2

Fick method. J. Clin. Monit. Comput. 22(4):285–292, 2008.27Pugsley, J., and A. B. Lerner. Cardiac output monitoring:Is there a gold standard and how do the newer technologiescompare? Semin. Cardiothorac. Vasc. Anesth. 14(4):274–282, 2010.

28Sackner, M. A. Measurement of cardiac output by alveolargas exchange. In: Handbook of Physiology: The Respira-tory System IV, edited by J. Field. Washington, DC:American Physiological Society, 1977, pp. 233–255.

29Sanderland, T. G., and P. do Nascimiento. Maintainingtissue perfusion in high-risk surgical patients: a systematicreview of randomized clinical trials. Anesth. Analg. 112:1384–1391, 2011.

Non-invasive Estimation of Cardiac Output in ICU 1789