1Department of Intensive Care, Leiden University Medical Centre, The Netherlands. 2Department of Cardiac Surgery, San Raffaele Hospital, Milan, Italy and CARIM, University of Maastricht, The Netherlands. 3Department of Anaesthesiology, University of Leuven, Belgium. 4Department of Anaesthesiology, University of California San Diego, CA, USA. 5TNO Biomedical Instrumentation, Academic Medical Centre, Amsterdam*Corresponding author
Financially supported in part by Edwards Co., Anaheim, CA, USA. Dr Wesseling holds a patent on the cardiac output method. TNO has no interest in the cardiac output method.
Accepted for publication: February 14, 2001
![]() |
Abstract |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Br J Anaesth 2001; 87: 21222
Keywords: heart, cardiac output; measurement techniques, thermodilution; arterial pressure; model, computer simulation
![]() |
Introduction |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Many methods have been proposed to do this, including arterial pulse contour analysis;13 transoesophageal,4 transtracheal,5 and intrapulmonary artery Doppler;6 the Fick principle;7 continuous thermodilution;811 and bioimpedance.12 13 To gain widespread acceptance, however, obstacles must be removed such as physiological limitations, limited reliability, cumbersome maintenance, insufficient precision, and slow responsive have to be overcome.
The Modelflow method maybe suitable.1417 This method derives an aortic flow waveform from arterial pressure by simulation of a non-linear three-element aortic input impedance model and integrates stroke volume from the flow waveform. The method is now fully automatic, self-recording and has a fast response, and its precision appears substantially improved compared with earlier methods.2 We studied bias, precision and tracking ability in 54 cardiac surgery patients, in three hospitals, by simultaneous comparison with right heart thermodilution estimates of cardiac output.
![]() |
Materials and methods |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
The anaesthesia regimen differed slightly between the three hospitals. Premedication was with lorazepam (Maastricht, Leuven) or midazolam (San Diego). Anaesthesia was induced with sufentanil and maintained by further continuous infusion of sufentanil (Maastricht), or by injections of midazolam and fentanyl (San Diego), or of etomidate and sufentanil (Leuven) as needed. Complete muscle relaxation was maintained with pancuronium bromide (Maastricht), pipecuronium (San Diego), or vecuronium (Leuven). Patients were ventilated without PEEP, at a rate of approximately 10 bpm. Ventilatory volume and/or frequency were adjusted to maintain between 32 and 42 mm Hg. To control arterial pressure after sternotomy, in some cases during dissection of the internal mammary artery, and after bypass, nitroglycerine or nitroprusside were given. Some of the patients received phenylephrine, dopamine, or dobutamine.
Study plan
Cardiac output was measured continuously from the arterial pressure and thermodilution series were performed at times when identifiable changes in a patients state occurred. Principally, a few minutes after the induction of anaesthesia, immediately after sternotomy, just before and after bypass, after sternal closure, after changes in drug infusion rate, after cardiac pacemaker rate changes, and after the completion of surgery. No measurements were made during bypass because of the absence of arterial pulsations. The number of measurements made in each patient depended on the duration of surgery and on the complexity of the surgical procedure. The series after induction of anaesthesia, however, was always obtained. Measurements were called and executed by an operator after permission from the anesthetist in charge and a statement of no objection from the surgeon. After pressing the start button, the cold liquid injections for the thermodilution estimates and the recording of all data was fully automatic by computer without human intervention thereby removing any operator bias or error. To improve haemodynamic stability, major surgery was suspended during the measurements.
Modelflow method physiologic background
Left ventricular contraction causes inflow of blood into the arterial system, but this inflow is opposed by arterial counter pressure and aortic and peripheral arterial input impedance. The Modelflow method simulates this behaviour. A haemodynamic model of arterial input impedance is used which is known to have realistic properties in computing stroke volume: the extended Windkessel model (Fig. 1).18 19 The model has three principal components: a characteristic impedance which represents the opposition of the aorta to pulsatile inflow, Windkessel compliance which represents the opposition of the aorta to volume increases, and peripheral resistance which represents the opposition of the vascular beds to blood flow. These components are not constant. Impedance and compliance depend on pressure itself,20 and total systemic peripheral resistance depends on many factors including circulatory filling, metabolism, sympathetic tone, and vasoactive drugs.
|
Use of the Modelflow method
Before the start of surgery the patients gender, age, height, and weight are entered into the Modelflow computer. These determine pressure-volume, pressure-compliance, and pressure-characteristic impedance relationships using Langewouters equations, which are population averages. Look-up tables are formed in computer memory for each pressure level from the equations. At each new pressure sample taken at 100 Hz the corresponding values are read from the table and entered into the model. The model is simulated digitally in real-time and supplied with the sampled arterial pressure waveform. The pressure waveform is taken from the monitor in use in the operating room. The simulation result is a sampled continuous aortic flow waveform (Fig. 1). The flow waveform is integrated during arterial systole to deliver stroke volume. Cardiac output is computed for each beat as the product of stroke volume and heart rate. The result is called model cardiac output in the remainder of the paper.
Model total systemic peripheral resistance is obtained as follows. For the first simulated beat, an initial value for peripheral resistance is assumed and mean arterial pressure and cardiac output are computed with this first value in place. The ratio of pressure to cardiac output for this first beat defines a new resistance value, which is used in the model for the next beat, and so forth. Within five beats from the start, the model resistance usually stabilizes to the systemic peripheral resistance value. This self-adaptation scheme remains permanently active so that changes in systemic peripheral resistance that occur are followed by the model. This is possible because systemic peripheral resistance changes slowly, with a time constant which is typically approximately 10 s.
Pressure transducers were clamped to the operating table to keep their correct hydrostatic height with respect to heart level. When height changes occurred they were noted but level correction was not attempted. Continuous flush devices prevented clotting at the catheter tips. The resonant frequency of the system in use at each location was measured in the laboratory and ranged between 15 and 25 Hz. Before each comparison with thermodilution the arterial waveform quality was visually inspected and the catheter was flushed if slow rising upstrokes took more than 100 ms to reach a maximum. The close observation of pressure pulsation quality was facilitated by a static on-screen display similar to the one shown in Figure 1. Damping of the waveform was continuously monitored by software and an alerting message was displayed whenever risetime increased beyond 150 ms. Occasionally, however, a slow rising waveform had physiological causes and led to a false alarm.
Thermodilution method
Thermodilution cardiac output measurements were performed with system controlled by a personal computer, and included an iced injectate container (CO-SET, Baxter-Edwards, Irvine, CA, USA), a proprietary, motor driven injectate syringe, a thermodilution Swan-Ganz catheter (Baxter), and a COM-2 cardiac output computer (Baxter). The start of the ventilatory cycle was read from the ventilator output or, if unavailable, detected from the capnogram waveform. At precisely timed, variable delays from the start of the ventilatory cycle injections of 10 ml of iced glucose solution 5% were triggered automatically. Delays were spread equally over the ventilatory cycle,21 22 each 25% of the cycle for a series of four injections (Maastricht), each 33% for a series of three injections (San Diego and Leuven). The three or four cardiac output determinations were averaged to obtain one single value for average cardiac output in that period. For this technique to work optimally, the haemodynamic state and respiratory rate must be stable during the series.21
Data acquisition
The software we used is an online real-time version of the BEATFAST offline program, called MODELFLO.EXE (TNO, Academic Medical Centre, Amsterdam, The Netherlands) dated May 1997. It automates all actions and records all haemodynamic data relevant to the study. As cardiac output in a patient can be quite variable, it is important to acquire the data from each method simultaneously. Computer storage of the sampled arterial pressure waveform and beat-to-beat derived haemodynamic data for each comparison started one full respiratory cycle before a thermodilution injection, continued for as many respiratory cycles until at least 18 s had passed, and stopped after one additional respiratory cycle. This resulted in an average recording time of 30 s for each single thermodilution measurement. The digital output of the COM-2 device was also stored, including values for cardiac output, blood temperature, injectate temperature, computation constant, and warning and error codes. This procedure was repeated for the remaining injections in a series with a period of five ventilatory cycles allowed between any two measurements. A series of three or four measurements thus took between 150 and 210 s. Although the MODELFLO monitor ran continuously, waveform samples were recorded only during thermodilution series.
Data analysis
To compare model and thermodilution cardiac output, model cardiac output was first averaged over the beats recorded during an injection. Then the three or four model and thermodilution values per series were each averaged to obtain one single data pair (COmf, COtd) for further analysis. Both values estimate the average cardiac output during a series.
The same set of comparison data pairs was analysed twice.
1. In a first analysis, model cardiac output (COmf) was used as is with the model based on the patients gender, age, height, and weight. This is called uncalibrated.
2. In a second analysis model, cardiac output (CO1) was made equal to thermodilution cardiac output in the first series after induction, through multiplication by a patient individual calibration factor,
1=COtd1/COmf1. This reduces the uncertainty in the patients aortic diameter, because of the high standard deviation of the age and gender based population average. As the first data pair in each patient, would, therefore, be defined to have zero difference it was excluded from further analysis.
With the second method, we mainly investigated the ability of the model to track changes in cardiac output.
A trend score was computed for each patient and also for the group of patients. The trend score is derived from the changes in consecutive cardiac output values. If both methods simultaneously indicate a positive trend, the changes compare positively and a positive score is counted. If both show a negative trend, they again compare positively. When the changes in cardiac output are in opposite directions they compare negatively and a negative score is counted (see Fig. 6 for examples). Ideally, only positive scores are present. Separate scores were made for all consecutive changes regardless of size, and also for changes where consecutive thermodilution cardiac output values differed by at least 0.5 litre min1, which is considered clinically relevant.
|
Haemodynamic stability was verified by analysis of mean arterial pressure and heart rate during a thermodilution series. Stability was considered absent if mean arterial pressure and heart rate averaged per injection period deviated more than 5% from their series average.22 Severe, persistent arrhythmias during thermodilution measurement was also considered to be absence of stability. If stability was not present, the series was excluded from further analysis. Periods with balloon pump counterpulsation had to be discarded as the model method did not properly recognize the heart beats.
The range ratio of haemodynamic values during an operation was computed as the ratio of the largest to the smallest value that was measured in a patient. Thus, if a lowest mean arterial pressure of 60 and a highest of 120 mm Hg were measured the range ratio was 120/60=2. For two-pressure ranges one 3090 mm Hg (difference=60, ratio=3) and another 80140 mm Hg (difference=60, ratio 1.75), the first is more important than the second, because of the non-linear properties of the aorta. The ratio was computed for cardiac output, heart rate, mean arterial pressure, and total peripheral resistance as it has been suggested that these parameters may affect the accuracy of methods such as the model method.
Statistical analysis
We gathered data in three clinical centres in three countries with slightly different anaesthetic regimen, in female and male patients, with changing patient condition during the operation. We studied how these factors might affect cardiac output comparison by testing correlations and differences between cardiac output methods. We used the statistical package BMDP version 7, program 1V and 5V (BMDP, Los Angeles, CA, USA). If either variable is not significantly dependent on centre or patient gender, the data can be pooled to get more reliable statistics.
Computation of limits of agreement was the principal method of analysis with differences in data pairs plotted against their average.23 The agreement between model and thermodilution cardiac output is computed as the bias (mean), with limits of agreement computed as bias ±2 SD23 when differences followed normal distributions. Normality was tested with the KolmogorovSmirnov one-sample test. The coefficient of variation was computed as CV=(SD/mean)x100%. Data averages are given as mean (SD). Statistical significance was considered present when P<0.05.
![]() |
Results |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Figure 2 shows an example of a trend graph of model and thermodilution cardiac output of each injection in patient 54. Although substantial changes in cardiac output occurred between the 14 series, the tracking of thermodilution by model cardiac output was close. Individual thermodilution measurements show scatter within some series of four measurements, but no measurement was rejected.
|
|
The KolmogorovSmirnov test for the difference between model and thermodilution cardiac output did not indicate a significant deviation from a normal distribution. No significant differences between clinical centre and between gender were found either before or after once calibrating the model. The data from both patient sexes and from the three clinical centres, could therefore, be pooled. Both methods, however, indicated a highly significant increase in cardiac output towards the end of surgery. This is shown as half hour averages pooled for the group in Figure 3. The difference between the methods is similarly plotted in the bottom trace but shows no significant trend away from zero. Of 436 trends in cardiac output, 354 (81%) were scored in the same direction by both methods. Of 204 trends in thermodilution cardiac output greater than 0.5 litre min1, 199 (97%) were indicated correctly in direction by model cardiac output.
|
|
|
Figure 7 summarizes in four panels the pooled results before and after calibration of the model. The two left hand panels show scatter plots of the 490 (top panels) and 436 series (bottom panels). The points lie closer to the line of identity after calibration. The right hand panels show BlandAltman plots. The 15 largest negative differences, those below the 2 sd line in the bottom right diagram, were obtained in eight patients and were all recorded immediately after bypass. They do not depend on the value of cardiac output.
|
![]() |
Discussion |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
A precise thermodilution method
An important factor in this research was a good reference for comparison. Individual thermodilution cardiac output estimates show substantial scatter in their values even under stable haemodynamic and ventilatory conditions.25 Four quickly repeated thermodilution estimates of cardiac output may differ more than 1 litre min1. In such circumstances, an average of a number of random injections does not allow a true mean cardiac output estimate. When injections are synchronized with ventilation, however, not by a fixed or a random but by a systematically varied phase equally spread over the ventilatory cycle, much closer estimates of true mean cardiac output can be obtained21 by eliminating ventiliation effects. Precise timing of the injections requires a trigger synchronized with ventilation, and computer control. Precisely measured injections over a repeatably short time period requires a motor driven syringe.26 We used a system for thermodilution cardiac output that is simple but not generally available.
Quality control of arterial pressure measurement
A second important factor was adequate control of the quality of measurement of intra-arterial pressure. All pressure transducers were calibrated against the same standard. Zeroing was done semi-automatically and occasionally checked, although never adjusted during a surgery. The pressure pulsation risetime was constantly monitored with a simple computer algorithm. Visual inspection before each thermodilution series was also the rule. Changes in the hydrostatic level of the pressure transducer with respect to the heart affect model cardiac output. Theoretically, when arterial pressure is artificially high because of a low position of the transducer, model cardiac output is slightly but measurably reduced, and vice versa. Tilting of the operating table caused height changes up to ±15 cm, or approximately 10 mm Hg, and which we did not correct. As height changes were in both directions, and as the errors introduced are small, they have little effect on overall statistics in view of the other sources of error.14
Patient selection
A third important factor was patient selection. All patients probably had patent aortic valves and no aortic aneurysms. An aneurysm affects a patients aortic compliance. A patent aortic valve is required for proper model cardiac output computation as the model computes forward flow into the aorta and in regurgitation ignores backward flow. Thus, model cardiac output will be systematically larger than thermodilution which estimates the net flow as forward minus backward flow. We cannot exclude the possibility that some patients had a small undetected valve leakage, or that such leakage occurred at some times during surgery. If so, this will increase the inaccuracy of the comparison.
Haemodynamic stability
Suspension of major surgery during a series was been a fourth important factor. Haemodynamic stability, however, is not a prerequisite for a correct model cardiac output. Changes in cardiac output are as reliably estimated by the model during arrhythmias and various cardiovascular manoeuvres as when heart rate and rhythm and circulatory volume distribution were stable (J.J. van Lieshout, personal communication).
Model calibration
Differences between uncalibrated model and thermodilution cardiac output in individual patients are small in most but substantial and unreliable in some. The standard deviation of the difference, 19%, is similar to that of a single thermodilution estimate,22 to that of bioimpedance,13 or ultrasound Doppler flow velocity in the aorta,16 or continuous thermodilution,10 11 but the differences are largely systematic, as this study shows (Figs 5 and 6). Thus, averaging a number of measurements tends not to improve the precision of model cardiac output much. When interest is in monitoring percentage changes in cardiac output from a control level, calibration is not needed and the fact that model errors are principally systematic is an advantage.16 In other circumstances calibration is necessary. As the trend plot of differences between the methods (Fig. 3) does not show a trend away from zero a single calibration per operation seems adequate. The calibration does not change on the second day of monitoring compared with from the first.17
Pharmacological agents
The administration of fluids, vasoactive drugs, and cardiac stimulants was not changed from normal routine for this study. These will change arterial pressure, heart rate, ejection time, contractility, cardiac output, and peripheral resistance. As a consequence, the variations in heart rate, cardiac output, and peripheral resistance were substantial (Table 1), yet tracking precision of model cardiac output appeared unaffected. Nitroglycerine or sodium nitroprusside were administered principally after sternotomy and after cardiopulmonary bypass. Thus, if these agents affected model error, it should be evident in these parts of the operation. No important trend in the differences away from zero is seen in Figure 3, and is confirmed by the absence of significance in the differences analyses (see Statistics methods).
In another study3 changes in total peripheral resistance were suggested to reduce the tracking ability of a commercial device. The close tracking of our model method during changes in peripheral resistance might have been facilitated by the self-adapting model peripheral resistance, as explained under Methods.
Exclusion of data
We performed 566 series of thermodilution estimates and rejected 76, based on pre-set criteria. This study was set up to investigate bias and precision of the model method. To achieve this goal the precision of the comparisons should not be impaired by a poor reference value or by inadequate arterial pressure recordings.
Single thermodilution estimates are inaccurate as they are strongly affected by baseline temperature changes and the waxing and waning of right heart cardiac output with ventilation.27 Thus, averaging of several random estimates is needed to obtain reasonable accuracy. Our goal for the cardiac output reference was to achieve near 5% precision. Given the precision of single estimates of 17%22 25 28 31 we would have to average some nine estimates as inaccuracy decreases in proportion to the square root of the number. However, using the ventilatory phase-spreading technique, inaccuracy reduces in proportion to the number of estimates and three estimates would improve precision from 17 to 6%.21 22 For the averaging technique to work, the haemodynamic, conditions should be stable during the series. To judge stability we could have used model cardiac output but this method was under test. We, therefore, used mean arterial pressure and heart rate as easily obtainable measures. Between the thermodilution estimates in a series, we allowed each parameter to deviate maximally 5% from the series average. Inspection of the rejected series suggest that the 5% criteria might have been too strict, as we rejected some apparently adequate series in terms of cardiac output comparison. In another study, under different circumstances, the limit was relaxed to 10% deviation, and acceptable results were still obtained.16
The arterial pressure waveform used as input for the haemodynamic model should ideally have been aortic pressure. This waveform differs from radial artery pressure in shape. As radial artery pressure is what is used clinically the model had to accept this. As shown in another study14 the use of radial artery pressure modifies the model flow waveform but by integrating the flow over systole, to obtain stroke volume, waveform purity becomes less relevant. Radial artery pressures in some patients, however, may deteriorate in mean level and pulse amplitude for a period after bypass in comparison to aortic pressures.29 The pressure difference between aorta and radial artery is usually almost negligible at 02 mm Hg. After bypass it can become as great as 10 mm Hg for mean and more than twice that for systolic pressures in six of 38 patients,29 caused by an increase in pressure gradient in the proximal arteries of uncertain cause. With 15 outliers in eight patients all recorded immediately after bypass, this emphasizes importance of a reliable arterial pressure as input for the model method. Due to warming of the patient post bypass, with substantial changes in thermal baseline, thermodilution may also be unreliable in this critical period.
Discussion of errors
In an editorial Gardner30 proposed that objective criteria be established for judging the precision of cardiac output measurement methods. Critchley and Critchley,31 in trying to establish such objective criteria state that: if a new method is to replace an older, established method, the new method should itself have errors not greater than the older method.
Thermodilution is the reference cardiac output method in almost all studies, as in the present one. A single thermodilution estimate of cardiac output has a probable percentage error standard deviation or coefficient of variation, further called error, of 1520%.22 25 28 31 A triplicate, randomly injected thermodilution has an error of 10% as the result of averaging.31 When the new methods error, n, is acceptable if it is the same as that of the method to be replaced, r, the error of the comparison, c, can be computed with Pythagoras law as c2=n2+r2 or c=n2+r2.31 This computation requires the errors to be statistically independent of each other. Usually, however, we have the error of the comparison and an estimate of the error of the reference method. The error of the new method is then computed as n=
c2r2.
Approximately, the error we found between the uncalibrated model and thermodilution cardiac output is 19% (Table 1). Our reference cardiac output used the phase-spreading injection technique, which has only 17/36% error. The conclusion is that uncalibrated model cardiac output has an error of
19262=18%. If a triplicate random thermodilution is to be replaced by another technique, uncalibrated model cardiac output is not the method of choice as it is not sufficiently precise (18 vs 10% required) as a standalone method.
After model calibration, the error of subsequent comparisons decreases to 9% (Table 1). With 6% error of the thermodilution reference, the calibrated model cardiac output has a probable error of 9262=7%. This is almost as good as a triplicate phase-spreading thermodilution (7 vs 6%) and could thus almost replace it.31 It is definitely better than a triplicate random thermodilution (7 vs 10%).
Positioning of the model method
Response time is a variable to be specified for a continuous cardiac output method.30 For the model method it is the duration of one beat, and even after eight-beat averaging, response time is measured in seconds, not minutes.9
Invasiveness is another aspect to be considered.27 Clearly, the model method as studied here, requires an invasive signal: radial artery pressure. Thus, although as a method it is not more invasive, it is not non-invasive. In a remarkable study, however, Hirschl and colleagues15 used non-invasive finger arterial pressure as input to the same model as used in the present study in critically ill patients in an emergency department. They also used the phase-spreading thermodilution technique as a precise reference although no mention is made of rejection of data if haemodynamic conditions were unstable. No mention is made of the possibility of calibration on the first series. Expressed as cardiac index, their results are quite similar to ours before calibration, including a similar bias and a small number of outliers beyond the limits of agreement.
The present study adds to Hirschl and co-workers results the use in cardiac surgery patients; emphasizes the tracking ability of the model method after calibration; indicates the insensitivity of the model method to changes in mean arterial pressure, heart rate, and systemic peripheral resistance as caused by vasoactive agents, cardiac stimulants or improved heart function after surgery; and confirms that differences between model and thermodilution cardiac output before calibration are not related to age, gender, underlying diagnosis, or body mass index. Differences are because of uncertainty about the diameter of the individual patients aorta, not apparently related to any obvious patient characteristic.20 By using invasive arterial pressure in the present study we avoided uncertainty about non-invasively obtained arterial pressure pulsations. Even though mean arterial pressure was measured well, non-invasively, in a similar setting in similar patients,32 this does not automatic guarantee that non-invasive pressure pulsations as input to the model are also correct. In view of the similarity in results between the present and Hirschl and colleagues study, however, we support their comment that differences between model and thermodilution cardiac output before calibration (our emphasis) are substantially caused by the uncertainty in the model calibration for each patient.
In a non-invasive study in young adult healthy volunteers who underwent a laboratory tilt table procedure, model stroke volume with non-invasive finger arterial pressure as input tracked changes in thermodilution to within 10% error during head-up tilt. Head-up tilt and standing induce blood volume shifts in the body with 50 and 30% reductions in stroke volume on average.16 Although not obtained in critically ill patients we interpret these results as confirmation that non-invasive tracking of changes in cardiac output may be valuable in the future.
The model simulation method could be tested for cardiac output tracking in other categories of patients, including children for which little information is yet available, and more extensively than is already done,15 16 with non-invasive finger pressure as input.
Conclusions
In patients without aortic abnormalities, undergoing coronary artery bypass surgery, the continuous monitoring of changes in cardiac output by simulation of a non-linear, self-adapting model of arterial input impedance is reliable and response to changes in cardiac output is almost immediate. After an initial thermodilution calibration for each individual patient, it has near zero bias and, a 7% error, and a precision sufficient to replace subsequent conventional triplicate thermodilution. Close control of the quality of peripheral artery pressure measurement is necessary. In our automated set-up this is facilitated by computer detection of damped waveforms. Vasoactive drugs and cardiac stimulants in the usual doses do not appear to affect the ability to track the changes in cardiac output thus induced.
![]() |
Acknowledgement |
---|
![]() |
References |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
2 Wesseling KH, De Wit B, Weber JAP, Smith NT. A simple device for the continuous measurement of cardiac output. Adv Cardiovasc Phys 1983; 5: 1652
3 Rödig G, Prasser C, Keyl C, Liebold A, Hobbhahn J. Continuous cardiac output measurement: pulse contour analysis vs thermodilution technique in cardiac surgical patients. Br J Anaesth 1999; 82: 52530
4 Shimaoto H, Kito H, Kawazoe K, Fujita T, Shimamoto Y. Transoesophageal Doppler echocardiographic measurement of cardiac output by mitral annulus method. Br Heart J 1992; 68: 5105[Abstract]
5 Siegel LC, Pearl RG. Noninvasive cardiac output measurement: troubled technologies and troubled studies. Anesth Analg 1992; 74: 7902[ISI][Medline]
6 Segal J, Gaudiani V, Nishimura T. Continuous determination of cardiac output using a flow directed Doppler pulmonary artery catheter. J Cardiothorac Vasc Anesth 1992; 5: 30915
7 Doi M, Morita K, Ikeda K. Frequently repeated Fick cardiac output measurements during anesthesia. J Clin Monit 1990; 6: 10712[ISI][Medline]
8 Yelderman ML, Ramsey MA, Quinn MD, Paulsen AW, McKown RC, Gillman PH. Continuous thermodilution cardiac output measurement in intensive care unit patients. J Cardiothorac Vasc Anesth 1992; 6: 2704[Medline]
9 Aranda M, Mihm FG, Garrett S, Mihm MN, Pearl RG. Continuous cardiac output catheters. Delay in in vitro response time after controlled flow changes. Anesthesiology 1998; 89: 15925[ISI][Medline]
10 Mihm FG, Gettinger A, Hanson III CW, et al. A multicenter evaluation of a new continuous cardiac output pulmonary artery catheter system. Crit Care Med 1998; 26: 134650[ISI][Medline]
11 Zöllner C, Polasek J, Kilger E, et al. Evaluation of a new continuous thermodilution cardiac output monitor in cardiac surgical patients: a prospective criterion standard study. Crit Care Med 1999; 27: 2938[ISI][Medline]
12 Shoemaker WC, Wo CCJ, Bishop MH, et al. Multicenter trial of a new thoracic electrical bioimpedance device for cardiac output estimation. Crit Care Med 1994; 22: 190712[ISI][Medline]
13 Haryadi DG, Westenskow DR, Critchley LAH, et al. Evaluation of a new advanced thoracic bioimpedance device for estimation of cardiac output. J Clin Monit 1999; 15: 1318[ISI]
14 Wesseling KH, Jansen JRC, Settels JJ, Schreuder JJ. Computation of aortic flow from pressure in humans using a nonlinear, three-element model. Appl Physiol 1993; 74: 256673[Abstract]
15 Hirschl MM, Binder M, Gwechenberger M, et al. Noninvasive assessment of cardiac output in critically ill patients by analysis of the finger blood pressure waveform. Crit Care Med 1997; 25: 190914[ISI][Medline]
16 Harms MPM, Wesseling KH, Pott F, et al. Continuous stroke volume monitoring by modelling flow from non-invasive measurement of arterial pressure in humans under orthostatic stress. Clin Sci 1999; 97: 291301[ISI][Medline]
17 Jellema WT, Wesseling KH, Groeneveld ABJ, Stoutenbeek CP, Thijs LG, van Lieshout JJ. Continuous cardiac output in septic shock by simulating a model of aortic input impedance. Anesthesiology 1999; 90: 131628
18 Westerhof N, Elzinga G, Sipkema P. An artificial arterial system for pumping hearts. J Appl Physiol 1971; 31: 77681
19 Toorop GP, Westerhof N, Elzinga G. Beat-to-beat estimation of peripheral resistance and arterial compliance during pressure transients. Am J Physiol 1987; 21: H127583
20 Langewouters GJ, Wesseling KH, Goedhard WJA. The static elastic properties of 45 human thoracic and 20 abdominal aortas in vitro and the parameters of a new model. J Biomech 1984; 17: 42535[ISI][Medline]
21 Jansen JRC, Versprille A. Improvement of cardiac output estimation by the thermodilution method during mechanical ventilation. Intensive Care Med 1986; 12: 719[ISI][Medline]
22 Jansen JRC, Schreuder JJ, Settels JJ, Kloek JJ, Versprille A. An adequate strategy for the thermodilution technique in patients during mechanical ventilation. Intensive Care Med 1990; 16: 4225[ISI][Medline]
23 Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 30710[ISI][Medline]
24 Altman DG, Royston JP. The hidden effect of time. Stat Med 1988; 7: 62937[ISI][Medline]
25 Stevens JH, Raffin TA, Mihm FG, Rosenthal MH, Stetz CW. Thermodilution cardiac output measurement. Effect of the respiratory cycle on its reproducibility. JAMA 1985; 253: 22402[Abstract]
26 Nelson LD, Houtchens BA. Automatic vs manual injections for thermodilution cardiac output determinations. Crit Care Med 1982; 10: 1902[ISI][Medline]
27 Popovitch MJ, Hoffman WD. Noninvasive cardiac output monitoring. Crit Care Med 1997; 25: 17834[ISI][Medline]
28 Stetz CW, Miller RG, Kelly GE. Reliability of the thermodilution method in the determination of cardiac output in clinical practice. Am Resp Dis 1982; 125: 10014
29 Pauca AL, Hudspeth AS, Wallenhaupt SL, et al. Radial artery to aorta pressure difference after cardiopulmonary bypass. Anesthesiology 1989; 70; 93541[ISI][Medline]
30 Gardner RM. Continuous cardiac output: how accurate and how timely? Crit Care Med 1998; 26: 13023[ISI][Medline]
31 Critchley LAH, Critchley JAJH. A meta-analysis of studies using bias and precision statistics to compare cardiac output measurement techniques. J Clin Monit 1999; 15: 8591[ISI]
32 Hirschl MM, Binder M, Herkner H, et al. Accuracy and reliability of noninvasive continuous finger blood pressure measurement in critically ill patients. Crit Care Med 1996; 24: 16848[ISI][Medline]