Accuracy of feedback-controlled oxygen delivery into a closed anaesthesia circuit for measurement of oxygen consumption

A. W. Schindler*,1, T. W. L. Scheeren1, O. Picker1, M. Doehn2 and J. Tarnow1

1 Department of Anaesthesiology, University-Hospital Düsseldorf, Moorenstrasse 5, D-40225 Düsseldorf, Germany. 2 Department of Anaesthesiology, Merheim Hospital, D-51109 Cologne, Germany

Corresponding author. Email: achim.schindler@uni-duesseldorf.de
{dagger}Results were presented in part at the ‘47. Deutscher Anaesthesiekongress’, Munich 2000, and at the 8th Annual Congress of the European Society of Anaesthesiologists, Gothenburg, Sweden 2001.

Accepted for publication: November 7, 2002


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
Background. Oxygen consumption (V·>O2) is rarely measured during anaesthesia, probably because of technical difficulties. Theoretically, oxygen delivery into a closed anaesthesia circuit (V·>O2-PF; PhysioFlexTM Draeger Medical Company, Germany) should measure V·>O2. We aimed to measure V·>O2-PF in vitro and in vivo.

Methods. Three sets of experiments were performed. V·>O2-PF was assessed with five values of V·>O2 (0–300 ml min–1) simulated by a calibrated lung model (V·>O2-Model) at five values of FIO2 (0.25–0.85). The time taken for V·>O2-PF to respond to changes in V·>O2-Model gave a measure of dynamic performance. In six healthy anaesthetized dogs we compared V·>O2-PF with V·>O2 measured by the Fick method (V·>O2-Fick) during ventilation with nine values of FIO2 (0.21–1.00). V·>O2-PF and V·>O2-Fick were also compared in three dogs when V·>O2 was changed pharmacologically [102 (SD 14), 121 (17) and 200 (57) ml min–1]. In patients during surgery, we measured V·>O2-PF and V·>O2-Fick simultaneously after induction of anaesthesia (n=21) and during surgery (n=17) (FIO2 0.3–0.5).

Results. Compared with V·>O2-Model, V·>O2-PF values varied from time to time so that averaging over 10 min is recommended. Furthermore, at an FIO2 >0.8, V·>O2-PF always overestimated V·>O2. With FIO2 <0.8, averaged V·>O2-PF corresponded to V·>O2-Model and adapted rapidly to changes. Averaged V·>O2-PF also corresponded to V·>O2-Fick in dogs at FIO2 <0.8. V·>O2 measured by the two methods gave similar results when V·>O2 was changed pharmacologically. In contrast, V·>O2-PF systematically overestimated V·>O2-Fick in patients by 52 (SD 40) ml min–1 and this bias increased with smaller arteriovenous differences in oxygen content.

Conclusion. V·>O2-PF measures V·>O2 adequately within specific conditions.

Br J Anaesth 2003; 90: 281–90

Keywords: oxygen, consumption; monitoring, oxygen; equipment, anaesthesia machines


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
Cardiac output (Qt) and oxygen consumption (V·>O2) are almost linearly related both in the awake state1 2 and during anaesthesia.3 Here, V·>O2 is considered the independent variable and determines blood flow. Therefore, if Qt is measured, it can be assessed in terms of its relation with V·>O2. Moreover, an increase in V·>O2 when Qt is augmented has been suggested as a method to indicate inadequate Qt.4 Consequently, the state of the cardiovascular system can be better assessed when V·>O2 is known. Despite the value of knowing oxygen uptake, it is rarely measured during anaesthesia, probably because measurement is difficult. V·>O2 can be calculated using the Fick principle as the product of Qt and the arterio–mixed-venous oxygen content difference C(a–v)O2. This method is not suitable for routine monitoring, since a pulmonary artery catheter is required. Mathematical coupling of a shared variable (Qt) can give a false relationship between Qt and V·>O2.5 Indirect calorimeters measure V·>O2 by analysing the respiratory gases. The indirect calorimeters currently marketed measure V·>O2 during spontaneous breathing or mechanical ventilation with non-rebreathing respirators,6 7 but their operating principles do not work with the circle breathing systems that are used in anaesthetic practice.

In addition to the above methods, V·>O2 can be measured in a closed circuit where the gases consumed are accurately replaced; in such a system oxygen replacement should reflect V·>O2. Automated quantitative oxygen delivery is used in the feedback-controlled closed circle system (PhysioFlexTM, Draeger Medical Company, Lübeck, Germany).8 9 The accuracy of this device for measuring V·>O2 (V·>O2-PF) is not known. We carried out three sets of experiments and compared V·>O2-PF with (i) V·>O2 simulated in a calibrated lung model, (ii) V·>O2 of anaesthetized healthy dogs under standardized experimental conditions, and (iii) V·>O2 of patients during general anaesthesia.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
Measurement of oxygen consumption in a closed circuit
V·>O2 was measured by the closed anaesthesia circuit PhysioFlexTM (Draeger Medical Company, Lübeck, Germany) described previously.8 9 Unless leakage occurs, gases leave the closed circuit only because of uptake by the patient. In the case of oxygen, this uptake reduces FIO2 and circuit volume, which are measured by a paramagnetic sensor and by the end-expiratory position of four membranes that sense the circuit volume.8 9 Exact quantities of oxygen and, if appropriate, a carrier gas (air or nitrous oxide) are replaced by electronic valves. These valves are regulated by a PID (proportional, integral, differential) algorithm so that decreases in FIO2 and circuit volume are corrected immediately. The PhysioFlex continuously estimates these quantities in volumes at standard temperature and pressure (STPD; temperature 273 K, pressure 760 mm Hg, water pressure 0 mm Hg) and stores the values on the internal hard disk.

In our experiments V·>O2-PF was measured at the same time as the other measures of V·>O2 described below. The data were transferred to a personal computer (FlexComTM distribution kit, Physio, Harlem, The Netherlands) for further analysis. Air was used as the carrier gas during all experiments, to detect any leakage. Continual air replacement indicates that nitrogen is being lost from the circuit. If the set value of FIO2 is reduced, this also requires that the nitrogen concentration increases. In such circumstances there is an initial deficit of nitrogen that is met by air replacement. After a short period this ceases and thus can easily be distinguished from leakage. If air replacement occurred when the set FIO2 was not reduced these measurements were excluded from analysis. Before the first experiment, the PhysioFlex was checked and calibrated by the manufacturer.

Simulation of oxygen consumption in a calibrated lung model
A modified dry-seal spirometer (Ohio 840, Ohio Medical Products, Madison, USA) was used to increase the volume of the system by gradually withdrawing the piston of the spirometer using a stepping motor. The nitrogen content of the withdrawn gas was replaced by a continuous flow of nitrogen from a roller pump (type PA-B1, Janke & Kunkel, Staufen, Germany). The oxygen deficit represented the oxygen uptake of the lung model. A rubber bag (1 litre) to accept the tidal volume (350 ml) completed the lung model. The stepping motor and roller pump were controlled by a computer program written by the authors. Volume enlargement and nitrogen flow were checked daily with a calibrated syringe to obtain calibration factors proportional to the respective flows. Since the system was open to the atmosphere between experiments, the calibration factors referred to the ambient conditions, and were corrected to STPD according to the formula:


where Camb, CSTPD, Tamb, PBaro and PH2O are the calibration factors for ambient and STPD conditions, ambient temperature (mercury thermometer), atmospheric pressure (mercury barometer), and absolute humidity (derived from temperature and relative humidity).

This model was set in random order to simulate oxygen uptakes of 50, 100, 200 and 300 ml min–1 (STPD). Each setting was separated by a phase of no uptake which served as a control. The uptake periods were maintained for exactly 20 min. The control phase was maintained until 10 min of steady state were completed, for at least 20 min. Each value of oxygen uptake and one control phase (chosen at random) were analysed. These different uptakes were studied at five different oxygen concentrations (FIO2 0.25, 0.3, 0.5, 0.75, 0.85) while the lung model was ventilated with the PhysioFlex at a respiratory rate of 12 bpm and at zero end-expiratory pressure. Finally, the whole program was run a second time to assure reproducibility. We analysed the last 10 min of each condition so that a total of 50 data sets each containing ten values were obtained: five settings of V·>O2 each measured twice at five values of FIO2.

The time required for V·>O2-PF to adapt to the changes in simulated oxygen uptake was measured as a measure of ‘promptness’. Adaptation was present if both (i) the running average of ten consecutive values showed no further consistent change and (ii) the SD of these ten values was the same as the SD before the change.

Oxygen consumption of anaesthetized dogs
With approval of the District Governmental Animal Research Committee we studied six healthy dogs (foxhounds, 25–33 kg). V·>O2 was measured simultaneously with the closed anaesthesia circuit (V·>O2-PF) and by the Fick principle (V·>O2-FickDOG=QtxC(a–v)O2). Anaesthesia was with either pentobarbital or propofol and sevoflurane (details below). Following intubation of the trachea with a cuffed tube, the lungs were ventilated with the PhysioFlex (respiratory rate 14 bpm, tidal volume adjusted to maintain normocapnia). Qt was measured continuously by ultrasound transit-time flow probes (S-series probes, Transonic, USA) implanted around the trunk of the pulmonary artery. The probes were calibrated after implantation, to compensate for possible errors from misalignment of the probe on the vessel axis.10 To measure Qt, the probe signal was averaged over 5 min. In the middle of this period, arterial and mixed venous blood were sampled simultaneously from indwelling catheters and oxygen content was measured in duplicate by galvanometry (Lex-O2-Con-TL, Lexington Instruments, Waltham, USA). V·>O2-PF was averaged over 10 min starting 5 min before the measurement of Qt so that both measurements ended simultaneously. Oxygen uptake by the Fick principle (V·>O2-FickDOG) was calculated as QtxC(a–v)O2. Since oxygen content measured by the Lex-O2-Con referred to STPD conditions, this uptake compared directly with V·>O2-PF.

Variation of the inspired oxygen fraction
The influence of different oxygen concentrations on the accuracy of V·>O2-PF was tested during pentobarbital anaesthesia (20–25 mg kg–1 i.v. for induction, followed by a continuous infusion of 35–45 mg kg–1 h–1) by adjusting FIO2 to 0.21, 0.25, 0.3, 0.5, 0.75, 0.85, 0.9, 0.95 and 1.0 in random order during mechanical ventilation and to 0.3 during spontaneous ventilation. Each condition was maintained for 20 min before V·>O2-PF and V·>O2-FickDOG were measured as described above.

Changes in oxygen consumption
Three dogs were anaesthetized with propofol 2–3 mg kg–1 i.v. for induction, followed by sevoflurane as specified below. V·>O2 was changed pharmacologically to assess the accuracy of the PhysioFlex. V·>O2 was measured under three conditions: (i) during basic anaesthesia with 1.5 MAC sevoflurane (end tidal concentration 3.0 vol%, when the difference between FI and FE' was <0.2 vol%); (ii) during deep anaesthesia with 2.5 MAC of sevoflurane to decrease V·>O2; iii) during sevoflurane anaesthesia (1.5 MAC) with epinephrine 0.8 µg kg–1 min–1 i.v. to increase V·>O2 about 60% above the basal metabolic rate.11 Each condition of V·>O2 was studied at an FIO2 of 0.3 and 0.75 so that a total of six conditions were studied. Each condition was maintained for 20 min for stabilization followed by further 10 min for measurement of V·>O2-PF and V·>O2-FickDOG (as described above).

Oxygen consumption of anaesthetized patients
With approval of the local ethics committee and after obtaining informed consent we studied 21 patients during major surgery. After intubation of the trachea, anaesthesia was with enflurane and fentanyl. V·>O2 was measured simultaneously using the Fick principle (V·>O2-FickPAT) and the PhysioFlex. Patients with acute lung disease were excluded. All patients had a risk of cardiovascular complications, so that pulmonary arterial and radial arterial catheters were required for clinical purposes.

After induction of anaesthesia the patients’ lungs were ventilated with the PhysioFlex (respiratory rate 10.5 (SD 1.4) bpm, tidal volume 8.6 (1.6) ml kg–1, PEEP 2.3 (1.4) mbar, FIO2 0.39 (0.06) to maintain normocarbia and normoxia). After haemodynamic and respiratory variables had been stable for 10 min, V·>O2-PF was measured for 10 min. Between the 6th and the 10th minute of the measurement, arterial and mixed-venous blood samples were collected simultaneously over 1 min for blood gas analysis (ABL 510, Radiometer Copenhagen, Denmark) and oximetry (OSM3, Radiometer Copenhagen, Denmark). Oxygen content (CO2) for each sample was calculated according to the formula:

CO2=HbxSO2x1.34+0.003xPO2

This formula refers to STPD.

Immediately after blood sampling Qt was measured by thermodilution (pulmonary artery catheter) with iced dextrose 5% 10 ml injected rapidly into the right atrium. Four injections evenly spaced over the respiratory cycle12 were averaged. V·>O2-FickPAT was calculated as QtxC(a–v)O2. V·>O2-PF and V·>O2-FickPAT were measured again later during surgery when no major surgical manipulation was taking place. This was not possible in four patients.

Measurement errors
Our study compared V·>O2-PF with a lung model and with V·>O2 measured using the Fick principle in dogs and in patients. To assess differences between the methods of measurement we estimated possible errors with each of the reference methods. The 95% confidence interval (95% CI) was used to estimate the maximum error, and the law of error propagation was used.

Errors of the lung model method
The calculations were based on volume enlargement and nitrogen flow, which were checked repeatedly against a calibrated syringe. In calibrations before the study the 95% CIs were 1.7% and 1.4% for volume enlargement and nitrogen flow. Repeated calibrations between days revealed maximum errors of 1.6% and 1.6%, respectively. Thus, the model had a total 95% CI of 3.1% for consecutive simulations and 3.2% between days.

Errors of Fick-derived V·>O2 in dogs
The 95% CIs were 6%10 13 for C(a–v)O2 measured by galvanometry and 2% for Qt measured by ultrasound flowmetry (Transonic Operator’s Manual). Thus, a maximum error of 8% resulted for changes of V·>O2-FickDOG. The absolute value of V·>O2 can also be affected by incorrect alignment of the flow probe on the vessel, which then underestimates Qt. We therefore calibrated the flow probes after implantation as described previously.10 This calibration procedure has a 95% CI of 10%, and this additional error has to be included in the absolute error of V·>O2. Hence the 95% CI of V·>O2-FickDOG was 18%.

Errors of Fick-derived V·>O2 in patients
An absolute standard for in vivo measurement of Qt is lacking, so we do not have a 95% CI for Qt data. However, the 95% CI can be derived from the SD (95% CI=2SD). Since the average deviation (AD) is an estimate of uncertainty by repeated measures, calculated in a similar way to the SD,14 we used twice the AD to estimate uncertainty of the Qt measurement. We determined the average AD for each measurement (ADi=|x–xi|) where x and xi are the mean of all measurements at one time point and each individual measurement, respectively. The ADi was then normalized by mean Qt for each measurement to be comparable between individuals (percentage ADi=(ADix100)/x). To obtain one overall AD for all measurements, we then summed all percentage ADi values and divided by (n–1)


Finally, we took 2·AD as an analogue of the 95% CI. This value was 14.4% for Qt measurements during steady state in our series. For C(a–v)O2, a cumulative 95% CI of 8.5% was obtained from routine quality check data. These values are similar to the values reported for thermodilution Qt15 16 and for calculated C(a–v)O2.17 Thus, we assumed a maximum error of 22.9% for V·>O2-FickPAT. This error range indicates the intra-individual random error of Qt measurement but not any possible systematic error.

Data analysis and statistics
Results are given as mean (SD) unless otherwise indicated. Differences between V·>O2-PF and V·>O2-Model were calculated for each of the 50 data sets described above. From each data set we obtained a mean difference to assess the agreement between V·>O2-PF and V·>O2-Model and a SD to assess the variability of the closed circuit measurement. V·>O2-FickDOG and V·>O2-FickPAT were compared with V·>O2-PF according to the Bland–Altman method.18 Percentage differences between V·>O2-PF and the Fick-derived values were calculated as percentage difference = absolute differencex100xmean V·>O2–1. Differences between V·>O2-PF and each of the other three methods, as well as the dependence of these differences on FIO2 and on the value of V·>O2, were tested by ANOVA for repeated measures followed by Fisher’s protected least squares differences post hoc test. P<0.05 was considered statistically significant.


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
Simulation of oxygen consumption in the lung model
A typical lung model experiment (Fig. 1) shows that (i) overall V·>O2-PF agreed with V·>O2-Model, although the single values were scattered, and (ii) V·>O2-PF adapted faster to increases than to decreases in V·>O2-Model. These results are confirmed in Table 1, which summarizes the lung model experiments. After averaging over 10 min V·>O2-PF and V·>O2-Model agreed well and reproducibly, except at an FIO2 of 0.85, where V·>O2-PF significantly exceeded V·>O2-Model (Table 1). The scattering of the single values (Fig. 1) is indicated by the large SD values in Table 1. Averaging values of V·>O2-PF over at least 10 min would reduce this variation. FIO2 affected the agreement between V·>O2-Model and V·>O2-PF but the absolute V·>O2-Model did not. The reproducibility of the results was confirmed by running the experimental program twice (Table 1).



View larger version (15K):
[in this window]
[in a new window]
 
Fig 1 A typical record of a lung model experiment. Simulated oxygen consumption (V·>O2-Model) was estimated by the closed anaesthesia circuit (V·>O2-PF). Simulated V·>O2 (continuous line) was set randomly to different values, each maintained for 20 min. Control periods of no oxygen consumption (V·>O2-Model 0 ml min–1) separated each of the other V·>O2 levels. Single values of V·>O2-PF (circles) showed substantial scatter. V·>O2-PF tracked increases of V·>O2-Model faster than decreases.

 

View this table:
[in this window]
[in a new window]
 
Table 1 Oxygen consumption (V·>O2) measured with a closed anaesthesia circuit (V·>O2-PF) compared with five levels of oxygen consumption simulated by a calibrated lung model (V·>O2-Model) at five oxygen concentrations (FIO2). Experiments were run twice to assure reproducibility. Under each condition, differences between V·>O2-PF and V·>O2-Model were calculated every minute for 10 min of steady state. Mean and SD of the differences were calculated for each 10-min period to indicate agreement between both methods and variability of the closed circuit’s measurement, respectively. Promptness of V·>O2-PF indicates the time required for V·>O2-PF to stabilize after sudden increases and decreases of V·>O2-Model. Data are given as means (SD). *P<0.05
 
As shown in Figure 1, V·>O2-PF adapted to increases in V·>O2-Model within 2–4 min, independently of FIO2, whereas the adaptation to decreases in V·>O2-Model was slower, particularly when FIO2 was greater (Table 1). Oxygen consumption did not affect the promptness of adaptation (data not shown).

Oxygen consumption in anaesthetized dogs
Influence of inspired oxygen fraction
The mean differences between V·>O2-FickDOG and V·>O2-PF increased with FIO2 (Fig. 2). At FIO2<=0.85 both methods differed by less than 15 ml min–1. In contrast, at FIO2 >0.9 the differences between both methods increased to about 40 ml min–1. The relations between V·>O2-PF and V·>O2-FickDOG varied markedly between dogs but were similar in each individual. V·>O2-PF was greater than V·>O2-FickDOG in four dogs and less in two dogs. This caused a small mean difference between methods with broad limits of agreement.



View larger version (21K):
[in this window]
[in a new window]
 
Fig 2 Bland–Altman comparisons of V·>O2 measured with a closed circuit (V·>O2-PF, values averaged over 10 min) and Fick-derived V·>O2 (V·>O2-FickDOG) measured simultaneously in six anaesthetized healthy dogs. Each symbol represents values obtained from the same dog at different levels of FIO2: 0.21–0.3 (upper panel), 0.5–0.85 (middle panel), 0.9–1.0 (lower panel). At lower FIO2 values the methods agreed well, but they differed significantly by about 40 ml min–1 when FIO2 exceeded 0.85. Note that variation within subject was less than variation between subjects.

 
During spontaneous breathing at an FIO2 of 0.3, V·>O2-PF and V·>O2-FickDOG differed by 5 (27) ml min–1. The difference between V·>O2-PF and V·>O2-FickDOG during controlled ventilation at the same FIO2 was 3 (26) ml min–1.

Changes in oxygen consumption
Compared with basic sevoflurane anaesthesia 1.5 MAC, when mean V·>O2 was 121 (17) ml min–1, V·>O2 decreased to 102 (14) ml min–1 during sevoflurane anaesthesia 2.5 MAC, and increased to 200 (57) ml min–1 during epinephrine infusion. V·>O2-PF and V·>O2-FickDOG concordantly followed the changes of V·>O2, at both 30% and 75% inspired oxygen (Fig. 3). The changes in V·>O2 were not identical, and V·>O2-PF varied less than V·>O2-FickDOG, especially with FIO2 0.3.



View larger version (10K):
[in this window]
[in a new window]
 
Fig 3 Linear correlations of V·>O2 measured with PhysioFlex (V·>O2-PF, values were averaged over 10 min) and using the Fick method (V·>O2-FickDOG) measured simultaneously in three anaesthetized dogs. V·>O2 was reduced by increasing the depth of anaesthesia from 1.5 to 2.5 MAC sevoflurane and increased by epinephrine infusion 0.8 µg kg–1 min–1. Each panel represents the results from one dog at 30% (open symbols) and 75% (filled symbols) of inspired oxygen. The broken line in each panel represents the line of identity (slope=1).

 
Oxygen consumption of anaesthetized patients
V·>O2-PF and V·>O2-FickPAT differed significantly by 52 (40) ml min–1 (Fig. 4) or 25 (18)% of V·>O2. The systematic difference correlated indirectly with C(a–v)O2 (r2=0.32, P<0.01), and directly with Qt (r2=0.28, P<0.01) and V·>O2 (r2=0.23, P<0.01) but it did not correlate with FIO2 (r2=0.02, P=0.45). The percentage difference (absolute differencex100xV·>O2–1) correlated indirectly with C(a–v)O2 (r2=0.40, P<0.01), directly with Qt (r2=0.22, P<0.01), but did not correlate with V·>O2 (r2=0.04, P=0.20), and FIO2 (r2=0.01, P=0.65).



View larger version (51K):
[in this window]
[in a new window]
 
Fig 4 Difference between V·>O2 measured with a closed circuit (V·>O2-PF, data averaged over 10 min) and calculated using the Fick method (V·>O2-FickPAT) in patients ventilated mechanically (FIO2 0.3–0.5) during anaesthesia alone (open circles, n=21) and during anaesthesia plus major surgery (filled circles, n=17). Data values from the same individual are connected. The grey area indicates the estimated maximum error of the Fick method, calculated from our own data on reproducibility of cardiac output (thermodilution) and arteriovenous oxygen content difference (calculated from blood gas analysis). V·>O2-PF exceeded V·>O2-FickPAT significantly by 52 (SD 40) ml min–1 (P<0.05), and surgery did not systemically affect V·>O2 or the difference between the two methods.

 
The grey area in Figure 4 indicates the maximum relative error of V·>O2-FickPAT, which was calculated from our own data on reproducibility of the Fick-derived values (see ‘Measurement errors’). This area shows where disagreement between the methods can be explained by imprecision of the Fick method. In 22 of 38 cases, the difference between both methods exceeded this range, indicating a systematic difference between both methods.


    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
We report the first systematic assessment of feedback-controlled oxygen replacement into a closed rebreathing circuit (V·>O2-PF) to measure V·>O2. V·>O2-PF was measured over V·>O2 of 0–300 ml min–1, the range of V·>O2 values usually expected during anaesthesia. We found (i) the PhysioFlexTM accurately measured V·>O2 of a calibrated lung model and of healthy dogs during steady-state conditions, with FIO2 <0.85, and when V·>O2-PF was averaged over about 10 min; (ii) the PhysioFlex followed stepwise increases and decreases in V·>O2-Model (changes plus or minus 50–300 ml min–1) within 2–4 and 5–12 min, respectively; (iii) in patients V·>O2-PF and Fick-derived V·>O2 differed randomly as well as systematically.

General limitations of V·>O2-PF
When FIO2 exceeded 0.85, we found a systematic overestimation of V·>O2-Model and V·>O2-FickDOG by V·>O2-PF. We can only speculate about the causes for this systematic error. In a closed circuit, oxygen uptake causes changes in FIO2 that are less and thus are detected less precisely than when FIO2 is greater.19 The smaller changes in FIO2 cause less precise oxygen substitution, increasing the probability for error. A lack of oxygen replacement can be easily corrected by further oxygen substitution, over-replacement causes loss of gas from the circuit and thus oxygen is lost, which results in a systematic overestimation of real V·>O2 by the closed circuit. Another mechanism may contribute to the systematic error of V·>O2-PF at very high levels of FIO2 because foreign gases (e.g. water vapour, carbon monoxide, methanol) accumulate during closed circuit anaesthesia.20 When their fractions together with the set level of FIO2 exceed 1.0, the set FIO2 cannot be achieved. As a result, the circuit is permanently flushed with fresh gas and thus oxygen is permanently wasted (personal communication with Mr W. Buschke, engineer of Draeger Medical Company, Lübeck, Germany). However, this mechanism applies only to in vivo measurements and not to the lung model, because the foreign gases come from the body and, in the case of water vapour, from carbon dioxide absorption. Taken together, above an FIO2 threshold of about 0.85 V·>O2-PF is not a good measure of V·>O2 because of systematic errors from both imprecise oxygen replacement and accumulation of foreign gases. However, we can not explain why the systematic error occurs abruptly at about 0.85.

Compared with the lung model, which simulated V·>O2 constantly and precisely (see ‘Measurement errors’), V·>O2-PF showed variation from minute to minute. However, this did not cause a systematic error when values were averaged over 10 min unless FIO2 exceeded 0.85. The scattering is probably caused by some tolerance of the PID algorithm that regulates oxygen replacement. Some tolerance is necessary since there is some delay between oxygen replacement and the detection of its effect—because of gas mixing inside the circuit, or because circuit volume is measured only at the end of expiration. If the regulation were too exact, this delay would cause too much oxygen replacement, which would cause loss of gas from the circuit, and this would again overshoot, leading to an unstable system. Taken together, averaging V·>O2-PF over 10 min is recommended, because short-term changes are caused more by tolerant regulation than by real changes in V·>O2.

Lung model experiments
Since the lung model simulated V·>O2 very precisely (see ‘Measurement errors’), the agreement between V·>O2-PF and V·>O2-Model is an important finding of our study. This model of oxygen uptake is based on a nitrogen dilution technique, which predictably and accurately simulates oxygen consumption during ventilation of a mechanical lung model with a non-rebreathing system.6 This technique was modified for the closed circuit of the PhysioFlex because adding nitrogen and the resultant oxygen replacement would have caused a gas overload of the circuit. We used a dry-seal spirometer to compensate for this gas overload, otherwise drainage of gas from the circuit would have resulted, with unpredictable effects on oxygen substitution. Both components of V·>O2-Model (i.e. volume enlargement and nitrogen flow) were repeatedly calibrated before and during the study. These calibrations indicated excellent accuracy and reproducibility of V·>O2-Model with 95% CI of 3.1% and 3.2% within and between days, respectively.

The dynamic performance of V·>O2-PF was assessed by varying V·>O2-Model in steps of plus or minus 50–300 ml min–1. V·>O2-PF adapted to increases and decreases of V·>O2-Model within 2–4 and 5–12 min, respectively. Basing on this rapid adaptation, we consider the dynamic performance of V·>O2-PF fast enough for clinical use, although the changes in V·>O2 are usually smaller in the clinical situation. However, a device that adapts rapidly to supraphysiological changes should also be able to adapt rapidly to physiological changes. The adaptation of V·>O2-PF to a small (physiological) change may be impaired by the inability of the device to detect such a small change. However, this is more a matter of accuracy and variability of the measurement than a matter of dynamic performance. The probable reason for the faster adaptation to increasing changes may be that increases in V·>O2 are answered with a faster action by the device for safety reasons: an increase in V·>O2 causes an oxygen deficit and thus includes the risk of potentially harmful hypoxia. In contrast, a decrease in V·>O2 is not potentially harmful so a slower response is justified.8 Running averages and SDs were used to estimate adaptation, although more conventional measures would have been desirable. However, V·>O2-PF scattered considerably even before V·>O2-Model was changed so that it was impossible to differentiate whether consecutive values differed because of scattering or adaptation. A quantitative measure of response time could suggest precision that was unjustified.

Dog experiments
Oxygen consumption calculated using the Fick equation is not a gold standard for V·>O2-measurements, mainly because the combination of different measurements with respective errors causes substantial error propagation that impairs precision, and also because pulmonary V·>O2 is excluded, which impairs accuracy. Compared with Fick-derived methods, indirect calorimeters provide more accurate and more precise measurements and therefore would be more suitable to evaluate a new method. But since indirect calorimeters and the closed anaesthesia circuit both use all of the respired gas, simultaneous measurements are not possible. Therefore, we had to use the less precise and less accurate Fick principle. To increase precision, Qt and C(a–v)O2 must be measured as accurately as possible. To do so, we measured Qt with chronically implanted and calibrated ultrasound transit-time flow probes10 and C(a–v)O2 directly by galvanometry.13 These methods are more precise than the methods commonly used in patients, (i.e. thermodilution Qt and calculated C(a–v)O2, as shown in ‘Measurement errors’.

Comparison of V·>O2-PF with V·>O2-FickDOG assumes that V·>O2, Qt and C(a–v)O2 are stable.21 This was assured by stability of the continuously measured haemodynamic variables during the dog experiments (data not shown).

The data for each animal are grouped around an individual, constant bias. The most likely reason for a specific bias in a dog is that the flow probe is not perpendicularly aligned with the vessel axis, so that Qt is underestimated. Such an underestimation was corrected for by calibrating the probes repeatedly, but the calibration has a relative error of 10% so that bias of up to 10% of Qt is possible in each dog.10 The constant bias in each dog supports this possibility.

If FIO2 was less than 0.85, the PhysioFlex measured V·>O2 adequately both in vivo and in the dog study. A bias between V·>O2-PF and V·>O2-FickDOG could have been expected, because only the first variable includes pulmonary V·>O2. However, in individuals without pulmonary diseases— dogs22 as well as humans23—pulmonary V·>O2 is less than 5% of whole body V·>O2 and thus within the random measurement error (95% CI) of V·>O2-FickDOG.

Nitrogen washout after an increase in FIO2 may have contributed to the bias between V·>O2-PF and V·>O2-FickDOG at high oxygen concentrations, because nitrogen had to be replaced by oxygen. This effect increases with the magnitude of the FIO2 step. After increasing FIO2 from 0.21 to 1.0 the end-tidal nitrogen concentration was less than 1% after 2–6 min,24 indicating almost complete nitrogen washout. In our study, FIO2 was allowed to equilibrate for 20 min before the measurements were performed so that an effect of nitrogen washout on the bias is unlikely. Furthermore, a bias was noted in each animal at FIO2 >=0.85 despite randomization (i.e. independent of the magnitude of the FIO2 increase).

Changing the dogs’ V·>O2 caused comparable changes of V·>O2-PF and V·>O2-FickDOG, which is not surprising since both methods measure the same physiological variable. Increasing sevoflurane from 1 to 2 MAC decreased V·>O2 by about 10%,3 and an infusion of epinephrine 0.8 µg kg–1 min–1 increases V·>O2 by more than 60%.11 Both methods detected changes in V·>O2 precisely enough to detect an increase in V·>O2 during an induced increase in Qt, which has been suggested to show an oxygen debt and to guide its therapy.4 These results support the results of the lung model experiments, which show that the device detects changes in V·>O2 with acceptable accuracy, but the dynamic performance in vivo remains unknown. V·>O2-PF was not measured as V·>O2 actually changed, but only after steady-state conditions were attained.

Patient study
In patients Qt was measured by thermodilution and C(a–v)O2 was calculated from blood gas analysis, oximetry and haemoglobin concentration. The maximum error of V·>O2-FickPAT was 22.9%. Hence, the poor agreement between V·>O2-PF and V·>O2-FickPAT observed in our study could be caused by the imprecision of V·>O2-FickPAT. Nevertheless, V·>O2-PF significantly exceeded the Fick-derived V·>O2 in patients by 52 (40) ml min–1, which is 25 (18)% of V·>O2 and thus more than the maximum error of V·>O2-FickPAT. The bias exceeded the range of imprecision attributable to V·>O2-FickPAT in 22 of 38 cases (Fig. 4), suggesting a systematic difference between V·>O2-PF and V·>O2-FickPAT. This difference increased with decreasing C(a–v)O2 as indicated by the indirect correlation. A small value of C(a–v)O2 increases the bias between Fick-derived V·>O2 and V·>O2 measured by indirect calorimetry in patients.25 The same may be true for the direct correlation between the systematic difference and thermodilution Qt, since when Qt is great the temperature change in the pulmonary artery is small so that the signal-to-noise ratio and precision decrease. This is substantiated by the finding that when Qt is great, small amounts of indicator reduce the reproducibility of thermodilution Qt.26 A systematic error of thermodilution Qt at high Qt is not reported in the literature, but the opposite is.27 28 The systematic difference correlated weakly with mean V·>O2, which was expected because usually any measurement is characterized by a constant percentage error. However, we do not think that the systematic error of V·>O2-PF increases with increasing V·>O2, because the lung model data do not confirm the correlation between bias and V·>O2, and there was no correlation between the percentage difference and mean V·>O2, indicating that the relative error is independent of V·>O2. In contrast, with C(a–v)O2 and Qt, a significant correlation was found for the percentage differences between V·>O2-PF and V·>O2-FickPAT.

A systematic error between Fick-derived V·>O2 and V·>O2 measured from pulmonary gas exchange can occur when pulmonary V·>O2 exceeds the random error of the methods. Pulmonary V·>O2 may be increased to more than 30% of the body’s V·>O2 in critically ill patients17 29 30 and to about 15% of V·>O2 in dogs with pneumonia of at least one lower lobe.22 However, it is very unlikely that our patients had pulmonary V·>O2 values of this magnitude.


    Conclusion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
The PhysioFlex agreed acceptably with V·>O2 simulated by a calibrated lung model and with Fick-derived V·>O2 of anaesthetized healthy dogs but not with Fick-derived V·>O2 of anaesthetized patients. This lack of agreement in patients probably comes from two factors: the methodological imprecision of the Fick-derived V·>O2 and an error caused by small C(a–v)O2 and high Qt. Of note, a good agreement between methods was obtained only when the FIO2 was below 0.85.


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Conclusion
 References
 
1 Douglas CG, Haldane JS. The regulation of the general circulation rate in man. J Physiol 1922; 56: 69–100

2 Proctor DN, Beck KC, Shen PH, Eickhoff TJ, Halliwill JR, Joyner MJ. Influence of age and gender on cardiac output-V·>O2 relationships during submaximal cycle ergometry. J Appl Physiol 1998; 84: 599–605[Abstract/Free Full Text]

3 Scheeren TWL, Schwarte LA, Arndt JO. Metabolic regulation of cardiac output during inhalation anaesthesia in dogs. Acta Anaesthesiol Scand 1999; 43: 421–30[CrossRef][ISI][Medline]

4 Bihari D, Smithies M, Gimson A, Tinker J. The effects of vasodilation with prostacyclin on oxygen delivery and uptake in critically ill patients. N Engl J Med 1987; 317: 397–403[Abstract]

5 Archie JP Jr. Mathematic coupling of data: a common source of error. Ann Surg 1981; 193: 296–303[ISI][Medline]

6 Takala J, Keinanen O, Vaisanen P, Kari A. Measurement of gas exchange in intensive care: laboratory and clinical validation of a new device. Crit Care Med 1989; 17: 1041–7[ISI][Medline]

7 Makita K, Nunn JF, Royston B. Evaluation of metabolic measuring instruments for use in critically ill patients. Crit Care Med 1990; 18: 638–44[ISI][Medline]

8 Verkaaik AP, Erdmann W. Respiratory diagnostic possibilities during closed circuit anesthesia. Acta Anaesthesiol Belg 1990; 41: 177–88[Medline]

9 Verkaaik AP, Van Dijk G. High flow closed circuit anaesthesia. Anaesth Intens Care 1994; 22: 426–34[ISI][Medline]

10 Picker O, Schindler A, Scheeren TWL. Accuracy and reproducibility of long-term implanted transit-time ultrasound flow probes in dogs. Intensive Care Med 2000; 26: 601–7[CrossRef][ISI][Medline]

11 Scheeren TWL, Arndt JO. Different response of oxygen consumption and cardiac output to various endogenous and synthetic catecholamines in awake dogs. Crit Care Med 2000; 28: 3861–8[CrossRef][ISI][Medline]

12 Vender JS. Clinical utilization of pulmonary artery catheter monitoring. Int Anesthesiol Clin 1993; 31: 57–85

13 Valeri CR, Zaroulis CG, Marchionni L, Patti KJ. A simple method for measuring oxygen content in blood. J Lab Clin Med 1972; 79: 1035–40[ISI][Medline]

14 Lindberg V. Uncertainties and error propagation – part I. http://www.rit.edu/~vwlsps/uncertainties/Uncertaintiesart1.html 2000

15 Chiolero R, Mavrocordatos P, Bracco D, Schutz Y, Cayeux C, Revelly JP. O2 consumption by the Fick method. Methodologic factors. Am J Respir Crit Care Med 1994; 149: 1118–22[Abstract]

16 Heerdt PM, Pond CG, Blessios GA, Rosenbloom M. Comparison of cardiac output measured by intrapulmonary artery Doppler, thermodilution, and electromagnetometry. Ann Thorac Surg 1992; 54: 959–66[Abstract]

17 Jolliet P, Thorens JB, Nicod L, Pichard C, Kyle U, Chevrolet JC. Relationship between pulmonary oxygen consumption, lung inflammation, and calculated venous admixture in patients with acute lung injury. Intensive Care Med 1996; 22: 277–85[CrossRef][ISI][Medline]

18 Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986; 1: 307–10[ISI][Medline]

19 Ultman JS, Bursztein S. Analysis of error in the determination of respiratory gas exchange at varying FIO2. J Appl Physiol 1981; 50: 210–6[Abstract/Free Full Text]

20 Versichelen L, Rolly G, Vermeulen H. Accumulation of foreign gases during closed-system anaesthesia. Br J Anaesth 1996; 76: 668–72[Abstract/Free Full Text]

21 Cohen IL, Roberts KW, Perkins RJ, Feustel PJ, Shah DM. Fick-derived hemodynamics. Oxygen consumption measured directly vs oxygen consumption calculated from CO2 production under steady state and dynamic conditions. Chest 1992; 102: 1124–7[Abstract]

22 Light RB. Intrapulmonary oxygen consumption in experimental pneumococcal pneumonia. J Appl Physiol 1988; 64: 2490–5[Abstract/Free Full Text]

23 Loer SA, Scheeren TWL, Tarnow J. How much oxygen does the human lung consume? Anesthesiology 1997; 86: 532–7[CrossRef][ISI][Medline]

24 Carmichael FJ, Cruise CJ, Crago RR, Paluck S. Preoxygenation: a study of denitrogenation. Anesth Analg 1989; 68: 406–9[ISI][Medline]

25 Walsh TS, Hopton P, Lee A. A comparison between the Fick method and indirect calorimetry for determining oxygen consumption in patients with fulminant hepatic failure. Crit Care Med 1998; 26: 1200–7[CrossRef][ISI][Medline]

26 Renner LE, Morton MJ, Sakuma GY. Indicator amount, temperature, and intrinsic cardiac output affect thermodilution cardiac output accuracy and reproducibility. Crit Care Med 1993; 21: 586–97[ISI][Medline]

27 Norris SL, King EG, Grace M, Weir B. Thermodilution cardiac output – an in vitro model of low flow states. Crit Care Med 1986; 14: 57–9[ISI][Medline]

28 Bourdillon PD, Fineberg N. Comparison of iced and room temperature injectate for thermodilution cardiac output. Cathet Cardiovasc Diagn 1989; 17: 116–20[ISI][Medline]

29 Epstein CD, Peerless JR, Martin JE, Malangoni MA. Comparison of methods of measurements of oxygen consumption in mechanically ventilated patients with multiple trauma: the Fick method versus indirect calorimetry. Crit Care Med 2000; 28: 1363–9[CrossRef][ISI][Medline]

30 Oudemans-van Straaten HM, Scheffer GJ, Eysman L, Wildevuur CR. Oxygen consumption after cardiopulmonary bypass–implications of different measuring methods. Intensive Care Med 1993; 19: 105–10[ISI][Medline]