Background: The aims of this study were to determine the agreement between pulmonary artery thermodilution (PA-TD), transpulmonary thermodilution (TP-TD) and the pulse contour method, and to test the ability of the pulse contour method to track changes in cardiac output.
Methods: Cardiac output was determined twice before cardiac surgery with both PA-TD and TP-TD. The precision (two standard deviations of the difference between repeated measurements) and agreement of the two methods were calculated. Post-operatively, cardiac output was determined with the PA-TD and pulse contour methods, and the bias and limits of agreement were again calculated. Finally, in patients with heart rates below 60 beats/min or a cardiac index of less than 2.5 l/min/m2, atrial pacing was started and the haemodynamic consequences were monitored with the PA-TD and pulse contour methods.
Results: Twenty-five patients were included. The precisions of PA-TD and TP-TD were 0.41 l/min [95% confidence interval (CI), ± 0.07] and 0.48 l/min (95% CI, ± 0.08), respectively. The bias and limits of agreement between PA-TD and TP-TD were – 0.46 l/min (95% CI, ± 0.11) and ± 1.10 l/min (95% CI, ± 0.19), respectively. Post-operatively, the bias and limits of agreement between the PA-TD and pulse contour methods were 0.07 l/min and ± 2.20 l/min, respectively. The changes in cardiac output with atrial pacing were in the same direction and of the same magnitude in 15 of the 16 patients.
Conclusion: The precision of cardiac output measurements with PA-TD and TP-TD was very similar. The transpulmonary method, however, overestimated the cardiac output by 0.46 l/min. Post-operatively, cardiac output measurements with the PA-TD and pulse contour methods did not agree, but the pulse contour method reliably tracked pacing-induced changes in cardiac output.
The estimation of cardiac output by the pulse contour method was re-introduced more than two decades ago by Wesseling et al. (1).
In the mid-1990s, Pulsion Medical Systems developed an algorithm and a monitor (PiCCO) that estimated the left ventricular stroke volume, beat-to-beat, from the pressure curve of the aorta. The PiCCOplus, however, requires a patient-specific in vivo calibration to compensate for the unknown compliance of the aorta. For this purpose, Pulsion Medical Systems elected to use the bolus transpulmonary thermodilution (TP-TD) cardiac output. Iced saline (15 ml) is injected into a central vein and, from the temperature change in the aorta, the cardiac output can be calculated and the system can be calibrated.
Most studies have found, a posteriori, ‘acceptable’ agreement between TP-TD, the pulse contour method and the de facto clinical standard – pulmonary artery thermodilution (PA-TD) (2–9). In none of the studies, however, has the precision of the methods been determined a priori, and therefore the notion that the methods agree and can be used interchangeably is not securely founded. Moreover, the ability of the PiCCO algorithm to track changes, induced or spontaneous, in cardiac output has not been determined unequivocally.
The issues to be presented and discussed in this paper are as follows: (i) the precision of PA-TD and TP-TD cardiac output measurements; (ii) the agreement between PA-TD and TP-TD cardiac output measurements; (iii) the agreement between PA-TD and pulse contour cardiac output measurements; and (iv) a comparison of the changes in cardiac output determined with the PA-TD and pulse contour methods.
The ethics committee of Copenhagen County approved the study, and all patients gave written informed consent.
Twenty-five patients without heart valve pathology and in sinus rhythm, scheduled for adult cardiac surgery [coronary artery bypass grafting (CABG) or off-pump coronary artery bypass (OPCAB)], were included in the study. The patients were anaesthesized with induction doses of fentanyl 5–10 μg/kg, midazolam 2–5 mg and pancuronium 0.1 mg/kg, and anaesthesia was continued with enflurane 0.8–1.2% or sevoflurane 1–2% inspired in oxygen. During extracorporeal circulation, anaesthesia was maintained with propofol 200–300 mg/h. All patients were normothermic and normoventilated (PaCO2 between 4.5 and 6.0 kPa) during the study.
A 7.5F PA-TD catheter (Paceport®, Edwards, Copenhagen, Denmark) was inserted via an internal or external jugular vein and advanced until a typical pulmonary artery pressure contour, measured from the tip of the catheter, was evident. The PA-TD cardiac output was determined by the Siemens Sirecust 1281 computer (Siemens, Ballerup, Denmark).
A 20-cm 5F thermistor-tipped arterial catheter (Pulsion Medical Systems, Munich, Germany) was advanced, at the same time, into the abdominal aorta via one of the femoral arteries. The catheter was connected to a pressure transducer from which the pressure signal was transferred to the PiCCO monitor (Pulsion Medical Systems).
After the induction of anaesthesia and haemodynamic stabilization, the cardiac output was determined twice with both PA-TD and TP-TD to determine the precision of the methods. The results from TP-TD were also used for a patient-specific calibration of the PiCCO monitor.
Post-operatively, the cardiac output was determined simultaneously using the pulse contour method and PA-TD. The four PiCCO readings matching the four thermal indicator injections for PA-TD were averaged to determine the pulse contour cardiac output. In patients with a cardiac index of less than 2.5 l/min/m2 or a heart rate of less than 60 beats/min, epicardial pacing (heart rate, 80 beats/min) from the right atrium was started and the cardiac output was again determined using PA-TD and the pulse contour method. In the present investigation, we have defined an increase/decrease in cardiac output as a change of more than one standard deviation (0.2 l/min) of the difference in replicate cardiac output determinations with PA-TD.
The thermodilution results are the average of four thermal indicator injections. Pre-operatively, 15 ml of iced saline was used to simultaneously calibrate the PiCCO system and to assess cardiac output with PA-TD. As the Sirecust 1281 computer does not have a constant for 15 ml, the results were multiplied by 1.5 to obtain the actual cardiac output. Post-operatively, the volume of thermal indicator injections was 10 ml.
The injections were by hand and always completed within 3 s (10 ml) and 5 s (15 ml). All injections were started as soon as the cardiac output computer indicated that the pulmonary and peripheral artery temperatures were stable (± 0.05 °C), without considering the relationship between the timing of the injection and the respiratory cycle. All curves of changes in temperature in the pulmonary artery or the aorta were inspected for irregularities and accepted/rejected before results were displayed on the monitor. Aborted attempts were replaced by a new injection.
From the data on the pre-operative cardiac output measurements with PA-TD and TP-TD, the precision (two standard deviations of the difference between replicate measurements) of the two methods was determined. Subsequently, PA-TD and TP-TD cardiac outputs were compared with bias and limits of agreement analysis according to Bland and Altman (10).
Post-operative cardiac output determinations with the PA-TD and pulse contour methods were likewise compared according to Bland and Altman (10).
Changes in cardiac output (ΔCO) after the initiation of pacing were analysed using a paired t-test. ΔCO values measured with the PA-TD and pulse contour methods were compared using correlation analysis.
Thirty patients were included in the study. Three were not studied due to malfunction of the PiCCO monitor and, in two patients, it was impossible to insert the femoral artery PiCCO catheter.
The investigation was completed in 20 men and five women (CABG, n = 20; OPCAB, n = 5). The mean age was 62 years (range, 42–79 years), mean weight 84 kg (range, 60–110 kg) and mean height 174 cm (range, 153–183 cm). At study entry, the mean heart rate was 58 beats/min (range, 39–74 beats/min), the mean arterial blood pressure was 71 mmHg (range, 54–115 mmHg), mean pulmonary artery pressure was 15 mmHg (range, 8–25 mmHg), mean central venous pressure was 8 mmHg (range, 1–16 mmHg), mean body temperature was 35.8 °C (range, 34.9–36.7 °C) and mean cardiopulmonary bypass time was 87 min (range, 43–135 min).
Precision of cardiac output measurements: PA-TD
The mean difference between replicate measurements of cardiac output was 0.09 l/min and the precision (two standard deviations of the difference between repeated measurements) was 0.41 l/min [95% confidence interval (CI), ± 0.07]. The difference between two determinations of cardiac output did not change throughout the range of the measurements (Fig. 1). The mean cardiac output was 4.2 l/min (range, 2.9–7.7 l/min) and the coefficient of variation was 5.2%.
Precision of cardiac output measurements: TP-TD
The mean difference between replicate measurements of cardiac output was 0.15 l/min and the precision was 0.48 l/min (95% CI, ± 0.08). The difference between two determinations of cardiac output did not change throughout the range of cardiac output measurements (Fig. 2). The mean cardiac output was 4.6 l/min (range, 3.3–6.9 l/min) and the coefficient of variation was 6.7%.
Agreement between cardiac output determined with PA-TD and TP-TD
Cardiac output measured with TP-TD was significantly higher than that measured with PA-TD. The bias was 0.46 l/min (95% CI, ± 0.11). The limits of agreement were 1.10 l/min (95% CI, ± 0.19) (Fig. 3) and the percentage error was 21.2%.
Agreement between cardiac output determined with the PA-TD and pulse contour methods post-operatively
The mean difference (bias) between the cardiac output measured with the PA-TD and pulse contour methods was 0.07 l/min (95% CI, ± 0.17) with limits of agreement of 2.20 l/min (95% CI, ± 0.38). The mean cardiac output was 4.8 l/min (range, 3.8–6.1 l/min) and the percentage error was 49.8% (Fig. 4).
Changes in cardiac output after atrial pacing
Of the 25 patients included in the study, 16 needed atrial pacing to fulfil pre-specified haemodynamic criteria.
The cardiac output measured with PA-TD increased significantly from a mean value of 4.85 l/min (range, 3.20–6.70 l/min) to 5.50 l/min (range, 4.10–6.98 l/min) with pacing. The cardiac output measured with the pulse contour method likewise increased significantly from 4.78 l/min (range, 2.72–6.81 l/min) to 5.42 l/min (range, 2.39–7.71 l/min) with pacing.
Comparison of the changes in cardiac output measured with the PA-TD and pulse contour methods reveals a significant correlation (P < 0.003) (Fig. 5).
Of the 16 patients who needed atrial pacing, 13 showed increased cardiac output according to both methods, one patient showed decreased cardiac output according to both methods, one patient showed decreased cardiac output measured with PA-TD (– 0.525 l/min) and increased cardiac output measured with the pulse contour method (+ 0.8 l/min), and one patient showed increased cardiac output of more than one standard deviation (0.4 l/min) according to the pulse contour method but less than one standard deviation (0.13 l/min) according to PA-TD (Fig. 5).
PA-TD vs. TP-TD cardiac output
This is the first study to demonstrate, with adequate methods, that the precisions (two standard deviations of the difference between replicate measurements) of cardiac output determinations with PA-TD and TP-TD are equally high.
Method comparison studies along the lines advocated by Bland and Altman (10) are essential in the development/validation of new techniques. Because the PA-TD cardiac output is an important parameter in most of the currently used haemodynamic treatment algorithms, it is of vital importance that new methods are as precise and agree with this de facto standard before they are considered and accepted as suitable substitutes.
When examining two ways of measuring cardiac output according to Bland and Altman, it is first necessary to determine the precision (random error) of the individual methods (10, 11). Without a knowledge of the precision, it is impossible to determine the acceptability of the limits of agreement found when the two techniques are compared. When the precision is unknown, a fair and unbiased assessment of a limit of agreement analysis is difficult and subjective.
In previous studies (2–5, 7, 9) comparing the PA-TD and TP-TD techniques, the limits of agreement varied from 1.2 to 2.3 l/min (median, 1.6 l/min). In one study (4), coefficients of variation were calculated but not used in the statistical analysis. The precisions of the two methods were not reported in any of the studies. Moreover, all the previous studies have used many repeated measurements in a few patients to determine the limits of agreement. This approach results in falsely low limits of agreement (10). Therefore, the conclusions and positive or negative recommendations from these papers should be viewed with caution and scepticism.
The limits of agreement when TP-TD was compared with PA-TD in this investigation were ± 1.1 l/min (95% CI, 0.9–1.3). From the analysis of the random errors (precisions) of the two methods, the limits should only be ± 0.7 l/min if the two methods are equivalent. If, however, the highest estimates of the random errors (0.28 and 0.32 l/min) are used in the calculation of the acceptable limits of agreement, the result is just within the 95% CI of the limits of agreement actually found in the study.
The cardiac output was significantly higher (bias, 0.46 ± 0.11 l/min) when the four thermal indicator injections were used to estimate the cardiac output from the temperature change in the aorta. The other studies of the two techniques (2–5, 7, 9) have reported similar results (0.15–0.62 l/min; median, 0.38 l/min).
From this study and the others mentioned above, it is not possible to judge which method is the most accurate; however, for the reasons alluded to above, it is probably most convenient to modify the PiCCO algorithm to reduce/eliminate the bias.
Taken together, it seems fair to conclude that PA-TD and TP-TD can be used interchangeably if one remembers, mentally, to correct for the higher cardiac output values returned from the PiCCO monitor.
PA-TD vs. pulse contour cardiac output
Post-operatively, the bias between the methods was negligible (0.07 l/min; 95% CI, ± 0.17), but the limits of agreement were very wide (± 2.2 l/min). The precision of the PA-TD method was not determined post-operatively. From previous post-operative investigations, however, it is known to be of the same order of magnitude (12, 13) as that determined pre-operatively in this investigation. The precision of the pulse contour method cannot be measured directly but, by subtracting the variation inherent in the PA-TD method from the total variation, one is left with an estimate of the precision of the pulse contour method of 2.1 l/min. Clearly, the methods are not in agreement and cannot be used interchangeably.
Previous authors (2, 5–9) have found limits of agreement ranging from 1.4 to 2.6 l/min (median, 2.1 l/min). Again, the limits have been determined in experiments in which few patients have been repeatedly tested. As no statistical correction for this has been used, the reported limits of agreement between PA-TD and the pulse contour cardiac output are falsely lowered (10).
A large part of the discrepancy between the two techniques, in our investigation, can be traced to two patients who had very unusual differences in cardiac output values with the two methods (patients 17 and 20; Fig. 4). After the study, these patients were carefully examined. No clinical feature distinguished the two patients from the rest, and a re-calibration of the PiCCO monitor did not eliminate the problem. If the results from these two patients were disregarded, incorrectly we believe, the limits of agreement would have been 1.3 l/min.
As a consequence of this investigation, we suggest that pulse contour cardiac output results should always be confirmed with the TP-TD method before attempting major treatment changes.
Pacing-induced changes in cardiac output and the pulse contour method
Both the PA-TD and pulse contour methods reliably tracked pacing-induced changes in cardiac output (Fig. 5). Of the 16 patients who needed pacing to reach pre-specified haemodynamic goals, 13 increased and one decreased their cardiac output by more than 210 ml (standard deviation determined during replicate PA-TD cardiac output measurements) with both methods. One patient had an increase in cardiac output with both methods, but it was less than 210 ml. In one patient, however, a very discrepant result was seen. The PA-TD cardiac output decreased unexpectedly (0.5 l/min), whereas the pulse contour cardiac output increased (0.8 l/min) as anticipated. We have re-examined the records and it is clear that the unusual result was caused by methodological problems with the PA-TD technique. The coefficient of variation of the four thermal indicator injections before pacing was 0.161. This is three times higher than the coefficients we have seen in studies of more than 100 patients (12, 13). The coefficient of variation after pacing was in the normal range (0.051), and so there is little doubt that the pre-pacing PA-TD cardiac output value is suspect, and the results from this patient should therefore not weaken our conclusion that the pulse contour method is able to track pacing-induced haemodynamic changes dependably. The ability of the PiCCO system to track haemodynamic changes during volume loading has recently been published (14).
Limitations of the study
Only patients in sinus rhythm with normal heart valves and left ventricular function were included in the study. All the patients were haemodynamically stable, normovolaemic and normothermic, and none received vasoactive agents. The patients in this investigation are therefore very different from the critically ill patients who might benefit from a reliable continuous cardiac output monitor. Before the pulse contour method has been validated in the relevant patient categories, it is no more than a promising technology to be cautiously embraced.
The main attraction of the pulse contour method is that it continuously monitors cardiac output with minimal need for operator interventions. In accordance with this, the PiCCO monitor was not calibrated in vivo before the post-operative measurements. A re-calibration could potentially have increased the agreement between the methods, but, in the two patients discussed above, re-calibration had no effect.
PA-TD and TP-TD cardiac output measurements are equally precise in haemodynamically stable patients without heart valve pathology. The agreement between the in vivo-calibrated pulse contour method and the PA-TD method is not convincing. The pulse contour method reliably tracks pacing-induced changes in cardiac output and, as such, can be used to indicate, continuously, the direction of haemodynamic changes. Before major haemodynamic interventions are undertaken, however, we recommend that the pulse contour-derived cardiac output values be checked by TP-TD cardiac output measurements.