Recently, a new device has been developed to measure cardiac output noninvasively using partial carbon dioxide (CO(2)) rebreathing. Because this technique uses CO(2) rebreathing, the authors suspected that ventilatory settings, such as tidal volume and ventilatory mode, would affect its accuracy: they conducted this study to investigate which parameters affect the accuracy of the measurement.
The authors enrolled 25 pharmacologically paralyzed adult post-cardiac surgery patients. They applied six ventilatory settings in random order: (1) volume-controlled ventilation with inspired tidal volume (V(T)) of 12 ml/kg; (2) volume-controlled ventilation with V(T) of 6 ml/kg; (3) pressure-controlled ventilation with V(T) of 12 ml/kg; (4) pressure-controlled ventilation with V(T) of 6 ml/kg; (5) inspired oxygen fraction of 1.0; and (6) high positive end-expiratory pressure. Then, they changed the maximum or minimum length of rebreathing loop with V(T) set at 12 ml/kg. After establishing steady-state conditions (15 min), they measured cardiac output using CO(2) rebreathing and thermodilution via a pulmonary artery catheter. Finally, they repeated the measurements during pressure support ventilation, when the patients had restored spontaneous breathing. The correlation between two methods was evaluated with linear regression and Bland-Altman analysis.
When VT was set at 12 ml/kg, cardiac output with the CO(2) rebreathing technique correlated moderately with that measured by thermodilution (y = 1.02x, R = 0.63; bias, 0.28 l/min; limits of agreement, -1.78 to +2.34 l/min), regardless of ventilatory mode, oxygen concentration, or positive end-expiratory pressure. However, at a lower VT of 6 ml/kg, the CO(2) rebreathing technique underestimated cardiac out-put compared with thermodilution (y = 0.70x; R = 0.70; bias, -1.66 l/min; limits of agreement, -3.90 to +0.58 l/min). When the loop was fully retracted, the CO(2) rebreathing technique overestimated cardiac output.
Although cardiac output was underreported at small VT values, cardiac output measured by the CO(2) rebreathing technique correlates fairly with that measured by the thermodilution method.
ALTHOUGH there is controversy over the cost benefit of pulmonary artery catheterization, 1,2cardiac output (CO) is commonly monitored when treating critically ill patients. Recently, a new device, the NICO2system (Novametrix Medical Systems Inc., Wallingford, CT), has been developed to measure CO noninvasively using partial carbon dioxide (CO2) rebreathing. 3,4This device uses periodic partial CO2rebreathing to create a CO2disturbance, which is then used in a differential Fick CO2equation to calculate CO. 3
There have been few studies to investigate how well the results obtained by CO2rebreathing correlate with those obtained by the conventional thermodilution technique. 5–7Furthermore, it remains to be clarified which ventilatory or hemodynamic parameters affect the measured values when the CO2rebreathing technique is used. Because noninvasive CO measurement depends on CO2rebreathing and assumes constant dead space and mixed venous CO2content through the CO2rebreathing procedure, 3,4we suspected that change in ventilatory settings might affect accuracy of the CO measurement. Consequently, we performed a prospective comparative study to evaluate the effects of tidal volume (VT), ventilatory mode, inspired oxygen fraction (Fio2), and positive end-expiratory pressure (PEEP) on the accuracy of the measurement. The NICO2system uses a rebreathing loop in which volume is adjustable according to tidal volume. We suspected that a too-short loop may affect the accuracy due to poor signal-to-noise ratio. Therefore, we investigated, as a factor of the machine itself, the effect of adjusting the length of the rebreathing loop.
Subjects and Methods
The study was approved by the institutional ethics committee of the National Cardiovascular Center (Osaka, Japan), and written informed consent was obtained from each patient.
Patients
Twenty-five adult patients aged 48–78 yr (median, 61 yr) who had undergone cardiac surgery (table 1) were enrolled in this study. Enrollment criteria were (1) insertion of a Swan-Ganz catheter; (2) stable hemodynamics in the intensive care unit; and (3) no leakage around the endotracheal tube. We excluded candidates who (1) had central nervous system disorders; (2) might be adversely affected by induced hypercapnia (risk of severe pulmonary hypertension or increased intracranial pressure); or (3) demonstrated severe tricuspid regurgitation on intraoperative examination of transesophageal echocardiography, which interferes with the accuracy of thermodilution CO measurement. Arterial blood pressure, heart rate, pulmonary artery pressure, central venous pressure, and pulse oximeter signal (PM–1000; Nellcor Inc., Hayward, CA) were continuously monitored in all patients. After waiting 1–3 h for hemodynamics to stabilize after surgery, we started the measurements.
Measurements
We measured CO using two methods. Values for CO derived from a thermodilution technique (COTD) were obtained using a Swan-Ganz catheter (7.5 French; Abbott Laboratories, North Chicago, IL). Injection of 10 ml cold saline (0°C) was performed in triplicate, and the values were averaged. Because the CO measurement varies depending on when in the respiratory cycle the measurement is initiated, 8we standardized the timing of bolus injection after the first half of the expiratory phase. We confirmed the injection timing by watching the waveform of airway pressure versus time on the graphic monitor of a ventilator (Bird Corp., Palm Springs, CA). Noninvasive measurement of CO (CONI) was performed with a NICO2system (software version 3.1, fast mode). This procedure has been presented in detail elsewhere. 3,4Briefly, on a breath-by-breath basis, CO2production (V̇co2) is calculated from the flow and CO2concentration at the airway opening. Then, to establish the relation between V̇co2and CO, the Fick principle is applied as follows:
where CV̄co2and Caco2represent the CO2content in mixed venous and arterial blood, respectively. In the NICO2system, CO2rebreathing is performed for 50 s every 3 min using a disposable sensor (Novametrix Medical Systems). A brief period of CO2rebreathing caused a change in Paco2and a change in V̇co2but little or no change in CV̄co2in anesthetized dogs, 3probably because the quantity of CO2stores in the body is large, and new equilibrium levels are attained after 20–30 min. 9Assuming that CO and CV̄co2remain constant during the CO2rebreathing procedure, the following equation can be substituted for the previous one:
where ΔV̇co2is the change in V̇co2between normal breathing and CO2rebreathing, and ΔCaco2is the change in arterial CO2content. Assuming here that dead space fraction (VD/VT) remains constant during the CO2rebreathing and that ΔCaco2is proportional to changes in arterial carbon dioxide pressure (Paco2) and end-tidal CO2pressure (PETco2), the following equation can be plotted:
where ΔPETco2is the change in PETco2between normal breathing and CO2rebreathing, and S is the slope of the CO2dissociation curve from hemoglobin. The constant S can be expressed as a function of hemoglobin concentration and Paco2as follows 3:
where [Hb] is hemoglobin concentration.
Before the start of the study protocol, the NICO2system was calibrated for zero CO2by opening the system to the atmosphere, according to the manufacturer's instructions. We entered the results of arterial oxygen pressure (Pao2), Paco2, Fio2(0.4–0.7), and hemoglobin concentrations (7.9–11.9 g/dl) into the machine when each patient was under the baseline ventilation. Inclusion of these parameters is used to calculate shunt fraction (Pao2and Fio2), alveolar dead space (Paco2), and the slope of the CO2dissociation curve (hemoglobin). 3,4
Study Protocol
We used Bird 8400 STi ventilators (Bird Corp.). At the time of admission to the intensive care unit, initial ventilatory settings were as follows: synchronized intermittent mandatory ventilation mode; volume-controlled ventilation (VCV); inspired VTof 10 ml/kg; decelerating flow pattern; respiratory rate of 10–12 breaths/min; and inspiratory time of 1.0 s. The Fio2was adjusted by attending physicians to maintain a Pao2greater than 100 mmHg. Baseline PEEP was set at 4 cm H2O in 23 patients; because of hypoxemia, the remaining 2 patients needed PEEP of 6 and 8 cm H2O, respectively. With the patients maintained in the supine position, sedated with continuous intravenous injection of propofol (2–3 mg · kg−1· h−1), and paralyzed with bolus administration of vecuronium bromide (4–8 mg), we started the measurement protocol.
In random order, we applied six ventilatory settings to all of the 25 patients, and then we applied three additional settings in a fixed order (table 2). To test the effects of ventilatory mode and VT, we chose VCV with inspired VTof 12 or 6 ml/kg and pressure-controlled ventilation (PCV) with the same VTsettings. The Fio2and respiratory rate were fixed identical to baseline. The PEEP was also fixed identical to the baseline measurement (4 cm H2O in 23 patients, 6 cm H2O in 1, and 8 cm H2O in 1). The inspiratory time was set to 1.0 s for both VCV and PCV. The level of pressure control was adjusted to obtain the same VTduring VCV. The rebreathing loop was sized according to the manufacturer's instructions recommended for a set VTof 12 ml/kg. To examine the effects of Fio2, we increased the Fio2to 1.0 with VCV and 12 ml/kg VT. To examine the effects of high PEEP, we increased PEEP to 12–15 cm H2O with VCV and 12 ml/kg VT, depending on the patient's hemodynamic stability. The order of these six conditions was randomized. Then, to examine the effects of varying the length of the rebreathing loop, in 17 patients, measurements were performed with the loop maximally expanded (400 ml) or fully retracted (150 ml) while VCV and 12 ml/kg VTwere used. After the measurements were completed, vecuronium infusion was stopped. When the patient recovered stable spontaneous breathing, we switched the ventilatory mode to continuous positive airway pressure of 4 cm H2O plus pressure-support ventilation (PSV) of 10 cm H2O.
After establishing steady-state conditions (approximately 15 min) at each setting, we measured both CONIand COTD. We limited ourselves perform only nine measurements (one measurement for each ventilatory setting) per patient. Arterial blood samples were analyzed with a calibrated blood gas analyzer (ABL 505; Radiometer, Copenhagen, Denmark). Hemodynamic data were also recorded. VD/VTand venous admixture fraction (Q̇S/Q̇T) were calculated using the following equations 10,11:
and
where V̇Eis minute ventilation, Cc′o2is oxygen content at the pulmonary capillary, Cao2is arterial oxygen content, and CV̄o2is mixed venous blood oxygen content. Assuming that pulmonary capillary blood is fully saturated with oxygen and that oxygen content is roughly proportional to oxygen saturation, the second equation can be revised as follows:
where Sao2and SV̄o2are oxygen saturation at the artery and mixed venous blood, respectively.
Statistical Analysis
Data are presented as mean ± SD. Using analysis of variance with repeated measures, mean values were compared across different settings. When significance was observed, the mean values were tested by multiple comparison with the Bonferroni correction. We evaluated the correlation between CONIand COTDwith linear regression and Bland-Altman analysis. 12,13To investigate which parameters contributed to the discrepancy between CONIand COTD, we also performed linear multiple regression analysis among Fio2, VT, VE, PEEP, peak inspiratory pressure, pH, Pao2, Paco2, PETco2, V̇co2, and SV̄o2. Statistical significance was set at P < 0.05.
Results
Blood gas and hemodynamic results are summarized in table 3. Minute ventilation was stable at all 12-ml/kg VTsettings. Regardless of ventilatory mode, the 6-ml/kg VTsettings resulted in higher Paco2, higher PETco2, and less V̇co2, compared with the 12-ml/kg VTsettings. During PSV, VTvalues (8.8 ± 2.6 ml/kg) decreased to between those for 12- and 6-ml/kg VTsettings, whereas minute ventilation was similar to that at the 12-ml/kg VTsettings. COTDvalues were similar at each 12-ml/kg VTsetting, although COTDvalues at the 6-ml/kg VTsettings were slightly larger in comparison. At high PEEP, COTDvalues were lower.
Levels of pressure control were 24 ± 7 (16–36) cm H2O with inspired VTof 12 ml/kg and 13 ± 4 (8–22) cm H2O with VTof 6 ml/kg. As a result, there was no difference in peak inspiratory pressure for VCV and PCV at either VTsetting (table 3).
Results of Bland-Altman analysis and linear regression analysis are shown in table 4for each ventilatory setting. When VTvalues were the same, Bland-Altman analysis characteristics between COTDand CONIwere almost identical (bias and precision: 12-ml/kg VTVCV, 0.18 and 1.04; 12-ml/kg VTPCV, 0.37 and 1.17; 6-ml/kg VTVCV, −1.67 and 1.06; and 6-ml/kg VTPCV, −1.64 and 1.19, table 4). Consequently, for the same VTvalues, CO data during both VCV and PCV were analyzed together.
When VTwas 12 ml/kg, a fair correlation was observed between CONIand COTD(fig. 1). The slope of linear regression was 1.02 (R = 0.63, fig. 1), and bias was small (0.28 l/min, fig. 2), although limits of agreement were wide (−1.78 to +2.34 l/min, fig. 2). This is the case with ventilatory setting of high Fio2or high PEEP (table 4). By contrast, when VTwas small (6 ml/kg), the CONIunderestimated the COTDwith a slope of 0.70 (fig. 1), a bias of −1.66 l/min, and limits of agreement of −3.9 to +0.58 l/min (fig. 2). During PSV, the correlation between CONIand COTDwas also close to identical (slope = 1.07, R = 0.63, bias = 0.52 l/min, table 4). With the loop maximally expanded, the CONIcorrelated moderately with COTD(slope = 1.05, bias = 0.48, table 4); however, with the loop fully retracted, CONIoverestimated COTD(slope = 1.23, bias = 1.30, table 4). Linear multiple regression analysis revealed that the setting most affecting the discrepancy between CONIand COTDwas minute ventilation (R = 0.616).
Figure 3shows a relation between changes in COTDand those in CONIwhen PEEP was increased during VCV and 12-ml/kg VT. When average PEEP was increased from 4.2 to 14.0 cm H2O, both COTDand CONIdecreased. Both values moved in identical directions in all patients but one. The value of CVP increased from 7.9 ± 2.5 to 10.4 ± 2.4 mmHg at higher PEEP, and pulmonary capillary wedge pressure also increased from 9.8 ± 2.2 to 11.8 ± 2.1 mmHg (table 3).
Discussion
The main findings of this study are as follows. (1) During mechanical ventilation with large constant VTor during PSV, CO measurements obtained by CO2rebreathing technique correlate with those obtained by thermodilution method. (2) When minute ventilation is large, the accuracy of the CO2rebreathing technique is not affected by a selection of VCV, PCV, spontaneous breathing (PSV), PEEP, or Fio2. (3) When VTand minute ventilation are reduced, the CO2rebreathing technique underreports CO. (4) CO measurements are accurate when the rebreathing loop is maximally expanded but is overestimated when the loop is fully retracted.
Clinical Implications
Using partial CO2rebreathing, CO can be measured noninvasively. 3,4However, there have been few clinical reports, on the accuracy of this technique. 5–7We need to confirm that it provides effective monitoring for critically ill patients and discover parameters that might affect accuracy. Our results suggest that at a large VTsetting and with constant minute ventilation, CO measurements obtained from this technology correlate fairly with those from the thermodilution method. When inspired VTis set at 12 ml/kg and respiratory rate is set at 10–12 breaths/min, which results in an actual minute ventilation of 0.13–0.14 l · min−1· kg−1, the linear regression slopes for CONIand COTDwere almost identical (1.01:1.05). Bias analysis also indicated small bias and moderate precision (fig. 2), while accuracy was consistent regardless of ventilatory mode (VCV or PCV), PEEP, or Fio2. Correlation of results from CONIand COTDwas also satisfactory during PSV (table 4). These observations suggest that this CO2rebreathing technique is reliable both with large constant VTand during PSV. In addition, because the maximally expanded loop did not affect accuracy (table 4), rather than it being necessary to strictly adjust the loops, there may be some leeway in adjusting them for the maximal expected VT. In contrast, when the rebreathing loop was set too short for a given VT, CONImeasurements had greater values than those obtained by COTD(table 4). This may be due to the small changes in PETco2that occur with the shortest loop during CO2rebreathing, when a slight amount of noise would likely generate large errors.
To our surprise, when VTwas small (6 ml/kg), CONImeasurements showed consistently lower values than those produced by COTD, resulting in a linear regression slope of 0.70 and a negative value of bias (figs. 1 and 2). Low VT(6 ml/kg) is currently recommended for ventilator management in acute respiratory failure, 14so attention needs to be drawn to the lack of reliable measurement using CONIat the low VTsetting. Reasons for these discrepant results have not been clarified, but there are several possible explanations.
First, after we adjusted the length of rebreathing loops for high VT, when VTwas decreased, results may have been affected because the loop had become relatively too long. However, we found that the maximally expanded loop did not make CONImeasurements less accurate (table 4). This finding suggests that the combination of long loop and small VTare unlikely to impair the accuracy of CONI.
Second, at small VTsettings, PETco2increased to almost 60 mmHg in several patients. The software (version 3.1) that we used suspends rebreathing when the baseline PETco2is greater than 65 mmHg or PETco2is greater than 80 mmHg during CO2rebreathing. It could be that the linearity between Caco2and PETco2is less accurate when PETco2is extremely high.
Finally, the assumed constancy of mixed venous CO2content may be false for some time after VTand minute ventilation are changed. The measured values of V̇co2were smaller at low VTthan at high VT(table 3). Although we waited for 15 min, this may not have been enough time for CO2stores to reach a steady state, which is 100 times larger than oxygen stores. 9In addition, the time course of the increase in Paco2after abrupt decrease of ventilation is much slower than the rate of decrease after abrupt increase of ventilation. 9These facts suggest that CO2stores and mixed venous CO2content may continue to change even after Paco2and PETco2seem to have reached plateau values. If this is the case, the accuracy of the CO2rebreathing technique may be compromised when there are abrupt changes in minute ventilation and V̇co2. Further study is needed to find out exactly what happens after these sudden changes and whether these mechanisms affect the accuracy of the CO2rebreathing technique.
Limitations
The current study has several limitations. First, the patients in our study were sedated and paralyzed initially, resulting in constant VTand stable V̇co2. Even during PSV, they breathed quietly with small variation in VT. Therefore, our results may not be directly extrapolated to populations of patients whose VTand V̇co2are changing. 6Secondly, our patients had relatively normal lung mechanics (respiratory system compliance, 45.4 ± 12.8 ml/cm H2O; resistance, 11.2 ± 4.1 cm H2O · s · l−1), and their hemodynamics had been stabilized at time of entry into the study. In more seriously compromised patients, the accuracy may be quite different. To corroborate the relevance of our findings for acutely ill and ventilator-dependent patients, it is prudent to perform further studies. Third, we did not examine how the ventilatory pattern alterations affect the assumptions underlying the fundamental equation of the NICO2technique:e.g. , constant VD/VT, constant CO, and constant mixed venous CO2content during the CO2rebreathing procedure. Finally, it remains to be clarified whether the impaired accuracy of CONIwith small VTresults from small VTitself or from reduced minute ventilation. During PSV, when VTwas smaller (8.8 ± 2.6 ml/kg) but minute ventilation was similar to that at the high VTsettings, CONIand COTDvalues correlated fairly (y = 1.07x); we speculate that if normocapnia is sustained by adjusting the respiratory rate, the accuracy of the CONItechnique can be maintained at small VT.
In conclusion, noninvasive measurement of CO using CO2rebreathing is reliable with a bias of less than 0.5 l/min and a precision of 1 l/min when the tidal volume is large and constant, regardless of ventilatory modes. However, at small tidal volume, the rebreathing system underreports CO, compared with the conventional thermodilution technique.