Pulmonary Capillary Blood Flow and Cardiac Output Measurement by Partial Carbon Dioxide Rebreathing in Patients with Acute Respiratory Distress Syndrome Receiving Lung Protective Ventilation

The protocol was approved by our local ethics committee (Marseille University Hospital, Marseille, France), and informed consent was obtained from next of kin. This study was performed in a 12-bed medical intensive care unit in a university teaching hospital. We prospectively enrolled twenty consecutive mechanically ventilated patients meeting the American-European Consensus Conference criteria of ARDS.24Exclusion criteria were age under 18 yr, pregnancy, a history of chronic respiratory failure, hemodynamic instability defined by rapid change in mean arterial pressure, renal replacement therapy, bronchopleural fistula, almitrine infusion, and inhaled nitric oxide.

All patients received volume-controlled mechanical ventilation in a semirecumbent position (Evita XL; Dräger Medical, Lübeck, Germany). Inspiratory gases were automatically heated and fully humidified with a MR 850 device (Fisher & Paykel, Auckland, New Zealand). An endotracheal aspiration was performed before the investigation by using a closed-suction system (Steri-cath® 16-French; Portex, Keene, NH). Pulmonary aspiration was not repeated during the study period. A recruitment maneuver was performed after suctioning (40 cm H²O for 30 s) to standardize the lung volume history. Respiratory settings included a tidal volume of 6 ml · kg⁻¹of predicted body weight with a 0.4-s end-inspiratory pause. Respiratory rate was adjusted to complete expiratory lung washout and to maintain Paco²at 35–55 Torr and pH greater than 7.25. After completion of a quasi-static pressure volume curve by using the Evita XL tool, the positive end-expiratory pressure level was set at + 2 cm H²O higher than the low inflection point when present or at 10 cm H²O if not. Continuous infusion of midazolam, sufentanil, and ketamine if needed, was adjusted to provide a Ramsay scale score of 6 without spontaneous breathing efforts.25Mean arterial pressure was recorded through a 20-gauge radial artery cannula (Plastimed 4 French; Saint Leu la Fôret, France) and maintained constant higher than 70 mmHg. Fluid requirement and the need for vasopressor were assessed before entering the study.

Cardiac output was measured by continuous thermodilution using a 7.5-French pulmonary artery catheter (Vigilance II Monitor; Edwards, Irvine, CA). The pulmonary artery catheter was introduced via  the right internal jugular vein into the pulmonary artery by using the pressure signal obtained at the distal port. The proximal pressure port displays the right atrial pressure. The correct position of the pulmonary artery catheter was confirmed if the pulmonary artery occlusion pressure remained below the diastolic pulmonary pressure and if the distal tip of the catheter was located in the proximal pulmonary artery on the chest radiography. A 10-cm thermal filament continuously delivered heat into the bloodstream according to a pseudorandom binary sequence.26The change in blood temperature was detected by a thermistor and correlated with the input sequence to produce a thermodilution washout curve.27Cardiac output corresponds to the area under the curve. In critically ill patients, continuous thermodilution provides good agreement with intermittent bolus thermodilution or with dye dilution method.28–32We thus considered for the purpose of this study that the continuous thermodilution method is the reference. The monitor was set in the trend mode, updating cardiac output values every minute. We considered the average value of the last 3 min for statistical analyses. Central venous pressure and pulmonary artery occlusion pressure were recorded in the semirecumbent position by using the midaxillary line as zero reference. Arterial and mixed venous blood gases were analyzed immediately after drawing blood with a RapidPoint® 405 instrument (Bayer Health Care, Sudbury, United Kingdom). After inclusion, the inspired oxygen fraction was increased to 100% during 15 min to determine the true pulmonary shunt fraction according to the Berggren equation5:

where Q^Sis the pulmonary shunt blood flow, Q^Tis the total pulmonary blood flow, C^c'O²is end-capillary oxygen content (C^c'O²= hemoglobin · 1.34 + P^AO²· 0.0033), C^aO²is arterial oxygen content (Cao²= hemoglobin · Sao²· 1.34 + Pao²· 0.0033), and C^vO²is mixed venous oxygen content (Cvo²= hemoglobin · SvO²· 1.34 + Pvo²· 0.0033). We assumed that the end-capillary oxygen saturation at an inspired oxygen fraction of 1 was 100%. Although breathing 100% oxygen for a short period may induce alveolar collapse by denitrogenation, it has been reported that a sufficiently high positive end-expiratory pressure level impairs the formation of atelectasis and thus prevents the increase in pulmonary shunt.33,34Respiratory settings were kept constant throughout the investigation, and the protocol duration was short; therefore, we assumed that the pulmonary shunt fraction remained constant, and we considered the initial shunt fraction for all other measurements. Thereafter, inspired oxygen fraction was resumed at baseline level. Pulmonary capillary blood flow measured by thermodilution was calculated as cardiac output minus shunt.

The NICO²monitor version 3.0 is a fully automated device placed between the Y piece of the ventilator and the endotracheal tube. It includes a calibrated mainstream infrared carbon dioxide sensor, a differential pressure flow sensor, and a pneumatically controlled loop acting as an adjustable rebreathing bag (from 150 to 400 ml). Complete function and procedure of the NICO²monitor have been described elsewhere.11,13Briefly, the additional dead space obtained during the rebreathing period raises Petco²slightly (typically by 3–5 mmHg) and lowers Vco². This rebreathing period is activated for 35 s before resuming basal condition for 60 s. Finally, a stabilization period of 85 s completes the NICO cycle, which is cyclically repeated every 3 min. Pulmonary capillary blood flow measured by partial carbon dioxide rebreathing corresponds to the difference between nonrebreathing and rebreathing periods such that:

PCBF^REB(ml · min⁻¹) =−ΔVco²/S.ΔPetco², where S is the slope of the carbon dioxide dissociation curve from hemoglobin.35Cardiac output is then calculated by adding a noninvasive estimation of the shunt. For this purpose, the NICO²software plotted the manually entered Fio², Pao²(or Spo²), and hemoglobin values on Nunn's iso-shunt line.36To compare the estimation of shunt by NICO²with the true shunt (Q^S/Q^T), we used the following formula:

Shunt^REB(%) =[1 − (PCBF^REB/CO^REB]· 100

We also determined the dead space fraction (V^D/V^T) according to a breath-by-breath analysis of volumetric capnography. Paco²was manually entered into the NICO²monitor, which continuously displayed the V^D/V^Tand other volumetric indices.

After a 2-h period of hemodynamic stability (i.e. , mean arterial pressure constant higher than 70 mmHg), we performed seven pairs of measurements of cardiac output and pulmonary capillary blood flow, repeated every 20 min over a 2-h period. Finally, 140 pairs of simultaneous measurements by thermodilution and partial carbon dioxide rebreathing were available. No other interventions were allowed during the study period.

Distribution of the data were determined by using a Kolmogorov-Smirnoff test. Data are expressed as mean ± SD or median ± interquartile range (25–75%) according to normality. We used analysis of variance on ranks for repeated measurements (RM ANOVA) to determine significant changes in hemodynamic variables through the study period. Correlation between methods for the determination of cardiac output and pulmonary capillary blood flow was calculated by using a Spearman Rank Order test (r²) because of the nonparametric distribution of data. Agreement was determined according to the method described by Bland and Altman for repeated measurements where the true value varies.37We calculated bias (mean difference), precision (SD of the difference), and limit of agreement (95% confidence interval [CI]) for cardiac output and pulmonary capillary blood flow. For this purpose, we used the R statistical software (The R Project for Statistical Computing, Vienna, Austria).38We also determined the percentage error of agreement (2 · SD/mean of the reference method).39 

Finally, we performed a linear mixed model analysis to identify variables associated with difference in the measurement of cardiac output and pulmonary capillary blood flow between methods.40The following variables were introduced in the two models: shock (as defined by the presence of norepinephrine), dead space, arteriovenous oxygen content difference (Davo²), mean pulmonary arterial pressure, baseline cardiac output, and total positive end-expiratory pressure. The estimated fixed effect of each parameter was quantified with the use of estimated value and SE. The dependant variables were the difference in cardiac output measurement for the first model and the difference in pulmonary capillary blood flow measurement for the second model. Significance level was set at P < 0.05. All statistical analyses were performed with SPSS 15 (SPSS Software, Chicago, IL).

Characteristics of the patients at inclusion are presented in table 1. Sixteen patients (80%) received norepinephrine, and none received dobutamine. Respiratory, hemodynamic, and NICO-related parameters at inclusion are presented in table 2. No significant changes in metabolic, respiratory, and hemodynamic parameters (heart rate, mean arterial pressure, central venous pressure, pulmonary artery occlusion pressure, cardiac output, and pulmonary capillary blood flow) were observed during the study period (analysis of variance for repeated measurements).

Cardiac output values ranged from 3.2 to 11.1 1 · min⁻¹with thermodilution and from 2.7 to 11.3 l · min⁻¹with partial carbon dioxide rebreathing. Mean cardiac output values were 6.7 ± 1.9 l · min⁻¹with thermodilution and 5.8 ± 1.7 l · min⁻¹with partial carbon dioxide rebreathing. The correlation coefficient between methods was r²= 0.42 (P < 0.001, fig. 1). Bias was 0.8 ± 1.2 l · min⁻¹, and limit of agreement ranged from −2.1 to 3.7 l · min⁻¹(fig. 2). Percentage error was 36%. In linear mixed analysis, dead space, Davo², and baseline cardiac output were independently associated with difference between methods (table 3).

Pulmonary capillary blood flow values ranged from 2 to 8.5 l · min⁻¹with thermodilution and from 2.2 to 9.9 l · min⁻¹with partial carbon dioxide rebreathing. Mean pulmonary capillary blood flow values were 4.6 ± 1.3 l · min⁻¹with thermodilution and 4.8 ± 1.4 l · min⁻¹with partial carbon dioxide rebreathing. The correlation coefficient between methods was r²= 0.45 (P < 0.001, fig. 3). Bias was −0.1 ± 0.8 l · min⁻¹, and limit of agreement ranged from −2.2 to 1.9 l · min⁻¹(fig. 4). Percentage error was 35%. In linear mixed analysis, dead space and mean pulmonary arterial pressure were independently associated with difference between methods (table 4).

Pulmonary shunt (Q^S/Q^Tat FIO²1) values ranged from 22 to 40% with thermodilution and from 6 to 28% with partial carbon dioxide rebreathing. Mean shunt values were 30 ± 6% with thermodilution and 18 ± 6% with partial carbon dioxide rebreathing. The correlation coefficient between methods was r²= 0.17 (P = 0.046). Bias and limit of agreement were, respectively, 12 ± 6% and −0.5 to 24%.

The main findings of this study are that, in a population of ARDS patients receiving lung protective ventilation, (1) partial carbon dioxide rebreathing cannot replace thermodilution for the measurement of cardiac output or pulmonary capillary blood flow because it slightly overpasses the boundary of clinical relevance, and (2) the main factors that contribute to differences between methods were the mean pulmonary arterial pressure, the dead space fraction, the arteriovenous oxygen content difference, and the baseline cardiac output.

In the current study, the partial carbon dioxide rebreathing method underestimated cardiac output. Most previous studies have reported similar results.15,17,18Because the partial carbon dioxide rebreathing method is underlaid by a number of assumptions, several factors may have contributed to this difference.41 

The change in Petco²during rebreathing is supposed to mirror the change in Caco². However, the relation between Petco²and Caco²is not a straight line but curvilinear, convex at higher values. Thus, depending on the nonrebreathing Petco²value, the change in Caco²will differ. For instance, a high Petco²will be associated with a lower change in Caco².

The difference between Petco²and Paco²should remain constant during the rebreathing and the nonrebreathing periods.11,13Basically, this difference depends on the magnitude of dead space, which is largely increased in ARDS patients.42In the setting of high dead space fraction, alveolar ventilation is reduced and the time required for equilibrium after a rebreathing period is longer. Thus the difference between Petco²and Paco²will not remain constant. It must been underlined that the low tidal ventilation employed in this study may have also increased dead space.21 

The change in mixed venous carbon dioxide content during rebreathing is supposed to be negligible. This assumption is supported by the large carbon dioxide storage capacity of the body.11,13Odenstedt et al.  reported a slight increase in Pvco²(2.5 ± 0.3%) during the rebreathing period.17In this study, we did not record Pvco², but we would expect at least similar variation. Nillson et al.  stated that the range of Pvco²variation is not large enough to influence cardiac output's measurement.16 

Finally, our data show a large and systematic underestimation of pulmonary shunt by using the partial rebreathing method as compared with venous admixture. Similar underestimation has been reported already in critically ill patients.17One of the sources of error in the shunt's estimation may be in the computation of arterioenous oxygen content difference. The NICO²software uses a constant value at 5 ml · dl⁻¹, which is largely above the range observed in the ARDS population. In the current study, the mean Davo²was 3.3 ± 0.6 ml · dl⁻¹. Hence, we are not surprised to identify Davo²as a factor contributing to inaccuracy in the measurement of cardiac output. The lower the Davo², the larger the error in measurement. Nevertheless, the contribution of shunt in the underestimation of cardiac output is weak, as confirmed by the poor weight of its estimated value in the linear mixed model. For instance, supposing a cardiac output value of 5 l · min⁻¹, a shunt value of 20%, and an error in the estimation of shunt of 10%, the change in the cardiac output measurement will be about 0.5 l · min⁻¹.

Because an increase in pulmonary shunt and dead space fraction is a common finding during ARDS,43De Abreu et al.  have investigated their respective effect on the precision of the method.44In an animal model, they demonstrated that increasing both shunt and dead space induces error in the estimation of Caco². They also demonstrated that during high cardiac output conditions, the higher Pvco²values tend to overestimate Petco²during rebreathing. By using computer models of emphysema and pulmonary embolism, Yem et al.  have identified four sources of systematic error: ventilation-perfusion imbalance, alveolar-proximal airway Pco²gradient, venous blood recirculation, and duration of the rebreathing period.45In the current study, we demonstrated in a clinical setting that dead space, Davo², and baseline cardiac output have a significant impact on the accuracy of the cardiac output determination by partial rebreathing. According to the estimated values of parameters entered in our linear mixed model, dead space seems to be the most important source of error.

Contrasting with the results on cardiac output, the measurement of pulmonary capillary blood flow is somewhat overestimated by the partial rebreathing method, but agreement with thermodilution is closer. Bear in mind that the partial carbon dioxide rebreathing algorithm has been elaborated to measure pulmonary capillary blood flow and not cardiac output. As a result, it is important to note that the accuracy of the algorithm is correct in most measurements. In the current study, we found that the accuracy of the method is reduced in the setting of high dead space fraction and/or a high mean pulmonary artery pressure; both are associated with the severity of the disease. Nevertheless, the clinical utility of the bedside measurement of pulmonary capillary blood flow in ARDS patients remains poorly investigated.12De Abreu et al.  have suggested a place in the titration of positive end-expiratory pressure, challenging the hypothesis that excessive positive end-expiratory pressure level will be associated with a worse pulmonary capillary blood flow.46Whether pulmonary capillary blood flow should be considered as a resuscitation endpoint remains to be investigated.

From a clinical point of view, one major question is whether the measurement of cardiac output or pulmonary capillary blood flow by partial rebreathing is sufficiently close to the “true value” such that it would provide helpful assistance to the physician. For this purpose, two major criteria have been proposed to address the clinical relevance of a new method: SD of bias less than 1 l · min⁻¹and percentage error of agreement no greater than 30%.39,47In our population, percentage error was 36% for cardiac output and 35% for pulmonary capillary blood flow. SD of bias overlapped 1 l · min⁻¹for cardiac output (1.2 l · min⁻¹) but not for pulmonary capillary blood flow (0.8 l · min⁻¹). On the basis of the limit of agreement that we observed, 95% of the pulmonary capillary blood flow values measured by partial rebreathing ranged between −2.2 and 1.9 l · min⁻¹around the true value. In the same way, 95% of the cardiac output values measured by rebreathing ranged between −2.1 and 3.7 l · min⁻¹around the true value. Taken together, these results indicate that pulmonary capillary blood flow and cardiac output measurement by partial rebreathing should not be recommended stricto sensu  for routine monitoring.

Some limitations must be highlighted in this study. We used continuous and not intermittent thermodilution measurement, but equivalence has already been addressed in critically ill patients, as discussed in the Materials and Methods section. Although the shunt fraction was only measured once at the beginning of the 2-h period, making this a potential limitation of the study, we controlled most of respiratory, hemodynamic, and metabolic variables. We investigated ARDS patient who all received standardized mechanical ventilation with low tidal volume and positive end-expiratory pressure. As a result, this study presents the advantage of testing a new device under homogeneous ventilator settings. However, these results cannot be extrapolated to other settings or breathing patterns. In the same way, our data were recorded under stable hemodynamic conditions of arterial pressure and cardiac output, and most of our patients received norepinephrine; these results may not be extrapolated to other hemodynamic situations. The crude mortality of our population was 60%, but it seems in accordance with the literature regarding the number of organ failure and the characteristics of our patients.

In conclusion, partial carbon dioxide rebreathing is a semicontinuous, noninvasive, and automated method to quantify pulmonary capillary blood flow. In the situation of ARDS patients ventilated with lung protective settings, its accuracy is not sufficient to replace thermodilution for the routine measurement of pulmonary capillary blood flow or cardiac output. The main clinical factors associated with the inaccuracy of the rebreathing method were baseline cardiac output, dead space, and Davo²for cardiac output measurement; only dead space and mean pulmonary arterial pressure were associated with the inaccuracy of the pulmonary capillary blood flow measurement. Further research will be required to improve the algorithm of partial carbon dioxide rebreathing for ARDS patients.